Skip to main content
No. of Recommendations: 4
The idea of mechanically picking a screen, and beating the market was very enticing to me. I spent a lot of time experimenting on it, and had a lot of fun. At this point in time, it appears it no longer works for most people. Attendance on the board is declining. It seems it is mainly people like myself, who have been coming for years, (some have been coming much longer than me), who still come here.
When a companies business model no longer works. What do they do. They change.
They are a number of things that still seem to work. Close end bond funds beat other forms of fixed income, preferred stocks beat bonds, I know people who do very well selling cash secured puts, and covered calls.

Then there are the more out there ideas. Crypto currency, NFTs, buying farmland, etc.
I have loved going to this group, and I fear that unless it changes with the times, that attendance will dwindle more and more.

I don’t know if my ideas have any merit, except one. The focus of the group needs to change to things that generate alpha.
Print the post Back To Top
No. of Recommendations: 32
Random notes---

I came to a conclusion that a lot of the standard screens no longer work, or no longer work very well.
I guess a lot of people have come to the same conclusion.

But I have long had the habit of developing screens myself.
The thing is, some of those have held up wonderfully. Especially those relying heavily on sales growth rates, for example.

The sad part from a community point of view is that I stopped posting most of them.
The dark speculation is that there is a correlation between posting a screen and its holding up,
but it's hard to believe our posts have any material effect on the markets.
I doubt we're that important to the world.

There's an old saying, though. Never mistake the cycle for the trend.
Maybe in this infinite-liquidity-fuelled bull (I consider the 2020 Covid dip to have been merely a blip in it),
we have to wait for the fat bear lady to sing before drawing any conclusions about what works longer term.
Markets have been weird in the last 5-7 years in a lot of ways. That might be a new normal, or it might be merely a temporary excursion into the twilight zone.
I can't help but think that things like earnings and earnings growth might actually matter again some day.

Jim
Print the post Back To Top
No. of Recommendations: 1
The thing is, some of those have held up wonderfully. Especially those relying heavily on sales growth rates, for example.
and
I can't help but think that things like earnings and earnings growth might actually matter again some day.


Sounds very much like the driving forces on this board:
https://boards.fool.com/sauls-investing-discussions-120980.a...
Print the post Back To Top
No. of Recommendations: 5
One thing that always troubled me about a number of the MI screens was that they seemed like overtuning. Yeah, they backtested great, and indeed worked quite well for a while.....but then they ran into a brick wall.

The ones that had 15% annual returns just seemed too good to be true. Guess they were. Guess they were only good for a certain era of the market. Has happened before. Back in the late 90's and early 2000's there was a logical sounding screen that I got from a writer on MS Investor (RIP). Worked fine for a while and churned out a lot of profit. Time went on, and I drifted off of it. Later I backtested it on Portfolio123, and it stunk.

Jim has tossed out a bunch of ideas for screens, both on the MI and the Berkshire board. Maybe on other boards, too, that I'm not on. Actually, so many screens that I don't have enough money to invest in them all. ;-(
Some of them can be quickly backtested at Portfolio Visualizer. Simple models only, not the complex rules the our typical screens have.


I've gotten some other ideas from some of his ideas. Buy & hold things, not active trading things.

One set of ideas is this:
Vanguard Health Care Fund (VGHCX) and
T. Rowe Price Health Sciences Fund (PRHSX)
also possible: Fidelity Select Health Care (FSPHX)

All better returns and MaxDD and Sortino than S&P500.
Print the post Back To Top
No. of Recommendations: 0
At least for me, it is ideas like Ray's that will keep this group useful.
Print the post Back To Top
No. of Recommendations: 20
But I have long had the habit of developing screens myself.
The thing is, some of those have held up wonderfully. Especially those relying heavily on sales growth rates, for example.
The sad part from a community point of view is that I stopped posting most of them...


In the spirit of community sharing---

How about this for KISS two-step?
Value line stock universe and data fields
Step 1: Reported price-to-book ratio top 100
Step 2: Five year sales per share growth rate top 10
That's it.

Note that neither of those criteria change all that fast, so the turnover is quite low.

May '97 through July '21, beat S&P by 10.77%/year.
Beat the S&P in 77% of rolling years and 86% of rolling two year stretches.

If anything, it would have worked better lately, not worse.
Beat the S&P by 11.1%/year in the last 10 years, and by 19.2%/year in the last five.

At least it's simple.

You may wonder why it might work--high P/B is not a common factor in screens.
It's not just looking for overpriced stocks to jump on while they're still going up, though there is some of that.
As mentioned in a recent thread, very high P/B more often denotes a very good business than it denotes an overpriced one.
Extremely good businesses make profits without a requirement for a lot of assets.
This means they tend to trade at a high multiple of book value.

10 stocks monthly, no friction, lately:
2003     51.3     23.0
2004 11.4 8.9
2005 32.2 7.5
2006 15.7 13.8
2007 14.7 1.9
2008 -29.7 -32.9
2009 67.6 25.5
2010 29.7 14.2
2011 0.4 2.5
2012 23.1 17.1
2013 52.5 27.6
2014 25.7 12.9
2015 23.7 1.9
2016 -8.9 14.4
2017 33.1 21.8
2018 28.8 -3.5
2019 30.9 29.9
2020 94.4 16.9


The screen is improved by overtuning, as most screens are.
Put in a middle step of keeping only the 85 stocks closest to their 52 weeks highs, and the backtest improves on every metric.
CAGR up by 1.7%, win rate up, risk metrics down.
This just eliminates the few high P/B firms that are most out of fashion at the trade date.

Interestingly, the picks are mostly not tickers I immediately recognized, with the exception of occasional incursions by Tesla, Amazon, Netflix and Baidu.

It's not obviously relying on just a few winners.
Detuning it to 20 stocks deep, quarterly, adding friction, '97 to '21--
Basic version still beat the S&P in backtest by 6.3%/year
The overtuned three-step version beat by 8.7%/year

Jim
Print the post Back To Top
No. of Recommendations: 0
Nice.
I believe it was called 'mound of toast' or something like that, the idea is to evaluate the robustness of results as parameters are changed.
In principle, one could report results for e.g. P/B in top 50 through top 150 with sales per share growth in e.g. top 5 to 15.
Would be looking for a broad mound of parameters that produce somewhat similar results, not a sharp mountain peak.

Not that I'm suggesting work for anyone (:->
Just wondering if parameter robustness could help identify screens that don't have some occult over-tuning.
Print the post Back To Top
No. of Recommendations: 1
How about this for KISS two-step?
Value line stock universe and data fields
Step 1: Reported price-to-book ratio top 100
Step 2: Five year sales per share growth rate top 10
That's it.


Aren't sales/share, and so sales/share growth, industry specific--that is some industries typically have greater (or lesser) sales/share?

If so, maybe modify Step 2 (or add a Step 1a) to the effect of sales/share > industry average and growing?

'Course, just Step 2 might gravitate toward particular industries.

Eric Hines
Print the post Back To Top
No. of Recommendations: 14
The OP suggested perhaps changing the focus of the group from mechanical screening to e.g. selling cash secured puts, crypto, ... most anything that could generate alpha. All interesting topics, and presumably everyone here likes alpha.

But having spent a career in basic research, one lesson I've learned is that if something doesn't work that you really think should work, then it's often very fruitful to find out why it doesn't.
Often, you'll learn something that's new and valuable, but perhaps valuable in ways you never ever expected.

Lots of people here have put lots of effort into developing mechanical screens. If the screens mostly don't work, it could be valuable to understand in more detail why they don't.
Would testing parameter robustness help to avoid over-tuning?
How about bagging, boosting etc?
Mechanical screens are decision trees. There are machine learning methods to generate decision trees, e.g. CART, where great attention has been paid to not over-tuning. Might approaches like these from the statistical/machine-learning literature help?

What if nothing helps?
It's still valuable to know, not guess, that such financial data is a noisy mess (at least as evaluated with current methods).
Print the post Back To Top
No. of Recommendations: 3
Mark, in response, I'll throw you a challenge. Take a theme that we know HAS been working the last 10-15 years, just not necessarily classic Stock Investor screens that are aging out, and try some refreshed backtesting / data mining out to explore the theme.

Here's two ideas:
* Hi growth stocks like Tech Growers / "Saul". Maybe start with resurrecting one or two of the old simple VL pure earnings and revenue growth screens, with or without momentum, and try out a new screen without T-ness or Safety rank.

* GTAA-4 and GTAA-6 regular and aggressive growth versions. Different rebalance frequencies, potentially different ETFs.
==> My GTAA-6 portfolio including a Saul/Bowley quarterly growth screen as the main "large cap momentum" component, and the "standard" timing signals discussed here, and cash/fixed income, has crushed the Vanguard Balanced 60/40 fund the last 2 years, and has caught up and passed that for trailing 5 year performance. At the beginning of this year I stopped using the last SIP blend I'd been using for the past few years, after 3 more years of underperformance.

You may have other ideas. I encourage you to explore and report on the results.

FC
Print the post Back To Top
No. of Recommendations: 4
Aren't sales/share, and so sales/share growth, industry specific--that is some industries typically have greater (or lesser) sales/share?

Yes and no.
Yes, you're right, sales per share can't be compared between companies in different industries and get a meaningful result.
Low P/S screens work occasionally, but mostly by luck.
It's the profits and their growth that ultimately matter, and that can come from a high margin (Hermes) or low margin (Walmart) business.

But the rate of growth of sales per share is a not-bad proxy for the growth of the business, whether sales per share are low or high.
You're comparing a firm against its older self, not to another firm, so it is a meaningful metric.
There is no need to compare it to others in the industry. To a certain extent, any firm whose sales per share are rising at 20%/year has some interesting characteristics, whether it's a flower shop or a mine.
The "per share" view also has the advantage that a total top line growing from mergers won't float to the top if the mergers increased the share count.
It suitably penalizes firms using dilution.

Sale growth allows one to look at growth rates among [currently] not very profitable businesses, which is sometimes of interest.
And whether profitless growth is of interest to me for my hand picked stocks (not that much),
it's *definitely* a factor of interest to market investors in the last 20 years.
Above all else, it seems.

Jim
Print the post Back To Top
No. of Recommendations: 2
The focus of the group needs to change to things that generate alpha.

Why necessarily concentrate on alpha? Why not more like the QTAA that matches alpha but decreases beta?

JLC
Print the post Back To Top
No. of Recommendations: 5
I believe it was called 'mound of toast' or something like that, the idea is to evaluate the robustness of results as parameters are changed.
In principle, one could report results for e.g. P/B in top 50 through top 150 with sales per share growth in e.g. top 5 to 15.


I honour your memory of the mound!
We should do it more often.

I tried 20,30,40...200, then 250,300...500, then trying just ensuring P/B ratio is positive, then skipping it entirely.

To avoid any "attractor" at previous tuning, I used a new depth (15), hold period (2 months), and date range (2000 to date), with friction.

CAGR, GSD, and Sharpe all have their best figures with P/B top 160.
Win rate has two peaks, one at 170 (almost tied with the 160), and another way down at 20.

Since the 160ish range seemed nice, I tried it by 5 around there.
Sure enough, 165 is a hair better on all of CAGR, GSD, and Sharpe.

So, that would leave the screen:
[value line set]
P/B top 165
5-year sales growth top N (e.g. 10,15,20, but 15 was tested here)

1997 to date, same version tested top 15 each two months with friction: beat the S&P by 11.8%/year, no better or worse in risk using DDD3 metric.
Very much better in the second half of the backtest than in the first.

Maybe just luck, but if you have only one thing to tune you might as well take the best neighbourhood.

Jim
Print the post Back To Top
No. of Recommendations: 0
Mark, in response, I'll throw you a challenge. Take a theme that we know HAS been working the last 10-15 years, just not necessarily classic Stock Investor screens that are aging out, and try some refreshed backtesting / data mining out to explore the theme.

Here's two ideas:
* Hi growth stocks like Tech Growers / "Saul". Maybe start with resurrecting one or two of the old simple VL pure earnings and revenue growth screens, with or without momentum, and try out a new screen without T-ness or Safety rank.


I would but I am ashamed to admit I don't have the skills in GTR1 necessary to do that.
Print the post Back To Top
No. of Recommendations: 0
* GTAA-4 and GTAA-6 regular and aggressive growth versions. Different rebalance frequencies, potentially different ETFs.
==> My GTAA-6 portfolio including a Saul/Bowley quarterly growth screen as the main "large cap momentum" component, and the "standard" timing signals discussed here, and cash/fixed income, has crushed the Vanguard Balanced 60/40 fund the last 2 years, and has caught up and passed that for trailing 5 year performance. At the beginning of this year I stopped using the last SIP blend I'd been using for the past few years, after 3 more years of underperformance.

You may have other ideas. I encourage you to explore and report on the results.


PS - I am glad you are having success.
Print the post Back To Top
No. of Recommendations: 47
The focus of the group needs to change to things that generate alpha.

I categorically, vehemently, disagree. There are hundreds if not thousands of message boards at TMF, and there are a thousand times more forums out on the internet about every conceivable investing strategy. You want to spread your wings, pursue other interests and strategies? Knock yourself out. Go explore. Don't limit yourself. This message board is defined by a subject matter. Nothing else. Nothing else brings us here or keeps us here. If its name or subject matter changes, it is no longer the same board or the same group of people. As far as I'm concerned, this board can continue to exist as long as at least two people come to visit and exchange ideas, and if not it will die a natural death.

Elan
Print the post Back To Top
No. of Recommendations: 1
I categorically, vehemently, disagree. There are hundreds if not thousands of message boards at TMF, and there are a thousand times more forums out on the internet about every conceivable investing strategy. You want to spread your wings, pursue other interests and strategies? Knock yourself out. Go explore. Don't limit yourself. This message board is defined by a subject matter. Nothing else. Nothing else brings us here or keeps us here. If its name or subject matter changes, it is no longer the same board or the same group of people. As far as I'm concerned, this board can continue to exist as long as at least two people come to visit and exchange ideas, and if not it will die a natural death.

I guess I was thinking about new strategies that are somewhat mechanical. we already have them here. Buying health care funds is somewhat mechanical, as is buying an equal weight ETF, as is buying preferred stocks based on set metrics. I meant non stock screen strategies.
Print the post Back To Top
No. of Recommendations: 14
I have followed this board closely since 1998, and have been an inconsequential contributor at various times in the past, though not in recent years, since I don't have the financial or technical chops to be genuinely helpful.

Regardless, I would be sorry to see this board disappear, and even sorrier to see it turn into a place where "things that [purportedly] generate alpha" are taken seriously absent a substantial, reproducible back-test.

I cannot overstate how important the mechanical part of mechanical investing has been to my own success as an investor.

Baltassar
Print the post Back To Top
No. of Recommendations: 1
Would testing parameter robustness help to avoid over-tuning?
How about bagging, boosting etc?
Mechanical screens are decision trees. There are machine learning methods to generate decision trees, e.g. CART, where great attention has been paid to not over-tuning. Might approaches like these from the statistical/machine-learning literature help?


Not sure how people validate/test screens, but the standard for time series like stock prices is cross validation done in a specific way. You would identify a screen using data from years (1, 2 to 10), for example, and then test/validate the screen on future years (11, 12, 13), for example. The main point is that by testing a screen on data not used to identify/create the screen, you protect against overtuning and increase robustness of the screen/model. See (https://otexts.com/fpp3/tscv.html) for more on time series cross validation (apologies that this is a somewhat technical link).

I haven’t tried to do any screen like what this board does, but I believe that in machine learning this would be what is called a classification problem. One could build a decision tree that models this classification problem, like as follows.

For a set of candidate stocks, the goal is to identify the stocks that have the highest future returns (rank in top 10% of returns or some other criterion for “very good stock”). Classify the highest future return stocks with label 1, and classify all others with label 0. The machine learning decision is to identify the attributes (features) of the stocks (P/B above some threshold, revenue growth above some threshold, etc - this could be a very long list) using past data that most correctly classifies the high return stocks in some future period. There will be two types of errors in the model: false positives (low return stocks that are incorrectly classified as 1) and false negatives (high return stocks that are incorrectly classified as 0). The goal of the model/screen is to minimize these errors, which is where the cross validation mentioned above comes in.
Print the post Back To Top
No. of Recommendations: 2
I/we encourage you again: do something - focused, and tactical - about it.

I meant non stock screen strategies.
Print the post Back To Top
No. of Recommendations: 0
I believe it was called 'mound of toast'

Can anyone tell me where the term 'mound of toast' came from? Is it thought to literally describe the ideal shape of a graph of returns vs 2 parameters?

John
Print the post Back To Top
No. of Recommendations: 14
I welcome a discussion about mechanical investing topics that might be of interest. I am mostly interested in US stocks, Options, Tactical Asset Allocation, and Timing. I would have to be convinced to care about Crypto, NFT, or other alternatives. Alpha (the unknown or error term) is not my focus. Understanding the market (including news reports), avoiding mistakes, and aiming for excess returns is more interesting to me. I think backtests are not always needed.

Graham wrote about mechanical investing in the 1973 edition of "The Intelligent Investor", and in Chapter 15
Stock Selection for the Enterprising Investor wrote "we would end up with about 150 companies meeting all six of our criteria of selection. The enterprising investor would then be able to follow his judgment—or his partialities and prejudices—in making a third selection of, say, one out of five in this ample list." Graham did not include a backtest.

Activity on the MI board has been low recently (with an average of 5 posts a day in the last month). Average number of posts per day has been declining for 20 years (roughly halving every 7 years).

   date     MI_post  posts_per_day
4/8/1997 1
11/21/1998 10000 17
12/16/1999 50000 103
4/27/2001 100000 100
9/19/2003 150000 57
6/17/2007 200000 37
8/4/2011 231937 21
4/10/2014 250000 18
9/24/2018 271937 13
10/23/2021 281937 9


I have not posted much recently, and find myself posting on other boards. Some of my activity on other boards is answering questions or adding a different viewpoint to a discussion. There's just not much naivete on the MI board to respond to. My tendendcy is to post when I disagree, and do nothing if I agree. And there's not much motivation to post when the US stock market 3-year CAGR is 21.

I find value in this board, and read many of the posts. There are many valuable posts from years ago that are well worth reading, but they can be difficult to find.

"They are as excited as us to try and figure out this problem around online communities," said Mr D'Aloisio. "It's not only a hugely interesting problem, but also a necessity. All groups have the potential to become genuine communities. But most groups suffer from problems in online communication that prevent community-building - things like awkward silences, conversations going off-topic, and vitriol. However, we learned over the past two years that a group can transform into a community if its members feel their participation is welcomed."
https://www.bbc.com/news/technology-58998315

--- data ----

          Board Name             30 Day Total  posts
Macro Economic Trends and Risks 2794
Political Asylum 1979
Conservative Fools 1912
Berkshire Hathaway 823
Saul’s Investing Discussions 546
Retirement Investing 456
Atheist Fools 379
Pandemic Disease 379
Humor and Urban Legends 297
Climate Change 228
Building / Maintaining a Home 171
Help with this STUPID computer! 158
Mechanical Investing 153
Advanced Micro Devices 150
Retire Early CampFIRE 150
Destiny Solutions Corporation 149
Where Angels Fear to Tread 140
Living Below Your Means 132
Health and Nutrition 123
Major League Baseball 121
Political Quagmire 114
Tax Strategies 111
British Invasion 96
Retired Fools 89
Best Travel Spots/Tips 88


                                                  5-year  3-year
Name Ticker CAGR CAGR
Invesco S&P MidCap Momentum ETF XMMO 25 24
Cambria Shareholder Yield ETF SYLD 19 23
Value Line Mid Cap Focused VLIFX 20 21
First Trust Rising Dividend Achiev ETF RDVY 20 21
Vanguard US Momentum Factor ETF VFMO 21
Barclays ETN+ Shiller Capet CAPE 19 21
Alpha Architect US Quantitative Momt ETF QMOM 17 21
Invesco S&P 500® Momentum ETF SPMO 21 21
iShares MSCI USA Momentum Factor ETF MTUM 21 21
AI Powered Equity ETF AIEQ 21
Vanguard Total Stock Market ETF VTI 18 21
iShares MSCI USA Quality Factor ETF QUAL 18 20
SPDR® S&P 500 ETF Trust SPY 18 20
iShares MSCI USA Size Factor ETF SIZE 17 19
DoubleLine Shiller Enhanced CAPE® N DSENX 18 19
Invesco S&P 500® Equal Weight ETF RSP 16 19
PIMCO RAE US Small Instl PMJIX 16 18
Vanguard US Quality Factor ETF VFQY 17
Schwab Fundamental US Large Company ETF FNDX 16 17
Schwab Fundamental US Large Company Idx SFLNX 16 17
Schwab Fundamental US Broad Market ETF FNDB 16 17
Invesco FTSE RAFI US 1000 ETF PRF 15 16
Invesco S&P SmallCap 600® Equal Wt ETF EWSC 15 16
iShares Core S&P Mid-Cap ETF IJH 15 16
Vanguard US Liquidity Factor ETF VFLQ 16
Invesco S&P MidCap 400® Equal Weight ETF EWMC 14 16
Vanguard Health Care ETF VHT 16 16
AdvisorShares DoubleLine Value Eq ETF DBLV 12 16
SPDR® MSCI World StrategicFactors ETF QWLD 14 16
Invesco Zacks Mid-Cap ETF CZA 14 15
Knowledge Leaders Developed World ETF KLDW 13 15
Schwab Fundamental US Small Company Idx SFSNX 14 15
Schwab Fundamental US Small Company ETF FNDA 14 15
PIMCO RAFI Dyn Multi-Factor US Eq ETF MFUS 15
First Trust Health Care AlphaDEX® ETF FXH 15 15
Invesco FTSE RAFI US 1500 Small-Mid ETF PRFZ 14 15
Invesco S&P 500® ex-Rate Snsv LowVol ETF XRLV 15 15
iShares MSCI USA Multifactor ETF LRGF 15 15
iShares Core S&P Small-Cap ETF IJR 15 14
Innovator IBD® 50 ETF FFTY 17 14
Vanguard US Value Factor ETF VFVA 14
iShares S&P 500 Value ETF IVE 13 14
iShares S&P Mid-Cap 400 Value ETF IJJ 13 14
O'Shares US Quality Dividend ETF OUSA 13 14
iShares MSCI USA Small-Cap Mltfctr ETF SMLF 14 14
DFA US Micro Cap I DFSCX 14 13
iShares Biotechnology ETF IBB 12 13
Vanguard US Multifactor ETF VFMF 13
Hartford Multifactor US Equity ETF ROUS 14 13
DFA US Small Cap Value I DFSVX 12 13
Hennessy Cornerstone Growth Investor HFCGX 11 13
Invesco Dynamic Large Cap Value ETF PWV 11 13
iShares MSCI Global Multifactor ETF ACWF 12 13
PIMCO RAE US Instl PKAIX 13 13
iShares MSCI Intl Small-Cap Mltfct ETF ISCF 11 12
Alpha Architect Intl Quant Momt ETF IMOM 8 12
iShares MSCI USA Value Factor ETF VLUE 13 11
Cambria Global Asset Allocation ETF GAA 8 11
Alpha Architect US Quantitative Val ETF QVAL 12 10
Vanguard US Minimum Volatility ETF VFMV 10
Cambria Global Momentum ETF GMOM 8 9
Hennessy Cornerstone Value Investor HFCVX 10 9
iShares Edge MSCI Multifactor Intl ETF INTF 8 8
Hartford Multifactor Dev Mkts (exUS) ETF RODM 8 7
Vanguard Global Minimum Volatility Admr VMNVX 8 7
Core Alternative ETF CCOR 7
Innovator Ldrd Fd of S&P500 PwrBffr ETFs BUFF 9 7
Cambria Global Value ETF GVAL 7 5
WBI BullBear Value 3000 ETF WBIF 9 3
Alpha Architect Value Momentum Trend ETF VMOT 2
Alpha Architect Intl Quant Val ETF IVAL 4 2
Print the post Back To Top
No. of Recommendations: 10
<< Can anyone tell me where the term 'mound of toast' came from? >>

"Mounds of Toast" was, IIRC, coined my mrtoast himself....an early, prolific poster on this board (dating back to 1998).

He would backtest screens by varying a given parameter over a reasonable range. It was his way of uncovering over-tuning a screen.

If a screen's results held up reasonably well while varying that parameter by several percent (for example), he termed those results a "mound of toast".

If the results fell off precipitously (as opposed to a gentle "mound"), he deemed that screen as over-tuned/data-mined.

Like so many others, he has appeared to have wandered off.


Alan
Print the post Back To Top
No. of Recommendations: 6
"Mounds of Toast" was, IIRC, coined my mrtoast himself....an early, prolific poster on this board (dating back to 1998).

It was indeed named for mrtoast, but I think it was coined in homage to him and his advocated test method, not by him.

Jim
Print the post Back To Top
No. of Recommendations: 2
Like so many others, he has appeared to have wandered off.

IIRC, mrtoast decided to go into the financial advisor and/or CFP business, and he felt that he needed to exit this forum to keep himself out of potential hot water.

Tails
Print the post Back To Top
No. of Recommendations: 1
Jim,

I'm confused by your 'mound of toast' post -- I don't see the mound, I see the peak:

You found P/B criterion of top 165 or 170 (also 20) was even better than the original P/B of top 100 criterion:
CAGR, GSD, and Sharpe all have their best figures with P/B top 160.
Win rate has two peaks, one at 170 (almost tied with the 160), and another way down at 20.
Since the 160ish range seemed nice, I tried it by 5 around there.
Sure enough, 165 is a hair better on all of CAGR, GSD, and Sharpe.

That seems to be peak finding (i.e. 'tuning' or 'curve-fitting').

But the issue is whether the results don't drop dramatically as you get a reasonable distance away from from the peak. You mentioned trying different P/B limits:
20,30,40...200, then 250,300...500, then trying just ensuring P/B ratio is positive, then skipping it entirely.
So what happened?
Is there a mound around using top N of 165 or 170, or a cliff?
I'm guessing that you found a mound, but you didn't say.

Similar question for sales growth criterion -- you changed to top 15.
Were the top 10, 15, 20 values that you mentioned all somewhat similar, or at least, no cliffs?
Print the post Back To Top
No. of Recommendations: 9
<< I categorically, vehemently, disagree.....This message board is defined by a subject matter. Nothing else. Nothing else brings us here or keeps us here. >>

I sure don't see it that way. For me, this board is much more than just its purported subject matter. It's liveliness and longevity is predicated on the people who show up here, not just on what they backtest.


<< If its name or subject matter changes, it is no longer the same board or the same group of people. >>

Sure, our focus should remain on Mechanical Investing. But I still visit this board daily (for over 20 years) as much for the side-bars as the backtests. And let's face it....this isn't really the same board or group of people. It may be hard to notice on a granular, day-to-day basis, but this board evolves as the years go rolling by. That's probably why is has (yet) died a "natural death".


Alan
Print the post Back To Top
No. of Recommendations: 11
Is there a mound around using top N of 165 or 170, or a cliff?
I'm guessing that you found a mound, but you didn't say.


Most mounds of toast are a bit bumpy--those crust edges!

This mount is not super smooth, but VERY broad.

It works OK with 20 best by P/B.
It works beast with 165 best by P/B.
It works OK with 500 best by PB.
It doesn't work particularly well without the test.

Here are some of the CAGR figures by cutoff of that step.
 50   13.4
60 13.7
70 12.9
80 13.6
90 12.8
100 13.2
110 11.6
120 13.0
130 15.0
140 15.8
150 16.1
155 16.7
160 18.4
165 18.6
170 18.0
180 16.8
190 16.1
200 14.8
250 14.1
300 15.6
300 15.6
350 15.6
400 15.2
500 14.2

So, on the one hand, it's not a beautiful smooth hump.
But the good news is
* it's not at ALL brittle on the tuning parameter, which is the bigger lesson from the toast theory.
Really, the gain seems to come from skipping low and medium P/B firms.
* all these returns are pretty good. SPY returned 7.20% in this stretch.
Merely requiring P/B be positive returned 10.9%.

Similar question for sales growth criterion -- you changed to top 15.
Were the top 10, 15, 20 values that you mentioned all somewhat similar, or at least, no cliffs?


Yup, I purposefully changed to top 15 for the toast-test so I'd be testing with a depth other than what I originally tuned to.
(which was top 20, 2003-2020).

It depends mainly on how many stocks you want.
CAGR by depth:
20   15.0
19 16.4
18 16.6
17 16.4
16 18.1
15 18.6
14 18.0
13 17.7
12 17.0
11 15.6
10 15.5
9 16.3
8 16.1
7 14.2
6 14.7
5 18.4

The main lessons from this seem to be:
* statistical support might be fading as you get down into the 6-11 range, and or we're hitting the limits of the quality of the final sort.
* ...though the final sort seems pretty adequate, overall. The steady rise from 20 to 15 is encouraging.
* it's a usefully deep screen.

I use an alternate final sort offering backtest returns still rising all the way down to "top 2".
(I would never use a two stock screen, but it's a nice test of the sort)

Jim
Print the post Back To Top
No. of Recommendations: 7
Step 1: Reported price-to-book ratio top 100
Step 2: Five year sales per share growth rate top 10
...Extremely good businesses make profits without a requirement for a lot of assets.
This means they tend to trade at a high multiple of book value.


Jim, you've finally come around to seeing the benefits of low Capex, something we discussed a number of times.
:-)

At any rate, from 10 years ago this month, a post that uses Capex/Sales
https://boards.fool.com/six-or-seven-years-ago-wotdabny-poin...
The five lowest Capex/Sales picks outperformed the S&P500 by 8 percentage points, and the five lowest underperformed the index.

DB2
Print the post Back To Top
No. of Recommendations: 0
Looks like a nice mound to me!

Something sounds wrong about that comment ... maybe change it to "looks reasonably robust to parameter changes to me!"
Print the post Back To Top
No. of Recommendations: 11
I'm here every day, been reading the board for over 20 years now.
So much amazing wisdom posted here, the threads over the years have influenced so much of my thinking, and still do.

Most of what I think is still centered around the archaic notion of an actively traded blend that one hopes will outperform the market over time.
Seems like the time for that line of thinking has passed.
But, I'm still stuck there mentally :)



Mark
Print the post Back To Top
No. of Recommendations: 4
Seems like the time for that line of thinking has passed.
But, I'm still stuck there mentally :)


I too seem to have a unreasonable mental desire to find that seemingly unobtainable set of parameters that outperform. I’m still playing with about 15% in a greatly modified ROE_CASH screen. The screen itself has only equaled the S&P but ‘unmechanically’ I have been spending time looking over the selections, comparing them to their competitors, seeing what the analysists point out and then rejecting around 30 to 40% of them. This works out much better on a very low turn over screen.

Another ~20% are stocks picked from the Fool Stock Advisor.

The other 65% is in an allocatesmartly optimization of 3 different Tactical Asset Allocation models.

But over especially the earlier years MI has made my retirement significantly more comfortable.

RAMc
Print the post Back To Top
No. of Recommendations: 19
I too seem to have a unreasonable mental desire to find that seemingly unobtainable set of parameters that outperform...

I'm not sure all the disillusionment and abandonment is deserved.

The key, I think, is realistic goals.
My observation is that many of the screens that have very few parameters, meaningful depth (15+, up to 50), and very modest expectations seem to keep doing fine.
More like a custom index than what we used to think of as a 4-5 stock screen.
Remember, if you can outperform long term by 2-3%/year you're beating almost all professional money managers.
I think that's much more likely with a 30 stock screen that backtests at beating by 3-5%/year than with a five stock screen that backtests at beating by 20-30%/year.

This one...https://boards.fool.com/a-simple-screen-34742375.aspx
...returned 68.8% in the last year (the 15 stock, hold-till-drop at 25 version), beating the S&P by 25%.
That's wildly anomalous, but at least the sign is correct.
Mix of backtest and out-of-sample shows an advantage of 4-5%/year in the last 15 years, 3.7% in the last three.

Another simple one of mine (sum of 4 one-factor sort ranks), 30 stocks deep, beat the S&P in all of the last four calendar years out of sample.
Overall advantage after friction 6.3%/year versus the S&P in the last 3.7 years.

I don't think we should conclude that stock screening is dead or doesn't work.
I think we just have to keep away from our old habits of picking too few high flying stocks in an overtuned screen.
Low expectations are often met.
Nothing ever returned 20%/year over the long run, and nothing ever will.

Sure, some things we thought were *really* reliable turned out not to be. YLDEARNYEAR is the one that springs first to mind.
Especially during the nuclear winter for dividend stocks. Ouch.
A portfolio of the highest yield 50% of the VL dividend stocks lost -6.9%/year from Dec 2016 to April 2020, while the S&P 500 rose +7.3%/year.
That was a portfolio of over 500 stocks--so much for the idea that all diversified portfolios must be index trackers!

Jim
Print the post Back To Top
No. of Recommendations: 11
I'm not sure all the disillusionment and abandonment is deserved.

Remember, if you can outperform long term by 2-3%/year you're beating almost all professional money
managers.


Nice down to earth perspective. By that measure I have done decently.

After twenty plus years of retirement my net worth has increased more than the S&P after living
expenses. And I haven’t had to share any of the proceeds with a financial advisor. So, I should have
nothing to complain about. The outstanding insight from members of this board have greatly helped me
through this journey.
RAMc
Print the post Back To Top
No. of Recommendations: 4
Remember, if you can outperform long term by 2-3%/year you're beating almost all professional money managers.
...
Nice down to earth perspective. By that measure I have done decently.



I think it's doubly important because my theory is that the screens with modest aims are more likely to hit their target.
It's a theory, anyway.

Jim
Print the post Back To Top
No. of Recommendations: 8
Like Elan said, I emphatically disagree with the idea that an actively traded blend that one hopes will outperform the market over time.
Seems like the time for that line of thinking has passed.


The Question is: an actively traded blend OF WHAT? and BASED ON WHAT?

Clearly, the 15-25 year old screen definitions of mostly micro-cap stocks, from a time before widely available factoring / breakdown software and factor-based ETFs creating sliced-up tranches of investments, are broken.

But that's just one theme. Think about it.

Just because the 1998 definition of SPARK or the 2002 definition of PEG-variant18 or Gentle Screamers doesn't work anymore, so what?

It's undeniable that large cap tech (FAANG / MAGA whatever) has dominated every other US equity asset class the last 12 years (except maybe Sensei).

In the bro retail gaming / competition psychology driven environment we're in now, with some research using the exponentially better tools that have evolved online recently I'm confident some worthwhile NEW MI ideas could be developed.

Like TechGrowers - not "strictly" mechanical but meh, close enough.
Print the post Back To Top
No. of Recommendations: 8
Just because the 1998 definition of SPARK or the 2002 definition of PEG-variant18 or Gentle Screamers doesn't work anymore, so what?


I'm suspicious of the "does not work anymore" line.
My intuition is that it's more properly "the history wasn't long enough to capture times when the screen/blend does poorly".
Maybe after capturing those periods, you conclude it never really worked, which is a bit different.
Could also be "I didn't understand the statistics and had the wrong expectations".

Surely there are screens that really do stop working; RS-26 comes to mind.
But I'm too busy or lazy to really check into those things.

My own blend had a spectacularly poor period from around 2018-2020, a solid three years or more.
Of course, since I stopped using it, it is up 110% CAGR.
Did it stop working and then start working again?
I'd say it more more a rough patch. Stopped working implies it will not work again.

It's all semantics I guess.
Print the post Back To Top
No. of Recommendations: 8
True, "stopped working" is up to the user's definition.
In my case, if a screen does not provide significant return above what one can get by simply investing in an ETF like the Small Cap Growth IJT, or IWM, it just doesn't matter.

DATA: IJT was up 54% from 5/1/2018 to 5/1/2021.
My last used blend (vg-horse, lowps and HRV) was up 5% over that time. Those screens were among the highest ranked in recent 6m-1y-3y performance up to 5/1/18.

When a blend that has a backtestable history of beating the S&P stops doing that, over a 3 year period, it creates a unrecoverably large performance gap that is impossible to ignore.

Specifically, an unacceptably large number of these small and micro cap stocks still get crushed to the tune of 20 to 30% in any given iteration (month, 3 weeks, selection period). As a small percentage of an ETF, that's no big deal. As 1/9th of a 9 stock / 3 screen blend, it's very damaging in sequence. In the context of one of the strongest and longest bull markets of all time, it's unacceptable.

And "not working" can mean "worked fine in history before the advent of low-cost ETFS". Which many of these did, as early adopter versions of what is now known as "factors".
Factors are going into and out of style faster and for unknown reasons than any of us can follow / understand. We need to understand "what's working lately". But that contribution to the community was lost a couple of years ago.
Print the post Back To Top
No. of Recommendations: 18
True, "stopped working" is up to the user's definition.
In my case, if a screen does not provide significant return above what one can get by simply investing in an ETF like the Small Cap Growth IJT, or IWM, it just doesn't matter.
DATA: IJT was up 54% from 5/1/2018 to 5/1/2021.
My last used blend (vg-horse, lowps and HRV) was up 5% over that time.


Ouch.
My condolences, seriously.

Perhaps the "stopped working" is more common among smaller stocks.
Also maybe in part because we often measure "not working" relative to the S&P, and the S&P is mainly very very big stocks.
Anything that has stocks far from that definition will have returns that diverge a lot...for better or worse.

But FWIW, as an example of my "aim low" theory, I posted the LargeCapCash screen a while back.
https://boards.fool.com/a-spy-alternative-screen-34516863.as...
"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."

First 16 months out of sample (and a wild time it has been), after friction:
S&P: CAGR 41.48%
LargeCapCash as in kickoff post (40 stocks, hold-till-drop at 45, two month holds): CAGR 45.6%
LargeCapCashDiv as in kickoff post (ditto, but stocks have to pay a div): CAGR 56.3%.
These are the screens before the consensus improvements later in the loooong thread.

The first, simplest one had an advantage of 4.1%/year over the S&P in this short post-discovery era.
The advantage in backtest Jan 2000-April 2020: 4.0%/year.

Certainly this is largely luck: big cash-rich firms have obviously been the best picks lately.
But, that's kind of the point of the screen...they're usually pretty good bets, it seems.

I mention this because of what it might say about screen selection.
My hypothesis (with little evidence) is that, since the 4% improvement was so modest, it was more likely to be met.
20 years ago nobody would have read a post about a backtest that improved on the market by only 4%/year.
Yet, if someone actually accomplished that in the last 20 years, they'd have 2.2 times as much money in the bank compared to the Boglehead.

Jim
Print the post Back To Top