A Graphical Model Applied to Hedge Fund Indices

Abstract

This thesis studies the relationships between major hedge fund indices over a sixteen-year period, from January 2000 to December 2015. We first separate this time period into normal' and crash' regimes using a trend filtering model, which we use to calculate the correlation matrix of the selected indices for both regimes. We then apply a graphical model and PCA clustering algorithm to the index returns over the entire period, over a 7-year rolling window that spans the period, and over both normal and crash regimes. We find that the Managed Futures index is consistently selected by the graphical model across both regimes, while the Equity Market Neutral and Dedicated Short Bias indices are selected by both models for most time windows.