A Hybrid Model for Portfolio Optimization Based on Stock Price Forecasting with LSTM Recurrent Neural Network using Multi-Criteria Decision Making Techniques and Mean-CVaR

The importance of a price forecasting issue due to the market volatility is very substantial. Investors have no desire to tolerate the high risk and invest in such a market due to avoiding ambiguity and mainly looking for a suitable solution for investing with high returns and low risk. The purpose of the research is to combine decision-making techniques with recurrent neural networks to create and develop a mathematical model for stock portfolio optimization due to different time horizons. Therefore, the top ten industries were selected using the Fuzzy Analytical Hierarchy Process (FAHP), according to effective criteria on the active industries in the stock market and using the opinions of active industries' experts, between May ۲۰۱۶ and May ۲۰۲۱. Then the price of stocks was forecasted in intended time periods using Long Short Term Memory RNN. In the next step, three stock portfolios with the short-term, mid-term, and long-term time horizons were created using a Combined Compromise Solution method, and then the optimized weights of each stock in the different portfolios were defined, and an efficient frontier was drawn by using Conditional Value at Risk (CVaR). The results showed that the provided model has high efficiency in stock portfolio optimization.

Keywords:

Optimization , Fuzzy Analytical Hierarchy Process (FAHP) , Combined Compromise Solution (CoCoSo) , Long Short Term Memory (LSTM) , Conditional Value at Risk (CVaR)