Volatility Forecasting Through Hybrid GARCHLSTM Models With WalkForward Validation and Multi-Horizon Evaluation

Abstract

This chapter, per the authors, examines volatility forecasting through a hybrid GARCH-LSTM framework designed to improve prediction accuracy under a rigorous and reproducible evaluation scheme. The purpose of this chapter is to assess whether combining the econometric structure of GARCH with the nonlinear sequence-learning capacity of LSTM enhances forecasting performance relative to a standard GARCH benchmark. The content includes the theoretical foundations of financial volatility, a review of hybrid forecasting literature, the construction of daily return-based volatility measures, the specification of the proposed model, and an empirical evaluation using S&P 500, EUR/USD, and Bitcoin data. Results from walk-forward and multi-horizon validation show that the hybrid model delivers more accurate short-term forecasts and remains competitive across different asset classes.