Purpose
The purpose of this paper is to evaluate the predictive capacity of market risk estimation models in times of financial crises.
Design/methodology/approach
For this, value-at-risk (VaR) valuation models applied to the daily returns of portfolios composed of stock indexes of developed and emerging countries were tested. The Historical Simulation VaR model, multivariate ARCH models (BEKK, VECH and constant conditional correlation), artificial neural networks and copula functions were tested. The data sample refers to the periods of two international financial crises, the Asian Crisis of 1997, and the US Sub Prime Crisis of 2008.
Findings
The results pointed out that the multivariate ARCH models (VECH and BEKK) and Copula-Clayton had similar performance, with good adjustments in 100 percent of the tests. It was not possible to perceive significant differences between the adjustments for developed and emerging countries and of the crisis and normal periods, which was different to what was expected.
Originality/value
Previous studies focus on the estimation of VaR by a group of models. One of the contributions of this paper is to use several forms of estimation.
Comparison of certain value-at-risk estimation methods for the two-parameter Weibull loss distribution
