March 1997
Letters to the Editor
Editor,
CHARTING BY HAND
As an overseas subscriber, I would like to remind the advertisers in your esteemed magazine that it is not possible for overseas readers to use toll-free phone numbers.I am searching feverishly for an accurate system that trades stocks and options, one that would also work in Europe. I prefer a system that only requires a piece of paper and pencil, allowing me to make diagrams by hand, since I have no computer. Can you provide any suggestions?
ADI ZARETZKIAs most of the trading world is computerized, it's difficult to find a system that doesn't require a computer. Besides, isn't everyone searching for an accurate, flexible system?! -- Editor
Stockholm, Sweden
Editor,
COT INDEX
I hope you will have more articles on the Commitments of Traders (COT) index, such as the June and September 1996 articles by Scott Barrie. As Barrie points out in those articles, the characteristics of each market certainly need to be considered. Winning With the Insiders by R. Earl Hadady, I. Lee Finberg and Dan Rahfeldt, published by Weiss Research, got me started on this subject, but Barrie's articles are much more helpful, because Winning doesn't consider different contract months and different times of the year. I use the Cot as one of my initial filters.FRANK THOMPSON
via E-mail
Editor,
INSYNC INDEX
In the January 1995 STOCKS & COMMODITIES, you published an article by Norm North on the Insync index, which is a consensus indicator made up of 10 components designed by North. I am very interested in finding out more about this index, as the article didn't provide a formula for working out this index.Can you help?
J. MANAK
Ayr, Queensland, AustraliaCompiling the Insync index entails much work. As was stated in the article, its values are the net sum of all its component indicators' buy and sell signal areas at any given time, expressed as an index. I suggest contacting Norm North about his products (North Systems, 503 364-3829, NorthSys@aol.com). -- Editor
Editor,
BREAKING THE RULES
I read Jeffrey Katz and Donna McCormick's December 1996 article titled "A rule-based approach to trading," and I was a bit disturbed by the testing procedures and presentation of the results. I find this especially discouraging in view of the fact that Katz is a Ph.D. My specific complaints and comments are as follows:It's not surprising that the "almost perfect entry points" are almost perfect because they pertain to the data that was used to optimize or "fit" the rule. The real question is, how well would the entry points have fared in out-of-sample tests? Without out-of-sample testing, there is no benchmark against which the system may be compared. By my rough calculations, had one simply been long the near S&P 500 contract over the same 171-trading day period as the study, the account would have garnered 94.45 bars, or the equivalent of $47,225 ($276.16 per trading day). This is well in excess of Katz's findings and subject to significantly lower transaction costs.
Although the system claims that 52% of the trades in the optimized dataset were winners, this figure is not nearly as good as it looks. With 85 trials, there is a 41.4% probability of receiving a winning percentage higher than 52%, assuming a binomial distribution with equal probabilities. Based on the above figures, the actual probability of placing a winning trade may have been understated, further eroding Katz's results.
I hope that in the future the contributing writers will be required to be more careful in their studies and the way their results are presented to the readers.
CARL MAIERJeffrey Katz replies (with input from Donna McCormick):
Anderson Graduate School of Mgmt.
University of California RiversideI regret having to say that Maier's points are not well taken.
Regarding his criticism of the "almost perfect entry points," although it was not discussed in the article, out-of-sample data on the bond market was examined. Trades that were essentially similar to those observed in-sample were considered. Furthermore, the large number of trades, relative to the small number of parameters being optimized, greatly mitigates the dangers of curve-fitting. Finally, the system was tried (with parameters unchanged) on a totally different market -- that is, the S&P 500 -- and similar trading results (although more profitable, due to the larger dollar-value swings in that market) were obtained. This last test can be construed as out-of-sample, since it involves a completely different market than that on which the system was optimized. In any case, my November 1996 article, "On developing trading systems," as well as my subsequent and forthcoming articles, present both in-sample and out-of-sample results.
Regarding the benchmark comparison: I like the idea of benchmarks. The one Maier suggests, however, is not a valid benchmark for several reasons. The system had certain design constraints, including the use of very tight stops, the restriction that no positions be held overnight and that positions are held for only a very small percentage of the total trading time of the period tested. If, as he suggests, one had been long the near S&P 500 contract for 171 trading days, one would indeed have made more money; however, one's exposure to the market would have been incredibly greater in terms of absolute proportion of time in the market (100% for his suggested benchmark), exposure to overnight gaps and risk due to the absence of stops. Yes, there would have been lower transaction costs. However, I am not trying to find the most profitable system; instead, I am trying to work in the direction of developing a practical, tradable system. A more realistic benchmark would be to take a number of samples, each consisting of a set of trades similar to those generated by my system, all exited on the close and using the same stops, but with random entries; this would provide a benchmark with equal market exposure, no positions held overnight, equal stops and a roughly equal percentage of time in the market. That is the only kind of benchmark that would provide a fair comparison; it's a fairly complicated, Monte Carlo-type benchmark, which I did not use because it cannot be computed in TradeStation and it would have pushed the article well beyond its intended scope. Such a benchmark might be worthwhile if you were evaluating a trading system that you fully intend to trade. In these articles, however, my intention is more to explore the kinds of results one gets using different approaches in a rather simplistic manner. As readers will see, in later articles I do use genetic algorithms, Monte Carlo simulations, tests of statistical significance and more. It should also be noted that, in addition to the long side, the short side was profitable; this implies that the results could not be exclusively due to the fact that the S&P 500 or the bond market was rising during the period of the test.
Regarding the probability of winning trades: Maier's comments are very confusing, given that I simply reported that 52% of the trades were winners. I did not discuss probabilities of the winning percentage being higher or lower than any threshold. Finally, the percentage of winning trades is not generally all that important: One can have a low percentage of winning trades and yet have a very profitable system. For example, a system using very tight stops and having occasional extremely large wins may lose 80% of the time, but the losses only amount to $100 or $200, an amount that would easily be covered and surpassed by the occasional $3,000 or $4,000 winning trade.
Finally, there are a lot of statistics that are easy to compute and that help evaluate a trading system, but many of them are simply not available in TradeStation. I plan to discuss some of these statistics in future articles. I encourage Maier to continue reading, as it may provide him with further understanding of such issues.
Editor,
TICK DATA ANALYSIS
I have a question concerning software. I need a program in which I can analyze tick data. But I also want to be able to change the periodicities when they are first set. As I want to backtest some futures, I need to change the calculations on those time frames in Dos, and MetaStock for Windows cannot display intraday charts.Do you know of any software that is able to change the periodocities as often as I want, where I can run calculations and tests in Dos, as you can in MetaStock's Explorer and System Tester for Windows, and where I can see the tick data charts with different time frames on the screen of my PC?
GERALD HOFER
Vienna, AustriaI recommend upgrading to Windows, which will give you more flexibility in your choice of software. -- Editor
Editor,
ONLINE INDEX
I have been a subscriber to S&C for some years now and have saved all the issues, plus I've bought all the Volume Books for years prior to the commencement of my subscription. That's a lot of information! An online index of all your back issues at your Web site would be extremely valuable. Some time ago, you promised it would be coming. Is it still in the works? It seems the index could be done essentially from your S&C on CD. Hard for me to see any disadvantage in putting such an index on the Web. What do you think? Here's hoping.Thanks to all the S&C staffers for a fabulous job over the years.
LARRY HANSON
via E-mailCurrently, we are converting S&C on CD into a format that can be posted at our Web site for full-text searching. One of the great things about S&C on CD is its ability to perform comprehensive searches. We want to port this capability to our Web site both as a research tool for our subscribers and as a sales tool for the magazine. -- Publisher
Editor,
CONSTRUCTIVE CRITICISM
Your publication is outstanding; I've been a subscriber for more than five years and will remain a subscriber, no matter what. However, I must make a constructive criticism.As you must know, Figure 1 on page 62 of the December 1996 STOCKS & COMMODITIES erroneously used the term "transportation stocks" on the lower part of the chart, when it should have read "technology" stocks. Most of us probably discerned this after some contemplation, but the problem with your publication is more serious than one simple error.
The January 1997 issue contains an excellent article by Dennis Meyers, but it is replete with reading problems. Minor problems are that Brs, Srs, nlbx, nlsx, nhsx, etc., are used with no explanatory note, while spx = S&P 500 is defined as if "spx" will be used within a major formula, which never happens. These are only minor problems, but a major problem is that we are shown Excel optimization reports in Figures 1 and 2 with headings that are never defined. Studying the article's contents is really a chore.
You have a wonderful practice of creating sidebars to amplify the use of formulas by such learned contributors as Meyers, and these are of significant help to those of us without doctorates in mathematics. Of great help in this article would have been a sidebar showing how the formulas are derived and used.
After three attempts, I've given up trying to decipher Meyers's approach, which is a shame because, according to his profit results, he's really on to something.
During my career, I developed corporate policy statements for the CEO of Lockheed Martin in Bethesda, MD. I learned to flight-test my draft policies with individuals unfamiliar with my research, to test the readability of the document on a layperson before publishing it. Perhaps it would benefit your publication if you tried this approach; that is, let a novice read the article to see if he or she can follow the content, intent and procedure.
Again, your publication is without peer, and I am and always will be a very loyal subscriber.
WILLIAM F. ODELL
Winter Park, FLFirst, you're correct about the error in Figure 1.
Second, let me try and explain the definitions for the parameters Brs, Srs, and so on. All of these parameters are values to be found in the development of the system. They represent constants. For example, Brs is a number that the ADRrs indicator must exceed for a buy signal to occur. Brs could have been called x or any other variable name, but we try to make the variable names meaningful and use acronyms where possible. You make a good point, though, that it would have been easier for readers to follow the article if we had produced a reference table or key as a reading aid. The article is advanced as it is, and the acronyms can hinder readability. In similar future articles, we'll try to incorporate a sidebar defining the terms.
Finally, the column headings in the Excel optimization reports are either parameters from the system or acronyms from the summary tables, such as the summary table in Figure 3. For example, Npft is net profit; L%P is long percentage profit; S%P is short percentage profit; and so on.
Thanks for writing. We always appreciate feedback. -- Editor
The TradeStation code for the Adrrs-NH-NL system, which appeared in the January 1997 Traders' Tips, contained an error in the calculation of the advance-decline ratio. The 10-day exponential average of the ratio should have read as follows:
ERATTA
ADRx=XAverage((AI-DI)/(AI+DI),10)In addition, note the author's optimal inputs for Brs=4 and Srs=-40.
JIM APPLEGATE
Ft. Myers, FL
In the product review of CycleTrader in the February 1997 issue, one of the prices was listed incorrectly. The price of the CycleTrader SC Real-Time version (for SuperCharts) should have read $695, not $349.
ERATTA
