Forecasting Composite Stock Price Index on Indonesia Stock Exchange Using Extreme Learning Machine

Abstract

Technological advances have driven active participation in digital economic activities, including capital market investment. Stocks remain a dominant instrument, with the Composite Stock Price Index or Indeks Harga Saham Gabungan (IHSG) serving as a primary benchmark for investment decisions in Indonesia. However, its high volatility—driven by economic, political, global, and market sentiment factors—demands accurate forecasting methods. Traditional approaches such as ARIMA and linear regression are limited in capturing the non-linear and complex patterns of stock market data. This study proposes the use of the Extreme Learning Machine (ELM), an artificial intelligence method considered more adaptive to market dynamics. To enhance prediction accuracy, hyperparameter optimization was performed using the grid search method. The research forecasts IHSG performance by incorporating exogenous variables, namely gold prices, the US dollar to rupiah exchange rate, and a COVID-19 dummy variable. The optimal model utilized a hidden layer configuration of nine neurons. Evaluation results indicate that the ELM models effectively perform multi horizon forecasting (t+1 to t+5), as evidenced by low MAE, MAPE, and RMSE values across horizons. The five-day IHSG forecasts are 7,242.28, 7,228.42, 7,211.02, 7,192.67, and 7,174.06, demonstrating the model’s potential in supporting investment decision-making with high accuracy.