<![CDATA[This study compares the performance of Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Prophet models in predicting real estate stock prices on the Indonesia Stock Exchange (2019–2024) and evaluates investment risks using Value at Risk (VaR) and Expected Shortfall (ES). Historical stock data underwent normalization and dataset splitting (ratios of 70:30, 80:20, and 90:10), with time steps of 40, 60, and 100, and three dense layers (25 and 50 neurons). Performance was evaluated using MSE, RMSE, MAE, and MAPE. Results indicate that GRU achieved the highest accuracy, especially for PWON, ASRI, and DILD stocks, with the lowest MSE values (PWON: 120.7436, ASRI: 26.3150, DILD: 28.9713). LSTM showed competitive performance, while Prophet had the lowest accuracy for short-term predictions. Risk analysis revealed Prophet had the lowest historical risk but the highest risk for 150-day forecasts. LSTM demonstrated superior long-term risk mitigation. Comparison with actual prices revealed that LSTM and GRU more accurately captured stock price fluctuations than Prophet, particularly during sharp price changes. GRU provided the closest predictions in the 150-day forecast scenario, making it the most effective model for real estate stock forecasting. This study offers valuable insights for investors and portfolio managers in understanding stock price movements and managing investment risks in the real estate sector.]]>
Copyrights © 2025
