LSTM for Return Prediction and Portfolio Optimization in America Stock Market

With the development of the financial sector, asset portfolios have become an important part of the financial industry but using the traditional tools to complete asset portfolios is inefficient. In the paper, an asset portfolios system which uses machine learning as main analysis method was designed, aiming to get asset portfolios quickly and precisely. More specifically, firstly, this study used random forest model to process data to get variables importance, observing every variable’ s contribution and importance in prediction part. Then, based on known data, Long short term memory (LSTM) model was employed to predict future stock returns which was used as data base of asset portfolios. After that, this study combined prediction result with real data and put it into equal-weight allocation model to obtain annual return and annual volatility. Finally, efficient frontier was carried out using Monte Carlo stimulation, calculating the sharp ratio and the best weight over chosen stocks. By this process, this paper finally completed the asset portfolios and built the asset portfolios system. The result shows that the asset portfolios system, combination of random-forest model and LSTM model, can do the asset portfolios efficiently and precisely.