ESG Volatility Forecasting using LSTM Networks and Hyperparameter Optimization Techniques
Incorporating Environmental, Social, and Governance (ESG) factors has become increasingly important in modern investment strategies for both individual and institutional investors. However, forecasting ESG market volatility remains challenging due to the nonlinear and dynamic nature of financial time series. This study proposes a deep learning framework for ESG volatility forecasting based on an optimized Long Short-Term Memory (LSTM) model. The proposed approach integrates historical financial data, technical indicators, and macroeconomic variables to capture complex market behavior and volatility dynamics.To improve model robustness and automate hyperparameter selection, a Genetic Algorithm (GA) was employed to optimize three key LSTM hyperparameters: number of neurons, batch size, and look-back window. In addition to the proposed ESG-Volatility-3GA-LSTM framework, baseline LSTM, Random Search, and Grid Search optimization strategies were evaluated under identical experimental conditions.Experimental results demonstrate that the GA-optimized LSTM achieves competitive forecasting performance with improved stability across repeated runs. Statistical analyses, including paired t-tests and Wilcoxon signed-rank tests, further confirm the robustness and reproducibility of the proposed framework. The findings highlight the importance of systematic hyperparameter optimization for ESG volatility forecasting under varying market conditions.