Forecasting Indian Stock Market Index Using Singular Spectrum Analysis: A Comparison with ARIMA and Artificial Neural Network

Abstract

In this paper we focus on analyzing the predictive accuracy of three different types of forecasting techniques, Autoregressive Integrated Moving Average (ARIMA), Artificial Neural Network (ANN), and Singular Spectral Analysis (SSA), used for predicting chaotic time series data. These techniques have different origins. ARIMA, ANN and SSA roots to Statistical Time Series Analysis, Computational Biology and Signal Processing respectively. The objectives of the paper can be explained in two parts: (1) To present the use of Singular Spectral Analysis (SSA) as a forecasting tool for predicting the index value of Indian Stock Market. (2) To compare the forecasting results from SSA in comparison to a parametric model, say Autoregressive Integrated Moving Average (ARIMA) and a non-parametric model, say Artificial Neural Network (ANN). In order to understand the processes of these techniques, we start with an example where, the SSA, ARIMA and ANN are provided with NSE Nifty 50 daily closing index data for 14 years from 1st January 1998 to 30th June 2014 that consists of 4123 data points. The Data is truncated into 4000 data points as input for above mentioned models and 123 data points as a scale for comparing the forecasting results from the above models. Later on we run Simulation to measure the Consistency and Accuracy of Performance of SSA, ARIMA and ANN. The accuracy and performances are validated by running the technique on 100 randomly generated time series with 2500 data points each. For each time series, the technique is compared on the basis of Root Mean Squared Error (RMSE). We find Predictability accuracy and performance of ANN better than SSA and ARIMA, and SSA better than ARIMA.