Tests de non-stationnarité : application au PIB réel
Institution:
DijonDisciplines:
Directors:
Abstract EN:
Since the seminal paper of nelson and plosser (1982), it has been widely admitted that the presence of a unit root in the stochastic process generating a time series has major implications in prevision and macroeconomic modelisation. Thus if a macroeconomic variable is trend stationary, then short-run shocks have only transitory impact on the long run evolution of the series. This behavior is consistent with traditional theories of business cycle. However, if a macroeconomic variable is difference stationary, then short-run shocks affect the level of the variable permanently. This is more compatible with real business cycle models of equilibrium output. We therefore present tests elaborated by dickey and fuller (1979, 1981), said and dickey (1984) and more recently by phillips and perron (1988). However all these procedures postulate that the deterministic component of the series follows a linear trend. Ouliaris, parck and phillips (1989) and perron (1989) show the pitfalls of such a postulate on "standard" test results and propose new tests where the deterministic component follows respectively a polynomial trend or a piecewise linear trend under the additional hypothesis of exogenous major shocks. We then suggest an empirical testing strategy which reduces the risk to reach false conclusion due to the misspecification of the deterministic component and we apply it to the analysis of the french real gnp. Some flexible specifications which allow us to reject the unit root hypothesis are then presented. These specifications may be considered as giving approximations of the true process generating the gnp series but can be also useful for other time series.
Abstract FR:
Depuis l'etude fondamentale de nelson et plosser (1982), on admet largement que la presence d'une racine unitaire, dans le processus stochastique engendrant une serie chronologique, a des implications majeures en matiere de prevision et de modelisation macroeconomique. Se pose alors le probleme de l'elaboration de procedures de test fiables qui permettraient la detection de telles racines unitaires et, par consequent, la determination de la nature de la non stationnarite - deterministe ou stochastique - caracterisant les series chronologiques macroeconomiques. De nombreuses procedures de test sont ainsi presentees : celles de dickey et fuller (1979, 1981), said et dickey (1984) et plus recemment celle de phillips et perron (1988). Cependant toutes ces procedures supposent que la composante deterministe de la serie consideree suit une tendance lineaire. Or les dangers de la mauvaise specification de cette composante deterministe ont ete mis en evidence par ouliaris, park et phillips (1989) et perron (1989) qui proposent des procedures de test ou elle suit respectivement une tendance polynomiale et une tendance lineaire par morceaux sous l'hypothese additionnelle de chocs majeurs exogenes. Nous avons tente de mettre au point une strategie empirique permettant de reduire le risque de parvenir a des conclusions erronees a la suite a une mauvaise specification de la composante deterministe et nous l'avons appliquee a l'etude de la non stationnarite caracterisant le pib reel marchand en france. Nous avons ainsi mis en evidence des specifications flexibles permettant de rejeter l'hypothese nulle de la racine unitaire fortement etayee dans la litterature empirique. Ces specifications peuvent etre considerees comme des approximations du vrai processus engendrant la serie du pib reel mais peuvent egalement se reveler utiles dans l'etude d'autres series chronologiques.