I just ran a Screen backtest and in the results I noticed a huge return for a particular week.
See the attached images.
Looking into the results for the particular week, we can see that DAN was bought at 0.01 and sold at 12.70. In reality, someone who had been holding DAN shares at 0.01 would probably not be able to sell them at 12.70 after the new company with the same symbol floats.
I’ve checked my own chart software data and can confirm that the data for DAN begins at around February 2008. This means that the data on Portfolio123 for DAN is probably including data of a delisted stock that had the same stock symbol.
The fact that new stocks can take on the same symbol as a delisted stock can certainly cause much frustration for both the developers and the users of stock market analysis software or web sites. We normally think of a stock symbol as a unique identifier and usually treat it that way, but when we also include delisted data the symbol can no longer be considered as a unique identifier.
How does Portfolio123 normally handle delisted stocks?
Is there some kind of set procedure that is carried out when a stock gets delisted?
Is there anything to prevent a situation like the one I’ve brought forward here?
How will this be fixed? If you get rid of the delisted stock data then that would introduce survivorship bias, wouldn’t it?
If we leave it as it is then our results for screen backtests, rank performances, and simulations will be distorted by this and other similar cases.
Note that the ranking system that I was using contains the Pr2SalesPTM factor, in which DAN ranks around the top 20 after mid 2007.
I know that I could filter it out using a price limit, but I wanted to do analysis with as little restriction as possible. The only rule I set was a liquidity rule, and even with AvgDailyTot(20) > 100000, DAN passes. I’m not sure why there is such liquidity for the stock during that time. Is it bad data?
Even if I did use a price filter, it does not address the issue. If a stock gets delisted at above 3$ then it can still pass the 3$ price filter.
Delistings, splits, name changes and symbol changes can be messy work when managing historical data for many thousands of securities, but it is crucial to manage them properly and timely in order to uphold the quality/integrity of the data. The procedures that Portfolio123 undertake for these actions should be documented somewhere.
We need to correct the data. There was a new stock issued but for some reason it did not generate a new separate price history. We’ll examine the data to see the cause. Thank you for reporting this. We’ll manually correct this problem.
We’re still waiting for a resolution from reuters. It appears the issue that went inactive was overwritten with the new issue, thereby combining both price histories. Looks like human error.
How would this be fixed? Would you simply delete the old data? If so, wouldn’t that introduce survivorship bias?
There are some other stocks whose data look similar. I came across them when running a screen backtest based on a price-to-sales
rank without price limits. Will they be fixed as well?
My concern is not so much for the data of this or other particular stocks, but is more general.
I raised a few issues in my previous messages, e.g. regarding your processes to ensure quality & integrity of the data in cases of delistings, name & symbol changes, and splits.
In particular, what do you do when a new company takes on the same symbol as a previously delisted company?
It seems that there is no way for users of the web site to distinguish between data of a delisted company and data of a new company with the same symbol.
The data is the foundation upon which everything else in your business is built. We must mend all cracks and put processes/policies in place to prevent future ones.
We rely on Reuters to tell us all the active/inactive issues for a company. In this case there was an error from them an only one issue was reported which caused our logic to simply continue the old one.
The old one is still in our database: http://www.portfolio123.com/stock.jsp?ticker_uid=27607 and it would still be used by historical sims. You are correct that it is hard to tell them apart because we show the last symbol for the company in the transaction list. We have a project under way to show the last known ticker (Reuters RIC code) for all inactive issues instead of the current ticker. Once implemented a sim that buys the previous inactive issue for DAN will show DCNAQ.PK in the transactions, thus making it more obvious. This will also resolve situations where a company assumes a previously delisted ticker since REuters appends characters to an issue when it is delisted, for ex JEM^D95. The code at the end tells you what month and year is was delisted.
