An Empirical Evaluation of Multivariate Temporal Convolutional Networks with Global Market Indicators for Forecasting the Indonesian LQ45 Index

This study develops a multivariate Temporal Convolutional Network (TCN) framework to forecast the LQ45 stock index using daily time-series data from January 2015 to January 2025. The objective is to examine whether incorporating global market indicators, namely the Volatility Index (VIX), Brent crude oil price, and the Effective Federal Funds Rate (EFFR), alongside lagged LQ45 values, improves forecasting performance in Indonesia’s equity market. Two comparison models are considered: an Autoregressive Integrated Moving Average with Exogenous Variables (ARIMAX) model as a statistical baseline and a univariate TCN as a deep learning benchmark. Data preprocessing includes normalization and a seven-day sliding-window framing. Forecasting accuracy is evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE), complemented by validation and interpretability analyses. The results show that while the multivariate TCN captures interactions among multiple temporal features, it does not provide measurable performance advantages over ARIMAX or the univariate TCN. The lagged LQ45 series exhibits the strongest predictive contribution, followed by EFFR with a stable secondary effect, whereas Brent oil prices and VIX display weak and unstable influences. These findings suggest that in short-horizon forecasting of relatively stable emerging markets, integrating exogenous variables showed no performance improvement and was associated with higher forecast error when their predictive structure was unstable, highlighting the trade-off between feature complexity and model robustness.

@article{Marakub2026,
  title     = {An Empirical Evaluation of Multivariate Temporal Convolutional Networks with Global Market Indicators for Forecasting the Indonesian LQ45 Index},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {2},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Yohanes Marakub and Muhammad Zarlis},
  doi       = {10.14569/IJACSA.2026.0170220},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170220}
}