Does the dream of excess returns still live?

As of today, there are 228 Designer Models that have been live for 2 years or longer.

Of those 228 models, only 43 of them have provided excess return above the S&P 500 Index. That’s about 19% of the models. Because models do not always produce stellar returns, it’s natural for any person to invest in multiple models simultaneously. When investing in 2-4 models (or more) at once, if most models aren’t performing well, you are obviously going to lag the market averages.

This does not include the many models that have already been consigned to the dustbin of history.

My question: at what point must we throw in the towel and use broadly-diversified index funds?

I understand the usual rebuttals that will come to this post, including:

– there will be folks that tell me I need to design my own models; that is the only way I can beat the market. However, I would love to see these folks post their legitimate, total-portfolio rates of returns over the past 3, 5, and 10 years (or whatever long time periods they have). Have these people actually beaten the S&P 500 Index over the past 10 years (in their total portfolio, not just in part of their portfolio)?
– the few folks that mention how their models are up 50% or more annualized the past 2 years.
– the folks that talk about how value has under-performed growth for most of this site’s history (and this is a rebuttal that has merit, I think).

I have long wanted to believe that I could beat the market through disciplined, quantitative individual stock investing. I am coming close to pulling the plug for good on that dream.

Will there come a day (some years from now) when we’ll look at Designer Model performance and see that 80% of the models have outperformed the broad market over the prior 2 years? I suppose it’s possible, but perhaps unlikely.

I’m sorry to hear you’re on the verge of pulling the plug. Just two thoughts, if I may.

  1. The S&P 500 is a very well-designed actively managed stock system. It IS a designer model. It has a lot of rules and managers with years of experience. It’s hard to beat. It’s quite a bit easier to beat the Russell 3000.

  2. I started using Portfolio123’s ranking systems in November 2015. Since then I have beaten the indexes every year by large margins. I know that there are other users here who have done the same. My five designer models have all beaten their benchmarks, and I’m sure I’m not the only designer with a decent track record. A lot of designer models were ill-specified or curve-fit, but some have indeed done well.

  3. Besides the designer models, also consider the free P123 models we’re offering. You can find them here: https://www.portfolio123.com/app/investment/add-new?browse=1

Has anyone noticed that 75% of 2019 contest portfolios are outperforming the market (see attachment). Makes me wonder if the slippage eating away the returns as crowding in same names by too many P123 folks churning in and out of positions on Mondays is the culprit of systematic under performance (that is worse than random). Thoughts?


Those are annual returns. Only 8 models are ahead of the benchmark. I think slippage is a problem on smallcaps for most people. My small cap models have never performed as good as the sims. I have done well with large cap models with limited turnover.

I agree with both of you about slippage. Look for models with turnover of 8X or (a lot) lower. Small-caps require more turnover than large caps, and the slippage is much higher for those already. Microcaps, with huge slippage, are the riskiest stocks, but have the potential for bigger gains, and to make a microcap model work you probably need a lot of turnover. But for large cap models, turnover can be very low.

Here’s the data I grabbed for Yuval’s designer models:

	        excess 1 year	excess 2 year

small caps 19.09% 13.94%
microcap model 13.39%
mid caps -9.33% 14.37%
25 stocks… -10.78%
sp 1500 -13.72% 0.82%

The models seem to go through short periods of strong under- or over-performance. We’ll see what happens long term.

Thanks KJ. Rookie mistake :frowning:

Great to see. However, to outperform the market in a 6-12 month time frame is not that difficult. The problem is that most algos fall apart over time (>12 month). You can see numerous examples of this in the designer model list.

I still didn’t find the holy grail behind models which can outperform over long time, but there are definitely some out there.

De Prado gives AN explanation for this in his book “Advances in Financial Machine Learning”

I do not buy it as the whole reason but he is just plain correct that he has part of the reason for this. He places an “embargo” on out-of-sample testing. It would be like if P123 waited a year after submission of a designer model before starting to look at it out-of-sample.

So I think de Prado validates your observation. To be honest it was not one of the more simple things to understand but once understood it seems basic. I will not attempt an explanation here. Embargo could be searched in the electronic version and could be in the index.

One does not have to wait a year to start testing a model when one uses his “walk-forward validation” method so the embargo does not cause as many problems as you would think should you want to consider using some of his ideas.

Anyone interested can read the book and email your thoughts to de Prado.

-Jim

It hasn’t been a great time to be a stockpicker over the last 2-3 years.

All the performance has been in the S&P 100.

Most of the P123 models look at much wider universes.

Over the last 3 years, the S&P equal weight trails the S&P cap weight by nearly 10%. Midcaps and Russell trail by more.

Midcaps and Russell trail SPY by 11% over the last 2 years.

P123 models, by and large, equal weight their positions. Over time, the S&P equal weight will outperorm the cap weight. Over time, the midcaps and small caps will outperform the S&P 500 cap weight. Over time, value will outperform momentum.

But there can be long stretches, like 1995-2000 and recently, where that’s not true.

Perhaps if P123 had data back to 1995, models would have been designed differently or better. Perhaps not. Either way, when the S&P 100 leads for an extended time, P123 models using wider universes will underperform.

From O’Shaughnessy Asset Management 2019 Q1 Letter…

The consolation is that if my models are overfit, then so must everyone else’s.