Assuming a directional strategy (no pairs or spread trades) is there a "standard" method for quantifying mean-reversion? Should auto-correlation, variance ratios, hurst exponent, or some other measure be preferred in all cases, or are there advantages to each given the context?
$\begingroup$
$\endgroup$
asked Feb 15, 2011 at 18:46
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
1
There is no standard method and many techniques can work well, including simple time series z-scoring. I'm many cases, I would recommend using the simpler approaches unless the added complexity can be justified.
However, the challenge with all techniques is the proper calibration, which is very much context sensitive. The parameter selection needs to be guided by the characteristics of the specific security and investment horizon. Furthermore, should the parameters themselves be dynamic? Nearly all processes are driven by non-stationary distributions. The amount of leverage used may also be an important consideration.
-
$\begingroup$ What do you mean by "simple time series z-scoring"? Thanks $\endgroup$ Commented Jul 22, 2019 at 3:02
$\begingroup$
$\endgroup$
As usual, no standard method. You can maybe test for unit root over a rolling window, but as usual (again) you are going to be lagging. It will all depend on your choice of window and of the persistence of mean reversion in the market you consider.
answered Feb 16, 2011 at 2:48