Support Vector Regression for Time Series Analysis

Abstract

In the recent years, Support Vector Machines (SVMs) have demonstrated their capability in solving classification and regression problems. SVMs are closely related to classical multilayer perceptron Neural Networks (NN). The main advantage of SVM is that their optimal weights can be obtained by solving a quadratic programming problem with linear constraints, and, therefore, standard, very efficient algorithms can be applied. In this paper we present a 0–1 mixed integer programming formulation for the financial index tracking problem. The model is based on the use of SVM for regression and feature selection, but the standard 2–norm of the vector w is replaced by the 1–norm and binary variables are introduced to impose that only a limited number of features are utilized. Computational results on standard benchmark instances of the index tracking problems demonstrate that good quality solution can be achieved in a limited amount of CPU time.