Application of ARIMA-GARCH-S model to short-term stock forecasting

Abstract

The performance of the stock market in the financial domain profoundly impacts the economic well-being of many individuals. Thus, accurately predicting stock prices is an essential task. Although traditional financial time series models such as ARIMA and GARCH play a crucial role in predictions, they may fail to capture all the market dynamics. This study explores a composite model combining ARIMA, GARCH, and Stacking techniques (ARIMA-GARCH-S) to enhance the accuracy of predictions. The research data are derived from the stock closing price time series data of “Amazon” from June 29, 2020, to April 12, 2022, with 653 entries and “Caterpillar” from February 8, 2021, to October 21, 2022, with 621 entries. The model’s fitting performance is evaluated by comparing the fitting residual plots and variance graphs, while predictive performance is determined by comparing the MAPE, RMSE, and EC statistical metrics. The results indicate that the ARIMA-GARCH-S composite model has a significant predictive advantage over the ARIMA model. This finding not only offers a new avenue for model innovation but also provides financial market participants with a more precise and stable prediction tool, aiding them in making more informed investment decisions.