The theme of this release deals with calibration of term structure models with focus on calibrating to different samples. We consider CIR and Vasicek models with different volatility specifications and we examine calibration to two samples, one consisting of ATM caps and swaptions and one consisting of caps only, both ATM and OTM.
We perform a series of experiments and reach an interesting conclusion. For relatively simple specifications of the models it is possible to calibrate to ATM caps and swaptions and afterwards we are able to give an acceptable fit to OTM caps. Instead, if we use a more complex model we do not get nearly as good a fit to OTM caps.
This indicates that a model with a simple volatility specification has some advantages compared to a more complex model which does not perform well out-of-sample. We find a similar result if we start by calibrating to OTM caps and afterwards price ATM caps and swaptions.