Features extraction based on particle swarm optimization for high frequency financial data

A novel stock trading system is developed in this paper. The trading system combines particle swarm optimization based clustering method and basic financial rules to discover the potential features of high frequency financial data. By analyzing the robustness of PSO based trading system and comparing with the classical buy and hold trading policy, the empirical study gives evidences that the newly proposed trading system can be used as a decision support system for stock investors.

I. Introduction

Particle swarm optimization (PSO) [1] is a newly natural computing method that optimizes a problem by iteratively trying to improve a candidate solution with regard to a given measure of quality. Plenty of academic work have been spent to analyze the potential of particle swarm optimization and it has undergone many changes since its introduction in 1995 [2], [3]. Almost every aspect of global optimization problem has been investigated by means of particle swarm optimization. In recent years, particle swarm optimization is employed to data clustering [4], [5], [6]. To the best of our knowledge, most of academic work about PSO based data clustering focus on theoretical study and comparing with other classic clustering methods based on the UCI machine earning database. Having this in mind, our motivation is apply the PSO based data clustering to financial markets.