Texas Aggie(*) I/Q getting in the way - why are my tranches so different?


I am a Portfolio 123 expert wannabe (i.e., a newbie), so I suspect I am doing something wrong. I hope you folks can help.

I am (was?) about ready to take my investing / trading system live, but I wanted to do one last sanity check before doing so. I think I remembered reading or seeing in a presentation where the presenter was discussing the screening (or ranking or simulation) tools available to the user. In this presentation, the presenter showed a buy rule that (I think) used something related to the stock (I (mis-?)remembered #tickerID) and the “Mod()” function to divide the stocks into tranches. I presumed that the presenter would use the different tranches to see how reliable the test results were versus pure luck in the stock selection. I reached out on the message tool and Paul DeMartino (thanks, Paul) responded that it is “StockID”. He mentioned “EvenID” but said that “Mod()” should work. So here is the “Buy rule” I used in my simulation:

Mod(StockID, 5)=0

The results I got were not quite as good as the results I had gotten without the buy rule. But when I tried "Mod(StockID, 5)=1, I got terrible results. In fact, here are the results for the different values:

(no rule) 26.93%
Mod(StockID, 5)=0 22.76%
Mod(StockID, 5)=1 16.76%
Mod(StockID, 5)=2 7.92%
Mod(StockID, 5)=3 11.13%
Mod(StockID, 5)=4 18.91%

I am selecting the top 5 stocks using a variation of the Core:Sentiment ranking system, the S&P 1500 for my universe, and a test period of 01/01/2002 to 07/02/2022.

So what the heck am I doing wrong?


(*) Texas Aggie jokes: https://duckduckgo.com/?t=ffab&q=texas+aggie+jokes&ia=web

Hello Cary, welcome to P123!

A few high level observations:

  1. It is normal for the smaller universes to show lesser performance, to a degree. The “full universe” is often optimized.

  2. Sims with only 5 stocks tend to be overoptimized. With so few stocks being turned over, sample size over the full test period is smaller, and does not always translate to out of sample performance. This is likely would you’re seeing when you split the universe into smaller groups. To test this, try your portfolio with say 15-25 stocks, then repeat with universe splitting exercise, and see how the performance varies.

That said, some have had good luck with few stocks per strategy; it also really depends on your underlying ranking system and buy/sell rules. In my experience out of sample results are very unpredictable with so few stocks.

  1. The Core: Sentiment system is a great complement to other factors (value, growth, momentum, etc), but may not hold up as a standalone system over time. I use it (and varations of it) in many of my ranking systems, but not purely on its own.

Good luck!


Thanks, Ryan, for your thoughtful reply. Because my hypothetical returns varied so much, I was quite certain that I was using the Mod() function or the stock ID incorrectly. But when I followed your suggestion to enlarge the holdings to 15-25 stocks, I tried 20 stocks and got a more consistent result:

no use of Mod(): 18.08%
Mod(StockID, 5) = 0: 15.31%
Mod(StockID, 5) = 1: 13.89%
Mod(StockID, 5) = 2: 13.08%
Mod(StockID, 5) = 3: 13.35%
Mod(StockID, 5) = 4: 13.94%

I am glad I performed this exercise.

Regarding using Sentiment alone, my results for using the different factor groupings (Growth, Value, Quality, etc.) was not performing as well individually or collectively as was Sentiment alone. (The cause may have been my small size of 5 stocks.) I did not necessarily find that troubling, as one of the best performing stock screens at AAII since its inception in 1998 is the Earnings Estimate Revision Up 5% screen, with a CAGR of 21.5%.

But I am definitely willing to learn from the smart people on this forum. I have watched every tutorial video, and have read the e-book “How I Made $1,000,000 With Portfolio123”, every posting on the blog, and pretty much anything else I can find. I even recently read a very interesting thread I stumbled upon titled “To Quant Or Not To Quant, That Is The Question”. But after reading that, I wondered if I am a paint-by-numbers guy in the midst of Rembrandt and Monet and Picasso painters – in other words, WAY out of my league. :open_mouth: But if you or anyone else can provide useful guidance, I will definitely listen.


Glad to see you got some more consistent results with the exercise.

Sentiment is a solid factor group, interesting you managed better performance than Q, V & G. That said, the off the shelf “core” Q, V & G groups are rather “garden variety”, once you add more nuance to them you can achieve better alpha.

As for guidance, you’re doing all the right things. Quant investing is definitely a journey, I would argue even more so than your non-quant investing due to the technicalities involved. Each of us invest differently, relate differently to various factors, perceive risk differently etc. So, it is difficult to pinpoint specifics, but keep turning over rocks. SSRN is a great resource for academic research as well. Also try Verdad Capital, they have some great research that can be designed into strategies on P123.