SERVING THE QUANTITATIVE FINANCE COMMUNITY

 
stc
Topic Author
Posts: 1
Joined: December 4th, 2018, 8:38 am

Out of sample selection

December 4th, 2018, 8:51 am

Hi everyone,

First  post so please be gentile.

I have created a simulation for a number of stocks.  I do all of my testing in-sample and rolling forward a little out of sample (to obtain the performance statistics). I move this window through time to create a performance time-series for each stock I look at  I have noticed that when I stick together the out-of-sample results a few stocks are very consistently under performing through the entire back test period. 

My question is: I don't want to introduce any bias into my simulation by selecting stocks from the out-of-sample window instead of the in-sample window, however why would I include these very poorly performing stocks if they have not ever performed out of sample?  My assumption being that this trend will continue.  I could exclude these stocks and  back test on a regular basis to see if the performance changes?

Any views on this would be much appreciated.

Thanks for any help 
 
User avatar
ppauper
Posts: 70239
Joined: November 15th, 2001, 1:29 pm

Re: Out of sample selection

December 4th, 2018, 9:02 am

If you think those stocks are going to continue to underperform consistently, then you may have found free money.
Short the underperformers as part of a long-short strategy

If you buy stocks that are performing well with the hope that they will continue to perform well, that's price-momentum

On the opposite side of the argument, there's a strategy, Dogs of the Dow, where one buys the stocks with the worst dividend yield (which are therefore underpriced according to that metric)
 
User avatar
ISayMoo
Posts: 1398
Joined: September 30th, 2015, 8:30 pm

Re: Out of sample selection

December 18th, 2018, 5:37 pm

It seems that for every off-the-shelf strategy there is another OTS strategy which say you should do the opposite.
 
User avatar
ppauper
Posts: 70239
Joined: November 15th, 2001, 1:29 pm

Re: Out of sample selection

December 18th, 2018, 5:49 pm

It seems that for every off-the-shelf strategy there is another OTS strategy which say you should do the opposite.
Bulls make money, bears make money, pigs get slaughtered
 
User avatar
katastrofa
Posts: 7052
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: Out of sample selection

December 18th, 2018, 9:55 pm

You can phrase some hypotheses why those particular stocks underperform and test it on new data. This could help you generalise your model to the whole market (e.g. build a binary classifier telling you to what kind of stock your model applies).
It depends on what you have there, but e.g. you can divide your basket randomly in half. Train your model and phrase the hypothesis about the result. Test the hypothesis on the other half.
 
User avatar
katastrofa
Posts: 7052
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: Out of sample selection

December 18th, 2018, 11:03 pm

It seems that for every off-the-shelf strategy there is another OTS strategy which say you should do the opposite.
Bulls make money, bears make money, pigs get slaughtered
Yeah, yeah... And the groundhog sits there and wraps them in tinfoil.
ABOUT WILMOTT

PW by JB

Wilmott.com has been "Serving the Quantitative Finance Community" since 2001. Continued...


JOBS BOARD

JOBS BOARD

Looking for a quant job, risk, algo trading,...? Browse jobs here...


GZIP: On