Deep Learning and Machine Learning Models enabled Mean-Variance Portfolio Optimisation in the Indian Stock Market

Abstract: This study uses deep learning and machine learning models to combine return prediction with mean-variance portfolio optimisation in the Indian stock market. The top 50 biggest stocks in India are included in the Nifty 50 index, which is the subject of this research. Each stock is weighted according to its market size. Evaluation of daily closing prices of 33 chosen Nifty 50 equities are examined using historical data over a ten-year timeframe from 2014 to 2024. Prediction of the stock prices is done using three deep learning models, i.e., LSTM, GRU, and MLP and one machine learning model, i.e., Random forest (RF). The five models that are applied are contrasted using performance measures like Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) with autoregressive integrated moving average (ARIMA) as a comparison. Further, each of the 33 stocks has a different model predicting the best return of each stock. The predicted value as per the stocks best model is compiled to then apply these values to form a portfolio using the mean-variance portfolio optimisation model. The best way to allocate each stock in the portfolio is to use the Sharpe ratio; investments are made according to the portfolio that has the highest Sharpe ratio. The research aims to shed light on how well these deep and machine learning models work in the Indian stock market for portfolio optimisation. Keywords: Deep Learning, Machine Learning, Portfolio Management, Portfolio Optimization, Portfolio Selection JEL Classification Number: G11, G12