Multi-factor Stock Forecasting Based on Regression and Machine Learning Models

Abstract

In some stock predictions, shallow learning is better than deep learning. With this in mind, this study will adopt a machine learning model and a regression model to predict Chinese stocks to help the stock timing strategy. Different from the previous usage of plenty of indicators as stock features for prediction, this study selected specified multi-factor data and compared the prediction effects of foreign multi-factor data models and China's multi-factor data models to verify. It is found that the prediction effect of the Chinese new salience multi-factor data model which has a better ability to explain market anomalies is better. However, factor models with higher correlations have poor results. Compared with randomly selecting factors or selected factors weight by model, choosing a specific factor model in advance can not only improve calculation efficiency, but also improve prediction accuracy. In addition, compared with the regression model, the prediction results of machine learning also have better performance results.