The forecasting of Stock Indices using various models is a highly recurrent, as well as difficult, theme in Finance and Investment. The payoff in profitability from a small increase in predictability in stock markets is enormous. Traditionally, the problem has been tackled by econometric models. However the ‘normality’ and ‘linearity’ assumption in these econometric models is questionable as revealed by empirical studies. In recent times, non-linear approaches have shown good promise in financial forecasting. Artificial Neural Networks are able to outperform most linear methods, however, their design and choice of various parameters is a ‘difficult art’. One alternative to this problem is the utilization of a Genetic Algorithm to determine these decisions so as to discover the ‘best’ Neural Network. In this paper, this hybrid Artificial Intelligence method for Sensex prediction is used. The ‘Naturally Evolved Network’ shows very good accuracy in both training as well as test periods, outperforming other models.

The forecasting of financial markets has always been an interesting challenge to academics and practitioners alike because of the sheer intellectual thrill and, of course, the accompanying lucre from any successful attempts. Many modeling approaches have originated from diverse fields such as Economics, Physics, Chaos Theory, Control Systems, Engineering and Biology, and newer fields continue to lend their tools to enrich the financial toolkit.Time series forecasting involves analyzing past data and projecting estimates of future data values. In other words, this method attempts to model a linear or nonlinear function by a recurrence relation derived from past values. The recurrence relation can then be used to predict new values in the time series, which might be a good approximation of the actual values.

Literature containing existing approaches in time series forecasting [Franses et al. (2000)] includes ARIMA due to Box-Jenkins, ARCH and GARCH family due to Engle, Granger and Bollerslev. Univariate models, like Box-Jenkins, contain only one variable in the recurrence equation. The equations used in the model contain past values of moving averages and prices. Box-Jenkins is good for short-term forecasting but requires a lot of data. It is also a complicated process to determine the appropriate model equations and parameters. Multivariate models are univariate models expanded to, as one scholar says, “discover casual factors that affect the behavior of the data”. These models, as the name suggests, contain more than one variable in their equations. The problem with the commonly used models in empirical finance is the linearity and normality assumptions in them. However, as noted by some authors, financial data exhibit nonlinear behavior and hence nonlinear models may be more appropriate for forecasting returns and volatility.