I am looking for help in parameterizing the CIR short rate model. I followed the procedure of using an OLS regression to estimate the mean reversion and long-run equilibrium terms, and then regressing the squared residuals from the OLS against the lagged (t-i) rate values toi estimate the volatility parameter. The intention is to use these results, combined with the closed form formula for the expected squared error to estimate the volatilites. From the regression E(et^2) = a1 + a2(rt-1) + wt, we get two estimates for the volatility: sigb = Sqr((a2 * a) / (Exp(-a) - Exp(-2 * a)))siga = Sqr((a1 * (2 * a)) / (beta * (1 - Exp(-2 * a)) ^ 2))The question I have is, when I run this resudial regression, either one of the intercept or the coefficient becomes negative for each data series I ma testing. Thus, one of my vol estimates is undecideable. For the dediceable estimate, I am coming out with unexpected number. Has anyone tried this methed before that can point me in the right direction? Thanks, - Mark

BumpMaDee: Do you have a source on the OLS method you're describing?

Check Sherev's notes. It's available online.