Accurate of VAR Calculated Using Empirical Models of the Term Structure

This work compares the accuracy of different measures of Value at Risk (VaR) of fixed income portfolios calculated on the basis of different multi-factor empirical models of the term structure of interest rates (TSIR). There are three models included in the comparison: (1) regression models, (2) principal component models, and (3) parametric models. In addition, the cartography system used by Riskmetrics is included. Since calculation of a VaR estimate with any of these models requires the use of a volatility measurement, this work uses three types of measurements: exponential moving averages, equal weight moving averages, and GARCH models. Consequently, the comparison of the accuracy of VaR estimates has two dimensions: the multi-factor model and the volatility measurement. With respect to multi-factor models, the presented evidence indicates that the Riskmetrics model or cartography system is the most accurate model when VaR estimates are calculated at a 5% confidence level. On the contrary, at a 1% confidence level, the parametric model (Nelson and Siegel model) is the one that yields more accurate VaR estimates. With respect to the volatility measurements, the results indicate that, as a general rule, no measurement works systematically better than the rest. All the results obtained are independent of the time horizon for which VaR is calculated, i.e. either one or ten days