<![CDATA[The movement of the Indeks Harga Saham Gabungan (IHSG) exhibits high volatility influenced by macroeconomic dynamics. This study aims to compare the performance of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) architectures in predicting IHSG closing prices using a multivariate approach integrating the USD to IDR exchange rate and Bank Indonesia (BI) interest rate. Daily historical data from January 2, 2020, to December 31, 2024, were processed through normalization and sequential sliding window formation. Evaluation results indicate that the LSTM model outperformed the RNN. The LSTM model achieved a Mean Absolute Percentage Error (MAPE) of 0.8158% (99.18% accuracy), while the RNN model yielded a MAPE of 0.9314% (99.06% accuracy). Furthermore, the forecasted prices were successfully implemented into a Moving Average Crossover strategy simulation (MA-5 and MA-20) to detect Golden Cross and Death Cross momentums as automated trading signals. In conclusion, the LSTM architecture proved to be more optimal and effective in capturing financial data volatility for IHSG forecasting compared to RNN.]]>
Copyrights © 2026
