Um, I’m about to modify my portfolio significantly, based on the new site (I’m super happy about the changes), so if you’re going to revert to the old site I hope you let us know soon.
Agree with everyone else that stakeholder management has been poor. This is reminiscent of the ValueLine change. Maybe someone at p123 could take a few PMP classes.
As for the beta site, it seems to me that the ‘Taxonomy Series and Industry-Based Scope Parameters’ change is unrelated to Multi-Country support. Perhaps that could be rolled-back and users could retry their sims. While my models don’t seem to be as affected, I do have great sympathy for those that invested many hours of their private time for naught.
Guys , it probably boils down to a couple of functions that need to have a version that does not include holidays to get back the old sims. I think that’s where it all lies. For those with big differences can you send us a list of systems that were affected so we can see what functions could be causing this?
We’ll try to have the old version running in parallel asap too. It’s tricky because the new framework is incompatible , but might be doable if we just focus on a few changes like the holidays.
I realize it was little time, but we needed to move this forward. The data cost has gone up. It’s been a huge expense for us and taken a lot longer.
Lastly, the new version should not destroy a robust idea. The curve fitting trap is always there if you take your eyes off of it.
I’ll check with devs today. they were up late last night.
Thanks for your feedback.
We are investigating the issue and considering a rollback on one specific change on the industry series that is affecting your strategies.
Thank you for your patience
For those seeing a large change in performance, I suggesting sticking to simulations with at least 20 positions, the fewer positions the more noise. At 10 positions the noise is completely dominating IMO. And perhaps focus more on rolling screeners with many positions (50ish at least). This will give a much stronger indication of the ability to capture real alpha.
Please note that I was attempting to understand the differences I was seeing with respect to the IndWeight rule. Since my first post was dismissed out of hand I was trying to do a thorough job of it before getting back to you. I no longer have the ability to do the thorough analysis that I would like to do so figured I would summarize my high level findings so far.
I run a few different strategies. On three of them I was recently investigating the effect of an IndWeight rule. Since I was doing this recently it was easy for me to check the difference the new changes made to backtested performance depending on whether an IndWeight rule was used or not. I have summarized the results in the attached table. Note that the differences when not using an IndWeight rule are relatively negligible, in fact for two of the strategies the annualized return improved with the changes. The only strategy that had decreased performance was the Momentum strategy, which isn’t surprising given the changes in the Date handling.
Unfortunately, when the same strategies were run when using an IndWeight rule the change in performance was much more significant and all negative. This is highly suspect to me, and while I am not able to really dig into the causes due to the changes being rolled out I would hope that P123 staff could look into potential unintended consequences of the changes that might effect this rule.
So people are putting fitted momentum into their strategies?
Maybe something else that works for people and would be affected by a holiday? I had an old system that I now consider overfitted but it actually did well for a while: recency bias with momentum actually helping of a while? At the end of the day I decided fitting momentum was mostly overfitting. But one could also argue the momentum might have worked very well for a while so I remain agnostic on using momentum. Maybe I will be convinced to use it again in the future. And certainly, if it is working out-of-sample don’t change a thing (and certainly not just because of my post).
Anyway for broad spectrum of effects from the changes: basically none on my main strategy.
Jim
Thank you, Marco.
You can be sure, for some of us at least, the intensity of unhappiness we experience during a change like this is directly related to the depth of our affection for p123 too.
This is very seat of the pants but I think the other big change - if not the biggest - is coming from the new treatment of sector, subsector, industry and subindustry functions. I could be wrong but a very quick review of changed results in my sims and screens seems to suggest that the more of these across-the-board industry functions are used in a sim or screen, the bigger the difference.
Hugh
Hugh,
I second that, Strategies that are not using industry funct. are not affected much (almost not at all!).
“We are investigating the issue and considering a rollback on one specific change on the industry series that is affecting your strategies.”
That would be super great if you could do that, thank you!!!
We’re going to release an old version that does not include holidays first. I think that’s the main source of differences bc it affects everything.
We can then go by process of elimination to see which functions might need versions that skip holidays (ex: SMAxHoliday, AvgVolxHoliday, etc)
Then we can see about tweaking industry series
We should have something soon.
Thanks
Great idea, however it might be best to create holiday functions. as in SMAwHoliday, as opposed to SMAxHoliday.
This way existing formulas won’t need to be updated with xHoliday types.
It’s very very unlikely that the changes we made could have affected the IndWeight rule. That rule simply adds up the weights of the stocks according to the industry they’re in. Nothing we’ve done on the back end would have changed that.
I must apologize for the fact that my initial post on this subject was incorrect. Cross-sectional functions using the taxonomy-based scope parameters (#Sector, #Subsector, #Industry, and #Subindustry) have not changed. Only the various industry taxonomy series and the -Ind factors (Pr4W%ChgInd, BetaInd, etc.) have changed. I’ve changed the Google document accordingly.
Also, we are considering lowering the liquidity limits of the taxonomy series so that they’re more compatible with the old series.
Can you eliminate the “USA” suffix at the end of each ticker domiciled in the US? It’s goofy and unnecessary, can’t think of another financial website that does this. Might make sense to do “CN” at the end of Canadian tickers but I think the USA suffix should be eliminated, ie no suffix = US stock.
“Also, we are considering lowering the liquidity limits of the taxonomy series so that they’re more compatible with the old series.”
—> Please roll back 100% of the changes on industries in terms of liquidity!
If you do not do this, then please tell us what is the advantage of the change?
We loose a ton of alpha if we do not include lower liquidity stocks within the industries!
If micro caps with low volumne of an industry are doing well (or bad!), this is a great tell, that the industry is doing well (bad!) too, even if bigger caps of the industry are doing less well.
My models capture that context, and they are hit substantially.
Please undo this change for good!
Thank you!
PLEASE follow the recommendation of mm123 and eliminate the suffix for US stocks, at least for inputting stock symbols (“tickers”). I regularly copy lists of symbols from spreadsheets and other sources into P123 lists. This morning 12% of the symbols I tried to add to a list were rejected for geographical ambiguity. Am I really going to have to reformat symbol lists in Excel every time I want to put them in a list??? This is a real time waster for users like me who will not be using the new data on non-US stocks.
If the old universes included U.S. traded foreign stocks (secondary listings), and the new universes exclude those stocks, then industry aggregates in the new site use a smaller sample size.
This may explain some of the differences being reported.
Yeah, this was really dumb in hindsight. We’ll have a fix . Perhaps a default country in your account settings that is used when you don’t specify one. Also anyplace where you download we will include two columns: one with the stripped ticker, one fully qualified.
Thanks
Well, the previous taxonomy series excluded ADRs, but yes, there are some additional exclusions now.
.