When Sustainability Meets Machine Learning: Reinforcement and Neural Evidence from an Emerging Market

This study examined how firm-level environmental and social performance relates to stock price volatility in an emerging market characterized by limited transparency and weaker institutional frameworks. While prior research largely relied on linear models and focused on developed economies, this study adopted a dynamic, data-driven perspective to capture potentially nonlinear and time-dependent sustainability–risk patterns. Using a panel of non-financial firms listed on Tehran Stock Exchange (TSE) over the period ۲۰۱۱–۲۰۲۳, the firm-level environmental and social indicators were constructed based on a systematic analysis of sustainability disclosures. Empirical results from conventional linear regressions indicated weak and statistically insignificant average associations between sustainability performance and stock volatility. However, learning-based models, including reinforcement learning (RL) and LSTM neural networks, demonstrated superior ability to capture nonlinear and dynamic volatility patterns conditional on sustainability-related information. These findings suggested that the sustainability disclosures contain predictive information for volatility dynamics, even when linear risk-reduction effects are not evident. The study highlighted the importance of flexible modeling frameworks when assessing the financial implications of environmental and social performance in emerging markets.

کلیدواژه ها:

ESG Performance ، Stock Volatility ، reinforcement learning (RL) ، LSTM Neural Networks ، Sustainable finance ، Nonlinear dynamics