A statistic to capture the degree of mean reversion

196 Views Asked by At

Given a realization of a stochastic process, $x_{t_1}, x_{t_2}, \ldots, x_{t_n}$, is there a simple statistic that captures the degree to which the stochastic process is mean reverting?

For example, such a statistic would give a high value for a realization of an Ornstein-Uhlenbeck process, and a low value for Brownian motion or geometric Brownian motion.

1

There are 1 best solutions below

0
On

In the quantitative finance literature, you would attempt to fit an OU process to the data to get an estimate of the mean reversion parameter ($\eta$).

$$X\sim OU(\eta,\bar{x},\sigma)\implies dX_t = \eta(\bar{X}-X_t)dt +\sigma dW_t$$

Note that Brownian Motion is simply the OU process with $\eta=0$ (in line with your expectations).

Actual estimation of $\eta$ is not trivial, but here is a paper that outlies some ways to approach measuring mean reversion in stochastic models: http://www.investmentscience.com/Content/howtoArticles/MLE_for_OR_mean_reverting.pdf