A Computational Study on Window-Size Selection in Stock Market RILS Interval Forecasting

The S & P 500 index is an important overall measurement of the stock market. Instead of using traditional point methods, He and Hu used rolling interval least squares(RILS) to forecast the annual variability of the index from 1940-2004 and obtained astonishing results L. He and C. Hu (2007) and C. Hu and L. He (2007). They used a ten-year rolling window without detailed justification. In this study, we apply Fourier analysis to investigate if any periodical properties reside in the input data. Then, we try to apply such property, if any, in window size selection and to possibly improve the overall quality of the stock market annual interval forecasts. Our computational results indicate that the rolling window size used in C. Hu and L. He (2007) is fairly reasonable. In other words, it can produce overall comparable quality forecasts against the window size selected through Fourier analysis.