A Comparison between MDPP and Kernel Regression Smoothing Techniques for Forecasting Time Series Data
No Thumbnail Available
Date
2005-02-07T02:56:23Z
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
We present a comparison between two smoothing techniques for forecasting time series, Minimal Distance Percentage Principle (MDPP) and non-parametric Kernel Regression. This comparison reveals some important aspects such as the choice of the smoothing parameters, the complexity of the algorithms and the accuracy of the smoothing results used in forecasting. We have chosen to make tests on the predictive power of "head-and-shoulder" pattern.
Financial analysts are overloaded with huge historical stock data which makes searching for chart patterns a difficult job. This is the reason why we need to take smoothing as an integral part of pattern definition, ie., to find, through the smoothing process, those landmarks which are important in the pattern formation. Our final goal is to find as many good instances of "head-and-shoulder" pattern using MDPP and kernel Regression smoothing techniques, and to forecast the future time series prices.