This paper evaluates the performance of eight alternative models for predicting the stock price volatility using S&P CNX Nifty and CNX Nifty Junior Index return series. The competing models contain both historical models (random walk, historical mean, MA, ES, EWMA, AR) and models from the GARCH family (Standard GARCH, GJR-GARCH). The study uses both symmetric error statistics (ME, MAE, RMSE, MAPE, Theil-U) and asymmetric loss function (LINEX, MME(U), MME(O)) to evaluate the forecasting accuracy. The results show that the GARCH family models perform extremely well in the asymmetric error statistics and also do well in symmetric error statistics. Among historical models, random walk is superior according to symmetric error statistics only. It is relatively an `unbiased' model, systematically neither over-predicting nor under-predicting the volatility. The study does not find support for exponential weighted moving average, exponential smoothing, regression model, and moving average method, in contrast to the results found in various other markets.

The expected volatility of the financial markets is a key variable in many financial investment decisions. For example, it is a common practice to reduce asset allocation decisions to a two-dimensional decision problem by focusing solely on the expected return and risk of an asset or portfolio, with risk being related to the volatility of the returns. The volatility of returns also plays central role in the valuation of the financial derivatives, such as options and futures. Considering the significance of the volatility forecasting in asset pricing, option pricing, in the economic literature, a large number of
volatility forecasting techniques have been evolved over the years and the effort is still on to search for new alternative techniques.

There is a large literature on volatility forecasting ability of the models, mainly based on the data of the developed economies. Many econometric models have been tested. However, no single model is found to be good. Using US stock market data, for example, Akgiray (1982), Pagan and Schwert (1989), and Brooks (1998) find GARCH models outperform most competitors. Brailsford and Fall (1996) examined competing models for Australian stock market and found that no single model is clearly superior but found some support for GARCH models. Tse (1991), and Tse and Tung (1992) questioned the superiority of the GARCH model in the Japanese and Singaporean markets respectively but found support for Exponentially Weighted Moving Average (EWMA) model. Balaban, Bayar and Faff (2003) carried out a study to evaluate forecast accuracy of 11 models in 14 stock markets—Belgium, Canada, Denmark, Finland, Germany, Hong Kong, Italy, Japan, the Netherlands, Philippines, Singapore, Thailand, UK and US.

Applied Finance Journal, Volatility Forecasting Techniques, Indian Capital Market, Financial Markets, Financial Derivatives, Econometric Models, Australian Stock Market, GARCH Models, UK Stock Market, Auto Regressive Conditional Heteroskedasticity Models, Stochastic Volatility Model.