Delisting return adjustment

[mk0417]

Line 43 in Signals/Code/DataDownloads/I_CRSPmonthly.do: replace ret = ret + dlret.
In this way, 62 observations of delisting adujsted ret are less than -1. The minimum ret is -1.99

Is it better to use replace ret = (1+ret) * (1+dlret) - 1 as suggested by WRDS? There will be no delisting adjusted return that is less than -1.

[chenandrewy]

Yes, I agree @mk0417 . Thanks! Here's a link to the WRDS suggestion:

https://wrds-www.wharton.upenn.edu/pages/support/support-articles/crsp/stock/using-ret-return-variable-and-dlret-delisting-return-determine-return-last-trading-date/

and I think we can add this to the to do list in the update for early 2022.

[tomz23]

Emailed suggestion:

The code does (dlstcd == 500 | (dlstcd >=520 & dlstcd <=584))

But Shumway (1997), Bali et al (2016) suggest that 500, 520, 574, 580, 584 and (551 to 584) are the relevant codes. I take that to mean code such as 575,581,582 and 583 are not be taken into account. There are companies in crsp with these delisting codes. From Jam 1980 to Dec 2020 there are around 260 Permnos with these codes.

Q. Is my understanding correct that codes 575,581,582 and 583 are not to be taken into account and if so should the codes not be made more explicit?

[tomz23]

Emailed suggestion 2 (with delisting in Sep 1992):
Notice that the delisting payment date 01 oct 1992 is in the month after the delisting month. In this particular case the next day, so not much of an issue. I understand your code does ret + dlret. For the month of September 1992.

Q1. In general, would (1+ret) (1+dlret) -1 be a better way to compute this ?
Q2. When the payment date is further out, should one take the return for September as -.20 and the return for October as -.33 ? Does the date of the delisting payment have no relevance?

[chenandrewy]

Delisting returns are used twice in the repo: (1) in Signals/ and (2) in Portfolios/. (2) is probably much more important, and we should probably close this issue out after updating (2).

[chenandrewy]

I think I disagree with the emailed suggestion 1 (on (dlstcd == 500 | (dlstcd >=520 & dlstcd <=584))). Shumway's (1997) table I looks like his adjustments matter for all dlistcd 500 or anything in 520-584:

The footnote at the bottom has the list the emailed suggestion points out (500, 520, 574, 580, 584 and (551 to 584)), but my read is that this list is for illustrative purposes only. I think the other delisting codes are omitted just because they're hard to describe in a short footnote. Here are some descriptions from crsp.org:

575 Delisted by current exchange - company request, offer rescinded, issue withdrawn by underwriter.

581 Delisted by current exchange - failure to register under 12G of Securities Exchange Act.

I also note that Shumway states:

so 581 should be a performance related problem.

Long story short: I'm not an expert on delistings, but the code is fine with regards to emailed suggestion 1.

[chenandrewy]

Regarding emailed suggestion 2:

Q1: yes, we should be using the compounding version (1+ret)(1+dlret)-1.

Q2: this is a tricky issue and my view is that we should not move the return further out. Because, on the performance-related delisting event, clearly the market value has dropped, but we don't have a market value (it's delisted). So even if the dlret payment doesn't come for a couple months, it's still the best proxy we have for the unobserved market value.

[chenandrewy]

@mk0417's comment is finally fixed for Signals/Code/DataDownloads/I_CRSPmonthly.do here: 1b25a90.

Still got to patch up the portfolios code

[chenandrewy]

Patched Portforlios/Code/11_ProcessCRSP.R 5e08b08