Dynamic Model Average for Stock Index Return

Abstract

The classical AR, MA, ARIMA, ARCH and GARCH models only carry out modeling and prediction on a single time series variable, ignoring the influence of other exogenous variables. It usually results in a large deviation of the prediction results. In view of the above problem of the models, this paper considers the effect of exogenous variables on the basis of the autoregressive model to make the model more practical. For the model uncertainty, this paper introduces in detail an efficient dynamic model average based on DMA, which adds the grid search of parametric forgetting factor, so as to make eDMA prediction more accurate. In the empirical part, the influence of macroeconomic factors and economic policy factors on the stock market is comprehensively considered for modeling and use eDMA method for parameters estimation comparing it with the results of OLS and BMA. Model average method fitting results are better than the result of a single linear model, and the eDMA with out-of-sample prediction performance is better.