Value at Risk Estimates for Brady Bond Portfolios

Abstract

Estimation of value at risk in the case of emerging market bonds is not straightforward for several reasons. This study addresses estimation of VaR statistics for Brady bond portfolios using several versions of the GARCH(1, 1) framework, which explicitly models time-varying volatility. In a hybrid approach univariate ARCH models are used jointly with rolling correlation matrices. Variance-covariance and GARCH methodologies all produce reasonably accurate one-day VaR estimates. A GARCH(1, 1) model with stochastic shocks yields the most accurate VaR measures regardless of horizon, but variance-covariance, and two other GARCH variant models appear to underestimate portfolio risk, especially over longer horizons.