Tests de causalité sur données de panel : une application à l'étude de la causalité entre l'investissement et la croissance

摘要

Causaliy Testing in Panel Data with an Application to the Question of Investment and Growth by Diana Weinhold This paper develops an approach for panel causality analysis which allows for flexibility in the causal relationship from unity to unity, allowing estimation of distribution of causality in a possibly heterogeneous panel. Two possible estimation techniques for "mixed fixed and random" coefficients model are explored. These methods both allow the coefficients of an orthogonalized causal variable to vary randomly while avoiding many of the problems associated with random coefficients and dynamics in panels. Using simulations to develop a method for interpreting the estimated distributions, it is then possible to predict the probabilities of causality associated with the panel. This approach is then applied to the question of causality between investment and growth in a panel of countries. The paper find that there is great instability and feedback among countries in this relationship. Therefore, it is clear that the restricting assumptions of traditional pooling models can be inappropriate, especially in simple models. Because the analysis signals this problem, the paper's results further underline the need for the interpretive flexibility that this proposed causality test imparts.