Role of variation and jump component in measure, modelling and forecasting S&P CNX NIFTY index volatility

摘要

Measuring and forecasting volatility plays an indispensable role in the theoretical developments and in various applications in finance. Volatility is latent and cannot be observed; however, recent studies have used high frequency data to construct ex post measures of daily volatility. In this study, we predict the volatility of the S&P CNX NIFTY market index using different heterogeneous autoregressive (HAR) specifications based on realised volatility, realised bipower variation, jump and continuous component and their logarithms. In doing so, we also examined the role of the jump and continuous component in enhancing the predictability of the HAR model. We show that the jump and continuous component do not contain any additional information. The results also provide the empirical evidence that, inclusion of realised bipower variance in the HAR models helps in predicting future volatility. We also propose a feedforward neural network-based averaging approach to forecast realised volatility. The result shows that neural network-based averaging approach provides modest improvement in the accuracy compared to the HAR models.