Financial index prediction is the key for advanced financial information services. The movement of financial indices depends on various factors. Though time series analysis has been used for addressing the problem of predicting the movement of indices, the performance has been far from satisfactory. Of late, machine-learning techniques are employed for addressing this problem. One such technique is Support Vector Regression (SVR). SVR is based on Support Vector Machines, the state-of-the-art machine-learning algorithm. SVMs' are strongly based on duality theory in optimization. The basic idea of the algorithm is to perform a regression in the kernel induced space using mapped data points or features. It turns out that the optimal solution is always a linear combination of all features. The feasible solutions lie within a subset of this feature space and the area of interest of this research is limited by this constraint. In the current research SVR for short-term prediction of NSE S&P CNX NIFTY index from 1-12-1995 to 30-10-2004 is applied and the performance is compared with conventional time-series prediction method. SVR also helps in capturing and modeling the long-term behavior of the system. The research strengthens that the radial basis function kernel suits time-series prediction better than many other kernels. It reveals that the performance of SVR is much better than that of conventional techniques in reducing the relative mean errors and root mean square error of predicted financial index.

Financial index prediction is vital for advanced financial information services. Among many applications of time series, predicting the financial time series is formidable and fertile for applied research. The movement of financial indices depends on various factors. Generally, the financial time series is characterized by predictable component(s), noise, non-stationarity and volatility [1,2]. Advanced investor information systems are key for successful financial information services. Chun-Hsin Wu, et al. [3] cite the traditional techniques such as Bayesian classification, Kalman Filters, ARIMA model, GARCH, neural networks and Monte-carlo simulation that have been employed for this purpose but with not much of satisfactory performance. Neural networks perform better than other traditional statistical methods because it is data driven and learning abilities (both offline and on-line), non-parametric (weak) models and hence lesser the chances of specifying the model incorrectly. Furthermore, they are capable of adaptive learning with different transient data sets and also very powerful in describing the dynamics of the financial time series. As the training of data requires non-linear optimization solvers, there is a risk of getting caught in the local optima. Support Vector Machines (SVM), in sharp contrast to this drawback, provide better results. Support Vector Regression (SVR) is a state-of-the-art machine learning technique that has drawn the attention of researchers in industry and academia as well.

Short-term Forecasting of NIFTY Index Using Support Vector Regression, financial, prediction, information, services, techniques, kernel, factors, technique, traditional, analysis, feasible, fertile, industry, investor, lesser, Montecarlo, predictable, academia, statistical, transient, volatility.