Accelerating the Multivariate SKEW T Parameter Estimation

This paper considers an acceleration scheme for the multivariate skew t (MST) parameter estimation. MST is a heavy-tailed distribution allowing also for skewness. The distribution is very convenient for many real-life applications where there data is heavy-tailed, there may be outliers, and data may be skewed. One example is financial data. The classical method to estimate the MST parameters is the maximum likelihood estimation (MLE) via the expectation-maximization (EM) algorithm, which is often criticized for slow convergence. In this paper, we propose a scale-adjusted MST (SA-MST) model to accelerate the MST parameter estimation. Numerical results show that our proposed acceleration scheme can converge much faster than the standard MST model.