Market Risk Measurement Models: Estimation of Volatility and Correlation

International audienceThe objective of the paper is to compare the capacity of various market risk measurement models to take into account some empirical properties of financial asset returns. Through an empirical study, we show that financial markets have some empirical characteristics known as "stylized facts" that conventional market risk measurements are unable to reproduce. We propose Value-at-Risk (VaR) measures of market risk, based on dynamic modeling of portfolio volatility and correlations between asset classes, using two risk measurement approaches: the univariate risk measurement approach and the multivariate risk measurement approach. We first describe how to estimate the parameters of these models and then test their quality predictive using backtesting procedures. The results obtained show a great ability of the different risk measurement models to capture the stylized facts characterizing financial markets. Multivariate VaR models integrating correlations dynamic outperform the univariate VaR measures