The present study aims to propose a predictive model to forecast the sustainable stock indices. For this, the Long Short-Term Memory (LSTM) neural network model is applied through Keras and TensorFlow to closing values of six developed and emerging markets: the US, the UK, Japan, Brazil, South Africa, and China. Further, the ‘Adam’ optimiser and mean squared error loss function are used to train the model. To gauge the superiority of the LSTM model, a rolling window Autoregressive Integrated Moving Average (ARIMA) model is also employed. The performance accuracy of both models is evaluated using the Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared (R2). The LSTM model, with two LSTM and two dense layers, yields the best results, achieving the highest precision in predicting the values of sustainable indices. The values of RMSE and MAPE confirmed this, and the accuracy is also verified by the R2 values. LSTM shows superior predictive accuracy and is indicated to be fit for non-linear market patterns than rolling window ARIMA. The study enables policymakers and practitioners to forecast these indices and design policies to motivate related investments.
Forecasting sustainable stock indices using deep learning algorithms provides valuable insights into the dynamics of responsible investment markets. By capturing complex temporal patterns and non-linear relationships, the proposed framework supports dynamic portfolio reallocation and improves risk-adjusted returns, enabling investors to better manage volatility. This research provides significant implications for ESG-conscious investors, who may leverage these insights to strengthen ethical investment strategies and support climate transition initiatives. Moreover, the adaptability of the proposed approach is particularly significant for emerging economies, where rapid structural transformations and market volatility necessitate advanced, data-driven portfolio management solutions.