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