2
$\begingroup$

Under the assumption of the normal distribution, I'm trying to create a single stock mean reversion strategy. I took the log returns because they are stationary, I standardized them and used the zscore as trading signals(buy when zscore is -2, take gains when the zscore is 0, sell when zscore is +2 and take gains when zscore goes to 0. This strategy comes from Ernest chan's books and blogs.

I backtested it and the results are good. But I'm unsure about the statistical logic behind it.

Moreover I did not find any serious research paper about this strategy, and Ernest Chan's books don't really detail why he uses the Zscore.

Does this strategy makes sense or is it dumb, and why ?

$\endgroup$
8
  • $\begingroup$ I'm not sure the assumption about the log-returns being stationary is valid. Looking at this I am seeing a lot of changing slopes, so you may be getting varying results based on underlying asset and time period. Look to see if you are getting a lot more buy signals than sell signals. $\endgroup$ Apr 29, 2022 at 16:11
  • 1
    $\begingroup$ Well I just checked quickly, I would say it's almost 50-50 on average across all the stocks I tested. $\endgroup$
    – Alec Ric
    Apr 29, 2022 at 19:21
  • $\begingroup$ When you compute the z-score, do you do it on the whole dataset, or on a rolling basis? $\endgroup$ Apr 29, 2022 at 22:03
  • 2
    $\begingroup$ The question is not well posed. Are you working in a price space or a return space? How are you determining your time frames? What types of markets are you in? How liquid, particularly depth, is the market you are in? Pluggin in a parameter to a z-score can always be dominated by other techniques, so do you have an opponent that can take advantage of you? Step by step, what is your behavior. What kind of measurements are you taking, when and why? $\endgroup$ Apr 30, 2022 at 4:14
  • $\begingroup$ – rubikscube09 . I do it on the whole data set $\endgroup$
    – Alec Ric
    Apr 30, 2022 at 16:36

1 Answer 1

1
$\begingroup$

I don't know of any statistical basis for the phenomenon of mean reversion. It really defies the rule of independence which is usually part of statistical methods. Other than that, it is also counter to momentum investing which go against precisely what you are intending to do. Countering that with the positive skew of long term stock price movement, I think that you might make some money from it, but would be much less that what you would make with trend following. Also you would be subject to the same risks as the general stock market. I would only trade it with stop losses.

$\endgroup$
2
  • 1
    $\begingroup$ I see. Indeed this strategy has no statistical value. Thank you for your answer. $\endgroup$
    – Alec Ric
    Apr 30, 2022 at 16:48
  • $\begingroup$ You're welcome. Indeed, the 'dumbest questions' are sometimes the one's that make you rethink the status quo. $\endgroup$ Apr 30, 2022 at 17:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.