Interval Forecasting of Stock Price Changes using the Hybrid of Holt’s Exponential Smoothing and Multi-Output Support Vector Regression

Given the importance of investment in stock markets as a major source of income for many investors, there is a strong demand for models that estimate the future behavior of stock prices. Interval forecasting is the process of predicting an interval characterized by two random variables acting as its upper and lower bounds. In this study, a hybrid method consisting of Holt’s exponential smoothing and multi-output least squares support vector regression is used to forecast the interval of the lowest and highest prices in a stock market. First, Holt’s smoothing method is used to smooth the two bounds of the interval and then the residuals of the smoothing process are modeled with multi-output vector support regression. The output of the regression step is the error of the two bounds of the interval. The method is implemented on the weekly data of the overall index of the Tehran Stock Exchange from 1992 to 2016, with the interval defined as the distance between the lowest and highest overall index values. The results demonstrate the high accuracy of the hybrid method in producing in-sample and out-of-sample forecasts for the movement of the two bounds of the interval, that is, the weekly highs and lows of the overall index. Also, the hybrid method has achieved a lower mean squared error than the Holt’s smoothing method, indicating that multi-output vector support regression has improved the performance of the smoothing method