SARILSTMAX: A Novel Hybrid Approach for Mutual Fund Price Prediction Using SARIMA, LSTM and Prophet Models

Abstract

Predicting the accurate value of mutual fund prices is essential for investors and financial analysts to make informed investment decisions. Mutual fund price prediction is a critical task in financial forecasting, with significant implications for portfolio management, risk assessment, and strategic planning. Traditional models such as SARIMA (Seasonal Autoregressive Integrated Moving Averages), along with machine learning approaches like Long Short-Term Memory (LSTM) networks, have been extensively utilized for this purpose. Additionally, the Prophet model, known for its capability in handling seasonal and trend-based data, has been widely applied. In this proposed work, SARILSTMAX, a novel hybrid methodology, integrates the strengths of SARIMA, LSTM, and Prophet models to enhance prediction accuracy and robustness for mutual fund price forecasting. The SARILSTMAX model was evaluated using real-world mutual fund price data and compared against the individual models. The results demonstrate that SARILSTMAX significantly outperforms the standalone models, achieving a Mean Absolute Error (MAE) of 4.6 and a Test MAE of 8.8, compared to SARIMA's Train MAE of 5.5 and Test MAE of 11.2, LSTM's Train MAE of 5.4 and Test MAE of 10.5, and Prophet's Train MAE of 4.8 and Test MAE of 9.2. Furthermore, SARILSTMAX recorded a Root Mean Squared Error (RMSE) of 5.1 and a Test RMSE of 10.2, outperforming SARIMA, LSTM, and Prophet, which had Test RMSEs of 12.5, 12.5, and 10.8, respectively. These findings underscore the effectiveness of the SARILSTMAX approach in capturing complex patterns in Indian mutual fund price movements, leading to more accurate and reliable predictions for real-world investment scenarios.