Hierarchical Agglomerative Clustering Based T-outlier Detection

Diversification is a technique to reduce portfolio volatility. In traditional financial domains, the correlation coefficient has been used as a basis for diversification. However, it is problematic in reality since it only captures a single dimension. This research introduces a unique similarity based framework to identify outliers among high dimensional time series objects in financial markets. As the similarity between two assets decreases in the portfolio, the benefits of diversification increase. The paper proposes an efficient hierarchical agglomerative clustering (HAC) algorithm based on vertical and horizontal dimension reduction algorithms. Finally, this paper proposes a unique similarity measurement definition/calculation based on the time-value function. This paper discloses a series of experiment results illustrating the effectiveness of the framework. The detected outliers can be used to monitor portfolio diversification and therefore mitigate risk