“EMPIRICAL INVESTIGATION OF INDIA VIX FORECASTING MODELS: DEEP LEARNING MODEL AND LEGACY TIME SERIES FORECASTING TECHNIQUES”

Abstract

The study aims to forecast India VIX using advanced machine learning technique along with traditional time series models to provide insights into the strengths and weaknesses of diverse modelling approaches for predicting India's Volatility Index. The research utilized a comprehensive dataset spanning eleven years from 2012 to 2022, which includes a highly volatile period during the COVID-19 pandemic. The study employed both legacy time series models (such as SARIMAX, EMA, GARCH, and ARMA) and advanced machine learning techniques, with a focus on LSTM modelling. The findings of this study reveal limitations in the application of traditional time series models, notably SARIMAX, EMA, GARCH, and ARMA, when forecasting the India VIX. These models struggle to capture the complexity of volatility and the non-normality of the India VIX. On the other hand, LSTM modelling, a machine learning technique, emerges as a robust and effective tool for forecasting the India Volatility Index. This suggests that advanced machine learning techniques outperform traditional models in this specific context.

Keywords: India VIX, Volatility forecasting, LSTM model, GARCH model, ARMA model, EMA model, SARIMAX model, Predictability 

JEL Classification Codes – G11, G17, G14