File Exchange

image thumbnail

White Reality Check

version 1.3.0.0 (2.23 MB) by Arnout Tilgenkamp
A reality check for data snooping by Halbert White, uses the stationary bootstrap by Politis&Romano

7 Downloads

Updated 15 Oct 2013

View Version History

View License

This is a test used for significant performance, while taking into account the data mining issue.

I have added the paper of Politis and Romano describing the bootstrap; the paper of White describing the test! These can also be found free of charge on the internet.

Finally I have added a paper where I am the co-author. This paper can only be found here, and it describes the use of the White reality Check for testing trading strategies for significant returns.

Note: While the code is publicly available to promote research on this topic, I would very much appreciate it if a footnote is used (mentioning the source of the code) when used for papers.

Kind regards

Cite As

Arnout Tilgenkamp (2021). White Reality Check (https://www.mathworks.com/matlabcentral/fileexchange/34306-white-reality-check), MATLAB Central File Exchange. Retrieved .

Comments and Ratings (8)

Giorgio Mirone

Angelos Pazaitis

I get different p values when I test the "best" model vs the benchmark and when I test 10 models in which the "best" model is included against the benchmark..

Is this correct?

Shouldn't be the pvalue for the 10 models at least as low as when using the only the one "best" model?

Arnout Tilgenkamp

Waiting for update to be accepted by Mathworks

Arnout Tilgenkamp

@ Eduard, after checking I think you are right, I seem to have made a mistake. Testing vs alternative models is therefore incorrect, however testing vs zero return is still valid since, there is no reason to bootstrap a nonvarying vector. I will make the adjustment shortly

thx for the feedback

Arnout Tilgenkamp

@ Huan , it is valid to resample the strategy returns. To circumvent the issue with zeros I would suggest to start your resampling from day 'i' where all strategies have had the possibillity to trade.

One can also remove all collective zeros, but I am not sure whether this is valid and would influence the asymptotic properties of the test statistic. I need to verify whether this is valid.

Thx for the feedback do

Huan

Think so, too. And plus, Got question about reality test itself:is it right to just resample the strategy returns instead of the original data set? Like for many strategies the first N days return will always be 0 due to the first signal can only be produced after the Nth day. If just resample the strategy return, this will be violated, isn't it?
BTW, thanks loads for sharing~~

Huan

Think so, too. And plus, is it right to just resample the strategy returns instead of the original data set? Like for many strategies the first N days return will always be 0 due to the first signal can only be produced after the Nth day. If just resample the strategy return, this will be violated, isn't it?
BTW, thanks loads for sharing~~

Eduard

I have a short question,
I had a look at the PolitisRomanoBootstrap-Function and I realized that you are only boostrapping the row index of the 'alternative' and not of the 'mat' (benchmark models), I think that's not right, you need to bootstrap the whole Lostfifferential, meaning that 'mat' should be bootstrapped with the same row index like the 'alternative'!
Am I right??

MATLAB Release Compatibility
Created with R2011b
Compatible with any release
Platform Compatibility
Windows macOS Linux

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!