compararea internaŢionalĂ a...
TRANSCRIPT
Revista Română de Statistică - Supliment nr. 4 / 2019
SUMAR / CONTENTS 4/2019REVISTA ROMÂNĂ DE STATISTICĂ SUPLIMENT
UNELE CONCEPTE ŞI MODELE ECONOMETRICE DE ANALIZĂ A PERFORMANŢEI MACROECONOMICE 3SOME CONCEPTS AND ECONOMETRIC MODES OF MACROECONOMIC PERFORMANCE ANALYSIS 12Prof. Constantin ANGHELACHE PhD Alexandra PETRE (OLTEANU) PhD StudentCristian OLTEANU PhD Student
PARTICULARITĂŢI ALE UTILIZĂRII CURBEI PHILLIPS ÎN STUDIUL EVOLUŢIEI ŞOMAJULUI 21PARTICULARITIES OF USING THE PHILLIPS CURVE IN THE STUDY OF UNEMPLOYMENT EVOLUTION 33Assoc. prof. Mădălina-Gabriela ANGHEL PhD Ștefan Gabriel DUMBRAVĂ Ph.D StudentDaniel DUMITRU Ph.D Student
PRINCIPALELE METODE DE PREVIZIUNEA ECONOMICĂ 45THE MAIN METHODS OF THE ECONOMIC PROVISION 53Andreea – Ioana MARINESCU PhD Student
ASPECTE TEORETICE PRIVIND IDENTIFICAREA STRUCTURII PARŢIALE DIN SUBMODELE 61THEORETICAL ASPECTS REGARDING THE IDENTIFICATION OF PARTIAL STRUCTURE IN SUBMODES 71Radu STOICA Ph.D Student
ELEMENTE ESENŢIALE PRIVIND MODELAREA ŞI COMBINAREA SUBSISTEMELOR 81ESSENTIAL ELEMENTS ON MODELING AND COMBINATION OF SUBSYSTEMS 92Prof. Radu Titus MARINESCU PhDProf. Constantin ANGHELACHE PhDAssoc. prof. Mădălina-Gabriela ANGHEL PhD
Alexandru BADIU PhD Student
www.revistadestatistica.ro/supliment
Romanian Statistical Review - Supplement nr. 4 / 20192
UTILIZAREA INDICILOR LASPEYRES ȘI PAASCHE ÎN DETERMINAREA FORMULEI INDICELUI DE PREȚ 102THE USING OF THE LASPEYRES AND PAASCHE INDICES IN THE DETERMINING OF THE PRICE INDEX FORMULA 113Doina BUREA PhD Student
COMPARAREA INTERNAŢIONALĂ A PERFORMANŢEI MACROECONOMICE 124THE INTERNATIONAL COMPARABILITY OF THE MACROECONOMIC PERFORMANCE 138Ștefan Gabriel DUMBRAVĂ Ph.D StudentAlexandra PETRE (OLTEANU) PhD StudentCristian OLTEANU PhD StudentTudor SAMSON PhD Student
Revista Română de Statistică - Supliment nr. 4 / 2019 3
Unele concepte şi modele econometrice de analiză a performanţei macroeconomice
Prof. univ. dr. Constantin ANGHELACHE ([email protected])
Academia de Studii Economice din București / Universitatea „Artifex” din BucureștiDrd. Alexandra PETRE (OLTEANU) ([email protected])
Academia de Studii Economice din BucureștiDrd. Cristian OLTEANU ([email protected])
Academia de Studii Economice din București
Abstract Pentru analiza performanțelor macroeconomice se utilizează un sistem de indicatori corelați care în urma intrerpretarii dau sensul evoluției viitoare a economiei. De regulă, se utilizează modele macroeconomice care, puse în aplicare asigură obținerea de parametri, pe baza cărora se poate prognoza perspectiva de evoluție macroeconomică. Un punct de plecare în defi nirea conceptului de model macroeconomic îl constituie ideea unui proces general stochastic care a generat datele observate. În mod practic este ușor de detaliat particularitățile respective. Astfel, pot fi construite modele pentru analiza consumului și a investițiilor sau modele bazate pe teoria statistică, cum este modelul autoregresiv cu vector dimensional redus. În construirea unui model macroeconomic se poate urma o procedură care cuprinde: defi nirea și analiza de subsectoare ale economiei pentru alegerea variabilelor; construirea de modele parțiale relevante și apoi combinarea acestor submodele în scopul de a obține o reprezentare modelară a întregii economii. În practică trebuie să procedăm gradual pentru a cuprinde toate elementele submodelelor considerate. De stabilirea modelelor macroeconomice s-au ocupat o serie de cercetători, context în care, în articol sunt analizate aceste teorii. Cuvinte cheie: indicator, model econometric, variabile, performanță macroeconomică, analiză Clasifi carea JEL: C10, C82, E70
Introducere
Autorii au efectuat un amplu studiu asupra conceptelor care stau la baza construirii modelelor de analiză a performanțelor macroeconomice luând în considerare evoluția în timp a acestor concepte. Sunt exprimate evoluțiile în acest domeniu, concretizându-se prin sinteză o serie de aspecte care apoi au stat la baza construirii de modele pentru analiza macroeconomică. În articol se precizează că un concept al modelării macroeconomice se bazează pe
Romanian Statistical Review - Supplement nr. 4 / 20194
ipoteza că dacă modelul A este parte a modelului B, este posibil să identifi căm date despre modelul B din modelul A. Sunt expuse puncte de vedere ale clasicilor teoriei modelării reliefându-se elementele de care trebuie să se țină seama în procesul modelării pentru analize macroeconomice. Rând pe rând sunt prezentate principalele aspecte ale modelelor lui Jan Timbergen apoi modelului Haavelmo sau de dată mai recentă elemente econometrice pe care se bazează modelele macroeconomice. În cadrul articolului se face o amplă prezentare a conceptelor care stau la baza construirii modelelor de analiză macroeconomică.
Literature review Anghelache, Anghel și Marinescu (2019) au analizat o serie de elemente referitoare la indicatorii avuției naționale care au implicații asupra performanțelor macroeconomice. Anghelache și Anghel (2018) au prezentat o principalele metode și modele econometrice utilizate în analizele macroeconomice. Anghelache, Anghel și Dumitru (2018) au evidențiat modul de calcul al indicatorilor macroeconomici de rezultate. Anghelache, Isaic-Maniu, Mitruț, Voineagu și Dumbravă (2007) au evidențiat principalele aspecte teoretice și practice ale analizei macroeconomice. Bardsen, Nymagen și Jansen (2005) au studiat utilizarea econometriei în modelarea macroeconomică. Benjamin, Herrard, Hanee-Bigot și Tavere (2010) au studiat utilizarea modelelor econometrice în prognozele macroeconomice. Carrasco, Florens și Renault (2004) au studiat probleme liniare inverse în econometria structurală. Dougherty (2008) a prezentat noțiunile fundamentale ale econometriei. Fernandez-Villaverde și Rubio-Ramirez (2009) au studiat aspecte esențiale ale macroeconomiei. Johansen și Nielsen (2010) au analizat probabilitatea de inferență pentru un model autoregresiv.
Metodologia cercetării, date, rezultate și discuții
Modelarea macroeconometrică are scopul de a evidenţia comportamentul empiric al unui sistem economic real. Asemenea modele sunt concepute ca ecuaţii interconectate estimate pe bază de serii dinamice folosind tehnici statistice sau econometrice. Un punct conceptual de plecare îl constituie ideea unui proces general stocastic care a generat datele observate privind economia şi că acest proces poate fi rezumat în termeni de distribuţie cu mai multe variabile de probabilitate ale variabilelor observabile aleatoriu într-un sistem de ecuaţii stochastic. Cu toate acestea, este posibil să facem o abordare agregată pentru a reprezenta comportamentul unor variabile importante (de exemplu rata infl aţiei, creşterea PIB, rata şomajului) într-un model. Estimarea unor
Revista Română de Statistică - Supliment nr. 4 / 2019 5
asemenea modele econometrice poate fi bazată pe teoria statistică (modele autoregresive cu vector dimensional scăzut VAR). În mod practic, este uşor pentru utilizator să detalieze particularităţi ale modelului (modelarea separată a consumului şi investiţiei) spre a modela simultan toate ecuaţiile. Modelele care sunt utilizate pentru analiza impactului bugetului guvernamental asupra economiei sunt sisteme ample de ecuaţii. Chiar în cazul în care modelul foloseşte numai o variabilă din ţintele de plecare, precum ţinta prognozată de infl aţie, alegeri de politici sunt făcute împotriva fondului unei analize mai largi a efectelor ratei dobânzii asupra economiei (cursuri de schimb nominale şi reale, creştere a producţiei şi stabilitate fi nanciară). Creatorii de modele au avut ca sarcină obiectivul de a stabili o bună practică şi a dezvolta proceduri operaţionale pentru construirea de modele care asigură că rezultatul fi nal al modelarii este solid şi util. Contribuţii importante în literatură pe tema modelelor macroeconometrice includ lucrările lui Christ (1966), Klein (1983), Fair (1984, 1994), Klein (1999) sau cercetările lui Bodkin (1991) şi Wallis (1994) . În construirea unui model macroeconomic sugerăm următoarea procedură operaţională: • defi nirea şi analiza de subsectoare ale economiei prin alegeri
relevante de variabile. • construirea (prin condiţionare) de modele parţiale relevante
pe care le putem numi modele de tip A, prin considerarea de variabile exogene şi endogene.
• combinarea acestor submodele în scopul de a obţine un Model B al întregii economii.
Conceptul se bazează pe ipoteza că, fi ind dat Model A ca parte a Model B, e posibil să afl ăm date despre Modelul B din Modelul A. Alternativa la aceasta ipoteză ar putea reprezenta un tip de creaţionism fără ca macroeconometria să fi e restricţionată la modele agregate. De exemplu sunt proprietăţi care pot fi descoperite folosind procedura de cointegrare din Model B. Aceasta are la bază un corolar al teoriei sistemelor cointegrate: „orice combinaţie liniară non-zero de vectori de cointegrare este de asemenea un vector de cointegrare”. În cazul cel mai simplu, dacă sunt doi vectori cointegratori în Modelul B, există întotdeauna o combinaţie liniara a acestor vectori care va elimina una din variabile. Analiza cointegrării unei submulţimi de variabile (Model A), excluzând acea variabilă, va avea ca rezultat un vector cointegrator corespunzător la acea combinaţie liniară. Astfel, în ciuda faptului că este o proprietate a Modelului B, analiza cointegrată a subsistemului (Model A) identifi ca un vector de cointegrare. Dacă această identifi care are sens economic sau nu rămâne o
Romanian Statistical Review - Supplement nr. 4 / 20196
problemă deschisă, orice soluţie trebuie argumentată în fi ecare caz. Analiza pe baza unui astfel de sistem înseamnă identifi carea funcţiei de consum, ca relaţie cointegratoare, identifi carea relaţiilor cointegratoare folosite la stabilirea de preţuri şi salarii, existenţa unei rate naturale a şomajului şi relevanţa termenilor orientaţi către viitor privind stabilirea preţului şi a salariului. După cum sublinia Johansen (2002), aceasta este o situaţie difi cilă a procedurii anterioare: o teorie generală pentru cele trei etape va conţine criterii şi condiţii care sunt formulate pentru întreg sistemul. Modelarea sofi sticată pe porţiuni poate fi considerată ca un mod de gradualism – căutând să stabilească submodele care reprezintă structura parţială; adică, modele parţiale care sunt unidimensionale la extinderi ale perioadei de sondaj, la modifi cări în altă parte în economie şi rămân aceleaşi pentru extensii ale mulţimii de informaţii. Gradualismul implică şi o eventuală revizuire a unui submodel. Revizuirile sunt uneori declanşate de eşecul de prognoză. Mai multe motive banale includ revizuiri şi extensii de date care permit specifi caţii de model mai precise şi îmbunătăţite. Dialogul între creatorii de modele şi utilizatorii acestora au ca rezultat, de asemenea, revizuiri. Astfel utilizatorii experimentaţi de modele sunt capabili de obicei să evidenţieze implicaţiile unei singure specifi caţii de ecuaţie (sau submodel) asupra proprietăţilor întregului model. În mod evident, gradualismul nu înlătură testarea aprofundată a unui submodel. Dimpotrivă, primele două etape în procedura operaţională de mai sus nu cere ca să cunoaştem întreg modelul şi testarea acelor condiţii are o anumită atracţie intuitivă din moment ce viaţa reală asigură “o nouă dovadă” prin noi date şi prin “experimente naturale”, prin schimbări de regim cum sunt modifi cări în dereglementarea guvernamentală sau fi nanciară. Pentru cele trei etape, am putea considera în principiu modelul complet ca ultimă extensie a mulţimii de informaţii şi astfel stabilirea structurii sau structurii parţiale, ceea ce reprezintă un mod de a continua constatarea lui Johansen. Practic, ştim că întreg modelul este greu de realizat. Vom reduce activitatea la trasarea grafi cului modelului sectorial cu aproximările simplifi cate ale Modelului B şi vom verifi ca ipotezele externe relevante ale modelului parţial în acelaşi cadru. În măsura în care funcţia asemănătoare a Modelului B simplifi cat reprezintă adecvat sau aproximează funcţia asemănătoare a Modelului complet B, nu e nici o problemă. E posibil de asemenea să confi rmăm întreaga procedura din moment ce este adevărat că Modelul A poate fi testat şi îmbunătăţit gradual cu noi informaţii, ceea ce este un mod de a căpăta acea cunoaştere care se compară cu darwinismul modern în ştiinţele naturale. Un motiv practic pentru concentrarea pe submodele e următorul: creatorii de modele pot avea un motiv întemeiat să studieze anumite ramuri ale economiei cu mai multă atenţie decât alţii. Pentru o bancă centrală care
Revista Română de Statistică - Supliment nr. 4 / 2019 7
ţinteşte infl aţia, se impune folosirea modelului procesului infl aţionist. Acesta
necesită modelarea atentă a aranjamentelor instituţionale sau condiţionate de
stabilire a preţurilor şi salariilor, evoluţia pe pieţele fi nanciare şi economia
reală cu scopul de a răspunde la un număr de întrebări importante: Există
o rată naturală a şomajului care indică şomajul precum şi infl aţia? Care e
importanţa probabilităţilor de infl aţie şi cum pot fi acestea modelate? Care
este rolul banilor în procesul infl aţionist?
Considerăm că pentru a răspunde la aceste întrebări – şi a proba
ipoteze concurente privind ştiinţa economiei tip ofertă (supply-side) e necesara
o modelare detaliată, bazată pe informaţii specifi ce economiei în studiu.
Luarea în considerare a simultaneităţii este într-o mare măsură o chestiune
de efi cienţă a estimării. Dacă există o compensare între această efi cienţă şi
problema reglementării mecanismelor economice, specialiştii modelării
macroeconometrice ar trebui să acorde prioritate ultimei probleme.
Istoria modelării macroeconomice începe cu economistul olandez Jan
Tinbergen care a construit şi estimat primele modelele macroeconometrice pe
la mijlocul anilor 1930 (Tinbergen 1937). Tinbergen a dovedit că cineva poate
transforma un sistem de ecuaţii intr-un model econometric al ciclului de afaceri
(business cycle) utilizând teoria economică pentru a deriva ecuaţii dinamice
motivate comportamental şi metode statistice pentru a le verifi ca. Totuşi, se
pare că există un acord universal conform căruia statistica este disciplină a
ştiinţei economiei şi econometriei bazat pe contribuţia economistului norvegian
Trygve Haavelmo în tratatul său “ Abordarea probabibilităţii în econometrie”.
Haavelmo a fost inspirat de către unii din cei mai mari statisticieni ai timpului.
Aşa cum evidenţiază Morgan (1990), a fost convertit la utilizarea ideilor de
probabilitate de către Jerzy Neyman şi a fost totodată infl uenţat de Abraham
Wald pe care Haavelmo l-a creditat ca sursă a înţelegerii teoriei statistice.
E important de subliniat că Haavelmo a recunoscut şi explicat în
contextul unui model economic că distribuţia cu mai multe variabile ale
tuturor variabilelor observabile pentru întreaga perioadă studiată asigură
cadrul general pentru deducţia statistică (Hendry 1989). Aceasta se aplică la
specifi carea/identifi carea, estimarea şi verifi carea ipotezei:
„Toate se reduc la unul şi acelaşi lucru, şi anume, studiul proprietăţilor
distribuţiei de probabilitate cu mai multe variabile ale variabilelor întâmplătoare
intr-un sistem stochastic de ecuaţii”.
Revoluţia probabilistică a lui Haavelmo a infl uenţat econometria.
Gândirea lui a fost imediat adoptată de Jacob Marschak – un savant de origine
rusă - care a studiat statistica cu Slutsky – ca agendă de cercetare pentru
Comisia Cowles din perioada 1943-47, prin reconsiderarea lucrărilor lui
Tinbergen despre ciclurile de afaceri citate anterior. Lui Marschak i s-a alăturat
Romanian Statistical Review - Supplement nr. 4 / 20198
un grup de statisticieni, matematicieni şi economişti, inclusiv Haavelmo. Munca lor a fost aceea de a stabili standardele econometriei moderne şi s-a regăsit în lucrările de econometrie de la Tintner (1952) şi Klein (1953) şi alţii. Activitatea Comisiei Cowles a stabilit bazele dezvoltării modelelor macroeconomice şi creării de modele care a devenit o largă „industrie” în SUA în următoarele trei decenii (Bodkin 1991 şi Wallis 1994). Aceste modele au fost în principal destinate prognozării pe termen scurt şi mediu, adică modelarea pentru prognozarea ciclurilor de afaceri. Primul model (Klein 1950) a fost creat cu scopul explicit de a implementa ideile lui Haavelmo în cadrul modelării lui Tinbergen pentru economia SUA. Ca şi modelul lui Tinbergen, acesta era un model în care Klein a pus accentul pe modelarea ecuaţiilor simultane. Mai târziu modelele au devenit sisteme ample în care se foloseau peste 100 ecuaţii pentru a descrie comportamentul unei economii industriale moderne. În asemenea modele, s-a luat în consideraţie mai puţin fi ecare specifi caţie econometrică şi simultaneitatea nu putea fi tratată în mod satisfăcător. Scopul de prognoza al acestor modele a însemnat că acestea erau evaluate în funcţie de performanţa lor. Când modelele au eşuat să prognozeze efectele asupra economiilor industriale ale şocurilor petroliere din 1973 şi 1979, modelarea macroeconomică şi-a diminuat poziţia sa, mai ales în SUA. În anii 1980, modelele macroeconomice au profi tat de progresele metodologice şi conceptuale în econometria bazată pe o serie dinamică (cronologică). Box şi Jenkins (1970) au creat şi popularizat un instrument pur statistic pentru modelarea şi prognoza de serii cronologice unidimensionale. Al doilea infl ux al metodologiei statistice în econometrie îşi are rădăcinile în studiul naturii nestaţionare a seriilor de date economice. Clive Granger în studiile sale de statistică a arătat într-o serie de lucrări valoroase, importanţa pe care o are faptul ca o ecuaţie econometrică să fi e echilibrată. O variabilă staţionară nu poate fi explicată printr-o variabilă nestaţionară şi viceversa (Granger, 1990). În plus, conceptul de cointegrare (Granger 1981, Engle and Granger 1987, 1991), respectiv o combinaţie liniară a două sau mai multe variabile nestaţionare poate fi staţionară (fi xă), s-a dovedit folositor şi important în modelarea macroeconometrică. În cadrul unui model VAR general, statisticianul Soren Johansen a creat (Johansen 1988, 1991, 1995b) modelele uzitate pentru testarea cointegrării într-un cadru multidimensional, trasând modul analitic al corelării canonice şi regresia de poziţie a lui Anderson (1951). De asemenea, s-a acordat o atenţie sporită rolului evaluării în econometria modernă (Granger 1990, 1999). Aşa-numita metodologie London School of Economics scoate în evidenţă importanţa modelelor de testare şi evaluare econometrică (Hendry 1993a, 1995a, Mizon 1995 şi Ericsson 2005).
Revista Română de Statistică - Supliment nr. 4 / 2019 9
Hendry (1989) pretinde că multe aspecte ale cercetării lui Haavelmo au fost ignorate o perioadă mare de timp. De exemplu, funcţia cu mai multe variabile pentru variabile observabile a fost recunoscută de Comisia Cowles ca esenţială în rezolvarea problemelor de deducţie statistică, dar ideile nu au infl uenţat strategiile de modelare empirică. În contrast, multe evoluţii în econometrie, după 1980, corespund cu aceasta şi cu alte aspecte ale programului de cercetare al lui Haavelmo. Aceasta este valabilă şi pentru rolul teoriei economice în econometrie, respectiv: „Modelele teoretice sunt instrumente necesare în încercările noastre de a înţelege şi a explica evenimente din viaţa reală (Haavelmo 1944). Dar orice explicaţii preferăm, nu se poate ignora că ele sunt toate invenţiile noastre artifi ciale într-o căutare de a înţelege viaţa reală, nefi ind adevăruri ascunse care urmează a fi descoperite”. ( Haavelmo 1944) Având acest punct de pornire, ne-am putea aştepta la fapte sau observaţii care corespund oricărei formulări precise derivate dintr-un model teoretic. Teoriile economice trebuie deci exprimate ca formulări probabilistice şi Haavelmo a considerat teoria probabilităţilor ca indispensabilă în formalizarea noţiunii de modele ca aproximări ale realităţii. Programul de cercetare al Comisiei Cowles s-a concentrat pe modele de ecuaţii simultane (SEMs) şi a pus accent pe problema identifi cării. În abordarea acestor aspecte, teoria economică joacă un rol important. Reprezentant de seamă al acestei tradiţii, Lawrence Klein scrie într-un studiu despre interacţiunea dintre statistică şi ştiinţa economiei, în contextul modelării macroeconometrice (Klein 1988), că abordarea creării de modele poate fi opusă analizei pure statistice, care e empirică şi nu strâns legată de teoria economică aşa cum e crearea de modele. Totuşi crearea de modele macroeconomice tradiţionale a fost şi criticată (Favero 2001). În timp ce metodologia LSE atribuie eşecul acelor modele macroeconomice timpurii absenţei dinamicii sau specifi caţiei greşite de model, alţi critici precum Robert Lucas şi Christopher Sims au pretins: „cauza este mai degrabă aceea că a avut o bază teoretică slabă”. Critica lui Lucas (Lucas 1976) pretinde că eşecul modelelor condiţionate e cauzat de modifi cări de regim ca rezultat al schimbărilor de politici şi modifi cări ale probabilităţilor (expectations). Critica se raportează la SEMs, dacă probabilităţile nu sunt modelate. Pe de altă parte, Sims (1980)a argumentat că SEMs include „restricţii de identifi care incredibile”: „restricţiile necesare pentru a pretinde exogenitatea anumitor variabile nu ar fi valabile într-un mediu unde agenţii optimizează ocazional”. Sims, pe de alta parte, susţine utilizarea unei autoregresii de vector de ordine joasă (low-order vector autoregression) faţă de analiza tip serie cronologică economică. Această abordare ar avea
Romanian Statistical Review - Supplement nr. 4 / 201910
avantajul că nu depinde de o separare “arbitrară” între variabile exogene şi endogene şi că nu cere „restricţii de identifi care incredibile”. În schimb Sims a introdus restricţii de identifi care pe structura de eroare a modelului şi această abordare a fost criticată, fi ind considerată în mod egal arbitrară. Dezvoltări ulterioare au condus la modele VAR structurale în care cointegrarea defi neşte relaţii de lungă durată între variabile nestaţionare şi unde sunt reintroduse variabile exogene (Pesaran şi Smith 1998), studiu în care se reanalizează un model timpuriu al lui King (1991). De la controversa Keynes – Tinbergen (Morgan 1990 şi Hendry şi Morgan 1995), rolul teoriei în modelare a reprezentat o controversă majoră în econometrie. La un capăt al liniei teorie – empirism, avem modele deduse teoretic care consideră de la sine înţeles teoria şi nu o verifi că Exemple importante sunt modelele de echilibru general, modelele reale sincronizate de ciclu de afaceri care au dobândit o poziţie dominantă în mediul academic (Kydland şi Prescott 1991). Există de asemenea o nouă serie de modele macroeconometrice care presupun posibilitatea de optimizare bazată pe probabilităţi raţionale, ducând la o mulţime de ecuaţii Euler (Poloz 1994, Willman 2000, Hunt 2000 şi Nilsson 2002 pentru modele ale băncilor centrale din Canada, Finlanda, Noua Zeelandă şi Suedia). La cealaltă extremă, există date bazate pe modele VAR care erau iniţial instrumente statistice ce utilizau la nivel minim teoria economică. Conform celor menţionate, în cazul modelelor structurale VAR, restricţiile teoretice pot fi impuse ca relaţii cointegratoare verifi cabile ca nivel sau pot fi impuse în structura de eroare a modelului. Abordarea pe care o susţinem are multe în comun cu metodologia LSE la care ne-am referit şi se concentrează asupra evaluării conform recomandării lui Granger (1999). Reprezintă un compromis între modele bazate pe date (pur statistice) şi teoria economică: pe de-o parte învăţăm din procesul de a încerca să luam în considerare în mod serios datele, pe de altă parte evităm să facem presupuneri teoretice – necesare a face teoriile „complete” - care nu ar avea relevanţă pe plan empiric, adică, ceva ce nu este susţinut de date. Rezultă că metodologia econometrică duce lipsă de consens şi astfel abordarea modelării macroeconometrice pe care o susţinem e controversată. Heckman (1992) pune la îndoială succesul dar nu importanta revoluţiei probabilistice a lui Haavelmo. De asemenea, Keuzenkamp şi Magnus (1995) oferă o critică a paradigmei Keyman-Pearson pentru verifi carea ipotetică şi pretind că econometria a exercitat o infl uenţă redusă asupra opiniilor economiştilor în ultimii 50 ani (Summers 1991). Pentru interpretări sceptice ale metodologiei LSE, amintim pe Hansen (1996) şi Faust şi Whiteman (1995, 1997) la care a replicat Hendry.
Revista Română de Statistică - Supliment nr. 4 / 2019 11
Concluzii
Din studiul care stă la baza acestui articol se desprind o serie de concluzii teoretice și practice. În primul rând, metodologia econometrică trebuie să fi e ancorată în realitățile practice economice. Pe baza agregatelor care fundamentează conceptul macroeconomic se pot identifi ca variabile statistice, care pot fi analizate pe bază de modele. Din conceptele prezentate se desprinde concluzia că numai prin analiză corelativă se pot obține parametrii pe baza cărora să se poată prognoza evoluția economică. Desigur, aceste concepte teoretice pot fi formalizate matematic pentru a obține mărimi (parametri) pe baza cărora să se construiască perspectivele de evoluție macroeconomică. Aceste concluzii teoretice conduc la ideea că pe baza bazelor de date devine facilă construirea de modele macroeconomice.
Bibliografi e
1. Anghelache, C., Anghel, M.G., Marinescu, A.I. (2019). The main indicators of national wealth that have an eff ect on macroeconomic performance, Romanian Statistical Review, Supplement, 3, 14-24
2. Anghelache, C., Anghel, M.G. (2018). Econometrie generală. Teorie și studii de caz, Editura Economică, Bucureşti
3. Anghelache, C., Anghel, M.G., Dumitru, D. (2018). The Content and Calculation of the Macroeconomic Results Indicators. Romanian Statistical Review Supplement, 9, 50-56
4. Anghelache, C., Isaic-Maniu, A., Mitruț, C., Voineagu, V., Dumbravă, M. (2007). Analiza macroeconomică – Sinteze şi studii de caz, Editura Economică, Bucureşti
5. Bardsen, G., Nymagen, R., Jansen, E. (2005). The Econometrics of Macroeconomic Modelling, Oxford University Press
6. Benjamin, C., Herrard A., Hanee-Bigot, M., Tavere, C. (2010). Forecasting with an Econometric Model, Springer
7. Carrasco, M., Florens, J.P., Renault, E. (2004). Linear inverse problems in structural econometrics, estimation based on spatial decomposition and regularization, Working paper, GREMAQ, University of Toulouse
8. Dougherty, C. (2008). Introduction to econometrics. Fourth edition, Oxford University Press
9. Fernandez-Villaverde, J., Rubio-Ramirez, J. (2009). Two Books on the New Macroeconometrics, Taylor and Francis Journals, Econometric Reviews
10. Johansen, S., Nielsen, M. (2010). Likelihood inference for a fractionally cointegrated vector autoregressive model, CREATES Research Papers 2010-24,
School of Economics and Management, University of Aarhus
Romanian Statistical Review - Supplement nr. 4 / 201912
SOME CONCEPTS AND ECONOMETRIC MODES OF MACROECONOMIC
PERFORMANCE ANALYSIS
Prof. Constantin ANGHELACHE PhD ([email protected])
Bucharest University of Economic Studies / „Artifex” University of BucharestAlexandra PETRE (OLTEANU) PhD Student ([email protected])
Bucharest University of Economic StudiesCristian OLTEANU PhD Student ([email protected])
Bucharest University of Economic Studies
Abstract
For the analysis of the macroeconomic performances, a system of correlated indicators is used, which after the interpretation gives the meaning of the future evolution of the economy. As a rule, macroeconomic models are used which, implemented, ensure the obtaining of parameters on which the macroeconomic evolution perspective can be predicted. One starting point in defi ning the concept of a macroeconomic model is the idea of a general stochastic process that generated the observed data. Practically, these particularities are easy to detail. Thus models for consumption and investment analysis or models based on statistical theory can be constructed, such as the autoregressive model with reduced dimensional vector. In constructing a macroeconomic model, a procedure can be followed that includes: defi nition and analysis of economic subsectors for the choice of variables; building the relevant partial models and then combining these sub-models in order to obtain a model representation of the whole economy. In practice, we must proceed gradually to include all elements of the considered submodels. Macroeconomic models have been dealt with by a number of researchers, a context in which theories are analyzed in the paper. Keywords: indicator, econometric model, variables, macroeconomic performance, analysis JEL Classifi cation: C10, C82, E70
Introduction
The authors conducted an extensive study on the concepts underlying the construction of macroeconomic performance analysis models taking into account the evolution of these concepts in time. Developments in this area are expressed, summarizing a series of issues that then formed the basis for building models for macroeconomic analysis. The article states that a concept of macroeconomic modeling is based on the hypothesis that if model A is a part
Revista Română de Statistică - Supliment nr. 4 / 2019 13
of model B, it is possible to identify data for model B in model A. Are points of view of the classics of modeling theory outlining the elements of which must be taken into account in the modeling process for macroeconomic analysis. One by one, the main aspects of Jan Timbergen’s models are then presented to the Haavelmo model or more recently the econometric elements on which macroeconomic models are based. The article provides a broad overview of the concepts underlying the construction of macroeconomic analysis models.
Literature review
Anghelache, Anghel, and Marinescu (2019) analyzed a series of elements related to national wealth indicators that have implications for macroeconomic performance. Anghelache and Anghel (2018) presented the main econometric methods and models used in macroeconomic analyzes. Anghelache, Anghel, and Dumitru (2018) highlighted how macroeconomic performance indicators were calculated. Anghelache, Isaic-Maniu, Mitruţ,
Voineagu and Dumbravă (2007) highlighted the main theoretical and practical
aspects of the macroeconomic analysis. Bardsen, Nymagen and Jansen (2005)
studied the use of econometrics in macroeconomic modeling. Benjamin,
Herrard, Hanee-Bigot and Tavere (2010) studied the use of econometric
models in macroeconomic forecasts. Carrasco, Florens and Renault (2004)
studied inverse linear problems in structural econometrics. Dougherty (2008)
presented the basic notions of econometrics. Fernandez-Villaverde and Rubio-
Ramirez (2009) studied key aspects of macroeconomics. Johansen and Nielsen
(2010) analyzed the probability of inference for an autoregressive model.
Research methodology, data, results and discussions
Macroeconomic modeling aims to highlight the empirical behavior of
a real economic system. Such models are designed as interconnected equations
based on dynamic series using statistical or econometric techniques.
A conceptual starting point is the idea of a general stochastic process
that generated the observed data on the economy, and that this process can be
summed up in terms of distribution with multiple probability variables of the
randomly observable variables in a system of stochastic equations. However,
it is possible to make an aggregate approach to represent the behavior of
important variables (eg infl ation rate, GDP growth, unemployment rate) in
a model. Estimation of such econometric models can be based on statistical
theory (autoregressive models with low VAR dimensional vector).
Practically, it is easy for the user to detail the particularities of the
model (separate modeling of consumption and investment) to simulate all the
equations simultaneously. The models that are used to analyze the impact of
Romanian Statistical Review - Supplement nr. 4 / 201914
the government budget on the economy are large systems of equations. Even if the model uses only a variable from the outgoers, such as the infl ation target,
policy choices are made against the background of a wider analysis of the
eff ects of the interest rate on the economy (nominal and real exchange rates,
fi nancial stability). Model makers have the task of establishing good practice
and developing operational procedures to build models that ensure that the
end result of modeling is solid and useful. Major contributions to literature
on macroeconomic models include the works of Christ (1966), Klein (1983),
Fair (1984, 1994), Klein (1999), or the research by Bodkin (1991) and Wallis
(1994).
In constructing a macroeconomic model we suggest the following
operational procedure:
• defi ning and analyzing the sub-sectors of the economy through
relevant variables;
• constructing (by conditioning) relevant partial models that we can
call Type A models by considering exogenous and endogenous
variables;
• combining these sub-models in order to achieve a B-Model of the
entire economy.
The concept is based on the hypothesis that given Model A as part of
Model B, it is possible to fi nd out about Model B in Model A. The alternative
to this hypothesis could be a type of creation without macroeconomy being
restricted to aggregate models.
For example, there are properties that can be discovered using the
Model B cointegration procedure. It is based on a cohort of the theory of
cointegrated systems: „Any linear non-zero combination of cointegration
vectors is also a cointegration vector.” In the simplest case, if there are two
cointegrators in Model B, there is always a linear combination of these vectors
that will remove one of the variables.
Analysis of the cointegration of a subset of variables (Model A),
excluding that variable, will result in a cointegrator vector corresponding
to that linear combination. Thus, despite being a property of Model B,
cointegration analysis of the subsystem (Model A) identifi es a cointegration
vector. If this identifi cation has an economic sense or is not an open issue, any
solution must be argued in each case. Analysis based on such a system means
the identifi cation of the consumption function as a cointegrative relationship,
the identifi cation of the cointegrative relations used to establish prices and
wages, the existence of a natural rate of unemployment and the relevance of
the future-oriented terms of price and wage setting.
As Johansen (2002) pointed out, this is a diffi cult situation in the
Revista Română de Statistică - Supliment nr. 4 / 2019 15
previous procedure: a general theory for the three stages will contain criteria and conditions that are formulated for the whole system. Sophisticated modeling on portions can be considered as a way of gradualism - seeking to establish the submodels that represent the partial structure; that is, partial models that are unidimensional to the extensions of the survey period, to changes elsewhere in the economy, and remain the same for extensions of the information set. Gradualism also implies a possible revision of a submodel. Revisions are sometimes triggered by the forecast failure. Several common reasons include revisions and data extensions that allow more accurate and improved model specifi cations. Dialogue between model makers and their users also results in revisions. Thus, experienced model users are usually able to highlight the implications of a single equation specifi cation (or submodel) on the properties of the entire model. Obviously, gradualism does not remove the in-depth testing of a submodel. On the contrary, the fi rst two steps in the above procedure do not require that we know the whole model, and testing those conditions has a certain intuitive attraction since real life provides „new proof” through new data and „natural experiments” through changes such as changes in government or fi nancial deregulation. For the three stages, we could basically consider the complete model as the ultimate extension of the set of information and thus the establishment of the partial structure or structure, which is a way to continue to fi nd Johansen. Basically, we know the whole model is hard to achieve. We will reduce the activity to plotting the sectoral model graph with the simplifi ed approximations of Model B, and will verify the relevant external assumptions of the partial model in the same frame. To the extent that the similar function of Simplifi ed Model B is adequate or approximates the similar function of Full B Model, there is no problem. It is also possible to confi rm the whole procedure since it is true that Model A can be tested and gradually improved with new information, which is a way of acquiring that knowledge that compares to modern Darwinism in natural sciences. A practical reason for focusing on the submodels is as follows: modelers may have a good reason to study some branches of the economy with more attention than others. For a central bank targeting infl ation, it is
necessary to use the model of the infl ationary process. It requires a careful
modeling of institutional or conditional arrangements for pricing and wages,
developments in fi nancial markets and the real economy in order to answer a
number of important questions: Is there a natural rate of unemployment that
indicates unemployment and infl ation? What is the importance of infl ation
probabilities and how can they be modeled? What is the role of money in the
infl ationary process?
Romanian Statistical Review - Supplement nr. 4 / 201916
We believe that in order to answer these questions - and to test competing hypotheses on supply-side economics, a detailed modeling based on information specifi c to the study economy is needed. Taking into account the simultaneity is to a large extent a matter of effi ciency of estimation. If
there is a correlation between this effi ciency and the problem of regulating
the economic mechanisms, macroeconomic modeling specialists should give
priority to the last problem.
The history of macroeconomic modeling begins with the Dutch
economist Jan Tinbergen who built and estimated the fi rst macroeconomic
models in the mid 1930s (Tinbergen 1937). Tinbergen has shown that
someone can transform a system of equations into an econometric model of
business cycle using economic theory to derive behavioral motivated dynamic
equations and statistical methods to verify them. However, there seems to be a
universal agreement according to which statistics is a discipline of economics
and econometrics based on the contribution of Norwegian economist Trygve
Haavelmo in his treaty „Approaching probability in econometrics”. Haavelmo
was inspired by some of the greatest time statisticians. As Morgan points out
(1990), it was converted to the use of probability ideas by Jerzy Neyman, and
was also infl uenced by Abraham Wald, which Haavelmo believed as a source
of understanding of statistical theory.
It is important to emphasize that Haavelmo admitted and explained in
the context of an economic model that the distribution with multiple variables
of all variables observable for the whole studied period provides the general
framework for statistical deduction (Hendry 1989). This applies to specifying
/ identifying, estimating and verifying the hypothesis:
„All are reduced to one and the same thing, namely, the study of the
probability distribution properties with several variables of random variables
in a stochastic system of equations”.
Haavelmo’s probabilistic revolution infl uenced econometrics. His
thinking was immediately adopted by Jacob Marschak, a scientist of Russian
origin who studied statistics with Slutsky, as a research agenda for the
Cowles Commission from 1943-47, by reconsidering Tinbergen’s business
cycles. Marschak was joined by a group of statisticians, mathematicians and
economists, including Haavelmo. Their work was to establish the standards
of modern econometry and was found in the works of economics at Tintner
(1952) and Klein (1953) and others. The Cowles Commission’s work has laid
the foundations for developing macroeconomic models and modeling that
has become a broad „industry” in the US over the next three decades (Bodkin
1991 and Wallis 1994). These models were mainly designed for short and
medium term forecasting, ie modeling for business cycle forecasting. The fi rst
Revista Română de Statistică - Supliment nr. 4 / 2019 17
model (Klein 1950) was created with the explicit purpose of implementing Haavelmo’s ideas in Tinbergen’s modeling for the US economy. Like the model of Tinbergen, this was a model in which Klein emphasized the modeling of simultaneous equations. Later models became extensive systems using more than 100 equations to describe the behavior of a modern industrial economy. In such models, less econometric specifi cation was considered and simultaneity could not be treated satisfactorily. The purpose of these models’ forecasts was that they were evaluated on the basis of their performance. When models failed to predict the eff ects on the industrial shocks of the 1973 and 1979 oil shocks,
macroeconomic modeling has diminished its position, especially in the US.
In the 1980s, macroeconomic models took advantage of methodological
and conceptual advances in econometrics based on a dynamic (chronological)
series. Box and Jenkins (1970) created and popularized a purely statistical
tool for modeling and forecasting one-dimensional chronological series. The
second infl ux of statistical methodology into econometrics is rooted in the study of the non-static nature of economic data series. Clive Granger in his studies of statistics has shown in a series of valuable papers the importance of having an econometric equation balanced. A stationary variable can not be explained by a non-stationary variable and vice versa (Granger, 1990). In addition, the concept of cointegration (Granger 1981, Engle and Granger 1987, 1991), respectively a linear combination of two or more non-static variables can be stationary (fi xed), has proven to be useful and important in macroeconomic modeling. In a general VAR model, the statistician Soren Johansen created (Johansen 1988, 1991, 1995b) the models used to test cointegration in a multidimensional framework, tracing the analytical mode of canonical correlation and Anderson’s position regression (1951). Greater attention has also been given to the role of evaluation in modern economics (Granger 1990, 1999). The so-called London School of Economics methodology highlights the importance of econometric testing and assessment models (Hendry 1993a, 1995a, Mizon 1995 and Ericsson 2005). Hendry (1989) claims that many aspects of Haavelmo’s research have been ignored for a long time. For example, the function with multiple variables for observable variables has been recognized by the Cowles Commission as essential in solving statistical deduction problems, but ideas have not infl uenced empirical modeling strategies. In contrast, many developments in econometrics since 1980 correspond to this and other aspects of Haavelmo’s research program. This also applies to the role of economic theory in econometrics, namely: „Theoretical models are necessary tools in our attempts to understand and explain events in real life (Haavelmo 1944). But any explanation we
Romanian Statistical Review - Supplement nr. 4 / 201918
prefer, one can not ignore that they are all our artifi cial inventions in a quest to understand real life, not hidden truths to be discovered. „ (Haavelmo 1944) Having this starting point, we could expect facts or observations that correspond to any precise formulation derived from a theoretical model. Economic theories must therefore be expressed as probabilistic formulations and Haavelmo considered probability theory as indispensable in formalizing the notion of models as approximations of reality. The Cowles Commission’s research program has focused on Simultaneous Equation Models (SEMs) and has focused on identifying issues. In addressing these issues, economic theory plays an important role. A remarkable representative of this tradition, Lawrence Klein writes in a study on the interaction between statistics and economics in the context of macroeconomic modeling (Klein 1988) that the approach to modeling may be opposed to pure statistical analysis that is empirical and not closely related of economic theory as is the creation of models. However, the creation of traditional macroeconomic models has also been criticized (Favero 2001). While the LSE methodology attributes the failure of those early macroeconomic models to the absence of model misinterpretation or specifi cation, other critics such as Robert Lucas and Christopher Sims have claimed: „The reason is rather that it had a poor theoretical basis.” Lucas’s critique (Lucas 1976) claims that the failure of conditional models is caused by regime changes as a result of policy changes and changes in expectations. Criticism refers to SEMs, if probabilities are not modeled. On the other hand, Sims (1980) argued that SEMs include „incredible identifi cation restrictions”: „the restrictions required to claim the exogenity of certain variables would not be valid in an environment where agents occasionally optimize.” Sims, on the other hand, supports the use of a low-order vector autoregression (autoregression) against the chronological economic series. This approach would have the advantage that it does not depend on an „arbitrary” separation between exogenous and endogenous variables and that it does not require „incredible identifi cation restrictions”. Instead, Sims introduced identifi cation restrictions on the model error pattern, and this approach was criticized, being equally arbitrary. Subsequent developments have led to structural VAR models where cointegration defi nes long-term relationships between non-static variables and where exogenous variables are returned (Pesaran and Smith 1998), a study in which an early model of King (1991) is reanalyzed. Since the Keynes-Tinbergen controversy (Morgan 1990 and Hendry and Morgan 1995), the role of modeling theory has been a major controversy in econometrics. At the end of the theory line - empiricism, we have theoretically
Revista Română de Statistică - Supliment nr. 4 / 2019 19
deduced models that consider the theory and do not check it out. Important examples are the general equilibrium models, the actual synchronized business cycle models that have acquired a dominant position in the academic environment (Kydland and Prescott 1991). There is also a new series of macroeconomic models that imply optimization based on rational probabilities, leading to a multitude of Euler equations (Poloz 1994, Willman 2000, Hunt 2000 and Nilsson 2002 for models of central banks in Canada, Finland, New Zealand and Sweden). At the other extreme, there are data based on VAR models that were initially statistical tools that used economic theory to a minimum. Accordingly, in the case of structural VAR models, the theoretical constraints can be imposed as verifi able cointegrating relationships as a level or may be imposed in the model error pattern. The approach we support has much in common with the LSE methodology that we have referred to and focuses on the assessment as recommended by Granger (1999). It is a compromise between data-based models (purely statistical) and economic theory: on the one hand, we learn from the process of trying to seriously consider data, on the other hand we avoid making theoretical assumptions - necessary to make theories „Complete” - which would not be empirically relevant, that is, something that is not supported by the data. It follows that the econometric methodology is lacking in consensus, so the approach to the macroeconomic modeling that we support is controversial. Heckman (1992) casts doubt on the success but not the importance of Haavelmo’s probabilistic revolution. Keuzenkamp and Magnus (1995) also criticize the Keyman-Pearson paradigm for hypothetical verifi cation and claim that econometrics has exerted little infl uence over economists’ views
over the last 50 years (Summers 1991). For skeptical interpretations of the
LSE methodology, we mention Hansen (1996) and Faust and Whiteman
(1995, 1997) to which Hendry replied.
Conclusion
From the study underlying this article, a series of theoretical and
practical conclusions emerge. Firstly, the econometric methodology must be
anchored in practical economic realities. Based on the aggregates underpinning
the macroeconomic concept, statistical variables can be identifi ed, which can
be analyzed on a model basis. From the presented concepts it is concluded
that only through correlational analysis can the parameters be obtained on the
basis of which it is possible to forecast the economic evolution. Of course,
these theoretical concepts can be mathematically formalized to obtain the
parameters (parameters) on which to build the prospects of macroeconomic
evolution. These theoretical conclusions lead to the idea that based on
databases it becomes easy to build macroeconomic models.
Romanian Statistical Review - Supplement nr. 4 / 201920
References
1. Anghelache, C., Anghel, M.G., Marinescu, A.I. (2019). The main indicators of national wealth that have an eff ect on macroeconomic performance, Romanian Statistical Review, Supplement, 3, 14-24
2. Anghelache, C., Anghel, M.G. (2018). Econometrie generală. Teorie și studii de caz, Editura Economică, Bucureşti
3. Anghelache, C., Anghel, M.G., Dumitru, D. (2018). The Content and Calculation of
the Macroeconomic Results Indicators. Romanian Statistical Review Supplement, 9, 50-56
4. Anghelache, C., Isaic-Maniu, A., Mitruț, C., Voineagu, V., Dumbravă, M. (2007).
Analiza macroeconomică – Sinteze şi studii de caz, Editura Economică, Bucureşti
5. Bardsen, G., Nymagen, R., Jansen, E. (2005). The Econometrics of Macroeconomic Modelling, Oxford University Press
6. Benjamin, C., Herrard A., Hanee-Bigot, M., Tavere, C. (2010). Forecasting with an Econometric Model, Springer
7. Carrasco, M., Florens, J.P., Renault, E. (2004). Linear inverse problems in structural econometrics, estimation based on spatial decomposition and regularization,
Working paper, GREMAQ, University of Toulouse
8. Dougherty, C. (2008). Introduction to econometrics. Fourth edition, Oxford
University Press
9. Fernandez-Villaverde, J., Rubio-Ramirez, J. (2009). Two Books on the New Macroeconometrics, Taylor and Francis Journals, Econometric Reviews
10. Johansen, S., Nielsen, M. (2010). Likelihood inference for a fractionally cointegrated vector autoregressive model, CREATES Research Papers 2010-24,
School of Economics and Management, University of Aarhus
Revista Română de Statistică - Supliment nr. 4 / 2019 21
Particularităţi ale utilizării curbei Phillips în studiul evoluţiei şomajului
Conf. univ. dr. Mădălina-Gabriela ANGHEL ([email protected])
Universitatea „Artifex” din BucureștiDrd. Ștefan Gabriel DUMBRAVĂ ([email protected])
Academia de Studii Economice din BucureștiDrd. Daniel DUMITRU ([email protected])
Academia de Studii Economice din București
Abstract Coefi cienții ecuației șomajului, de regulă, nu fac obiectul exprimării ratei naturale. În varianta curbei Phillips, defi nirea ratei naturale a infl ației depinde de parametrii stabilirii prețurilor și salariilor. Acest model este de fapt o curbă Phillips pentru prețuri care desigur infl uențează rata salariului real. Este de înteles că rata șomajului ridicat presupune modele care să țină seama de lag-uri în cadrul ajustării în jurul unei rate naturale constante și permit schimbarea (modifi carea) echilibrului. Conceptul care relefează schimbările permanente a șomajului sunt modifi cări intense, cu manifestări intermitente. Din studiul curbei Phillips rezultă că diferența dintre rata naturală și parametrul independent de timp, nu mai este realistă fi ind vorba de un parametru stochastic care are o evoluție aleatoare. Rezultă că, în cazul ipotezei unei superexogenități, rezultatele pentru un model economic convențional, cum este curba Phillips convențională, nu sunt invariabile la o renormalizare. Trebuie avut în vedere că formularea regresiei clasice este de genul comportamental determinat de probabilități și nu de probabilități bazate pe modele mici cu parametrii estimați. Cuvinte cheie: șomaj, curba Phillips, salarii, prețuri, model econometric Clasifi carea JEL: E24, E31, J31
Introducere În acest articol, autorii s-au preocupat de evidențierea particularităților pe care le presupune utilizarea curbei Phillips în studiul evoluției șomajului. Pornind de la funcțiile pe care le are utilizarea curbei Phillips în analiza șomajului, se evidențiază că numai prin combinarea celor două funcții se pot obține rezultate certe. Studiul curbei Phillips oferă numai o compatibilitate
Romanian Statistical Review - Supplement nr. 4 / 201922
limitată de explicare economică a schimburilor care sunt determinate de rata infl ației. În fond, rezultatele macroeconomice corectate cu eliminarea efectelor infl ației dau o expresie asupra efctului pe care îl ale infl ația. Conceptele prezentate, reliefează că schimbările permanente ale șomajului sunt modifi cări de amploare cu manifestare asupra ratei șomajului. În continuare, se face o prezentare atentă cu privire la estimarea incertitudinii curbei Phillips de tip NAIRU. Motivul absenței intervalelor de încredere în majuritatea calculelor cu privire la NAIRU sunt correlate cu faptul că acest model este o funcție neliniară a coefi cienților de regresie. În articol se fac prezentări ample privind testarea statistică ale modelului și parametrilor obținuți. În fi nal se precizează că utilizarea unui model dVAR trebuie să recurgă în mod permanent la corecții de segment și alte corecții care pot apărea.
Literature review
Anghelache, Anghel, Dumbravă și Ene (2018) au efectuat un studiu cu privire la corelația dintre indicatorii rata de ocupare a populației, rata șomajului și rata locurilor de muncă vacante în economia românească. Anghelache și Anghel (2018) au expus principalele metode și modele de analiză econometrică. Anghelache și Anghel (2016) au prezentat noțiuni fundamentale ale statisticii economice. Deleplace (2008) a abordat aspecte referitoare la absorbția macroeconomiei de către microeconomie. Ewing și Seyfried (2003) au studiat aspecte ale modelării curbei Phillips. Ftiti (2010) a examinat elementele ale performanței macroeconomice a politicii de direcționare a infl ației. Isard, Laxton și Eliasson (2001) au studiat repere ale infl ației. Jansen (2002) a studiat probleme statistice în modelarea macroeconomică. Ferri (2000) a analizat aspecte ale dinamicii salariilor și curba Phillips. Stanley (2002) a studiat incertitudinea curbei Phillips de tip NAIRU.
Metodologia cercetării, date, rezultate și discuții
• Coefi cienţii ecuaţiei şomajului nu intră în exprimarea ratei naturale. În altă variantă a curbei Phillips, exprimarea pentru rata naturală depinde de parametrii stabilirii preţurilor şi salariilor, adică modelul este specifi cat mai degrabă ca o curbă Phillips pentru preţuri decât ca o curbă Phillips pentru salarii. Dar exprimarea ratei dintr-o curbă Phillips pentru preţuri rămâne independentă de parametrii calculați. Faptul ca o proprietate importantă de sistem (şomajul) poate fi estimată printr-o singură ecuaţie conduce într-un fel către explicarea popularităţii modelului de curbă Phillips. Cu toate acestea, rezultatele bazate pe analiza unui sistem incomplet oferă informaţii limitate. În particular, analiza bazată
Revista Română de Statistică - Supliment nr. 4 / 2019 23
pe o singură ecuaţie oferă informaţii insufi ciente despre proprietăţile dinamice ale sistemului. În ultimii ani ai secolului anterior, ratele europene ale şomajului au crescut vertiginos şi nu au arătat semne de revenire la nivelurile anilor 1960 şi 1970. Este de înţeles că şomajul ridicat a necesitat modele care îndeplinesc următoarele funcţii: - ţin seama de lag-uri în cazul ajustării în jurul unei rate naturale
constante; - permit schimbarea echilibrului. O combinaţie a celor două funcţii este de asemenea posibilă. În mod simplu, în virtutea de a fi un sistem dinamic, modelul de curba Phillips cuprinde dinamici lente). În principiu, coefi cientul de ajustare βu1 în ecuaţia şomajului poate fi mic în mod arbitrar, atât timp cât nu e zero, uphil corespunde formal stării stabile a sistemului. Totuşi, se pune problema cât de lentă poate fi viteza ajustărilor înainte ca ideea de echilibru să devină subminată “din interior”. Conform argumentelor lui Phelps (1995) şi Friedman, rata naturală trebuie să fi e aproape stabilă şi ar trebui să fi e un punct puternic de atracţie pentru rata efectivă a şomajului. Totuşi, experienţa anilor 1980 şi 1990 a demonstrat că în cel mai bun caz rata naturală este un punct slab de atracţie. Sunt aspecte practice ale problemei, importante: guvernanţii, analizând asupra perspectivelor după un şoc negativ al economiei, vor constata că rata şomajului va reveni eventual la nivelul său natural dar numai după 40 de ani sau mai mult. În plus, studiul curbei Phillips oferă numai o capabilitate limitată de explicare economică a schimbărilor care se manifestă prin rata şomajului. Pare, uneori, incredibil că modifi carea ratei reale de creştere ga poate explica singură creşteri profunde şi persistente ale ratei şomajului, cum s-a întâmplat în Europa. O rată nominală de creştere poate bineînţeles să sufere creşteri majore, dar pentru ca acele schimbări să aibă un impact asupra ratei de echilibru e nevoie de curba Phillips pe termen lung cu pantă descendentă. Astfel, curba Phillips e mai bine adaptată unui regim stabil caracterizat printr-un lag de ajustare redusă în jurul unei rate medii stabile a şomajului, decât la modifi cările şomajului în Europa anilor 1980 şi 1990. S-au dezvoltat în consecinţă noi modele în anii 80, modele care au promis că pot explica modifi cările ratei de echilibru a şomajului, şi există informaţii concrete despre modul în care caracteristicile structurale ale pieţelor forţei de muncă şi de mărfuri afectează echilibrul şomajului. Noile modele nu au ajuns la stadiul de a fi unanim acceptate, aşa cum a fost rolul curbei Phillips. Conceptele prezentate anterior reliefează că schimbările permanente ale şomajului sunt modifi cări de amploare, cu manifestare intermitentă, în
Romanian Statistical Review - Supplement nr. 4 / 201924
linie cu concepţia asupra ratei şomajului ca | (0). O opinie alternativă este aşa numita rată naturală variabilă în timp (Gordon 1997, Gruen et al. 1999 şi Staiger et al. 1997). Ideea de bază este că rata naturală a şomajului reacţionează la infl uenţe de mică amploare care se manifestă frecvent. Conform acestei opinii, rata naturală a şomajului este dată de relaţiile:
(1)
(2) Diferenţa este că rata naturală nu mai este un parametru independent de timp, ci un parametru stocastic care urmează calea aleatorie (2), şi o abatere
care reprezintă infl uenţele de mică amploare. În estimarea acestei pereche de ecuaţii (Kalman), eroarea standard este limitată de la început, altfel
s-ar modifi ca în sus şi în jos şi ar absorbi toate abaterile din lăsate neexplicate prin variabile explicative. Metodologia implică o unitate de bază, atât în rata observată a şomajului cât şi în rata naturală a şomajului. Relevanţa practică a acestui cadru pare să fi e limitată la SUA, unde există puţine modifi cări majore ale ratei şomajului. Legat de rata naturală a şomajului variabilă în timp este conceptul de hysteresis. Economiştii au invocat termenul de şomaj “hysteresis” pentru cazul în care rata de echilibru ar putea să devină identică ratei întârziate a şomajului. Totuşi, se face distincţie între întârzierea reală ca fenomen de echilibru neliniar şi multiplu şi proprietatea liniară a unei unităţi rădăcină, aşa cum reiese din cercetările lui Roed (1994). În plus, s-a demonstrat că hysteresis nu este o întârziere efectivă (în adevăratul sens al cuvântului, ca fenomen neliniar) şi ca fenomenul de hysteresis efectiv conduce la o abordare a şomajului care nu corespunde cu ipoteza ratei naturale.
• Estimarea incertitudinii curbei Phillips tip NAIRU
Aceasta descrie trei abordări ale estimării unui “domeniu de încredere” a curbei Phillips tip NAIRU. Motivul absenţei intervalelor de încredere în majoritatea calculelor legate de NAIRU sunt corelate cu faptul că NAIRU este o funcţie neliniară a coefi cienţilor de regresie. Există trei metode care pot fi utilizate pentru construirea intervalelor de încredere pentru NAIRU: metoda Wald, metoda Fieller şi metoda coefi cienţilor de probabilitate. Abordarea lui Fieller şi formulele coefi cienţilor de probabilitate sunt preferabile datorita proprietăţilor fi nite ale eşantionului. Prima şi cea mai intuitivă abordare se bazează pe eroarea standard asociată şi proporţia t pentru coefi cienţii estimaţi şi corespunde unei statistici Wald.
Revista Română de Statistică - Supliment nr. 4 / 2019 25
În cazul unei treceri totale a creşterilor productivităţii în salarii şi lipsei “iluziei banilor”, uphil din curba Phillips tip NAIRU este βw0/ βw1 şi valoarea sa estimata este . Putem scrie ca:
(3) unde uphil poate fi estimat direct prin metoda celor mai mici pătrate neliniare. Rezultatul este numeric echivalent cu coefi cientul derivat din estimări liniare ( ). În fi ecare caz, o eroare standard pentru poate fi calculată, din care intervalele de încredere pot fi obţinute direct. Un interval de încredere include estimarea nerestricţionată a uphil, care este şi un anumit domeniu în jurul acelei valori. În mod euristic, intervalul de încredere conţine fi ecare valoare a coefi cientului care nu respectă ipoteza: (4) Considerăm că este statistica F bazată pe Wald pentru verifi carea HW, iar Pr(·) este probabilitatea argumentului său. Un interval de
încredere de (1-α)% este defi nit de
pentru .
Dacă βw1, elasticitatea ratei şomajului în curba Phillips este estimată cu precizie, abordarea Wald este satisfăcătoare. Mărimi mici ale eşantionului pun în pericol în mod clar precizia estimării, dar dimensiunea eşantionului depinde de volumul de informaţii pe bază de observare. Totuşi, dacă βw1 este estimat imprecis (nu este semnifi cativ statistic), această abordare poate să inducă în eroare în mare măsură. Abordarea Wald ignoră modul în care
se comportă pentru valori ale apropiate relativ de zero, unde “relativ” refl ectă incertitudinea în estimarea lui βw1. Pentru curbele Phillips europene, estimările βw1 sunt tipic nesemnifi cative statistic, încât această preocupare este legată de calcularea ratelor naturale prin curba Phillips pentru Europa. În esenţă, problema apare deoarece este funcţie neliniară a estimatorilor ( ) care sunt repatizati normal. A doua abordare evită această problemă prin transformarea ipotezei neliniare într-una liniară, adică:
(5)Ipoteza şi statistica F corespunzătoare sunt notate HF şi . Deoarece ipoteza este liniară în parametrii βw0 şi βw1, verifi cările acestei ipoteze sunt corespunzătoare, chiar dacă este aproape de zero. Determinarea
Romanian Statistical Review - Supplement nr. 4 / 201926
intervalelor de încredere se face exact ca în abordarea Wald, exceptând faptul
că statistica-F este construită pentru . A treia abordare foloseşte coefi cientul de verosimilitate statistic pentru a calcula intervalul de încredere pentru ipoteza HW. Relaţia este estimată atât nerestrictiv cât şi sub restricţia HW, probabilităţi corespondente (sume reziduale ale pătratelor pentru ecuaţii singulare) sunt obţinute, iar intervalul
de încredere este construit din valori ale pentru care verosimilitatea este mai mică decât valoarea critică dată. Dacă modelul original este liniar în parametrii săi, soluţia lui Fieller este echivalentă numeric cu soluţia bazată pe verosimilitate, oferind o justifi care generică anterioară. În cazul în care curba Phillips estimată nu prezintă
omogenitate dinamică, este numai o componentă a estimării NAIRU care ar fi potrivită cu teoria de bază. Aceasta conduce la calcule complexe pentru NAIRU în continuare, respectiv trebuie luată în considerare codispersia
termenilor precum şi . Totuşi, în afara de cazul în care depărtarea de omogenitate are o valoare numerică mare,
poate fi reprezentativă pentru gradul de incertitudine asociat
cu rata naturală estimată a curbei Phillips. Probleme statistice identice apar în alte zone de aplicare a macroeconomiei, de exemplu în forma unui “indice de condiţii monetare”. • Reversul variabilelor dependente şi independente reprezintă o controversă continuă în literatura despre modelarea infl aţiei. Modul în care curba Lucas transformă cauzalitatea curbei Phillips convenţionale a fost prezentat anterior. Critica Lucas afi rmă de asemenea că modelele curbei Phillips convenţionale vor experimenta discontinuităţi structurale ori de cate ori se modifi că perspectivele economice (de exemplu, urmarea unei schimbări în politica economică). În cadrul unei ipoteze a unei superexogenităţi, rezultatele pentru un model econometric convenţional, cum ar fi curba Phillips convenţională, nu sunt invariabile la o renormalizare, conform formulei:
(6) unde: ryx = coefi cientul de corelaţie;
= coefi cientul estimat de regresie când y este variabila dependentă şi x este factor de regresie; = coefi cientul estimat în regresia inversată.
Revista Română de Statistică - Supliment nr. 4 / 2019 27
Schimbările de regim atrag alterări ale structurilor de corelaţie, deci modifi cări de ryx. Dacă, datorită superexogenităţii, este totuşi constant, atunci nu poate fi constant. Ecuaţia se aplică în general cu ryx interpretat drept coefi cient de corelaţie parţial. Deci, dacă (de exemplu) curba Phillips este estimată prin OLS, atunci constatarea că este stabil atrage după sine faptul că pentru ecuaţia re-normalizată (la rata şomajului) este instabil. Constatarea că o curbă Phillips stabilă pe o perioadă de sondaj care conţine modifi cări în corelaţiile parţiale respinge orice pretenţie că modelul are o interpretare a curbei cererii a lui Lucas. Această procedură simplă se aplică de asemenea la estimarea prin variabile instrumentale (datorită endogenităţii şi/sau ), cu condiţia numărul de variabile instrumentale să fi e mai mic decât numărul de variabile endogene din curba Phillips. Modelele econometrice condiţionale sunt predispuse la instabilitate şi eşec ori de câte ori probabilităţile nemodelate se schimbă. Critica Lucas poate fi confi rmată sau respinsă empiric. În general, se consideră o singură variabilă de serie cronologică (aleatoare) yt, care poate fi divizată într-o parte explicată şi o parte neexplicată independentă :
(7) Considerăm ca un plan atribuibil agenţilor şi - diferenţa între rezultatul planifi cat şi cel real al yt. Astfel:
(8) este o inovaţie aferentă planului, deci:
(9) Presupunem în continuare că agenţii folosesc o mulţime a informaţiilor Іt-1 pentru probabilităţi raţionale pentru variabila xt, adică: (10) şi probabilităţile sunt conectate la: (11)ceea ce se motivează prin teoria economică. Din formula , presupunând că din (7) este o inovaţie, obţinem:
(12) şi deci: (13)
Romanian Statistical Review - Supplement nr. 4 / 201928
Iniţial, se presupune că urmează un proces AR de prim ordin
(nestaţionaritatea / instabilitatea este considerată în continuare):
(14)
Astfel,
sau
(15)
Pentru simplifi care, presupunem că şi sunt independente. Vom presupune în continuare că singurul parametru de interes în ecuaţia (1) este β. Forma redusă a yt este obţinută din ecuaţiile (7), (11) şi (14):
(16) unde xt este slab exogen pentru ξ = α1β, dar parametrul de interes β este neidentifi cabil numai din (16). În plus, ecuaţia de formă redusă (16), în timp ce permite estimarea ξ într-o stare a naturii caracterizată prin stabilitate, este susceptibilă la critica Lucas deoarece ξ nu este invariabilă la schimbări în parametrul autoregresiv al modelului marginal (14). Practic, critica Lucas are ca scop “ecuaţii comportamentale” în sisteme de ecuaţii simultane, de exemplu, (17) cu termenul de abatere:
(18) În mod direct se poate arăta că estimarea (17) prin OLS pe un eşantion t = 1,2....T, conduce la: (19)
stabilind că, regresia yt la xt nu reprezintă opusul din (11). În mod specifi c, în loc de β, estimarea şi modifi carea parametrului α1 afectează stabilitatea estimărilor, confi rmând astfel critica Lucas. Totuşi, aplicabilitatea criticii se bazează pe ipotezele iniţiale. De exemplu, dacă se modifi că ipoteza din în α1 = 1, încât xt să aibă o rădăcină unitară, dar este cointegrată cu yt, critica Lucas nu se aplică: în cadrul cointegrării, întrucât parametrul cointegrării este unic
şi poate fi estimat în mod consecvent prin OLS. Ca un alt exemplu al importantei mulţimii exacte de ipoteze făcute, se considera înlocuirea (11) cu o altă teorie economică, anume planul contingent: (20)
Revista Română de Statistică - Supliment nr. 4 / 2019 29
Ecuaţiile (20) şi (7) conduc la: (21)
unde şi β pot fi estimate prin
OLS în cazul staţionar al .
• Probabilităţi bazate pe model comparate cu cele bazate pe date
Aparent, se uită adesea că formularea regresiei “clasice” (21) corespunde cu ideea: „comportamentul este condus de probabilităţi, dar nu de probabilităţi bazate pe model nici de probabilităţi raţionale cu parametri necunoscuţi care necesită să fi e estimaţi”. Pentru a interpreta probabilităţile din (21), relaţia (10) se înlocuieşte prin:
(22)
şi se presupune că se rezolvă pentru a obţine .
Substituirea lui şi folosirea (7) pentru conduc la (21). este un exemplu de regulă a predicţiei invariabile fără parametri, bazată pe proprietăţi ale datelor, sau probabilităţi bazate pe date. În mod realist, se poate alege utilizarea predictorilor bazaţi pe date, datorită costurilor colectării şi prelucrării informaţiilor asociate cu predictorii bazaţi pe modele. E adevărat că abordările bazate pe utilizează un model greşit specifi cat al procesului x (15) şi astfel prognozele lor nu vor atinge eroarea medie pătratică de prognoză. Deci, în condiţii de stabilitate, sunt câştiguri din estimarea α1 din (15). Totuşi, practic, nu există garanţia că parametrii procesului x rămân constanţi până la orizontul de prognoză şi starea instabilă dintr-o prognoza bazată pe model nu poate fi considerată ca mai bună decât prognoza derivată din regula simplă . În fapt, în funcţie de datarea schimbării de regim legate de prognoza “producţiei”, prognoza bazată pe date va fi mai buna decât prognoza bazată pe model în termeni de interferenţă.Introducem astfel un termen de creştere în (15), adică:
(23) şi presupunem ca există o modifi care în α0 (α0) în perioada T+1. Considerăm doi agenţi, A şi B, care prognozează xT+1. Agentul A colectează date pentru o perioadă t = 1,2,3,....,T şi poate descoperi valori reale ale {α0,α1} pentru acea perioadă. Totuşi, din cauza modifi cării neprevizionate
în perioada T+1, eroarea de prognoză a lui A va fi :
(24)
Romanian Statistical Review - Supplement nr. 4 / 201930
Agentul B, folosind prognoza bazată pe date , va întâlni o eroare de prognoză: care poate fi exprimată sub forma: (25)
unde înseamnă media xT (adică pentru segmentul anterior modifi cării ), . Comparaţia între relaţiile (24) şi (25) arată că singura diferenţă între cele două erori de prognoză este termenul
din (25). Astfel, ambele prognoze sunt afectate de modifi carea de regim care are loc după efectuarea prognozei. Mediile condiţionate şi dispersiile (repartiţiile) celor două erori sunt:
(26)
(27)
(28) stabilind că în acest exemplu (care ia în considerare o unei schimbare de regim post-prognoză), nu este nici o corespondenţă între cele două metode de prognoză privind primele două momente ale erorii de prognoză. Dispersiile (repartiţiile) de eroare de prognoză condiţionată sunt identice şi infl uenţele prognozei bazate pe model nu sunt mai mici decât cele ale predictorului bazat pe date; să presupunem, de exemplu, că , în acelaşi timp , interferenţa bazată pe date poate fi mai mică decât interferenţa bazată pe model. În plus, în mod necondiţionat, cei doi predictori au aceeaşi infl uenţă şi dispersie:
(29)
(30) În continuare considerăm prognozele făcute pentru perioada T+2, cu condiţia T+1 ca exemplu al unei modifi cări de regim ante-prognoză ( în perioada T+1). În afară de cazul în care A descoperă modifi carea în α0 şi corectează cu succes prognoza, infl uenţa erorii va fi din nou:
(31) Infl uenţa erorii de prognoză a agentului B pe de altă parte devine:
, (32)
Revista Română de Statistică - Supliment nr. 4 / 2019 31
unde reprezintă media necondiţionată de schimbare post-regim a x, adică
. În mod clar, interferenţa predictorului bazat pe date poate fi mai mica decât eroarea previziunii bazate pe model (dar şi opusul poate fi adevărat. Totuşi:
(33)
Erorile de prognoză necondiţionată sunt întotdeauna mai mici pentru previziunea bazată pe date în cazul modifi cării de regim ante-prognoză.
Analiza generalizează cazul unui rădăcini unitare în procesul x, în fapt se observă din cele anterioare că erorile prognozelor bazate pe date au
proprietăţi mai bune pentru cazul α1 = 1, de exemplu
din (21). Dacă xt este , atunci rezolvarea pentru a obţine va rezulta într-o prognoză cu aceeaşi soliditate privind modifi cările de regim, aşa cum s-a ilustrat în exemplul anterior. Această clasă de predictori aparţine modelelor de prognoză care sunt răspândite în termeni de diferenţe ale datelor originale, deci, autoregresii diferenţiate de vector, denumite dVAR. Ele sunt instrumente de tradiţie în macroeconomie, începând din anii 1970, apoi sub forma analizei de serii cronologice Box-Jenkins şi modele ARIMA. Un fi r comun al evaluărilor de prognoză publicate ale modelelor macroeconometrice în analiză este abordat de Granger şi Newbold (1986). Se înţelege acum mai bine de ce dVAR (model de autoregresie diferenţiată cu vector) tinde să câştige în competiţiile de prognoză, datorită lucrărilor lui Clements şi Hendry (1996, 1998, 1999). Explicaţia rezidă în comparaţia formării de previziuni de mai sus: dVAR asigură prognoze solide ale seriilor cronologice nestabile care sunt supuse la modifi cări intermitente de regim. Pentru a le întrece, utilizatorul unui model macroeconometric trebuie sa recurgă în mod regulat la corecţii de segment şi corecţii al altor opinii.
Concluzii
Din studiul cu privire la particularitățile utilizării curbei Phillips în studiul evoluției șomajului, se desprinde concluzia, că dupa ajustările impuse de evoluția șomajului, se pot obține parametrii utilizabili la analiza macroeconomică. Curba Phillips de tip NAIRU se poate utilize pentru obținerea coefi cienților de probabilitate în cadrul studiilor prin sondaj statistic. Rezultă clar că la dimensionarea eșantioanelor trebuie să se țină seama de faptul că acesta trebuie să răspundă realității pentru a elimina eroarea de
Romanian Statistical Review - Supplement nr. 4 / 201932
reprezentatibilitate. Modelele econometrice condiționale sunt predispuse la instabilitate și chiar eșec ori de câte ori probabilitățile nemodelate se modifi că.
Bibliografi e 1. Anghelache, C., Anghel, M.G., Dumbravă, Ș.G., Ene, L. (2018). Analyzing the
employment rate of the population, unemployment and vacancies in the economy,
Theoretical and Applied Economics, XXV (2018), No. 2(615), Summer, 105-118
2. Anghelache, C., Anghel, M.G. (2018). Econometrie generală. Teorie și studii de caz, Editura Economică, Bucureşti
3. Anghelache, C., Anghel, M.G. (2016). Bazele statisticii economice. Concepte teoretice şi studii de caz, Editura Economică, Bucureşti
4. Deleplace, G. (2008). La absorción de la macroeconomía por la microeconomía.
Lecturas de Economía, 69 (69)
5. Ewing, B.T., Seyfried, W.L. (2003). Modeling the Philips Curve: A Time-Varying Volatility Approach, Euro-American Association of Economic Development in
Applied Econometrics and International Development
6. Ftiti, Z. (2010). The Macroeconomic Performance of the Inflation Targeting
Policy: An Approach Based on the Evolutionary Co-spectral Analysis, Economic Modelling, 27 (1), January, Elsevier
7. Isard, P., Laxton, D., Eliasson A.C. (2001). Infl ation Targeting with NAIRU Uncertainty and Endogenous Policy Credibility. International Monetary Fund in
IMF Working Papers
8. Jansen, E.S. (2002). Statistical issues in macroeconomic modelling (with
discussion). Scandinavian Journal of Statistics, 29 9. Ferri, P. (2000). Wage Dynamics and The Phillips Curve, in Blackhouse, R. E.
and A. Salanti (eds.), Macroeconomics and the Real World. Volume 2: Keynesian Economics, Unemployment and Policy, chap. 5. Oxford University Press, Oxford
10. Stanley, T.D. (2002). When All Are NAIRU: Hysteresis and Behavioural Inertia,
Taylor and Francis Journals in Applied Economics Letters
Revista Română de Statistică - Supliment nr. 4 / 2019 33
PARTICULARITIES OF USING THE PHILLIPS CURVE IN THE STUDY OF
UNEMPLOYMENT EVOLUTION
Assoc. prof. Mădălina-Gabriela ANGHEL PhD ([email protected])
„Artifex” University of Bucharest Ștefan Gabriel DUMBRAVĂ Ph.D Student ([email protected])
Bucharest University of Economic StudiesDaniel DUMITRU Ph.D Student ([email protected])
Bucharest University of Economic Studies
Abstract The coeffi cients of the unemployment equation, as a rule, are not subject
to natural rate expression. In the Phillips curve, defi ning the natural rate of
infl ation depends on the parameters of price and wage setting. This model is
actually a Phillips curve for prices that of course infl uences the real wage rate.
It is understandable that the high unemployment rate involves models that take
into account lags in the adjustment around a constant natural rate and allow the
change of balance. The concept that releases permanent changes to unemployment
is intense changes, with intermittent manifestations. From the Phillips curve
study it follows that the diff erence between natural rate and time independent
parameter is no longer realistic as it is a stochastic parameter that has a random
evolution. It follows that, in the case of a superexogenic hypothesis, the results for
a conventional economic model, such as the conventional Phillips curve, are not
invariant to a re-normalization. It should be borne in mind that the formulation
of classical regression is of behavioral nature determined by probabilities rather
than probabilities based on small models with estimated parameters.
Keywords: unemployment, Phillips curve, wages, prices, econometric
model
JEL Classifi cation: E24, E31, J31
Introduction In this article, the authors were concerned with highlighting the peculiarities of the Phillips curve in the study of unemployment. Starting from the functions of the Phillips curve in the analysis of unemployment, it is clear that only by combining the two functions can certain results be obtained. The study of the Phillips curve off ers only limited compatibility of the economic
Romanian Statistical Review - Supplement nr. 4 / 201934
explanation of exchanges that are determined by the infl ation rate. In fact,
macroeconomic outcomes corrected with the elimination of infl ationary
eff ects give an expression of the eff ect of infl ation. The concepts presented
highlight that permanent changes in unemployment are large-scale changes
in the rate of unemployment. Next, a careful presentation is made of the
uncertainty estimate of the NAIRU Phillips curve. The reason for the absence
of confi dence intervals in the NAIRU calculations is correlated with the fact
that this model is a non-linear function of the regression coeffi cients. The
article provides ample presentations on the statistical testing of the model and
the parameters obtained. Finally, the use of a dVAR model must permanently
use segment corrections and other fi xes that may occur.
Literature review
Anghelache, Anghel, Dumbravă and Ene (2018) conducted a study on
the correlation between the population occupancy rate, the unemployment rate
and the vacancy rate in the Romanian economy. Anghelache and Anghel (2018)
exposed the main methods and models of econometric analysis. Anghelache
and Anghel (2016) presented fundamental notions of economic statistics.
Deleplace (2008) addressed issues related to macroeconomic absorption by
microeconomics. Ewing and Seyfried (2003) studied aspects of modeling
the Phillips curve. Ftiti (2010) examined the elements of macroeconomic
performance of infl ation targeting policy. Isard, Laxton and Eliasson (2001)
studied infl ation. Jansen (2002) studied statistical problems in macroeconomic
modeling. Ferri (2000) analyzed aspects of wage dynamics and the Phillips
curve. Stanley (2002) studied the uncertainty of the NAIRU Phillips curve.
Research methodology, data, results and discussions
The coeffi cients of the unemployment equation do not enter into
the expression of the natural rate. In another variation of the Phillips curve,
expressing the natural rate depends on the price and wage setting parameters,
ie the model is specifi ed as a Phillips price curve rather than as a Phillips
salary curve. But the expression of the rate in a Phillips price curve remains
independent of the calculated parameters.
The fact that an important system property (unemployment) can be
estimated by a single equation leads in some way to explaining the popularity
of the Phillips curve. However, results based on an incomplete system analysis
off er limited information. In particular, single-based analysis provides
insuffi cient information about the dynamic properties of the system.
In the last years of the previous century European unemployment
rates have risen sharply and have not shown signs of return to the levels of the
Revista Română de Statistică - Supliment nr. 4 / 2019 35
1960s and 1970s. It is understood that high unemployment required models that fulfi ll the following functions: - take into account lags when adjusting around a constant natural rate; - allow change of balance. A combination of the two functions is also possible. Simply by virtue of being a dynamic system, the Phillips curve model includes slow dynamics). In principle, the adjustment coeffi cient βu1
in the unemployment equation can be arbitrarily small, as long as it is not
zero, uphil corresponds formally to the stable state of the system. However,
there is the question of how slow the speed of adjustments can be before the
idea of equilibrium becomes undermined „from within.” According to Phelps
(1995) and Friedman’s arguments, the natural rate should be almost stable and
should be a strong point of attraction for the eff ective rate of unemployment.
However, the experience of the 1980s and 1990s demonstrated that at best the
natural rate is a weak point of attraction. These are practical aspects of the
issue: Governors, looking at prospects after a negative economic shock, will
fi nd that the unemployment rate will eventually return to its natural level, but
only after 40 or more years.
In addition, the study of the Phillips curve provides only a limited
capability of economic explanation of changes in the unemployment rate.
It sometimes seems incredible that the change in the real growth rate can
only explain profound and persistent increases in the unemployment rate,
as happened in Europe. A nominal growth rate may of course suff er major
increases, but for those changes to have an impact on the equilibrium rate, the
long-term Phillips curve with downward slope is needed.
Thus, the Phillips curve is better suited to a stable regime characterized
by a low adjustment lag around a stable average rate of unemployment than to
the unemployment changes in Europe in the 1980s and 1990s. Consequently,
new models have developed in the 1980s models promised to explain the
changes in the unemployment equilibrium rate and concrete information on
how the structural characteristics of labor and commodity markets aff ect the
balance of unemployment. The new models have not reached the stage of
being unanimously accepted, as was the role of the Phillips curve.
The concepts presented above highlight that permanent changes in
unemployment are large-scale changes, with intermittent manifestations, in
line with the conception of unemployment . An alternative view is the
so-called natural variable rate over time (Gordon 1997, Gruen et al., 1999 and
Staiger et al., 1997). The basic idea is that the natural rate of unemployment
reacts to small-scale infl uences that occur frequently. According to this
opinion, the natural rate of unemployment is given by relationship:
Romanian Statistical Review - Supplement nr. 4 / 201936
(1) (2)
The diff erence is that the natural rate uν is no longer a time independent
parameter, but a stochastic parameter that follows the random path (2), and a
deviation which represents small scale infl uences. In estimating this pair
of equations (Kalman), standard error is limited from the beginning,
otherwise would change up and down and absorb all deviations from
left unexplained by explanatory variables. The methodology involves a
basic unit, both in the observed unemployment rate and in the natural rate of
unemployment. The practical relevance of this framework appears to be limited
to the US, where there are few major changes in the unemployment rate.
Regarding the natural rate of variable unemployment over time is
the concept of hysteresis. Economists have invoked the term „hysteresis” for
the case where the equilibrium rate could be the same as the delayed rate
of unemployment. However, there is a distinction between real delay as a
nonlinear and multiple equilibrium phenomenon and the linear property of
a root unit, as evidenced by Roed’s research (1994). In addition, it has been
shown that hysteresis is not an eff ective delay (in the true sense of the word as a
nonlinear phenomenon), and that the phenomenon of eff ective hysteresis leads
to an approach of unemployment that does not correspond to the hypothesis of
the natural rate.
• The estimation of the NAIRU Phillips curve uncertainty
This describes three approaches to estimating a „trust area” of the
NAIRU Phillips curve. The reason for the absence of confi dence intervals in
most NAIRU calculations is related to the fact that NAIRU is a non-linear
function of regression coeffi cients.
There are three methods that can be used to build NAIRU trusted
intervals: the Wald method, the Fieller method, and the probability coeffi cient
method. Fieller’s approach and probability coeffi cient formulas are preferable
due to the fi nite sample properties.
The fi rst and most intuitive approach is based on the associated
standard error and the t ratio for the estimated coeffi cients and corresponds to
a Wald statistic.
In the case of a full shift in productivity gains in wages and the lack
of „money illusion”, uphil from the Phillips curve type NAIRU is βw0/ βw1 and
its estimated value este .
We can write that:
Revista Română de Statistică - Supliment nr. 4 / 2019 37
(3) where uphil can be estimated directly by the least square nonlinear method. The result is numerically equivalent to the coeffi cient
derived from linear estimates ( ). In each case, a standard error for
u ^ phil can be calculated, from which confi dence intervals can be obtained
directly.
A confi dence interval includes the unrestricted estimate of uphil,
which is and a certain range around that value. Heuristically, the
confi dence interval contains each coeffi cient value that does not follow the
hypothesis:
(4)
We consider that is the Wald-based F-statistic for HW
verifi cation, and Pr(·)is the probability of its argument. A confi dence interval
of (1-α)% is defi ned by for
.
If βw1, the elasticity of the unemployment rate in the Phillips curve
is accurately estimated, the Wald approach is satisfactory. Small sample sizes
clearly jeopardize the precision of the estimate, but the size of the sample
depends on the amount of observation-based information. However, if βw1
is imprecisely estimated (not statistically signifi cant), this approach can be
misleading to a large extent. The Wald approach ignores how
behaves for near-zero values of where „relatively” refl ects the uncertainty
in estimating βw1. For European Phillips curves, βw1 estimates are typically
statistically insignifi cant that this concern is related to the calculation of
natural rates through the Phillips curve for Europe. In essence, the problem
arises because is a nonlinear function of the estimators ( ) that are
normally repaired.
The second approach avoids this problem by transforming the
nonlinear hypothesis into a linear one, that is:
(5)
The hypothesis and the corresponding F statistic are noted HF and
. Since the hypothesis is linear in the parameters βw0 and βw1, the
checks of this hypothesis are appropriate, even if is close to zero. The
determination of confi dence intervals is done exactly as in the Wald approach,
except that the F-statistic is built for .
Romanian Statistical Review - Supplement nr. 4 / 201938
The third approach uses the statistical verosimility coeffi cient to
calculate the confi dence interval for the HW hypothesis. The relationship is
estimated both non-restrictively and under the HW restriction, the corresponding
probabilities (residual amounts of squares for singular equations) are obtained,
and the confi dence interval is constructed from values of for which the
verosimity is less than the given critical value.
If the original model is linear in its parameters, Fieller’s solution is
numerically equivalent to the solution based on verosimilitude, providing a
previous generic justifi cation. If the estimated Phillips curve does not exhibit
dynamic homogeneity, is only a component of the NAIRU estimate
that would be appropriate to the underlying theory. This leads to more complex
calculations for the NAIRU, and the codispersion of terms like and
. However, unless the homogeneity distance has a
high numerical value, may be representative of the degree of
uncertainty associated with the estimated natural rate of the Phillips curve.
Similar statistical problems occur in other areas of macroeconomic application,
for example in the form of a „monetary conditions index”.
The reversal of dependent and independent variables is a continuing
controversy in the literature on infl ation modeling. The way in which the Lucas curve transforms the causality of the conventional Phillips curve was previously shown. Lucas’s critique also states that conventional Phillips Curve models will experience structural discontinuities whenever economic prospects change (for example, following a change in economic policy). In a hypothesis of superexogenicity, the results for a conventional econometric model, such as the conventional Phillips curve, are not invariant to a renormalization according to the formula: (6) where: ryx = correlation coeffi cient;
= the estimated regression coeffi cient when y is the dependent
variable and x is the regression factor;
= the estimated coeffi cient in the inverse regression.
The regime changes bring about alteration of the correlation structures,
so changes of ryx. If, due to superexogenity, is still constant, then can not
be constant.
The equation generally applies to ryx interpreted as a partial correlation
coeffi cient. So if (for example) the Phillips curve is estimated by OLS, then
fi nding that is stable entails that p for the re-normalized equation
Revista Română de Statistică - Supliment nr. 4 / 2019 39
(at the unemployment rate) is unstable. Finding that a stable Phillips curve on a sampling period that contains changes in partial correlations denies any claim that the model has an interpretation of Lucas’s demand curve. This simple procedure also applies to estimation by instrumental variables (due to endogenity and/or ), provided the number of instrumental variables is less than the number of endogenous variables in the Phillips curve. Conditional econometric models are prone to instability and failure whenever unmodifi ed probabilities change. Lucas criticism can be confi rmed or rejected empirically. Generally, a single chronological (random) yt variable is considered,
which can be divided into an explicated part and an unexplained independent part :
(7) We consider a plan attributable to agents and - the diff erence
between the planned and the real yt results. So:
(8)
is an innovation related to the plan, so:
(9)
We further assume that agents use a lot of I-1 information for rational
probabilities for the xt variable, ie:
(10)
and the probabilities are linked to:
(11)
which is motivated by economic theory.
From the formula , supposing that of (7) is an
innovation, we obtain:
(12)
And so:
(13)
Initially, it is assumed that follows a fi rst order AR process (non-static / instability is still considered): (14)
Romanian Statistical Review - Supplement nr. 4 / 201940
So, and
(15) For simplicity, suppose that and are independent. We will further assume that the only parameter of interest in equation (1) is β. The reduced form of yt is obtained from equations (7), (11) and (14):
(16)
where xt is weak exogenous for ξ = α1β, but the parameter of interest
β is unidentifi able only from (16). In addition, the reduced form equation (16),
while allowing estimation ξ in a state of nature characterized by stability, is
susceptible to Lucas criticism because ξ is not invariant to changes in the
autoregressive parameter of the marginal model (14).
Practically, Lucas’s critique aims at „behavioral equations” in systems
of simultaneous equations, for example,
(17)
with the term of deviation:
(18)
Directly it can be shown that the estimate (17) by OLS on a sample t
= 1,2....T, leads to:
(19)
establishing that the regression yt to xt does not represent the opposite
of (11).
Specifi cally, instead of β, the estimate of and the modifi cation of
α1 aff ects the stability of the estimates, thus confi rming the Lucas critique.
However, the applicability of the critique is based on the initial
assumptions. For example, if the hypothesis from in α1 = 1, is changed
so that xt has a unitary root but is cointegrated with yt, Lucas criticism does
not apply: in cointegration, since the cointegration parameter
is unique and can be consistently estimated by OLS.
As another example of the importance of the exact set of hypotheses
made, it is considered to be a replacement (11) of another economic theory,
namely the contingent plan:
(20)
Revista Română de Statistică - Supliment nr. 4 / 2019 41
Equations (20) and (7) lead to:
(21) where and β can be estimated
by OLS in the stationary case of .
• Model-based probabilities compared to data based
Apparently, it is often forgotten that the formulation of the „classical”
regression (21) corresponds to the idea: „behavior is driven by probabilities,
but not by model-based probabilities or rational probabilities with unknown
parameters that need to be estimated.”
To interpret the probabilities in (21), the relation (10) is replaced by:
(22)
and supposed to solve to obtain . The substitution of
and the use (7) to lead to (21).
is an example of a non-parametric invariant prediction
rule based on data properties or data-based probabilities. Realistically, it is
possible to choose the use of data-based predictors due to the costs of collecting
and processing information associated with model-based predictors.
It is true that approaches based on use a specifi ed model of the x (15) process and thus their prognoses will not reach the predicted average predictive error. So under the conditions of stability, there are gains from the estimate α1 of (15). However, there is virtually no guarantee that
the parameters of process x remain constant until the forecast horizon and the
unstable state of a model-based forecast can not be considered as better than
the prognosis derived from the simple rule . In fact, depending on
the date of the change in the „production” forecast, the data forecast will be
better than the model-based forecast in terms of interference.
We introduce a growth term in (15), ie:
(23)
and we assume there is a change in α0 (α0) during the T+1 period.
We consider two agents, A and B, predicting xT+1. Agent A collects
data for a period t = 1,2,3,....,T and can fi nd real values of {α0,α1} for that
period. However, because of the unpredicted change during the T+1
period, the predictive error of A will be:
(24)
Romanian Statistical Review - Supplement nr. 4 / 201942
Agent B, using the forecast based on data , will encounter a forecast error: which can be expressed in form: (25)
where means xT media (ie for the segment before the change
), . The comparison between relations (24) and (25) shows that the only diff erence between the two forecast errors is the term
of (25). Thus, both forecasts are aff ected by the regime
change that occurs after the forecast is made. Conditioned environments and
dispersions (distributions) of the two errors are:
(26)
(27)
(28)
by establishing that in this example (which takes into account a
post-prognosis regime change) there is no correspondence between the two
forecasting methods for the fi rst two moments of the forecast error. Conditional forecast error dispersions (distributions) are identical, and model-based forecasting infl uences are not lower than those based on data; suppose, for example, that, , at the same time , data-based interference may be less than model-based interference. In addition, unconditionally, the two predictors have the same infl uence and dispersion: (29)
(30) We further consider the forecasts made for the T + 2 period, provided T + 1 is an example of an ante-prognosis regime change ( durind the T+1 period). Unless A discovers the change in α0 and successfully corrects the forecast, the impact of the error will be again: (31)
The infl uence of forecasting error on agent B on the other hand becomes: , (32) where represents the unconditioned post-regime change x, ie
.
Revista Română de Statistică - Supliment nr. 4 / 2019 43
Clearly, the data-based predictor interference may be less than the predictive model-based error (but the opposite may be true.) However:
(33)
The unconditional forecast errors are always lower for the data-based forecast in the case of ante-progression mode change. The analysis generalizes the case of a unit root in the x process, in fact it is remarked from the previous ones that the data-based forecast errors have better properties for the case α1 = 1, for example din
(21). If xt is , then solving to obtain will result in a
prognosis with the same regime change as illustrated in the previous example.
This class of predictors belongs to forecasting patterns that are spread in terms
of diff erences in original data, so vector-diff erentiated autoregressions, called
dVAR. They are tools of tradition in macroeconomics since the 1970s, then in
the form of box-Jenkins time series and ARIMA models. A common thread
of published forecasts of macroeconomic models in analysis is addressed
by Granger and Newbold (1986). It is now better understood why the dVAR
(vector-diff erentiated autoregression model) tends to win in the forecast
competitions due to the works of Clements and Hendry (1996, 1998, 1999).
The explanation is based on the comparison of the above predictions: dVAR
provides solid forecasts for unstable time series that are subject to intermittent
regime changes. In order to overcome them, the user of a macroeconomic model
must regularly apply segment corrections and corrections of other opinions.
Conclusion
From the study on the particularities of the use of the Phillips curve in
the study of unemployment evolution, it is concluded that after the adjustments
imposed by the evolution of unemployment, the parameters useful for the
macroeconomic analysis can be obtained. The Phillips NAIRU curve can be
used to obtain probability coeffi cients in statistical surveys. It is clear that when
sampled, it must be taken into account that it must respond to reality in order to
eliminate the error of representativeness. Conditional econometric models are
prone to instability and even failure whenever undefi ned probabilities change.
Romanian Statistical Review - Supplement nr. 4 / 201944
References
1. Anghelache, C., Anghel, M.G., Dumbravă, Ș.G., Ene, L. (2018). Analyzing the employment rate of the population, unemployment and vacancies in the economy, Theoretical and Applied Economics, XXV (2018), No. 2(615), Summer, 105-118
2. Anghelache, C., Anghel, M.G. (2018). Econometrie generală. Teorie și studii de caz, Editura Economică, Bucureşti
3. Anghelache, C., Anghel, M.G. (2016). Bazele statisticii economice. Concepte teoretice şi studii de caz, Editura Economică, Bucureşti
4. Deleplace, G. (2008). La absorción de la macroeconomía por la microeconomía. Lecturas de Economía, 69 (69)
5. Ewing, B.T., Seyfried, W.L. (2003). Modeling the Philips Curve: A Time-Varying Volatility Approach, Euro-American Association of Economic Development in Applied Econometrics and International Development
6. Ftiti, Z. (2010). The Macroeconomic Performance of the Inflation Targeting Policy: An Approach Based on the Evolutionary Co-spectral Analysis, Economic Modelling, 27 (1), January, Elsevier
7. Isard, P., Laxton, D., Eliasson A.C. (2001). Infl ation Targeting with NAIRU Uncertainty and Endogenous Policy Credibility. International Monetary Fund in IMF Working Papers
8. Jansen, E.S. (2002). Statistical issues in macroeconomic modelling (with discussion). Scandinavian Journal of Statistics, 29
9. Ferri, P. (2000). Wage Dynamics and The Phillips Curve, in Blackhouse, R. E. and A. Salanti (eds.), Macroeconomics and the Real World. Volume 2: Keynesian Economics, Unemployment and Policy, chap. 5. Oxford University Press, Oxford
10. Stanley, T.D. (2002). When All Are NAIRU: Hysteresis and Behavioural Inertia, Taylor and Francis Journals in Applied Economics Letters
Revista Română de Statistică - Supliment nr. 4 / 2019 45
Principalele metode de previziunea economică
Drd. Andreea-Ioana MARINESCU ([email protected])
Academia de Studii Economice din București
Abstract
În cadrul acestui articol, se urmărește ilustrarea modului de utilizare practică a principalelor metode de previziune economică. Având în vedere faptul că, nevoia de a anticipa evenimente viitoarea este din ce în ce mai prezentă, rolul previziunii economice și al metodelor utilizate în acest sens, devine din ce în ce mai important, atât la nivel macroeconomic, cât și la nivel microeconomic. În scopul ilustrării modului de utilizare practică a patru dintre cele mai des întâlnite metode de previziune economică, atât la nivel național, cât și la nivel internațional, acest articol prezintă utilizarea metodelor de previzionare a nivelului veniturilor în cazul unei mici întreprinderi, luând în considerare diferite variabile care exercită o infl uență, fi e această directă sau indirectă, asupra veniturilor. Cuvinte cheie: anticipare, prognoză, trend, regresie, evoluție, metodă Clasifi carea JEL: C20, C30, C53
Introducere
Previzionarea este principalul obiectiv al analizelor economice și econometrice alături de modelarea bazelor fenomenelor economice. Fără îndoială, previzionarea este una dintre puținele activități efectuate de economiști academicieni și profesioniști care reprezintă un scop în sine, în sensul în care, a încerca să descoperi ce îți rezervă viitorul este un efort uman universal. Având în vedere preocupările tot mai mari în ceea ce privește anticiparea unor evenimente viitoare, au fost depuse eforturi considerabile, la nivel academic, pentru a pune bazele și pentru a dezvolta instrumente pentru o previzionare efi cientă. Pornind de la faptul că previziunea reprezintă cea mai cuprinzătoare expresie a anticipării, ea poate fi realizată, în general, prin două tipuri de metode și tehnici. Pe de o parte se afl ă metodele și tehnicile explorative (descriptive), utilizate pentru înțelegerea viitorului pe baza tendințelor conturate, iar pe de altă parte, metodele și tehnicile normative (prescriptive). În cazul metodelor explorative se urmărește observarea unei evoluții probabile, pe baza unor date istorice, respectiv a evoluțiilor conturate în trecut și prezent. În schimb, în cazul metodelor normative, sensul evenimentelor este direcționat, pe baza unor obiective prestabilite, care ajustează tendințele constatate în trecut și prezent, urmând a le înlocui cu alte tendințe. La nivel general, există patru metode de previzionare pe care analiștii economici le utilizează pentru a anticipa nivelul veniturilor, cheltuielilor și
Romanian Statistical Review - Supplement nr. 4 / 201946
costurilor de capital viitoare pentru o companie. Deși, la nivel global, există o mare varietate de instrumente de prognozare a bugetului, în acest articol ne vom concentra pe top patru metode, după cum urmează:
Nr. crt. Metoda Descriere Date necesare
1. Metoda liniară Se bazează pe ritmul constant de creștere al unei variabile Date istorice
2. Metoda mediilor mobile Se bazează pe prognoze repetate Date istorice
3. Regresia liniară simplă
Are la bază compararea unei variabile independente cu o variabilă dependentă
Eșantion de observații relevante
4. Regresia liniară multiplă
Are la bază compararea a mai mult de o variabilă independentă cu o variabilă dependentă
Eșantion de observații relevante
Literature review
Anghelache, Anghel și Manole (2015), Anghelache şi Anghel (2014) au prezentat principalele instrumente caracteristice modelării economice. Benjamin, Herrard, Houée-Bigot, Tavéra (2010) au aprofundat importanța utilizării econometriei ca instrument de prognoză. Anghelache și Anghel (2014) au aprofundat modelarea economică și principalii indicatori macroeconomici. Andrei şi Spătaru (2010) s-au concentrat asupra creării unor aplicații utile în studiul econometriei. Corbore, Durlauf și Hansen (2006) prezintă atât din perspectivă teoretică, cât și din perspectivă practică, principalele instrumente econometrice. Anghelache (2015) reprezintă o lucrare de referință în domeniul previziunii macroeconomice. Anghelache și Anghel (2015), Anghelache, Mitruţ şi Voineagu (2013), Anghelache (2008), Anghelache (coord., 2007), Anghelache, Isaic-Maniu, Mitruţ, Voineagu (2007), Dodge (ed., 2006) abordează statistica macroeconomică din perspectiva metodele și instrumentele care o defi nesc. Anghelache, Pagliacci și Prodan (2013) elaborează un model de analiză macroeconomică utilizând metoda regresiei. Anghel și Diaconu (2016), Anghelache (2009), Anghelache și Anghelache (2010) prezintă principalele modele dedicate previziunilor macroeconomice. Andreou, Ghysels și Kourtellos (2013) abordează importanța datelor fi nanciare în realizarea de prognoze macroeconomice. Anghelache, Panait, Marinescu și Niță (2017) au aprofundat unele modele și indicatori macroeconomici dedicați realizării de previziuni macroeconomice.
Metodologia cercetării, date, rezultate și discuții
În cadrul acestei cercetări ne vom axa pe detalierea principalelor patru metode utilizate în realizarea de previziuni economice, utilizând situații ipotetice descrie în Microsoft Excel.
Revista Română de Statistică - Supliment nr. 4 / 2019 47
• Metoda liniară
Metoda liniară este una dintre cele mai simple și ușor de folosit metode de previzionare. Un analist fi nanciar poate utiliza cifre și trenduri istorice în vederea anticipării nivelului veniturilor viitoare. Un exemplu în acest sens se regăsește mai jos:
Tabelul nr. 1
Date istorice Prognoză
2017 2018 2019 2020 2021 2022 2023Procentul de
creștere a vânzărilor5% 5% 5% 5% 5% 5% 5%
Venituri 583.400 612.000
Conform datelor istorice, în anul 2017 vânzările au crescut cu 5% față de 2016. Așa cum se poate observa în tabelul de mai sus, în anul 2018, creșterea vânzărilor a rămas la același nivel, respectiv 5%. Astfel, utilizând metoda liniară, vom presupune că nivelul creșterii vânzărilor va rămâne constant în viitor și vom folosi aceeași rată pe următorii 5 ani (2019-2023). Având în vedere rata de creștere a vânzărilor pentru anii 2019-2023, prognoza vânzărilor este următoarea:
Tabelul nr. 2
Date istorice Prognoză
2017 2018 2019 2020 2021 2022 2023Procentul de creștere a vânzărilor
5% 5% 5% 5% 5% 5% 5%
Venituri 583.400 612.000 612.001 612.002 612.003 612.004 612.005
• Metoda mediilor mobile Metoda mediilor reprezintă o tehnică prin care se analizează modelul (tiparul) unui set de date pentru a estima valori viitoare. Cele mai des întâlnite tipuri sunt mediile la 3 luni și cele la 5 luni. În general, metoda mediilor mobile se utilizează atunci fenomenul supus analizei prezintă oscilații sezoniere, ciclice etc. În tabelul de mai jos se poate observa modul de utilizare a mediilor mobile în vederea realizării unor de prognoze cu privire la nivelul veniturilor obținute de o companie.
Romanian Statistical Review - Supplement nr. 4 / 201948
Tabelul nr. 3Luna Venituri (RON) MM la 3 luni MM la 5 luni
Ianuarie 150.000
Februarie 175.000
Martie 183.240 169.413
Aprilie 166.780 175.007
Mai 154.000 168.007 165.804
Iunie 162.000 160.927 168.204
Iulie 165.000 160.333 166.204
August 167.450 164.817 163.046
Septembrie 165.200 165.883 162.730
Octombrie 167.000 166.550 165.330
Noiembrie 168.450 166.883 166.620
Decembrie 169.800 168.417 167.580
Mediile la 3 luni se calculează ca medie a veniturilor obținute în luna curentă și a celor obținute în ultimele două luni. Prima prognoză trebuie să înceapă în luna martie, calculând media veniturilor din lunile ianuarie – martie. În mod similar, mediile la 5 luni se calculează începând cu luna a cincea (în cazul nostru luna mai). Pentru a observa diferența dintre nivelul actual și cel previzionat, este
utilă realizarea unui grafi c, astfel:
Figura nr. 1
�
��
������
������
�����
������
�������
�������
�������
������
�������
�������
� � � � � � � �� �� ��
�������������� ������������ ������������
• Regresia liniară simplă
Revista Română de Statistică - Supliment nr. 4 / 2019 49
Regresia liniară simplă este o metoda foarte des utilizată în vederea prognozării relației dintre existente între două variabile. Pentru a continua exemplul nostru cu privire la metodele de prognozare a nivelului veniturilor, vom utiliza drept exemplu relația existentă între numărul anunțurilor radio și veniturile generate de către acestea, pe baza următoarelor date:
Tabelul nr. 4
LunaAnunțuri radio (nr.)
Venituri (EUR)
Ianuarie 31 9.460
Februarie 190 23.866
Martie 60 14.500
Aprilie 200 22.100
Mai 105 16.000
Iunie 56 13.400
Iulie 182 21.800
August 145 20.845
Septembrie 132 17.223
Octombrie 88 14.120
Noiembrie 118 16.700
Decembrie 210 26.400
Total 1.517 216.414
Medie 126 18.035
Relația existentă între numărul anunțurilor radio și veniturile generate
Figura nr. 2
�
����� �� � �!�� �����"����� ���
��
�����
������
������
������
������
������
� �� ��� ��� ��� ���
;���������.D-#
C������������������
Romanian Statistical Review - Supplement nr. 4 / 201950
Pe lângă crearea unei drepte utilizând modelul regresiei liniare simple, cu ajutorul Microsoft Excel se poate prognoza și nivelul venitului. De exemplu, să presupunem că dorim să prognozăm venitul generat în urma a 100 de anunțuri radio înregistrate. Utilizând funcția FORECAST asupra datelor prezentate în tabelul de mai sus, obținem următoarele rezultate:
Tabelul nr. 5 Anunțuri radio (nr.) Venituri (EUR)
Prognoză100 15.935150 21.061200 28.423
• Regresia liniară multiplă
O companie poate utiliza modelul regresiei liniare multiple în scopul analizării relației existente între două sau mai multe variabile independente și o variabilă dependentă. În exemplul de mai jos, vom realiza o regresie multiplă utilizând drept variabile costurile cu promoțiile realizate, costurile de promovare și veniturile generat, pe baza următoarelor date:
Tabelul nr. 6Luna Promoții (EUR) Promovare (EUR) Venituri (EUR)
Ianuarie 70 132 587Februarie 123 244 1.100Martie 173 332 1.300Aprilie 121 254 935Mai 118 251 887Iunie 121 312 780Iulie 215 254 1.640August 121 344 798Septembrie 178 254 687Octombrie 121 344 977Noiembrie 219 232 1.730Decembrie 70 197 598
Total 1.650 3.150 12.019
În vederea ilustrării modului practic de aplicare a metodei regresiei liniare multiple, vom considera drept variabilă dependentă (Y) nivelul veniturilor, iar ca variabile independente (X), costurile aferente organizării de vânzări promoționale (promoții) și cele aferente promovării gamei de produse, în cazul unei companii private. Astfel, utilizând funcția de regresie disponibilă în Microsoft Excel, obținem următoarele rezultate:
Revista Română de Statistică - Supliment nr. 4 / 2019 51
Regression Statistics
Multiple R 0.841411576
R Square 0.707973441
Adjusted R
Square0.64307865
Standard Error 226.3745445
Observations 12
ANOVA
df SS MS FSignifi cance
F
Regression 2 1118130.007 559065.0036 10.90956 0.00393
Residual 9 461208.9095 51245.43439
Total 11 1579338.917
Coeffi cients Standard Error t Stat P-value Lower 95%Upper
95%
Lower
95.0%
Upper
95.0%
Intercept 181.4819502 315.1316284 0.575892528 0.578802 -531.395 894.3592 -531.395 894.3592
Promoții (EUR) 6.623633195 1.451057677 4.564693258 0.001357 3.341113 9.906154 3.341113 9.906154
Promovare (EUR) -0.345326404 1.129070955 -0.305850047 0.766675 -2.89946 2.20881 -2.89946 2.20881
Rezultatele obținute în urma utilizării metodei regresiei liniare multiple pot fi folosite, de exemplu, pentru prognozarea nivelului veniturilor, pentru un nivel dat al costurilor cu promoțiile și a celor aferente promovării produselor. Tabelul nr.7
Promoții (EUR) 150 90 300 Promovare (EUR) 300 120 500
Venituri 1.071 736 1.996
Astfel, dacă o companie anticipează faptul că nivelul costurilor aferente vânzărilor promoționale va fi de 150 EUR, iar cel aferente promovării va fi de 300 EUR, utilizând coefi cienții obținuți prin intermediul utilizării
regresiei liniare multiple, obținem un nivel al veniturilor de 1.071 EUR.
Formula utilizată pentru prognozarea nivelului veniturilor în tabelul de mai
sus este următoarea:
Venituri = intercept + (cost promoții*coefi cient promoții) + (cost
promovare*coefi cient promovare)
Romanian Statistical Review - Supplement nr. 4 / 201952
Concluzii
Având în vedere nevoia din ce în ce mai mare de anticipare a evenimentelor viitoare și implicit de diminuare a riscurilor aferente, previziunea economică și metodele și instrumentele utilizate în acest sens, capătă un rol primordial în asigurarea succesului unei companii. Printre metodele de previziune cele mai răspândite la nivel global se numără metoda liniară, metoda mediilor mobile, metoda regresiei liniare simple și metoda regresiei liniare multiple.
Bibliografi e
1. Anghelache, C., Panait, M., Marinescu, I.A., Niță, G. (2017). Modele și indicatori utilizați în prognoza macroeconomică / Models and indicators used in macroeconomic forecast, Romanian Statistical Review, Supplement, no. 3, pp. 29-39 / 40-48;
2. Anghelache, C., Anghel, M.G. (2016). Bazele statisticii economice. Concepte teoretice și studii de caz, Editura Economică, București;
3. Anghelache, C. (2015). Previziune economică. Note de curs, format electronic, Editura Artifex, București;
4. Anghelache, C. and Anghel, M. (2014). Modelare economică. Concepte, teorie și studii de caz, București: Editura Economică;
5. Anghelache, C., Pagliacci, M.G.R. and Prodan, L. (2013). Model for macroeconomic - analyse based on the regression function. Romanian Statistical Review, 1, 18-30;
6. Anghelache, C. (2009). Modele de previziune economică, Conferința a 57-a „Statistica – trecut, prezent și viitor”, ISBN 978-90-73592-29-2, Durban 2009;
7. Anghelache, C. (2008). Tratat de statistică teoretică şi economică, București, Editura Economică;
8. Anghelache, C. (coordonator) (2007). Analiza macroeconomică – Sinteze și studii
de caz, Editura Economică, București; 9. Anghelache, C., Isac-Maniu, A., Mitruţ, C., Voineagu, V., Dumbravă, M. and
Manole, A. (2006). Analiza macroeconomică – Teorie şi studii de caz, București, Editura Economică;
10. Clark, T. and Ravazzolo, F. (2015). Macroeconomic Forecasting Performance under Alternative Specifi cations of Time-Varying Volatility, Journal of Applied
Econometrics, 30 (4), 551–575; 11. Corbore, D., Durlauf, S. and Hansen, B. (2006). Econometric Theory and
Practice – Frontieres of Analysis and Applied Research, Cambridge University Press, United Kingdom;
12. Manole, A., Anghel, M.G., Badiu, A. and Avram, D. (2017). Theoretical considerations regarding the main macroeconomic proportions and correlations. Romanian Statistical Review Supplement, 3, 169-181.
Revista Română de Statistică - Supliment nr. 4 / 2019 53
THE MAIN METHODS OF THE ECONOMIC PROVISION
Andreea – Ioana MARINESCU PhD Student ([email protected])
Bucharest University of Economic Studies
Abstract
This article aims to illustrate the practical use of the main methods of economic forecasting. Considering that the need to anticipate future events is increasingly present, the role of the economic forecast and the methods used in this respect is becoming increasingly important, both at macroeconomic and microeconomic level. In order to illustrate the practical use of four of the most common methods of economic forecasting at both national and international levels, this article presents the use of revenue forecasting methods for a small enterprise, taking into account diff erent variables that exert an infl uence,
whether direct or indirect, on revenue.
Keywords: anticipation, prognosis, trend, regression, evolution,
method
JEL Classifi cation: C20, C30, C53
Introduction
Forecasting is the main objective of economic and econometric analyzes along with the modeling of economic phenomena. Undoubtedly, forecasting is one of the few activities of academic and professional economists that are an end in itself, in the sense that trying to fi nd out what your future holds is a universal human eff ort. Given the growing concerns about anticipating
future events, considerable eff orts have been made at the academic level to lay
the foundations and develop instruments for eff ective anticipation. Starting
from the fact that the prediction is the most comprehensive expression
of anticipation, it can be generally achieved by two types of methods and
techniques. On the one hand, there are the exploratory (descriptive) methods
and techniques used to understand the future based on the outlined trends,
and on the other hand the prescriptive methods and techniques. In the case
of exploratory methods, it is intended to observe a probable evolution, based
on historical data, respectively on the past and present trends. Instead, in
the case of normative methods, the meaning of events is directed, based on
predetermined objectives, which adjusts the trends found in the past and
present, and will replace them with other tendencies.
Generally, there are four methods of forecasting that economic analysts
use to predict the level of revenue, expense, and future cost of a company.
Romanian Statistical Review - Supplement nr. 4 / 201954
Although globally there is a wide variety of budget forecasting tools, in this article we will focus on four methods as follows:
No. Method Description Required data
1.The linear method
It is based on the constant growth rate of a variable
Historical data
2.Mobile Media Method
It is based on repeated forecasts Historical data
3.Simple linear regression
It is based on the comparison of an independent variable with a dependent variable
Sample of relevant observations
4.Linear multiple regression
It is based on comparing more than one independent variable with a dependent variable
Sample of relevant observations
Literature review
Anghelache, Anghel and Manole (2015), Anghelache and Anghel (2014) presented the main tools characteristic of economic modeling. Benjamin, Herrard, Houee-Bigot, Tavéra (2010) have deepened the importance of using econometrics as a forecasting tool. Anghelache and Anghel (2014) have deepened economic modeling and main macroeconomic indicators. Andrei and Spataru (2010) focused on creating useful applications in the study of econometrics. Corbore, Durlauf and Hansen (2006) present both the theoretical and practical perspectives, the main econometric tools. Anghelache (2015) is a reference work in the fi eld of macroeconomic forecasting. Anghelache and Anghel (2015), Anghelache, Mitruţ and Voineagu (2013), Anghelache (2008), Anghelache (coord., 2007), Anghelache, Isaic-Maniu, Mitruţ, Voineagu macroeconomic statistics from the perspective of the methods and tools that defi ne it. Anghelache, Pagliacci and Prodan (2013) develops a model of macroeconomic analysis using the regression method. Anghel and Diaconu (2016), Anghelache (2009), Anghelache and Anghelache (2010) present the main models for macroeconomic forecasting. Andreou, Ghysels and Kourtellos (2013) address the importance of fi nancial data in making macroeconomic forecasts. Anghelache, Panait, Marinescu and Niţă
(2017) have deepened some models and macroeconomic indicators dedicated
to macroeconomic forecasting.
Research methodology, data, results and discussions
In this research we will focus on detailing the main four methods
used in the realization of economic forecasts, using hypothetical situations
described in Microsoft Excel.
Revista Română de Statistică - Supliment nr. 4 / 2019 55
• The linear method
The linear method is one of the most simple and easy to use forecasting methods. A fi nancial analyst can use historical fi gures and trends to anticipate future income levels. An example of this can be found below.:
Table no. 1
Historical data Forecast
2017 2018 2019 2020 2021 2022 2023Percentage of sales growth
5% 5% 5% 5% 5% 5% 5%
Income 583.400 612.000
According to historical data, in 2017 sales increased by 5% compared
to 2016. As can be seen in the table above, in 2018, sales growth remained at
the same level, ie 5%. Thus, using the linear method, we will assume that the
level of sales growth will remain constant in the future and we will use the
same rate over the next 5 years (2019-2023).
Considering the sales growth rate for the years 2019-2023, the sales
forecast is the following:
Table no. 2
Historical data Forecast
2017 2018 2019 2020 2021 2022 2023
Percentage of
sales growth5% 5% 5% 5% 5% 5% 5%
Income 583.400 612.000 612.001 612.002 612.003 612.004 612.005
• Mobile Media Method
The media method is a technique that analyzes the pattern (pattern) of
a data set to estimate future values. The most common types are the 3-month
and the 5-month averages. Generally, the mobile media method is used then
the phenomenon under analysis presents seasonal, cyclic oscillations, etc.
The table below shows how to use mobile environments to produce
forecasts of revenue levels achieved by a company.
Romanian Statistical Review - Supplement nr. 4 / 201956
Table no. 3
MonthIncome
(RON)
MM at 3
months
MM at 5
months
January 150.000
February 175.000
March 183.240 169.413
April 166.780 175.007
May 154.000 168.007 165.804
June 162.000 160.927 168.204
July 165.000 160.333 166.204
August 167.450 164.817 163.046
September 165.200 165.883 162.730
October 167.000 166.550 165.330
November 168.450 166.883 166.620
December 169.800 168.417 167.580
The 3-month averages are calculated as the average of the earnings achieved in the current month and those obtained over the past two months. The fi rst forecast should begin in March, calculating the average of January to March revenues. Similarly, the 5-month averages are calculated from the fi fth month (in our case, the month of May). To see the diff erence between the current and the predicted levels, it is
useful to make a chart as follows:
Figure no. 1
�
��
������
������
�����
������
�������
�������
�������
������
�������
�������
� � � � � � � �� �� ��
�������������� ������������ ������������
Revista Română de Statistică - Supliment nr. 4 / 2019 57
• Simple linear regression
The simple linear regression is a very common method for predicting the relationship between the two variables. To continue our example of revenue-forecasting methods, we will use as an example the relationship between the number of radio ads and the revenue generated by them based on the following data
Table no. 4
MonthRadio
announcements (no.)
Income
(EUR)
January 31 9.460
February 190 23.866
March 60 14.500
April 200 22.100
May 105 16.000
June 56 13.400
July 182 21.800
August 145 20.845
September 132 17.223
October 88 14.120
November 118 16.700
December 210 26.400
Total 1.517 216.414
Average 126 18.035
The existing relationship between the number of radio announcements
and generated venues
Figure no. 2
�
����� �� � �!�� �����"����� ���
��
�����
������
������
������
������
������
� �� ��� ��� ��� ���
;���������.D-#
C������������������
Romanian Statistical Review - Supplement nr. 4 / 201958
In addition to creating a straight line using the linear linear regression model, Microsoft Excel can also predict income levels. For example, suppose we want to forecast revenue generated from 100 registered radio ads. Using the FORECAST function on the data presented in the table above, we get the following results:
Table no. 5 Radio announcements (no.) Income (EUR)
Forecast100 15.935150 21.061200 28.423
• Multiple linear regression
A company can use the multiple linear regression model to analyze the relationship between two or more independent variables and a dependent variable. In the example below, we will achieve a multiple regression using as variable the costs of promotions, promotion costs and revenue generated, based on the following data:
Table no. 6
MonthPromotional sales
(EUR)
Promoting of the
product range (EUR)Income (EUR)
January 70 132 587
February 123 244 1.100
March 173 332 1.300
April 121 254 935
May 118 251 887
June 121 312 780
July 215 254 1.640
August 121 344 798
September 178 254 687
October 121 344 977
November 219 232 1.730
December 70 197 598
Total 1.650 3.150 12.019
In order to illustrate the practical way of applying the multiple linear regression method, we will consider the income level as a variable variable, and as independent variables (X), the costs of organizing promotional sales (promotions) and those related to the promotion of the product range, in the case of a private company. Thus, using the regression function available in Microsoft Excel, we get the following results:
Revista Română de Statistică - Supliment nr. 4 / 2019 59
Regression Statistics
Multiple R 0.841411576
R Square 0.707973441
Adjusted R Square 0.64307865
Standard Error 226.3745445
Observations 12
ANOVA
df SS MS F Signifi cance F
Regression 2 1118130.007 559065.0036 10.90956 0.00393
Residual 9 461208.9095 51245.43439
Total 11 1579338.917
Coeffi cientsStandard
Errort Stat P-value Lower 95%
Upper 95%
Lower 95.0%
Upper 95.0%
Intercept 181.4819502 315.1316284 0.575892528 0.578802 -531.395 894.3592 -531.395 894.3592
Promotional sales (EUR) 6.623633195 1.451057677 4.564693258 0.001357 3.341113 9.906154 3.341113 9.906154
Promoting of the product range (EUR)
-0.345326404 1.129070955 -0.305850047 0.766675 -2.89946 2.20881 -2.89946 2.20881
The results obtained from the use of the multiple linear regression method can be used, for example, for forecasting the level of revenue, for a given level of promotional and promotional costs.
Table no.7Promotional sales (EUR) 150 90 300
Promoting of the product range (EUR) 300 120 500
Income 1.071 736 1.996
Thus, if a company expects that the cost of promotional sales will be 150 EUR and the promotion rate will be 300 EUR, using the coeffi cients
obtained through the use of multiple linear regression, we get a level of
revenue of 1,071 EUR. The formula used to forecast revenue levels in the
table above is as follows:
Income = intercept + (cost of the Promotional sales *coeffi cient of the
Promotional sales) + (cost of the Promoting of the product range *coeffi cient
of the Promoting of the product range)
Romanian Statistical Review - Supplement nr. 4 / 201960
Conclusion
Considering the increasing need to anticipate future events and implicitly mitigate the related risks, the economic forecasting and the methods and tools used in this respect take a leading role in ensuring the success of a company. Among the most widely spread predictive methods are the linear method, the moving average method, the linear linear regression method, and the multiple linear regression method.
References
1. Anghelache, C., Panait, M., Marinescu, I.A., Niță, G. (2017). Modele și indicatori utilizați în prognoza macroeconomică / Models and indicators used in macroeconomic forecast, Romanian Statistical Review, Supplement, no. 3, pp. 29-39 / 40-48;
2. Anghelache, C., Anghel, M.G. (2016). Bazele statisticii economice. Concepte teoretice și studii de caz, Editura Economică, București;
3. Anghelache, C. (2015). Previziune economică. Note de curs, format electronic, Editura Artifex, București;
4. Anghelache, C. and Anghel, M. (2014). Modelare economică. Concepte, teorie și
studii de caz, București: Editura Economică; 5. Anghelache, C., Pagliacci, M.G.R. and Prodan, L. (2013). Model for macroeconomic
- analyse based on the regression function. Romanian Statistical Review, 1, 18-30; 6. Anghelache, C. (2009). Modele de previziune economică, Conferința a 57-a
„Statistica – trecut, prezent și viitor”, ISBN 978-90-73592-29-2, Durban 2009; 7. Anghelache, C. (2008). Tratat de statistică teoretică şi economică, București,
Editura Economică; 8. Anghelache, C. (coordonator) (2007). Analiza macroeconomică – Sinteze și studii
de caz, Editura Economică, București; 9. Anghelache, C., Isac-Maniu, A., Mitruţ, C., Voineagu, V., Dumbravă, M. and
Manole, A. (2006). Analiza macroeconomică – Teorie şi studii de caz, București, Editura Economică;
10. Clark, T. and Ravazzolo, F. (2015). Macroeconomic Forecasting Performance under Alternative Specifi cations of Time-Varying Volatility, Journal of Applied
Econometrics, 30 (4), 551–575; 11. Corbore, D., Durlauf, S. and Hansen, B. (2006). Econometric Theory and
Practice – Frontieres of Analysis and Applied Research, Cambridge University Press, United Kingdom;
12. Manole, A., Anghel, M.G., Badiu, A. and Avram, D. (2017). Theoretical considerations regarding the main macroeconomic proportions and correlations. Romanian Statistical Review Supplement, 3, 169-181.
Revista Română de Statistică - Supliment nr. 4 / 2019 61
Aspecte teoretice privind identifi carea structurii parţiale din submodele
Drd. Radu STOICA ([email protected])
Academia de Studii Economice din București
Abstract
Studiul asupra aspectelor teoretice privind identifi carea structurii parțiale din submodele scoate în evidență modul în care trebuie acționat pentru construirea de modele și formalizarea matematică a ecuațiilor corespunzătoare. În mod practic, în modelarea economiei naționale se construiesc mai întâi submodele specifi ce anumitor sectoare (ramuri) ale economiei naționale. În baza studiului efectuat se identifi că variabilele care sunt interdependente urmărindu-se in fi nal construirea unui model care să sugereze perspectiva evoluției. În practică se urmărește transformarea unui model ipotetic într-un model empiric. Teoria economică reprezintă principalul ghid în formularea modelelor economice. Este o distincție importantă între considerarea teoriei ca reprezentare corectă a realității și considerarea teoriei ca ghid de defi nire a unui model care să cuprindă elementele empirice. La baza construirii modelelor stă un număr de cinci criterii care urmărite și respectate, asigură crearea de modele apte de a releva prin parametrii obtinuți perspectiva de evoluție macroeconomică. Cuvinte cheie: parametru, submodel, procesele cointegrate, funcţie,
eroare
Clasifi carea JEL: C36, C50
Introducere
Studiul efectuat de autori cu privire la aspectele teoretice privind identifi carea structurii parțiale din submodele prezintă o serie de aspecte
care sunt esențiale în formularea modelelor macroeconomice. Se explică pe
larg etapele care stau la baza construirii modelelor în funcție de scopul fi nal
urmărit. Sunt abordate aspecte esențiale privind: identifi carea structurilor;
erorile care pot aparea în alegerea variabilelor; alegerea datelor admisibile;
identifi carea variabilelor exogene; identifi carea parametrilor care trebuie
obținuți; fi nalizarea modelului care elimină orice altă posibilitate opusă. Se
fac referiri cu privire la erorile segvențiale, dezvoltarea modelului de bază,
stabilirea exactă a structurii modelului considerat. Se fac referiri și la faptul
că un model congruent nu este neapărat unul cert, dar constituie o bază de
plecare în analiză. Modelul trebuie să aibă o funcție de distribuție, care pleacă
Romanian Statistical Review - Supplement nr. 4 / 201962
de la analiza econometrică empirică. Procesul descompunerii secvențiale în modele condiționate și marginale stau la baza subsistemelor RIMINI. Studiul parcurge în mod succinct și prezintă principalele elemente importante pentru modelare rezultate din analiza multor cercetători în acest domeniu.
Literature review
Anghelache, Anghel, M.G. et al. (2019) au studiat legătura dintre agregatele economice pe baza modelelor econometrice. Anghelache și Anghel (2017) au analizat utilizarea modelelor econometrice în activitatea macroeeconomică. Anghelache (2008) a prezentat indicatorii statistici utilizați în analizele economice. Clements și Hendry (1999) au abordat o serie de aspecte cu privire la prognoza economică. Colander (2009) a studiat elemente ale CVAR. Eitrheim, Jansen și Nymoen (2002) s-au referit la unele progrese de la eșecul prognozării. Florens (2003) a studiat variabilele instrumentale. Hendry (2003) a analizat originea metodologiei Econometrice LSE. Hendry (2002) a prezentat elementele importante ale econometriei. Lettau și Ludvigson (2005) a studiat aspecte referitoare la erorile din modelele statistic-econometrice.
Metodologia cercetării, date, rezultate și discuții
Procesul de creare a modelelor se confruntă adesea cu cereri de la utilizatorii de modele incompatibile cu un model format închis cu 3-5 ecuaţii. Adesea se lucrează cu submodele pentru diferite sectoare ale economiei. Astfel este folositoare abordarea în termeni de simplifi care a distribuţiei cu mai multe variabile, a tuturor variabilelor observabile din model prin calcularea de factori, condiţionare şi restricţii. • Vom considera distribuţia cu mai multe variabile, de forma:
(1) t = 1, ..., T şi
Determinarea factorilor funcţiei de densitate rezultă din: în
(2)care s-a consacrat ca distribuţia Haavelmo (Spanos 1989). Apreciem xt ca funcţie a , având condiţiile iniţiale x0 şi un vector parametru de timp unidimensional‚ . Presupunem că aceasta se apropie de procesul generator de date – DGP - Data Generator Process, proces generator de date (Hendry 1995a), care necesită termeni de eroare,
Revista Română de Statistică - Supliment nr. 4 / 2019 63
pentru a fi un proces inovativ. Această abordare este denumită “teoria reducţiei”, deoarece încearcă să explice originea modelelor empirice în termeni de operaţii de simplifi care (reducţie) implicite în DGP, pentru a induce modelul empiric relevant. A doua etapă în reducţia datelor este condiţionarea şi simplifi carea. Considerăm împărţirea (divizarea) , şi calculul factorilor pentru funcţia de densitate cu mai multe variabile într-o funcţie de densitate condiţionată pentru yt | zt şi o funcţie de densitate marginală pentru zt, respectiv:
(3)
În practică, se simplifi că ulterior prin utilizarea aproximărilor (procese Markov de ordin Kth) şi se dezvoltă modele pentru:
(4)
pentru t > k.
O clasă dinamică liniară generală de modele cu un număr fi nit de întârzieri care este folosită uzual, pentru a modela procesul n-dimensional, xt este VAR de ordin kth cu eroare gaussiană, adică:
, (5) unde e normal, independent şi reapartizat identic, N.i.i.d. (0, Λε). Un model VAR este un punct de plecare pentru analizarea relaţiilor cointegratoare care pot fi identifi cate în vectorul xt. Vom presupune, pentru simplifi care, că elementele xt sunt variabile nestaţionare (ce devin staţionare după ce sunt diferenţiate). Apoi, dacă există cointegrare, sistemul VAR are întotdeauna o reprezentare a unui model de corectare a echilibrului cu vector, ceea ce poate fi scrisă ca diferenţe şi niveluri (ignorând prezenţa posibilă a variabilelor determinate ca tendinţe) în următorul mod: , (6)unde α şi β sunt matrici n-r de nivel r (r<n), iar (β’xt-1) conţine relaţii cointegratoare r. Considerăm că procesele cointegrate defi nesc o traiectorie de echilibru de lungă durată şi presupun renunţarea la corecţiile de echilibru induse, care infl uenţează economia pe o cale stabilă. Acestea sunt utile deoarece permit interpretarea economică a proprietăţilor modelului, iar proprietăţile lor stabile
Romanian Statistical Review - Supplement nr. 4 / 201964
pot fi considerate ca o interpretare a unor echilibre pe termen lung între variabile economice rezultate din teoria economică. Consecvenţa teoretică, conform căreia modelul conţine structuri identifi cabile care pot fi interpretate în lumina teoriei economice, este doar un criteriu pentru o reprezentare satisfăcătoare a economiei. • Dacă se ţine cont de toate operaţiile de reducţie implicate în procesul de transformare a unui DGP ipotetic într-un model empiric, este puţin probabil ca un model econometric să coincidă cu DGP. Un model econometric poate, totuşi, să posede anumite proprietăţi care vor reda o reprezentare apropiată a DGP. Conform metodologiei LSE, un asemenea model ar trebui să satisfacă următoarele criterii: - Modelul conţine structuri identifi cabile şi interpretabile în lumina
teoriei economice. - Erorile trebuie să fi e inovaţii secvenţiale de variabile aleatorii, pentru
ca modelul să fi e o simplifi care valabilă a DGP; - Modelul trebuie să reprezinte date admisibile pe bază de observaţii
exacte; - Variabilele condiţionate trebuie să fi e exogene pentru parametrii din
model; - Parametrii trebuie să fi e constanţi în timp şi să rămână invariabili
la anumite categorii de infl uenţe (depinzând de scopul pentru care modelul urmează a fi folosit);
- Modelul trebuie să poată cuprinde modele rivale. Un model Mi cuprinde alte modelele (Mj, j≠i), dacă poate explica rezultatele obţinute de alte modele.
Modelele care satisfac primele cinci criterii sunt considerate congruente, în timp ce modelul cuprinzător congruent satisface toate cele şase criterii. Teoria economică este principalul ghid în formularea modelelor economice. O interpretare clară permite comunicarea de idei şi rezultate între grupurile de cercetători şi structurează dezbaterea despre probleme economice. Totuşi, în timp ce teoriile economice sunt abstracte şi construite pe ipoteze simplifi catoare, o transformare directă a conceptelor teoretice într-un model econometric nu va conduce în mod automat la un model satisfăcător. În ciuda interpretării lor structurale, aceste modele nu vor prezenta proprietăţi structurale specifi ce. Este o distincţie importantă între considerarea teoriei ca reprezentare corectă a realităţii (lăsând estimarea parametrilor la latitudinea econometricianului) şi considerarea teoriei ca ghid în defi nirea unui model care, de asemenea, cuprinde repere instituţionale, încercări de a reprezenta
Revista Română de Statistică - Supliment nr. 4 / 2019 65
heterogenitatea agenţilor, exprimă caracteristice temporale pentru seturile de date, etc. În mod asemănător, există o diferenţă semnifi cativă între o procedură de simplifi care secvenţială (în timp ce se controlează erorile de inovaţie) şi practica adoptării unui axiom a priori al unei specifi caţii corecte care implică prin ipoteză erori de zgomot alb. Erorile de inovaţie cu secvenţe de variabile aleatorii conduc la restricţia conform căreia resturile nu pot fi prevăzute numai din informaţiile proprii ale modelului. Această proprietate este consecinţa logică a unui proces de reducţie şi este o cerinţă necesară pentru modelul empiric derivat din DGP. Dacă erorile nu au această proprietate, adică nu sunt erori de zgomot alb, defi nirea datelor nu include regularitate. Cerinţa ca modelul să aibă date admisibile presupune că modelul nu trebuie să producă predicţii care nu sunt logic posibile. Criteriul patru înseamnă că parametrii sunt funcţii ale θy׀z care
variază independent de θx. Această proprietate este corelată cu efi cienţa estimării: exogenitatea slabă a variabilelor condiţionate zt este necesară pentru estimarea modelului condiţionat pentru yt, fără pierderea informaţiilor legate de estimarea modelului cu mai multe variabile pentru yt şi zt. Pentru a construi prognoze condiţionate pe baza modelului condiţional fără pierderi de informaţii, se cere o exogenitate puternică, ceea ce se defi neşte ca întâlnire între exogenitate slabă şi noncauzalitate Granger, reprezentând absenţa unui feedback de la yt la zt, adică x1
t-1 în funcţia de densitate marginală pentru zt, Dz (zt|x
1t-1, x0; λz), prin (2), nu include valori decalate ale yt.
Al cincilea criteriu din listă este dezvoltat în mod formal şi concis pe detalii de Hendry (1995a). El defi neşte structura ca o mulţime de repere
permanente de bază ale mecanismului economic. Un vector de parametri
defi neşte o structură, dacă este invariabilă şi caracterizează direct relaţiile
supuse analizei. Un parametru poate fi structural numai dacă este: - constant şi invariabil la extinderi ale perioadei studiate; - neschimbat de modifi cări în economie şi, astfel, invariabil la
modifi cări de regim, etc.;
- rămâne neschimbat pentru extinderi ale setului de informaţii şi,
astfel, invariabil la adăugarea mai multor variabile în analiză.
Această proprietate de invariabilitate are o importanţă specială pentru
un program de cercetare progresiv: în mod ideal, modelarea empirică este un
proces cumulativ în cadrul căruia modelele sunt în mod continuu inlocuite
de unele noi şi mai utile. Modelele considerate utile sunt acele modele care
posedă proprietăţile structurale defi nite anterior, în special modele care sunt relativ invariabile la schimbări ale economiei, adică, conţin parametri autonomi. Modelele cu un înalt grad de autonomie sunt modele structurale,
Romanian Statistical Review - Supplement nr. 4 / 201966
care rămân invariabile la schimbări de politici economice şi alte şocuri ale sistemului economic. Totuşi, caracterul structural este limitat de două considerente: în primul rând, autonomia este un concept relativ deoarece un model econometric nu poate fi invariabil la orice şoc posibil; al doilea, este improbabil ca toţi parametrii ai unui model econometric să aibă caracter invariabil. Parametrii cu cel mai înalt grad de autonomie reprezintă structura parţială. De exemplu: elementele unui vector β într-o ecuaţie cointegratoare, care reprezintă adesea
structura parţială. Chiar dacă e improbabil ca submodelele să conţină structura
parţială în acelaşi grad, este plauzibil ca modelele foarte agregate să fi e mai puţin autonome decât submodelele, deoarece acestea se pot construi pe o mulţime mai bogată de informaţii. Congruenţa datelor, respectiv, abilitatea de a caracteriza datele, rămâne o calitate esenţială a modelelor econometrice. În acest sens, strategia noastră de cercetare este aceea de a verifi ca orice model general ipotetic care este ales ca punct de plecare a unei căutări de specifi caţie pentru congruenţa datelor şi să decidem asupra modelului fi nal, după o cercetare de la general la specifi c. Datorită progreselor recente în teoria şi practica construirii modelelor bazate pe date, este cunoscut faptul că prin utilizarea algoritmilor Gets, un cercetător are o şansă de a ajunge la o aproximare de înaltă acurateţe a procesului generator de date, iar pericolul de erori aleatorii este redus. Un model congruent nu este un model neapărat cert. O inovaţie poate fi predictibilă din alte informaţii, nu numai din setul de informaţii propriu. Rezultă că poate fi dezvoltată o secvenţă de modele congruente, fi ecare dintre acestea poate cuprinde toate modelele anterioare. O strategie care pune accent pe comportamentul prognozat, fără a evalua atent cauzele eşecului prognozei ex post, atrage riscul unor modele diferite care conţin importante elemente structurale. Principala cauză a eşecului de prognoză o reprezintă evoluţiile deterministe (de exemplu rata de echilibru a economiilor populaţiei) şi nu modifi carea unor coefi cienti precum înclinaţia spre consum, preocupare de prim rang în analiza politicilor. Discontinuităţile structurale sunt o preocupare majoră în modelarea econometrică, dar singurul mod de a evalua calitatea unei discontinuităţi ipotetice rezultă din confruntarea cu datele. În plus, deoarece urmează o abordare cuprinzătoare, un eşec de prognoză reprezintă o potenţială îmbunătăţire. • Funcţia de distribuţie completă nu este uşor de utilizat, de aceea nu este un punct de plecare operaţional pentru analiza econometrică empirică. În practică, trebuie să divizăm sistemul în subsisteme de variabile şi să le analizăm separat. Modelarea cu mai multe variabile e considerată numai în
Revista Română de Statistică - Supliment nr. 4 / 2019 67
cadrul subsistemelor. Apare însă riscul ca anumite infl uenţe posibile asupra subsistemelor să fi e ignorate, ceea ce conduce la condiţionare nevalabilă (prezumţia de exogenitate slabă nu este satisfacută) şi marginalizare nevalabilă (prin omiterea variabilelor explicative relevante din analiză), care implică estimare şi inferenţă statistică inefi cientă. Implementarea practică a acestor principii este exemplifi cată prin modelarea sectorului gospodăriilor populaţiei tip RIMINI (Model macroeconomic dezvoltat de Norges Bank – Banca Centrală a Norvegiei). Procesul descompunerii secvenţiale în modele condiţionate şi marginale se repetă în cadrul subsistemelor RIMINI. În subsistemul sectorului gospodăriilor populaţiei, cheltuielile totale de consum sunt modelate ca funcţie a venitului real disponibil şi avuţiei reale a populaţiei. Avuţia totală este formată din valoarea reală a stocului de capital imobiliar, plus avuţia fi nanciară netă. Suma activelor fi nanciare nete reale este egală cu diferenţa dintre activele fi nanciare brute reale şi împrumuturile (creditele) reale (Mt-Lt): (7) unde: Ht - volumul stocului imobiliar rezidenţial, (PH)t/Pt - preţul real al imobilelor, Pt - defl atorul pentru cheltuielile de consum.
Funcţia de distribuţie cu mai multe variabile pentru acest subsistem poate fi defi nită prin relaţia xt = (cht, yht, wht). Submodelul condiţionat pentru cheltuielile totale reale de consum este de forma:
(8) bazându-ne pe funcţia de densitate condiţionată corespondentă ca o reprezentare valabilă a DGP. Modelul RIMINI conţine submodele pentru yht şi pentru toate componentele individuale din wht. De exemplu, submodelul condiţionat pentru determinarea simultană a preturilor imobilelor, pht, şi împrumuturilor reale ale populatiei, lt, este:
, (9) unde: Rlt - rata dobânzii la împrumuturi (credite) De asemenea, în modelul RIMINI se regăsesc submodele pentru adaosul net la stocul de capital imobiliar şi preţul capitalului imobiliar, phnt:
Romanian Statistical Review - Supplement nr. 4 / 201968
sau
(10) unde: pjt - defl atorul investiţiilor brute în clădiri.
Modelul pentru consum agregat, descris de Brodin şi Nymoen (1992) satisface criteriile prezentate anterior. În cadrul modelului, analiza cointegrării stabileşte că relaţia liniară: cht = constant + 0,56yht + 0,27wht, (11) este relaţia cointegratoare şi că rangul cointegrării este unu. Rezultă că, în timp ce variabilele individuale din relaţia liniară se presupun a fi nestaţionare şi integrate, combinaţia liniară a celor trei variabile este staţionară cu o valoare medie constantă care arată discrepanţa dintre consum şi nivelul de echilibru pe termen lung (0.56yht +0.27wht). Modelele marginale estimate pentru venit şi avuţie constituie un argument pentru studiul discontinuităţilor structurale. Apariţia simultană a unui model condiţionat stabil (funcţia de consum) şi a unor modele marginale instabile este dovada invariabilităţii coefi cienţilor modelului condiţionat şi, de aici, variabile condiţionate super exogene (venit şi avuţie). Rezultatul invariabilităţii este coroborat, folosind o metodă alternativă bazată pe modele de tranziţie. Funcţia empirică de consum s-a dovedit a fi relativ stabilă pentru mai mult de un deceniu şi se aplică de regulă unei părţi cointegratoare a ecuaţiei. Comparaţia cu modele alternative este importantă pentru studiile consumului. Dereglementarea fi nanciară de la mijlocul anilor 1980 a condus la o creştere puternică a consumului agregat faţă de venit în mai multe ţări europene. Utilizarea funcţiilor macroeconometrice empirice de consum, funcţii care, în mod tipic, au explicat consumul agregat prin venit, în prognozare şi în explicarea datelor ex post, au condus la rezultate nesatisfăcătoare. O concepţie asupra acestor eşecuri de prognoză se constituie într-o demonstraţie directă în favoarea probabilităţilor raţionale rivale, a ipotezei venitului permanent; ca răspuns la dereglementarea fi nanciară, consumatorii şi-au previzionat venituri permanente în creştere, creându-se astfel o discontinuitate în relaţia condiţionată dintre consum şi venit. De asemenea, apariţia acestor eşecuri a fost interpretată ca o confi rmare a relevanţei criticii lui Lucas. Se poate compara efi cienţa a două modele concurente: funcţia empirică de consum, cu condiţia unui venit pe termen lung, şi ecuaţia Euler derivată dintr-un model de probabilitate. În timp ce funcţia de consum condiţionată cuprinde ecuaţia Euler pe o
Revista Română de Statistică - Supliment nr. 4 / 2019 69
perioadă de sondaj din 1968 până în 1984, ambele modele eşuează în prognoza creşterii anuale a consumului în următorii ani. Vom deduce proprietăţile teoretice ale prognozelor bazate pe cele două modele. Considerând că ecuaţia Euler este modelul real şi că funcţia de consum este un model cu specifi caţie greşită, ambele seturi de prognoze sunt imune la discontinuităţi (de exemplu, modifi carea ratei de echilibru a economiilor populaţiei) care se manifestă după realizarea prognozei. În plus, eşecul prognozelor funcţiei empirice de consum este posibil dacă funcţia este un model real. Ca urmare, eşecul de prognoză al funcţiei confi rmă demonstraţia în favoarea unei funcţii condiţionate pentru perioada anterioară producerii evenimentului. Totuşi, funcţia de consum respecifi cată, care a introdus avuţia ca o nouă variabilă, a avut succes în considerarea căderii ex post, în acelaşi timp susţinând nivelul constant al parametrilor în anii de consolidare fi nanciară care au urmat după reducerea iniţială a ratei economiilor populaţiei. Modelul respecifi cat a putut fi luat în considerare, în mod corespunzător, pentru variabilitatea ridicată a ratei economiilor, în timp ce modele anterioare nu au avut succes în acest demers. S-a explicat de ce critica Lucas a fost slabă în acest caz: în primul rând, discontinuitatea observată a funcţiilor condiţionate de consum în 1984-85 corespunde criticii Lucas, în sensul că interpretarea este respinsă prin identifi carea unui model condiţionat cu parametri constanţi. În al doilea rand, rezultatul invariabilităţii este similar modelelor tip ecuaţie Euler (derivate dintr-un model stochastic/aleatoriu de venit permanent) şi nu poate fi un model cuprinzător. Chiar dacă abordarea Euler are ca sprijin parametri empiric constanţi, nu se poate explica de ce un model condiţionat este stabil. În al treilea rând, constatarea că invariabilitatea este aproximată empiric formează o bază importantă pentru utilizarea funcţiei dinamice de consum în prognozarea şi analiza politică.
Concluzii
În urma studiului efectut de autori rezultă că metodologia de constituire a modelelor este una foarte importantă, deoarece numai prin considerarea variabilelor correlate se pot obține rezultatele dorite. Modelele utilizate trebuie să se bazeze pe submodele care să facă parte din modelul general utilizat. În urma formalizării matematice a modelelor rezultă posibilitatea de a calcula parametrii pentru stabilirea evoluției empirice neafectată de infl uența factorilor sau dinpotrivă infl uențată de aceștia. Pe baza acestor aspecte teoretice, se pot identifi ca structurile și submodelele care pot fi considerate în analiza pe baza modelului general. În activitatea practică se pot utilize modele econometrice care dau siguranță analizelor macroeconomice.
Romanian Statistical Review - Supplement nr. 4 / 201970
Bibliografi e 1. Anghelache, C., Anghel, M.G. et al. (2019). Econometric model used for GDP
correlation analysis and economic aggregates. Economic Computation and Economic Cybernetics Studies and Research, 53 (1), 183-197
2. Anghelache, C., Anghel, M.G. (2017). Econometric methods and models used in the analysis of the factorial infl uence of the gross domestic product growth. Network Intelligence Studies, V (9), (1), 67-78
3. Anghelache, C. (2008). Tratat de statistică teoretică şi economică, Editura Economică, Bucureşti
4. Clements, M.P., Hendry D.F. (1999). On Winning Forecasting Competitions in Economics. Spanish Economic Review, 1(2), 123-160
5. Colander, D. (2009). Economists, incentives, judgment, and the European
CVAR approach to macroeconometrics, Kiel Institute for the World Economy in Economics: The Open-Access, Open-Assessment E-Journal
6. Eitrheim, Ø., Jansen, E., Nymoen, R. (2002). Progress from forecast failure-the Norwegian consumption function. Econometrics Journal, 5
7. Florens, J.P. (2003). Inverse problem and structural econometrics: the example
of intstrumental variables, in Dewatripont, M., Hansen L.P., and Turnovsky S.J. (Eds.). „Advances in economics and econometrics: theory and applications”, Eighth World Congress, Vol. II
8. Hendry, D.F. (2003). J. Denis Sargan and the Origins of LSE Econometric Methodology. Econometric Theory, 19(3), 457-480
9. Hendry, D.F. (2002). Applied econometrics without sinning. Journal of Economic
Surveys, 16 10. Lettau, M., Ludvigson,S.C. (2005). Euler Equation Errors, National Bureau of
Economic Research, Inc in NBER Working Papers
Revista Română de Statistică - Supliment nr. 4 / 2019 71
THEORETICAL ASPECTS REGARDING THE IDENTIFICATION OF PARTIAL STRUCTURE
IN SUBMODES
Radu STOICA Ph.D Student ([email protected])
Bucharest University of Economic Studies
Abstract
The study of the theoretical aspects regarding the identifi cation of the partial structure of the submodels highlights the way in which it is necessary to act for the construction of models and the mathematical formalization of the corresponding equations. Practically, in the modeling of the national economy, the sub-models specifi c to certain sectors (branches) of the national economy are fi rst built. Based on the study, we identify the variables that are interdependent, aiming fi nally to construct a model to suggest the evolution perspective. In practice, it is intended to transform a hypothetical model into an empirical model. Economic theory is the main guide in formulating economic models. It is an important distinction between considering theory as the correct representation of reality and considering theory as a guide to defi ning a model that includes empirical elements. At the basis of the model construction, there are fi ve criteria that are followed and observed, ensuring the creation of models able to reveal the prospects of macroeconomic evolution through the obtained parameters. Keywords: parameter, submodel, cointegrated processes, function, error JEL Classifi cation: C36, C50
Introduction
The author’s study of the theoretical aspects of identifying the partial structure of sub-models presents a number of issues that are essential in formulating macroeconomic models. It explains extensively the stages underlying the construction of the models according to the fi nal goal pursued. Essential aspects are addressed on: identifying structures; errors that may occur in the choice of variables; the choice of admissible data; identifi cation of exogenous variables; identifying the parameters to be obtained; completing the model that eliminates any other opposed possibility. Reference is made to segmented errors, the development of the basic model, exact determination of the structure of the considered model. Reference is also made to the fact that a matching model is not necessarily a certain one, but it is a starting point in the analysis. The model must have a distribution function, starting from
Romanian Statistical Review - Supplement nr. 4 / 201972
the empirical econometric analysis. The process of sequential decomposition in conditional and marginal models is the basis of the RIMINI subsystems. The study briefl y presents the main important elements for modeling resulting from the analysis of many researchers in this fi eld.
Literature review
Anghelache, Anghel, M.G. et al. (2019) studied the link between economic aggregates based on econometric models. Anghelache and Anghel (2017) analyzed the use of econometric models in macroeconomic activity. Anghelache (2008) presented the statistical indicators used in economic analyzes. Clements and Hendry (1999) addressed a number of aspects of economic forecasting. Colander (2009) studied CVAR elements. Eitrheim, Jansen and Nymoen (2002) referred to some progress from the failure of forecasting. Florens (2003) studied instrumental variables. Hendry (2003) analyzed the origin of the Econometrice LSE methodology. Hendry (2002) presented the important elements of econometrics. Lettau and Ludvigson (2005) studied aspects relating to errors in statistical and econometric models.
Research methodology, data, results and discussions
The modeling process often faces requests from incompatible model users with a closed-ended model of 3-5 equations. Often work with submodels for diff erent sectors of the economy. Thus, it is useful to approach in terms of
simplifying the multi-variable distribution of all variables observable in the
model by calculating factors, conditioning and restrictions.
• We will consider distribution with several variables of the form:
(1)
t = 1, ..., T
and
Determination of density function factors results from:
in
(2)
which became the Haavelmo distribution (Spanos 1989).
We consider xt as the function of , having the initial conditions x0
and a one-dimensional time parameter vector‚ . Suppose it approaches the
data generating process - DGP - Data Generator Process, a data generator process
(Hendry 1995a), which requires error terms,
to be an innovative process. This approach is called the „reduction theory”
because it attempts to explain the origin of the empirical models in terms of
Revista Română de Statistică - Supliment nr. 4 / 2019 73
implied simplifi cation operations in the DGP to induce the relevant empirical model. The second step in reducing data is conditionality and simplifi cation. We consider the division , and the calculation of factors for the multi-variable density function in a conditional density function for yt | zt and a function of marginal density for zt, respectively: (3) In practice, it is further simplifi ed by using approximations (Markov processes of Kth order) and developing models for:
(4)for t > k.
A linear generalized dynamic class of models with a fi nite number of delays that is commonly used to model the n-dimensional process, xt is VAR, kth with Gaussian error, ie:
, (5) where is normal, independent and repaired identically, N.i.i.d. (0, Λε).
A VAR model is a starting point for analyzing cointegrating
relationships that can be identifi ed in the xt vector.
We will assume, for simplicity, that the xt elements are non-static
variables (which become stationary after they are diff erentiated). Then, if
there is cointegration, VAR always has a representation of a vector equilibrium
correction model, which can be written as diff erences and levels (ignoring the
possible presence of variables determined as trends) in the following way:
, (6)
where α and β are n-r matrices r (r<n), and (β’xt-1) contain cointegration
relationships r.
We consider cointegrated processes to defi ne a long-lasting equilibrium
trajectory, and suppose the abandonment of induced equilibrium corrections,
which infl uence the economy in a stable way. These are useful because they
allow the economic interpretation of the model properties, and their stable
properties can be considered as an interpretation of long-term equilibria
between economic variables resulting from economic theory. The theoretical
Romanian Statistical Review - Supplement nr. 4 / 201974
consistency, according to which the model contains identifi able structures that can be interpreted in light of economic theory, is only a criterion for a satisfactory representation of the economy. • If all the reduction operations involved in the transformation of a hypothetical DGP into an empirical model are taken into account, an econometric model is unlikely to coincide with DGP. An econometric model may, however, possess certain properties that will give a close representation of DGP. According to the LSE methodology, such a model should meet the following criteria: - The model contains identifi able and interpretable structures in the
light of economic theory. - Errors must be sequential innovations of random variables, so that
the model is a valid simplifi cation of DGP; - The model should be admissible data based on exact observations; - Conditional variables must be exogenous for model parameters; - Parameters must be constant over time and remain invariant to
certain categories of infl uences (depending on the purpose for which the model is to be used);
- The model must be comprised of rival models. A Mi model includes other models (Mj, j ≠ i), if it can explain the results obtained by other models.
Models meeting the fi rst fi ve criteria are considered congruent, while the compgressional model satisfi es all six criteria. Economic theory is the main guide in formulating economic models. A clear interpretation allows the communication of ideas and results between groups of researchers and structure the debate on economic issues. However, while economic theories are abstract and built on simplifying assumptions, a direct transformation of theoretical concepts into an econometric model will not automatically lead to a satisfactory pattern. Despite their structural interpretation, these models will not have specifi c structural properties. It is an important distinction between considering theory as the correct representation of reality (leaving the estimation of parameters to the econometrician’s latitude) and considering theory as a guide in defi ning a model that also includes institutional benchmarks, attempts to represent the heterogeneity of agents, express temporal characteristics for sets data, etc. Similarly, there is a signifi cant diff erence between a sequential simplifi cation procedure (while controlling innovation errors) and the practice of adopting an a priori axiom of a correct specifi cation that implicitly implies white noise errors. Innovation errors with random variable sequences result in the restriction that debris can not be predicted only from the model’s own
Revista Română de Statistică - Supliment nr. 4 / 2019 75
information. This property is the logical consequence of a reduction process and is a necessary requirement for the empirical model derived from DGP. If errors do not have this property, ie there are no white noise errors, data defi nition does not include regularity. The requirement for the model to have admissible data implies that the model should not produce predictions that are not logically possible. Criterion four means that the parameters are functions of θy׀z which
varies independently of θx. This property correlates with the estimation
effi ciency: the weak exogenity of conditional variables zt is required to
estimate the conditional model for yt without losing the information related
to estimating the multi-variable model for yt and zt. To construct conditional
forecasts based on the conditional model without loss of information, a
strong exogenity is required, which is defi ned as the encounter between weak
exogenity and Granger noncausality, representing the absence of feedback
from yt to zt, ie, x1t-1 in the function of marginal density for zt, Dz (zt|x
1t-1, x0;
λz), by (2), does not include the yt decal values.
The fi fth criterion in the list is formally and concisely developed
on details by Hendry (1995a). It defi nes the structure as a set of permanent
permanent benchmarks of the economic mechanism. A parameter vector
defi nes a structure if it is invariant and directly characterizes the relationships
underlying the analysis. A parameter can only be structural if it is:
- constant and invariable on the extensions of the studied period;
- unchanged from changes in the economy and thus invariably in
regime changes, etc.;
- remains unchanged for expanding the set of information and thus
invariably adding more variables to the analysis.
This invariability property is of special importance for a progressive
research program: ideally, empirical modeling is a cumulative process in which
patterns are continually replaced by new and more useful. Models considered
useful are those models that possess the previously defi ned structural properties,
especially models that are relatively invariable to economic changes, that is,
they contain autonomous parameters. Models with a high degree of autonomy
are structural models that remain invariable for economic policy changes and
other shocks of the economic system.
However, the structural character is limited by two considerations:
fi rst, autonomy is a relative concept because an econometric model can not
be invariably invoked at any possible shock; the second, it is unlikely that
all the parameters of an econometric model will have invariable character.
Parameters with the highest degree of autonomy represent the partial structure.
For example: the elements of a vector β in a cointegration equation, which
Romanian Statistical Review - Supplement nr. 4 / 201976
often represents the partial structure. Even if it is unlikely that the submodels contain the partial structure in the same degree, it is plausible that the highly aggregated models are less autonomous than the submodels, because they can build on a richer set of information. Congruence of data, respectively, the ability to characterize data, remains an essential quality of econometric models. In this sense, our research strategy is to check any hypothetical general model that is chosen as the starting point of a specifi cation search for data congruence, and to decide on the fi nal model after a general-to-specifi c research. Due to recent advances in the theory and practice of building data-based models, it is known that by using Gets algorithms, a researcher has a chance to achieve a high accuracy approximation of the data generation process, and the danger of random errors is low. A matching model is not necessarily a certain pattern. An innovation can be predicted from other information, not just from its own set of information. It follows that a sequence of congruent models can be developed, each of which can include all previous models. A strategy that focuses on the predicted behavior, without carefully assessing the causes of the ex post forecast failure, draws the risk of diff erent
models that contain important structural elements.
The main cause of the forecast failure is represented by the deterministic
evolutions (eg the equilibrium rate of the population’s economies) and not the
modifi cation of some coeffi cients such as the propensity for consumption, a
prime concern in the policy analysis. Structural discontinuities are a major
concern in econometric modeling, but the only way to assess the quality
of hypothetical discontinuity results from data confrontation. In addition,
because it follows a comprehensive approach, a forecast failure is a potential
improvement.
The full distribution function is not easy to use, so it is not an operational
starting point for empirical econometric analysis. In practice, we need to divide
the system into subsystems of variables and analyze them separately. Multiple
variable modeling is only considered within subsystems. However, there is a
risk that some possible infl uences on the subsystems will be ignored, leading to invalid conditioning (the assumption of poor exogenity is not satisfi ed) and invalid marginalization (by omitting the relevant explanatory variables in the analysis), involving ineffi cient estimation and inferential inference. The
practical implementation of these principles is exemplifi ed by the modeling of the RIMINI households sector (Macroeconomic model developed by Norges Bank - Norway Central Bank). The process of sequential decomposition in conditional and marginal
Revista Română de Statistică - Supliment nr. 4 / 2019 77
patterns is repeated within the RIMINI subsystems. In the subsystem of the households sector, total consumption expenditure is modeled as a function of real disposable income and real wealth of the population. Total wealth is made up of the real value of the stock of real estate plus the net fi nancial wealth. The amount of real net fi nancial assets is equal to the diff erence between the real
gross fi nancial assets and the real loans (loans) (Mt-Lt):
(7)
where:
Ht - the volume of residential real estate stock,
(PH)t/Pt - real estate price,
Pt - defl ator for consumer spending. The multi-variable distribution function for this subsystem can be defi ned by the relation xt = (cht, yht, wht). The conditional submodel for actual total consumption expenditure is of the form:
(8) based on the corresponding conditional density function as a valid representation of DGP. The RIMINI model contains sub-models for yht and for all individual components in wht. For example, the conditional submodel for the simultaneous determination of real estate prices, pht, and real borrowings of the population, lt is:
, (9) where: Rlt - interest rate on loans (loans). Also in the RIMINI model are the submodels for the net addition to the stock of real estate capital and the price of the real estate capital, phnt: or
(10) where: pjt - defl ator of gross investments in buildings. The Aggregate Consumption Model described by Brodin and Nymoen (1992) meets the criteria outlined above. Within the model, cointegration analysis determines that the linear relation: cht = constant + 0,56yht + 0,27wht, (11) is the cointegrative relationship and the rank of cointegration is one. It follows that while the individual variables in the linear relationship are assumed to be non-static and integrated, the linear combination of the three
Romanian Statistical Review - Supplement nr. 4 / 201978
variables is stationary with a constant average value showing the discrepancy between consumption and the long-term equilibrium level (0.56yht +0.27wht). Estimated income and wealth margins are an argument for the study of structural discontinuities. The simultaneous occurrence of a stable conditioning model (consumption function) and of unstable marginal models is proof of the invariance of the conditional model coeffi cients and hence of
super exogenous conditional variables (income and wealth). The result of
invariance is corroborated using an alternative method based on transition
models.
The empirical consumption function has proven to be relatively stable
for more than a decade, and is usually applied to a cointegrate part of the
equation. Comparison with alternative models is important for consumption
studies.
Financial deregulation in the mid-1980s led to a strong increase in
aggregate consumption versus income in several European countries. The
use of empirical consumer macroeconomic functions, functions that typically
explained aggregate income, forecasting, and explanation of ex post data,
have led to unsatisfactory results.
A conception of these forecast failures is a direct demonstration in
favor of rival rational probabilities, of the permanent income hypothesis;
in response to fi nancial deregulation, consumers predicted steady revenue growth, thus creating a discontinuity in the conditioned relationship between consumption and income. Also, the occurrence of these failures was interpreted as a confi rmation of the relevance of Lucas’s critique. One can compare the effi ciency of two competing models: the
empirical consumer function, with a long-term income, and the Euler equation
derived from a probability model.
While the conditional consumption function includes the Euler
equation over a sampling period from 1968 to 1984, both models fail to predict
the annual increase in consumption in the coming years. We will deduce the
theoretical properties of forecasts based on the two models. Considering that
the Euler equation is the real model and that the consumption function is a
misleading model, both sets of forecasts are immune to discontinuities (eg,
the change in the equilibrium of population economies) that occur after the
prognosis. In addition, the empirical consumption function prognosis failure
is possible if the function is a real model. As a result, the forecast failure of the
function confi rms the demonstration in favor of a conditional function for the period preceding the occurrence of the event. However, the respected consumption function, which introduced wealth as a new variable, was successful in considering the ex post fall, while
Revista Română de Statistică - Supliment nr. 4 / 2019 79
maintaining the constant level of parameters in the fi nancial consolidation years following the initial reduction in the population’s savings rate. The corresponding model could have been duly taken into account for the high variability of the savings rate, while previous models were not successful in this approach. It was explained why Lucas’s criticism was weak in this case: fi rst, the observed discontinuity of consumer-conditioned functions in 1984-85 corresponds to Lucas’s criticism, meaning that interpretation is rejected by identifying a conditional model with constant parameters. Secondly, the result of invariance is similar to the Euler equation models (derived from a stochastic / random income pattern) and can not be a comprehensive model. Even though Euler’s approach supports constant empirical parameters, one can not explain why a conditioning model is stable. Third, the fi nding that invariance is approximated empirically forms an important basis for using the dynamic consumption function in political forecasting and analysis.
Conclusion
As a result of the authors’ study, the modeling methodology is very important, because only by considering the correlated variables the desired results can be obtained. The models used must be based on the submodels that are part of the general model used. The mathematical formalization of the models results in the possibility to calculate the parameters for the determination of the empirical evolution unaff ected by the infl uence of the factors or infl uenced by them. Based on these theoretical aspects, we can identify the structures and submodels that can be considered in the analysis based on the general model. In practice, econometric models can be used to ensure macroeconomic analysis.
References
1. Anghelache, C., Anghel, M.G. et al. (2019). Econometric model used for GDP correlation analysis and economic aggregates. Economic Computation and Economic Cybernetics Studies and Research, 53 (1), 183-197
2. Anghelache, C., Anghel, M.G. (2017). Econometric methods and models used in the analysis of the factorial infl uence of the gross domestic product growth. Network Intelligence Studies, V (9), (1), 67-78
3. Anghelache, C. (2008). Tratat de statistică teoretică şi economică, Editura Economică, Bucureşti
4. Clements, M.P., Hendry D.F. (1999). On Winning Forecasting Competitions in
Economics. Spanish Economic Review, 1(2), 123-160
5. Colander, D. (2009). Economists, incentives, judgment, and the European
CVAR approach to macroeconometrics, Kiel Institute for the World Economy in
Economics: The Open-Access, Open-Assessment E-Journal
6. Eitrheim, Ø., Jansen, E., Nymoen, R. (2002). Progress from forecast failure-the
Norwegian consumption function. Econometrics Journal, 5
Romanian Statistical Review - Supplement nr. 4 / 201980
7. Florens, J.P. (2003). Inverse problem and structural econometrics: the example of intstrumental variables, in Dewatripont, M., Hansen L.P., and Turnovsky S.J. (Eds.). „Advances in economics and econometrics: theory and applications”, Eighth World Congress, Vol. II
8. Hendry, D.F. (2003). J. Denis Sargan and the Origins of LSE Econometric Methodology. Econometric Theory, 19(3), 457-480
9. Hendry, D.F. (2002). Applied econometrics without sinning. Journal of Economic Surveys, 16
10. Lettau, M., Ludvigson,S.C. (2005). Euler Equation Errors, National Bureau of Economic Research, Inc in NBER Working Papers
Revista Română de Statistică - Supliment nr. 4 / 2019 81
Elemente esenţiale privind modelarea şi combinarea subsistemelor
Prof. univ. dr. Radu Titus MARINESCU ([email protected])
Universitatea „Artifex” din BucureștiProf. univ. dr. Constantin ANGHELACHE ([email protected])
Academia de Studii Economice din București / Universitatea „Artifex” din BucureștiConf. univ. dr. Mădălina-Gabriela ANGHEL ([email protected])
Universitatea „Artifex” din BucureștiDrd. Alexandru BADIU ([email protected])
Academia de Studii Economice din București
Abstract
În activitatea practică se pune problema ca modelele construite să se bazeze pe combinarea subsistemelor care sunt în concordanță cu structura fenomenului supus analizei. În dezvoltarea modelelor macroeconomice pe scară largă, abordarea tradițională este aceea de a exprima printr-o ecuație posibilitatea stabilirii rezultatului. Prin adoptarea metodelor de probabilitate maximă cu informații limitate, se pot obține parametrii unei ecuații, în timp ce parametrii altor ecuații rămân restricționați. După construirea submodelelor relevante prin marginalizare și condiționare, modele reduse de subsectoare pot fi aggregate într-un model larg macroeconomic. Modelele macroeconomice sunt utilizate ca bază în stabilirea strategiilor economice, de aceea este important să se modeleze coefi cienții tuturor variabilelor explicative. Deaceea obținerea proprietăților pe termen lung ale subsistemelor capătă o importanță deosebită. Astfel, putem vorbi de analiză pe baza modelului infl ației și a curbei Phillips. Urmare a descoperirii de către Phillips a unei regularități empirice între rata șomajului și creșterea salariului, aceasta a fost o noutate preluată de mai toți cercetătorii de renume din perioada următoare. Cuvinte cheie: subsistem, curba Phillips, critica Lucas, coefi cient, ecuație Clasifi carea JEL: C44, C61, E31
Introducere În acest articol, pe baza studiului efectuat de autori, se urmărește identifi carea elementelor esențiale utilizate apoi în modelarea și combinarea subsistemelor. Sunt prezentate în mod argumentat metodele utilizate de diverși cercetători care au condus la realizarea de modele macroeconomice pe baza
Romanian Statistical Review - Supplement nr. 4 / 201982
cărora se pot stabili strategii de evoluție macroeconomică. În continuare, se face o amplă analiză a modelului infl ației și a curbei Phillips care sunt utilizate și în present. Deasemenea, sunt reliefate aspectele de care depinde aplicația curbei Phillips. Relația între creșterea salariului și a activitatea economică a fost analizată la nivel macroeconomic încă din perioada neo-clasică. O versiune empirică a curbei Phillips denumită modelul triunghiular al infl ației s-a răspândit fi ind utilizată și în condițiile actuale, bineînteles cu ajustările care se impuneau. Sistemul curbei Phillips pentru salarii intr-o economie deschisă reprezintă o specifi cație completă a modelului dinamicii infl ației.
Literature review
Anghelache, Anghel et al (2017) au evidențiat o serie de elemente privind utilizarea curbei Philips în analizele macroeconomice. Anghelache și Anghel (2014), precum și Anghelache (2012) au analizat aspectele principale ale modelării economice. Bjerkholt (2005) a studiat noțiuni referitoare la modele și planifi care. Dickens (2008) a examinat metode de estimare a varianței de timp. Ewing și Seyfried (2003) au examinat concepte ale modelării curbei Phillips. Johansen (2002) au studiat testele ipotezelor privind vectorii de cointegrare. Karanassou și Snower (2007) au abordat aspecte privind revizuirea Curbei Philips. Lee și Aaker (2006) au prezentat un studiu Monte Carlo al regresiilor de creștere. Levy (2004) au studiat elemente ale cointegrării în domeniul frecvențelor. Mandel și Tomšík (2003) au prezentat principalele aspecte privind funcția de consum.
Metodologia cercetării, date, rezultate și discuții
• În dezvoltarea de modele macroeconometrice pe scara largă, abordarea tradiţională a fost aceea de a estima o ecuaţie (sau submodel) odată şi a stabili rezultatele simultan. Adesea, aceasta s-a facut fără a se verifi ca adecvarea procedurilor. Abordarea ar putea să fi e justifi cată din punct de vedere al estimărilor. Prin adoptarea metodelor de probabilitate maximă cu informaţii limitate, se pot estima parametrii unei ecuaţii, în timp ce parametrii altor ecuaţii rămân nerestrictionaţi, aşa cum arată studiile lui Anderson şi Rubin (1949) şi Koopmans şi Hood (1953). S-a argumentat că metodele cu informaţii limitate au fost mai puternice faţă de ecuaţiile eronat specifi cate din sistem, în situaţiile în care au existat teorii mai bune sau informaţii demne de încredere despre o submulţime de variabile. Metodele bazate pe informaţii limitate au fost adoptate din considerente practice, spre a evita complexitatea calculelor, specifi că metodelor cu informaţii complete – precum probabilitatea maximă bazată pe informaţii complete. După construirea submodelelor relevante prin marginalizare şi
Revista Română de Statistică - Supliment nr. 4 / 2019 83
condiţionare, modelele mici de subsectoare sunt agregate într-un model larg macroeconometric. O teorie generală pentru cele trei etape va conţine invariabil criterii şi condiţii, care sunt formulate pentru întregul sistem, ceea ce ridică problema dacă există căi de rezolvare atunci când modelul complet este prea complex pentru o modelare simultană. O soluţie ar fi folosirea modelelor cu grad ridicat de agregare care sunt sufi cient de mici pentru a fi analizate ca sistem complet. Într-o asemenea abordare sunt omise mai multe mecanisme economice, importante şi relevante pentru demersul de a descrie economia în mod adecvat. Abordarea generală poate fi vazută ca una de gradualism – urmărind stabilirea structurii în submodele. Primele două condiţii din defi niţia structurii parţiale nu obligă la cunoaşterea modelului complet. Excluderea din model a unor variabile explicative importante conduce la nevalabilitatea submodelelor, care se detectează în cadrul submodelului atunci când se modifi că modalitatea de corelare între variabilele incluse şi cele care au fost omise. Pentru ultimele condiţii este posibil, cel putin în principiu, să se dezvolte un model, ca ultimă extensie a informaţiilor, stabilind astfel că structura sau structura parţială reprezintă o modalitate de a rezolva problema lui Johansen. Practic, se ştie că modelul complet este inaccesibil. Totuşi, funcţia condiţionată de consum este constantă atunci când intervalul de sondaj este extins cu nouă ani de observaţii trimestriale; rămâne nealterată pe perioada de dereglementare fi nanciară şi susţine modelarea simultană a consumului privat, a venitului disponibil al gospodăriilor populaţiei, avuţiei populatiei şi preţurilor reale ale imobilelor. Astfel se poate valida dovada inductivă că funcţia condiţionată de consum va reprezenta o structură parţială. Modelul simultan este în acest caz cu greu un substitut ideal pentru un model mai bun pentru efecte tip “ofertă”, care operează prin piaţa muncii, cu toate acestea oferă protecţie faţă de opiniile alternative. De exemplu, venitul este în fapt corecţia echilibrului, nu consumul. Modelele macroeconomice sunt utilizate ca bază în stabilirea politicilor economice. Din acest punct de vedere, este important să se modeleze coefi cienţii tuturor variabilelor explicative relevante prin condiţionarea tuturor cunoştintelor (relevante şi aplicabile) despre condiţiile instituţionale în societatea studiată. Ne bazăm pe specifi caţii cu un nivel de agregare mai ridicat, unde coefi cientii bruţi creeaza efecte combinate ale variabilelor incluse, iar variabilele corelate omise pot duce la recomandări politice greşite. Asemenea interferenţe sunt mai dăunătoare pentru guvernanţi decât interferenţa de simultaneitate care poate apare prin combinarea submodelelor.
Romanian Statistical Review - Supplement nr. 4 / 201984
Dacă aceasta rămâne valabilă sau nu este o problemă interesantă, care poate fi explorată prin intermediul simulărilor Monte Carlo pe specifi caţii de model particulare. Rezultă că obţinerea proprietăţilor pe termen lung ale submodelelor este de o importanţă specială. Odată ce o ecuaţie cointegratoare este determinată, este invariabilă la extensii ale mulţimii de informaţii. Pe de altă parte, aceasta este o proprietate care necesită a fi stabilită în fi ecare caz. Noţiunile de separare în sisteme cointegrate sunt descrise de Granger şi Haldrup (1997). Ideea este de a descompune fi ecare variabilă într-o componentă persistentă şi o componentă tranzitorie. În cadrul unui model de corectare a echilibrului cu vectori, se consideră două subsisteme, unde variabilele unui subsistem nu includ ecuaţii cointegratoare ale celuilalt subsistem (separare de cointegrare). Totuşi pot apare efecte pe termen scurt ale variabilelor dintr-un subsistem asupra variabilelor celuilalt şi ecuaţiile cointegratoare ale unui sistem pot afecta dezvoltarea pe termen scurt al variabilelor celuilalt. Absenţa ambelor tipuri de interacţiune se numeşte separare completă, în timp ce, dacă numai unul dintre acestea este prezent, devine separare parţială. Aceste concepte sunt strâns legate de exogenitatea variabilelor într-un subsistem privind parametrii celuilalt. Ambele submodele, parţial sau complet separate, sunt ipoteze verifi cabile care trebuie să fi e testate ca parte a analizei cointegrării. Hecq et al. (2002) extind rezultatele lui Granger şi Haldrup (1997). Concluzia lui Hecq et al. (2002) este, totuşi, că verifi carea separării necesită întregul sistem să fi e cunoscut, ceea ce corespunde cu observaţia lui Johansen. • Modelul infl aţiei şi curba Phillips îşi au originea în aceeaşi epocă a macroeconomiei. Dar în timp ce modelul lui Aukrust s-a depărtat de mediul academic, literatura referitoare la curba Phillips s-a dezvoltat în anii 1960 şi a realizat un impact imens în următoarele patru decenii. În continuare sunt redate câteva din etapele semnifi cative în dezvoltările curbei Phillips. În anii 1970, curba Phillips şi modelul lui Aukrust au fost considerate ca alternative, reprezentând modelul de infl aţie tip “cerere” şi “ofertă” (Frisch 1977). Totuşi, aşa cum arată Aukrust (1977), diferenţa între considerarea pieţei muncii ca sursă importantă a infl aţiei şi atenţia deosebită acordată de curba Phillips pieţei produsului este mai mult o chestiune de accent decât de principiu, cele două mecanisme pot opera împreună. În continuare, se arată modul în care cele două abordări, din punct de vedere formal, pot fi combinate, acordând curbei Phillips rolul de relaţie pe termen scurt a creşterii salariului nominal, în timp ce teza principală e valabilă pe termen lung. În cadrul capitolului sunt abordate probleme esenţiale pentru aplicarea curbei Phillips în contextul actual, reprezentarea sa într-un sistem de variabile
Revista Română de Statistică - Supliment nr. 4 / 2019 85
cointegrate; consecvenţa şi modifi cările ratei şomajului; incertitudinea curbei Phillips estimate NAIRU (rata naturală a şomajului, Non-Accelerating Infl ation Rate of Unemployment); şi statutul curbei Phillips inversate, respectiv curba ofertei a lui Lucas. • Urmare a descoperirii, de către Phillips, a unei regularităţi empirice între rata şomajului şi creşterea salariilor în Marea Britanie, curba Phillips a fost integrată în macroeconomie printr-o serie de lucrări în anii ‘60. Samuelson şi Solow (1960) au interpretat-o ca un compromis cu care se confruntă guvernanţii, şi Lipsey (1960) a fost primul care a estimat curbele Phillips cu tehnici multivariate de regresie. Lipsey a interpretat relatia din perspectiva dinamicii clasice a preţurilor cu rata şomajului acţionând ca intermediar între cererea în exces şi fricţiunile de pe piaţa forţei de muncă. Important, Lipsey a inclus creşterea preţurilor de consum ca variabilă explicativă în regresiile sale şi astfel a formulat ceea ce a devenit cunoscut drept curba Phillips a probabilităţilor. Dezvoltări ulterioare includ distincţia între curba Phillips pe termen scurt unde infl aţia se abate de la infl aţia previzionată şi curba Phillips pe termen lung, unde previziunile de infl aţie se îndeplinesc. În fi nal, conceptul unei rate naturale a şomajului a fost defi nit ca rata stabilă a şomajului, corespunzând unei curbe verticale pe termen lung (Phelps 1968 şi Friedman 1968). Relaţia între cresterea salariului şi activitatea economica fi gurează de asemenea în macroeconomia neo-clasică (Lucas şi Rapping, 1969 şi 1970 şi Lucas 1972). Totuşi, în economia neo-clasică, relaţia de cauzalitate din modelul original al lui Phillips a fost răsturnată: dacă există o corelaţie între infl aţie şi şomaj în mod absolut, cauzalitatea merge de la infl aţie la nivelul activităţii şi şomaj. Inversiunea lui Lucas şi Rapping se bazează pe aprecierea că nivelul preţurilor este ancorat în relaţia teoriei cantitative şi a stocului autonom de bani. Creşterea preţurilor şi salariilor este determinată în afara curbei Phillips, astfel încât formularea corectă ar fi aceea că pe de-o parte se situează rata şomajului, iar creşterea salariilor (şi/sau infl aţia) pe cealaltă parte. Lucas (1972) lansează o altă abordare cunoscută, bazată pe probabilităţi raţionale despre preţurile relativ incerte ale produselor. Dacă probabilităţile sunt îndeplinite (în medie), oferta agregată este nemodifi cată faţă de ultima perioadă. Totuşi, dacă sunt preţuri-surpriză, apare o distanţare faţă de nivelul mediu al producţiei pe termen lung. Astfel se stabileşte o relaţie tip ofertă “surprise only”. Funcţia de ofertă a lui Lucas este inversul curbei verticale Phillips pe termen lung, mărită cu previzionările lui Lipsay, dar derivată cu ajutorul teoriei microeconomice şi a ipotezei probabilităţilor rationale.
Romanian Statistical Review - Supplement nr. 4 / 201986
În plus, pentru specifi caţiile convenţionale ale cererii agregate (Romer 1996), modelul implică o asociere pozitivă între producţie şi infl aţie sau o relaţie negativă între rata şomajului şi infl aţie. Astfel, există o corespondentă neo-clasică pentru curba Phillips pe termen scurt. Totuşi, curba cererii a lui Lucas, atunci când se aplică asupra unor date şi este estimată prin metoda celor mai mici pătrate simple (OLS), nu reprezintă o relaţie cauzală care poate fi exploatată de guvernanţi. Dimpotrivă, se va schimba atunci când, de exemplu, oferta de bani a crescut pentru a stimula producţia într-un mod care lasă politica fără posibilitatea de a infl uenţa producţia reală sau şomajul. Acest concept este denumit critica Lucas (Lucas 1976), care a fost formulată ca o critică a compromisului infl aţie-şomaj al curbei Phillips, tratată în literatura academică precum şi în modelele macroeconometrice ale anilor 1970 (Wallis 1995). Forţa criticii provine totuşi din generalitatea sa: este un pericol potenţial pentru toate modelele econometrice condiţionate. Problema cauzalităţii se manifestă în legătură cu ultimele versiuni ale curbei Phillips – varianta neo-keynesiană. În SUA, o versiune empirică a curbei Phillips, denumită “modelul triunghiular al infl aţiei”, s-a răspândit în pofi da criticii Lucas (contribuţii în acest sens se regăsesc în lucrările lui Gordon 1983 şi 1997, Staiger şi alţii 2001). Aşa precum vom argumenta în continuare, o explicaţie a valabilităţii curbei Phillips americane constă în faptul că şocurile ratei şomajului au fost în general de o magnitudine mai mică decât în ţările europene. • Aşa cum am arătat anterior, există multe moduri prin care o curbă Phillips pentru o economie deschisă poate fi dedusă din teoria economică. Aprecierea noastră asupra curbei Phillips se bazează pe Calmfors (1977), care a reconciliat curba Phillips cu modelul scandinav al infl aţiei. Ne propunem să facem un pas înainte, totuşi, şi să încorporăm curba Phillips într-un cadru care ţine cont de serii de date pentru salarii şi preţuri. Reconstruirea modelului în termeni de cointegrare şi cauzalitate relevă că versiunea curbei Phillips a modelului principal impune un mecanism de corectare a echilibrului asupra sistemului. Astfel, în timp ce este consecventă cu teoria principală a lui Aukrust, curba Phillips este şi un model special deoarece include numai unul din mecanismele de stabilire a salariilor discutate de Aukrust. Fără a ne îndepărta de generalitate, ne vom concentra asupra salariilor în curba Phillips şi amintim că, în concordanţă cu teoria lui Aukrust, se presupune că: 1. , posibil după îndepărtarea schimbărilor deterministe; şi 2. structura cauzală este “o cale” reprezentată de H4mc şi H5mc.
Revista Română de Statistică - Supliment nr. 4 / 2019 87
Consecvenţa cu cointegrarea asumată şi cauzalitatea necesită existenţa
unui model de corecţie a echilibrului pentru rata salariilor nominale în sectorul
expus. Presupunând o dinamică de prim ordin pentru simplifi care, un sistem
tip curba Phillips se defi neste prin următoarele două ecuaţii:
•
(1)
•
unde (notaţia se simplifi că prin renunţarea (reducerea) la “e”).
εwt şi εut sunt inovaţii privind o informaţiile disponibile în perioada
t-1, notată Іt-1. Ecuaţia (1) reprezintă ideea de bază că profi tabilitatea (în
sectorul e) este un factor care explică modifi carea ratei şomajului.
Zut reprezintă alte variabile (şi termeni determinişti) care, în
condiţiile în care ceilalţi factori nu se modifi că, va conduce la scăderea ratei
şomajului. Factorul zut va include în mod tipic o măsurare a ratei de creştere a
economiei, şi alţi factori legaţi de oferta de forţă de muncă.
Pentru stabilirea ratei principale a şomajului de echilibru, ecuaţia (1)
se rescrie sub forma:
(2)
unde
este rata şomajului care nu infl uenţează creşterea salariilor.
Folosind medii necondiţionate, notate prin E, pe ambele părţi ale (3) obţinem:
(3)
Folosind ipoteza unei ponderi staţionare a salariilor, partea stângă
este zero. Astfel, folosind ga şi gf ca notaţii ale ratelor de creştere constantă a
productivităţii şi preţurilor externe, obţinem:
(4)
ca soluţie pentru rata şomajului de echilibru, notată uphil. Media pe
termen lung a ponderii salariilor este în consecinţă:
(5)
Romanian Statistical Review - Supplement nr. 4 / 201988
În plus, uphil şi wshphil reprezintă starea stabilă şi unică a perechii corespunzătoare de ecuaţii de diferenţe deterministe. Forma cunoscută a curbei Phillips e ilustrată în Figura 1. Se presupune că în economie se manifestă iniţial un nivel scăzut al şomajului, adică uo din fi gură. Curba Phillips pe termen scurt determină rata creşterii salariilor ∆w0. Ponderea salariilor conform ecuaţiei (1) este deasupra valorii de echilibru pe termen lung, implicând că şomajul începe să crească şi creşterea salariilor se diminuează de-a lungul curbei Phillips. Forma abruptă a curbei Phillips se defi neşte pentru cazul ∆wt = ∆qt + ∆at. Panta curbei este dată de –βw1/(1- βw3), fi ind denumită drept curba Phillips în literatură. Echilibrul stabil se atinge când creşterea salariilor este egală cu creşterea constantă în stare de echilibru , adică, gf+ga iar nivelul şomajului este dat de uphil.
Dinamica şi echilibrul curbei Phillips într-o economie deschisă
Figura 1
Problema pantei curbei Phillips pe termen lung este văzută ca depinzând de coefi cientul βw3, elasticitatea creşterii salariilor determinată fără a se ţine seama de preţul produselor. În fi gură, curba pe termen lung are o tendinţă descrescătoare, respectiv βw3 < 1, ceea ce în mod convenţional face referire la o neomogenitate dinamică în stabilirea salariilor. Contrariul, omogenitatea dinamică, implică βw3 = 1 şi o curbă Phillips verticală. Cu condiţia omogenităţii dinamice, rata de echilibru uphil este independentă de infl aţia mondială gf. Panta curbei Phillips pe termen lung a reprezentat una din cele mai dezbătute probleme în macroeconomie în anii 1970 şi 1980. Un argument în
Revista Română de Statistică - Supliment nr. 4 / 2019 89
favoarea curbei Phillips verticale pe termen lung este acela că s-a observat evident că lucrătorii sunt capabili să obţină compensare deplină pentru infl aţie. Rezultă ca βw3 = 1 este o restricţie normală asupra curbei Phillips, cel puţin dacă ∆qt e interpretat ca o variabilă a probabilităţilor. Panta descendentă a curbei Phillips pe termen lung a fost contestată pe motiv că oferă un tablou prea optimist pentru politica economică: şi anume că guvernul poate reduce permanent nivelul şomajului sub rata naturală prin “fi xarea” unui nivel ridicat al infl aţiei. În cadrul unei economii deschise, această discuţie apare a fi oarecum exagerată deoarece compromisul pe termen lung dintre infl aţie şi şomaj nu apare din premisa unei curbe pe termen lung cu pantă descendentă. În schimb, conform Figurii 1, nivelul stabil al şomajului e determinat de rata infl aţiei importate gf şi de creşterea productivităţii exogene ga. Aceşti doi indicatori nu sunt consideraţi mijloace (sau ţinte intermediare) de politică economică. În economia reală, connsideraţiile privind costul vieţii joacă un rol semnifi cativ în stabilirea salariilor. Astfel, în cercetarea econometrică aplicată se include de obicei infl aţia actuală şi întârziată, refl ectând accentul pus pe consideraţii privind costul vieţii în cadrul negocierilor salariale. Pentru reprezentarea acestei posibilităţi, se consideră următorul sistem:
(6)
(7)
(8) Pentru a se realiza o distincţie formală se utilizează un semn distinctiv deasupra celorlalţi coefi cienţi (şi deasupra termenului de abatere). A doua ecuaţie este identică cu ecuaţia şomajului (1). Ultima ecuaţie, a preţurilor stocastice, combină defi niţia preţurilor de consum cu ipoteza identică a stabilităţii ponderii salariilor în sectorul acoperit şi creşterea salariilor în sectorul expus. Folosind (8) pentru a elimina Δpt în (6), ne întoarcem la ecuația cu coefi cienți, cu coefi cienţi şi εwt redefi nite corespunzător. Este utilă exprimarea uphil în termeni de coefi cienţi ai unui sistem extins (6)-(8):
(9) deoarece sunt două restricţii de omogenitate necesare pentru curba Phillips pe termen lung, şi anume: =1 şi . Sistemul curbei Phillips pentru salarii într-o economie deschisă reprezintă o specifi caţie completă a dinamicii modelului infl aţiei. În mod
Romanian Statistical Review - Supplement nr. 4 / 201990
clar, proprietăţile dinamice ale modelului se aplică altor versiuni ale curbei Phillips. În special, toate sistemele tip curbă Phillips presupun că rata naturală a şomajului (NAIRU) este o soluţie stabilă. Ca o singură ecuaţie, ecuaţia
curbei Phillips este instabilă pentru o rată dată a şomajului. Stabilitatea în dinamică a ponderii salariilor şi ratei şomajului depinde de mecanismul de echilibrare integrat în ecuaţia pentru rata şomajului. În acest sens, o defi niţie a formării salariilor bazată pe curba Phillips nu poate fi adaptată unei politici economice care ţinteşte nivelul (rata) şomajului întrucât numai rata naturală a şomajului corespunde unei ponderi salariale stabile. Orice alt nivel (ţintit) duce la creştere sau descreştere continuă a ponderii salariale. Stabilitatea în dinamică a ratei naturale este un subiect de interes şi nu poate fi adresată într-un sistem incomplet tip curbă Phillips, adică, prin estimarea unui model de curbă Phillips cu o singură ecuaţie. Însă există abordări practice prin care estimarea ratei naturale a şomajului de bazează pe astfel de sisteme incomplete. Staiger şi alţii (1997) este un exemplu de studiu important care urmează tradiţia estimării numai a curbei Phillips, considerând mecanismul de echilibrare implicit (1). Pentru alte ţări, în special europene, unde instabilitatea ratei şomajului se manifestă mai puternic, problema despre corespondenţa dintre ratele estimate şi stabilitate este o problemă de actualitate.
Concluzii
În urma studiului efectuat, printr-o amplă analiză a lucrărilor unor mari cercetători în domeniul modelării și analizelor mactroeconomice s-au obținut elemente relevante care pot fi utilizate în modele actuale de analiză. Din studiul curbei Phillips rezultă aspecte cum ar fi aceea a dinamicii și echilibrului curbei Phillips într-o economie deschisă guvernată de legile pieței libere. Stabilirea în dinamică a ratei naturale este o temă de interes și deaceea trebuie estimată printr-un model Phillips cu o singură ecuație. Deși analizele privind rata naturală a infl ației se bazează pe sisteme incomplete, totuși se poate identifi ca trendul evolutiv al ratei șomajului. Se desprinde concluzia că în economia reală, modele rezultate din combinarea subsistemelor conduc la obținerea de parametrii care pot fi utilizați în analizele macroeconomice.
Bibliografi e
1. Anghelache, C., Anghel, M.G. et al (2017). Using the Philips Curve in
Macroeconomic Analysis, Romanian Statistical Review, Supplement, no. 7, pp. 3-15 / 16-28,
2. Anghelache, C., Anghel, M.G. (2014). Modelare economică. Concepte, teorie şi
studii de caz, Editura Economică, Bucureşti 3. Anghelache, C. (2012). Elemente de modelare economică, Editura Artifex,
Bucureşti
Revista Română de Statistică - Supliment nr. 4 / 2019 91
4. Bjerkholt, O. (2005). Markets, models and planning: the Norwegian experience, Oslo University, Department of Economics in series Memorandum
5. Dickens, W.T. (2008). A New Method to Estimate Time Variation in the NAIRU, Federal Reserve Bank of Boston in Conference Series ; [Proceedings]
6. Ewing, B.T., Seyfried, W.L. (2003). Modeling the Philips Curve: A Time-Varying Volatility Approach. Applied Econometrics and International Development, vol 2-3, 7-24
7. Johansen, S. (2002). A small sample correction for tests of hypotheses on the cointegrating vectors. Journal of Econometrics, Elsevier, 111(2), 195-221
8. Karanassou, M., Snower, D. (2007). Infl ation Persistence and the Philips Curve Revisited, Kiel Institute for the World Economy in Kiel Working Papers series
9. Lee, A.Y., Aaker, J.L. (2006). A Monte Carlo Study of Growth Regressions, Stanford University, Graduate School of Business in Research Papers series
10. Levy, D. (2004). Cointegration in Frequency Domain, EconWPA in its series Econometric Haldane, A., Quah, D. (2000) – “UK Philips Curves and Monetary Policy”, Centre for Economic Performance, London School of Economics in CEP Discussion Papers
11. Mandel, M., Tomšík, V. (2003). The Consumption Function and Ricardian Equivalence in a Small Open Economy, University of Economics, Prague in Politická ekonomie.
Romanian Statistical Review - Supplement nr. 4 / 201992
ESSENTIAL ELEMENTS ON MODELING AND COMBINATION OF SUBSYSTEMS
Prof. Radu Titus MARINESCU PhD ([email protected])
„Artifex”University of BucharestProf. Constantin ANGHELACHE PhD ([email protected])
Bucharest University of Economic Studies / „Artifex” University of BucharestAssoc. prof. Mădălina-Gabriela ANGHEL PhD ([email protected])
„Artifex” University of BucharestAlexandru BADIU PhD Student ([email protected])
Bucharest University of Economic Studies
Abstract
In practical work, it is a problem for the built models to be based on the combination of subsystems that are in line with the structure of the phenomenon under analysis. In the development of large-scale macroeconomic models, the traditional approach is to express through an equation the possibility of determining the outcome. By adopting maximum probability methods with limited information, the parameters of an equation can be obtained, while the parameters of other equations remain restricted. After building the relevant submodels by marginalization and conditioning, reduced models of subsectors can be aggregated into a broad macroeconomic model. Macroeconomic models are used as the basis for establishing economic strategies, so it is important to model the coeffi cients of all explanatory variables. Therefore,
obtaining long-term properties of subsystems is of particular importance.
Thus, we can speak of analysis based on the infl ation model and the Phillips
curve. Following the discovery by Phillips of an empirical regularity between
unemployment and wage growth, this was a novelty taken over by all of the
most prominent researchers in the period ahead.
Keywords: subsystem, Phillips curve, Lucas critique, coeffi cient,
equation
JEL Classifi cation: C44, C61, E31
Introduction In this article, based on the authors’ study, it is intended to identify the essential elements used in the modeling and combining of the subsystems. Methods used by various researchers that led to the development of macroeconomic models based on which macroeconomic evolution strategies can be established are presented. Next, a comprehensive analysis of the infl ation
Revista Română de Statistică - Supliment nr. 4 / 2019 93
model and the Phillips curve is being used. Also, the aspects that depend on the Phillips curve application are highlighted. The relationship between wage growth and economic activity has been analyzed at macroeconomic level since the neo-classical period. An empirical version of the Phillips curve, called triangular infl ation model, has also been used in current conditions, of course, with adjustments being made. The Phillips Salary Curve system in an open economy is a full specifi cation of the infl ation dynamics model.
Literature review
Anghelache, Anghel et al (2017) highlighted a series of elements regarding the use of the Philips curve in macroeconomic analyzes. Anghelache and Anghel (2014) and Anghelache (2012) analyzed the main aspects of economic modeling. Bjerkholt (2005) studied concepts and planning. Dickens (2008) examined methods of estimating the time variation. Ewing and Seyfried (2003) examined concepts of modeling the Phillips curve. Johansen (2002) studied hypothesis tests on cointegration vectors. Karanassou and Snower (2007) addressed aspects of the Philips Curve Review. Lee and Aaker (2006) presented a Monte Carlo study of growth regressions. Levy (2004) studied elements of frequency cointegration. Mandel and Tomšík (2003) presented the main aspects of the consumer function.
Research methodology, data, results and discussions
• In the development of macroeconomic models on a large scale, the traditional approach was to estimate an equation (or submodel) once and to establish the results simultaneously. Often, this was done without checking the suitability of the procedures. The approach could be justifi ed in terms of estimates. By adopting the maximum probability methods with limited information, the parameters of an equation can be estimated, while the parameters of other equations remain unrestricted, as shown by Anderson and Rubin (1949) and Koopmans and Hood (1953). It was argued that methods with limited information were stronger than the erroneous equations specifi ed in the system, in situations where there were better theories or reliable information about a subset of variables. Limited information methods have been adopted for practical reasons to avoid the complexity of calculations specifi c to full-information methods - such as full-scale probability based on complete information. After the construction of the relevant submodels by marginalization and conditioning, small models of subsectors are aggregated into a broad macroeconomic model. A general theory for the three steps will invariably contain criteria and
Romanian Statistical Review - Supplement nr. 4 / 201994
conditions that are formulated for the entire system, which raises the question if there are ways to solve when the complete model is too complex for a simultaneous modeling. One solution would be to use models with a high degree of aggregation that are small enough to be analyzed as a complete system. In such an approach, several economic mechanisms, important and relevant to the description of the economy are omitted. The general approach can be seen as a gradualism - following the establishment of the structure in the submodels. The fi rst two conditions in the defi nition of the partial structure do not require full knowledge of the model. Excluding important explanatory variables from the model leads to the invalidity of the submodels, which is detected in the submodel when the correlation between the included variables and those that were omitted is changed. For the latter, it is possible, at least in principle, to develop a model as the last extension of the information, thus establishing that the structure or partial structure is a way to solve Johansen’s problem. Basically, it is known that the complete model is inaccessible. However, conditional consumption is constant when the survey interval is extended by nine years of quarterly observations; remains unaltered over the period of fi nancial deregulation and supports the simultaneous modeling of private consumption, household disposable income, population wealth, and real estate prices. Thus, the inductive proof can be validated that the conditional consumption function will represent a partial structure. The simultaneous model is hardly an ideal substitute for a better model for “bid” eff ects that operate through the labor
market, yet it provides protection against alternative views. For example,
income is in fact the correction of balance, not consumption.
Macroeconomic models are used as the basis for economic policy
making. From this point of view, it is important to model the coeffi cients of
all relevant explanatory variables by conditioning all knowledge (relevant
and applicable) on institutional conditions in the studied society. We rely on
higher aggregation specifi cations where gross coeffi cients create combined
eff ects of the included variables, and correlated omitted variables can lead to
wrong policy recommendations. Such interference is more damaging to the
governors than the simultaneity interference that can occur by combining the
submodels. Whether it remains valid or not an interesting issue that can be
explored through Monte Carlo simulations on particular model specs.
It follows that obtaining the long-term properties of the submodels
is of special importance. Once a cointegration equation is determined, it is
invariant to extensions of the set of information. On the other hand, this is a
Revista Română de Statistică - Supliment nr. 4 / 2019 95
property that needs to be determined in each case. The concepts of separation in cointegrated systems are described by Granger and Haldrup (1997). The idea is to break down each variable into a persistent component and a transient component. In a vector equilibrium correction model, two subsystems are considered, where the variables of a subsystem do not include cointegrating equations of the other subsystem (cointegration separation). However, short-term eff ects of variables in a subsystem on the other variables may occur and
co-integrative equations of a system may aff ect the short-term development
of the other’s variables. The absence of both types of interaction is called
complete separation, whereas if only one of them is present, it becomes partial
separation. These concepts are closely related to the exogenity of the variables
in a subsystem on the parameters of the other. Both submodels, partially or
completely separated, are verifi able assumptions that need to be tested as part of cointegration analysis. Hecq et al. (2002) expand the results of Granger and Haldrup (1997). The conclusion of Hecq et al. (2002) is, however, that verifi cation of separation requires the whole system to be known, which corresponds to Johansen’s observation. • The infl ation model and the Phillips curve originated in the same era of macroeconomics. But while Aukrust’s model departed from academia, literature on the Phillips curve developed in the 1960s and made a tremendous impact over the next four decades. Below are some of the signifi cant steps in Phillips curve developments. In the 1970s, the Phillips curve and the Aukrust model were considered as alternatives, representing the “demand” and “supply” infl ation model (Frisch 1977). However, as Aukrust (1977) points out, the diff erence between considering the labor market as an important source of
infl ation and paying particular attention to the product market Phillips curve is more a matter of principle than the principle, the two mechanisms can work together. Next, it shows how the two approaches formally can be combined, giving the Phillips curve the short-term relationship of nominal wage growth, while the main thesis is valid in the long run. Within the chapter are addressed key issues for the application of the Phillips curve in the current context, its representation in a system of cointegrating variables; consistency and changes in the unemployment rate; the uncertainty of the Phillips curve estimated NAIRU (Non-Accelerating Infl ation Rate of Unemployment); and the inversed Phillips curve status, respectively Lucas’s supply curve. • Following the discovery by Phillips of an empirical regularity between unemployment and wage growth in the UK, the Phillips curve was integrated into macroeconomics through a series of works in the 1960s. Samuelson and Solow (1960) interpreted it as a compromise faced by the
Romanian Statistical Review - Supplement nr. 4 / 201996
governors, and Lipsey (1960) was the fi rst to estimate the Phillips curves with multivariate regression techniques. Lipsey interpreted the relationship from the perspective of classical price dynamics with the unemployment rate acting as an intermediary between excess demand and labor market friction. Importantly, Lipsey has included rising consumer prices as an explanatory variable in its regressions and so formulated what became known as the Phillips curve of probabilities. Subsequent developments include the distinction between the short-term Phillips curve where infl ation deviates from projected infl ation and the long-term Phillips curve where infl ation forecasts are met. Finally, the concept of a natural rate of unemployment was defi ned as the stable unemployment rate, corresponding to a long-term vertical curve (Phelps 1968 and Friedman 1968). The relationship between wage growth and economic activity also appears in neo-classical macroeconomics (Lucas and Rapping, 1969 and 1970 and Lucas 1972). However, in the neo-classical economy, the causal relationship of Phillips’s original model was overthrown: if there is a correlation between infl ation and unemployment in absolute terms, causality goes from infl ation to activity and unemployment. The inversion of Lucas and Rapping is based on the appreciation that the price level is anchored in the relationship between quantitative theory and the autonomous stock of money. The rise in prices and wages is determined outside the Phillips curve, so the correct wording is that on the one hand there is the unemployment rate and wage increases (and / or infl ation) on the other. Lucas (1972) launches another known approach, based on rational probabilities about relatively uncertain product prices. If the probabilities are met (on average), the aggregate off er is unchanged over the last period.
However, if there are surprise prices, there is a distance from the average
level of long-term production. This establishes a “surprise only” bidding
relationship.
Lucas’s bidding function is the opposite of the long-term Phillips
vertical curve, augmented by Lipsay’s predictions, but derived from
microeconomic theory and the rational probability hypothesis.
In addition, for the conventional aggregate demand specifi cation (Romer 1996), the model involves a positive association between output and infl ation or a negative relationship between unemployment and infl ation. Thus, there is a neo-classic correspondence for the short-term Phillips curve. However, Lucas’s demand curve, when applied to data and is estimated by the least squares (OLS) method, is not a causal relationship that can be exploited by the governors. On the contrary, it will change when, for example, money supply has increased to stimulate production in a way that leaves politics
Revista Română de Statistică - Supliment nr. 4 / 2019 97
without the possibility of infl uencing real output or unemployment. This concept is called Lucas’s critique (Lucas 1976), which was formulated as a critique of the infl ation-unemployment compromise of the Phillips curve, which was dealt with in the academic literature as well as in the 1970’s macroeconomic models (Wallis 1995). The force of criticism, however, comes from its generality: it is a potential hazard for all conditioned econometric models. The issue of causality is manifested in connection with the latest versions of the Phillips Curve - the Neo-Keynesian version. In the US, an empirical version of the Phillips curve, called the “Triangular Infl ation Model,” spread despite Lucas’s critique (contributions to this eff ect are found in Gordon 1983 and 1997, Staiger et al., 2001). As we
will continue to argue, an explanation of the validity of the American Phillips
curve is that unemployment rate shocks were generally of less magnitude than
in European countries.
As we have shown before, there are many ways in which a Phillips curve
for an open economy can be deduced from economic theory. Our appreciation of
the Phillips curve is based on Calmfors (1977), which reconciled the Phillips curve
with the Scandinavian infl ation model. We intend to take a step forward, however, and incorporate the Phillips curve into a framework that takes into account wage and price data series. Reconstructing the model in terms of cointegration and causality reveals that the Phillips curve version of the master model imposes a mechanism to correct the balance on the system. Thus, while consistent with Aukrust’s main theory, the Phillips curve is also a special model because it includes only one of the wage setting mechanisms discussed by Aukrust. Without departing from generality, we will focus on salaries in the Phillips curve and recall that, in accordance with Aukrust’s theory, it is assumed that: 1. , possibly after removal of deterministic changes; and 2. the causal structure is „a path” represented by H4mc and H5mc. Consistency with assumed cointegration and causality requires a balance correction model for the nominal wage rate in the exposed sector. Assuming a fi rst order dynamic for simplifi cation, a Phillips curve system is defi ned by the following two equations:
•
(1) •
Romanian Statistical Review - Supplement nr. 4 / 201998
where (notation is simplifi ed by renunciation to „e”). εwt and εut are innovations regarding information available during t-1,
marked Іt-1. Equation (1) is the basic idea that profi tability (in sector e) is a
factor explaining the change in the unemployment rate.
Zut represents other variables (and deterministic terms) which, if
the other factors do not change, will lead to the decrease of the unemployment
rate. The factor will typically include a measure of the rate of growth of the
economy, and other factors linked to the supply of labor.
To determine the main rate of equilibrium unemployment, equation
(1) is rewritten as a form:
(2)
where
is the unemployment rate that does not aff ect wage growth.
Using unconditional environments, denoted by E, we get on both sides of (3):
(3)
Using the assumption of a fi xed wage weight, the left is zero. Thus,
using ga and gf as notations of steady growth rates of productivity and external
prices, we obtain:
(4)
as a solution for the equilibrium unemployment rate, noted uphil. The
long-term average of the salary weight is accordingly:
(5)
In addition, uphil and wshphil represent the stable and unique state of
the corresponding pair of deterministic diff erence equations.
The known shape of the Phillips curve is illustrated in Figure 1. It is
assumed that the economy initially exhibits a low level of unemployment, ie
uo in the fi gure. The short-term Phillips curve determines the wage growth
rate ∆w0. The share of wages according to equation (1) is above the long-term
equilibrium value, implying that unemployment is rising and wage growth
decreases along the Phillips curve. The steep shape of the Phillips curve is
defi ned for the case ∆wt = ∆qt + ∆at.
The slope of the curve is given by –βw1/(1- βw3), being referred to
as the Phillips curve in the literature. The stable balance is reached when the
salary increase equals constant steady growth, ie, gf+ga and the unemployment
level is given by the uphil.
Revista Română de Statistică - Supliment nr. 4 / 2019 99
Dynamics and balance of the Phillips curve in an open economy
Figure 1
The problem of the long-term Phillips curve is seen as depending on the βw3 coeffi cient, the elasticity of wage growth determined without regard
to the price of the products. In the fi gure, the long-term curve has a decreasing tendency, ie βw3 <1, which conventionally refers to a dynamic inhomogeneity
in wage setting. Conversely, dynamic homogeneity implies βw3 = 1 and a
vertical Phillips curve. Given the dynamic homogeneity, the uphil equilibrium
rate is independent of global infl ation gf. The long-term slope of the Phillips curve was one of the most debated problems in macroeconomics in the 1970s and 1980s. One argument in favor of the long-term vertical curve Phillips is that it has clearly been noticed that workers are able to obtain full compensation for infl ation. It follows that βw3
= 1 is a normal restriction on the Phillips curve, at least if Δqt is interpreted as a variable of probabilities. The downward slope of the Phillips curve in the long run has been challenged on the grounds that it off ers a picture too optimistic about economic policy: that the government can permanently reduce unemployment levels below the natural rate by „fi xing” a high level of infl ation. In an open economy, this discussion appears to be somewhat exaggerated because the long-term compromise between infl ation and unemployment does not come from the premise of a long-term downward curve. In contrast, according to Figure 1, the stable level of unemployment is determined by the imported infl ation rate gf and the increase in exogenous productivity ga. These two indicators are not considered as means (or intermediate targets) of economic policy.
Romanian Statistical Review - Supplement nr. 4 / 2019100
In the real economy, cost-of-living constraints play a signifi cant role in wage setting. Thus, applied econometric research usually includes current and delayed infl ation, refl ecting the emphasis on cost-of-living considerations in wage negotiations. To represent this possibility, the following system is considered: (6)
(7)
(8) In order to achieve a formal distinction, a distinctive sign is used above the other coeffi cients (and above the term deviation). The second
equation is identical to the unemployment equation (1). The last equation of
stochastic prices combines the defi nition of consumer prices with the identical assumption of the stability of the wage weight in the covered sector and the increase in wages in the exposed sector. Using (8) to remove Δpt in (6), we return to the equation with coeffi cients, with coeffi cients and εwt redefi ned accordingly. It is useful to
express uphil in terms of coeffi cients of an extended system (6) - (8):
(9)
because there are two homogeneous restrictions required for the long-
term Phillips curve, namely: =1 and .
The Phillips salary system in an open economy is a complete
specifi cation of the dynamics of the infl ation model. Clearly, the dynamic
properties of the model apply to other versions of the Phillips curve. In particular,
all Phillips curve systems assume that the natural rate of unemployment
(NAIRU) is a stable solution. As a single equation, the equation of the Phillips
curve is unstable for a given rate of unemployment. The dynamic stability
of wage and unemployment rates depends on the integrated equalization
mechanism in the unemployment rate equation. In this respect, a defi nition of
wage formation based on the Phillips curve can not be adapted to an economic
policy that targets the unemployment rate, since only the natural rate of
unemployment corresponds to a stable wage weight. Any other level (targeted)
leads to a continuous increase or decrease of the salary weight.
Dynamic stability of the natural rate is a subject of interest and can
not be addressed in an incomplete Phillips curve system, that is, by estimating
a Phillips curve model with a single equation. But there are practical
approaches where estimating the natural rate of unemployment is based on
such incomplete systems. Staiger et al. (1997) is an important study example
following the tradition of estimating only the Phillips curve, considering the
implicit balancing mechanism (1). For other countries, especially European
Revista Română de Statistică - Supliment nr. 4 / 2019 101
ones, where the instability of the unemployment rate is stronger, the problem of the correspondence between estimated rates and stability is a topical issue.
Conclusion
Following the study, a wide-ranging analysis of the work of large researchers in the fi eld of modeling and macroeconomic analyzes has produced relevant elements that can be used in current analysis models. From the Phillips curve, aspects such as Phillips curve dynamics and balance are emerging in an open economy governed by free market laws. Dynamic determination of the natural rate is a topic of interest and must therefore be estimated by a Phillips model with a single equation. Although the analysis of the natural rate of infl ation is based on incomplete systems, it is still possible to identify the evolutionary trend of the unemployment rate. It is concluded that in the real economy models resulting from the cohabitation of the subsystems lead to obtaining parameters that can be used in macroeconomic analyzes.
References
1. Anghelache, C., Anghel, M.G. et al (2017). Using the Philips Curve in Macroeconomic Analysis, Romanian Statistical Review, Supplement, no. 7, pp. 3-15 / 16-28,
2. Anghelache, C., Anghel, M.G. (2014). Modelare economică. Concepte, teorie şi
studii de caz, Editura Economică, Bucureşti 3. Anghelache, C. (2012). Elemente de modelare economică, Editura Artifex,
Bucureşti 4. Bjerkholt, O. (2005). Markets, models and planning: the Norwegian experience,
Oslo University, Department of Economics in series Memorandum 5. Dickens, W.T. (2008). A New Method to Estimate Time Variation in the NAIRU,
Federal Reserve Bank of Boston in Conference Series ; [Proceedings] 6. Ewing, B.T., Seyfried, W.L. (2003). Modeling the Philips Curve: A Time-Varying
Volatility Approach. Applied Econometrics and International Development, vol 2-3, 7-24
7. Johansen, S. (2002). A small sample correction for tests of hypotheses on the cointegrating vectors. Journal of Econometrics, Elsevier, 111(2), 195-221
8. Karanassou, M., Snower, D. (2007). Infl ation Persistence and the Philips Curve Revisited, Kiel Institute for the World Economy in Kiel Working Papers series
9. Lee, A.Y., Aaker, J.L. (2006). A Monte Carlo Study of Growth Regressions, Stanford University, Graduate School of Business in Research Papers series
10. Levy, D. (2004). Cointegration in Frequency Domain, EconWPA in its series Econometric Haldane, A., Quah, D. (2000) – “UK Philips Curves and Monetary Policy”, Centre for Economic Performance, London School of Economics in CEP Discussion Papers
11. Mandel, M., Tomšík, V. (2003). The Consumption Function and Ricardian
Equivalence in a Small Open Economy, University of Economics, Prague in Politická ekonomie.
Romanian Statistical Review - Supplement nr. 4 / 2019102
Utilizarea indicilor Laspeyres şi Paasche în determinarea formulei indicelui de preţ
Drd. Doina BUREA ([email protected])
Academia de Studii Economice din București
Abstract
În contextul teoriei microeconomice, nevoia de identifi care a unui model economic care să refl ecte comportamentul producătorului
sau consumatorului, precum şi problema indicilor numerici se rezumă la
întrebarea: care este „cea mai bună” metodă pentru a construi setul de
agregate ale acestui model? Indicii numerici pot fi priviţi ca o descompunere
a valorii unui set bine defi nit de tranzacţii dintr-o perioadă de timp într-o
valoare agregată a preţurilor şi o valoare agregată a cantităţilor. Această
abordare poate fi realizată şi la nivelul indicilor, astfel că indicele valoric
poate fi reprezentat ca fi ind produsul unei componente care măsoară evoluţia
de ansamblu a preţurilor din cele două perioade (indicele de preţ) şi o valoare
ce măsoară evoluţia volumului fi zic pentru aceleaşi perioade (indicele
cantitativ).
Cuvinte cheie: indice de preţ, tranzacţii, infl ație, indice armonizat al
prețurilor, indicele Laspeyres, indiciele Paasche.
Clasifi carea JEL: E31, E60
Introducere
Indicele cu bază fi xă este considerat cel mai simplu indice de preţ, obţinut ca raport dintre două valori calculate pentm acelaşi set de cantităţi, dar folosind seturi de preţuri diferite pentru cele două perioade comparate 0,
perioadă de bază, şi 1, perioadă curentă. Selectarea opţiunilor privind setul constant de cantităţi se refl ectă fi e către utilizarea setului cantităţi tranzacţionate în perioada 0, sau cele din perioada 1.
Având în vedere cele prezentate, voi detalia indicii de tip Laspeyres (187l) şi indicii de tip Paasche (1874). În acest sens, Principiul Laspeyres constă în utilizarea ponderilor constante ale unei perioade considerate de bază, iar principiul Paasche presupune utilizarea ponderilor variabile referitoare la perioadă curentă. În cazul celor două principii problema este ca deşi sunt considerate acceptabile aplicarea lor duce la obţinerea de rezultate diferite, mai ales atunci când ponderile aplicate suferă modifi cări rapide şi considerabile.
Revista Română de Statistică - Supliment nr. 4 / 2019 103
Literature review
Anghel, Mirea și Badiu (2018) au studiat elemente fundamentale ale indicilor prețurilor utilizați în măsurarea infl ației. Anghel (2015), precum și
Anghelache, Niţă și Badiu, A. (2016) au realizat studii cu privire la evoluția
indicelui prețurilor în România. Anghelache și Sacală (2015) au prezentat
o serie de noțiuni de bază ale infl ației. Anghelache, Voineagu și Gheorghe
(2013) au analizat cei doi indicatori Laspeyres şi Paasche, care au stat la baza
calculului indicilor preţurilor de consum (IPC) din toată lumea. Armantier și
colaboratorii (2015) au abordat o serie de aspecte privind previziunile asupra
infl ației. Ascari și Ropele (2009) au studiat tendințele infl ației, o temă similară
fi ind analizată și de Cogley și Sbordone (2008). Hornstein și Wolman (2005)
au cercetat corelația dintre tendințele infl ației și capitalul specifi c fi rmei. Kim
și Henderson (2005) au abordat aspecte referitoare la țintirea infl ației și a
creșterii veniturilor nominale.
Metodologia cercetării, date, rezultate şi discuţii
Cea mai simplă metodă pentru deteminarea formulei indicelui de preţ
a fost descrisă detaliat de către Lowe (1823). Metoda sa pentru măsurarea
modifi cării preţului între momentele 0 şi 1 a constat în defi nirea unui set
reprezentativ de produse al vectorului cantitativ q =[q1,......, qn] apoi în
estimarea agregatului valoric, folosind preţurile din perioada 1 şi perioada 0:
å=
n
iii qp
1
1, reprezintă costul setului reprezentativ de produse din perioada 1,
å=
n
iii qp
1
0 costul din perioada 0 a coşului. (1)
Metoda cu bază fi xă pentru determinarea indicelui de preţ se referă
la alegerea vectorului q pentru setul reprezentativ de produse. De asemenea,
economiştii şi statisticienii de preţuri au căutat să selecteze vectorul q cu ceva
mai multă precizie în timp. Opţiunile pentru stabilirea setului de referintă sunt:
- vectorul mărfurilor din perioada de bază q0 sau
- vectorul mărfurilor din perioada curentă q1. Aceste două opţiuni
conduc la indicele de preţ Laspeyres (1871) PL defi nit prin ecuaţia (2) şi la
indicele de preţ Paasche (1874) Pp defi nit prin ecuaţia (3).
å
å
=
==n
iii
n
iii
L
qp
qpqqppP
1
00
1
01
1010 ),,,(* (2)
Romanian Statistical Review - Supplement nr. 4 / 2019104
å
å
=
==n
iii
n
iii
P
qp
qpqqppP
1
10
1
11
1010 ),,,(*( (3)
Formulele (2) şi (3) prin rescrierea alternativă sunt mai utile pentru ofi ciile statistice. Ponderea cheltuielilor pentru marfa i din perioada t se defi neşte astfel:
å=
=n
j
tj
tj
ti
ti
ti qpqps
1
/ (4)
unde i=1,….., n şi t = 0, 1
Rescrierea indicelui Laspeyres (1) se poate realiza astfel:
001
1
00000
1
1
1
00
1
011010 )/(/)//),,,(* iii
n
jiiiii
n
ii
n
iii
n
iiiL sppqpqpppqpqpqqppP ååååå ===
====
(5)
De asemenea, indicele de preţ Laspeyres PL poate fi scris ca şi medie
aritmetică a celor n rapoarte de preţ, 01 / ii pp , ponderate cu cotele de cheltuieli
din perioada de bază. De precizat că această formulă de tip Laspeyres a servit
ca bază de calcul pentru indicii preţurilor de consum (IPC) din toată lumea.
Pentru aplicarea acesteia un ofi ciu statistic trebuie să colecteze informaţii cu
privire la cotele de cheltuieli 0
ns pentru domeniul de defi niţie al indicelui din
perioada de bază, ulterior să colecteze lunar numai informaţii despre preţurile
produselor. Astfel, Indicele Preţului de Consum (IPC-ul) de tip Laspeyres se
poate calcula fără a fi necesare informaţiile din perioada curentă. Acest indice
Paasche mai poate fi scris şi sub formă de raport între cotele de cheltuieli şi
preţ astfel:
1
110
1
1
1
1101
1
11101
1
0
1
11
1
10
1010
)/(
)/(
1
/)/(
1
/
1),,,(*
-
-
=
=
-
====
þýü
îíì
=
þýü
îíì
=
þýü
îíì
=
þýü
îíì
=
å
ååååå
ii
n
ii
n
iiii
n
iiiiii
n
ii
n
iii
n
iii
P
spp
sppqpqpppqpqp
qqppP (6)
Indicele de preţ Paasche PP mai poate fi scris şi ca medie armonică a
celor n rapoarte de preţ, 01 / ii pp , ponderate cu cotele de cheltuieli din perioada
1. Din cauza lipsei de informaţii privind cantităţile din perioada curentă ofi ciile
statistice sunt restricţionate să producă indici de tip Paasche în mod regulat.
Dacă înlocuim în ecuaţia ),,,(/)/(),,,( 1010011010 qqppPVVqqppQ = P
cu PL defi nit de ecuaţia (2), atunci se va obţine următorul indice cantitativ:
Revista Română de Statistică - Supliment nr. 4 / 2019 105
(PQ
å
å
=
==n
iii
n
iii
qp
qpqqpp
1
01
1
11
1010 ),,,( (7)
Dacă P este înlocuit cu Pp , atunci se obţine următorul indice cantitativ:
å
å
=
==n
iii
n
iii
L
qp
qpqqppQ
1
00
1
10
1010 ),,,( (8)
Formulele indicilor Laspeyres şi Paasche prezintă o sincopă, chiar dacă aceştia sunt plauzibili, vor da în general răspunsuri diferite. Ofi ciul de statistică pentru atingerea mai multor scopuri trebuie să ofere răspunsuri la întrebarea: Care este „cea mai bună” măsură sintetică a modifi cării preţului valorii agregate pe parcursul celor perioade avute în vedere? sau care este relaţia „normală” dintre indicii Paasche şi Laspeyres? Relaţia Bortkiewicz poate realiza comparaţia indicilor Laspeyres şi Paasche. Calculul indicelui Laspeyres este simplu având în vedere că estimarea numitorului este necesară o singură dată, dar aplicarea acestei fromule nu ia în considerare scăderea cererii sau modifi carea producţiei ceea ce poate duce la o supraestimare a creşterii preţurilor. De cealaltă parte, indicele Paasche folosind ponderile perioadei curente tinde să subestimeze creşterea preţurilor. În practică se poate observa ca preţurile şi cantitătile nu se modifi că cu acelaşi procent şi de aceea relaţia dintre două sisteme depinde de corelaţia dintre modifi carea preţurilor şi a cantitătilor, care normal ar trebui sa fi e negativă. Gradul de divergenţă dintre indicii Laspeyres şi Paasche, indiferent de direcţia arătată, depinde de intesitatea corelaţiei dar si de dispersia preţurilor şi cantităţilor relative, masurată prin coefi ecntul de variaţie. În problema clasică a puterii de cumpărare a banilor, de exemplu, nivelul indicelui de preţ Laspeyres sau Paasche este stabilit în primul rând de factori monetari în timp ce divergenţa dintre cele două formule depinde mai mult de infl uenţele nonmonetare care lucrează la dispersia preţurilor relative în jurul „standardului”. Principiul Laspeyres constă în utilizarea ponderilor constante ale unei perioade considerate de bază, în timp ce Principiul Paasche presupune utilizarea ponderilor variabile referitoare la perioadă curentă. Problema în cazul celor două principii este ca deşi sunt considerate acceptabile aplicarea lor duce la obţinerea de rezultate
Romanian Statistical Review - Supplement nr. 4 / 2019106
diferite, mai ales atunci când ponderile aplicate suferă modifi cări rapide şi considerabile. Acest capitol furnizează o analiză a relaţiei matematice fundamentale care există între cei doi indici. Utilizând cele două formule de calcul pentru măsurarea
indicilor de preţ şi de volum
å
å
=
==n
iii
n
iii
L
qp
qpqqppP
1
00
1
01
1010 ),,,( respectiv
å
å
=
==n
iii
n
iii
L
qp
qpqqppQ
1
00
1
10
1010 ),,,( pentru indicii de tip Laspyres şi
å
å
=
==n
iii
n
iii
P
qp
qpqqppP
1
10
1
11
1010 ),,,( respectiv
å
å
=
==n
iii
n
iii
P
qp
qpqqppQ
1
01
1
11
1010 ),,,(
pentru indicii de tip Paasche, se poate ilustra faptul că raportul dintre indicele de preţ Paasche şi cel Laspeyres este egal cu raportul dintre Indicele de volum Paasche şi cel indicelui Laspeyres:
÷÷÷÷
ø
ö
çççç
è
æ
÷÷÷÷
ø
ö
çççç
è
æ
==
å
å
å
å
=
=
=
=n
iii
n
iii
n
iii
n
iii
L
P
L
P
qp
qp
qp
qp
Q
Q
P
P
1
10
1
11
1
01
1
00
Ecuaţia de mai sus ne arată că sensul divergenţei existent între indicii de volum este acelaşi şi în cazul indicilor de preţ, astfel că dacă QP este mai mare decât QL atunci şi PP este mai mare decât QL. Folosind ecuaţia de mai
sus rezultă că
÷÷÷÷
ø
ö
çççç
è
æ
==
å
å
=
=n
iii
n
iii
PLLP
qp
qpQPQP
1
00
1
11
ceea ce ne arată că atât indicele de
tip Laspeyres cât şi cel de tip Paasche pică testul de inversare a factorului.
Ecuaţia precedentă ne arată că QP poate fi exprimat ca funcţie de PL
şi de cheltuielile perioadei curente şi a celor din perioadă de bază, după cum
Revista Română de Statistică - Supliment nr. 4 / 2019 107
urmează: L
n
iii
n
iii
P Pqp
qpQ
1
1
00
1
11
÷÷÷÷
ø
ö
çççç
è
æ
=
å
å
=
= .
Pentru simplifi care defi nim =0V å=
n
iii qp
1
00 şi å=
=n
iii qpV
1
111 de unde
rezultă că .1
0
1
LP PV
VQ ÷÷
ø
öççè
æ=
Folosind această funcţie în cadrul ecuaţiei L
P
L
P
Q
Q
P
P= rezultă că
raportul indicilor de preţ Paasche şi Laspeyres poate fi scris astfel:
÷÷ø
öççè
æ=
úúúúú
û
ù
êêêêê
ë
é÷÷ø
öççè
æ÷÷ø
öççè
æ
==LLL
L
L
P
L
P
QP
VV
Q
PV
V
Q
Q
P
P 010
1
/
1
Astfel, ponderea fi ecărui produs i în totalul cheltuielilor poate fi
calculată pentru orice perioadă t folosind formula: å=
=n
j
tj
tj
ti
ti
ti qpqps
1
/ .
Folosind aceste ecuaţii matematice fundamentale pentru descrierea relaţiei existente între indicii depreţ Laspeyres şi Paasche, se poate defi ni statistic, corelaţia care există între cei doi indici prin examinarea distribuţiei produs cu produs şi covarianţei dintre preţurile şi cantităţilor relative. Altă relaţie algebrică dintre indicii de preţ Laspeyres şi Paasche a fost analizată de Ladislaus Josephowitsch Bortkiewitcz. Scultz nota în anul 1997, că Bortkiewitcz nu a văzut descopunerea celor doi indici ca un instrument de analiză a rezultatelor numerice furnizate, exceptând un exemplu furnizat pentru ilustrarea propriei teoreme. În plus mulţi dintre urmaşi lui Bortkiewitcz au folosit ideea acestuia pentru studii abstracte, fi e pentru a examina anumite atribuţii generale ale formulelor indicilor, fi e pentru speculaţii cu privire la rezultatele comparative furnizate de diverse formule de calcul. Pentru ilustrarea teoremei lui Bortkiewitcz este ncesar de la bun început să defi nim covarianţa dintre vectorul preţurilor relative 01 / pp şi cel al cantităţilor 01 / qq ponderată cu vectorul 0
is . Utilizând notaţiile consistente cu indicii de preţ şi de volum Laspeyres, media ponderată a preţuirlor şi cantităţilor relative este calculată
Romanian Statistical Review - Supplement nr. 4 / 2019108
prin ponderarea atât a preţurilor relative cât şi a cantităţilor cu ponderile din perioada de bază 0
is :
Ln
iii
n
iii
n
ii
n
i i
ii
i
i Pqp
qp
s
p
ps
p
p===
å
å
å
å
=
=
=
=
1
00
1
01
1
0
10
10
0
1
Ln
iii
n
iii
n
ii
n
i i
ii
i
i Qqp
qp
s
q
qs
q
q===
å
å
å
å
=
=
=
=
1
00
1
01
1
0
10
10
0
1
Media ponderată a preţurilor şi cantităţilor relative sunt pur şi simplu egale cu indicele corespunzător de preţ şi de volum Laspeyres. De aceea covarianţa ponderată dintre vectorul preţurilor şi cel al cantităţilor relative utilizând vectorul 0
is poate fi defi nită astfel:
å ååå
å ååå
å
= ===
= ===
=
+--=
+--=
÷÷ø
öççè
æ-÷÷
ø
öççè
æ-=
n
i
n
iiLL
i
iLiL
n
i i
iiL
n
i ii
iii
n
i
n
iLLi
i
iLi
n
iL
i
ii
n
i ii
iii
Li
in
iL
i
iipqs
sQPq
qPsP
p
psQ
qp
qps
QPsq
qPsQ
p
ps
qp
qps
qP
p
ps
1 1
0
0
10
10
10
100
110
1 1
0
0
10
10
10
100
110
0
1
10
10
0s
considerând că 11
0 =å=
n
iis
ååå===
+--=n
iLL
i
iLiL
n
i i
iiL
n
i ii
iii QP
q
qPsP
p
psQ
qp
qps
10
10
10
10
100
110
considerând că
Li
in
ii P
p
ps =å
=0
1
1
0
åå==
+--=n
iLL
i
iiLLL
n
i ii
iii QP
q
qsPPQ
qp
qps
10
10
100
110
considerând că
å=
=n
iL
i
ii Q
q
qs
10
10
Revista Română de Statistică - Supliment nr. 4 / 2019 109
LL
n
i ii
iii QP
qp
qpså
=
-=1
00
110 , considerând că
å=
=n
iii
iii
qp
qps
1
00
000
LLi
i
i
in
jn
iii
ii QPq
q
p
p
qp
qp-÷÷ø
öççè
æ÷÷ø
öççè
æ
÷÷÷÷
ø
ö
çççç
è
æ
=åå=
=
0
1
0
1
1
1
00
00
LLn
iii
n
iii
QPqp
qp-=
å
å
=
=
1
00
1
11
, considerând că å=
=n
iii qpV
1
000 şi å=
=n
iii qpV
1
111
LLpqs
QPV
V-=
0
1
0s
Astfel,că raportul indicilor de Laspeyres poate fi scris folosind formula:
÷÷ø
öççè
æ=
úúúúú
û
ù
êêêêê
ë
é÷÷ø
öççè
æ÷÷ø
öççè
æ
==LLL
L
L
P
L
P
QP
VV
Q
PV
V
Q
Q
P
P 010
1
/
1
Adăugând şi scăzând unu în partea dreaptă a ecuaţiei de mai sus raportul indicilor de preţ Paasche şi Laspeyres poate fi rescris folosind formula:
11/ 01
-+÷÷ø
öççè
æ=
LLL
P
QP
VV
P
P
LL
LL
LLL
P
QP
QP
QP
VV
P
P-+÷÷
ø
öççè
æ= 1
/ 01
1/ 01
+÷÷ø
öççè
æ -=
LL
LL
L
P
QP
QPVV
P
P
Se observă că în partea dreaptă a ecuaţiei precedente, conţine la numărator covarianţa dintre preţurile şi cantităţile relative, prezentată mai sus. Astfel, raportul indicilor de preţ Paasche şi Laspeyres poate fi exprimat astfel:
Romanian Statistical Review - Supplement nr. 4 / 2019110
10
+÷÷ø
öççè
æ=
LL
pqs
L
P
QPP
P s
Bortkiewitcz susţine că un coefi cient negativ al covarianţei dintre preţurile şi cantităţile relative, care susţine legea statistică fi rească a legăturii
cererii şi ofertei, reprezintă o conditie sufi cientă pentru ca L
P
P
P<1. Semnul
covarianţei 0pqss determină sensul diveregenţei dintre indicii de preţ Paasche
şi Laspeyres. Indicele de preţ Paasche este mai mare decât indicele de preţ Laspeyres atunci când 0pqs
s este mai mică ca zero. Covarianţa nu este o marime statistică folositoare pentru măsurarea precisă a relaţiei dintre două variabile, deoarece este dependentă de unităţile de măsură folosite pentru cele două variabile. Corelaţia, pe de altă parte, este independentă de unităţile de măsură folosite pentru cele două variabile, deoarece legătura lineară dintre cele două variabile prin normalizarea produsului deviaţiei celor două variabile de la deviaţia lor standard. Ca urmare, coefi cientul de corelaţie dintre doua variabile x şi y, rxy , este delimitat între -1 şi 1 şi este mult mai folositor pentru compararea gradului de legătură dintre două variabile. Legătura statistică dintre covarianţă şi corelaţie este data de formula:
�/
/
/ ��
��
�
,
unde rpq reprezintă corelaţia dintre preţuri şi cantităţi
Înlocuind covarianţa +�������� ���������� � ���� /� � /� /�
��� � � �� �������� în formula
10
+÷÷ø
öççè
æ=
LL
pqs
L
P
QPP
P s, rezultă că raportul indicilor de preţ Paasche şi Laspeyres
poate fi defi nit după cum urmează:
���
�
�
�
�
��
--
/� � /�
-
.
2.
.
. ���
�
�
�
���
--
/� � /�
-
.
2.
.
. ���
���
�
�
�
�
�
��
�
--
/� � /�
-
-.
2.
.
.. ��� ���
�
Revista Română de Statistică - Supliment nr. 4 / 2019 111
Astfel că declajul dintre indicii Paasche şi Laspeyres este determinat de trei factori: 1. Coefi cientul de corelaţie 0pqs
r dintre preţurile şi cantităţilor relative; 2. Deviaţia standard a preţurilor relative ca raport a preţurilor medii
�������������������-
�
.
����������
�
(coefi cient de variaţie). 3. Deviaţia standard a cantităţilor relative ca raport a ���� ��
-
/�
2
�������������� �����cantitătilor
medii (coefi cient de variaţie). Cei doi coefi cienţi de variaţie sunt totdeauna pozitivi şi de aceea sensul divergenţei dintre indicii Paasche şi Laspeyres este determinat numai de coefi cientul de corelaţie. Decalajul dintre indicii Paasche şi Laspeyres poate fi sumarizat urmărind trei scenarii:
Concluzii
În urma analizei efectuate se desprind o serie de concluzii teoretice, sistematizate sub forma a trei scenarii. Scenariul 1: PL > PP şi QL > QP. Indicele Laspeyres este mai mare decât indicele Paasche, refl ectă situaţia unei pieţe tipice în care cererea este predominantă. În acest caz, preţurile şi cantităţile relative sunt corelate negative, şi astfel indicele Laspeyres depăşeste sistematic indicele Paasche. Aceasta este o situaţie normală a unei pieţe orientate spre cerere unde cumpărătorul domină şi vânzătorul se confruntă cu o curbă a cererii negative. Scenariul 2: Indicele Laspeyres este mai mic decât indicele Paasche. refl ectă situaţia în care factorul ofertă este cel care domină. În acest caz preţurile şi cantităţile relative se corelează pozitiv şi astfel indicele Laspeyres este în mod sistematic sub indicele Paasche. Această situaţie este probabil să apară atunci când piaţa este dominată de furnizori, unde consumatorii sunt incapabil să reacţioneze la creşterea preţurilor, fi e prin reducerea consumului sau prin substituirea produselor. Situaţia de faţă este comună, în special, în cazul mărfurilor vândute de exportatori pe o piaţă internaţională mare şi în cazul locuinţelor şi a produselor energetice. Scenariul 3: PL = PP şi QL=QP, indicele Laspeyres este egal cu indicele Paasche, refl ectă situaţia in care niciunul dintre factori, cerere sau ofertă, nu predomină şi nu există corelaţie între preţurile şi cantităţile relative. Scenariul acesta poate fi descris prin faptul că furnizorii vând în piaţă produse pe care consumatorii le tratează cu indiferenţă.
Romanian Statistical Review - Supplement nr. 4 / 2019112
Bibliografi e 1. Anghel, M.G., Mirea, M., Badiu, A. (2018). Analysis of the Main Aspects Regarding
the Price Indices Applied in the Determination of Infl ation. International Journal of Academic Research in Accounting, Finance and Management Sciences, 8 (2), 28-36
2. Anghel, M.G. (2015). The Infl ation (Consumer Price) Evolution. Romanian Statistical Review Supplement, 1, 128-132
3. Anghelache, C., Niţă, G., Badiu, A. (2016). The Infl ation (Consumer Prices) in the Romanian Economy. Romanian Statistical Review Supplement, 1, 99-102
4. Anghelache, C., Sacală, C. (2015). Some Theoretical Aspects regarding the Infl ation. Romanian Statistical Review Supplement, 6, 5 – 11
5. Anghelache, C., Voineagu, V., Gheorghe, M. (2013). Metode și modele de măsurare
şi analiză a infl aţiei, Editura Economică, Bucureşti 6. Armantier, O., Bruine de Bruin, W., Topa, G., Klaauw, W., Zafar, B. (2015).
Infl ation Expectations and Behavior: Do Survey Respondents Act on their Beliefs?. International Economic Review, 56 (2), 505-536
7. Ascari, G., Ropele, T. (2009). Trend Infl ation, Taylor Principle, and Indeterminacy. Journal of Money, Credit and Banking, 41, 1557–1584
8. Cogley, T., Sbordone A. (2008). Trend Infl ation, Indexation, and Infl ation Persistence in the New Keynesian Phillips Curve. American Economic Review, 98
(5), 2101–2126 9. Hornstein, A., Wolman, A.L. (2005). Trend Infl ation, Firm-Specifi c Capital, and
Sticky Prices. Economic Quarterly, 91 (4), 57–83 10. Kim, J., Henderson, D.W. (2005). Infl ation Targeting and Nominal-Income-
Growth Targeting: When and Why are They Suboptimal?. Journal of Monetary
Economics, 52, 1463 – 1495
Revista Română de Statistică - Supliment nr. 4 / 2019 113
THE USING OF THE LASPEYRES AND PAASCHE INDICES IN THE
DETERMINING OF THE PRICE INDEX FORMULA
Doina BUREA PhD Student ([email protected])
Bucharest University of Economic Studies
Abstract
In the context of microeconomic theory, the need to identify an economic model that refl ects the behavior of the producer or consumer, as
well as the numerical index issue, is the question: what is the „best” method
to build the set of aggregates of this model? Numeric indices can be seen as a
decomposition of the value of a well-defi ned set of transactions over a period
of time into an aggregate price and an aggregate amount of the quantities.
This approach can also be done at the index level, so that the value index
can be represented as the product of a component that measures the overall
price evolution of the two periods (price index) and a value that measures the
evolution of physical volume for the same periods ( quantitative index).
Keywords: price index, transactions, infl ation, Harmonized Index of
Prices, Laspeyres index, Paasche index.
JEL Classifi cation: E31, E60
Introduction
The fi xed base index is considered to be the simplest price index obtained as a ratio of two values calculated for the same set of quantities but using diff erent price sets for the two compared periods 0 base period and 1
current period. The selection of options for the constant set of quantities is
refl ected either in the use of the set of quantities traded in period 0 or period 1. In view of the above, I will detail the Laspeyres (187l) and Paasche type indices (1874). In this respect, the Laspeyres Principle consists in using the constant weights of a considered base period, and the Paasche principle assumes the use of variable weights for the current period. In the case of the two principles, the problem is that although they are considered acceptable their application leads to diff erent results, especially when the applied weights suff er rapid and
considerable changes.
Romanian Statistical Review - Supplement nr. 4 / 2019114
Literature review
Anghel, Mirea and Badiu (2018) studied fundamental elements of price indices used in infl ation measurement. Anghel (2015) and Anghelache, Niţă and Badiu, A. (2016) conducted studies on the evolution of the price index in Romania. Anghelache and Sacal (2015) presented a series of basic notions of infl ation. Anghelache, Voineagu, and Gheorghe (2013) analyzed the two Laspeyres and Paasche indicators, which were the basis of the calculation of consumer price indices (CPIs) worldwide. Armantier et al. (2015) addressed a number of aspects of infl ation forecasts. Ascari and Rope (2009) studied infl ation trends, a similar theme being analyzed by Cogley and Sbordone (2008). Hornstein and Wolman (2005) investigated the correlation between infl ation trends and company-specifi c capital. Kim and Henderson (2005) addressed issues related to infl ation targeting and nominal income growth.
Research methodology, data, results and discussions
The simplest method for determining the price index formula was described in detail by Lowe (1823). Its method for measuring the price change between moments 0 and 1 consisted in defi ning a representative set of products of the quantitative vector q =[q1,......, qn] then in estimating the value
aggregate using the prices of the period 1 and period 0: å=
n
iii qp
1
1 , represents
the cost of the representative set of products during the period 1, å=
n
iii qp
1
0 cost
of the basket of products for the period 0. (1)
The fi xed base method for determining the price index refers to the choice of vector q for the representative set of products. Economists and price statisticians also sought to select the vector q with some accuracy over time. The options for setting the benchmark set are: - the commodity vector of the base period q0 or - the goods vector of the current period q1. These two options result in the Laspeyres (1871) PL price index defi ned by equation (2) and the Paasche (1874) Pp price index defi ned by equation (3).
å
å
=
==n
iii
n
iii
L
qp
qpqqppP
1
00
1
01
1010 ),,,(* (2)
Revista Română de Statistică - Supliment nr. 4 / 2019 115
å
å
=
==n
iii
n
iii
P
qp
qpqqppP
1
10
1
11
1010 ),,,(*( (3)
The formulas (2) and (3) by alternative rewriting are more useful for statistical offi ces. The share of expenditure for commodity i from period t is
defi ned as follows:
å=
=n
j
tj
tj
ti
ti
ti qpqps
1
/ (4)
where i=1,….., n and t = 0, 1
The rewriting of the Laspeyres index (1) can be done as such:
001
1
00000
1
1
1
00
1
011010 )/(/)//),,,(* iii
n
jiiiii
n
ii
n
iii
n
iiiL sppqpqpppqpqpqqppP ååååå ===
====
(5)
Also, the Laspeyres PL price index can be written as the arithmetic
mean of the n price reports, 01 / ii pp , weighted by the base-year expenses.
It should be noted that this Laspeyres formula has served as a basis for
calculation of consumer price indices (CPIs) worldwide. For its application,
a statistical offi ce must collect information on the share of expenditure 0
ns
for the base defi nition domain of the base period, and then collect monthly
product price information only. Thus, the Laspeyres Consumer Price Index
(CPI) can be calculated without the need for information from the current
period. This Paasche index can also be written as a ratio between cost and
price rates as follows:
1
110
1
1
1
1101
1
11101
1
0
1
11
1
10
1010
)/(
)/(
1
/)/(
1
/
1),,,(*
-
-
=
=
-
====
þýü
îíì
=
þýü
îíì
=
þýü
îíì
=
þýü
îíì
=
å
ååååå
ii
n
ii
n
iiii
n
iiiiii
n
ii
n
iii
n
iii
P
spp
sppqpqpppqpqp
qqppP (6)
The Paasche PP price index may also be written as a harmonious
mean of n price reports, 01 / ii pp , ponderate cu cotele de cheltuieli din
perioada 1. weighted by the share of expenditure in period 1. Due to the lack
of information on the quantities of the current period, statistical offi ces are
restricted to producing Paasche-type indices on a regular basis. If we replace
in equation ),,,(/)/(),,,( 1010011010 qqppPVVqqppQ = P with PL
defi ned by equation (2) then the following quantitative index will be obtained:
Romanian Statistical Review - Supplement nr. 4 / 2019116
(PQ
å
å
=
==n
iii
n
iii
qp
qpqqpp
1
01
1
11
1010 ),,,( (7)
If P is replaced by Pp, then the following quantitative index is obtained:
å
å
=
==n
iii
n
iii
L
qp
qpqqppQ
1
00
1
10
1010 ),,,( (8)
Laspeyres and Paasche’s formula formulas show a syncope, even if they are plausible, will generally give diff erent answers. The statistical
offi ce to reach more goals must answer the question: What is the „best”
synthetic measure of modifying the price of aggregate value over the periods
considered? or what is the „normal” relationship between the Paasche and
Laspeyres indices?
The Bortkiewicz relationship can compare the Laspeyres and Paasche
indices. The calculation of the Laspeyres index is simple given that denominator
estimation is required only once, but the application of this clause does not
take into account the fall in demand or the change in production, which can
lead to an overestimation of price increases. On the other hand, the Paasche
index using the weights of the current period tends to underestimate the rise
in prices. In practice, it can be seen that prices and quantities do not change
with the same percentage, and therefore the relationship between two systems
depends on the correlation between price and quantity changes, which should
normally be negative.
The degree of divergence between the Laspeyres and Paasche indices,
regardless of the direction shown, depends on the intensity of the correlation
but also on the dispersion of relative prices and quantities, measured by the
coeffi cient of variation.
In the classical issue of purchasing power, for example, the Laspeyres
or Paasche price index is primarily determined by monetary factors, while
the divergence between the two formulas depends more on non-monetary
infl uences that work on the relative price dispersion around „standard”. The Laspeyres principle consists in using the constant weights of a considered base period, while the Paasche principle assumes the use of variable weights for the current period. The problem with the two principles is
Revista Română de Statistică - Supliment nr. 4 / 2019 117
that although they are considered acceptable their application leads to diff erent
results, especially when the applied weights suff er rapid and considerable
changes.
This study provides an analysis of the fundamental mathematical
relationship that exists between the two indices.
Using the two calculation formulas to measure price and volume indices
å
å
=
==n
iii
n
iii
L
qp
qpqqppP
1
00
1
01
1010 ),,,( respectively
å
å
=
==n
iii
n
iii
L
qp
qpqqppQ
1
00
1
10
1010 ),,,(
for Laspyres and indices
å
å
=
==n
iii
n
iii
P
qp
qpqqppP
1
10
1
11
1010 ),,,( respectively
å
å
=
==n
iii
n
iii
P
qp
qpqqppQ
1
01
1
11
1010 ),,,(
for the Paasche type indices, it can be illustrated that the ratio between the
Paasche and the Laspeyres price index is equal to the ratio between the
Paasche volume index and the Laspeyres index
÷÷÷÷
ø
ö
çççç
è
æ
÷÷÷÷
ø
ö
çççç
è
æ
==
å
å
å
å
=
=
=
=n
iii
n
iii
n
iii
n
iii
L
P
L
P
qp
qp
qp
qp
Q
Q
P
P
1
10
1
11
1
01
1
00
The above equation shows that the meaning of the divergence
between volume indices is the same for price indices, so if QP is higher
than QL then PP is higher than QL. Using the above equation results that
÷÷÷÷
ø
ö
çççç
è
æ
==
å
å
=
=n
iii
n
iii
PLLP
qp
qpQPQP
1
00
1
11
which shows that both the Laspeyres and the
Paasche type indexes the factor inversion test
Romanian Statistical Review - Supplement nr. 4 / 2019118
The previous equation shows that QP can be expressed as a function of PL and the current period and base period expenses, as follows:
Ln
iii
n
iii
P Pqp
qpQ
1
1
00
1
11
÷÷÷÷
ø
ö
çççç
è
æ
=
å
å
=
= .
For simplicity, we defi ne =0V å=
n
iii qp
1
00 şi å=
=n
iii qpV
1
111 and from
which it follows that .1
0
1
LP PV
VQ ÷÷
ø
öççè
æ=
Using this function within equation L
P
L
P
Q
Q
P
P= it follows that the ratio
of Paasche and Laspeyres price indices can be written as follows:
÷÷ø
öççè
æ=
úúúúú
û
ù
êêêêê
ë
é÷÷ø
öççè
æ÷÷ø
öççè
æ
==LLL
L
L
P
L
P
QP
VV
Q
PV
V
Q
Q
P
P 010
1
/
1
Thus, the weight of each product i in the total expense can be calculated
for any period t using formula: å=
=n
j
tj
tj
ti
ti
ti qpqps
1
/ .
Using these fundamental mathematical equations to describe the relationship between Laspeyres and Paasche, it is possible to defi ne statistically the correlation between the two indices by examining product distribution and the covariance between prices and relative quantities. Another algebraic relationship between the Laspeyres and Paasche price indices was analyzed by Ladislaus Josephowitsch Bortkiewitcz. I draw the note in 1997 that Bortkiewitcz did not see the two indices as an instrument for analyzing the numerical results provided, except for an example provided for illustrating their own theorems. In addition, many of Bortkiewitcz’s followers used his idea for abstract studies either to examine some general attributions of index formulas, or to speculate on the comparative results provided by various formulas. To illustrate Bortkiewitcz’s theorem, it is necessary from the very beginning to defi ne the covariance between the relative price vector 01 / pp and the quantity vector 01 / qq weighted by the vector 0
is . Using the labels consistent with the Laspeyres price and volume indices, the weighted average price and relative quantities are calculated by
Revista Română de Statistică - Supliment nr. 4 / 2019 119
weighting both the relative prices and the weights with the weights of the base period 0
is :
Ln
iii
n
iii
n
ii
n
i i
ii
i
i Pqp
qp
s
p
ps
p
p===
å
å
å
å
=
=
=
=
1
00
1
01
1
0
10
10
0
1
Ln
iii
n
iii
n
ii
n
i i
ii
i
i Qqp
qp
s
q
qs
q
q===
å
å
å
å
=
=
=
=
1
00
1
01
1
0
10
10
0
1
The weighted average of prices and relative quantities are simply equal to the corresponding price and volume index Laspeyres. Therefore, the weighted covariance between the price vector and the relative quantity using the vector 0
is can be defi ned as follows:
å ååå
å ååå
å
= ===
= ===
=
+--=
+--=
÷÷ø
öççè
æ-÷÷
ø
öççè
æ-=
n
i
n
iiLL
i
iLiL
n
i i
iiL
n
i ii
iii
n
i
n
iLLi
i
iLi
n
iL
i
ii
n
i ii
iii
Li
in
iL
i
iipqs
sQPq
qPsP
p
psQ
qp
qps
QPsq
qPsQ
p
ps
qp
qps
qP
p
ps
1 1
00
10
10
10
100
110
1 1
00
10
10
10
100
110
0
1
10
10
0s
considering that 11
0 =å=
n
iis
ååå===
+--=n
iLL
i
iLiL
n
i i
iiL
n
i ii
iii QP
q
qPsP
p
psQ
qp
qps
10
10
10
10
100
110 considering that
L
i
in
ii P
p
ps =å
=0
1
1
0
åå==
+--=n
iLL
i
iiLLL
n
i ii
iii QP
q
qsPPQ
qp
qps
10
10
100
110 considering that
å=
=n
iL
i
ii Q
q
qs
10
10
Romanian Statistical Review - Supplement nr. 4 / 2019120
LL
n
i ii
iii QP
qp
qpså
=
-=1
00
110 , considering that
å=
=n
iii
iii
qp
qps
1
00
000
LLi
i
i
in
jn
iii
ii QPq
q
p
p
qp
qp-÷÷ø
öççè
æ÷÷ø
öççè
æ
÷÷÷÷
ø
ö
çççç
è
æ
=åå=
=
0
1
0
1
1
1
00
00
LLn
iii
n
iii
QPqp
qp-=
å
å
=
=
1
00
1
11
, considering that å=
=n
iii qpV
1
000 şi å=
=n
iii qpV
1
111
LLpqs
QPV
V-=
0
1
0s
Thus, the ratio of the Laspeyres indices can be written using the formula:
÷÷ø
öççè
æ=
úúúúú
û
ù
êêêêê
ë
é÷÷ø
öççè
æ÷÷ø
öççè
æ
==LLL
L
L
P
L
P
QP
VV
Q
PV
V
Q
Q
P
P 010
1
/
1
Adding and lowering one to the right of the above equation, the Paasche and Laspeyres price index ratios can be rewritten using the formula:
11/ 01
-+÷÷ø
öççè
æ=
LLL
P
QP
VV
P
P
LL
LL
LLL
P
QP
QP
QP
VV
P
P-+÷÷
ø
öççè
æ= 1
/ 01
1/ 01
+÷÷ø
öççè
æ -=
LL
LL
L
P
QP
QPVV
P
P
It is noted that on the right side of the previous equation, it contains in the numerator the covariance between the prices and the relative quantities, presented above.
Revista Română de Statistică - Supliment nr. 4 / 2019 121
Thus, the ratio of Paasche and Laspeyres price indices can be expressed as follows:
10
+÷÷ø
öççè
æ=
LL
pqs
L
P
QPP
P s
Bortkiewitcz argues that a negative coeffi cient of the covariance
between prices and relative quantities, which supports the natural statistical
law of supply and demand, is a suffi cient condition for L
P
P
P<1. The sign
of covariance 0pqss determines the meaning of the diff erence between the
Paasche and Laspeyres price indices. The Paasche price index is higher than
the Laspeyres price index when 0pqss is less than zero.
The covariance is not a statistically useful value for accurately
measuring the relationship between two variables because it is dependent on the
units of measurement used for the two variables. The correlation, on the other
hand, is independent of the units of measurement used for the two variables,
because the linear relationship between the two variables by normalizing the
product of the deviation of the two variables from their standard deviation.
As a result, the correlation coeffi cient between the two variables x and y, rxy
is delimited between -1 and 1 and is much more useful for comparing the
degree of binding between two variables. The statistical relationship between
covariance and correlation is given by the formula:
�/
/
/ ��
��
�
,
where rpq represents the correlation between prices and
quantities.
Replacing the covariance +�������� ���������� � ���� /� � /� /�
��� � � �� �������� in the formula
10
+÷÷ø
öççè
æ=
LL
pqs
L
P
QPP
P s, it follows that the ratio of Paasche and Laspeyres price
indices can be defi ned as follows:
���
�
�
�
�
��
--
/� � /�
-
.
2.
.
. ���
�
�
�
���
--
/� � /�
-
.
2.
.
. ���
���
�
Romanian Statistical Review - Supplement nr. 4 / 2019122
�
�
�
�
��
�
--
/� � /�
-
-.
2.
.
.. ��� ���
� Thus the declamation of the Paasche and Laspeyres indices is determined by three factors: 1. The correlation coeffi cient 0pqs
r between relative prices and
quantities;
2. The standard deviation of relative prices as a ratio of average prices
�������������������-
�
.
����������
�
( coeffi cient of variation).
3. Deviaţia standard a cantităţilor relative ca raport a ���� ��-
/�
2
�������������� �����
cantitătilor medii (coeffi cient of variation).
The two coeffi cients of variation are always positive and therefore
the meaning of the diff erence between the Paasche and Laspeyres indices is determined only by the correlation coeffi cient. The gap between the Paasche
and Laspeyres indices can be summarized by following three scenarios.
Conclusion
Following the analysis, a series of theoretical, systematized
conclusions in the form of three scenarios are revealed.
Scenario 1: PL > PP şi QL > QP. The Laspeyres index is higher than
the Paasche index, refl ecting the situation of a typical market where demand
is predominant. In this case, relative prices and volumes are negatively
correlated, and the Laspeyres index is systematically outperforming the
Paasche index. This is a normal situation of a demand-driven market where
the buyer dominates and the seller faces a negative demand curve.
Scenario 2: The Laspeyres index is lower than the Paasche index.
refl ects the situation in which the supplying factor is the dominant one. In
this case, relative prices and quantities correlate positively and the Laspeyres
index is systematically below the Paasche index. This is likely to occur when
the market is dominated by suppliers, where consumers are unable to react to
price increases, either by reducing consumption or by substituting products.
The present situation is common, in particular, in the case of goods sold by
exporters in a large international market and in the case of housing and energy
products.
Scenario 3: PL = PP şi QL=QP, the Laspeyres index is equal to the
Paasche index, refl ects the situation where none of the factors, demand or
supply prevails and there is no correlation between prices and relative
Revista Română de Statistică - Supliment nr. 4 / 2019 123
quantities. This scenario can be described by the fact that vendors sell products in the market that the consumers treat them with indiff erence.
References
1. Anghel, M.G., Mirea, M., Badiu, A. (2018). Analysis of the Main Aspects Regarding
the Price Indices Applied in the Determination of Infl ation. International Journal of Academic Research in Accounting, Finance and Management Sciences, 8 (2), 28-36
2. Anghel, M.G. (2015). The Infl ation (Consumer Price) Evolution. Romanian Statistical Review Supplement, 1, 128-132
3. Anghelache, C., Niţă, G., Badiu, A. (2016). The Infl ation (Consumer Prices) in the Romanian Economy. Romanian Statistical Review Supplement, 1, 99-102
4. Anghelache, C., Sacală, C. (2015). Some Theoretical Aspects regarding the Infl ation. Romanian Statistical Review Supplement, 6, 5 – 11
5. Anghelache, C., Voineagu, V., Gheorghe, M. (2013). Metode și modele de măsurare
şi analiză a infl aţiei, Editura Economică, Bucureşti 6. Armantier, O., Bruine de Bruin, W., Topa, G., Klaauw, W., Zafar, B. (2015).
Infl ation Expectations and Behavior: Do Survey Respondents Act on their Beliefs?. International Economic Review, 56 (2), 505-536
7. Ascari, G., Ropele, T. (2009). Trend Infl ation, Taylor Principle, and Indeterminacy. Journal of Money, Credit and Banking, 41, 1557–1584
8. Cogley, T., Sbordone A. (2008). Trend Infl ation, Indexation, and Infl ation Persistence in the New Keynesian Phillips Curve. American Economic Review, 98
(5), 2101–2126 9. Hornstein, A., Wolman, A.L. (2005). Trend Infl ation, Firm-Specifi c Capital, and
Sticky Prices. Economic Quarterly, 91 (4), 57–83 10. Kim, J., Henderson, D.W. (2005). Infl ation Targeting and Nominal-Income-
Growth Targeting: When and Why are They Suboptimal?. Journal of Monetary
Economics, 52, 1463 – 1495
Romanian Statistical Review - Supplement nr. 4 / 2019124
Compararea internaţională a performanţei macroeconomice
Drd. Ștefan Gabriel DUMBRAVĂ ([email protected])
Academia de Studii Economice din BucureștiDrd. Alexandra PETRE (OLTEANU) ([email protected])
Academia de Studii Economice din BucureștiDrd. Cristian OLTEANU ([email protected])
Academia de Studii Economice din BucureștiDrd. Tudor SAMSON ([email protected])
Academia de Studii Economice din București
Abstract Problema comparabilității internaționale este una de mare actualitate, în sensul că, pe baza indicatorilor macroeconomici de rezultate se pot efectua ierarhizări, clasamente pe baza cărora să se poată efectua studii comparative. Compararea internațională a gradului de dezvoltare economică a diferitelor țări se poate efectua pe baza indicatorilor în expresie fi zică cum ar fi consumul pe locuitor, investițiile, indicii calității vieții etc. Pentru a desprinde concluzii cu privire la nivelul de dezvoltare, se alege o țară de referință, iar nivelurile înregistrate de celelalte țări se compară cu nivelul luat drept etalon. În contextul comparării internaționale a PIB se ajunge la concluzia că, din mai multe motive, cursul de schimb nu poate fi considerat întotdeauna un element decisiv, deoarece modifi cările agegatelor depind și de foarte mulți factori interni. În aceste condiții, comparabilitatea se realizează pe baza parității puterii de cumpărare ale valutelor țărilor care se compară. Paritatea puterii de cumpărare exprimă numărul de unități de valută necesare pentru cumpărarea într-o țară a aceluiași volum de bunuri. În cadrul Uniunii Europene există un program de ierarhizare/clasifi care a țărilor membre după acest indicator. În acest sens, țările Uniunii Europene sunt împărțite în trei grupe după valoarea indicelui parității de cumpărare. În această ierarhizare, România se afl ă în grupa a doua alături de țări cum sunt Bulgaria, Slovenia, Albania, Austria, Belarus, Moldova sau Ucraina. Pe plan mai larg european, tot pe baza parității puterii de cumpărare se fac ierarhizări ale statelor. În astfel de comparații internaționale s-au identifi cat și statuat unele procedee de comparare a indicatorilor, cum sunt: compararea binară, compararea de tip stea, metoda asterisk, metoda Geary – Khanis, metoda Contry – Product Dummy, metoda EKS, utilizarea parității de cumpărare și altele. Cuvinte cheie: indicatori macroeconomici de rezultate, comparare internațională, performanță, paritatea puterii de cumpărare, procedeu Clasifi carea JEL: C10, C82, E31
Revista Română de Statistică - Supliment nr. 4 / 2019 125
Introducere
În cadrul studiului efectuat, autorii au pus accentul pe identifi carea posibilităților de comparare internațională a performanțelor macroeconomice ale diferitelor țări. Comparara internațională macroeconomică nu este doar un aspect teoretic, ci o problemă practică legată de comparațiile între țări, aspect care este o preocupare a instituțiilor naționale și internaționale de statistică. Autorii analizează pe larg particularitățile pe care le presupune utilizarea parității puterii de cumpărare în comparațiile internaționale. Compararea se poate efectua utilizând indicatorii macroeconomici de rezultate calculați în cadrul Sistemului Cornturilor Naționale, deoarece țările care utilizează acest sistem de evidență macroeconomică, calculează indicatorii macroeconomici după o concepție și o metodologie unitare. În continuare, se evidențiază faptul că, pentru realizarea comparării internaționale sunt urmate, în principiu trei etape, respectiv: PIB-ul fi ecărei țări este descompus într-un număr de grupe primare de cheltuieli; în cadrul fi ecărei grupe primare, sunt stabilite prețuri selectate care trebuiesc să asigure comparabilitatea și reprezentativitatea; valorile naționale ale grupelor primare sunt transformate în valori comparabile internațional prin intermediul parității puterii de cumpărare. În asigurarea comparabilității, există două programe, unul pentru cooperarea europeană și altul pentru programul de comparare internațională. Autorii acordă importanță prezentării principalelor procedee de comparare internațională a indicatorilor. Procedeele menționate se pot utilize cu ușurință pentru a face amalize practice pe baza datelor existente.
Literature review
Anghel, Anghelache și Samson (2017) au abordat o serie de aspecte referitoare la metodologia privind comparabilitatea indicatorilor de rezultate. Anghelache, Mitruţ și Voineagu (2013), precum și Anghelache şi colaboratorii (2007) au analizat și au prezentat elemente fundamentale ale statisticii macroeconomice. Anghelache (2009) a acordat o atenție deosebită calculului principalilor indicatori macroeconomici care utilizaţi în comparabilitatea internaţională. Anghelache (2008) a prezentat noțiuni de bază ale statisticii economice, precum și indicatorii de calcul utilizat în activitățile economice. Goodwin (2008) a analizat aspecte ale macroeconomiei specifi ce secolului actual. Chetty (2009) a studiat metode de analiză a bunăstării. Farhi și Werning (2016) au cercetat teorii ale politicilor macroprudențiale. Ftiti (2010) a analizat elemente ale performanței macroeconomice a politicii de direcționare a infl ației. Öllera, și Teterukovsky (2007) au studiat modalități de cuantifi care a calității variabilelor macroeconomice.
Romanian Statistical Review - Supplement nr. 4 / 2019126
Metodologia cercetării, date, rezultate și discuții Compararea agregatelor macroeconomice de rezultate ridică probleme legate de exprimarea bănească, a sferei de cuprindere, a metodologiilor de calcul, a destinaţiei de utilizare a indicatorilor macroeconomici de rezultate.
• Utilizarea parităţii puterii de cumpărare în comparaţiile
internaţionale
Compararea indicatorilor macro- nu este doar o chestiune teoretică, ci, în mai mare măsură, o problemă practică legată de comparaţiile între ţări, afl ată în preocupările instituţiilor naţionale şi internaţionale de statistică. Compararea internaţională a gradului de dezvoltare economică a diferitelor ţări se poate efectua pe baza indicatorilor în expresie fi zică prin compararea producţiei şi a consumului pe locuitor înregistrat la o serie de produse de bază. Compararea presupune alcătuirea a două liste: una cu ţările care se compară şi cu produsele de bază ce intră în comparaţie. Este necesar ca aceste produse să fi e reprezentative pentru toate ţările cuprinse în comparaţie şi cât de cât similare din punct de vedere calitativ. În continuare, pentru a desprinde concluzii cu privire la nivelul dezvoltării, se alege o ţară de referinţă, iar nivelurile atinse de celelalte ţări se compară cu nivelul ţării luate drept etalon. În acest scop, se utilizează mărimile relative de coordonare (Y). Compararea se poate efectua utilizând indicatorii macroeconomici de rezultate calculaţi în cadrul SCN. Compararea internaţională a nivelului, structurii şi ritmului de dezvoltare pe baza indicatorilor macroeconomici de rezultate necesită rezolvarea a două probleme esenţiale, şi anume: realizarea comparabilităţii indicatorilor din punctul de vedere al conţinutului acestora, al sferei lor de cuprindere şi asigurarea comparabilităţii indicatorilor din punctul de vedere al unităţilor monetare în care se exprimă. Ţările care utilizează SCN determină indicatorii macroeconomici după
o concepţie şi o metodologie unitare. Evaluarea indicatorilor macroeconomici într-o valută unică se poate efectua prin următoarele procedee: - Pe baza cursurilor de schimb ofi ciale, cursul de schimb reprezintă preţul ce trebuie plătit pe piaţa devizelor în valută străină pentru a obţine o unitate de valută naţională sau cantitatea de monedă străină ce se primeşte în schimbul unei unităţi de monedă naţională. În contextul comparării internaţionale a PIB, dezbaterile teoretico-metodologice au concluzionat că, din mai multe motive, cursul de schimb nu poate fi considerat element decisiv în asigurarea comparabilităţii indicatorilor în expresie absolută şi nu poate fi folosit la transformarea indicatorilor sintetici din valuta naţională, deoarece: nu refl ectă structura reală a PIB; este infl uenţat
Revista Română de Statistică - Supliment nr. 4 / 2019 127
de factori interni şi internaţionali, determinând fl uc tuaţii ale acestuia pe perioade scurte de timp; există cursuri valutare multiple. Acesta este un procedeu relativ uşor de aplicat, dar care poate infl uenţa negativ valorile reale, întrucât cursurile de schimb nu refl ectă, de
regulă, puterea de cumpărare a valutelor. Acest lucru se explică prin faptul că,
pe de o parte, stabilirea cursului de schimb este infl uenţată numai de un număr
restrâns de bunuri, şi anume acelea care fac obiectul relaţiilor comerciale
dintre ţările respective, şi, pe de altă parte, cursul de schimb este puternic
infl uenţat de diferitele operaţii speculative de la bursă. Rezultă că exprimarea
indicatorilor într-o singură valută pe baza cursurilor de schimb nu este o
metodă recomandată pentru efectuarea unor comparaţii internaţionale cât mai
corecte privind nivelurile, structurile şi ritmurile de dezvoltare ale diferitelor ţări.
Pe baza evaluării parităţii puterii de cumpărare (PPC) a valutelor ţărilor
care se compară. Această metodă de evaluare a indicatorilor macroeconomici într-o
singură valută este mai indicată spre utilizare, comparativ cu metoda cursurilor
de schimb ofi ciale. Ea constă în folosirea unor indici ai preţurilor calculaţi pe
baza preţurilor bunurilor materiale şi serviciilor din ţara care face calculele de
comparaţie şi ai preţurilor aceloraşi bunuri materiale şi servicii din ţara a cărei
valută s-a hotărât să fi e utilizată pentru exprimarea indicatorilor macroeconomici.
PPC exprimă numărul de unităţi de valută necesare pentru cumpă-
rarea într-o ţară a aceluiaşi volum de bunuri care se poate obţine cu o unitate
monetară a ţării bază de comparare sau cu o unitate monetară comună unui
grup de ţări. PPC pot fi calculate pentru total Produs Intern Brut, dar şi pentru
niveluri diferite de agregare (de exemplu, pentru formarea brută de capital fi x).
Fundamentarea acestor indici de preţuri se face pe baza principiilor
generale ale calculării indicilor de preţuri. Este ştiut că indicele agregat al
preţurilor se calculează utilizând preţurile bunurilor din perioada curentă şi
de bază (p1 şi po), iar ca ponderi pentru ambele perioade fi e cantităţile din
perioada curentă (q1), fi e cele din perioada de bază (q0).
Paritatea puterii de cumpărare reprezintă raportul mediu între preţurile
practicate în ţara care se compară şi cele ale ţării luate ca bază de comparare,
la un număr mare de categorii de cheltuieli din compunerea PIB, defalcate
din elementele globale (consumul fi nal al populaţiei, consumul colectiv
guvernamental, formarea brută a capitalului fi x, modifi carea stocurilor şi
exportul net).
Determinarea PPC pe baza preţurilor bunurilor şi serviciilor cuprinse
într-un eşantion de articole comparabile ce sunt selectate cu scopul de a
reprezenta o clasifi care bine defi nită şi de a fi reprezentative pentru modelul
de consum al ţărilor participante la comparare poate fi asimilată unui indice de
preţ de tip teritorial.
Romanian Statistical Review - Supplement nr. 4 / 2019128
Pentru realizarea comparării internaţionale, sunt urmate, în principiu, trei etape, respectiv: - PIB-ul fi ecărei ţări este descompus într-un număr de grupe primare de cheltuieli. - În cadrul fi ecărei grupe primare sunt preţuri pentru articole selectate care ar trebui să respecte două proprietăţi esenţiale: comparabilitatea (care înseamnă că factorii ce infl uenţează formarea preţului efectiv sunt identici în ţările care se compară) şi reprezentativitatea (care înseamnă că articolele selectate sunt caracteristice pentru modelul cheltuielilor şi au o pondere semnifi cativă în grupa primară a ţărilor participante la comparare). Preţurile culese (inregistrate/observate) sunt folosite pentru a obţine rapoarte de preţ pentru bunurile şi serviciile individuale, apoi rapoartelor de preţuri li se calculează media pentru a obţine parităţi neponderate la nivel de grupă primară; în fi nal, utilizând ca ponderi structurile de cheltuielile PIB ale ţărilor care sunt comparate în formula mediei geometrice ponderate, se obţine o PPC pentru fi ecare nivel de agregare – până la nivel de PIB. - Valorile naţionale ale grupelor primare sunt transformate în valori „comparabile internaţional” prin intermediul parităţilor puterii de cumpărare. Rezultatele obţinute sunt valori reale, spre a le deosebi de valorile nominale (exprimate în valută naţională). Adaptarea presupune ca preţurile produselor celor două ţări să fi e ponderate fi e cu cantităţile unei ţări (de exemplu, ţara A), fi e cu cantităţile celeilalte ţări (sau ţara de referinţă, ţara B). Evident, între rezultatele obţinute prin utilizarea celor doi indici de preţuri există anumite diferenţe datorate aplicării unor ponderi diferite. În scopul eliminării acestor infl uenţe se calculează un indice de preţ de tip Fischer ca medie geometrică a indicilor de tip Paasche şi Laspeyres. Indiferent de variantă (Paasche, Laspeyres sau Fischer), calcularea unui astfel de indice general este o operaţiune foarte greu de realizat, putând fi efectuată numai de organismele naţionale şi internaţionale de statistică. Există mai multe metode de abordare şi soluţionare a procedurilor de comparare la paritatea puterii de comparare (PPC), unele dintre ele fi ind promovate de către instituţii de statistică internaţională. Principalele metode şi programe se vor prezenta în continuare.
• Programul de Comparare Europeană
Primele acţiuni sistematice de comparare în Europa s-au efectuat începând cu anul 1980, sub auspiciile Comisiei Economice pentru Europa, în contextul regionalizării Programului de Comparare Internaţională. România a
Revista Română de Statistică - Supliment nr. 4 / 2019 129
participat la lucrările Programului European de Comparare în anii de referinţă: 1990, 1993, 1996 şi 1999. Pentru runda 1993, ţările participante au fost împărţite în trei grupe: grupa I acoperind ţările Europei de Vest, grupa a II-a incluzând ţări ale Europei Centrale şi de Est şi grupa a III-a cuprinzând noile state independente ale fostei Uniuni Sovietice. Acestea au fost comparări bilaterale, fi ecare ţară fi ind comparată direct cu Austria. Deci atât comparările între ţările grupei a II-a, cât şi comparările între ţările grupei I (şi alte ţări OECD) şi ţările grupei a II-a s-au realizat prin intermediul Austriei, „ţară punte”, iar în grupa a III-a, s-a adoptat modelul comparării multilaterale. Runda de comparare 1993 a coincis cu modifi cările politice ale Europei Centrale şi de Est, cu destrămarea fostei Uniuni Sovietice. 23 de ţări au făcut parte din grupa a II-a. România a fost inclusă într-un program de comparare bilaterală: România – Republica Moldova şi România – Moldova prin Austria (ţara-punte). Runda de comparare 1996 a fost organizată sub forma unei serii de comparaţii multilaterale înlănţuite între ţări, făcându-se, de data acesta distincţia între trei grupe, a căror structură geografi că diferă faţă de runda 1993. - grupa I: cele 15 ţări membre ale UE (din 1996): Islanda, Norvegia, Elveţia, Turcia, Israel, Federaţia Rusă şi Republica Slovacă la care se adaugă Polonia, Ungaria, Republica Cehă. Legături directe cu celelalte grupe au fost asigurate printr-o comparaţie cu Austria (cu grupa a II-a) şi prin Turcia (cu grupa a III-a de ţări); - grupa a II-a: Estonia, Letonia, Lituania, Bulgaria, România, Slovenia, Croaţia, fosta Republică Iugoslavă a Macedoniei, Albania, Austria, Belarus, Moldova, Ucraina şi Federaţia Rusă. - grupa a III-a (grupa CSI): Federaţia Rusă, Armenia, Azerbaidjan, Georgia, Mongolia, Kazahstan, Republica Kârgâstan, Tadjikistan, Turkmenistan, Uzbekistan, Turcia, Belarus şi Republica Moldova. Fără a exista o schimbare de metodologie faţă de runda anterioară, 1996, EUROSTAT a hotărât componenţa a trei grupe coordonate de către un stat membru al Uniunii Europene, apartenenţa ţărilor la un grup sau altul fi ind judecată în funcţie de: similaritatea pieţelor (în vederea atât maximizării numărului de articole evaluate în cel puţin două dintre ţările partenere, cât şi a apropierii geografi ce. Ţările coordonatoare în cazul rundei 1999 sunt:
Finlanda, Austria şi Italia. România face parte din grupa coordonată de Italia, alături de Bulgaria, Cipru, Franţa, Grecia, Portugalia, Spania.
Romanian Statistical Review - Supplement nr. 4 / 2019130
• Programul de Comparare Internaţională (PCI)
PCI este o lucrare organizată la nivel internaţional, care permite realizarea de comparări economice internaţionale prin eliminarea diferenţelor între ţări în ceea ce priveşte nivelul preţurilor. Lucrările PCI au fost organizate de către Comisia de Statistică a Naţiunilor Unite. Şase runde au marcat existenţa proiectului, demarat în 1968; începând cu faza a treia, proiectul a căpătat caracteristici de program organizat o dată la cinci ani. Compararea are două etape principale, respectiv: - PIB al fi ecărei ţări este descompus într-un număr de grupe primare. Descompunerea cheltuielilor pe grupe primare este considerată o clasifi care a cărei raţiuni este aceea de a asigura estimări mai corecte ale agregatelor de cheltuieli de nivel mai înalt, care decurge din considerentele: este asigurată sfera de cuprindere a tuturor agregatelor de cheltuieli; dispersia rapoartelor individuale de preţ este mai mică în cadrul grupei primare decât între grupele din cadrul agregărilor de nivel mai înalt; calculul mediilor ponderate ale parităţilor se poate face la nivel relativ detaliat. - Grupa primară de cheltuieli PIB reprezintă nivelul primar pentru care sunt calculate parităţile puterii de cumpărare. Grupa primară este cea mai detaliată entitate a clasifi cării cheltuielilor PIB. Grupa primară se defi neşte ca fi ind grupa de produse omogene din care va fi selectat un număr de articole pentru a fi urmărite prin anchete de preţ, articole reprezentative atât pentru categoria de produse, cât şi pentru ţara în cauză participantă la comparare. În delimitarea grupelor primare, este necesar să se respecte ca principii: omogenitatea, interpretată în termenii dispersiei rapoartelor individuale de preţ, şi faptul că datele de cheltuieli pentru grupele primare sunt folosite drept ponderi. Aşa, de exemplu, PIB este descompus în agregate principale de cheltuieli, iar „consumul individual al gospodăriilor” este unul dintre acestea. La rândul său, acesta este defalcat pe grupe de cheltuieli: „Alimente, Băuturi şi Tutun” este un exemplu pentru care „Alimente” reprezintă o subgrupă. În cadrul acestora, sunt categorii de cheltuieli: de exemplu, „Pâine şi cereale”. „Orez” şi „Făină şi alte cereale” sunt grupe primare. Reprezentarea grupelor primare prin intermediul articolelor individuale înseamnă, în cazul orezului, a evalua cel puţin una dintre specifi caţiile: orez, boabă lungă; orez, boabă medie; orez, boabă lungă, prefi ert; orez pentru gătire rapidă. În cadrul fi ecărei grupe primare sunt culese preţuri pentru articolele selectate care respectă proprietăţile de comparabilitate şi reprezentativitate. Preţurile culese sunt folosite pentru a obţine rapoarte de preţuri pentru bunurile şi serviciile individuale; apoi, rapoartelor de preţuri li se calculează media pentru a obţine parităţi neponderate la nivel de grupă primară; în fi nal,
Revista Română de Statistică - Supliment nr. 4 / 2019 131
utilizând ca ponderi structurile de cheltuieli PIB ale ţărilor care sunt comparate
în formula mediei geometrice ponderate, se obţine o PPC pentru fi ecare nivel de agregare – până la nivel de PIB. Valorile nominale sunt transformate în valori reale prin intermediul PPC. Principalele informaţii statistice care se obţin din calculele de comparaţii sunt: - Rapoartele de preturi (PPC) folosite pentru determinarea rapoartelor de volum; nu este vorba doar de o singură paritate generală a puterii de cumpărare, ci de un set de PPC-uri pentru defl atarea diferitelor grupe/categorii de bunuri şi servicii; - Indicii de volum general şi pe locuitor; cifrele de volum pot fi utilizate
pentru a crea totaluri zonale, regionale şi a determina ponderea fi ecărei ţări în total (se poate stabili o ierarhizare a ţărilor faţă de o anumită ţară luată ca bază (100,0%); - Indicele comparativ al nivelului preţului; - Indicatori de structură: de exemplu, împărţind indicele de volum al consumului populaţiei la indicele de volum al PIB, se determină indicele de volum relativ între două ţări, care, în analiză, se poate interpreta astfel încât se poate aprecia că, faţă de PIB, volumul relativ în gospodării, sau ceea ce se numeşte „rata consumului real” este mai mare sau mai mică într-o ţară faţă de cealaltă.
• Principalele procedee de comparare internaţională a indicatorilor
Practica statistică a SCN a confi rmat mai multe procedee de comparare, mai cunoscute fi ind: compararea binară, compararea multilaterală, stea, metoda Geary – Khanis, metoda Country-Product – Dummy şi altele. A. Compararea binară este efectuată între două ţări, iar analiza nu implică alte date decât ale celor două ţări participante la program. În practică, etapele de calcul sunt următoarele: - calculul Parităţii Puterii de Cumpărare la nivel de grupă primară se efectuează prin mai multe modalităţi, utilizând, după caz: a) indici de tip Laspeyers:
nn
i
iBA
LBA RPMPPC
1
1
÷ø
öçè
æ= Õ=
(1)
unde: n - număr de articole reprezentative pentru ţara de bază „A” ale grupei primare respective
=LBA RPM raportul preţurilor medii ale articolului „i” = PM i
A / /PM iB/
Romanian Statistical Review - Supplement nr. 4 / 2019132
b) indici de tip Paasche:
nn
i
iAB
PBA RPMPPC
1
1
÷ø
öçè
æ= Õ=
(2)
unde: n = număr de articole reprezentative pentru ţara „B” ale grupei primare respective;
=LAB RPM raportul preţurilor medii ale articolului „i” = PM i
B/ / PM iA /
c) indici de tip Fisher:
PBA
LBA
FBA PPCPPCPPC ´= (3)
- calculul Parităţii Puterii de Cumpărare la nivel agregat a) folosind indici de tip Laspeyres:
( )
åå =
=
´´=n
i
iA
iBAn
iA
LBA WPPC
WPPC
1
1
1
(4)
unde: n – numărul de grupe primare ale agregatului economic comparat
PPCPPC iBA = pentru grupa primară „i” a ţării de bază „A”
WA = cheltuieli PIB aferente grupei primare „i” a ţării de bază „A”
b) folosind indici de tip Paasche:
( )
( )å
å
=
=
´=
n
i
iB
iAB
n
iB
PBA
WPPC
WPPC
1
1
(5)
unde:
iBA PPC = PPC a grupei primare „i”
Wb = cheltuieli PIB aferente grupei primare „i” a ţării „B” c) folosind indici de tip Fischer:
PBA
LBA
FBA PPCPPCPPC ´= (6)
Revista Română de Statistică - Supliment nr. 4 / 2019 133
B. Compararea de tip stea şi multilaterală
Compararea stea este cea în care un set selectat de comparări binare este utilizat pentru a deduce comparările pentru alte perechi de ţări. Fiecare ţară este comparată în mod (direct) cu ţara de bază, iar comparările între oricare două ţări rezultă din rapoartele corespunzătoare ale ţărilor cu ţara de bază. În baza indicilor de preţ sau cantităţi Ij/k şi Il/k sunt ţările participante la comparare (j, k, l) cu ajutorul metodei de comparare tip stea se calculează indicele Ij/il ca raport între indicii Ij/k şi Il/k. Rundele 1990 şi 1993 ale PCE din grupa a III-a au fost comparări binare de tipul stea, fi ecare ţară fi ind comparată direct cu Austria ca ţară-punte. Compararea multilaterală simultană a mai mult de două ţări având drept rezultat relaţii valabile între toate perechile de ţări, verifi cându-se astfel proprietatea de circularitate sau tranzitivitate. Interpretarea rezultatelor, chiar pentru două ţări, se construieşte în acest caz pe baza datelor tuturor ţărilor participante la exerciţiu. Elementele de care este necesar a se ţine seama pentru ca o astfel de abordare complexă să fi e accesibilă sunt: un nivel asemănător de dezvoltare a pieţelor naţionale un standard tehnic echivalent între ţări; o infrastructura statistică adecvată pentru a asigura datele de preţ şi de cheltuieli (imputate) PIB; o listă comună de bunuri şi servicii care urmează a fi evaluate, listă ce se alcătuieşte prin consultarea tuturor ţărilor participante, astfel încât, în fi nal, aceasta să fi e echireprezentativă. C. Metoda asterisk este un procedeu simplu care presupune indicarea, de către fi ecare ţară participantă, prin intermediul unui asterisk, a articolelor „reprezentative pentru modelul său de consum. Acestea pot fi : articole pe care ţara le-a propus la alcătuirea listei pentru ţările participante la programul de comparare; articole disponibile pe scară largă pe piaţa internă şi uşor de identifi cat pentru a fi evaluate – decizia este în funcţie de volumul relativ mare al vânzărilor articolului în cauză; articole prezente în calculul indicelui naţional al preţurilor de consum. Omologarea unui articol reprezentativ, semnalată prin asterisk (*), generează observaţiile de preţ incluse în calcul prin faptul că indicii de tip Laspeyres se calculează luând în considerare preţurile pentru acele articole reprezentative pentru ţara de bază, în timp ce indicii de preţ Paasche se determină pe baza articolelor reprezentative pentru ţara parteneră în programul de comparare. D. Metoda G – K (Geary – Khanis)
Este o metodă de comparare multilaterală prin care categoriile de cheltuieli exprimate în valută naţională sunt reevaluate prin intermediul parităţilor puterii de cumpărare calculate ca medii aritmetice ponderate
Romanian Statistical Review - Supplement nr. 4 / 2019134
ale preţurilor predominante într-o regiune. Rezultatele obţinute răspund principiului tranzitivităţii şi aditivităţii şi verifi că testul reversibilităţii indicilor teritoriali. Calculul are la bază parităţile puterii de cumpărare ale grupelor primare – considerate preţuri fi ctive – şi cantităţile fi ctive calculate ca raport între
valorile nominale şi parităţile corespunzătoare.
Notând preţurile internaţionale cu pij şi parităţile globale cu fj, valorile
acestora se determină prin rezolvarea unui sistem de ecuaţii:
"����������� �������� ��������������"���������� ���������0��#�
� � � � �
�=
/
=
/
�/�//�/� >>�&�
- -
E ��Q�-)�3)�/�Q�-��=� ���.�� (7)
� � � � � � �
�3
�
3
�
�/�/�/�/ >&>�- -
E ��Q�-)�3)�/�Q�-��=��� ��
!�"������0����� ���0������������������������������ ����������
(8)
Nivelul preţului internaţional al grupei primare (i) reprezintă o medie
aritmetică ponderată cu cantităţile fi ctive ale preţurilor fi ctive ajustate cu
paritatea globală a puterii de cumpărare pentru fi ecare ţară, unde:
Notaţiile din formulele anterioare indică:
Pij = preţul fi ctiv pentru grupa primara (i) în ţara (j) (în fapt, paritatea
puterii de cumpărare pentru grupa primară (i)
Qij = cantitatea fi ctivă
Fj = paritatea globală a puterii de cumpărare, pentru PIB-ul total al ţării (j)
9M�Q���� � �������������� ������������������
�
�=
��/� >?
-
�Q���� � � ��� � ������� � �� �� 0�����F� ������� = cantitatea totală pentru toate ţările – în cazul grupei
primare (i)
N = numărul de ţări
M = numărul de grupe primare
Paritatea globală a puterii de cumpărare a ţării (j) se calculează ca
raport al costului total exprimat în preţuri internaţionale şi costul total în
preturi naţionale.
E. Metoda < Country – Product – Dummy (CPD) >
Este un model de comparare de tip stea generalizat, care apelează la
analiza de regresie pentru comparări de preţ tranzitive pentru fi ecare grupă
primară. Datele de bază pentru o anume categorie constau din toate preţurile
disponibile pentru diferite specifi caţii pentru toate ţările participante la
comparare. Fiecare preţ este judecat în funcţie de ţara de observare şi de articolul
la care se referă. Logaritmul natural al preţului devine variabila dependentă
într-o ecuaţie de regresie lineară în care două seturi de variabile binare sunt
variabile independente: unul referitor la ţări (mai puţin ţara numerar), celălalt
Revista Română de Statistică - Supliment nr. 4 / 2019 135
referitor la articole. Parităţile tranzitive ale puterii de cumpărare pentru grupele primare rezultă din calculul coefi cienţilor de regresie a variabilelor binare reprezentând ţările.
F. Metoda EKS
Acest procedeu reprezintă o metoda multilaterală de comparare internaţională a PIB dezvoltată de Elteto, Koves şi Szulef. Se calculează indicele de preţ (indice Fisher multilateral) ca medie geometrică neponderată a tuturor indicilor Fisher posibil a fi determinaţi pentru n ţări.
( )
nn
k,ll
jkEKSk/j FFFPPC
1
j1
l/l.2
j/k /
úúú
û
ù
êêê
ë
é= Õ
¹=
unde: Fk/k = 1 (9)
De exemplu, într-o aplicaţie acoperind 4 ţări şi notând indicii Fisher pentru fi ecare pereche de ţări cu FB/A, FC/A, …, indicii de tip EKS (FB/
A*,FC/A*,… ) sunt calculaţi conform formulei:
( )[ ] ( )[ ]{ } 4/12B/AC/BD/AC/DC/AC/A FFFFFF ´´´´= (10)
Metoda se aplică în calculul parităţilor grupelor primare, dar şi în cazul agregării parităţilor.
G. Utilizarea parităţii puterii de cumpărare în cadrul Uniunii Europene
Introducerea monedei unice presupune determinarea agregatelor economice în EURO. Programele de comparare vor continua deoarece: - nu toate ţările europene membre UE sunt şi membre ale Uniunii
Monetare; - comparaţiile sunt necesare şi între ţările UE şi cele din afara grupului; - utilizarea EURO nu înseamnă automat şi eliminarea diferenţelor de
preţ existente între statele membre. În compararea unei ţări UE şi o ţară din afara grupului se poate aplica standardizarea puterii de cumpărare. Paritatea standard a puterii de cumpărare (PSP) reprezintă o noţiune specifi că lucrării de comparare internaţională a Produsului Intern Brut. Numerarul este astfel defi nit încât pentru fi ecare agregat de cheltuieli totalul UE obţinut din transformarea valorilor exprimate în moneda naţională cu ajutorul parităţilor este egal cu totalul UE pentru acel agregat exprimat în euro. Numerarul este o noţiune specifi că lucrării de comparare internaţională a Produsului Intern Brut. De regulă, valuta naţională a unei ţări este aleasă drept
Romanian Statistical Review - Supplement nr. 4 / 2019136
numerar pentru exprimarea cheltuielilor reale şi a parităţilor puterii de cumpărare. Numerar poate fi , de asemenea, media pentru un anume grup de ţări, procedeu frecvent în Uniunea Europeană până la introducerea euro. Calculul practic presupune parcurgerea mai multor etape, astfel: - parităţile sunt iniţial calculate, făcându-se referinţă la o ţară de bază; - valorile parităţilor se standardizează prin recalculare ca media
geometrică simplă a parităţilor pentru statele membre UE; - se calculează ��� ���������� �"
%%,%.� � ����� $%&
�������� �0����� ��������������� ���� ����3�
(unde PIBPPC este PIB real recalculat pe
baza parităţilor standard de la punctul anterior şi �������� �0����� ��������������� ���� ����3�� �"
",�%.� ��
.
- factorul de conversie se determina ca raport ��� ���������"���������� ������������ � �"
%%,%.� �#� �"
",�%.� �
3�� �� �� ����� � ��� � �������� "������� $%&� ������� �� �� �M ��� - şi este utilizat în stabilirea valorilor PIB transformate cu ajutorul parităţilor standard în valori comparabile;
- împărţind valorile obţinute la punctul anterior la PIB exprimat în valuta naţionala se obţin parităţile standard ale puterii de cumpărare (PSP).
Concluzii
Pe baza elementelor cuprinse în acest articol, rezultă că problema comparabilității internaționale este una de actualitate. Organismele economice
internaționale, care lucrează pe baza unor principii, țin seama în aplicarea
regulamentelor lor de poziția pe care o ocupă fi ecare țară în ierarhia mondială.
Studiile de comparabilitate internațională se pot realize utilizând o serie
de modele econometrice, care țin seama de variabilele unitare considerate
pentru toate țările respective. Modelul cel mai expresiv și usor de aplicat este
cel al utilizării parității puterii de cumpărare. Pe plan European, în cadrul
uniunii, introducerea monedei unice presupune determinarea agregatelor
macroeconomice în această monedă, realizându-se astfel o comparabilitate
rapidă și corectă, atâta timp cât țările member exprimă întreaga activitate în
moneda Euro.
Bibliografi e
1. Anghel, M.G., Anghelache, C., Samson, T. (2017). The Main Methodological
Elements Regarding the Comparability of the Result Indicators, Romanian Statistical Review, Supplement, no. 7, 45-59
2. Anghelache, C., Mitruţ, C., Voineagu, V. (2013). Statistică macroeconomică. Sistemul Conturilor Naţionale, Editura Economică, Bucureşti
3. Anghelache, C. (2009). Indicatori macroeconomici utilizaţi în comparabilitatea
internaţională, Conferinţa a 57-a „Statistica – trecut, prezent şi viitor”, ISBN 978-
90-73592-29-2, Durban, articol cotat ISI
4. Anghelache, C. (2008). Tratat de statistică teoretică şi economică, Editura
Economică, Bucureşti
Revista Română de Statistică - Supliment nr. 4 / 2019 137
5. Anghelache, C. şi colaboratorii (2007). Analiză macroeconomică Teorie şi studii
de caz, Editura Economică, Bucureşti
6. Goodwin, N.R. (2008). Macroeconomics for the Twenty-First Century, Tufts
University, series GDAE Working Papers no. 03-02
7. Chetty, R. (2009). Suffi cient Statistics for Welfare Analysis: A Bridge Between
Structural and Reduced-Form Methods. Annual Review of Economics, Annual
Reviews, 1(1), 451-488
8. Farhi, E., Werning, I. (2016). A Theory of Macroprudential Policies in the Presence
of Nominal Rigidities. Econometrica, 84, 1645-1704
9. Ftiti, Z. (2010). The Macroeconomic Performance of the Inflation Targeting Policy: An Approach Based on the Evolutionary Co-spectral Analysis, Economic
Modelling, 27 (1), January, Elsevier 10. Öllera, L.E., Teterukovsky, A. (2007). Quantifying the quality of macroeconomic
variables. International Journal of Forecasting, 23 (2), April–June, 205-217
Romanian Statistical Review - Supplement nr. 4 / 2019138
THE INTERNATIONAL COMPARABILITY OF THE MACROECONOMIC PERFORMANCE
Ștefan Gabriel DUMBRAVĂ Ph.D Student ([email protected])Bucharest University of Economic StudiesAlexandra PETRE (OLTEANU) PhD Student ([email protected])
Bucharest University of Economic StudiesCristian OLTEANU PhD Student ([email protected])
Bucharest University of Economic StudiesTudor SAMSON Ph.D Student ([email protected])
Bucharest University of Economic Studies
Abstract The issue of international comparability is a major one, in the sense that, on the basis of the macroeconomic results indicators, it is possible to make hierarchies, rankings on the basis of which comparative studies can be carried out. International comparisons of the economic development of diff erent
countries can be made on the basis of physical expression indicators such as
per capita consumption, investment, quality of life indices, etc. In order to draw
conclusions on the level of development, a reference country is chosen and the
levels recorded by the other countries compare to the level taken as a benchmark.
In the context of the international comparison of GDP, it is concluded that, for
several reasons, the exchange rate can not always be considered as a decisive
factor, because the changes in the age gap also depend on many internal factors.
Under these circumstances, comparability is based on the purchasing power
parity of the currencies of the comparing countries. Purchasing power parity
expresses the number of units of currency required to buy the same volume of
goods in one country. Within the European Union there is a hierarchy / ranking
program for member countries after this indicator. In this respect, the countries
of the European Union are divided into three groups according to the value of the
purchasing parity index. In this hierarchy, Romania is in the second group with
countries such as Bulgaria, Slovenia, Albania, Austria, Belarus, Moldova or
Ukraine. On a broader European level, the hierarchies of states are also based
on purchasing power parity. In such international comparisons or identifi ed and
statuated some methods of comparing such indicators as binary comparison,
star comparison, asterisk method, Geary - Khanis method, Contry - Product
Dummy method, EKS method, purchase parity usage and others.
Keywords: macroeconomic results indicators, international
comparability, performance, purchasing power parity, process
JEL Classifi cation: C10, C82, E31
Revista Română de Statistică - Supliment nr. 4 / 2019 139
Introduction
In the study, the authors focused on identifying the possibilities for
international comparison of the macroeconomic performance of diff erent
countries. International macroeconomic comparisons are not only a theoretical
aspect, but a practical problem related to comparisons between countries,
which is a concern of national and international statistical institutions. The
authors analyze the particularities of purchasing power parity in international
comparisons. The comparison can be made using the macroeconomic indicators
of results calculated in the National Corrupt System, because countries using
this macroeconomic evidence system calculate the macroeconomic indicators
according to a unitary concept and methodology. It is further emphasized that,
in order to achieve the international comparison, three phases are basically
followed, namely: each country’s GDP is decomposed into a number of
primary expenditure groups; within each primary group, selected prices are
set to ensure comparability and representativeness; national values of primary
groups are transformed into internationally comparable values through
purchasing power parity. In ensuring comparability, there are two programs,
one for European cooperation and another for the international comparison
program. The authors attach importance to presenting the main international
comparison of indicators. These methods can be easily used to make practical
amalgamations based on existing data.
Literature review
Anghel, Anghelache and Samson (2017) addressed a number of
issues related to the methodology for comparability of outcome indicators.
Anghelache, Mitruţ and Voineagu (2013) and Anghelache et al (2007) analyzed and presented fundamental elements of macroeconomic statistics. Anghelache (2009) paid special attention to the calculation of the main macroeconomic indicators you use in international comparability. Anghelache (2008) presented the basics of economic statistics as well as the calculation indicators used in economic activities. Goodwin (2008) analyzed aspects of macroeconomics specifi c to the present century. Chetty (2009) studied welfare analysis methods. Farhi and Werning (2016) investigated macroprudential policy theories. Ftiti (2010) analyzed elements of the macroeconomic performance of infl ation targeting policy. Öllera, and Teterukovsky (2007) studied ways of quantifying the quality of macroeconomic variables.
Romanian Statistical Review - Supplement nr. 4 / 2019140
Research methodology, data, results and discussions
The comparing macroeconomic aggregates of results raises issues
related to monetary expression, coverage, calculation methodologies, and the
use of macroeconomic performance indicators..
• The using purchasing power parity in international comparisons
The comparison of macro indicators is not just a theoretical issue
but, to a greater extent, a practical problem related to comparisons between
countries in the concerns of national and international statistical institutions.
International comparisons of the economic development of diff erent
countries can be made on the basis of physical expression indicators by comparing
production and consumption per capita with a number of commodities. The
comparison involves the compilation of two lists: one with the comparison
countries and the commodities that compare. It is necessary for these products
to be representative of all countries comparing and qualitatively similar.
Next, to draw conclusions on the level of development, a reference
country is chosen and the levels reached by the other countries compare with
the country level taken as a benchmark. For this purpose, the relative co-
ordination sizes (Y) are used.
The comparison can be made using the macroeconomic indicators
of results calculated in the SCN. The international comparison of the level,
structure and pace of development based on the macroeconomic results of the
results requires the solving of two key issues, namely: achieving comparability
of the indicators in terms of their content, their scope and ensuring comparability
of the indicators from the point of view of the monetary units in which it is
expressed.
The countries using SCNs determine macroeconomic indicators
according to unit design and methodology. The assessment of macroeconomic
indicators in a single currency can be done by the following processes:
- Based on the offi cial exchange rates, the exchange rate is the price
to be paid on the foreign currency market to obtain a unit of national currency
or the amount of foreign currency received in exchange for a unit of national
currency.
In the context of international comparison of GDP, the theoretical
and methodological debates concluded that for several reasons the exchange
rate can not be considered a decisive element in ensuring the comparability of
the indicators in absolute terms and can not be used to convert the synthetic
indicators from the national currency because : does not refl ect the real GDP structure; is infl uenced by internal and international factors, causing fl uctuations over short periods of time; there are multiple currency exchange rates.
Revista Română de Statistică - Supliment nr. 4 / 2019 141
This is a relatively easy procedure to apply, but which can negatively
infl uence real values, as exchange rates usually do not refl ect the purchasing power of currencies. This is due to the fact that, on the one hand, the fi xing
of the exchange rate is infl uenced only by a small number of goods, namely those which are the subject of trade relations between the respective countries, and, on the other hand, the exchange rate is strongly infl uenced by various speculative operations on the stock market. It follows that the expression of indicators in a single currency based on exchange rates is not a recommended method for making the most accurate international comparisons on the levels, structures and pace of development of diff erent countries.
On the basis of the purchasing power parity (PPC) evaluation of
the currencies of the countries that compare. This method of assessing
macroeconomic indicators in one currency is more appropriate for use than
the offi cial exchange rate method. It consists in using price indices calculated
on the basis of the prices of goods and services in the country making the
comparison and price calculations of the same material and service goods in
the country whose currency has been decided to be used for the expression of
macroeconomic indicators.
PPC expresses the number of units of currency required to buy in a
country the same volume of goods that can be obtained with a monetary unit
of the benchmark country or a monetary unit common to a group of countries.
PPCs can be calculated for total Gross Domestic Product but also for diff erent
aggregation levels (eg gross fi xed capital formation).
The basis of these price indices is based on the general principles of
calculating price indices. It is known that the aggregate price index is calculated
using current and basic (p1 and po) commodity prices, and as weights for both
periods either the current (q1) or baseline (q0).
Purchasing power parity is the average ratio between the prices in the
comparison country and those of the country taken as the basis of comparison,
to a large number of categories of expenditure in the GDP composition, broken
down by global elements (fi nal consumption of the population, collective
government consumption, gross fi xed capital formation, changes in stocks and
net exports).
PPP determination based on the prices of goods and services contained
in a sample of comparable items that are selected to represent a well-defi ned
classifi cation and to be representative of the pattern of consumption of the
countries participating in the comparison can be assimilated to a price index
Territorial type.
In order to achieve the international comparison, three phases are
followed, in principle:
Romanian Statistical Review - Supplement nr. 4 / 2019142
- The GDP of each country is decomposed into a number of primary
spending groups.
- Within each primary group, there are prices for selected items that
should have two essential properties: comparability (which means that the
factors infl uencing actual price formation are identical in the comparing countries) and representativeness (meaning selected articles are characteristic of expenditure model and have a signifi cant share in the primary group of
the competing countries). Gathered (observed / observed) prices are used to
obtain price reports for individual goods and services, then price reports are
averaged to obtain unmatched parities at the primary group level; Finally, using
as weights the structures of the country’s GDP spending that are compared in
the weighted geometric mean formula, a PPP is obtained for each level of
aggregation - up to GDP.
- National values of primary groups are translated into „international
comparable” values through purchasing power parities. The obtained results
are real values, to distinguish them from the nominal values (expressed in
national currency).
Adaptation implies that the prices of the products of the two countries
are weighted either with the quantities of a country (for example, country A)
or the quantities of the other country (or country of reference, country B).
Obviously, there are some diff erences between the results obtained
using the two price indices due to the application of diff erent weights. For the
purpose of eliminating these infl uences, a Fischer price index is calculated as the geometric mean of the Paasche and Laspeyres type indices. Regardless of the variant (Paasche, Laspeyres or Fischer), the calculation of such a general index is a very diffi cult operation to be performed
only by national and international statistical bodies.
There are several methods of approaching and solving comparative
procedures to PPP, some of which are being promoted by international
statistical institutions. The main methods and programs will be presented
below.
• European Comparison Program
The fi rst systematic comparison actions in Europe have been carried
out since 1980 under the auspices of the Economic Commission for Europe in
the context of the regionalization of the International Comparison Program.
Romania participated in the work of the European Comparison Program in the
reference years: 1990, 1993, 1996 and 1999.
For the 1993 round, the participating countries were divided into three
groups: Group I covering the countries of Western Europe, Group II including
Revista Română de Statistică - Supliment nr. 4 / 2019 143
Central and Eastern European countries and Group III comprising the new
independent states of the former Soviet Union.
These were bilateral comparisons, each country being compared
directly with Austria. So both comparisons between Group II countries and
comparisons between Group I countries (and other OECD countries) and
Group II countries were made through Austria, the „bridge country”, and in
Group III, a model of multilateral comparison was adopted.
The 1993 comparison round coincided with the political changes of
Central and Eastern Europe, with the dissolution of the former Soviet Union.
23 countries were in Group II.
Romania has been included in a bilateral comparison program:
Romania - Moldova and Romania - Moldova through Austria (bridge country).
The 1996 benchmarking round was organized as a series of cross-
country multilateral comparisons, this time being a distinction between three
groups whose geographic structure diff ers from the 1993 round.
- Group I: the 15 EU Member States (since 1996): Iceland, Norway,
Switzerland, Turkey, Israel, the Russian Federation and the Slovak Republic,
plus Poland, Hungary and the Czech Republic. Direct links with the other
groups were provided by comparison with Austria (with Group II) and Turkey
(with Group III countries);
- Group II: Estonia, Latvia, Lithuania, Bulgaria, Romania, Slovenia,
Croatia, the former Yugoslav Republic of Macedonia, Albania, Austria,
Belarus, Moldova, Ukraine and the Russian Federation.
- Group III (CIS Group): Russian Federation, Armenia, Azerbaijan,
Georgia, Mongolia, Kazakhstan, Kyrgyz Republic, Tajikistan, Turkmenistan,
Uzbekistan, Turkey, Belarus and the Republic of Moldova.
Without a change in methodology over the previous round, 1996,
EUROSTAT decided on the composition of three groups coordinated by
a Member State of the European Union, the membership of countries in a
group or another being judged by: market similarity (for both maximizing the
number of articles evaluated in at least two of the partner countries as well as
their geographical proximity The coordinating countries for the 1999 round
are: Finland, Austria and Italy Romania is part of the coordinated group of
Italy alongside Bulgaria, Cyprus, France, Greece, Portugal, Spain.
• International Comparison Program (PCI)
PCI is an internationally organized work that allows international economic
comparisons to be made by eliminating price diff erences across countries.
PCI work has been organized by the United Nations Statistical
Commission. Six rounds marked the existence of the project, started in 1968;
Romanian Statistical Review - Supplement nr. 4 / 2019144
starting with the third phase, the project gained program features once every fi ve years. The comparison has two main stages, namely: - The GDP of each country is decomposed into a number of primary groups. - Decomposition of expenditure by primary groups is considered a classifi cation whose purpose is to provide more accurate estimates of higher-level expenditure aggregates resulting from the following: the coverage of all expenditure aggregates is ensured; the dispersion of individual price reports is lower in the primary group than among the groups within the higher aggregates; the calculation of weighted averages of parities can be done at a relatively detailed level. - Primary Gross Gross Domestic Product (GDP) is the primary level for which purchasing power parities are calculated. Primary group is the most detailed entity of GDP expenditure classifi cation. The primary group is defi ned as the homogeneous product group from which a number of articles will be selected to be tracked through price surveys, representative articles for both the product category and the country involved in the comparison. In delimiting primary groups, it is necessary to observe as principles: homogeneity, interpreted in terms of the dispersion of individual price reports, and that expenditure data for primary groups are used as weights. Thus, for example, GDP is decomposed into main aggregates of expenditures, and „individual consumption of households” is one of these. In turn, it is broken down into spending groups: „Foods, Drinks and Tobacco” is an example for which „Food” is a subgroup. Within these, there are categories of expenditure: for example, „Bread and cereals”. „Rice” and „Flour and other cereals” are primary groups. The representation of primary groups by means of individual articles means, in the case of rice, at least one of the specifi cations: rice, long grain; rice, medium grain; rice, long grain, pre-cooked; rice for quick cooking. Within each primary group, prices are collected for selected items that meet the comparability and representativeness properties. The collected prices are used to obtain price reports for individual goods and services; then, the price reports are averaged to obtain unmatched parities at the primary group level; Finally, using as weights the GDP expenditure structures of countries that are compared in the weighted geometric mean formula, a PPP is obtained for each level of aggregation - up to GDP. Nominal values are converted to real values through PPC. The main statistical information obtained from the comparison calculations is:
Revista Română de Statistică - Supliment nr. 4 / 2019 145
- Price reports (PPCs) used to determine volume reports; it is not just
a general purchasing power parity but a set of PPCs to defl ate diff erent groups
/ categories of goods and services;
- General and per capita indicators; volume fi gures can be used to
create regional, regional totals and determine the share of each country in total
(a country ranking may be established for a given country based on 100.0%);
- Comparison of the price level;
- Indicators of structure: for example, dividing the volume index
of population consumption into GDP volume index, determine the relative
volume index between two countries, which in the analysis can be interpreted
in such a way that it can be estimated that, relative to GDP , the relative volume
in households, or what is called „real consumption rate” is higher or lower in
one country than the other.
• The main international comparison of indicators
The statistical practice of SCN has confi rmed several comparisons, most notably: binary comparison, multiparallel comparison, star, Geary-Khanis method, Country-Product-Dummy method and others. A. Binary comparison is conducted between two countries and the analysis involves data other than the two countries participating in the program. In practice, the calculation steps are as follows: - The calculation of the Purchasing Power Parity at the primary group level is done in several ways, using, as appropriate: a) Laspeyres type indices:
nn
i
iBA
LBA RPMPPC
1
1
÷ø
öçè
æ= Õ=
(1)
where: n – number of articles representative of the base country „A” of the respective primary group
=LBA RPM average cost ratio of item „i” = PM i
A / /PM iB
b) Paasche type indices:
nn
i
iAB
PBA RPMPPC
1
1
÷ø
öçè
æ= Õ=
(2)
where: n = number of articles representative of Country „B” of the respective primary group;
Romanian Statistical Review - Supplement nr. 4 / 2019146
=LAB RPM the average price ratio of the item „i” = PM i
B / PM iA /
c) Fisher-type indices:
PBA
LBA
FBA PPCPPCPPC ´=
(3)
- Purchasing Power Parity calculation at aggregate level
a) using Laspeyres-type indices:
( )å
å =
=
´´=n
i
iA
iBAn
iA
LBA WPPC
W
PPC1
1
1 (4)
where:
n – the number of primary groups of the compared economic aggregate
PPCPPC iBA = for the primary group „i” of the base country „A”
WA = GDP expenditure for the primary group „i” of the base country
„A”
b) using Paasche-type indices:
( )
( )å
å
=
=
´=
n
i
iB
iAB
n
iB
PBA
WPPC
W
PPC
1
1
(5)
where:
iBA PPC = PPC of primary group „i”
Wb = GDP expenditures related to primary group „i” of country „B”
c) using Fischer indexes:
PBA
LBA
FBA PPCPPCPPC ´= (6)
B. Compararea de tip stea şi multilaterală
The star comparison is one in which a selected set of binary comparisons
is used to infer comparisons for other country pairs. Each country is compared
(directly) with the base country, and comparisons between any two countries are
the result of corresponding country reports with the base country.
Based on the price / quantity indices Ij / k and Il / k are the countries
participating in the comparison (j, k, l) using the star comparison method, the
Revista Română de Statistică - Supliment nr. 4 / 2019 147
index Ij / il is calculated as the ratio between the indices Ij / k and Il / k . The 1990s
and 1993s of the Group III PCE were binary star comparison, each country being
compared directly with Austria as a bridge country.
Multiple simultaneous comparison of more than two countries resulting
in valid relationships between all pairs of countries, thus verifying the circularity
or transitivity property. Interpretation of results, even for two countries, is based
on data from all countries participating in the exercise.
The elements that need to be taken into account for such a complex
approach to be accessible are: a similar level of development of national markets
an equivalent technical standard between countries; an adequate statistical
infrastructure to provide the price and expense (imputed) GDP data; a common
list of goods and services to be evaluated, a list of all participating countries, so
that it is ultimately equivocal.
C. The asterisk method is a simple procedure involving the
indication by each participating country, by means of an asterisk, of the
articles „representative of its consumption pattern. These may be: articles that
the country has proposed to compile the list for countries participating in the
comparison program; widely available articles on the domestic market and
easily identifi able for evaluation - the decision is based on the relatively large volume of sales of the item concerned; articles present in the calculation of the national consumer price index. The approval of a representative article, signaled by an asterisk (*), generates the price observations included in the calculation by the fact that the Laspeyres-type indices are calculated taking into account prices for those items representative of the basic country, while the Paasche price indices are determined on the basis of the articles representative of the partner country in the comparison program. D. G – K Method (Geary – Khanis)
It is a multilateral comparison method by which the expenditure categories denominated in national currency are revalued through purchasing power parities calculated as weighted arithmetic averages of prices prevailing in a region. The results obtained meet the principle of transitivity and additivity and check the reversibility test of territorial indices. The calculation is based on the purchasing power parities of the primary groups - considered fi ctitious prices - and the fi ctitious quantities calculated as a ratio between the nominal values and the corresponding parities. By pointing to international prices with pij and global parities with fj, their values are determined by solving a system of equations:
Romanian Statistical Review - Supplement nr. 4 / 2019148
"����������� �������� ��������������"���������� ���������0��#�
� � � � �
�=
/
=
/
�/�//�/� >>�&�
- -
E ��Q�-)�3)�/�Q�-��=� ���.�� (7)
� � � � � � �
�3
�
3
�
�/�/�/�/ >&>�- -
E ��Q�-)�3)�/�Q�-��=��� ��
!�"������0����� ���0������������������������������ ����������
(8)
The international primary price level (i) represents an arithmetic mean
weighted by the fi ctitious quantities of fi ctitious prices adjusted to the overall purchasing power parity for each country where: The notes in the previous formulas indicate: Pij = the fi ctitious price for the primary group (s) in country (j) (in fact, purchasing power parity for the primary group (i) Qij = the fi ctitious amount Fj = the overall purchasing power parity for the country’s total GDP (j)
�
�=
��/� >?
-
�Q���� � � ��� � ������� � �� �� 0�����F� ������� = total quantity for all countries - for primary group (i)
N = number of countries M = number of primary groups The global purchasing power parity of the country (j) is calculated as the ratio of the total cost expressed in international prices and the total cost in national prices. E. < Country – Product – Dummy (CPD) > Method
It is a generic star comparison model, which uses regression analysis for transient pricing comparisons for each primary group. Basic data for a given category consists of all available prices for diff erent specifi cations for all countries participating in the comparison. Each price is judged according to the country of observation and the article to which it refers. The natural logarithm of the price becomes the dependent variable in a linear regression equation in which two sets of binary variables are independent variables: one referring to countries (less the cash country), the other to articles. Transitional purchasing power parities for primary groups result from calculating the regression coeffi cients of binary variables representing countries.
F. EKS Method
This process is a multilateral method of international comparison of
GDP developed by Elteto, Koves and Szulef. Calculate the price index (Fisher
multilateral index) as the unplanned geometric mean of all Fisher indexes that
can be determined for n countries.
Revista Română de Statistică - Supliment nr. 4 / 2019 149
( )
nn
k,ll
jkEKSk/j FFFPPC
1
j1
l/l.
2
j/k /
úúú
û
ù
êêê
ë
é= Õ
¹=
where: Fk/k = 1 (9)
For example, in an application covering 4 countries and noting the
Fisher indexes for each pair of countries with FB/A, FC/A, …, type indices
EKS (FB/A*,FC/A*,… ) are calculated according to the formula:
( )[ ] ( )[ ]{ } 4/12
B/AC/BD/AC/DC/AC/A FFFFFF ´´´´= (10)
The method applies to the calculation of primary group parities, but
also to parity aggregation.
G. Use of purchasing power parity within the European Union
The introduction of the single currency involves the determination of
economic aggregates in the euro. Comparison programs will continue because:
- not all EU Member States are also members of the Monetary Union;
- comparisons are also needed between EU and non-EU countries;
- the use of EURO does not automatically mean the elimination of
existing price diff erences between Member States.
In comparing an EU country and a country outside the group, purchasing
power standardization can be applied.
Purchasing Power Parity Standard (PSP) is a specifi c notion of the
International Gross Domestic Product comparison.
Cash is defi ned so that for each aggregate of expenditure the total EU
obtained from the conversion of the values expressed in national currency by
parities is equal to the total EU for that aggregate expressed in euro.
Cash is a notion specifi c to the international comparison of the Gross
Domestic Product. As a rule, the national currency of a country is chosen as cash
for expressing actual expenditure and purchasing power parities. Cash can also
be the average for a certain group of countries, often in the European Union, until
the introduction of the euro.
Practical computation involves going through several stages, as follows:
- the parities are initially calculated, referring to a base country;
- the parity values are standardized by recalculating as a simple
geometric mean of parities for EU Member States;
- is calculated ��� ���������� �"
%%,%.� � ����� $%&
�������� �0����� ��������������� ���� ����3�
(where PIBPPC is real GDP recalculated
based on the standard parities from the previous point and �������� �0����� ��������������� ���� ����3�� �"
",�%.� ��
.
Romanian Statistical Review - Supplement nr. 4 / 2019150
- the conversion factor is determined as a ratio ��� ���������"���������� ������������ � �"
%%,%.� �#� �"
",�%.� �
3�� �� �� ����� � ��� � �������� "������� $%&� ������� �� �� �M ���
and is used to establish the GDP values converted using standard parities in
comparable values;
- by dividing the values obtained at the previous point to the GDP
expressed in the national currency, the standard purchasing power parities (PSP).
Conclusion
Based on the elements contained in this article, it emerges that the issue of
international comparability is a matter of topicality. International economic bodies,
working on principles, take into account in the application of their regulations the
position of each country in the world hierarchy. International comparability studies
can be carried out using a series of econometric models that take into account unit
variables considered for all countries concerned. The most expressive and easy to
apply model is the use of purchasing power parity. At the European level, within
the union, the introduction of the single currency involves the determination
of macroeconomic aggregates in this currency, thus achieving rapid and fair
comparability, as long as the member countries express all their activity in the euro.
References
1. Anghel, M.G., Anghelache, C., Samson, T. (2017). The Main Methodological
Elements Regarding the Comparability of the Result Indicators, Romanian Statistical Review, Supplement, no. 7, 45-59
2. Anghelache, C., Mitruţ, C., Voineagu, V. (2013). Statistică macroeconomică. Sistemul Conturilor Naţionale, Editura Economică, Bucureşti
3. Anghelache, C. (2009). Indicatori macroeconomici utilizaţi în comparabilitatea
internaţională, Conferinţa a 57-a „Statistica – trecut, prezent şi viitor”, ISBN 978-
90-73592-29-2, Durban, articol cotat ISI
4. Anghelache, C. (2008). Tratat de statistică teoretică şi economică, Editura
Economică, Bucureşti
5. Anghelache, C. şi colaboratorii (2007). Analiză macroeconomică Teorie şi studii
de caz, Editura Economică, Bucureşti
6. Goodwin, N.R. (2008). Macroeconomics for the Twenty-First Century, Tufts
University, series GDAE Working Papers no. 03-02
7. Chetty, R. (2009). Suffi cient Statistics for Welfare Analysis: A Bridge Between
Structural and Reduced-Form Methods. Annual Review of Economics, Annual
Reviews, 1(1), 451-488
8. Farhi, E., Werning, I. (2016). A Theory of Macroprudential Policies in the Presence
of Nominal Rigidities. Econometrica, 84, 1645-1704
9. Ftiti, Z. (2010). The Macroeconomic Performance of the Inflation Targeting Policy: An Approach Based on the Evolutionary Co-spectral Analysis, Economic
Modelling, 27 (1), January, Elsevier 10. Öllera, L.E., Teterukovsky, A. (2007). Quantifying the quality of macroeconomic
variables. International Journal of Forecasting, 23 (2), April–June, 205-217