August 2006 Letters To The Editor
or return to August 2006 Contents
The editors of S&C invite readers to submit their opinions and information on subjects relating to technical analysis and this magazine. This column is our means of communication with our readers. Is there something you would like to know more (or less) about? Tell us about it. Without a source of new ideas and subjects coming from our readers, this magazine would not exist.
Address your correspondence to: Editor, STOCKS & COMMODITIES, 4757 California Ave. SW, Seattle, WA 98116-4499, or E-mail to email@example.com. All letters become the property of Technical Analysis, Inc. Letter-writers must include their full name and address for verification. Letters may be edited for length or clarity. The opinions expressed in this column do not necessarily represent those of the magazine. -Editor
BROKERAGE RATINGS AND AWARDS
It is my understanding you publish an article each year rating online brokerage products and services. In which 2006 issue can I find that article?
You are probably referring to our annual "Readers' Choice Awards" section published in our annual Bonus Issue, which is mailed to subscribers in February and throughout the year to new subscribers. In that section, we present the results of an online reader survey on investment-related products and services in 20 categories, including analysis software and data services. In the survey, we ask readers to vote for their favorite products and services, and then we tally the results and present the listing.
So while we do not publish editorial ratings on brokerage services, this section in the Bonus Issue may be of interest. Please contact our subscription department at firstname.lastname@example.org or 800-Technical about how to get a copy.--Editor
CODE FOR THE RELATIVE VOLATILITY INDEX?
I am trying to get the computer code for the relative volatility index for the article "Refining The Relative Volatility Index" by Donald Dorsey, which I purchased from your Online Store. The program I use is Trade Navigator from Genesis Financial Technologies. I understand that you often provide such code at the back of your magazine. 1) Is the code I need available? 2) If so, how do I acquire it? 3) Does Trade Navigator from Genesis Financial Technologies usually provide such code at the back of your magazine?
Port Melbourne, Australia
Code implementing this technique was provided in Dorsey's article for TradeStation (in EasyLanguage), and code for several other programs implementing the technique was also given in our Traders' Tips section in that issue, which is the section you are referring to that appears toward the back of each issue. However, Trade Navigator was not yet contributing code to our magazine at that time, so it is not available through us. We suggest contacting Genesis Financial and submitting the formula and logic to the Trade Navigator support department to ask whether they can provide the Trade Navigator code to implement this technique.
Incidentally, you can search Traders' Tips sections that have been published in past issues at our website using the search feature at http://www.traders.com/S&C/SiteSearch.html. Select the "Traders' Tips" area from the pulldown menu to search in.
In addition, subscribers might like to know that you can log into the Subscriber Area at our website at http://technical.traders.com/sub/sublogin.asp using your last name and subscriber number from your magazine mailing label to check for code from any recent issue. We post code there if it appeared in an article. We do this as a convenience to subscribers, so that the code can be copied and pasted into their program.--Editor
MARKET (MIS)BEHAVIOR AND THE BROWNIAN MODEL
In your June 2006 issue you published the article "Harnessing The (Mis)Behavior Of Markets" by Rick Martinelli. I found this article fascinating, but not for the reason the author intended. I actually found it a classic demonstration of the dangers of strategy development through backtesting optimization.
In the article, Martinelli very clearly states the three defining characteristics of white noise, or Brownian motion. The first characteristic is "Price changes are statistically independent." This property means that there is no predictor, indicator, or measure that can give any insight about future behavior, beyond the overall statistics of the process. Yet the author then uses a simple straight-line predictor, wrapped in a trading strategy, optimized to a set of data, which makes a profit from trading a truly Brownian data series. In essence, he is successfully making predictions about the behavior of a data series that is by definition unpredictable. How is this possible?
The answer lies in an area the author does not discuss. One must ask, is the observed past profitability of an optimized strategy the result of the intrinsic behavior of the price data, or is it the result of finding some incidental characteristic of the particular data that is unrelated to the overall behavior or future profitability?
To put this into stark relief, consider a trading strategy that says I will simply take a position on the nth day after some starting point, and hold it for one day. Then I use an optimizer to find the value of n for a particular data series that gives the largest return. Obviously, the optimizer will find the one day in the data series with the single largest one day change. Will this strategy work going forward? More than likely it will not.
To avoid this type of situation, there are a few different tests that I use to sanity-check an apparently profitable optimized strategy:1) Is it predictive? If it is optimized over an earlier time frame, will it still make money subsequently? For example, cut the price data series for a stock into two halves. Train on one, test on the other. Is the strategy still profitable?
2) Is it portable? Can the same optimization be used to trade other similar stocks or commodities?
3) Are its parameters sensitive? Will small changes in parameter values drastically alter the performance of a system? If small parameter changes cause large changes in profitability, the system will probably fail.
4) Is the system profitable in different market conditions? Does it work in bull, bear, and transitional markets? If it does not work in all markets, can you identify the market-type and consequent behavior before suffering significant losses?
5) Unintentional use of future knowledge. This is the most insidious and hardest to detect. For example, all free price data sources do not offer data for stocks that have been delisted. Thus it is not possible to test a strategy on the price data for a company that is known to have failed.
I hope other readers find these criteria helpful.
Thank you for sharing your methodology. I am sure readers will find them helpful.--Editor
I was interested to read Rick Martinelli's June 2006 article "Harnessing The (Mis)Behavior Of Markets," but I have some questions and comments.
1) I looked at GM data in Excel over the same one-year period that you used in your Figure 1 and got basically the same ANN (-26.6%) but got a standard deviation of 2.40% looking back one year. That value seems to "fit," since the maximum ROC return was +18.1% (on the day you cited), so dividing 18.1% by 2.4% is 7.55, which is consistent with Figure 1. In addition, the minimum ROC is -13.99%, which is a sigma of -5.82, which is again consistent with the chart in Figure 1. The question is, then, where does the S=0.727 come from, since even looking back seven days (from 5/4/2005) results in 1.32%? Why this number?
2) How did you calculate the mean at -0.039%? Excel gives me -0.086% (or roughly -0.105% dividing -26.6% by 252 days).
3) I think Figure 4 in the article is most misleading. As you state, it is a "fake chart" (and I understand the implications of that) and is computer-generated using the same mean and standard deviation as Figures 1 and 3. The results have to be an anomaly, since I calculated only 43.6% up days versus 56.4% down days. Thus, one would expect most random runs to result in a negative buy & hold (as in a coin toss with that bias). The fact that it has a gain on the order of 40% is pure luck, and if one were to repeat the random run, say, 10 times, I'll bet most would have a negative buy & hold. The point here is that running a random sequence over a short time period like 253 datapoints is a crap shoot, and no conclusions can really be drawn from it or similar-type runs. Thus, I question the relevance of BM data in Figure 11.
Thank you for the article.
Norman J. Brown
Rick Martinelli replies:
Thank you for your interest in our research. Please note that data in the article is Yahoo-adjusted closes.
Our research was focused on the competing theses of Louis Bachelier and Benoit Mandelbrot. As such, random sequences, indeed true Brownian motions, were required to test these theses. Applying this method to an investment scheme to seek out market misbehavior is completely ancillary. During our research, I was as surprised as anyone to see results like the chart in Figure 6 showed. I assure you, no attempt was made to mislead anyone.
I have some comments on Rick Martinelli's article in the June 2006 issue, "Harnessing The (Mis)Behavior Of Markets." My comments have to do with his assumption of the underlying characteristics of the Brownian model. His assumptions do not properly characterize Brownian motion. I wonder if others have already written you with such a critique, and if so, I need not elaborate here. But if you would like, I can elaborate further by citing references that describe Brownian motion as 1/f noise or pink noise rather than white noise. The difference is subtle but crucial in the proper description of market price dynamics. For example, the Black-Scholes option pricing formulas properly treat Brownian motion dynamics.
Rick Martinelli replies:
It is not clear to me which of my assumptions you find improper. Brownian motion is a complex idea, has many features, and may be characterized in several ways. The defining assumptions I used in the article are the standard ones available in most books on probability or stochastic processes, such as Leo Brieman's classic, Probability, Chapter 12, Section 1. For a web reference, see http://mathworld.wolfram.com/WienerProcess.html.
These are mathematical definitions of Brownian motion and (I believe) were the ones considered by both Bachelier and Mandelbrot. It is true that the motion may be generalized in several ways, although these are procedures that are studied mostly within the engineering community. "White noise" refers to a flat (constant) process spectrum. A "pink noise" version is one in which the spectrum is no longer flat, and it may yield better results in some settings. But due to the nature of the article, I did not consider this or any other generalization of the plain-vanilla version.
I hope this adequately addresses your comment.
COVERED CALL WRITING VS. BUY & HOLD
I would like to issue a note of warning about the article by Gunter Meissner and Sandra Wu, which addresses covered call writing ("Does Consecutive Covered Call Writing Beat Buy & Hold?"), published in the May 2006 issue.
The authors are calculating the option premium by inputting the volatility index (VIX) implied volatility into a Black-Scholes calculator. Unfortunately, this leads to nonsensical option premiums, which will bear no reflection whatsoever of the true market value of these options. The VIX incorporates the whole SP skew, and is therefore around three percentage points above ATM implied volatility. (For example, see: "Time-Changed Lévy Processes And Option Pricing" by Peter Carr and Sandra Wu, Journal Of Financial Economics, 2004).
The premium calculated by the authors is therefore very likely to be much higher than what call sellers would have actually received, thus inflating the overall returns of their strategy and invalidating their conclusions.
PhD student in finance
CASS Business School, London
Gunter Meissner replies:
You state, "The VIX incorporates the SP skew, and is therefore around five percentage points above ATM implied volatility." The first part of your statement, "The (new) VIX incorporates the SP skew," is correct. However, due to a new algorithm, the new VIX is on average lower than the original one, as demonstrated in the following table:
Year Orig VIX New VIX High Low High Low 1990 38.07 15.92 36.47 14.72 1991 36.93 13.93 36.20 13.95 1992 21.12 11.98 20.51 11.51 1993 16.90 9.04 17.30 9.31 1994 22.50 9.59 23.87 9.94 1995 15.72 10.49 15.74 10.36 1996 24.43 12.74 21.99 12.00 1997 39.96 18.55 38.20 17.09 1998 48.56 16.88 45.74 16.23 1999 34.74 18.13 32.98 17.42 2000 39.33 18.23 33.49 16.53 2001 49.04 20.29 43.74 18.76 2002 50.48 19.25 45.08 17.40 2003 39.77 19.23 34.69 17.75
In conclusion, we used a low implied volatility, and hence, our results are conservative.
Gunter Meissner, PhD
Associate Professor of Finance (www.hpu.edu)
President, Derivatives Software (www.dersoft.com)
Back to August 2006 Contents
Originally published in the August 2006 issue of Technical Analysis of STOCKS & COMMODITIES magazine.
All rights reserved. © Copyright 2006, Technical Analysis, Inc.