![IconGems-Print](images/IconGems-Print.png)
411 – Copula Model and Maximum Likelihood Estimation
Multi-Level Time Series Clustering: Issues With Traditional Risk Management Frameworks
Michael Kotarinos
University of South Florida
Kim Doo Young
Sam Houston State University
Chris P. Tsokos
University of South Florida
Multi-Level Time Series Clustering (MLTC) is a distance based technique which allows for efficient clustering of time series data. While MLTC has previously been used to study the characteristics of different sectors and form a diversified portfolio it has not been used in context of broader problems in asset allocation. In particular, there is no appealing automated procedure for the generation of portfolios based on complex preferences over risk. In this study the authors first reviewed historical literature relating to asset allocation and sector weighting with a focus on optimization under the Sharpe ratio and current practices in the industry. After briefly reviewing the relevant literature in the machine learning field and the properties of MLTC clusters, the authors proceeded to generate an alternative to portfolios generated under the CAPM framework by modifying the clustering criteria of MLTC based on a decision theoretic Multi Criteria Decision Analysis (MCDA) Framework representing preferences over risk.