Penalized logistic regressions with technical indicators predict up and down trends.

Area of Science:

• Financial econometrics
• Machine learning in finance
• Time series analysis

Background:

• Accurate stock price trend prediction is crucial for financial market success.
• Existing methods may not fully capture complex market dynamics.
• Enhancing prediction accuracy can lead to significant investor benefits.

Purpose of the Study:

• To introduce and evaluate five penalized logistic regression models for stock price trend prediction.
• To compare the performance of these novel methods against traditional machine learning algorithms.
• To identify the most effective penalized regression technique for improving stock return prediction accuracy.

Main Methods:

• Application of five penalty functions (ridge, LASSO, elastic net, SCAD, MCP) to logistic regression.
• Utilizing 19 technical indicators for prediction.
• Employing tenfold cross-validation and coordinate descent algorithms for parameter estimation.
• Assessing performance using confusion matrices and ROC curves.

Main Results:

• Penalized logistic regressions were successfully formulated and implemented.
• Binomial deviation and cross-validation error were used to select optimal tuning parameters.
• Minimax concave penalty (MCP) logistic regression outperformed other methods, including logistic regression, SVM, and ANN, for Google stock price prediction.
• The proposed methods showed improved prediction accuracy for stock returns.

Conclusions:

• Penalized logistic regression, particularly with MCP, offers a robust approach to stock price trend prediction.
• The integration of technical indicators and advanced penalty methods enhances predictive power.
• These findings suggest potential for improved investment strategies and economic benefits for investors.