Portfolio Construction Based on Combination of Individual Stocks and LSTM

With the development of idea portfolio optimization, it has become one important topics in field modern finance for achieving balance between maximizing return assets and minimizing risk. This study selected ten stocks from WIND used LSTM Neural Network as fundamental forecasting model. Based on method cyclic prediction, we predicted asset’s forecast value next day by rolling multi-dimensional historical features. Five with best outcomes were using both component volatility, covariance matrix stocks’ returns to build efficient frontier optimize weighting. The Monte Carlo Simulation was maximum Sharpe ratio minimum volatility portfolio. showed that Midea Mindray had potential profit dramatically at risk high volatility. However, relatively means possibility profiting is questionable, so a Minimum Volatility Portfolio also conducted, which indicated conservative investors usually reduce their investment much possible. Whereas max showing risk-seeking are more willing take risks if can generate profit. Overall, hybrid intelligent algorithm based artificial neural network proposed, new applications machine learning mathematical methods asset allocation promoted.