Portfolio optimization with return prediction using LSTM, Random forest, and ARIMA

In today's world, optimizing investment portfolios has received increasing attention. While predicting the expected returns of investment options and incorporating them into the objective function for profit maximization is a common practice, the most significant innovation in current research is the minimization of prediction error as the objective function. This innovation advises investors to emphasize not only on profit and risk but also on the predictability of investment options when forming an investment portfolio. Integrating return prediction from traditional time series models into portfolio formation can enhance the performance of the primary portfolio optimization model. Since machine learning and deep learning models have demonstrated a significant superiority over time series models, this paper combines return prediction in portfolio formation with machine learning models, namely Random Forest, and deep learning model, Long Short-Term Memory (LSTM). To evaluate the performance of the proposed model, five years of historical data from 2017 to 2021 are used for five industry sectors: banking, automotive, pharmaceutical, metal, and petroleum. The experimental results demonstrate that the mean-variance optimization models perform better when return prediction is done using Random Forest