Marches financiers, memoire longue et processus chaotiques
Institution:
Paris 10Disciplines:
Directors:
Abstract EN:
Despite the vast literature concerning the efficient capital market hypothesis, none unambiguous conclusion seems to emerge. This absence of clear conclusion can be the result of the low power of tests traditionally used. Thus, this thesis has been consecrated to the study of two approaches, longterm memory and chaotic processes, in order to reexamine the efficient capital market hypothesis. The first part had proposed a critical study of the efficient market theory and of the rationality of investors. Inadequacies of traditional tests has led us, in a second part, to the analysis of two different alternatives which appear particularly consistent in order to study efficient market theory: long-term memory processes and chaotic processes. These two approaches have been tested in a third part devoted to an empirical study of stock returns series. We showed that the deterministic chaotic hypothesis is rejected. However, the long-term memory hypothesis constitutes an hypothesis that is generally accepted. Arfima processes have been estimated and their predictive capacities have been studied and compared to the random walk model. We have showed that arfima processes have led to better forecasts than naive ones. Finally we have tempted to model and to predict the risk by the mean of a new model which takes into account both long-term memory in stock returns and variability of their volatility. Results have suggested a possible prediction of volatility. Both predictability of stock returns and risk seem to be hardly coherent with weak-form efficient capital market hypothesis.
Abstract FR:
Malgre le nombre considerable de travaux visant a tester l'hypothese d'efficience des marches financiers, aucune conclusion claire ne semble ressortir. Cette absence de resultats unanimes nous a semble provenir en partie de la faiblesse des outils econometriques utilises. Cette these s'est alors attachee a reexaminer la theorie de l'efficience des marches financiers sous l'angle double des processus a memoire longue et des processus chaotiques. La premiere partie de la these a propose une etude critique de la theorie de l'efficience des marches financiers ainsi que de son concept indissociable qu'est la rationalite des investisseurs. Les insuffisances attachees aux approches traditionnelles nous ont conduit a etudier, au sein d'une deuxieme partie, deux alternatives qui nous sont apparues particulierement pertinentes pour apprehender l'efficience : les processus a memoire longue et les processus chaotiques. Ces deux approches ont ensuite ete testees au cours d'une troisieme partie consacree a l'etude empirique des rentabilites des indices boursiers des pays du g7. Nous avons montre que si l'hypothese de chaos deterministe ne pouvait etre retenue, la memoire longue constituait quant a elle une hypothese generalement acceptee. La qualite des estimations des processus arfima a ete apprehendee au travers de leurs capacites predictives. Nous avons a cet egard montre que les processus arfima conduisaient a de meilleures previsions qu'un simple modele de marche aleatoire. Nous avons enfin cherche a modeliser et a prevoir le risque au travers d'une nouvelle modelisation prenant en consideration non seulement la dynamique de long terme des series de rentabilites mais egalement la variabilite de leur volatilite. A nouveau, nos resultats ont montre que la volatilite pouvait etre prevue. La double previsibilite des rentabilites et du risque nous parait difficilement conciliable avec l'hypothese d'efficience des marches au sens faible.