Stock market trend prediction using a sparse Bayesian framework

Abstract

The aim of this study is to develop a relevance vector machine-a RVM classifier for trend prediction of the BELEX15 index of the Belgrade Stock Exchange. In addition, the RVM model is compared to two `similar' methods: support vector machines - SVMs and least squares support vector machines - LS-SVMs to analyze their classification precisions and complexity. The test results indicate tha tRVMs outperform benchmarking models and are suitable for short-term stock market trend predictions.