T-outlier and a novel dimensionality reduction framework for high dimensional financial time series

Abstract

Outlier trading strategy is one of the more popular strategies in the financial markets. In reality, the outlier trading strategy tends to provide a higher return and is accompanied by a respective higher risk. Even so, more fund managers and investment companies turn to outlier trading strategies to maximize their return. The key is to minimize the risk by finding the most accurate and relevant outliers. The framework presents a T-outlier concept in high dimensional time series data space. The framework generalizes the characteristics of time series outliers and proposes a novel dimensionality reduction framework in high dimensional time series data space of financial markets to support further mining. The framework consists of horizontal dimensionality reduction and vertical dimensionality reduction algorithms. On the vertical dimension, an Attribute Selection algorithm reduces the vertical dimension by identifying significant dimensions among a given set of attributes that closely represent the data characteristics. On the horizontal dimension, a Segmentation and Pruning algorithm reduces the selected attributes’ horizontal data points to concisely represent the data object in piecewise linear representation form. This paper discloses a series of experiment results that illustrate the effectiveness of the framework.