Published Online:March 2026
Product Name:The IUP Journal of Financial Risk Management
Product Type:Article
Product Code:IJFRM040326
DOI:10.71329/IUPJFRM/2026.23.1.55-63
Author Name:Tushar Kumar and Chandravesh Chaudhari
Availability:YES
Subject/Domain:Finance
Download Format:PDF
Pages:55-63
Effective risk management, derivative pricing, and strategic asset allocation in emerging financial markets are all about accurate volatility forecasting. This paper makes a methodical comparison of the volatility forecasting models in Indian financial markets, comparing the old econometric models with machine learning models. The predictive performances of GARCH, VAR, LSTM, GRU and XGBoost models are compared by analyzng daily data from 2010 to 2024 of five asset classes: Nifty 50, Gold, USD/INR, 10-Year Bond Yield and India VIX. The findings indicate that machine learning approaches, especially GRU, can be more accurate when the regime is in the high volatility state, whereas conventional methods can be used when the market is stable. The improved performance of the GRU model is attributed to the fact that it is able to model nonlinear dependencies and volatility spillovers using parameter efficiency. The findings provide useful information on the choice of financial models and the development of future volatility prediction strategies in emerging markets. The findings give empirical recommendations on the application of the right modeling structures in various market environments.
Structural instabilities and intricate cross-asset linkages present a methodological challenge to volatility forecasting in emerging markets. Conventional econometric models, especially GARCH and VAR, offer theoretically sound models of volatility prediction, and machine learning methods, long short-term memory (LSTM), gated recurrent unit (GRU), and XGBoost, are more flexible to nonlinear dynamics and regime changes (Gu et al., 2020). Nevertheless, an empirical comparison of these approaches in an Indian financial setting is rather strict and has not yet been made in the literature.