Nieliniowa identyfikacja rzędu autozależności w stopach zmian indeksów giełdowych

Abstract

The mutual information coefficient is one of the most important generalizations of the Pearson correlation coefficient. Its advantage is that it is able to measure all kinds of dependencies, also nonlinear ones. Like the Pearson correlation coefficient, the mutual information coefficient may be applied to a single time series in order to identify serial dependencies. In this paper the log-returns of the selected Polish and foreign stock indices have been analyzed. By comparing the results obtained for the raw returns and the residuals of the fitted ARMA-GARCH models, the nature of the identified dependencies has been determined. Moreover, the bootstrap procedure has been applied to verify significance of the mutual information coefficients and, in consequence, to determine the number of lags in the analyzed series. In the most investigated indices, nonlinear dependencies different from the ARCH effect have been detected and lag = 1 was the dominant time delay in such situations.