Self-projecting time series forecast - An online stock trend forecast system

Abstract

This paper explores the applicability of time series analysis for stock trend forecast and presents the Self-projecting Time Series Forecasting (STSF) System we have developed. The basic idea behind this system is online discovery of mathematical formulas that can approximately generate historical patterns from given time series. SPTF offers a set of combined prediction functions for stocks including Point Forecast and Confidence Interval Forecast, where the latter could be considered as a subsidiary index of the former in the process of decision-making. We propose a new approach to determine the support line and resistance line that are essential for market assessment. Empirical tests have shown that the hit-rate of the prediction is impressively high if the model were properly selected, indicating a good accuracy and efficiency of this approach. The numerical forecast result of STSF is superior to normal descriptive investment recommendation offered by most Web brokers. Furthermore, SPTF is an online system and investors and analysts can upload their real-time data to get the forecast result on the Web. Keywords: Self-projecting, forecast, Box-Jenkins methodology, ARIMA, time series, linear transfer function.