Value at Risk for Corporate Bond Portfolios

Abstract

Risk measurement systems (particularly daily value at risk (VaR) systems) typically reflect a compromise between the competing objectives of delivering cost-effective and timely but reasonably accurate results. Judicious choices with respect to the data used and computational aspects can be made to reduce the overall costs and computation time. An important question is how these design choices affect the quality of the risk measures produced. An examination of producing a daily VaR for portfolios of credit-risky bonds and loans indicates that rating-level index returns exhibit time series properties consistent with GARCH-like effects. Incorporation of these time series properties in index test portfolios leads, on balance, to improved VaR forecasts. For test portfolios constructed from individual bonds, however, VaR performance relying on a single risk factor is somewhat deficient. Incorporating a second “generic idiosyncratic” risk factor leads to significantly improved VaR performance.