Shoutout to our Model Designers and mechanical, “boring” market beating investing

With the recent melt-down of growth (sexy) stocks, a discretionary stock picking approach has never been more stressful. I for one was managing about ⅔ of my portfolios using a combination of screening, back of the envelope analysis, and whether “I liked” the products being produced. It has been a disaster compared to my other ⅓ mechanical approach where I know very little about what I’m buying. The difference is probably 2-3 Million had I just done mechanical investing. See images below.

So, are boring, mechanical approaches to investing finally showing their merits? It’s never too late to re-evaluate your approach. Here are the top 15 mechanical designer models ranked by a combination of metrics I like.

[size=3]The link to this post in a google doc is here

Share this tweet if you agree



I don’t have a clear opinion about your specific question. But I know your interests are broad include developing some machine learning at P123.

Broadening your discussion a bit I have been looking at reinforcement learning.

Sometimes there is no real difference between machine learning and reinforcement learning. Reinforcement learning is sometimes just the cool new thing. But also reinforcement learning can have “policy” methods and value methods. The polices can be truly different than value methods at times (and therefore different from machine learning’s value methods).

Policy methods CAN get around the problem of changing market regimes or the very serious “non-stationary” problems of machine learning data (using machine learning jargon). That and the related mathematical topic of “prediction with expert advice” can deal with the problem of changing market regimes.

Anyway, not to bore anyone with my particular policy methods, I am having success in this market using “policies” intended to work in many market regimes.

That having been said, I would be the first to say how a policy method differs from a value method for stock investing is not well defined (anywhere as far as I can tell).

Perhaps not related to what you are doing, but my present take on machine learning and beyond (including reinforcement learning and prediction with expert advice). And specifically methods of addressing changing market regimes (or “non-stationary” data).

BTW, Fidelity’s StarMine method could be called “Prediction with Expert Advice” and attempts to mitigate the problem of changing market regimes. First, by using a type of prediction with expert advice but also by looking back only 2 years. So I am not alone in exploring/using this.

It is an open question as to whether Fidelity’s method should be called a “policy,” a value method or something else, however.



Marco - the boring mechanical approach is great so long as you are not chasing past results that are cherry-picked, and are prepared to underperform for years without switching strategies. That is difficult to do.

The biggest problem with discretionary trading (and I’m going to throw in the discretionary choice of trading systems) is fighting your own demons. One demon is chasing what worked in the past instead of what will likely work in the future. Another is knowing when to quit. In 2000 and 2008, I didn’t stop when I should have and lost significantly more money than I made in previous runups. It is like Las Vegas. Casinos invite winners back because they know they will eventually lose all their money :slight_smile:

This time around, I learned my lesson. I stopped more than a year ago because I realized that growth was not sustainable in the current economic conditions. TipRanks has its issues (manipulation, etc) but shows what can be accomplished with market timing and a little bit of luck. [url=][/url]

Sure, curve-fitting , cherry picking, survivorship bias, are all problems with mechanical strategies. Only long out-of-sample performance proves a mechanical system’s worth. I’m starting to feel like Designer Models are finally showing their worth now that many have decent out of sample periods. We put a lot of effort in the DM platform, some was misguided, and frankly it seemed doomed. But it just takes time; and DMs being powered by a machine just kept going.

As far as helping people succeed with sticking to a strategy, the only way I see this possible is for us to fully automate the rebalancing. It’s something we really want to do , but we needed a strong DM marketplace for it to succeed. So things are falling into place, we’ll see. (it won’t be fully automated because of regs, but maybe just pushing a single button in a phone app to rebalance will suffice)

BTW congrats on your great cherry picking stats. Only problem with following a real person is 1) finding one 2) people change! :wink:

I don’t think that following someone else works. I hope I wasn’t implying that. I’m also not sure that following mechanical systems that someone else designed works in general either. Chances are, all or most of the systems you listed will eventually go through a downturn, and quite possibly at the same time if their past success is a result of similar system properties. If they are well diversified from each other then you have a shot at slowing growing your capital.

Kumar isn’t on there. Just saying.

I assume this means there may be some survivorship bias.

Just to be clear, P123 technically maintains control/owns the models even if they are removed by the designer? You don’t happen to have the data on all of the designer models without survivorship bias do you?

If so can we see that?

BTW, I fund a port. We will see how it does out-of-sample. But I think there may be some value at P123.

My point is that, taking my surviving ports that are on auto, sorting them to show the best ones that I am still running would not be very good proof that I can make a good model.

It would not be proof that I cannot make a good model either. Actually, it would not show much of anything.

Taking the results of all of one one’s models that are funded (as Marco does for his ports here) is good evidence, however. Not proof perhaps but evidence that he might be able to make good models. And I would certainly hope that would be the case.

Thank you for sharing that Marco.

Can we get some more solid evidence? See my comments above about the possibility of getting data with no survivorship bias (somehow).

Maybe start a good sample of models from Yuval Dan and Marco that are never removed if P123 will not show us designer models with no survivorship bias?

BTW, Yuval’s designer models have little survivorship bias and he has removed zero models recently. But also he is just one model designer and there is a bit if a multiple comparison problem if you just look at his models. Still, there is some evidence for the value of P123 in Yuval’s results.

P123 does have ways to prove there is value in its ports (assuming there is value there). And to even quantitate that value in a rigorous way.

BTW, to be balanced (with Yuval’s generally good results) and to see the multiple comparison problem as well as the problem of looking at just the last 2 years, has anyone looked at Marc’s results recently? Alert: Serious survivorship bias there.


One of the issues with many designer models is that many are extremely high turnover. Not really feasible in a taxable account IMO.

That’s what I used to think too. I hated paying taxes , and disregarded signals to sell winners.

If you are paying taxes on a high turnover model then you (very likely) made money overall for the year. How is that a bad thing?

The other advantage of high turnover is a constantly fresh set of holdings. It’s very hard to anticipate sector rotations until it’s too late. Nobody wants to leave a great party early, then the cops show up. Not talking about real investors who constantly watch/read things and apply their experience. Just regular investors with short term horizon who think they know better bc of past success (like me).

This was a brutal crash. Likely near the bottom now. It’s the third time for me: 2000, 2008. But I always forget the lessons. Investing is hard. 97% of the people should not be trading. Regular folks are throwing in the towel now and hedge funds are starting to buy. It’s terrible. P123 needs to come up with automated products that send orders to accounts with boring, machine driven strategies.

Thanks for responding, Marco. Assume my hurdle is VOO on a long term hold. Short term gains means that you are only getting about 60-70% of the net gain. If the SP returns 10% annualized, then I have a 5.38X on my money on the VOO on a 20 year hold. i may have more if I do not liquidate. To get equivalent returns with a high turn portfolio I should get 15-18% annualized over 20 years. I would love to find such a portfolio, but am not sure that it exists and has the same volatility or less.

PS - you forgot 2020, but that was a quick recovery :slight_smile:


P123 could try to make the case that using P123 works for the average Joe by taking the equally-weighted returns of the active designer models. That way survivorship bias would not be a problem.

I am not sure I would recommend doing that.

P123 could also EASILY make the case that P123 can work for someone who has a good amount of experience by creating some models by Marco, Yuval and Dan and following the aggregate returns. Marco, Yuval and Dan could even define how much weight to put on each of their models at the beginning of each month and be free to add or remove models at any time.

We could at least hope that 3 professionals with a great deal of experience with P123’s platform would thrive doing this.

Let me just assume that in a relatively short period of time this portfolio would be killing the market. Then it is true that P123 could easily document its usefulness in a short amount of time. P123 could document its usefulness without having to sort thorough a bunch of cherry-picked data that people decide to present in the forum or sort through the P123 designer models with the survivorship bias and make post hoc decisions with flawed data that has a huge selection bias.

I think I would do that if I were at P123 and I had confidence in the usefulness of my product and if I thought that all that talk about an efficient market comes from a bunch of egghead academics.

But it would be easy to do. And we would know for sure (whatever the result might be).


I don’t agree with that. I think it’s more like 11%-13.5% to be worth it. Here’s why:

SP pays around 1.8% on average in yield, which is taxable. So that reduces the returns to around 9.5% This is equivalent to 13.5% in a high turnover model with a 30% tax rate. BUT… with a high turnover model, tax loss harvesting is always available to you, for the entire 20 years. With buy & hold tax loss harvesting is not possible.

This is huge. Lets say you have a rule to sell everything and switch to SP when your yearly loss is -15%. Then after 31 days you go back to your model to avoid wash sale rule. This loss would cover your gains for about 1.5 years. There have been 9 years in the past 30 years that would trigger this. This covers your taxes for 13 years (9*1.5), or about 1/3 of the time. And if you get a little lucky, maybe you lock in much bigger losses like -30, or -40 in the 5 worse years bc your SP position might also have lost more.

In other words, 13.5% high turnover model is far superior than SP in taxable accounts provided you do tax-loss harvesting. Maybe even a 11-12% high turnover model is better than the SP.

Sorry it’s all back of the envelope math. But seems right. Here’s the intra-year data I used

1 Like

Thanks, let me work thru this!

From this site (and others I think): [url=][/url]

“Investors are allowed to claim only a limited amount of losses on their taxes in a given year. You’re allowed up to $3,000 per year to offset taxable income ($1,500 if you’re married, filing separately).”

So that is a maximum $60,000 over a 20 year period, I think. Not a trivial number. and this should not be ignored. This should probably be used by every P123 member earning income in the United States. The relative impact of this, and the percent of assets rationally dedicated to this strategy could vary widely for each member.

“The tax code allows joint filers to apply up to $3,000 a year in capital losses to reduce ordinary income”

Ordinary income is the key. You should be able to offset a big capital loss the following year with a big capital gain.

A 50K capital loss in one year only deducts your ordinary income tax liability by $3K, but you can carry forward 47K in capital loss the next year to offset capital gain entirely.

So it works perfectly, for any account size. Any tax experts here?

1 Like


I think I am beginning to understand what you are doing with taxes. Thank you for taking the time to explain it.


Great discussion and I started implementing “tax alpha” recently (locking in many small losses and holding out for long term gains).

Great discussion.

I have clients who pay 50% in taxes on short term gains. Trading those accounts was very difficult to justify. My thinking was that I would need to make double the benchmark in order to make trading worthwhile. I have been beating the market significantly, but not by double.

I am trying to wrap my brains around this.

If we can model tax fees into the sim I might be much more comfortable with the idea. Is that a feature that can be implemented?

This is very good, thanks Marco.

Along these lines… I have been toying around with the idea recently of incorporating and trading inside of a C-Corp. I haven’t run the traps on this or spoken to a professional, but in theory with the tax changes under the last administration wouldn’t capital gains be taxed at an ~20% rate whether short-term or long-term?

Anyone else given this any thought?