Market Style Discrimination via Ensemble Learning

Abstract

Market styles represent patterns of stock trends in a period. Investors can discriminate and leverage current market styles to improve their stock price forecasts and make more profits. However, existing studies only discriminate the market styles by features’ scale values, which exists two deficiencies. One is ineffective features and the other one is weak connections between the market styles and classifiers. Inspired by the multiple-factor model in quantitative finance and the Gini index in ensemble learning, we propose a novel approach that discriminates the market styles by features’ contribution to stock trend forecasts, which strongly bridges the discrimination of the market styles and the properties of classifiers. Experimental results of 12 stocks on the Hong Kong Exchange demonstrate that our method outperforms baselines in terms of F1 score over most stocks. Our source code is available at: github.com/Pangjing-Wu/FC-MSD.