Applying of Long-Memory Processes Modeling to the Prices of Russian Stocks

In this final qualifying work we consider the definition of long memory as property of time series and models that allow it to be taken into account. We estimate such models (ARFIMA – GARCH, ARFIMA – EGARCH, ARFIMA – TGARCH, ARFIMA – APARCH) and compare them between themselves and with other models of ARMA class by fit quality and predictive power basing on information criteria and root-mean square error (RMSE) respectively. This modeling is applied to the logarithmic returns of several Russian «blue chips». The obtained results indicate that at the present time the long-memory processes modeling allows us to identify shares of some top Russian companies in the stock market for which the forecasting of its returns with using of these methods is more effective than performance of other methods, in particular of «random walk» model.