Hi Grant,

You're right. It looks like we incorrectly have the split occurring on 5/7/2015. A fix is in the works. Thanks for reporting the problem.

Hello Jamie -

Thanks for the follow-up. A few questions/comments:

1. Any idea how missing split errors end up in the data? It is very mysterious, looking in from the outside. Presumably, splits are public knowledge, and are handled openly and systematically (and efficiently, since https://www.splithistory.com/ offers up the data for free, without even ads on their website). Is there something about how Quantopian works that makes addressing splits particularly challenging?
   
2. At one point, I started to do a head-scratch on how the research platform might be used to search for such errors across all stocks and time. Apparently there are some challenges (or you or one of your 90,000 users would have done it by now). Any guidance? Naively, I'd think that correlating changes in shares outstanding with overnight price jumps would do the trick, but maybe since the data aren't exactly 1:1, one needs a "fuzzy" algorithm and every case needs to be reviewed in detail (but then, data such as is available on https://www.splithistory.com/ could be used to firm up the decision).
   
3. At the risk of beating a dead horse, I continue to recommend a user-facing database of known gross data errors, and the means to access the database from within the research platform and algos. Is this on the engineering task list, or is it a non-starter? And if the latter, why is it a bad idea (it just seems so logical and practical in my mind)?
   

Hi Grant,

1. Unfortunately, the data from our provider is simply just missing a split sometimes. We are in the process of improving the robustness of our corporate actions data. Regarding https://www.splithistory.com, it's a good source for confirming splits, but in my experience, it should not be relied on as a primary source for what happened. The sensitivity is good on that site (they don't miss many events), but they frequently list events as splits that aren't really splits (for example, spinoffs). I also cannot speak to whether the data they post is good live. I also think that the text on each symbol is auto-generated so it's not that descriptive.
   
2. Tracking down missed/incorrect events simply by looking at pricing data is not trivial, but it is something that we would like to do. It's not something that we're doing yet, but it's on our radar.
   
3. This is not something that we are really considering right now. Instead, we're focusing our efforts on fixing the existing problems and doing what we can to prevent these from happening in the future. The fix to PPG also includes fixes to KGC, BIS, HBI, XHS, XHE, XSW, XTN, NTN, and VBFC which are all the tickers with incorrect/missing split info that we are aware of.
   

Thanks Jamie -

This is all still very mysterious to me. When a company decides to do a stock split, presumably it has to file paperwork, and the corporate action ends up in an authoritative database. It's not like the market opens and a company could say "Oops, we forgot to mention that we decided to do a stock split today." What am I missing here?

Regarding data errors, it seems that it would be a very simple change to your API that could be cranked into https://www.quantopian.com/help#ide-trading-guards . For example, add:

security_lists.missing_splits  

As Dan says on https://www.quantopian.com/posts/discrepancy-in-quantopian-input-data-sets , "Data errors are always present in financial data; I'm sure we haven't found them all." I'd encourage you to at least have it on the list as a project to be more systematic in publishing problems as they are found, and making the information accessible to users in their code. Seems pretty straightforward, but I can also appreciate that you have lots of competing priorities.

Grant