The nature of the correlation structure of traditional and alternative asset classes

• Main Author: Elliott, Keith Eric William
• Published: University of Reading 2012
• Subjects: 332
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.602538

The goal of this thesis is to increase the understanding of 'alternative assets' and their interaction with other asset classes. This is a relevant area of focus as there are currently more assets available to investors than at any other time. Firstly several assets are reviewed to see if they should be considered for further analysis. This process examines the philosophical question of what is an asset class and also considers the ease of investing in each. The asset's return drivers are analysed using statistical and macroeconomic factor models. Most assets are found to be explained by up to seven macroeconomic factors; however, assets such as real estate, gold and wine are not explained well and thus may have portfolio diversification benefits. I then focus my study on the correlation structures of the asset returns. These are examined using rolling correlations and statistical testing of the stability of correlation matrices and correlations are found to time vary. Semicorrelation is adopted to differentiate correlations between those in outperforming and underperforming markets. I find that for many assets, correlations increase in underperforming markets and thus diversification fails when it is needed the most. Government bonds' diversification power is found to improve during underperforming markets and thus these are important for diversification. The final section applies an AG-DCC model to retrieve conditional correlations and study their driving factors using macroeconomic factor models. This model proves that correlations change over time and are asymmetric. I correct for overestimation of goodness-of-fit and my models show an average ability to explain changes in conditional correlations of approximately 7.5%, in some cases this is up to 30%. Two key factors that are found to drive correlations are dividend yield and the oil price; correlations response to factors implies that higher correlations occur during periods of economic underperformance.