CFE 2019: Start Registration
View Submission - CMStatistics
Title: Financial systemic risk prediction with non-Gaussian orthogonal-GARCH models Authors:  Marc Paolella - University of Zurich (Switzerland)
Patrick Walker - University of Zurich (Switzerland) [presenting]
Abstract: Several financial systemic risk indicators have been proposed after the great financial crisis with the goal of quantifying risks inherent in the markets and to anticipate future crises. One of the most popular stress indicators is based on the leading eigenvalues of the covariance matrix of a set of returns and describes the level of interconnectedness of financial assets. Originally, this stress indicator is computed from the sample covariance matrix and its dynamics are thus mainly determined by the sample size. Moreover, the sample estimator is sensitive to outliers, leading to distorted systemic risk measures. We investigate computing the risk indicator based on the forecasted conditional covariance matrices from various MGARCH models, such as CCC-, DCC- and O-GARCH. An alternative measure derived from the conditional correlation matrix is discussed. We propose a novel asymmetric, fat-tailed O-GARCH model and present an EM-algorithm for maximum likelihood estimation. Using this new robust O-GARCH model, we compute the systemic risk indicator from the implied predicted conditional correlations and achieve realistic dynamics and out-of-sample forecasts. Finally, an application to tactical asset allocation shows the economic value of the risk indicator in anticipating equity market drawdowns.