Research on Stock Selection Strategy Based on AdaBoost Algorithm

Abstract

In this paper, we propose a new direction of stock selection strategy, coining the AdaBoost-Decision Tree (Ada-DT) model, which uses the AdaBoost lifting algorithm and Decision Tree to predict stock movements and select superior stocks. As for financial investment, people are more sensitive to losses than the same amount of gains. Therefore, we add this consideration to the use of AdaBoost algorithm. We use quality indicators, growth indicators, per-share indicators, and sentiment indicators, 24 in total, as the feature space, and try to identify the stocks that may outperform the broader market and have high yield from the Shanghai Stock Exchange (SSE) 50 constituent stocks. The stock selection result of Ada-DT is that the average correct rate of the out-of-sample test set is 72.84%, and the AUC is 0.634. At the same time, we compare the Ada-DT with AdaBoost-Support Vector Machine (Ada-SVM) prediction model and find that Ada-DT provides a higher correct rate and better cumulative returns, indicating that the Ada-DT algorithm is reasonable for complex nonlinear stock markets.