A Smart Approach to Portfolio Management using LSTM

Abstract

This work is aimed at analysing the efficacy of the Long Short Term Memory algorithm on adaptive allocation of financial assets which includes cash, stocks, bonds, mutual funds, by building a system that automates the process of recommending asset values based on improved risk-adjusted returns. A limited set of assets, such as stocks is chosen and then this system will perform fund risk analysis and risk management to determine how to optimally allocate funds to those assets. Our model estimates the correlation among the different assets and stochastically optimizes the diversification of the assets for personalized recommendations. The agent is trained on both historic and real-time market data. After comparing it’s performance with traditional portfolio management algorithms we conclude that the LSTM model can generate recommendations which accounts for better profitability.