ENHANCING LQ45 STOCK PRICE FORECASTING USING LSTM MODEL

Abstract

Stocks listed in the LQ45 index represent companies with high liquidity, large market capitalization, and strong fundamentals, making them pivotal to the movements of the Indonesian capital market. This study selects eight LQ45-listed stocks from the energy and mining sectors, as well as the banking sector. Historical data spanning a 10-year period from February 28, 2015, to February 28, 2025. This research aims to mitigate the impact of stock market dynamics, a significant challenge for investor decision-making. The Long Short-Term Memory (LSTM) method was employed to forecast stock prices using four variables: opening, highest, lowest, and closing prices. The LSTM architecture was chosen because its gated memory cells can effectively capture long‑term dependencies and nonlinear patterns in financial time series, thereby aligning with the research objective of minimizing forecasting error under volatile market conditions. Evaluation results using the Mean Absolute Percentage Error (MAPE) showed prediction errors below 2.5%, indicating relatively low forecasting error. Root Mean Squared Error (RMSE) values varied depending on stock price volatility. Companies exhibiting higher stock prices, such as Indo Tambangraya Megah Tbk (ITMG), demonstrate larger RMSE values. For opening prices, predictive accuracy was notably strong, with MAPE values consistently below 1.26%. This suggests that opening prices, influenced by pre-market sentiment and historical data, are more stable and easier to predict compared to other price variables.