MA200 and HMM in Regime - Specific Stock Markets Prediction

This study investigates how using regime-dependent forecasting in stock market prediction will affect the prediction accuracy compared with traditional methods that ignore market states. In this paper, we are interested in predicting the rise and fall of stock prices and using the log return for each day as a criterion for this. We assume there are two regimes: Bull and Bear, and two methods are used for including them in forecasting. The MA200 method directly classifies each day into bull or bear regimes and estimates separate transition matrices to calculate the probability for the next day being up or down. While the Hidden Markov Model (HMM) uses the ExpectationMaximisation(EM) algorithm to estimate regime transitions and assumes regime-dependent normal distributions for log returns to find the probability. Afterwards, we make a prediction to label each day as Up or Down and compare with the actual data to find the hitting rate as accuracy. This paper has used 3 different stock markets and found out that both methods perform better than the traditional one. Additionally, HMM outperforms in one-step forecasts, while due to the complexity of this method, MA200 will be good to perform N-step forecasts by getting the nth power of P until n = 10.