Selecting Mean-Reverting Pairs and Managing Slow Reversions
Summary
The discussion addresses how to estimate or manage the time a mean-reversion trade takes to return toward its mean. It notes the Ornstein–Uhlenbeck process as one way to model half-life, but shifts attention to whether a candidate spread has a credible reason to revert and whether the apparent behavior is statistically significant. For identified reverting episodes, a trader can model their durations and compare an open trade’s elapsed time with a chosen percentile of historical episode lengths.
An illustrative pairs-trading example marks entries when a spread crosses a one-standard-deviation threshold and exits when it returns to its mean. Some marked trades profit, while others fail to revert promptly; closing those positions incurs costs that must be weighed against successful trades. The examples are conceptual and do not establish performance. Pair selection is presented as central, while no formal duration estimator, validation procedure, or complete stop rule is specified.
Key ideas
- An Ornstein–Uhlenbeck model is one approach to estimating mean-reversion half-life.
- A plausible mechanism and statistically significant evidence for reversion matter when selecting a spread.
- Historical reversion durations can provide a benchmark for deciding whether a trade is taking unusually long.
- Threshold entries and mean-crossing exits can produce both profitable reversions and failed trades.
- Pair selection and the costs of exiting failed trades affect whether a strategy is viable.
Tags
Full text
# Mean Reversion Time Frame # Mean Reversion Time Frame I am running a mean reversion strategy. I have question with regards to half-life; I have heard of OU process to determine the half-life but it's not giving me that kind of result. Can anyone advise another way to identify half-life or else any other indicator that I can use to trade to support my Z score? ## Answer by bill_080 (score 8, accepted) https://quant.stackexchange.com/a/1240 The OU process is: http://en.wikipedia.org/wiki/Ornstein-Uhlenbeck_process Here's an example of the use of the OU method. http://epchan.blogspot.com/2007/01/what-is-your-stop-loss-strategy.html To me, the problem is identifying processes that actually have a reason to bleed down to the mean, and that show statistically significant results. If you can identify the "bleeds", then you can build a model of a "bleed". From that, you can determine if the current "bleed" is taking longer than 95% (or whatever % you like) of the "bleeds" that you modeled. Edit 1 ================================= To illustrate some considerations, I'll use the picture from one of Ernie's examples at this link: http://epchan.blogspot.com/2006/11/cointegration-of-oih-with-spot-oil.html As shown below, I added green "entry points" (1 through 7) where you would "get in" as the blue line drops down through the top one-sigma red line. I added red "exit points" (A through C) where you "get out" as the blue line drops through the mean zero line. As you can see, trades 1-A, 2-B and 5-C are all profitable "bleeds". However, entries 3, 4, 6, 7, (and others that I didn't label) are garbage. You'll have to come up with a way to get out of those positions without wasting too much time and money. The costs of closing out the garbage have to be subtracted from the gains of the bleeds to determine if this pair-trade is worth trading. Now look at another potential pair: http://epchan.blogspot.com/2006/11/reader-suggested-possible-trading.html Using the same general idea, this trade might have less garbage to deal with. My point is, choosing the right pair is probably the most important part of pairs trading.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.