20254

Tests de racines unitaires et performance prévisionnelle des modèles AR: application sur les variables du transport en France

Mourad Mahmoud نويسنده

113

131

19

This paper examines the nature ofthe non-stationarity in twenty-five monthly time series that cover the different sectors of the transportation in France. Tests of unit roots have been used to discriminate the deterministic or stochastic of the trend and the seasonality. A sensitivity in the answers of the DHF procedure has been demonstrated if we vary the AR order p of12 to 24. These answers sometimes contradictory drive us to consider four types of autoregressive models (AR) including trend and / or seasonal factor. The optimal order for the different models has been identified while using the automatic criteria FPE, AIC, RIC and HQ. The prediction performance of the estimated models has been measured with the help of the RMSE (Mean Root Square Error) and MAPE (Mean Absolute Percentage Error) criteria. Among the AR models chosen, there are eight time series having a deterministic seasonal factor, seven for a stochastic seasonal factor, five for a deterministic trend and seasonal factors and finally five time series for a stochastic trend and seasonal factors

1204283