Screening for the top few percent

Jim has generously shared numerous screens that produce high returns. Successful screening criteria include things such as ROE, revenue growth and P/E. The common denominator in all of these backtests is screening for the top few percent of stocks in the database using the given criterion. Table 7.2 in “The Little Book That Beats the Market” shows the returns by decile of 2500 stocks as ranked by the magic formula, a screen that used two criteria, one for profitability (return on invested capital) and one for cheapness (earnings yield). The returns were:

top decile, 17.9%
2nd decile, 15.6%
3rd, 14.9%
4th, 14.2%
5th, 14.1%
6th, 12.7%
7th, 11.3%
8th, 10.1%
9th, 5.2%
bottom decile, 2.5%
median, 13.1%
S&P 500, 12.4%

Simply by screening for the top 10% by the selected criteria, the return increased 5 percentage points relative to the median. If one plots these returns by decile and extrapolates to the top 5%, the implied return is 6% above the median.

I’m really just stating the obvious, but the basic trick is to select one or more criteria upon which return logically depends, and then to buy the stocks in the top 5% or 10%. In mechanical investing one generally buys all the stocks that pass the screen, but it’s not absolutely necessary to do so. One can start with a universe of, say, the largest 2000 stocks by market cap, and then screen by one or more criteria to get the top 100 stocks (5% of the universe), and then choose the 25 that one likes the most after applying other criteria such as quality of product or service, market share, debt/equity or what have you. The important thing is to stay within the top 5% to 10% by one’s primary criteria.

Three criteria that work well are (1) some measure of profitability (such as 5-year average ROE or ROA), (2) some measure of growth (such as 5-year average revenue growth) and (3) some measure of cheapness (such as normalized earnings yield), but as Jim has shown us, using a single criterion works very well.

Like I say, I’m just restating what others have already shown, but such screens are useful. They generally turn up some good companies, such as Alphabet.

5 Likes

“In mechanical investing one generally buys all the stocks that pass the screen, but it’s not absolutely necessary to do so.”

Granted, buying the top 20 stocks using only the primary criteria may give a higher return than picking 20 stocks from the top 100 after applying additional criteria. Some statistical studies have shown this to be the case. However I like the flexibility of being able to eliminate a company with, say, high debt or declining revenues, even if my primary screening criterion is simply ROE.

Table 7.2 in “The Little Book That Beats the Market” shows the returns by decile of 2500 stocks
as ranked by the magic formula, a screen that used two criteria, one for profitability (return on
invested capital) and one for cheapness (earnings yield). The returns were:…

One might add “reported returns”.

It should probably be noted that nobody, and I mean NOBODY, has found the same results that he author presented, nor anything close.
Same set of stocks, same criteria, same date range.
I have done my own tests, and lots of other people have too, some very capable. Not there, not close.
Better than an index, but nothing worth writing a book about.

Some theories are that he simply made it all up, which is a bit unkind.
The other explanation is that he is simply very bad at backtesting. Insulting, but it does not impugn his ethics.
A third explanation is that the system he tested is not the one he described. If that is the real explanation, it must have differed in extremely material ways.

One can absolutely forgive a model that worked in a given date range for not working in a later date range.
Sometimes the world changes, and sometimes a predictive model turns out to have been a fluke.
But if it doesn’t work in the original date range when tested by someone else, it’s just plain wrong.
That’s his book.

Jim

36 Likes

“It should probably be noted that nobody, and I mean NOBODY, has found the same results that he author presented, nor anything close.”

The reported returns for the top 1%, as shown in Table 6.1, don’t seem to match the returns by decile as shown in Table.2. The returns by decile show a smooth curve that extrapolates to about a 20% return for the top 1%, not the 30.8% return for the top 1% in Table 6.1.

1 Like

The reported returns for the top 1%, as shown in Table 6.1, don’t seem to match the returns by decile
as shown in Table.2. The returns by decile show a smooth curve that extrapolates to about a 20%
return for the top 1%, not the 30.8% return for the top 1% in Table 6.1.

I wouldn’t see that as a problem. A curve might not extrapolate and still be valid.

The bigger problem is that (probably) the 20% figure and (especially) the 30% figure are simply wrong.
As in, not true.

For sure, there are some backtests that show that kind of returns.
But…
(1) The future will be different from the backtest in that situation.
There is no simple quant method that returns over 20% a year over long time frames.
Never was, never will be.
…and…
(2) The numbers the author quotes are apparently simply wrong even for that time frame.
Nobody can reproduce his results for that same era. No 30% returns to be found, even then.
Maybe it was all a practical joke. I suspect a badly constructed test.

Jim

7 Likes

Greenblatt’s funds performance, e.g. value fund, aren’t eye-popping either
https://www.gothamfunds.com/default.aspx#

Philosophical question: what defines a ‘quant method’?
Buffet’s early returns were certainly great.
But his later returns, as the fund grew larger were less great, generating much debate on the BRK board of whether it beat the S&P (a start/end point dependent question).
Is WEB using a ‘quant method’?
He’s certainly thinking, i.e. presumably he’s not throwing darts at a dartboard.
But we apparently don’t know, and perhaps he doesn’t know, how to reduce his thoughts to a relatively simple algorithm (I qualified that sentence because presumably one could cook up something complicated that matched his returns year for year, but odds are it wouldn’t extrapolate)

1 Like

Philosophical question: what defines a ‘quant method’?

Doesn’t need much philosophical debate.
A quant investment method is one that picks securities deterministically using only available data, with no human gut feel assessment or filtering.

Ironically it doesn’t remove the human judgment, it generally just displaces it.
All the work that would have gone into assessing the quality of a security for investment is moved to assessing the quality of the model to be used.
The work you might put into assessing a stock for 10% of your portfolio might go into assessing a model to put 1/2% of your portfolio into each of 20 stocks it picks.
In some ways assessing a model is much harder. The past lies.

Jim

9 Likes

A quant investment method is one that picks securities deterministically using only available data, with no human gut feel assessment or filtering.

So far has quant investment as a group outperformed the market index?

So far has quant investment as a group outperformed the market index?

It’ s like asking whether individual stock selection has beat the index.
Whose?
There are thousands of different quant metrics and strategies, and practitioners.
As with individual stock selection, many don’t work, some do.

We can say that some quant strategies have worked very well for a very long time. And you don’t have to be Mr Simons.
A random example:
One fellow on the MI board has been following pretty much exactly the same quant strategy for over 22 years.
He runs that strategy in a separate account, making the results easy to track.
The account has returned 16.7%/year compounded in that time, as of late last year. The S&P 500 returned about 7.5%/year in the same stretch.
The returns on the individual positions for this particular strategy are wildly variable, so he targets 2/3 cash.
The 16.7% return is on the account as a whole, counting the cash drag. The balance in the account is up by about a factor of 30.
He trades once a month, generally replacing four of the 12 existing positions with four new ones, each position held three months.
Other than (I believe) one stretch during the credit crunch that he eased towards cash by not
opening new positions for a while, I think he has done pretty much the exact same thing the whole time.

One advantage of quant techniques is that two people doing the same approach in the same time frame will get the same returns.
It’s not like you’ll match Mr Buffett’s stock picking results by using the same strategies he does.
If you wanted to do what this quant fellow has been doing, you could.

Jim

17 Likes

Greenblatt’s funds performance, e.g. value fund, aren’t eye-popping either

True, though it’s been a tough few years to be a value investor. Buffett hasn’t shot the lights out either.

Some of Greenblatt’s funds have done OK, the Index plus (GINDX) and Enhanced S&P 500 Index Fund (GSPFX) have done slightly better than Berkshire over 5 years ending June 30. The Gotham funds not based on the S&P500 index haven’t done well.

I got the Longleaf Partners report last night. Used to have money with them, dumped them years ago.
Longleaf Partners Fund LLPFX 5 year CAGR 2.9% Ouch! Berkshire 10%, GINDX 10.7%, GSPFX 12%, SPY 11.2%

I like Greenblatt. He talks sense. Disappointing about the Magic Formula. Has anyone ever got a straight answer from him about not being able to replicate his results?

Not knocking Greenblatt specifically, he makes sense, I like him.

It occurred to me that many of the people/funds that I like and think make sense actually have mediocre performance.

Greenblatt https://www.gothamfunds.com/default.aspx
Performance - Meh

Pabrai https://pabraifunds.com/
Performance - Meh

AlphaArhcitects https://www.gothamfunds.com/default.aspx
Performance - Meh

Mebane Faber https://cambriafunds.com/
Performance - Meh

For individual screens, I liked the BlueCheaps screen, it made sense.
Performance (last time I was able to easily check) - Meh

Not sure what to make of that. Perhaps start/end point effects?

5 Likes

It’ s like asking whether individual stock selection has beat the index.
Whose?
There are thousands of different quant metrics and strategies, and practitioners.
As with individual stock selection, many don’t work, some do.

Right. I should rephrase it as “Has the well-known quant funds as group outperformed the market index?”

Oops, alphaarchitects link is https://etfsite.alphaarchitect.com/

I’m a long time MI board and general amateur follower of investing. My take is there’s quite a bit of difference between quant sold to the masses and quant you can do by yourself. What an individual can do easily and quite profitably is also likely to not be scalable for commercial purposes. However, the individual most likely has to be able to deal with a higher level of volatility, sometimes significantly so. When trying to implement something commercially there tends to be way more stocks held than what an individual can chose to hold which of course mathematically is going to drive the big guys to be more market like in their returns.

In practical terms as has been mentioned, you just do not have an absolute assurance that something that worked or backtested well in the past is going to work in the future. Accordingly, one needs to have a system in place that can monitor the system(s) going forward to see if its edge is being maintained and sufficiently robust that one can distinguish performance in different market types. Bad results from a good system in a bad market != failure of the system. Finally, one can find a really excellent system that is insanely volatile which can be tamed to taste with a hedging strategy.

It’s a lot of work though with no guarantees…but if you can find something that does work it can definitely be worth the effort.

1 Like

What an individual can do easily and quite profitably is also likely to not be scalable for commercial purposes.

This is true of the much discussed equal-weighting.

Equal weighting the S&P500 has a practical limit of about $67bn (if I did my math right, assuming no more than 10% owned of one company).

1 Like

In practical terms as has been mentioned, you just do not have an absolute assurance that something that worked or backtested well in the past is going to work in the future.

That’s very hard to endure during bear market without knowing the true worth of the holding.

Jim Simons in the interview video said that they mainly spot and act on abnormalities. That makes more sense to me, because abnormalities do occur from time to time.

For individual screens, I liked the BlueCheaps screen, it made sense.
Performance (last time I was able to easily check) - Meh
Not sure what to make of that. Perhaps start/end point effects?

My guess: Badly designed screen. (by me). A poor predictive model as it lacked out-of-sample validation.
The ever-present danger of trusting a backtest too much…it is FAR too easy to pick something that happens to match the quirks of history.
For real money, it is better to use something that has worked for at least a few years AFTER somebody invented it.

But even though that helps, it isn’t bullet proof.
It’s true that certain styles of stocks go in and out of fashion.
Since most quant screens have a focus on a certain style, they too can go in and out of fashion.
(Outright failure is probably more common!)
On of my long term favourite screens, based almost entirely on earnings yield and dividend yield, beat the market by 10.6%/year in the first 10.6 years after it was published.
That makes for a screen that seems to deserve a whole lot of confidence, especially since it made intuitive sense.
What would one guess would have happened in the next 6.25 years? Underperformance by 20%/year,
mainly because dividend stocks went WAY out of fashion and screening in that pool magnified the effect.
In the last 2.25 years it has returned 43.6%/yr (2.25x the money), outperforming the S&P by 24.2%/year, having nothing but 10 high yield low P/E stocks.
So… which stretch was the fluke? What will the future hold?
I would expect, at a guess, it will do OK during periods that dividends and earnings aren’t considered trash concepts.

Jim

10 Likes

For individual screens, I liked the BlueCheaps screen, it made sense.
Performance (last time I was able to easily check) - Meh

BlueCheaps is such a wonderful idea and I would love to use it, but the performance doesn’t make it worth it.

That’s very hard to endure during bear market without knowing the true worth of the holding.

There are many methods to quant by. A bear market is a completely normal market state which generally means no matter what you are holding you are losing money assuming a long only strategy. The issue for your system isn’t losing money; it’s whether or not it’s departing “normality” from how it lost money in the past.

For individual screens, I liked the BlueCheaps screen, it made sense.
Performance (last time I was able to easily check) - Meh

BlueCheaps is such a wonderful idea and I would love to use it, but the performance doesn’t make it worth it.

Yes, it is a disappointment. Certainly a high earnings yield has not been the key to riches in recent years.
Maybe that’s the reason for the dreary returns, or more likely the good returns prior to that were more like a fluke.

I think the later attempt is much better, the “LargeCapCash” screen, introduced here https://discussion.fool.com/a-spy-alternative-screen-34516863.as…
It’s intended as an alternative to SPY.
“The goal is a screen which is as safe as the S&P 500 but with the hope of somewhat higher returns over the long run.”

This has indeed beat the S&P by a small amount since its introduction 2.2 years ago, with a high correlation.
So, in that sense it’s meeting its goals: so far so good, anyway.

General description: An equally weighted slate of 40 large cap dividend payers with high ROE and lots of cash on hand.

There are two variations mentioned in the kickoff post, with and without dividends.
Perhaps to my surprise, the one which requires dividends from all picks has always tested just a little better, and this remains the case.
There are more elaborations later in the thread, but the version I track is one of the simplest:

Start from the Value Line 1700 set of stocks.
Eliminate stocks with no “Timeliness” rating (just a sanity check to skip those just listed or in the middle of M&A).
Eliminate any stocks that don’t pay a dividend
Of those remaining with a reported ROE, take the top 30% or 32% sorted on ROE.
For each stock remaining, calculate its cash balance in excess of long term debt and sort on that.
(Note: this uses largest cash balances in absolute terms, not largest cash balances as fraction of market cap…that’s why it’s a large cap screen)
Buy equal dollar amounts of the top 40 sorted on [cash-debt], hold two months, repeat.
Since the criteria don’t change often, there isn’t much trading.
Rebalance all positions to equal weight annually.

It has had a very unpleasant stretch since around the start of November, but still market beating overall so far by 2.91%/year after trading costs.
Of course 2.2 years isn’t enough to draw any firm conclusions, but at least it has not taken the opportunity to blow up.
The backtest January 1989 to April 2000 showed outperformance of 5.15%/year, which is as always probably too optimistic.

Top picks this year have generally included Microsoft, TSMC, Cisco, Nvidia, Accenture, Costco.
Microsoft has been the top pick every month since 2009.
But, as this is equally weighted, it doesn’t really matter: unlike an index fund, there is no concentration in the top pick.

Jim

8 Likes