Dude, where’s my mean reversion?

Posted by on Sep 8, 2011 in ArbMaker News! | No Comments

What is the point of knowing cointegration is present if the mean reversion coefficient is unknown?


The mean reversion coefficient, denoted by “MRC” in ArbMaker, indicates how fast a cointegrated pair of companies’ prices come back to their long-term mean equilibrium. It is derived from an “error correction mechanism” equation built into the Engle-Granger 2 step method (which is actually 4 steps but that’s another post).

A valid MRC (which in turn validates the cointegration) is negative indicating adjustment by the dependent variable/company back to equilibrium in response to changes in the independent variable/company.

Positive MRC suggests either (i) that you are looking at the independent variable/company or (ii) that although the cointegrated time series may be stationary, it may also be auto-correlated.

Finally, if the cointegrated relationship is stable MRC will be found between the values of zero and absolute 1.

Bottom line: if you don’t know the MRC you don’t know enough about the pair. Saying “Dude, it’s cointegrated!” can be kind of like saying “Dude, it’s a llama!”.

Leave a Reply