UnThreaded | Threaded | Whole Thread (8) | Ignore Thread Prev Thread | Next Thread
Author: Zeelotes Big red star, 1000 posts Feste Award Nominee! Old School Fool Add to my Favorite Fools Ignore this person (you won't see their posts anymore) Number: of 254198  
Subject: Annual Hold Strategy Date: 12/29/2007 4:28 AM
Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Recommendations: 19
Introduction

I've been debating whether to employ an annual hold strategy for a particular account. This would be based on holding a fixed # of stocks for exactly one year + one day for tax purposes. The difference between what I did this time and what I've done in the past is that I used the same approach as I've done for building blends employing a monthly rebalance, i.e., sort based on various measures to select the screens to use at the end of each year. Here are the two other posts detailing research I've done on longer than monthly rebalances.

Small Port: Quarterly Rebalance
http://boards.fool.com/Message.asp?mid=23744437

Longer Rebalances: Blends
http://boards.fool.com/Message.asp?mid=23787636

I began by running tests on all screens using various start years each holding four stocks -- 1969, 1986, 1987, 1989, 1997 and those in the SIPRO universe starting in 1997. Each year is based on a single start date so this is far from an exhaustive test, but I believe it is sufficient to guide me to the best solution.

I then set up a backtest where all the various measures were tried based on both an ascending and descending sort. I chose three screens from each universe -- three from VL and three from SIPRO -- resulting in a hold of six screens and 24 stocks. Because SIPRO was included I limited the test to 1999 to 11/7/2007.

The S&P 500 as a Benchmark

The advantage of this period for testing is that the market has not done too well over this period so it is a difficult period for any strategy to significantly beat the market. There are certainly very few mutual funds that have market beating returns. The test period includes three years of a major bear market and nearly six years of a bull. During this time the S&P 500 has a return of just under 2% CAGR without dividends. But more importantly, it has a huge Ulcer Index # and a very high maximum daily drawdown. Even the sharpe is negative.
             S&P 500
CAGR %
1.91%
GSD 19.34
Sharpe -0.01
Ulcer Index 22.76%
Drawdown -46.42%

The Best Predictive Measure for Annual Holds

The first thing that is clear is that the best measures for annual holds are not necessarily the ones that were best for monthly rebalances. Downside Deviation pops right to the top when the sort is UI, which for me, is the optimal means of selecting the best measure in this case. In fact, Sharpe and UI pick the same measure as best.
        Based On         Sort   CAGR   GSD   Sharpe  Ulcer Index  Troughs  Sortino  Yrs >= Index  Yrs >= 0  Drawdown
Downside Deviation ASC 22.75% 15.9 1.22 4.08% 3 1.78 100.00% 100.00% -17.13%
Ulcer Index ASC 19.04% 15.9 1.02 4.85% 2 1.47 88.89% 88.89% -21.02%
Sleep Ratio ASC 19.10% 16 1.01 4.93% 3 1.47 88.89% 88.89% -21.02%
Sharpe/(GSD^x) DESC 23.01% 18.4 1.10 4.95% 5 1.63 88.89% 100.00% -18.53%
GSD ASC 17.68% 14.6 1.00 5.30% 3 1.46 88.89% 100.00% -18.20%
Upside Potential ASC 15.79% 13.7 0.93 5.30% 3 1.33 88.89% 88.89% -19.82%
Normalized Trough Count ASC 18.47% 14.9 1.03 5.67% 4 1.49 88.89% 100.00% -18.79%
CAGR/(UPI^x) ASC 19.24% 20.6 0.84 7.22% 5 1.22 88.89% 77.78% -30.09%
Correlation DESC 26.46% 24.3 1.02 7.47% 10 1.53 77.78% 77.78% -25.54%
Treynor DESC 16.85% 18.9 0.78 7.49% 8 1.12 88.89% 100.00% -21.20%
GSD Ratio DESC 24.80% 23.5 0.98 7.84% 10 1.56 100.00% 100.00% -24.95%
Treynor/(GSD^x) DESC 12.36% 17.6 0.58 8.61% 7 0.82 77.78% 88.89% -25.32%
Treynor/(Beta^x) DESC 14.30% 18.9 0.65 9.12% 5 0.93 77.78% 77.78% -26.16%
Upside Potential Ratio ASC 19.07% 21.3 0.82 9.72% 5 1.19 77.78% 77.78% -29.23%
Upside Potential Ratio DESC 17.78% 27.4 0.64 12.26% 11 0.97 77.78% 77.78% -38.59%
Beta ASC 17.99% 23.8 0.71 12.85% 8 1.08 66.67% 66.67% -38.46%
Correlation ASC 12.81% 22 0.53 13.32% 11 0.76 77.78% 77.78% -39.07%
Ulcer Performance Index ASC 23.94% 27.3 0.86 13.45% 9 1.28 88.89% 77.78% -35.39%
Sortino DESC 20.73% 30.4 0.71 14.55% 14 1.04 77.78% 77.78% -49.77%
CAGR/(SF^x) DESC 11.03% 24.7 0.42 16.20% 9 0.62 66.67% 77.78% -43.05%
Sharpe DESC 19.04% 29.3 0.67 16.47% 13 0.99 88.89% 88.89% -49.77%
Sortino ASC 21.90% 28 0.78 16.60% 10 1.15 88.89% 77.78% -39.99%
Sharpe/(GSD^x) ASC 21.73% 27.7 0.78 16.93% 10 1.15 88.89% 77.78% -39.99%
CAGR/(UI^x) DESC 14.52% 26 0.55 17.23% 11 0.80 77.78% 77.78% -43.05%
Upside Potential DESC 18.33% 33.4 0.60 17.29% 15 0.88 66.67% 55.56% -47.57%
Normalized Trough Count DESC 23.49% 34.3 0.74 17.94% 14 1.08 77.78% 66.67% -50.43%
CAGR/(GSD^x) DESC 15.74% 28.7 0.56 18.21% 13 0.83 88.89% 88.89% -49.77%
GSD Ratio ASC 26.22% 34.7 0.81 18.68% 16 1.20 88.89% 77.78% -53.95%
Treynor/(Beta^x) ASC 19.74% 27.4 0.72 18.82% 10 1.04 88.89% 77.78% -42.63%
Jensen ASC 19.48% 26.3 0.73 18.99% 10 1.06 88.89% 77.78% -46.42%
CAGR/(GSD^x) ASC 20.66% 26.8 0.76 19.13% 9 1.12 88.89% 66.67% -44.92%
GSD DESC 14.24% 33.3 0.48 19.20% 12 0.70 66.67% 55.56% -50.15%
Alpha ASC 18.39% 27 0.68 19.23% 11 0.98 88.89% 77.78% -46.42%
Ulcer Performance Index DESC 14.12% 26.6 0.53 19.60% 11 0.77 77.78% 77.78% -43.47%
Sharpe ASC 20.15% 27.8 0.72 20.07% 9 1.07 88.89% 66.67% -47.12%
Calmar ASC 20.44% 28.2 0.73 20.11% 9 1.07 88.89% 66.67% -47.12%
Downside Deviation DESC 18.22% 32.6 0.61 20.60% 14 0.89 77.78% 66.67% -52.41%
Treynor ASC 20.06% 28.9 0.71 21.13% 10 1.04 88.89% 88.89% -48.47%
Treynor/(GSD^x) ASC 17.97% 29.2 0.63 21.18% 14 0.91 77.78% 66.67% -50.03%
Calmar DESC 12.74% 28.5 0.46 21.73% 12 0.69 66.67% 77.78% -56.04%
CAGR/(UI^x) ASC 18.35% 26.5 0.68 22.02% 10 1.01 88.89% 77.78% -46.81%
Jensen DESC 20.64% 33.3 0.67 22.89% 15 1.00 88.89% 88.89% -59.06%
Sleep Ratio DESC 13.36% 31.2 0.47 22.95% 13 0.67 66.67% 66.67% -53.29%
Alpha DESC 20.27% 33 0.66 23.58% 16 0.99 88.89% 88.89% -59.06%
CAGR/(SF^x) ASC 17.89% 27.6 0.65 24.94% 10 0.96 77.78% 55.56% -53.21%
CAGR/(UPI^x) DESC 14.82% 26.7 0.55 28.11% 10 0.79 88.89% 77.78% -56.33%
Ulcer Index DESC 14.06% 33.6 0.47 36.24% 15 0.69 77.78% 66.67% -68.01%
Beta DESC 8.62% 32.9 0.31 39.45% 16 0.44 77.78% 77.78% -68.19%

If the highest CAGR is your goal, according to this test, Correlation is best with a descending sort. What this means is that screens that are most highly correlated with the overall market do better than those that are not correlated. In fact, correlation has one of the biggest spans between the two sorts.

As you can see, no matter what method you employed you'd have beaten the market easily. This is more due to the nature of the screens selection for inclusion in the weekly ranks than in their potential to win in the future IMO. Which is all the more reason why it is important to have a good strategy for screen selection going forward. Choosing a blend of screens for annual holds based on just one snapshot is not advisable in my view. It is better to test it as I've done here -- employing a strategy for screen selection that is used at the end of each year and seeing how it worked out over the full test period.

The only problem with this whole approach is that for measures like Downside Deviation, you have very little turnover in the screens employed over the full history -- just 16% in this case. For this reason alone I'm tempted to select a measure that ranks very high, but has a much higher turnover rate -- Normalized Trough Count and Correlation fill the bill in that case.
Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Print the post Back To Top
Author: mungofitch Big gold star, 5000 posts Top Favorite Fools Top Recommended Fools Feste Award Winner! Old School Fool Add to my Favorite Fools Ignore this person (you won't see their posts anymore) Number: 205666 of 254198
Subject: Re: Annual Hold Strategy Date: 12/29/2007 8:29 AM
Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Recommendations: 2
Wonderful ranking, as always.

A couple of random thoughts---

Just to be clear, if I understood,
The downside deviation (as with all your metrics) is the daily calculation
from the daily portfolio values of that screen held 4 stocks deep for the
whole period of available history for that screen up to the ranking day,
running the ranking a total of nine times at the start of each year
starting with 1999?

Given the desire for a long hold, and the recent work showing topmost
ranks really are pretty good on average, this might be a nice "dozens".
Each month, calculate the rankings for the entire period so far, and buy
the rank=1 stock from all the screens calculated as optimal so far.
For 6 screens you'd end up with anywhere from 6 to 72 distinct stocks, but
it would usually be somewhere in the middle ground.

For the statistics, I'd be tempted to use the "all start dates" approach,
where you use the average of the statistics calculated 12 times, once for
each rolling month of the year. If the picks are a little different,
it might knock a little off the error bounds on the various metrics.

lastly...
Would I be correct in speculating that the screens which got selected
ended up being concentrated in the value and yield end of things?
I would expect this to be somewhat more true for the longer holds than for
the shorter holds.

Jim

Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Print the post Back To Top
Author: mungofitch Big gold star, 5000 posts Top Favorite Fools Top Recommended Fools Feste Award Winner! Old School Fool Add to my Favorite Fools Ignore this person (you won't see their posts anymore) Number: 205668 of 254198
Subject: Re: Annual Hold Strategy Date: 12/29/2007 9:00 AM
Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Recommendations: 7
Sorry for the digression, but here's a screen.

I read once that one of the best predictors of future earnings
growth is past growth in book value per share, not earnings growth.
(from the Rule #1 book, if I remember correctly).
Since an annual screen probably needs long run growth rather than short
run fashionability, this inspires the following screen for annual hold.

Timeliness 1-2 (based on recent post on even RS26 working well as an annual!)
Half of those (200 stocks) with lowest P/E ratio. Why pay more?
Book value growth 10-year top 10, using 5-year if 10-year not available.
No momentum criterion at all.

With annual hold, top 10, 1986-2006: CAGR 28, GSD 30, Sharpe 0.87
With annual hold, top 5, 1986-2006: CAGR 31, GSD 38, Sharpe 0.81
backtest.org/FastBookAnnual:h12SBT12XcesG0XpriDcesB200XObvgtVbg5CT10
Note, the great performance at high ranks is somewhat biased due to
a few huge wins in the topmost rank. (3.4x in 1990 and a 6.2x in 2001)
Take out 2-3 of those, and there isn't any particular falloff by rank.

Seems to work a heck of a lot better starting early in the year
compared to later in the year. Probably random noise, but maybe not.
Average of Jan/Feb/Mar starts 22.3%, average of Oct/Nov/Dec starts 12%.
I wouldn't count on it working better early in the year in future,
but nor would I try it starting later in the year!

Jim

Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Print the post Back To Top
Author: Zeelotes Big red star, 1000 posts Feste Award Nominee! Old School Fool Add to my Favorite Fools Ignore this person (you won't see their posts anymore) Number: 205670 of 254198
Subject: Re: Annual Hold Strategy Date: 12/29/2007 9:23 AM
Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Recommendations: 0
Jim asked:
The downside deviation (as with all your metrics) is the daily calculation from the daily portfolio values of that screen held 4 stocks deep for the whole period of available history for that screen up to the ranking day, running the ranking a total of nine times at the start of each year starting with 1999?

Daily for twenty years of history if available. Nine times? Not sure what you mean by that.

For the statistics, I'd be tempted to use the "all start dates" approach, where you use the average of the statistics calculated 12 times, once for each rolling month of the year. If the picks are a little different, it might knock a little off the error bounds on the various metrics.

I agree, but there comes a point where a backtester has to say "give." And on this one, I've reached that point.

Would I be correct in speculating that the screens which got selected
ended up being concentrated in the value and yield end of things?


Yes, that is true, but it is also true for the VL side with the monthly rebalance, at least for those with the lowest GSD and highest Sharpe.

Print the post Back To Top
Author: mungofitch Big gold star, 5000 posts Top Favorite Fools Top Recommended Fools Feste Award Winner! Old School Fool Add to my Favorite Fools Ignore this person (you won't see their posts anymore) Number: 205674 of 254198
Subject: Re: Annual Hold Strategy Date: 12/29/2007 10:14 AM
Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Recommendations: 3
Nine times? Not sure what you mean by that.

I meant, screens were selected once at the beginning of 1999, once at the
beginning of 2000, 2001, 2002, 2003, 2004, 2005, 2006, and 2007.

there comes a point where a backtester has to say "give."
Gee, about time!
You still put us all to shame.

Jim

Print the post Back To Top
Author: BigBunkler Big funky green star, 20000 posts Old School Fool Add to my Favorite Fools Ignore this person (you won't see their posts anymore) Number: 205688 of 254198
Subject: Re: Annual Hold Strategy Date: 12/29/2007 3:11 PM
Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Recommendations: 1
With annual hold, top 10, 1986-2006: CAGR 28, GSD 30, Sharpe 0.87
With annual hold, top 5, 1986-2006: CAGR 31, GSD 38, Sharpe 0.81
backtest.org/FastBookAnnual:h12SBT12XcesG0XpriDcesB200XObvgtVbg5CT10
Note, the great performance at high ranks is somewhat biased due to
a few huge wins in the topmost rank. (3.4x in 1990 and a 6.2x in 2001)
Take out 2-3 of those, and there isn't any particular falloff by rank.

Seems to work a heck of a lot better starting early in the year
compared to later in the year. Probably random noise, but maybe not.
Average of Jan/Feb/Mar starts 22.3%, average of Oct/Nov/Dec starts 12%.
I wouldn't count on it working better early in the year in future,
but nor would I try it starting later in the year!
Jim
=== === ===
Quite a spike in CAGR noted for January starts in Gritton's All Months tester:
backtest.org/FastBookAnnual:h12AM(SBT12XcesG0XpriDcesB200XObvgtVbg5CT10)
--BigBunk

Print the post Back To Top
Author: gritton Big red star, 1000 posts Feste Award Winner! Old School Fool Add to my Favorite Fools Ignore this person (you won't see their posts anymore) Number: 205693 of 254198
Subject: Re: Annual Hold Strategy Date: 12/29/2007 5:49 PM
Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Recommendations: 6
Quite a spike in CAGR noted for January starts in Gritton's All Months tester:
backtest.org/FastBookAnnual:h12AM(SBT12XcesG0XpriDcesB200XObvgtVbg5CT10)


Here's your spike:
http://backtest.org/9999h12ST(FastBookAnnual:SBT12XcesG0XpriDcesB200XObvgtVbg5CT10)11

That, my friends, is QCOM going 25x during 1999. 1999's picks 2-10 actually lost a little on the aggregate. An outlier like that is enough to make a backtest (especially an annual one) pretty near worthless.

- Jamie

Print the post Back To Top
Author: mungofitch Big gold star, 5000 posts Top Favorite Fools Top Recommended Fools Feste Award Winner! Old School Fool Add to my Favorite Fools Ignore this person (you won't see their posts anymore) Number: 205703 of 254198
Subject: Re: Annual Hold Strategy Date: 12/30/2007 7:19 AM
Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Recommendations: 0
That, my friends, is QCOM going 25x during 1999.

Ah, I wondered what it was!
I spotted the return, but of course I didn't know which stock it was.

It's interesting to note that a screen like this can catch the
occasional zoomer---I don't think it's entirely by chance, in that it's
looking for very high increases in book value per share, which is the
sort of thing that often catches the public's imagination.
Note the 239% in 1990, 517% in 2001, and 94% in 2006, all at rank 1.
I don't think average returns anywhere near this high can be expected,
but I do think it may be more than chance that several such did show up.

If true, then call options on rank 1 might qualify for part of the
"zippy" bit of the swan-avoidance "barbell" investing strategy (80-90%
zero-risk and 10-20% exposure to potential extreme upside).
Even though the expected return might not be great, the exposure
to the rare upside might be useful. Returns are blah (16-20%) but still
better than the S&P for variants avoiding QCOM (all month start
averages, 11 and 13 month holds, or dozens strategies any of which
dilute the one-lucky-pick effect). So the upside exposure has a chance
of having a positive expected value even in the absence of a moon shot.

Not a very good screen. But then not many annuals are particularly great.

Jim

Post New | Post Reply | Reply Later | Create Poll Report this Post | Recommend it!
Print the post Back To Top
UnThreaded | Threaded | Whole Thread (8) | Ignore Thread Prev Thread | Next Thread
Advertisement