Independent variable selection: Application of independent component analysis to forecasting a stock index

Abstract

Forecasting of financial time series requires the use of a possibly large set of input (explanatory) variables drawn from a very large set of potential inputs. Selection of a meaningful and useful subset of input variables is a formidable task. How to find a reasonable transformation for a large set of multivariate data is a very common problem in many areas of science. This paper proposes a technique called independent component analysis (ICA) to extract the independent components (ICs) from monthly time series on a wide range of economic variables. This procedure will reduce the number of explanatory variables by reducing the set of financial and economic information to a much smaller subset of significant or dominant ICs, which it is hoped will capture most of the useful information. Removal of the non-significant components representing random elements in each of the sets of economic data should make it much easier to identify relationships between the ICs and the stock indexes. Properly estimated ICs are independent of each other. The ICs from the explanatory variable datasets are then used to perform and test forecasts of the S&P500 stock index, using neural network procedures. Numerous studies have shown neural networks to be very useful in non-linear forecasting. Independent component analysis has been employed in relatively few applications to finance. Kiviluoto and Oja (Independent Component Analysis for Parallel Financial Time Series, Helsinki University of Technology, Laboratory of Computer and Information Science, 895–8, 1998) used ICA in an application to parallel cash flow time series. Back and Weigand (‘A First Application of Independent Component Analysis to Extracting Structure from Stock Returns’, International Journal of Neural Systems, 8(4), 473–84, 1997) used ICA to extract estimates of the structure from a set of common stock returns. This research will contribute to the identification and understanding of the major economic factors affecting stock prices.