A critical challenge in managing quantitative funds is the computation of volatilities and correlations of the underlying financial assets. We present a study of Kendall's t coefficient, one of the best-known rank-based correlation measures, for computing the portfolio risk. Incorporating within risk-averse portfolio optimization, we show empirically that this correlation measure outperforms that of Pearson's in our out-of-sample testing with real-world financial data. This phenomenon is mainly due to the fat-tailed nature of stock return distributions. We also discuss computational properties of Kendall's t, and describe efficient procedures for incremental and one-time computation of Kendall's rank correlation.
Robust and Sparse Estimation of the Inverse Covariance Matrix Using Rank Correlation Measures