No. of Recommendations: 22
I have a correction to make to my last post: The results I gave for SmallMoVL going to cash whenever EightDaySignal < 4 were indeed for a 1-day lag in the signal, as stated. However, the link I cited for the results (the second link in the post labeled "2", which should have been labeled "4") is wrong. It should have been http://gtr1.backtest.org/2012/?lf0lp0h15::tim.v:am4:mcp:bn61... . The incorrect link I did give imports EightDaySignal from a bad URL for the signal, which contains Field File Lag Adjustments and Price Lag Adjustments both set to 1, violating my recommendation to always compute signals with zero lag and apply any lag, if desired, during importation, not during signal computation, in order to avoid exactly this kind of mistake (which I warned about in my post LOL). As a result of the mistake, the signal is computed with a lag of 1 market day, and it is then imported with an additional day of lag, meaning the total signal lag in the backtest produced by the URL is 2, not 1. I caught the error while I was composing the post, but forgot to correct the URL after correcting the data table.

More importantly, I'd like to point out a big advantage of Jim's version of the 8-day NHNL signal over all other NHNL varieties I am aware of, and that is this: Whether there will be a change in Bull/Bear state at the next day's close can be predicted 96.6% of the time.

On those 3.4% of market days when the next market day's closing Bull/Bear state cannot be predicted, it would not be at all impractical to calculate the signal with a spreadsheet and real-time quotes during the next day's trading session shortly before the close, thereby allowing one to respond to a signal change with zero lag. 3.4% of market days (about once every six weeks) is less frequently than most of us trade.

To see how I came up with this 3.4% figure, observe that the only way there can be a change in Bull/Bear state at the next market day's close is if the latest seven-day sum of daily signals is exactly three, i.e., only when the number of highs has been at least the number of lows for exactly three out of the last seven market days. Why? Because if the seven-day sum is at most two (meaning the eight-day sum is at most three, a Bearish state), then the eight-day sum can be at most three at the next market close as well, which is still Bearish. And if the seven-day sum is at least four (meaning the eight-day sum is at least four as well, a Bullish state), then the eight-day sum must be at least four at the next close as well, which is still Bullish. In both cases, there will be no change in Bull/Bear state. QED.

Thus, to come up with the 3.4% figure, all I had to do was create a field that computes the seven-day sum of daily signals and count how many times it has equaled 3 since 1974, which I have done here: http://gtr1.backtest.org/2012/?h1::EightDaySignal:al0:SevenD... (I have left the EightDaySignal in there, but it has no effect on the screen.) If you click "Count Stocks", you see that an average of 0.0340741 stocks pass the "SevenDaySignal = 3" step. Since the universe consists of only one investment (^GSPC), that means it passes the screen 3.4% of the time.

As for why one would be so keen to trade on this signal with zero lag, here are the results of SmallMoVL, first with no timing, then with EightDaySignal applied with 1-day lag (as already reported in my last post), and finally with EightDaySignal applied with zero lag:

19860106 to 20140220
(1) (2) (3)
SmallMoVL 1-5, 15-day hold, no friction SmallMoVL 1-5, 15-day hold, no friction SmallMoVL 1-5, 15-day hold, no friction
No timing Go to cash any day where EighDaySignal (lag=1) < 4 Go to cash any day where EighDaySignal (lag=0) < 4
Avg Min Max SD Avg Min Max SD Avg Min Max SD

CAGR: 37.45 30.34 42.79 3.33 CAGR: 29.96 22.82 34.30 3.22 CAGR: 31.47 24.68 35.19 3.01
TR: 911417.19 167796.88 2165397.25 564478.50 TR: 187077.77 31693.42 389041.91 101686.02 TR: 250695.86 48317.98 467298.81 120492.70
GSD(20): 40.33 37.95 41.99 1.22 GSD(20): 27.30 26.38 28.04 0.55 GSD(20): 26.73 25.93 27.83 0.55
DD(20): 24.20 21.87 26.63 1.43 DD(20): 13.29 11.88 14.78 1.10 DD(20): 12.70 11.59 14.02 0.95
MDD: -58.20 -65.44 -51.79 3.95 MDD: -36.36 -47.58 -29.20 5.15 MDD: -36.79 -43.21 -29.86 4.54
UI(20): 12.58 11.14 14.28 0.89 UI(20): 9.55 7.43 11.67 1.39 UI(20): 8.88 6.83 11.00 1.26
Sharpe(20): 1.01 0.87 1.13 0.08 Sharpe(20): 1.03 0.83 1.18 0.09 Sharpe(20): 1.09 0.90 1.21 0.08
Beta(20): 1.43 1.36 1.51 0.05 Beta(20): 0.63 0.59 0.67 0.02 Beta(20): 0.58 0.54 0.62 0.02
TI(20): 24.02 20.95 27.11 2.01 TI(20): 41.01 31.87 46.63 4.32 TI(20): 46.63 36.53 52.02 4.57
AT: 9.47 9.30 9.63 0.10 AT: 8.53 8.39 8.70 0.09 AT: 8.53 8.34 8.71 0.10
(1) http://gtr1.backtest.org/2012/?h15::tim.v:am4:mcp:bn610:dolv...
(2) http://gtr1.backtest.org/2012/?h15::tim.v:am4:mcp:bn610:dolv...
(3) http://gtr1.backtest.org/2012/?h15::tim.v:am4:mcp:bn610:dolv...

You can decide whether chasing that possible 1.5 point CAGR advantage is worth the effort (which wouldn't be much with enough automation), but if I were using SmallMoVL with timing, I think it would be.

Robbie Geary
Print the post  

Announcements

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.
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.
Advertisement