Quantitative Finance Stack Exchange is a question and answer site for finance professionals and academics. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I'm in the process of building a quantitative trading model, I want to improve on the way in which I decide upon a look back length for the indicators. I understand the different pros/cons for very short and very long look backs, but rather I want to assess what the optimal length is between say 20-days, 30-days, or in between. I feel the choice of 20 or 30 is done without thinking to much and just because it is round, which seems lazy to me. Is there a better method?

share|improve this question
up vote 5 down vote accepted

I think you are having it backwards: Optimising your lookback period is a sure recipe for disaster because it introduces data snooping bias.

To develop a robust trading strategy you have to check whether it is sufficiently stable with different lookback periods (e.g. in a certain range). If results differ significantly that is a good sign that your system won't work out-of-sample!

I agree with you that many people do these things without thinking and are indeed lazy... this is one of the reasons why so many trading systems fail under real-world conditions.

share|improve this answer
Thanks @vonjd that was my problem, I didn't want to optimise for a specific look back because of the various issues it introduces, but also felt that just choosing any period without thinking/analysing was default thinking. But you have definitely cleared up my thinking – Celeste Jul 24 '14 at 10:18
I disagree with this answer. Optimising lookback does not imply that this was used to guide variable selection (which is the assumption made in this answer). Even then, it's only true for low frequency strategies. – user2763361 Jul 25 '14 at 11:47
@user2763361: The only one who is talking about "variable selection" is you. Why should optimising other parameters (like lookback periods) be any better? And why should something that is a statistical problem in low frequency trading magically vanish in the HF domain?!? – vonjd Jul 25 '14 at 14:24
The OP was referring to indicators (i.e. variables), then you said that the strategy needs to work with different lookback periods. So for his strategy, this would be the features. Anyway, this doesn't matter, my comment applies generally. I don't see any statistical issue with creating a model that works for multiple lookbacks, then optimising for a single lookback after the model has been finalised. The number of hypotheses that you tested in your research is equivalent, so your "test size" is the same (test size being a mere analogy to hypothesis testing, not to be taken literally). – user2763361 Jul 26 '14 at 8:41
@user2763361 The OP is not referring to indicators as such but to lookback lenghts of indicators (this is even the title of the question). Furthermore he is not talking about models that work for multiple lookbacks (you made this up) but about optimising lookback periods in the range of very long and very short (he mentions that explicitly). A model will not be finalized until you haven't decided on all parameters used (lookback period being one of them). So at best you are addressing a completely different question. – vonjd Jul 26 '14 at 16:37

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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