ENHANCING GARCH FORECAST ACCURACY THROUGH A HYBRID DECOMPOSITION FRAMEWORK

In recent years, there has been a greater emphasis on the forecasting accuracy of heteroscedastic models. Instead of estimating the returns volatility using a generalised autoregressive conditional heteroscedastic model (GARCH model), this study separates the returns internal components from the external trend first using a decomposition method called “external trend and internal components analysis method” (ETICA), then estimates the returns volatility using a GARCH (1,1). The study's goal is to determine whether this separation has an effect on the prediction accuracy of the volatility of S&P 500, NASDAQ and Dow Jones stock indices. To explore the ETICA method effect, the root mean squared error has been used to compare the prediction accuracy before and after decomposition. The findings show that on average, the RMSE results were found to be lower before decomposition which means that stock returns had a higher prediction accuracy.