Index | Recent Threads | Who's Online | Search

New Thread
This topic has been viewed 110 times and has 5 replies
abwillingham
Rank performance screen

I have tested several ranking systems, both mine and others, and more often than not I see a graph like this, where the bottom tiers (0 - 30) out perform the upper tiers (70 - 100).

Am I misunderstanding the output? Are the ranking systems I am testing just that wonky?

Attachment B286CEE2-9B7F-40CD-BB7A-91FA0BA42AD2.jpeg (520995 bytes) (Download count: 54)


Oct 11, 2020 9:32:42 AM       
yuvaltaylor
Re: Rank performance screen

What's the time frame of the performance test? If it's five years or less, you're going to see a lot of "factor inversion," particularly with value-heavy ranking systems. Also, it's a good idea not to use "All Fundamentals" when testing. That often produces some pretty strange results. Lastly, check to make sure the ranking system has the top arrow (next to 100%) pointing up.

Yuval Taylor
Product Manager, Portfolio123
invest(igations)
Any opinions or recommendations in this message are not opinions or recommendations of Portfolio123 Securities LLC.

Oct 11, 2020 7:49:35 PM       
abwillingham
Re: Rank performance screen

Yuval, yes, it is almost always less than 5 years. Seems to me that anything that worked well more than 5 years ago is ho-hum now. Even more so for anything before 2010. Tell me more about this "factor inversion". Does this mean if I want to test a rank with young data not to use this tool?

Thanks

Oct 11, 2020 9:42:49 PM       
ETFOptimize
Re: Rank performance screen

Yuval: "Also, it's a good idea not to use "All Fundamentals" when testing. That often produces some pretty strange results."


For which Universe are you advocating, Yuval?

Chris

Oct 12, 2020 12:59:45 AM       
SpacemanJones
Re: Rank performance screen

ab, re: inversion. What you're showing is what a factor rank plot can look like if you've got the ranking inverted or upside down (don't know enough to say if it is, but seems like a possibility given that plot). I think the specific arrow Yuval is referring to is the arrow at the very top folder for the ranking system. You can completely invert the ranking system by toggling that top-most arrow up and down. What Yuval is referring to is in the ranking systems there is an arrow next to the factor that shows whether an variable is ranking ascending or descending.

(If higher is "better" for a factor then you want the arrow pointing up. If higher is "worse", then you want the error pointing down.)

Example: If you feel like lower PE are better than higher PE (not saying it is, just as an example), then lower would be better in that context and you'd want to put a down arrow next to PE so the lower PE will rate higher than higher PE.

As mentioned, an entire system can be inverted by toggling the very top arrow.

hope this helps. I also agree with setting up a custom universe to encapsulate the universe you wish to test/and or trade, constrained by exchanges and liquidity requirements. Depending on what you're doing, All Fundamentals might have some stocks that are too small or too illiquid to be part of your potential system. For example, here's lines from one of my base testing universes (not saying this is one to use, just giving an example because I'm concrete and prefer concrete examples, and these are main filters I use):

Universe(NOOTC) and Universe($ADR)=False and Universe(MasterLP)=False
!GICS(FINANCIAL) and !GICS(REALESTATE) //this has changed to RBICS, so maybe update this. I think it still maps though.
MedianDailyTot(63)> (200000) and MktCap > 75 and Price > 3 //this is close to what I use for liquidity but also have a scaling factor for mkt value.
IssueChange(8) = 0 and DaysLate <=5 and DaysFromMergerAnn = NA
Country("CHN,RUS") = False

edit: the new 2nd line might be something like
!RBICS(3010,3015,3020,302520,303010)

other folks probably have other constraints, but that's one I use to test with. I have some other universes where I add a line for maximum liquidity to screen out the biggest companies when looking at small caps only.

Oct 12, 2020 4:38:18 AM       
yuvaltaylor
Re: Rank performance screen

Yuval: "Also, it's a good idea not to use "All Fundamentals" when testing. That often produces some pretty strange results."


For which Universe are you advocating, Yuval?

Chris

See Michael's post above this one for a decent suggestion.

Yuval Taylor
Product Manager, Portfolio123
invest(igations)
Any opinions or recommendations in this message are not opinions or recommendations of Portfolio123 Securities LLC.

Oct 12, 2020 9:12:42 AM