No. of Recommendations: 70
May, 2006

What is MI? part 3 (of 6): Screens, Switching

---- Screens
What is a “stock universe”?
What are some resources available to conduct backtesting of MI strategies?
How do you create a stock screen?
Ok, what are these famous “screens”?
What is WER?
What is BI?
What is BMW?
Where did all these screens COME from?
How do I choose among all these screens?
So do you really just follow the screen?
What is a short screen?
What is “HTD”?
What is “SOS”?
What is “Dozens”?
What is a Bullet?
What is blending?
How do you construct a blend of screens for a stock portfolio?
---- Switching
What is an ETF?
Are some screens better for different parts of the business cycle?
Is there a January effect?
Is there seasonality in screens?
What is the Rotor3 approach?
What's working lately?


“Watching the portfolio trades will be like watching grass grow. Once a month you mow the lawn. ;-)”
Elan in FW # 22151, 12/5/2000


What is a “stock universe”?

It's a shorthand phrase we use to indicate a data provider. The data providers we use most on this board are the ValueLine Investment Survey, and Stock Investor Pro from AAII. The set of stocks covered in SIP is not the same as is covered in VL: so we often ask “Which universe?” when someone talks about talks about taking “all stocks” and filtering them somehow.

There are advantages and disadvantages to each. VL does more of the legwork for you in terms of filtering out crap; SIP includes more stocks, many many more stocks, so you have the potential to find opportunities there that aren't in VL. VL has the longer history, which is a huge advantage in assessing the robustness of backtested screens.

There are other “stock universes”, of course: the stocks available at MSN or at Zacks or elsewhere. If you have an account at FolioFN, the “window stocks” comprise another universe, which overlaps pretty well with the VL screens but less so with SIP. The thing you want to be aware of is, working with any particular data provider means that you're looking at the list of stocks they cover, and they don't all cover the same stocks. So two people can each create a list of “all stocks sorted by price/sales ratio descending”, and the lists can be completely different if the two people are working with different providers. “All stocks” does not always mean “all stocks”.

What are some resources available to conduct backtesting of MI strategies?

Remember we defined MI very broadly, to include non-stock strategies. There are mutual funds, ETFs, bonds, indexes, options, futures. You can usually get historical data for ETFs and bonds and indices off a free source like Yahoo or MSN. You can download that to a local CSV file and get to work. Historical options data is pretty much non-existent, to the best of my knowledge: we don't have any. When we've examined options strategies, it's usually been by simulation rather than with actual historical data. In terms of futures, not much is freely available to the nonprofessional for backtesting of strategies involving futures trading, to the best of my knowledge. There is a big market of products aimed at developing & backtesting systems for trading futures: you can buy lots of stuff.

For stock screens, on this board we have Jamie for VL; Keelix for SIP; and tradesim for WER

VL; Jamie

SIP; Keelix

WER; tradesim

More tidbits about the backtesters tucked in among the answers to other questions, below.

How do you create a stock screen?

LAPropDoc fielded this question very very well, in a little tutorial he wrote years ago, so I'm just going to reference that post. Basically you start with an idea and some data to backtest with…

Rough Draft Lesson Two, DG # 1369 7/19/2003
(Scroll down about 1/3 of the way, to “2.4 The ScreencBuilder”)

Here are three other instructive threads, case studies in screen building: CAPRS, RSPEG and YieldEarnYear. They illustrate the process: idea, backtest, questions, tweaks, more questions, acceptance. This is how you go about doing it.

The Monster Screens, MI # 70434 5/31/2000

New RS Screen, MI # 100857 5/8/2001

Interesting Metric, MI # 145850 6/8/2003

Ok, what are these “screens” that the MI board talks about?

You know that aphorism about today's technology, where if you buy a computer or an mp3 player today, it will be obsolete in 6 months? Or maybe 6 weeks? The same is emphatically true about stock screens on this board. Someone is always coming up with a new wrinkle, others get excited about it, it gets included in the list of stuff we track. Part of the challenge of writing an intro to MI is coming up with something that will have a shelf life of longer than couple weeks: any list of screens I gave here would be sadly out of date very quickly. So I'm not going to attempt any listing of screens here.

Instead, go to the Datahelper MI search page here:

Click the “Rankings” button on the lower left of the search page. We've got weekly posts with the latest & greatest rankings for the standard screens using the VL universe, the alternate screens using RRS on the VL universe, the SIPro universe, alternate screens run on the SIPro universe, the XG family of screens,

Screen rankings 10MAR06

Screen definitions 2/18/2006:

Base screens pt1

Base screens pt2; Shorts; SOS

Base screens that use IBD data

SI rankings 24FEB06

Exponential Growth Rates 3/10/06

Alternate Screen Rankings 20060310

Alternate SI rankings: 10MAR06

ETF Rankings 2/25/05

ETF - RRSXG Screen Ranks 2/22/2006

What is WER?

The Investors Business Daily WeekEnd Review.

IBD mechanical screen, MI # 33088 7/30/1999

BD Mech Screen: IBD WER RS-AD, MI # 36007 8/29/99

So they dug up the old back-issues

Weekend Review Start Date Found, MI # 41599 10/15/1999

The group who was doing the work of digging into the WER stuff moved to Yahoo groups, in order to more tightly coordinate their efforts.

Throwing in the towel, MI # 91925 2/1/2001

Mechanical_Investors, MI # 92710 2/8/2001

WERST-10 Update Sep 01, WER # 1185 2/10/2002

And they never moved back!

Decentralize = Devolve, MI # 93134 2/11/2001

Decentralize = Devolve, MI # 93155 2/11/2001

I think most of us in the MI board community lost track of what was going on with WER; how many screens have been developed, what the performance characteristics are, what the volatility and robustness look like. WER and TradeSim are almost completely overlooked. I think there are historical factors at play there: it wasn't right in front of the faces of the mainline MI community because it wasn't right here on this board, and its development occurred at or near a time when we were concerned about ValueLine, and WER seemed like another black box method to produce a stock list. Anyway: my impression from comments I've seen is that the WER picks are quite profitable, even more volatile than the VL picks: hold on to your hat. But there are backtests to look at. And there is a fabulous user's guide, written by Alan Flum:

MrToast walks us thru his thinking on choosing from the WER screens:

Wow on the WER, MI # 185597 3/19/2006
go here:
click on “Search Screens” and get here:
set a “Field” to “Stocks” and its
“Value” to “5” (or 10).
I set another “Field” to “HoldingPd”
and its value to “0” (which means monthly holds).
I set “Maximum Number of Filter/Sort Criteria**:” to 2 or 3.
I sort by “CAGR”; or “Sharpe Ratio”.
I hit the “Submit” button.
This gives me a list of the best-performing screens. I check out the ones without UVs (user variables). I see if they hold up with different start weeks.

And ProbTrader goes into more detail:

Subject: WER screens menu, MI # 185605 3/19/2006

What is BI?

Benchmark Investing. Value/Contrarian method of screening stocks. Toward the end of this intro there's a “Book Club” (under the “Further” section): Trouncing the Dow is mentioned there. That's the book where this method of screening is elucidated. There's a board here devoted to the method, and the author posts regularly to that board.

Benchmark Investing:

What is BMW?

A high-end German car: also an acronym for a user at TMF, BuildEmWell, who developed a Value/Contrarian stock screening method which is named after him. The method is basically Technical Analysis, buying on the dips: but this isn't cup-and-handle charting and these aren't short-term price movements. What distinguishes this TA method is its use of very long-term price history, one or two or even three decades worth; also its use of mathematical techniques to rigorously set signals. They use regression analysis (RRS!) to determine a long-term trend, and they use standard deviations to determine when a dip has actually occurred. This makes it “objective” and “repeatable” in the MI sense, which is great.</p

There are two weaknesses with the method from the MI sense, one of them fairly minor and one potentially very serious. The minor one is, they haven't (as of this writing) done a lot of work to optimize their sell criteria. They've done some (return to the CAGR line), and it looks pretty good: it would be nice if they could do more. The potentially serious issue is, all of the backtesting done to date has involved serious Survivorship Bias. Since the method is to buy stocks that have recently plummeted, if the only stocks that remain in your backtest data are the survivors then all of the worst falling knives have been eliminated. So the batting average of the method is artificially inflated. It's almost impossible to overstate the seriousness of this issue.

Two things mitigate this risk a little, at least to my eyes. One, the method is formalized off of the work done by one investor over the past dozen years or so, so there is real-world experience backing it up. If this were a newly-developed screen I'd be shouting from the treetops about it being an invalid backtest: but as an algorithm for doing what a real investor has actually been doing, and succeeding with, I'm less concerned about it. Two, the BMW traders try to do a certain amount of due diligence on each stock in the “pool” that they're monitoring, so that they have some assessment of whether the company is really the “same” company that generated the price history. This of course moves them out of the territory of MI and into the area of subjectivity; but we occasionally skip a stock here if there are questions or irregularities around it. The BMW method also restricts itself to stocks with “long enough” price histories; not less than 15 years and in some cases requiring 30 years. This helps ensure that the companies are more stable and move visible; though of course an Enron could certainly happen.

Anyway, it's an interesting method, worth looking into.


There's like a zillion stock screens! Where did they all COME from?

We made 'em up. We've been talking about mechanical stock screening on this board for a long time – since 1997, or maybe earlier if you go back to when TMF was on the AOL message boards: an eternity in internet time – and we've invented a very large number of screens. I wanted to find a place to stick in an overview of the history of this board, and this seemed like a good one. So here we go:

This board migrated from AOL in 1997, centered around the “Unemotional Growth” screen popularized by Robert Sheard. (UG took the 100 stocks ranked Timeliness 1 by ValueLine, and sorted descending by IDB's Earnings Growth ranking. Volatile as hell, this is not a screen we trade anymore.) Many of the early posts here are concerned with replicating approaches from O'Shaughnessy books What Works on Wall Street and Invest Like the Best. There wasn't a whole lot of data lying around, so people here generally relied on “manual” backtests: results that people reported from their own historical trades with a given strategy.

Sparfarkle arrived on the board in the Summer of 1998, with a strategy that he said beat UG: this screen was eventually named Spark. More important than the screen, Sparfarkle told the board that he had been able to wangle historical data out of Valueline, going back to 1984. This meant that historical backtesting was possible, and really opened the way for the board as we know it today. Due to efforts that can be accurately described as Herculean, on the part of a few people, posters on this board acquired data on the Valueline T1 stocks going back to 1986. This is two types of data: the information needed to generate picks, free of Survivorship Bias; and subsequent performance data to see how those picks did. I do not know who all was involved in this, beyond Jamie and Sux2BeU: but that work is the bedrock of what we do here.

RS Backtest Update: Complete, MI # 74150 7/11/2000

Backtesters have more months, MI # 74704 7/19/2000

Latest backtester edition, MI # 84654 11/2/2000

For those of you who didn't make it to MICon, MI # 97656 4/1/2001

That last post introduced Jamie's ScreenBuilder. With that, you could take any set of parameters delivered with the Valueline data since 1986, concoct your own idea for a screen, let it run and see how it did. Screen development really exploded from there. But I'm skipping ahead a little.

By August 1998 the known screens included:

Formula 90
“Investing for Growth – Classic”
“Investing for Growth – Relative Strength”
Low Price/Sales
Unemotional Growth

Note that these are all growth/momentum screens, except for Low Price/Sales which while it sounds like a Value/Contrarian screen is really a GARP screen. (It filters for companies with a P/S ratio <= 1.5, but its other components are all growth/momentum; VL T1, EPS Growth, RS26.) The mechanical value stuff was pretty much confined to Benchmark Investing and the Foolish Four, which were off-board: they had their own boards devoted to them. We were just starting to develop our repertoire of techniques: and what we were coming up with was predominantly momentum. This is symptomatic of the times, of course. Read some of the posts from back then: the returns people were getting in backtest results were just silly. Unbelievable. This post gives some of the flavor of the times:

Just kill the canary now, MI # 45890 11/17/1999

I don't want to give the impression it was ALL irrational exuberance, all the time. There definitely were voices of wisdom, advocating a balanced portfolio, careful attention to GSD, diversification among styles, etc. Those posters who paid due attention to that advice probably did ok for themselves in the awful correction that began in 2000. But those CAGRs! Heady stuff: impossible to ignore. Perennial self-appointed gadfly Vizcacha quipped that “MI” stood for “Momentum Investing”. (He still posts, under the screen name cynicinvestor.) And since the mechanical value stuff was off on other boards, he wasn't far off: MI wasn't by definition Momentum Investing, but maybe de facto it was.

As 1999 ended and 2000 began, the board added some mechanical value to its repertoire. Net-net, Div Yield and LowPB (aka Value_A) were all born right around then:

Cigar Butts: A Mechanical Value Screen, 12/5/1999

BV per Share Price Screen, 12/29/1999

A Retirement Screen?, MI #53254 1/5/2000

This was definitely the right idea at the right time: but it's unclear how many people actually adopted the new stuff. As noted elsewhere, true mechanical value is hard; I think our value techniques have never truly caught up to our momentum techniques. Many MI'ers were invested pretty heavily in momentum in 2000, and rode the tech implosion down. Another thing that happened around that same time, is that DrBob started developing stock screens using another data provider (not ValueLine):

A simple earnings surprise screen, MI # 48165 12/4/1999
Market Guide supplies weekly stock data disks back to 1996… Using the limited data I have on hand (the disks are expensive, $100 a pop) I looked at the following simple screen:

Value screen that doesn't use T=1&2, MI # 54435 1/11/2000

This started out very small, and re-reading those threads it's striking how deferential DrBob was, to people telling him that the parameters he'd found didn't seem to work with the longer backtests available in the ValueLine universe. What we understand now is that DrBob was using a different dataset from a different provider, with its own pluses and minuses. The data DrBob was using is what we now think of as the Stock Investor Pro data, and (thanks mainly to him) we know have a whole set of screens developed for that universe. On the plus side, many many more stocks are covered by SIP than by ValueLine: so there are some opportunities there, and you can also avoid any possible swamping by a bunch of MI'ers all jumping into the same pool. On the minus side, the data is available only back to 1997, which is a huge deficit compared to the available ValueLine data; and you don't get the VL Timeliness ranking, on which so many of our VL screens depend. We have experimented with various different “crap filters” for the SIP universe, to varying degrees of success: we don't have one single standard that works across screens. Keelix developed a backtester that runs on the SIP universe: an amazing accomplishment. Between Jamie and WER and Keelix, we have a truly great array of backtesting tools available to us, on which to develop a nicely diverse set of strategies.

As you know, the market went to hell from mid-2000 to late 2002. Deep into the bear market, we got creative and added a new class of screens based on institutional/insider ownership.

% Institutional Holdings, MI # 99240 4/17/2001

TR4W with %IH, MI # 120365 3/15/2002

Shrinkage, MI # 141725 2/22/2003

PIH_CSO, MI # 150142 9/23/2003

And finally, some 50+ years after Graham wrote The
Intelligent Investor, we discovered dividends:

OPTIMAN Dividend Growth Screen, MI # 95695 3/7/2001

YieldYear, MI # 137927 1/3/2003

YieldEarnYear, MI # 145752 6/8/2003

That pretty much brings us up to date on the various screen families. Note as soon as we had a couple screens in each family, people would go off and create variations, so that three defined screens become thirteen screen variants. Some of these we track in the weekly posting; many people have their own pet variants that they use, just slightly different from the posted screen.

The other major theme in any history of MI, has been the examination of market timing systems. This heats up from time to time, basically after corrections (especially the tech crash). A market timing system can be something you use to get in and out of the market entirely; or to tell you to go long or go short; or it can be something you use for your asset allocation, so that sometimes you're 100% stocks (long) and sometimes you have a greater proportion in cash, to mitigate risk. Timing didn't seem all that important when the market was going up and up and up in the 90s; it loomed more significantly when some people gave all their money back in 2000-2002. More on timing below.

Also last year a couple MI'ers started looking at mechanical systems for trading futures. For purposes of an intro / FAQ, the two things I want to say about that are (A) it absolutely is “mechanical investing”, about as purely mechanical as you can get; and (B) it is not a fit topic for something directed at newbies. For MI on futures, you're on your own.

So that's where the set of screens came from. For any specific screen, I would recommend using the Datahelper search engine to find the earliest references to it: read that thread to see the discussions around the screen.

How on earth do I choose among all these screens???

This is like the central question of MI. How do you know which screens are worth investing in? I see three main dimensions along which to
evaluate stock screens.

1. Robustness of algorithm. Is the screen likely to be a datamining artifact? Or is there reason to suspect it might have a real, lasting advantage?
2. Performance characteristics. Basically return vs volatility. Is it too volatile to stomach?
3. Family or type, for blending purposes.

Robustness of algorithm
We look for mounds of toast, evidence of canaricide, degree of complication, over-reliance on a few outsized returns, etc: markers for a screen whose performance should be regarded with suspicion. Over-reliance on a few outsized returns: I always think of that as the “buy Microsoft at 5” test. If a screen shows a great backtest, but all its returns are due to picking one great stock early on and holding it for ages, then we would wonder whether that screen would be likely to show good performance going forward. We also look for some understandable explanation of how stocks which meet the screening parameters would outperform. Are the screen's market-beating backtest returns due solely to one year where it returned 500%, while the rest of the time it was a loser? That would be a bit of a red flag. Is the backtest too short to really take seriously? We're looking for some reason to judge the screen as reliable, or unreliable. But it is a fact that we have very few hard, objective guidelines for why one screen should be rejected while another screen is accepted. We rely a great deal on our experience and judgment, from looking at a lot of screens over a number of years and from using stock screens in the market.

It would be a fair criticism to point out that this remains largely a subjective evaluation. Many of us are engineers or software guys, so we have a sense of algorithm design (or we feel like we do, anyway ;-). Algorithm robustness is real: in the field of software design there are quantitative ways of assessing this: number of branching paths etc. I don't want to give the impression that we've applied any of those formal methods to screens: we have not, and I'm not sure it's possible in the field of stock investing But I am trying to make the case that the notion of “degree of complication” is a real thing, that algorithm “over-complexity” can be pointed out, even if we don't have easy ways of quantifying it.

Performance characteristics
CAGR, of course: we want the return to be high “enough”. But that's not usually an issue: the screens that get posted to the board all show backtested CAGRs that beat “the market”; indeed, that's part of our Multiple Hypothesis problem. GSD is a big factor: we want volatility we can live with. The Sharpe Ratio (and its reward/risk variants) is probably the most important single number to look at. Though as a footnote, a screen with mediocre numbers in one or more areas could be a valuable add to the repertoire if it had low (or negative!) correlation with other, better screens.

Family or type
This has to do with putting together a balanced portfolio. If you already have a couple of good screens picked out, from a couple different families, then you may be looking to round out a blend: so you'd restrict your final choices to screens that are good complements to the screens you've already chosen, if in the process you pass over other screens with better solo performance numbers. More on this process later, when we talk about blending.

I don't want to downplay the importance of this topic, by reducing it to three bullet points. The topic of screen selection is so central to what MI is, that really it underlies almost all of our conversations, in one way or another. You could interpret this whole board as one long discussion about which screens to choose. And to the extent that it involves predicting the future, we bring to bear all the art, hunch, judgment and intuition we can on it.

Using Outliers to detect Datamining, MI # 100276 5/1/2001
William Eckhardt, who is also a very good investor with a strong statistics background, argues that intuition and synthesis are big parts of evaluating whether overfitting is present:
“The best way is to look at hundreds of examples. Add degrees of freedom and see how much you can get out of them. Add bogus ones and see what you can get. Try systems that make sense to you and ones that don't. Try systems that have very few parameters and ones that are profligate with them. After a while you develop an intuition about the trade-off between degrees of freedom and the reliability of past performance as an indicator of future performance.”

Question about Sharpe ratios, MI # 109623 10/22/2001
First, I wouldn't base the decision on only the Sharpe ratios or the CAGRs. I would also include subjective factors in my decision. These would primarily relate to my confidence in the repeatability of the returns and volatilities, and would start with my assessment of their “dataminedness” (assuming these were MI screens). Issues included here might be things like: complexity of variables and steps; the intuitiveness of the variables used; the landscape of the backtests (was only once-a-month data used, or are results for all start days?; how sensitive to parameter changes are the results (mounds of toast)?, etc.). The MI board has listed lots of other screen aspects to include in this subjective decision process.
Second, after filtering out any screens due to the subjective criteria, the level of risk I seek would drive the decision. And even if I share my desired level of risk with you, that doesn't point you in a direction appropriate for your own level of risk aversion.

A Taxonomy of MI Screens, MI # 114592 1/6/2002
there are three evolutionary steps to how you form a sensible investment strategy based on Value Line MI:
1: Identify superior screens, across a sytle-diversified spectrum;
2: Cross-correlate the most elite style-diversified screens in a portfolio that does not lean precariously in any stylistic
direction, where “elite” looks first to the Sharpe Ratio;
3: Employ a money management strategy that disallows constant compounding in any one part of the portfolio.

What Constitutes a Good Screen? MI # 124004 4/29/2002

Newbie Q?: Is This Screen Any Good? MI # 144018 4/30/2003

Workshop Articles By Subject - Part 2 MI # 144021 4/30/2003
Guidelines to Good Screens: Which One To Use?

Screen depth when optimizing blends, MI # 185239 3/12/2006
You want a robust screen, because a truly robust screen's good returns are NOT a fluke - they'll repeat. We don't have a way to absolutely identify a truly robust screen (wish we did), but we have a few ways to test a screen and decide that it probably isn't robust. One of these is depth: not only do the top 5 picks have a nice set of statistics (CAGR and GSD), but the next 5 are nearly as good, and so on for maybe 3 more iterations (total about 25 stocks). If a screen fails this test, it's a sign that the screen may be rather less trustworthy than we would like. Suppose that the top 5 stocks do okay on average, but the next 5 stocks routinely tank. How much of a change in market conditions would it take so that only the top 4 stocks do okay, and the #5 stock tanks? Probably not much. But if a screen passes this test (and passes some other tests as well), then it looks like a good 5-stock screen. And it's reasonable to use it as a 5-stock screen, in a blend or otherwise. On the other hand, it may or may not be a good 10-stock screen. We didn't examine it for that. Without such an examination we shouldn't use it as a 10-stock screen.

Chasing that Last Stubborn Stock, MI # 188172 5/11/2006
One way I have compensated for this is to judge screens by their "worst 4 of 5 stocks." This shows how screens do without their best performer each month. This is one of my robustness tests. If you have a screen that does well without its best performer…

Zeelotes has done a lot of work recently on evaluating screens, coming up with a list of monthly screens which he refers to as the “gold” screens. Here are links to several posts:

Which measure will predict the best screen? MI # 175260 9/18/2005

Pre-Bear Screens -- Which are Olympic Winners? MI # 175447 9/23/2005

TD Results for the Top Gold Screens in 2005, MI # 176007 10/3/2005</p

MI Investors: Favorite Screens Examined, MI # 176415 10/13/12005

Finding the Optimal Screen Blend for Ranks 1-5, MI # 176664 10/19/2005

SIPRO vs Value Line: Limited Ranking, MI # 177266 10/30/2005

The Best Predictor of Future Return, MI # 178461 11/25/2005

Small Port: Quarterly Rebalance, MI # 184345 2/23/2006

Those can get you started: but as noted, pretty much the whole board is a protracted discussion on screen selection.

So do you really just follow the screen?

Yep, pretty much.

In its purest form, a mechanical investor using stock screens might not even know the name of the companies he just bought! Just the ticker symbols. If it's your trading month, you generate the new list of symbols, sell the last bunch were holding, and buy the new bunch.

And this is about where your careful research-driven investor chokes.

If your screens give you a basket of stocks that as a group are likely to outperform over the holding period, then you generally want to hold the whole basket. Unless you can somehow tell ahead of time which stocks in the basket are going to do better than the others. We don't generally tend to know anything about individual stocks here: “we pick strategies and procedures, not stocks.” If we want to capture the “actuarial” advantage that (we hope) the screens can give us, the advantage shown in the backtest results, then we want to stick with the screen output as closely as possible: take the good and the bad, because we're not sure ahead of time which is which. Theoretically the Mechanical Investor doesn't second-guess the system, or use his own discretion when the system gives an order. The investor has exercised a ton of judgement and forethought in setting up the system: but in regular execution, the investor just follows the plan, mechanically.

But of course there are valid reasons for skipping a particular stock that pops up on a screen. A stock shows up on a value screen, the same week you see in the news that the company has filed for bankruptcy; or a stock shows up on a momentum screen, the same week that a merger/acquisition offer is made at a certain price, and the stock zooms right to that price. There are valid reasons for skipping a stock on a screen.

The advice here is to determine ahead of time what the conditions are under which you will deviate from the screen. For example, if you decide that inverted yield curves are a signal you want to incorporate; or rumors of financial wrongdoing on the part of the company, like mis-stating earnings or something; or the government announces it's going to Federalize all your energy producers – note it down. Make a list of the kinds of things that are valid exceptions to the screen. That way they are officially incorporated into your method. Then when you are tempted to deviate, you can see whether the current conditions are the kind of thing you had in mind as valid exception conditions, or whether they represent something you need to add to the list – or whether you are just getting spooked by volatility. If you encounter a situation you did not pre-plan for, and it represents a “valid” reason for deviating from the screen, make a note of it: now it gets officially incorporated into your method.

But yeah, in general you just follow the screen. You wouldn't believe how hard that can be, sometimes.

What is a short screen?

Well, a regular screen is for buying stocks that are expected to go UP: for taking a “long” position in the market. A short screen is the opposite. It is supposed to identify stocks that are going to go DOWN. These are stocks that you would either sell short, or on which you would buy put options. We generally don't do sell-to-open positions here: the put is probably safer, though it perhaps has less profit potential.

Some examples of short screens include SHORT_ALTMAN_Z and Negative_FCF.

What is “HTD”?

“Hold Til Drop”. It's a method of reducing transaction costs and turnover with monthly screens. The essential insight is that with monthly screens there is a tendency to get whipsawed. For example, the stock NVR appeared on some of our PEG screens in January 2001. It fell out of the top 5 in March, but was back in April. It fell out of the top 5 again in May; but the following May it was back. If you had followed the screen every month, then you'd have bought it, sold it, rebought it, resold it, rebought it again, resold it again… You'd actually have done pretty decently for yourself, picking up 25% here, 8% there, 33%, etc. But you'd have spent more on transaction fees than was strictly necessary: and if you'd bought it in Jan 2001 and just held the stupid thing for two years, you'd have nearly tripled your investment: something like a 70% CAGR. (As long as we're kicking ourself for the road not taken: if you'd held the thing til July 2005, you'd have nearly nine-tupled your investment, for a 4-1/2 year CAGR in the neighborhood of 60%).

Now suppose that when NVR fell out of the top 5 on this screen, it didn't fall completely off the screen but just fell to a lower position: say it was still in the top 10, or top 15, or top 25. Then if you were running this screen as a HTD 25, you would have bought NVR when it appeared in the top 5 of the screen in Jan 2001, but when you rebalance each month you would only sell it if it fell out of the top 25. I present this as a hypothetical because NVR did in fact fall completely off this screen for periods of time: but you get the point. With a HTD x, you only sell stocks

So: HTD sounds like a nice idea. Reduce transaction costs and turnover: hold on longer to stocks that are making long sustained moves. But does it help? Does it increase returns? For that we have to turn to the backtest results. Jamie Gritton's backtest site has a “Hold Til Drop” backtester, where you can see if HTD is a good idea or not. I believe you will find that some screens benefit from adding a HTD criterion, and others do not. As an example of one that might, check out RS26 1-5 from 1969 to 2005, both straight and with a HTD 25 component. If you can cut the number of trades in half and increase the CAGR & Sharpe, then you've really got something.

More on Hold Til Drop:

Momentum Screens and Non Standard Holding Period, MI # 52748 1/2/2000

Hold-Till-Drop: Insights and Corrections, MI # 55087 1/15/2000

What is “SOS”?


Not really. As used here, it's an acronym for “Screen of Screens”. If a stock appears on one screen, it might be good: if it appears on several screens, it's probably better. Right? To set up an SOS you'd take some number of base screens, let's say 5: and let's say you're looking at those screens 5 deep. Each screen is “voting” for a stock, in exactly the way that college football or college basketball teams are ranked during the season by voters: the #1 stock gets a certain number of points, the #2 stock gets a smaller number of points, etc. Tally up the points, and that gives you a ranking for all the stocks on that set of screens.

LAPropDoc illustrates it well here:

TTM, YTD, RS-O, PEG-O? FW # 12204 4/10/2000</p

Difference between SOSs & Blends, DG # 16 6/12/2001

Some well-known SOS'es over the years have included “SOS Elan”, which appeared in numerous annual configurations until he moved to a different method of blending a couple years ago, and “SOS Ancer”, for anti-ulcer.

What is “Dozens”?

Term coined by Robert Sheard back in the day, to describe a tactic of spreading out screen purchases across an intended multi-month holding period. To me this is one of those elegant and powerful ideas, that seem simple and utterly obvious after someone has articulated it. I'd like to give Sheard full credit: I also think there's great value in seeing the historical context of ideas. So rather than just describing it in a short & simple manner, as might be consistent with a concise FAQ approach, instead here's a series of links:

Robert Sheard, The Daily Workshop Report 2/25/1998
I've begun following two new simple portfolios: the Keystone Dozen and the Dow Dozen. In both cases, the portfolios select the highest-ranked stock each month that's not already in the portfolio and add a new position. Then as each position becomes a year old, it's replaced with the latest top stock.

Robert Sheard, Foolish Four Report 3/24/1998
Investors convinced that regular savings and investment in top-flight companies are the two prongs of a life-long strategy for building tremendous wealth often ask for advice on the best ways to add money to the market regularly without getting nailed on commissions. If we assume a deep-discount commission rate of somewhere in the $7-to-$10-per-trade range, it makes sense to keep all of our positions at a minimum of $1,000. (Expect two trades per stock each year, one to sell it and one to replace it.) That minimum investment keeps us at or below 2% of our original investment each year, a goal worth keeping in mind. And there are a number of possible methods for adding new money. The one I've become enamored of recently is the use of a “Dozens” portfolio. This is a strategy where one buys (or updates) one stock each month, holding a total of a dozen great positions. By staggering the purchase points across the twelve months, and then holding each position a year (or, if it's a taxable account, a year and day), you give yourself twelve opportunities to add new money to your portfolio each year without running up extra trades. (You're updating a position each month already; just add the new money when you buy your monthly replacement stock.)

Robert Sheard, Foolish Four Report 3/25/1998
A number of readers asked me to write a follow-up column today about the process one would use a year later when it becomes necessary to start replacing the oldest stocks. … the rebalancing dilemma is a little bit more difficult. You'll either have to accept some slight imbalances when positions are over- or under-valued relative to the whole portfolio, or you'll have to keep a small cash slush-fund to add to or draw on when a stock is out of line with its ideal value. Don't get too wrapped up in getting every position perfectly in line, however. The real goal is to be as fully invested as possible at all times in a dozen top appreciation prospects. If you'll do that as simply as possible and add new money at every opportunity, your long-term success will be…

Robert Sheard, The Daily Workshop Report 4/6/1998
For those of you unclear on the whole Dozens concept, it's a way to build a portfolio of 12 stocks using whatever screen, or combination of screens, you like the best. Each month, you buy (or update) one stock, holding each one for a full year (a year and a day in taxable accounts). The theory is that by spreading out your purchases, you're pulling stocks from higher in the rankings (because the rankings change throughout the year), thus boosting your long-term returns over buying the entire group of stocks at one time. In addition, you're getting the diversity of a 12-stock portfolio over four or five stocks and trading costs and taxes are still under control. And if you frequently add new money to your portfolio from regular savings, this gives you twelve opportunities each year to do so without increasing costs or wondering where to invest subsequent deposits.

Robert Sheard, Foolish Four Report 4/20/1998
One of my favorite saving/investment techniques is what we've begun to call the Dozens method here in the Fool. Many investors are just starting out and don't have enough saved yet to fund a full portfolio, but for a variety of reasons still want to begin with individual stocks. Alternately, an investor may have enough to invest in 10 or more stocks already, but adds new money every month and wants a method to do so without getting chewed up in trading costs. (Where do you put that extra $250 this month?) The Dozens method has the investor buy one stock every month, choosing the top-ranked stock not already in the portfolio. What rankings you use are up to you, but the idea is that you're always buying the highest-ranking stock that you don't already own and, by the end of one year, you'll have a full portfolio of twelve top stocks. Then, as each stock grows to be a year old (a year and a day if your account is taxable and you're in a high tax bracket), you will re-evaluate it and replace it if necessary. … Think of the calendar as a dance card, with one dance open per month. Eventually you'll have all twelve dances filled in, and each year the names on the dances simply change as you pick the highest-ranked stocks each time around.

Robert Sheard, The Daily Workshop Report 5/29/1099
Today completes our fifth month with the Dozens portfolios, so at today's closing bell, we'll be adding a hypothetical $1,000 to each of our half-dozen Dozens groups and will add stock #6 to each portfolio.

Robert Sheard, The Dozens at Mid-Year 6/30/1998
Today closes out the first half for 1998, so let's recap the start for our Dozens models and look toward our seventh purchase for each screen.

Robert Sheard, Two Doubling Dozens 7/23/1998
With the market hiccuping this week, let's see how the Dozens Portfolios are holding up. … Two of the six portfolios are still doing extremely well, with annualized returns on pace to record doubles in 1998. … A third strong performer, the Formula90 Dozen, is also doing well… The remaining three portfolios are all lagging the Spyder benchmark, and in fact, one is in the red.

Robert Sheard, Funding the Dozens 7/31/1998
I've discussed the Dozens strategies, both here and in the Workshop area. Usually, those scenarios are based on the idea of starting slowly and adding positions throughout the first year as one saves more money. But what about a case where you have enough money to establish the twelve positions already (at least $12,000)? How do you set up the rotation then? The first year, as you make the transition to the twelve monthly positions, is going to be a little awkward, and you'll have a couple of short-term trades in all likelihood, but this doesn't have to be a huge burden. Let's look at one way to set up the portfolio. In month one, rather than leaving 11/12 of your money in cash or the Standard & Poor's 500 Index for each month's purchase…

Robert Sheard, A Dozens Wrap-Up 8/27/1998
Since I'm closing out my tenure with the Motley Fool tomorrow, let me use today's column to review the Dozens models one last time.

So you can see that the name came from staggering purchases of annual screens across the 12 months of the year. But the concept readily stretches to other holding periods. Someone might say they're running a screen as “top 6 quarterly dozens”: that would mean they're buying the top 2 stocks in January, then in February buying the top 2 they don't already hold, then in March the next 2; then each month selling their oldest 2 picks and buying the highest 2 stocks on the screen that they don't already hold. “Top not held”, we say. Other terms used to refer to this are time diversification and time averaging.
(Fireballs prefers that name “time averaging” for it. ;-)

Thinking of this as “time diversification” involves the observation that there can be great variance in screen returns based on what date you start. The stock market is all over the place: some weeks up, some weeks down, some weeks sideways. Two investors can run the very same screens, make the same picks and everything, and experience very different returns just due to the luck of when in the week (or the day!) they make their trades. Think of a quarterly screen. This has a January cycle (Jan-Apr-Jul-Oct), a February cycle (Feb-May-Aug-Nov) and a March cycle (Mar-Jun-Sept-Dec). If you backtest a quarterly screen, and compare each of these cycles, then by chance one of them is going to have the best return and one of them the worst. I don't think there's any reason to think that that relationship will hold up: the February cycle (say) will not necessarily be better than (say) the March cycle going forward. But if possible you'd like not to be locked-in to the underperforming cycle, if there is one. A Dozens approach can address that: you're likely to get the average result across the start-month-cycles. This can apply to monthly trading also. Which week is better, for trading a monthly screen: week 1, week 2, week 3, week 4? Which week is going to be better over the next few years? You don't know: but you can sidestep the question by using “weekly dozens” with a monthly screen: say top 4 monthly, but buying 1 stock a week, then each week sell the oldest and buy the top-not-held off the screen.

BarryDTO, Time Diversification - My Logic for It 12/6/2000

“Time diversification” can be important. But I think an even more important benefit for the various Dozens approaches is, they can be a great way to manage the investor. Suppose you have an itchy trigger finger, and want to be trading very actively, but your account size is too small to safely support monthly trading. With a Dozens approach, you can trade monthly, even though your holding period is longer and you're incurring only a fraction of the transaction fees a monthly screen would generate. You get the thrill of monthly trading, without the same costs and possibly without the same volatility. Psychological issues are so important in this kind of investing; I think anything that helps, anything that makes it easier to stick with it, might be worth adopting even if there were shown to be a small degradation in returns. It's much better to stick with a system that'll give you 85% of a possible (high) return, if you can stay with it over a long period, than to try a high-return high-difficulty system and then give up and let it slide after just a few months. If a Dozens approach helps keep the investor engaged and active and interested, or if conversely it helps the investor stay calm and controlled by giving him something to do (something systematic and backtested) when he's just itching to do something, then that's great.

In Sheard's original articles, he touched on one aspect of using Dozens that can get complicated. In real life, the stocks you own across the various months are not going to grow uniformly. One of them will be up 20%, one of them down 7%, etc. So at rebalance time, the stock you sell may give you more than 1/12th of your account size; or your sale may give you noticeably less than the amount required to make an even 1/12th purchase. If it's a taxable account approved for margin, then you could just not worry about it: one month you might dip a little bit into margin to make your purchase, two months later you pay it off. But in a cash account (and all IRAs are cash accounts), these problems can be thorny to plan for.

One approach is to designate an “extra” position, and hold that amount in cash. So if you're doing a classical 12-stock annual Dozens, then the amount you buy with each month is 1/13th of your account size. Or if you're trading 24 stocks, each position would be 1/25th of your account size. This gives you a slush fund for handling commissions etc; also can help make up an occasional shortfall. I actually do this with an account where I'm trading 24 stocks quarterly/Dozens: each buy is about 1/25th of the account. But note that means only 4% or so of the account is in cash; you're likely to have around that much in cash anyway, no matter what you do, just because of how the prices of the different stocks multiply out at purchase time: you'll never be able to establish perfectly even positions. Also, with that many stocks rebalancing each month, it's unlikely the whole month will have a big shortfall for me to make up. So setting aside an extra position seems ok, in my situation. But if you're only trading 12 stocks, I have my doubts about setting aside a whole position in cash, for fees and such. That's 7.7% of your account. Seems high. One thing you might do, is set aside half a position. So each month, determine your purchase amount by dividing your account size by 12.5. If you make 12 purchases, that leaves 4% in cash for fees and such: precisely the amount I'm using.

That still leaves you with the problem of what to do if, at rebalance time, the stock you sell leaves you with cash that is significantly more than, or less than, the fraction of your account that you're supposed to buy with. I have a novel suggestion for that problem: don't worry about it.

• If the stock you sell gives you “significantly” more cash than you're supposed to use this month: well you just sold a winner! Congratulations! That's not a “problem”, that's the result you're trying to achieve! And the situation will take care of itself: over the coming months, just keep on buying with 1/12.5, and the extra cash will get spread out among your other stocks picks. That's what you want.

• If the stock you sell gives you less cash than you're supposed to use this month: well, you just unloaded a loser. Congratulations! You have other stocks that are doing better, and those winners are running a little longer, and that's fine. The only real issue you have is that this position you're about to establish is underfunded, relative to your portfolio. If that really bothers you, then in a couple months when you sell a larger position, you might want to circle back and add to this position. (But be aware of possible tax implications, as this new money in that stock will be held for less than 12 months.) Also, if this is an annual hold, then you will probably know a couple months ahead of time that one of your positions is a loser. So you might want to slightly underfund the couple picks prior to that one, like with say 1/13th or 1/14th of your account, so you have a few extra dollars to add to the weak month at rebalance date, thus making each pick “close” to the correct amount, rather than having one stand out like a sore thumb.

Getting “close” is the best you're ever going to be able to do, because these amounts will fluctuate all the time. Remember, this method is supposed to make things easier for you, not harder: so don't obsess over small differences here and there. Get it right in broad strokes. As Sheard wrote (above): “Don't get too wrapped up in getting every position perfectly in line, however. The real goal is to be as fully invested as possible at all times in a dozen top appreciation prospects … do that as simply as possible.”

What if I have a large enough account to buy 12 stocks for an annual hold right now, but I want to do 12 stocks annual Dozens? How do I start, and transfer to a Dozens rotation?

Hell if I know, and I'm a little uncomfortable giving explicit instructions on what you should buy and sell. It'll be clunky; annual Dozens might not even be the right approach for you. As Baltassar noted back in 1998 :

Spark or EG Dozens, MI # 10810 12/2/1998
I view the Dozens approach, regardless of the underlying screen, as a solution to a specific (and rather nice) problem: you've got new money to invest every month, in sufficient quantity to justify the trading costs of [new] purchases. If this is not your problem, the Dozens method isn't your solution.

Here's one suggestion:

• Buy the top 6 now. Leave the rest in cash (or RSP, the Rydex
S&P 500 Equal Weight ETF; or IWM, the Russell 2000 Index ETF) for 6 months.
• In six months, buy the top 6 not held.
• Next year, each month sell 1 of your oldest picks, and buy 1 stock.
So that first group of 6 stocks would be broken up over the first six months of next year. One will have been held for 12 months, one for 13 months, one for 14 months, etc. The second group of 6 would be broken up over the second six months of next year.

There might be better ways to do this: that's just an idea off the top of my head.

What is a Bullet?

A 1-stock screen. If the top 5 from any particular screen show a better CAGR than the top 10, and if the top 3 show and even better CAGR than that, then why not just sink all your money into the #1 stock? I hope you notice that in the discussions of blending we were moving in exactly the opposite direction from that, toward holding more stocks and having less volatility in our returns. The notion of a Bullet is left over from the go-go raging bull 90s; we're smarter now. (Hopefully.)

I think it's possible to get confused between Dozens and Bullets. If you're running a screen as a 12-stock annual Dozens, doesn't that mean that each month you're buying the Bullet? Well, sort of, I guess: but the point is that's not all you're holding. With Dozens you're getting time diversification: you're holding several stocks in a screen, you're just not buying them all at once. With a Bullet, you get no diversification, because you never hold more than one stock from a screen.

What is blending?

The screen-based equivalent to diversification.

If you buy stocks from just one screen, you can theoretically get enough stocks (say 10 or 12) to diversify against “company risk”. But the stocks you hold will all be similar in some important way: obviously, because they'll all fit the screening criteria. If your screen is a momentum screen, then your stocks will all be high-flying momentum stocks; if it's a value screen, then your stocks will all be beaten-down contrarian stocks; etc.
This suggests that your stocks will tend to be affected similarly by changing market conditions. If that “style” of stock goes out of favor, your whole portfolio is hit the same way. You're actually more vulnerable than the number of stocks you're holding would suggest.

What we do instead is diversify by holding stocks from a number of different screens. So if you're holding 12 stocks, rather than 12 from 1 screen we would look for 4 each from 3 screens, or 3 each from 4 screens. This gives you “style”-based diversification: some of your stocks can be momentum stocks, some of them high-yield stocks, some of them value stocks, etc. This portfolio would tend to be less volatile in the face of changing market conditions, than a portfolio of 12 stocks which all come from the same screen.

Diversification the Workshop Way, 6/27/2000
So here's where we get back to diversification. One way to potentially reduce the volatility of a mechanical portfolio is by using several different screens. For example, let's say that Strategy A returns 40% in odd-numbered years, but returns -5% in even-numbered years. By contrast, Strategy B returns -10% in odd-numbered years, and 45% in even-numbered years. Obviously, investing in one of these strategies is a wild roller-coaster ride. However, if you were to split your funds between the two screens, your results would look something like this:

If there's one thing we talk about even more than screen construction, it's blending screens to create a balanced portfolio. “What blend do you use?” is the “What's your sign?” of the MI board. Blending is kind of amazing, because – well, this is where the correlation coefficient comes in. A few years ago I did a “thought experiment” on blending based on the correlation coefficient:

Uncorrelated investments/rebalancing, FW # 29538 1/2/2002
The LowPB and Mod1, 2 and 3 columns all contain the same values, yearly returns for LowPB: they are just rearranged to change the correlation coefficient with CAPRS. Mod1 is more strongly correlated to CAPRS than straight LowPB is; Mod2 has almost no correlation; Mod3 has a strong inverse correlation to CAPRS. …
The results are pretty clear: increasing the correlation of returns decreased CAGR and the risk/reward relationship from straight LowPB, even though the Mod1 returns are identical to the LowPB returns. Dropping the correlation to near zero improved both CAGR and the risk/reward measure. And here's the point I was wondering about: dropping the correlation even further, from near 0 to an inverse correlation, further improves both CAGR and the risk/reward measure - dramatically.
Bottom line: an inverse correlation is better than a zero correlation.

The important thing about that “experiment” is that it shows (to me: everyone else already knew it, probably, but I needed to see for myself) that if you have an investment and you're looking to blend it with something else, and you have a choice of three investments A&B&C, then you get an improved return and reward/risk ratio from blending your first investment with the most negatively-correlated of the other investments, even if the total returns for A&B&C are equal. The correlation alone will improve the return and the reward/risk ratio. Catch that: it's not just that the volatility is reduced, which would be a good result all by itself. The total return is increased also! Just from the correlation of the two investments! That's kind of amazing. Here's another example along the same lines:

Re: Interesting Metric, MI # 145798 6/9/2003
Blend YEY/CAPRS quarterly … Gotta love negative correlation.

That last post really illustrates the power of blending. You take two screens with a somewhat negative correlation, blend them in a portfolio, and wind up with a CAGR higher than either of them! With a manageable GSD, usually right in the middle of the GSDs of the component screens. It's really, really, really amazing. I'm not sure we've made a more powerful contribution to the understanding of practical investing – not that we invented diversification, because we certainly did not. But our work in creating blend after blend after blend, of high-flying momentum screens and solid value screens, curve-fitting just about every permutation you can think of, created a very graphic, easy-to-grasp illustration of what happens to CAGR and GSD and Sharpe as you blend investments. We demonstrated it over and over and over, to the point that looking for low-correlation alternative investments for blending is just automatic for us now. More immediate and practical and in-your-face than any academic study.

Here's the exact same lesson again – I just can't get enough of this:

Gentle Screamers questions, MI # 184319 2/22/2006
blend a momentum screen with a value- or dividend-based screen, so as to get a stabilizing blend effect. … use some blending, even if (especially if?) you're trading a small account. As an example of the power of blending, here are the component screens of the above blend, shown traded separately: …
Blending really is magic! Blend two good, uncorrelated investments and you can get a gross return higher than either! I mean, not always: but it is possible, and this is an example. … You absolutely need to get some of that magic working in your own portfolio of screens, even if you can only afford to hold a small number of stocks.

Note that the screen YieldEarnYear was used in each of the last two linked posts. It tends to show up when people discuss blending, not only because it has good risk/reward properties and a nice low GSD, but also because it shows fairly low correlation with most of the rest of our screens.

Note also that the last few paragraphs strongly imply that you want to find screens that are negatively correlated. In practice, that doesn't really happen. The market has some affect on all stocks: if the market is rising, then most stocks are going up; and if it's falling, then most stocks are too. I don't think we have any pairs of decently-performing long screens that really show negative correlation. Even YieldEarnYear, mentioned above, doesn't really tend to show negative correlation when you get down to month-by-month returns: I was using quarterly rebalancing in the links above.

(As a footnote, if you feel very strongly about a certain style of investing, let's say value investing, then you might prefer not to be diversified in this way. Remember that Warren Buffet advocates focus rather than diversification; and he makes the case that volatility is not the same thing as risk. So there are a number of ways to skin this cat. If you're going to be focused in only one style, then probably value is the safest one to use. We have some value screens here. Perhaps not enough: if you'd like to help us by developing some more, that would be fantastic. Note that the game Warren Buffett plays is a little different from the game we tend to play here, and the tool set he can bring to bear is a little different from the set most of us can bring to bear.)

How do you construct a blend of screens for a stock portfolio?

Now you're talking.

A little over 5 years ago, there was briefly an “official” MI portfolio, called the Workshop Portfolio. We went thru a long protracted discussion of how to construct the blend that would “officially” represent MI. The official port vanished soon after: but that discussion still represents the most thorough examination of the various considerations for putting together a blend. Here are the links: I think the third is the meatiest one.

Foolish Workshop Official Portfolio, FW # 19861 10/19/2000

FW Portfolio: Allocation, FW # 19863 10/19/2000

The Official Workshop Portfolio, FW # 21682 11/27/2000

It's interesting to note that probably none of the screens
chosen 5 years ago would make the cut if we went thru the same exercise now. But
that's still possibly the most educational examination of MI that we've done.

If we were to construct a blend today, to me there are two different “directions” to come at the problem from. Think of the difference between a “top down” and a “bottom up” approach. On the one hand you can look at the steps of different screens, classify the screens into families based on the filters, and then choose screens from each family; on the other hand you can look at the performance numbers of different screens, and construct a blend empirically.

For the first approach, classifying into families based on the screening steps, recall that we said there are five basic “types” of screens here:

• Value/Contrarian
• Income/Yield
• Institutional or Insider Ownership
• Growth at a reasonable price (GARP)
• Momentum or Relative Strength

So a first cut at a decent blend would be to take one screen from each category. If you want to hold 15 stocks then take the top 3 from each screen; if 20, then the top 4 from each screen; if 25, then the top 5 from each screen. However, you may not need to have all 5 families represented. The GARP screens tend to be fairly strongly correlated with the momentum screens, so you could pick one screen to represent both groups. And the “ownership” screens – I don't think we've done a whole lot of work with those yet, to the point where we've got a number of them and a handle on how they tend to behave as a group (if indeed they have any shared group behavior). Basically, you could go a long way toward constructing a good blend, by choosing a representative screen from each of the three groups Value/Contrarian, Income/Yield, and Momentum/RS.

For the second approach, using screen performance to group screens, Emintz did some phenomenal work using the correlation coefficient to divide screens into families, for blending:

Screen Families, MI # 181229 1/7/2006
There are clearly three major groupings. The High Yield value screens on the top, the other value screens on the bottom, and the growth/momentum screens in the middle. I've found that combining three 5-stock screens provides most of the improvement in Sharpe ratio, usually, adding more screens beyond that, while they may result in improvements, usually only result in minor increases.
I see three major families of screens: high-dividend at the top of the chart (14 screens), deep-value at the bottom of the chart (9 screens), and a very large blob of mostly momentum screens in the middle. Every screen in one family differs from any screen in any other family by 0.3 or more - and the biggest difference the chart shows is only 0.38.
LowPE_ZLTD and and H52EarnPS are both in the deep-value family.
Screamers and Beta are both in the momentum family, and therefore quite a bit different from the other two.
(I'll admit, it puzzles me that deep-value is more closely related to momentum than it is to high-dividend. But the software used forces a strictly binary branching, so some arbitrariness is to be expected.

Pick one good screen each from the high, middle and low sections of Emintz's chart, and you've gone a long way toward constructing a good blend.

You could spend the next month reading about different blending ideas here. In fact that's not a bad idea. Here are some places to start:

Another simple screen, MI # 58548 2/14/2000
Are we even looking for that? My impression is that we are looking for a screen (or set of screens) which exhibits returns that are considerably better than the indices, but are not closely correlated to our current batch of screens. Since most (all?) of our screens are “growth” screens, we would expect such screens to be “value” screens. But they don't have to be. … We're not looking for a “value” screen per se, we're just looking for a screen that does as well or better than the market overall, that is not strongly correlated with most of our current screens. This will allow us to do a “blend” that will (hopefully) have lower volatility (and downside risk) than any one screen.

Re: Brass In Pocket, 4/19/2000
On the RS/Momentum side of the stock market, the investor is practicing a radically different religion than what Buffett, Lynch and Graham concentrated on with their buy-and-hold, “wait for the weighing mechanism to kick in” thinking. There's absolutely nothing wrong with that thinking, and I am a big fan of value investing. But it's important to know what game you're playing before you enter the arena, and we're not playing the buy-and-hold game with RS-O monthly. … Capturing discounts to earnings is important in a portfolio, I believe, and I also believe RS is important. Bucause of the opposing ends of a continuum that RS and earnings tend to sit on, I think the mixture of the two concepts in a portfolio is a great way to protect risk-adjusted return.

PGY: Probably good year blend, MI # 165803
As with many investors, for me the most annoying thing about investing is the part where the prices go down.

30 vs. 3, MI # 175158 9/15/2005
What is the benefit of having 30 holdings in your portfolio vs., say, 3?

SiPro Index Fund, MI # 181425 1/9/2006
I wanted to see what would happen if you bought five stocks from 20 screens, thus holding 100 stocks

New Optimized Blends, MI # 182576 1/28/2006
In the past I optimized my blend to find the highest Sharpe ratio. This time I've added other measures - Sortino Ratio, Information Ratio, Upside Potential Ratio. Each optimization target produced different blends.

Bumpy Road, MI # 183831 2/15/2006
I counted the number of times that the current portfolio ($ amount) had declined from the portfolio six months earlier. The smaller the number, the “smoother” the climb upward of the portfolio, at least in my definition/perception. The number preceding the url is that count, starting from 1989. As you can see, most blends and SOS screens seem to outdo an individual screen.

Small Port: Quarterly Rebalance, MI # 184345 2/23/2006
I needed to put together a portfolio for a friend with a small account, and thought it may help out a few to see what tests I ran before making a recommendation. My stated goals were:
1.) Minimize trading costs to about …

FOLIOfn blend, MI # 184583 2/27/2006</BR>
The 2006 FOLIOfn real money port is a 12 screen blend (4 VL, 8 SI) held 3 deep.

Longer Rebalances: Blends, MI # 184855 3/4/2006
I went ahead and extended the previous research on Quarterly Rebalancing to include other longer rebalance periods. The primary purpose being to find a blend that would not be hurt by limiting the trades even further than just quarterly.

Blends list, MI # 184948 3/5/2006
I compiled a list of blends [mentioned in previous posts] and sorted it by Rebalance Period, then by Stocks Held.


What is an ETF?

Exchange Traded Fund: a basket of stocks like a mutual fund, but traded on an exchange so you can buy & sell it just like a stock. There's been a real proliferation of ETFs in recent years, and that's nice because it gives a trader the opportunity to invest in broad sectors or in overseas stocks, without drilling into details of specific companies and without paying high loads to mutual funds. We have some MI strategies that look for momentum in ETFs, and buy for a short (say 3 month) holding period; however, the really huge proliferation of ETFs is such a recent phenomenon that there's been no chance to really backtest those strategies: they should be considered as “in development”. There's a board at TMF:

More information on ETFs:

ETF basics
For your U.S. exposure, the Vanguard Total Stock Market VIPERs (VTI:AMEX - commentary - research - Cramer's Take) and the iShares S&P 500 Index (IVV:AMEX - commentary - research - Cramer's Take) are relatively cheap. The iShares Russell 3000 Index (IWV:AMEX - commentary - research - Cramer's Take) is also worth checking out, since its expense ratio is just 0.20%. On the international front, consider the iShares MSCI EAFE Index Fund (EFA:AMEX - commentary - research - Cramer's Take), suggests Culloton. This ETF mirrors the Morgan Stanley Capital International Europe, Australia and Far East Index (MSCI EAFE) index and is composed of about 1,000 companies that trade on 20 stock exchanges around the world (not including the USA, Canada, and Latin America). Can't get much more international than that. For emerging markets, there's the Vanguard Emerging Markets Stock VIPERs (VWO:AMEX - commentary - research - Cramer's Take), which is cheaper than its iShares counterpart.

Are some screens better for different parts of the
business cycle?


A related idea is ETF switching, usually on the basis of RS.

Using Relative Strength of Sector Funds, MI # 65135 4/3/2000<BR

Sector fund RS, MI # 96337 3/15/2001

ETFs - response to Buckaroo, MI # 166533 11/25/2004

Capitalizing on Sector Rotation, MI # 172847 6/6/2005

Dread's - Portfolio Performance, MI # 185770 3/24/2006

ETF - RRSXG Screen Ranks, MI # 32421 3/29/2006

Is there a January effect?


Is there seasonality in screens?


The Halloween Indicator, MI # 2408 4/28/1998

BACKPOSTS-Statistics & Seasonality Update, MI # 62559 3/16/2000
(scroll about 2/3 down to find the links on seasonality)

Timing Canaries in the Data Mine, MI # 83497 10/20/2000

Renting airline stocks, MI # 131314 8/8/2002

More on Seasonality by Hulbert, MI # 150971 10/15/2003

Last/First of Month and Other Effects, MI # 160831 4/25/2004

Halloween and the Dows, MI # 163794 7/30/2004

PGY: Probably good year blend, MI # 165803 10/19/2004

Seasonal Screen Allocation Periods, MI # 169401 2/8/2005

The Superior Eight: Seasonal Study, MI # 169898 2/19/2005

Going Fishing Seasonality, MI # 182396 1/25/2006

What is the Rotor3 approach?

A version of “What's Working Lately”, mentioned in passing post 165659, and in more detail here:

RRS question, 4/4/2004

What's working lately?

“What's Working Lately” is, along with “What's Working Now,” shorthand for a concept also known as Relative Strength of Screens. So the acronyms you'll see for this concept are WWL, WWN and RSS. The idea is, look at the recent performance among a group of screens, to decide which screens to invest in for the next few months. A related concept is mechanical screen switching. Here's a selection of links on mechanical screen switching:

MSS - Collection of Backposts, 6/28/2000

For RSS/WWL itself, here are some links:

Relative Strength of Screens, 10/27/1999

1969-1999 Limited Data Backtest Beats RSO, 4/18/2000
The market does alternate between periods in which value investing yields superior returns to growth investing, as well as the reverse. … Figuring all this out is most of the market's ongoing obsession. Doing so is very hard. Your approach would seem to be rather more straightforward: Let The Market Tell You! When the market as measured by an appropriate index does poorly for a year, take that as a signal that a switch is occurring from what has been in favor to what has been out of favor.

Screen selection, 5/19/2003

What's working now, 6/10/2003

WWL: What's Working Lately, 10/13/2004

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.
Community Home
Speak Your Mind, Start Your Blog, Rate Your Stocks

Community Team Fools - who are those TMF's?
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.