I am using CIR to model the term structure in the "real world" (similar to dynamic financial analysis in the insurance industry), so that I can analyze things like exposure to interest rate risk in a VaR framework. As such, I will not be calibrating my parameters to the current term structure or option prices, but to time series data. My question is : given that movements in yields in the past 3 years, in some cases, to 40 year lows, when I calibrate my model to long series of data, the current environment does not even show up in the tails (1st or 5th percentile) of the resulting simulation. Similarly, if I callibrate to a shorter series, then I miss out on the high interest rates of the 1980s and early 1990s. Has anyone dealt with this problem? Any ideas? I considered adding a time varying mean, or moving from a square root volatility process to a process that is more sensitive to levels of rates. Anybody have any thoughts??

Post your e-mail address and I will get in touch. I also have some pretty good references to papers written about the whole issue of utilizing simulation to model a sovereign's exposure to interest rate risk as well as optimal "fixed/floating" mix based on risk and expected cost.

Hi - I have a similar problem trying to calibrate it to HW. The problem I have is if I use something else than a AR(1) process I no longer comply with the HW process - so am not sure if I can use the parameters directly. It will be helpful if you can get in touch entchepl@hotmail.com