UnThreaded | Threaded | Whole Thread (5) | 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 253129  
Subject: Blending at a Whole New Level Updated Date: 12/13/2012 12:43 PM
Post New | Post Reply | Reply Later | Create Poll . Report this Post | Recommend it!
Recommendations: 37
Back in July of 2008 some of you may recall that I did extensive backtesting to come up with a better solution than using Excel's Solver. The idea was simple: Use the running balance of all Gold, Silver, and Bronze Value Line screens to find a measure that is more predictive of higher, and safer, returns than simply a dart board. The conclusion of all that testing was:

1. Optimal to employ a simple timing system on top to limit downside risk during bearish periods -- I suggested the Nasdaq NH-NL -- which from those posts till now has become quite well known on this forum. For the sake of my backtests today I'm using this to go wholly into Cash during the bearish periods, since my inclination and practice is this.

2. Sharpe/GSD is an excellent measure for choosing the optimal blend.

These tests are built on the premise of holding FIVE screens with each of them invested in FOUR stocks, for a total hold of TWENTY stock positions.

Original Research Links:

Blending at a Whole New Level - 7/22/2008
http://boards.fool.com/blending-at-a-whole-new-level-2683434...

The Best Measure for the Best Blend - 11/16/2007
http://boards.fool.com/the-best-measure-for-the-best-blend-2...

Optimizing Blends with Sharpe/(GSD^x) - 11/17/2007
http://boards.fool.com/optimizing-blends-with-sharpegsdx-261...

Since I first shared about using Sharpe/GSD in the middle of November 2007, these tests are based on a backtest from 1/1/2008 to the present. In other words, I am seeking to validate that the method outlined back then ended up being a good one for selecting the screens to invest in to create a blend.

Indexes to Compare To

These indexes are being tested over the same time period and will serve as a benchmark for the blends that follow.

Symbol                Name                 CAGR    GSD   Sharpe   DDD3    BBW   Drawdown
^BTK Amex Biotech Index 14.90% 34.37 0.56 6.78% 2.28 -38.89%
^RLX Retail - Rlx 10.43% 35.95 0.46 9.09% 1.19 -49.68%
^IIX Amex Internet 7.68% 34.39 0.38 9.78% 0.84 -54.44%
^PSE Tech: Pse Tech 6.39% 29.66 0.37 10.21% 0.67 -45.04%
^NDX Nasdaq 100 Index 5.68% 31.31 0.3 10.87% 0.56 -49.56%
^XAL Amex Airline Index 5.37% 67.54 0.32 15.30% 0.37 -67.10%
^SML S&p 600 Small Cap 3.74% 37.29 0.25 12.61% 0.32 -54.79%
^CMR Amex Ms Consumer Index 3.18% 21.83 0.23 8.63% 0.4 -41.13%
^IXIC Nasdaq Composite 3.07% 31.63 0.22 11.36% 0.29 -51.43%
^XCI Amex Computer 2.99% 30.91 0.21 10.32% 0.31 -50.22%
^HCX S&p Healthcare Index 2.74% 22.94 0.2 8.14% 0.36 -40.32%
^MXY Amex Mexico Index 2.48% 38.39 0.21 15.93% 0.17 -64.61%
^DRG Pharmaceutical 2.36% 22.16 0.18 6.82% 0.36 -38.65%
^MSH Morgan Stanley Technology Index 2.35% 34.18 0.2 11.62% 0.22 -53.74%
^XNG Energy - Natural Gas 2.30% 47.88 0.26 14.40% 0.17 -60.69%
^RUT Russell 2000 2.11% 39.52 0.21 12.88% 0.18 -55.03%
^XTC Telecommunications 1.58% 31.31 0.17 10.81% 0.16 -52.82%
^CYC Amex Ms Cyclical Index 0.81% 42.02 0.18 18.38% 0.05 -72.29%
^DJI Dow Jones Industrial Average 0.32% 27.15 0.11 12.07% 0.03 -49.86%
^RUA Russell 3000 0.23% 30.99 0.12 13.05% 0.02 -53.47%
^DJA Dow 65 Composite 0.14% 28.34 0.12 12.74% 0.01 -51.95%
^RUI Russell 1000 0.07% 31.4 0.12 13.07% 0.01 -53.41%
^GSPC S&p 500 -0.28% 30.24 0.1 12.97% -0.02 -53.27%
^OEX S&p 100 -0.78% 28.98 0.07 12.50% -0.07 -52.36%
^SOXX Phlx Semiconductor Sector -1.12% 42.43 0.14 13.70% -0.09 -59.37%
^XII Institutional Ndx -1.26% 28.46 0.05 12.48% -0.11 -51.22%
^XAU Phlx Gold And Silver Sector -2.24% 58.66 0.16 14.11% -0.17 -68.15%
^NYA Nyse Composite -2.79% 32.34 0.02 14.24% -0.21 -56.23%
^VLIC Value Line -3.50% 39.93 0.05 16.62% -0.23 -64.73%
^UTY Phlx Utility Sector -3.81% 24.62 -0.08 11.30% -0.37 -47.63%
^HKX Amex Hong Kong 30 Index -4.17% 33.09 -0.02 14.87% -0.31 -59.64%
^XOI Energy - Oil -4.60% 42.8 0.03 14.02% -0.35 -53.30%
^JPN Amex Japan Index -9.41% 33.01 -0.22 14.32% -0.71 -53.55%
^BIX Finance: S&p Bank -10.45% 77 0.08 21.89% -0.52 -83.31%
^BKX Phlx - Kbw Bank Index -10.62% 72.77 0.06 22.34% -0.52 -80.63%
^IUX Finance: Insurance -11.03% 53.59 -0.07 21.89% -0.55 -78.64%
^XBD Finance: Broker -15.48% 61.35 -0.11 20.56% -0.82 -75.00%

The blends that follow are all sorted based on the CAGR. Amazingly, the absolute best measure ends up being the exact same one that was determined to be best back in 2007 and 2008. A very nice confirmation of all the work that went into those posts back then.

         Measure           Sort   CAGR    GSD   Sharpe  DDD3    BBW   Drawdown
Sharpe/(GSD^x) DESC 8.85% 16.69 0.68 3.29% 2.84 -24.90%
Treynor/(Beta^x) DESC 8.70% 17.23 0.64 4.64% 2.01 -27.13%
Treynor/(GSD^x) DESC 6.85% 14.67 0.58 3.58% 2.03 -25.22%
Best of Both Worlds DESC 7.40% 16.48 0.57 4.11% 1.92 -27.67%
Calmar DESC 7.01% 16.52 0.57 5.22% 1.45 -26.79%
Ulcer Performance Index DESC 7.09% 16.99 0.55 5.89% 1.31 -28.59%
Sortino DESC 7.32% 18.22 0.55 4.72% 1.66 -27.10%
Jensen DESC 8.04% 19.44 0.54 6.13% 1.42 -31.10%
Sharpe DESC 6.68% 18.1 0.51 5.69% 1.27 -28.05%
Treynor DESC 6.62% 16.64 0.51 5.00% 1.43 -27.82%
CAGR/(UPI^x) ASC 6.23% 17.24 0.51 4.21% 1.59 -23.28%
Alpha DESC 6.91% 19.23 0.49 6.91% 1.09 -31.08%
CAGR/(GSD^x) DESC 6.33% 17.01 0.49 5.41% 1.26 -27.31%
Double Downside Deviation ASC 5.14% 14.95 0.49 4.25% 1.29 -26.20%
CAGR/(UI^x) ASC 6.66% 23.47 0.44 4.14% 1.69 -28.50%
Correlation ASC 5.59% 18.57 0.44 3.23% 1.83 -24.31%
GSD DESC 6.30% 26.7 0.43 5.23% 1.29 -31.30%
CAGR/(SF^x) DESC 4.67% 15.73 0.42 5.85% 0.87 -29.07%
Sleep Ratio ASC 4.38% 15.4 0.42 4.16% 1.13 -20.91%
CAGR/(UI^x) DESC 4.76% 15.49 0.42 6.14% 0.84 -29.33%
Sleep Ratio DESC 5.02% 24.67 0.41 4.79% 1.11 -29.49%
Downside Deviation DESC 5.03% 25.97 0.39 6.24% 0.87 -34.46%
Ulcer Performance Index ASC 4.99% 21.84 0.36 3.94% 1.33 -27.36%
Treynor/(Beta^x) ASC 5.42% 21.44 0.36 3.65% 1.56 -27.76%
Sortino ASC 4.05% 21.62 0.32 3.90% 1.09 -27.36%
Upside Potential ASC 3.28% 14.95 0.32 2.89% 1.18 -16.78%
Calmar ASC 4.25% 21.39 0.32 4.11% 1.09 -27.36%
Sharpe ASC 4.05% 21.62 0.32 3.90% 1.09 -27.36%
Ulcer Index ASC 2.37% 14.49 0.31 4.97% 0.52 -22.93%
Treynor ASC 3.92% 21.21 0.31 4.53% 0.92 -27.20%
GSD Ratio ASC 2.76% 18.73 0.31 3.90% 0.74 -25.85%
GSD Ratio DESC 2.78% 17.63 0.31 5.33% 0.56 -26.30%
CAGR/(UPI^x) DESC 3.66% 22.51 0.3 5.93% 0.66 -30.71%
Double Downside Deviation DESC 3.19% 23.71 0.29 5.78% 0.59 -32.95%
CAGR/(GSD^x) ASC 3.84% 21.2 0.29 3.74% 1.08 -27.53%
GSD ASC 2.59% 14.23 0.28 3.24% 0.84 -17.53%
Upside Potential DESC 2.35% 25.43 0.27 5.94% 0.42 -32.31%
Sharpe/(GSD^x) ASC 3.37% 20.85 0.27 4.13% 0.86 -29.17%
Upside Potential Ratio DESC 2.42% 18.23 0.26 6.45% 0.41 -28.74%
Beta DESC 2.03% 26.72 0.26 7.77% 0.28 -36.69%
Correlation DESC 2.16% 17.48 0.25 5.54% 0.42 -23.08%
Downside Deviation ASC 1.85% 14.73 0.23 3.77% 0.52 -18.83%
CAGR/(SF^x) ASC 2.65% 21.34 0.22 4.67% 0.6 -28.50%
Normalized Trough Count ASC 1.69% 15.37 0.21 4.66% 0.39 -22.89%
Best of Both Worlds ASC 1.86% 21.06 0.21 4.80% 0.41 -28.50%
Jensen ASC 1.61% 22.2 0.2 5.30% 0.33 -29.04%
Alpha ASC 1.90% 21.94 0.2 5.21% 0.39 -28.77%
Treynor/(GSD^x) ASC 2.16% 20.4 0.2 4.83% 0.48 -29.06%
Upside Potential Ratio ASC 1.55% 19.92 0.2 4.51% 0.36 -27.42%
Ulcer Index DESC 0.33% 23.41 0.17 5.92% 0.06 -30.54%
Normalized Trough Count DESC -1.41% 25.82 0.11 8.18% -0.19 -32.67%
Beta ASC 0.29% 14.98 0.09 4.49% 0.07 -23.43%
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: 240535 of 253129
Subject: Re: Blending at a Whole New Level Updated Date: 12/13/2012 1:17 PM
Post New | Post Reply | Reply Later | Create Poll . Report this Post | Recommend it!
Recommendations: 4
The following takes the top ten measures and applies one of my better timing systems. In this case, there are only two bearish periods from 7/1/2007 to the present. I extended it back in time to take in the full brunt of the bear market from 2007 to 2009... in other words, to raise the bar and make it more difficult to have a position return. In this case, as with the previous backtest, during the bearish period I'm just sitting in One Year Treasuries. The S&P 500 drops over -5% during this time period.

        Measure          Sort   CAGR    GSD   Sharpe  DDD3    BBW   Drawdown
Sharpe/(GSD^x) DESC 19.88% 20.12 1.05 0.81% 19.42 -22.66%
Sharpe DESC 18.48% 22.42 0.92 3.63% 5.24 -29.61%
Sortino DESC 18.33% 22.05 0.92 2.72% 6.7 -30.26%
Treynor/(Beta^x) DESC 18.27% 21.01 0.94 1.60% 10.83 -21.14%
Treynor/(GSD^x) DESC 18.22% 21.5 0.93 1.24% 13.14 -22.64%
Jensen DESC 16.14% 24.86 0.78 5.41% 3.16 -36.76%
Ulcer Performance Index DESC 15.55% 19.9 0.86 3.59% 4.5 -30.79%
Best of Both Worlds DESC 15.49% 20.21 0.85 3.42% 4.65 -30.79%
Calmar DESC 15.26% 19.39 0.86 3.64% 4.35 -30.79%
Treynor DESC 12.56% 20.98 0.69 4.63% 2.87 -31.69%


Post New | Post Reply | Reply Later | Create Poll . Report this Post | Recommend it!
Print the post Back To Top
Author: rgearyiii Three stars, 500 posts Old School Fool Add to my Favorite Fools Ignore this person (you won't see their posts anymore) Number: 240536 of 253129
Subject: Re: Blending at a Whole New Level Updated Date: 12/13/2012 1:44 PM
Post New | Post Reply | Reply Later | Create Poll . Report this Post | Recommend it!
Recommendations: 30
Zeelotes:
Back in July of 2008 some of you may recall that I did extensive backtesting to come up with a better solution than using Excel's Solver.

I do a lot more than recall. That post was the beginning of what's turned out to be a 4+ year project to turn the GTR1 backtester into a tool for both backtesting your WWL system and generating daily stock picks for it, which is finally just around the corner.

The GTR1 version will of course exploit daily data to the max, allowing for averages across daily-staggered trading cycles when calculating the various metrics, perhaps leading to sharper signals and even better results. You'll be able to throw ETFs (or any ticker symbols from Yahoo! Finance) into the mix as well, and import market timing signals generated in other backtests.

Robbie Geary

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: 240537 of 253129
Subject: Re: Blending at a Whole New Level Updated Date: 12/13/2012 2:37 PM
Post New | Post Reply | Reply Later | Create Poll . Report this Post | Recommend it!
Recommendations: 5
Wonderful!

If there is a moral to the story, it might be this:
If you're going to pick stocks mechanically and use backtests to
determine what methods work well, it makes sense to pick your screens
mechanically and use backtests to determine which of those work well.

As I recall, the test details include the following
- you use daily GSD (GDSd not GSDm)
- GSD^x is always squared in this list (x=2)
- four week hold cycle assumed unless there is a timing signal hit
- no reduction in the weight of duplicate picks, instead you reduce high-overlap
situations by pre-eliminating screens that have high average overlap ratios
- you recalculate the set of screens only once per year, based on all
the history up to the end of the year just ending for use the whole of the next
- Though I believe you have in the past done this separately for VL and
SIpro screens, I imagine this table of results based on selection from
the unified screen list of all "standard" long screens other than SOSs
and those eliminated due to high overlap of picks with another screen.

Does that all sound about right?

Jim

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: 240538 of 253129
Subject: Re: Blending at a Whole New Level Updated Date: 12/13/2012 2:52 PM
Post New | Post Reply | Reply Later | Create Poll . Report this Post | Recommend it!
Recommendations: 4
Jim wrote:
If you're going to pick stocks mechanically and use backtests to
determine what methods work well, it makes sense to pick your screens
mechanically and use backtests to determine which of those work well.


Absolutely! This was and is the whole point in a nutshell.

- you recalculate the set of screens only once per year, based on all
the history up to the end of the year just ending for use the whole of the next


In the case of these tests I recalculate at each signal. Furthermore, I did not use the full history since I didn't want to redo the full history in order to run these tests. So I ran them from 2005 to present, but actually only had the measure look at the past 12 months to determine the screens to choose.

Though I believe you have in the past done this separately for VL and
SIpro screens, I imagine this table of results based on selection from
the unified screen list of all "standard" long screens other than SOSs
and those eliminated due to high overlap of picks with another screen.


This is for Value Line screens only. Only Gold, Silver, and Bronze screens are considered. This is a total of 47 screens.

Print the post Back To Top
UnThreaded | Threaded | Whole Thread (5) | Ignore Thread Prev Thread | Next Thread
Advertisement