Skip to main content
No. of Recommendations: 7
From the overtuning department:

For this SI implementation, it looks like the sweet spot for the "top 30%" step is around 27-28% for returns (27% best top 40 without HTD, 28% with HTD 45),
With returns perhaps a whisker steadier with slightly higher number like 29.5%.
The risk difference is so small that one might as well pick the number that maximized returns, so 28% sounds sensible.
Certainly pointless tuning, but it seems a harmlessly small difference so you might as well go with what backtests best.
This backtests 19870302 - 20200529 at CAGR 14.80 (versus the "top 30%" at 14.28)
Very consistent with the VL version: without the dividend requirement a number a bit lower than 30% is optimal.

For a "with dividend" version, I added replaced step 1 with dps12m.s > 0
In that instance, the highest returns in backtests are with a cutoff around 27.5%, and steadiest returns with a cutoff around 24.5%.
These figures are a fair bit lower than what seemed optimal with VL.
You could pick any figure between the two and probably do fine, but I'd pick 27%.
This backtests 19870302 - 20200529 at only CAGR 12.58, not so interesting.

For the version with a greedier final sort meant for smaller number of holdings, e.g., 20HTD25 or 20HTD30,
First, a deep hold-till depth with very low turnover seems best for returns, with 20 HTD 32 or 33.
There's little difference in steadiness.
For something in that vicinity, 20HTD30, the ROE cutoff seems best at about 24%.
Here is a link for 30HTD33, cutoff 24%.
This backtests 19870302 - 20200529 at CAGR 15.41

You can manage a slightly higher return with 15 stocks, e.g. 15 HTD 30.
This backtests 19870302 - 20200529 at CAGR 16.47

One problem is that the turnover is quite low, so you might end up with the issue that the picks on the first day get stuck there for a very long time.
Since the (one) first day is a bit of a random thing, you're doing a lot of testing on what was picked that day.
Since the way the GTR1 backtester works, I believe it does the test for the maximum possible period, and only reports on the subset date range you ask for if that's different.
So, you can't pick a later date and get some variation by having a different starting slate of picks via a different start date.
(there might be some elaborate way to force all fresh picks on some other later day, but I don't know what it might be)
All of this is just a long way of saying that perhaps the statistical support of these HTD tests might be weaker than typical.

Print the post  


What was Your Dumbest Investment?
Share it with us -- and learn from others' stories of flubs.
When Life Gives You Lemons
We all have had hardships and made poor decisions. The important thing is how we respond and grow. Read the story of a Fool who started from nothing, and looks to gain everything.
Contact Us
Contact Customer Service and other Fool departments here.
Work for Fools?
Winner of the Washingtonian great places to work, and Glassdoor #1 Company to Work For 2015! Have access to all of TMF's online and email products for FREE, and be paid for your contributions to TMF! Click the link and start your Fool career.