<![CDATA[This study compares the performance of traditional statistical arbitrage and Long Short-Term Memory (LSTM)-based deep arbitrage strategies in generating returns and risk-adjusted performance in the Indonesian stock market. A quantitative approach is employed using long-only trading simulations on daily closing prices of blue-chip financial sector stocks listed on the Indonesia Stock Exchange from April 2015 to April 2025. Stock pairs are selected based on correlation and cointegration criteria, while spread volatility is modeled using a GARCH (1,1) framework. To ensure a genuine out-of-sample evaluation, the sample is divided into an in-sample period from April 2015 to August 2021 for model training and parameter optimization, and an out-of-sample period from September 2021 to April 2025 for performance assessment. Strategy performance is evaluated using portfolio return and Sharpe ratio. The empirical results show that both strategies are feasible in the Indonesian market; however, the LSTM-based deep arbitrage strategy significantly outperforms the traditional statistical arbitrage approach, achieving a higher out-of-sample portfolio return (735% versus 482%) and a superior Sharpe ratio (1.67 versus 0.69). These findings indicate that deep learning-based arbitrage can provide substantial improvements in both return and risk-adjusted performance under long-only trading constraints in an emerging market context.]]>
Copyrights © 2025
