Modeling spread risk for interest rate products, i.e., changes of
the yield difference between a yield curve characterizing a class
of equally risky assets and a riskless benchmark curve, is a
challenge for any financial institution seeking to estimate the
amount of economic capital utilized by trading and treasury
activities. With the help of standard tools this contribution
investigates some of the characteristic features of yield spread
time series available from commercial data providers. From the
properties of these time series it becomes obvious that the
application of the parametric variance-covariance-approach for
estimating idiosyncratic interest rate risk should be called into
question. Instead we apply the non-parametric technique of
historical simulation to synthetic zero-bonds of different
riskiness, in order to quantify general market risk and spread
risk of the bond. The quality of value-at-risk predictions is
checked by a backtesting procedure based on a mark-to-model
profit/loss calculation for the zero-bond market values. From the
backtesting results we derive conclusions for the implementation
of internal risk models within financial institutions.