Why was this stock not sold 4 weeks ago? My sell rule is RankPos >41 and this is well below it.
Is this a case of p123 updating information (and ranks) retroactively or did it just screw up?
Either scenario lowers my trust in the system as a whole.
Is this a sim?
Does the sim’s end dates match up?
How frequently does it reconstitute and rebalance?
Does it use the same ranking system?
Does it operation within the same universe?
another thing to look at: rankpos > 41 (your sell rule) does not mean the same as rank<41 (which I think is the chart you are showing)
It does seem that this stock ranks from 3/9/19 - 3/29/19 were much higher than what they are now.
We examined what could have caused this on our end, and we hit a dead end.
We’re asking S&P if there’s any chance that they did a data correction without changing the effective dates for the data. It’s a no-no in point-in-time datasets, but sometimes it does occur and could be justifiable.
Thanks
Sorry I suppose I should have provided more information
-this is a live strategy
-It trades weekly
-It uses the same ranking system and universe as above
-The universe is around 800 stocks so a rank of 7 is well above Rankpos of 41
Only 2 scenarios that would explain this:
-simulation malfunctioned/screwed up
-Data was retroactively changed and the rank was still 99 until today
I wouldn’t have even noticed were it not for a rank drop to a single digit which made me look deeper into this.
My #1 concern with p123 is that the backtests are not trustworthy as they may be backtesting based on information that wasn’t available live. This adds to those concerns.
We found the problem. Has to do with the Cashflow statement DivPaid not being point in time for period 12/31/2018. When DivPaid is pulled from the annual statement it is point in time. We think Compustat overwrote the Cashflow data and caused many factors in the past to go to N/A because the data no longer became visible to the point-in-time date.
-For Dividends Paid from the Cashflow statement today we only see one value today that was “effective” as of 5/6/2019 (the final data)
-For Dividends Paid from the Annual statement today we see two values “effective” on 3/14/19 (the preliminary data) and 5/6/2019 (the final data)
The values never changed but the effective dates have huge implications for our engine since they make the data visible or not in the past. When calculating DivPaidTTM we only look at the Cashflow tables because that should be enough. Since the effective date is recent we resolve to NA for a few months, which causes the erroneous drop in Rank when calculated today in the past. We have reported this to Compustat and are waiting for confirmation that this is an isolated case. You can see their data in the image below. The top two rows are from the Annual table. The bottom one is the Cashflow.
For this particular case the drop in Rank was wrong. The Rank now is back above 90, so probably not in you top 41 stocks. I think the “real” sell signal should have been quite recently, and not a few months ago like the rank shows. When exactly I can’t say until they fix the problems with the cashflow data.
Thanks for reporting this I’ll keep you posted
NOTE: there’s a chance that they will tell us that when dealing with Q4 we should be looking at the annual instead. I hope that’s not the case because that would mean that they will not admit it’s an error, and other past historical data is affected.
Please note. I only used DivPaid as an example. Other cashflow items could show similar issues.
We’re still investigating. Might be a combination of reasons: 1) annual cashflow figures not being carried over to interim datasets by S&P 2) a rare problem with our fallback mechanism where final filings are filed & processed so late that they show up with the next period . Both these problems should not affect live systems in the present. They affect the past which does not behave the same as it did when it was the present. Also this is more likely with very small companies.
Thanks for looking into it. The rank chart looks even worse today.
Would have been nice to have sold it in Oct 2018!
This could be one of the reasons why back-tests always look so good 
This issue should be resolved now , please see https://www.portfolio123.com/mvnforum/viewthread_thread,11816
The present was not the problem. The “CompleteStmt” flag was being set to true in the past which turned off the fallback mechanism . If you re-run the rank chart you should see a higher rank in April
Sorry it took so long