Deep learning in factor investing: an application to Indonesian Islamic and conventional equities

Abstract

This study aims to examine the efficacy of integrating machine learning (ML) architectures and feature selection protocols within a traditional asset pricing framework to enhance equity return predictability. Leveraging a methodological pipeline that synergizes artificial neural networks (ANN) with sequential feature selection (SeFS) and Least Absolute Shrinkage and Selection Operator (LASSO) regularization, this study analyzed the momentum, value and quality risk premia across 949 conventional and 621 Islamic equities in the Indonesian market from 2016 to 2025. To isolate robust signals, this study further uses complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) for price denoising. Empirical results indicate that momentum factors, particularly those with a one-month horizon, exhibit superior predictive power. Feature selection consistently identifies one-month momentum, earnings-to-price and gross profit-to-total assets as primary predictors. Notably, Islamic equities exhibit greater sensitivity to valuation anomalies, with EBIT/EV and gross profit-to-enterprise value providing additional predictive power. The ANN models achieve robust forecasting performance, with forecasting performance metrics ranging from 70% to 85%. The predictive outcomes exhibit significant invariance to the number of hidden layers, suggesting that factor risk premia possess an inherent structural stability that is not materially enhanced by increasing the complexity of the deep network. The findings provide actionable insights for portfolio managers, Islamic fund managers and quantitative investors by demonstrating that ML models combined with factor investing strategies can substantially improve equity return forecasting in emerging markets. The study also highlights the relevance of short-term momentum and value-related factors for Islamic equities, supporting the development of more efficient Shariah-compliant investment strategies and AI-driven portfolio allocation systems. This study advances the asset pricing and Islamic finance literature by integrating ANN, SeFS, LASSO and CEEMDAN within a unified equity return forecasting framework. It provides novel comparative evidence from 949 conventional and 621 Islamic Indonesian equities, highlighting the predictive dominance of short-term momentum and value-related factors. The findings also reveal the structural stability of factor risk premia across different neural network complexities in an emerging market context.