# Difference between Ornstein-Uhlenbeck, Vasicek and Geometric Mean Reversion

I am currently reading the following 3 paper:

Each of them is terming what to me seems the same stochastic process differently. Paper 1 p. 387 Equation 4 is terming the process Geometric Mean Reversion (GMR) and Equation 5 the exponential version Exponential Mean Reversion (EMR), Paper 2 p. 27 Equation 47 is terming it Vasicek Model and on p. 31/32 Equation 62 Exponential Vasicek Model and Paper 3 p. 4 Equation 3 is terming it Exponential Ornstein-Uhlenbeck process.

Am I correct, that all of these 3 essentially are the same stoachstic process?

Also, for the exponential version two paper seem to offer a closed form solution, where I spotted a small difference. Paper 1 p. 387 Equation 6 has the following solution:

$$S_{T,i}=\exp\left(\ln(S_0)e^{-\alpha T}+\left(\theta-\frac{\sigma^2}{4\alpha}\right)(1-e^{-\alpha T})+\sqrt{(1-e^{-2\alpha T})\frac{\sigma^2}{2\alpha}}\epsilon_i\right)$$

Paper 3 p. 5 Equation 9 on the other hand has the following solution:

$$\ln(X_{t})=\ln(X_{t-1})e^{-\theta \Delta t}+\left(\mu-\frac{\sigma^2}{2\alpha}\right)(1-e^{-\theta \Delta t})+\sigma\sqrt{\frac{1}{2\alpha}(1-e^{-2\theta \Delta t})}\epsilon_i$$

Now different variables aside, the first equation has $$\frac{\sigma^2}{4\alpha}$$ and the second equation has $$\frac{\sigma^2}{2\alpha}$$. Can anyone tell me which one is correct?

• Welcome to Math.SE! I have tried to improve the readability of your question by improving the $\rm \LaTeX$ code. It is possible that I unintentionally changed the meaning of your question. Please proofread the question to ensure this has not happened. Commented Jan 8, 2021 at 10:10

• Ornstein-Uhlebeck = Vasicek (only difference being that Vasicek can sometimes have (deterministic) non-constant reversion speed parameter, whilst Ornstein-Uhlenbeck has all parameters constant by definition).

• Exponetial Vasicek = EMR.

• The term $$4 \alpha$$ in the denominator looks like a typo, should be $$2 \alpha$$

Ornstein-Uhlenbeck process is defined as (I use $$\theta$$ instead of your $$\alpha$$, otherwise the notation is identical):

$$\tag{1} X_t:=X_0+\int_{h=0}^{h=t}\theta(\mu- X_h)dh + \int_{h=0}^{h=t}\sigma dW_h$$

Or in "short-hand" notation we can write:

$$dX_t=\theta (\mu-X_t)dt+\sigma dW_t$$

In an answer to another question on Math SE here, I show how the Ornstein-Uhlenbeck can be solved step-by-step (the bottom section of my answer titled "Why not to use Short-hand notation")

The final step I don't show is to complete the Riemann integral which solves as follows:

$$\int_{h=0}^{h=t}\left(e^{\theta(h-t)}\theta\mu\right)dh=\mu(1-e^{\theta t})$$

So the full solution is:

$$X_t=X_0e^{-\theta t}+\mu(1-e^{\theta t})+\sigma\int_{h=0}^{h=t} e^{\theta(h-t)} dW_h$$

Ornstein-Uhlenbeck and Vasicek (and GMR) processes are the same thing (except that Vasicek process can have the parameter $$\theta$$ non-constant, it can be a deterministic function of $$t$$).

In the first reference you provide, the author introduces the EMR process (or Exponential Vasicek) for $$S_t$$ (i.e. what I call $$X_t$$) as:

$$dS_t=\alpha(\theta - ln(S_t))S_tdt + \sigma_t S_t dW_t$$

Or in long-hand notation:

$$S_t=S_0+\int_{h=0}^{h=t}\alpha(\theta - ln(S_h))S_hdh+\int_{h=0}^{h=t}\sigma_hS_s dW_h$$

We can see right away that this is not the same process as the Ornstein-Uhlenbeck defined in (1).

As for your last question, the $$4\alpha$$ in the denominator looks like a typo, should be $$2\alpha$$ I believe (indeed the author who has $$4 \alpha$$ doesn't derive the solution, he just quotes Brigo. Also see the Wikipedia link at the bottom for why it should be $$2 \alpha$$).

Last but not least, note that Ito integral of any deterministic function is a Normally distributed random variable. Variance of the term $$\sigma\int_{h=0}^{h=t} e^{\theta(h-t)} dW_h$$ is derived here on Wikipedia, and this variance and the Normal distribution property of the Ito Integral are used to rewrite the solution to the EMR process in the form you show in the two equations you quote.

• The first part of your answer is great, however I don't understand the second part yet. I think you took the wrong equation from Paper 1. The equation you took is the EMR. The equation for the GMR (Equation 4 in the paper) is $dS_t=\alpha(\theta-S_t)dt+\sigma dW_t$ which then looks exactly like the short hand version of your equation 1. Also note that paper 1 quotes paper 2 for the exponential version, which yields me further to believe GMR is indeed Vasicek/Ornstein-Uhlenbeck. Commented Jan 8, 2021 at 14:37
• Ok now I understood where the difference lies between both even though you did by mistake use the EMR equation. I didn't see that I have to multiply the equation by $S_t$ which then of course makes it different. What I still did not understand, is the closed form solution in my question the solution of the EMR or the exponential Vasicek/Ornstein-Uhlenbeck? Because in paper 1 it is presented as the solution of the EMR and in paper 3 as the solution of the exponential Vasicek/Ornstein-Uhlenbeck, seems like one of them has to be wrong. Commented Jan 8, 2021 at 15:03
• @Tharmis: you are correct, I took the EMR, not the GMR. I corrected my answer. Commented Jan 8, 2021 at 15:38
• @Tharmis: EMR and exponential Vasicek are the same thing. The solution to the exponential Vasicek is given in paper 2 in equation (63). That can be transformed into the equations you show by using the fact that the last term in equation (63) in paper 2: $$\sigma \alpha^{-\alpha t}\int_s^te^{au}dW_u$$ is normally distributed and has variance $$\frac{\sigma2}{2\alpha}\left(e^{-\alpha|t-u|}-e^{-\alpha (t+u)}\right)$$ to see how that is done, see how Variance can be computed on Wikipedia Commented Jan 10, 2021 at 8:25
• @Tharmis: To be clear: (i) Vasicek = Ornstein-Uhlenbeck = GMR. (ii) Exponential Vasicek = EMR Commented Jan 10, 2021 at 8:28