No. of Recommendations: 19
Picked up from this post
http://boards.fool.com/Message.asp?mid=26820148

This is the only long screen that I know of that relies totally on negative momentum.
So, even though the results aren't the most fantastic in the world,
I figure it ought to have a low correlation with almost all our other
screens, and good blending properties. It did very well compared
to the S&P in 2000-2002, and if nothing else it certainly won't pick
the same stocks as RS26 or Screamers.

Timelines 1-5 (just to avoid weird stuff like buyouts)
Safety=1 (because "safe" stocks bounce, on average)
Price / 52 week high bottom 20 (things that have fallen a long way)
Price / 52 week low bottom 5 (things still very near their bottoms)

backtest.org/Safety_Floor:8608SBT15XsftE1XpriDh52B20XpriDl52...
         Screen     S&P
CAGR 21.77 11.82
GSD 17.69 15.77
Sharpe 1.01 0.54

1986 28.0 23.9
1987 34.6 5.8
1988 39.1 10.4
1989 16.5 35.2
1990 7.1 -4.8
1991 32.0 30.8
1992 19.3 7.1
1993 11.8 10.2
1994 -1.2 1.5
1995 58.1 38.7
1996 34.6 23.2
1997 22.9 27.9
1998 45.8 34.5
1999 18.0 18.4
2000 56.1 -10.8
2001 19.5 -9.3
2002 -6.8 -22.1
2003 38.1 28.5
2004 29.4 10.2
2005 -2.9 7.5
2006 19.8 13.8
2007 -10.1 5.6
2008 -8.1 -11.9 (first half)


My speculation is that this should do well right after a market bottom.
Like, maybe, now?
These probably won't bounce as well as a growth stocks, but then this
won't usually have you in stocks that crash either. This past year
is probably the perfect torture test, when tons of "safe" stocks have
fallen without end---FRE anyone?---but that isn't usually the case.

Anyone think it might be a keeper?

Jim
Print the post Back To Top
No. of Recommendations: 1
Don't you think you're going against the flow for no reason.

The very simple

Timeliness™ 1-5; Stock Price / High Price 52-Week top 5
http://backtest.org/Simple_h52:8608SBT15XpriDh52T5

gives better results for the same period of time

Screen S&P

CAGR 27 11
GSD 23 16
Sharpe Ratio 1.06 0.48
Print the post Back To Top
No. of Recommendations: 0
Not bad at all, I think it might be hard to stick with it looking at the last 4 years however.

Terry
Print the post Back To Top
No. of Recommendations: 3
Don't you think you're going against the flow for no reason.

Against the flow, yes. But not for no reason.
(the reason may not be right, but there is a reason!)

The screen you mention will be very much like so many others, and
will pick the same old momentum stocks, generally at high valuations,
and with a big vulnerability to a drop in a bad storm. A blend of
such screens isn't really going to offer true diversification.

The one I suggest is chosen specifically to be contrarian: it will
tend to pick stocks which are very much out of fashion, a little more value
oriented, and often at pricse which are below each stock's typical
valuation. And of course, all the stocks have very strong financials,
or they wouldn't be rated safety=1. Note the perfomrance in October 1987:
-4.4% isn't so bad for that particular month. Out of fashion stocks
tend to react very little to bad surprises, and very strongly to
good surprises, so you get positive skew on your volatility.

So, my thinking is that if you can get even modestly market-beating returns
with below-market risk, it might make sense as part of a diversified portfolio.

Another reason I suggest this screen is that it might be quite timely in the next year. Maybe.

Jim
Print the post Back To Top
No. of Recommendations: 0
Nice find Jim.

Assuming i've entered it in correctly, this is the equivalent on the GTR1 backtester:

http://backtest.org/gtr1/tim.v:am5:sft.v:et1:ph253.g:2bn20:r...

For a 5 stock hold, i'm seeing a CAGR/GSD of 19.10/17.96. Slightly torpedoed, but not badly.
Print the post Back To Top
No. of Recommendations: 9
Assuming i've entered it in correctly, this is the equivalent on the GTR1 backtester:...

Looks right to me.

A more interesting view (to me) is how it blends, since that's a lot of the reason for suggesting it.

The following figures are from the monthly-start backtester (sorry).

Monthly volatility isn't the best metric of risk, especially for value
screens (ref skew comment), so as a risk metric I use downside deviation.
I use MAR=0 for rolling-3-month periods because I don't like losing capital,
and MAR=10% for rolling-12-month periods because I need to make money.
The average of the two is a handy single-number risk metric.

I'll take the H52 screen as a proxy for a typical positive-momentum screen.

On that basis,
the simple H52 screen (which I like incidentally) is CAGR 28.1, risk 6.60
the Safety_Floor screen is CAGR 20.6%, risk 5.10
a 50-50 blend of the two is CAGR 25.3%, risk 3.38
So, that's only about half the risk of the momentum screen alone,
while knocking CAGR down only 2.8 points.

The reason for the unusual risk metric is that these screens correlate
reasonably well at the month-to-month level, but are almost opposites
in many full year periods, so this metric captures the improved
smoothness.

For example, other rolling-year metrics show it different ways.
Probability of a positive rolling year:
H52 85.1%, Safety_Floor 87.6%, blend 95.6%
Worst rolling year
H52 -30.9%, Safety_Floor -24.2%, blend -13.7%
20th percentile rolling year
H52 6.0%, Safety_Floor 5.4%, blend 10.6%
Standard deviation of rolling year returns (simple, not geometric)
H52 27.2%, Safety_Floor 18.2%, blend 17.5%
Median year minus 2 standard deviations
H52 -24.7%, Safety_Floor -12.5%, blend -9.4%

But, for those who like the monthly stats, the following are through end-2007 only
GSDm
H52 23%, Safety_Floor 18%, blend 16%
...and Sharpe
H52 1.07, Safety_Floor 1.01, blend 1.29

Nothing earth shattering. The idea is simply to pick something
very uncorrelated to the rest of the things you're doing which still
has a positive expectation.

Jim
Print the post Back To Top
No. of Recommendations: 0
I looked at this and thought "it ought to do even better if you put in a check for stocks that have turned up".

It doesn't. In fact it significantly decreases return.

So I said to myself "Okay, so how about stocks that have NOT turned up?"

That's even worse.

I tried several variations of this: a final check to pick the 3 stocks with highest (lowest) TR4W, same thing only widening limits to 30/15/5 and 100/20/5, checking at the end for positive (negative) TR4W among the five stocks picked, checking up front for positive (negative) TR4W. They all pushed the Sharpe down under 0.7. In some cases quite a bit under.
Print the post Back To Top
No. of Recommendations: 3
They all pushed the Sharpe down under 0.7...

Yes, I had a hard time thinking of an extra step that would improve it.
Requiring positive current earnings? Worse.
Requiring negative or undefined current earnings? Worse.
Low PE? Worse.
Earnings growth five year positive? Worse.
Earnings growth five year negative? Worse.
Book value growth five year positive? Worse (but close).
Book value growth five year negative? Worse.
Still above the price a year ago? Worse.
Price below the price a year ago? Worse.
Even changing the trading-day lag didn't make a difference.

So, it's somewhere on that fuzzy line between optimal and brittle.
At least it's simple!

Jim
Print the post Back To Top
No. of Recommendations: 1
Requiring positive current earnings? Worse.
Requiring negative or undefined current earnings? Worse.
Low PE? Worse.
Earnings growth five year positive? Worse.
Earnings growth five year negative? Worse.
Book value growth five year positive? Worse (but close).
Book value growth five year negative? Worse.
Still above the price a year ago? Worse.
Price below the price a year ago? Worse.
Even changing the trading-day lag didn't make a difference.


In a general sense at what point do you declare a screen is walking the edge of a razor?
Print the post Back To Top
No. of Recommendations: 0
I use MAR=0 for rolling-3-month periods because I don't like losing capital,
and MAR=10% for rolling-12-month periods because I need to make money.
The average of the two is a handy single-number risk metric.

I'll take the H52 screen as a proxy for a typical positive-momentum screen.


What is MAR?
What is H52?

John
Print the post Back To Top
No. of Recommendations: 1
What is MAR?

MAR is short for "Minimum Acceptable Return", which is the key input
to calculating a risk metric called "Downside Deviation". This is not
the return you desire to hit (which would always be a really high number),
but the number below which your investment goal has failed.
If you're a pension fund trustee and have to hit 8%/year on average
in order to keep your sponsor from going bankrupt, your MAR is 8%.
The downside deviation is a single measure capturing the frequency and
magnitude of returns in a one year period which fell short of your goal,
with a squared penalty on bigger shortfalls. i.e., if your MAR is 8%,
then a 0% return is four times as bad as a 4% return, since the
shortfall is twice as big.

An excellent single measure of real world investment histories is
the Upside Potential Ratio (UPR), which is the Upside Potential (UP)
divided by the Downside Deviation.
http://sortino.com/htm/Upside%20Potential.htm
I add the phrase "real world" because this metric tends to give
pretty nonsensical results on our backtests, since the results of
MI backtests are substantially better than what one gets in real life.

What is H52?
Sorry for not being clear, this is just the Simple "52-week high" screen
that FrenchorFoe mentioned in the second post in this thread.
http://boards.fool.com/Message.asp?mid=26826373

Jim
Print the post Back To Top
No. of Recommendations: 0
In a general sense at what point do you declare a screen is walking the edge of a razor?

I think it is partly that, but partly that the screen simply has to
be redesigned if you add another filter. There aren't that many
"safety=1" stocks, so any other filter cuts that down a huge amount,
and the two cutoffs of 20 and 5 probably have to be changed at the very least.
A lot of screens aren't helped by more filters, since they rely on
getting the absolute extremes of the one thing they're looking for.
When you cut down the number of stocks among which it is comparing,
the benefit of the "extreme seeking" fades. So, this result of
nothing else helping is much more common than you might expect.
The screens that get posted here are usually played with a lot before
getting posted, and the screen designer often stops when he/she has
run out of things to add which help. As an excercise, try adding
an extra filter setp to the start of a few of the standard screens---
it isn't often you can find something which helps.

Jim
Print the post Back To Top