Short Interest Data

[SeongDeokKo]

Hello

I downloaded the short interest data.

According to the definition in the signal document, it is the ratio of the number of stocks yet to be covered to the number of shares outstanding.

Because of the definition, I think it is reasonable the values are within (0,5).

But I found that values are really large.
I checked whether it indicates the value in percent but found it false.

Please let me know if I misunderstood something.

Thank you

[chenandrewy]

Hi @SeongDeokKo : I just want to thank you for contributing this issue and the other one. Sorry but I think Tom and I are bogged down with all sorts of other work right now. I'm struggling to get the 2025 update to this dataset on the calendar 😥. But we will get back to you!

[SeongDeokKo]

Hello. Investigating the source of mismatch I found something.
To double check the shortinterest, I used the shortinterest file from FINRA (https://www.finra.org/finra-data/browse-catalog/equity-short-interest/files)

I found that we just need to divide the value by 10^6

When I multiply the shrout to the ShortInterest
I found the shortinterest value is really similar to that of FINRA reported

We just need to adjust the value by 100,000

[tomz23]

Thank you @SeongDeokKo for the issue and the hint! I looked into it a bit, what happens is:

1. Compustat Short Interest data (identical to FINRA) are reported in raw units
2. We get shares outstanding from CRSP which reports it in thousands but we change it to millions after downloading the data in I_MonthlyCRSP.do

So it really makes sense to report short interest in millions as well which I did in the commit here: c403f68

Notably, this does not change any of the portfolio returns of the affected predictors (ShortInterest, IO_ShortInterest, Recomm_ShortInterest) because rescaling does not change the ordering.

Thanks for investigating the issue!

[SeongDeokKo]

Hello It's me again.
I thought everything is done.
But recently I found out that there are 10 cases where the 10^6 divided short interests exceed 10.(1000%)

I cannot find what is the problem.

In the data, I found that weird cases happened after 2020.