Answers are grounded in this paper and cite their sources.
The Jakarta Composite Index (JCI) is an important indicator of the Indonesian capital market, characterized by fluctuating movements influenced by various economic factors, making JCI forecasting particularly challenging, especially in multi-step forecasting. This study aimed to implement a Bidirectional Gated Recurrent Unit (Bi-GRU) Encoder–Decoder model with an Attention mechanism for multi-step JCI forecasting. Historical JCI data consisting of Open, High, Low, and Price variables were divided into training and testing datasets using an 80:20 ratio. The research process included data normalization, sequence generation using a sliding window technique with 30 input time steps and 10 output time steps, and the construction of three-dimensional tensors as model inputs. The model was developed using a Bi-GRU Encoder–Decoder architecture with Scaled Dot-Product Attention and trained for 100 epochs with a batch size of 32. The model was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results showed that the model achieved an MAE of 196.02, an RMSE of 252.37, and a MAPE of 2.7%, indicating very good performance. In addition, the model effectively captured the actual data patterns and generated 10-day-ahead JCI forecasts that remained relatively stable within the range of 8.534–8.543 points. The Attention mechanism enabled the model to focus on important information from historical data, thereby improving the accuracy of multi-step JCI forecasting. Therefore, it can be concluded that the implementation of the Bi-GRU Encoder–Decoder model with an Attention mechanism is effective in improving the accuracy of multi-step JCI forecasting.
No takes yet. Share an insight, caveat, or question.
Mooduto et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: