I've estimated the parameters for the Vasicek model $$ dr(t) = a(b - r(t))dt + \sigma dW(t) $$ and the CIR model $$ dr(t) = a(b - r(t))dt + \sigma\sqrt{r(t)} dW(t) $$ to one-year Treasury yield data from 1974 (which were around 8% then!). Let's say the estimates I got were $$ Vasicek: a = 3.2, b = 8.1, \sigma = 6.0 \qquad (1)\\ CIR: a = 3.2, b = 8.1, \sigma = 2.3. \qquad (2) $$ N.b. these values correspond to $r(t)$ in percent, not decimal. So, my dimensions are short rate level measured in percent (%), and time, say, measured in seconds (s). The units of the parameters are then $$ a = s^{-1}, b = \%, \sigma = \%/\sqrt{s} $$ My question is, how does one intuitively interpret the estimated values in (1) and (2)? I.e., I'm trying to think of the process as a physical process, and so what does a "mean reversion speed of $3.2 / s$" mean, e.g.? Actually, it seems that calling $a$ a "speed" is a misnomer, given the units.

Any insights welcome!


The processes revert towards their mean with the speed $E(dr(t)/dt) =a*(b-r(t))$ so $a$ is not the speed itself, only one factor of it. If $\sigma$ would vanish then $a$ would be $ln (2)$ times the inverse of the time - hence its unit $s^{-1}$- it would take for $b-r(t)$ to halve. Think of a as the decay-factor of the deviation from the mean!

| improve this answer | |
  • $\begingroup$ @M Lind, could you please be so kind to use the Latex formatting capabilities in your answer? Thanks. $\endgroup$ – Quantuple Aug 8 '16 at 12:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.