Predicting return reversal through a two-stage method

In the stock market, return reversal happens when investors sell overbought stocks and buy oversold stocks, making the trends of the stocks' prices reverse. While existing studies mainly focused on developing theories to explain the cause of return reversal, we aim at predicting return reversal by proposing a two-stage method in this paper. In the first stage, we employ dynamical Bayesian factor graph (DBFG) to select key factors correlating with return reversal closely from a comprehensive set of economic factors. In the second stage, we input the key factors into artificial neural network (ANN), support vector machine (SVM) and hidden Markov model (HMM) respectively, to accomplish the prediction of return reversal. Through extensive experiments on the US stock market, we establish that the key factors influencing return reversal generally change from year to year, yet factors related to the economic theory of liquidity effect consistently emerge as part of the key factors. Besides, DBFG-ANN achieves the most accurate prediction among the models, leading to precisions above 60%.