Stock Market Forecasting Using the Random Forest and Deep Neural Network Models Before and During the COVID-19 Period

Abstract

Stock market forecasting is considered the most challenging problem to solve for analysts. In the past 2 years, Covid-19 has severely affected stock markets globally, which, in turn, created a great problem for investors. The prime objective of this study is to use a machine learning model to effectively forecast stock index prices in three time frames: the whole period, the pre-Covid-19 period, and the Covid-19 period. The model accuracy testing results of mean absolute error, root mean square error, mean absolute percentage error, and r 2 suggest that the proposed machine learning models autoregressive deep neural network (AR-DNN(1, 3, 10)), autoregressive deep neural network (AR-DNN(3, 3, 10)), and autoregressive random forest (AR-RF(1)) are the best forecasting models for stock index price forecasting for the whole period, for the pre-Covid-19 period, and during the Covid-19 period, respectively, under high stock price fluctuations compared to traditional time-series forecasting models such as autoregressive moving average models. In particular, AR-DNN(1, 3, 10) is suggested when the number of observations is large, whereas AR-RF(1) is suggested for a series with a low number of observations. Our study has a practical implication as they can be used by investors and policy makers in their investment decisions and in formulating financial decisions and policies, respectively.