A Study of Stock Portfolio Strategy Based on Machine Learning

Abstract

At present, artificial intelligence is a hot topic in the field of finance. With the continuous development of domestic quantitative investment technology, it is increasingly difficult to obtain excess returns from traditional quantitative investment methods. Artificial intelligence, as a new data analysis and forecasting tool, has excellent processing capability for high-dimensional and serial data in the field of quantitative investment. As a result, quantitative investment has become one of the key areas where artificial intelligence is empowering the financial industry. In this paper, the data of listed companies in the New York Stock Exchange was used as the fundamental dataset. Twelve factors were selected as input variables for machine learning training. In terms of research methodology, portfolios were first selected based on different model algorithms, then the actual performance of each algorithm was back-tested, and investors were simulated to hold the portfolios for a long period. To ensure that the conclusions are better guided in practice, this paper attempts to apply the emerging machine learning algorithms and classical machine learning algorithms to the study of New York stock market returns, and to compare and discuss the predictive power of the algorithms on portfolio performance. The results of the study show that the portfolios selected by support vector regression and neural networks outperform the Dow Jones Index in the face of high noise and small sample space. In contrast, the emerging machine learning algorithms Adaboost regression and Bayesian Ridge regression performed slightly worse than the Dow Jones Index.