GARCH Models for Predicting Volatility in Limited Data

Volatility forecasting is crucial in economic and agricultural markets, particularly in emerging economies where data availability is often limited. This paper investigates the performance of Generalized Autoregressive Conditional Heteroskedasticity (GARCH) models, including Glosten-Jagannathan-Runkle GARCH (GJR-GARCH), EGARCH, and Markov-Switching GARCH (MS-GARCH), in predicting price volatility. The models are evaluated using Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) to assess their forecasting accuracy. Empirical analysis based on real daily stock return data from the Nairobi Securities Exchange (NSE), demonstrates that the MS-GARCH model outperforms other variants, achieving the lowest AIC and BIC values and effectively capturing regime shifts in volatility. These findings highlight the importance of considering structural breaks in volatility modeling. The study also emphasizes the potential of integrating GARCH models with machine learning techniques to enhance forecasting accuracy and adaptability in dynamic market environments.

Keywords:

Volatility Forecasting , GARCH , GJR-GARCH , EGARCH , MS-GARCH