Tests of Noncausality under Markov Assumptions for Qualitative Panel Data

For many years, social scientists have been interested in obtaining testable definitions
of causality (Granger [ 1 2 ] , Sims [ 2 8 ] ) .Recent works include those of Chamberlain [7]
and Florens and Mouchart [ a ] .The present paper first clarifies the results of these latter
papers by considering a unifying definition of noncausality. Then, log-likelihood ratio (LR)
tests for noncausality are derived for qualitative panel data under the minimal assumption
that one series is Markov. LR tests for the Markov property are also obtained. Both test
statistics have closed forms. These tests thus provide a readily applicable procedure for
testing noncausality on qualitative panel data. Finally, the tests are applied to French
Business Survey data in order to test the hypothesis that price changes from period to
period are strictly exogenous to disequilibria appearing within periods.

1. INTRODUCTION AND SUMMARY
FORMANY YEARS, social scientists have been interested in obtaining a testable
definition of causality. Earlier contributions include the works of Simon [27],
Strotz and Wold [29]. Alternative definitions of causality which heavily rely on
the stochastic nature of the variables and the principle that the future does not
cduse the past were then proposed and studied by Granger [12] and Sims [28].
Recently, Chamberlain [7] and Florens and Mouchart [8] extended these latter
definitions to possibly nonstationary nongaussian processes. The present paper
first clarifies the results of these two recent papers, and second, derives some
tests for noncausality under minimal assumptions on the process generating the
qualitative panel data, and finally, applies the tests to an empirical example.