Group penalized logistic regressions predict up and down trends for stock prices

Stock prices are influenced by many economic factors, investors psychology and expectations, movement of other stock markets, political events, etc. Therefore, correctly predicting up and down trends for stock prices is an important puzzle in the financial field. In this paper we combine technical analysis with group penalized logistic regressions, and propose group SCAD/MCP penalized logistic regressions with technical indicators to predict up and down trends for stock prices. Firstly, we screen out 24 important technical indicators, divide them into the five different indicator groups, and construct group SCAD/MCP penalized logistic regressions for the three listed companies. Secondly, we apply the training set to learn the parameter estimators and the probability estimators for the two group penalized logistic regressions, adopt the test set to obtain confusion matrices and ROC(Receiver Operating Characteristic) curves to assess their prediction performances, and found that the AUC values to the three companies all exceed 0.78. Finally, we compare group SCAD/MCP penalized logistic regressions with SCAD/MCP penalized logistic regressions, and found that the two group penalized logistic regressions perform better than the two penalized logistic regressions in terms of prediction accuracy and AUC. Therefore, in this paper we develop a new prediction method by combining group SCAD/MCP penalized logistic regressions with technical indicators to improve the prediction accuracy and bring huge economic benefit for investors.