Forecasting Indian Financial Markets Using Echo-State Networks

Financial time-series prediction is a challenging problem due to the noisy, chaotic, and non-stationary dynamics characteristic of stock markets. Echo-State Networks who are a part of RC (Reservoir Computing), offer a computationally efficient alternative to traditional recurrent neural networks by training only a linear readout layer while maintaining a rich nonlinear dynamical reservoir. In this work, we apply the ESN model to forecast the NIFTY 50 index using 10 years of historical data. The model achieves a Mean Absolute Percentage Error of 0.164% under an open-loop evaluation setting, demonstrating the capability of ESNs to capture temporal dependencies in financial data. The results indicate that ESNs are competitive, lightweight, and well suited for short-term financial forecasting.