Data Collaboration and Predictability of Financial Index

This paper investigates the possibility of improving the predictability of financial time series by exploiting the effects of long-memory and cross market correlation. By using support vector machines (SVM) to predict S&P 500, we compare the forecasting performances of different kinds of data collaboration. Our results indicate that it is hard to improve the predictability of financial index by incorporating correlated time series into forecasting models. For a given forecasting horizon, the predictive performance could be improved, provided that the historical information is well organized. Furthermore, the directions of predictive errors of S&P 500 are almost contrary to the directions of the financial index return, regardless of daily return or accumulated multi-day return.