Causality tests in dynamic macroeconometric models.

Summary This thesis is devoted to the study of granger causality tests. One of the important characteristics of these tests is the assumption of stationarity of the series. This work considers non-stationary (integrated series of order one), admitting a finite vector autoregressive (var) representation whose coefficients are constant in time. In this case, the absence or presence of cointegration influences the causality tests. In the presence of cointegration, toda and philips propose a new method, consisting in constructing statistics in an error correction model (ecm), however, the formulas of their statistics are complicated. We then propose simpler formulas. They are built from the ecrm model. They converge to a chi-square distribution. Moreover, this thesis extends the research to series with deterministic tendency of degree one. This leads to propose three definitions of cointegration: stochastic cointegration, deterministic cointegration and global cointegration. Our work considers three ecm forms, derived from a finite var model, according to the level of consideration of the constraints on the coefficient of the time term. This work performs: (1) the comparison of these three ecm forms on the theoretical economic level, (2) the study of the consequences of the bad estimation of the johansen ecm model (based on the specific deterministic trend), (3) the proposal of the statistics of the global cointegration test, (4) the modifications of the formulas of the proposed statistics concerning the causality tets. Various simulations are also carried out, in order to compare the considered statistics.