Machine learning applied to stock index performance enhancement

Abstract

The project constructs a stock selection model by machine learning methods to enhance the performance of the benchmark index for individual investors. Stock returns prediction is a highly researched topic. However, it is a difficult problem because the stock prices are complex, non-linear, and chaotic. Moreover, overfitting is always an important issue in machine learning field. In this article, it shows that how to solve these problems by dealing with time series data, feature engineering, and model construction. We apply the stock selection model on S&P 500 index and FTSE 100 index. The result shows that the portfolios with stock selection model outperform the benchmarks, and 2% of the number of constitution stocks is the best choice for the stock selection model. Besides, feature importance analysis shows that the stock selection model can measure import features appropriately, which means it has the ability to adapt to different economic environments. In addition, the portfolios with fewer stocks usually outperform the portfolios with more stocks shows the good prediction of the stock selection model. The results imply that machine learning techniques have a good application in stock markets.