July 1999 Newsletter

I. Time Running Out for the NeuroShell Trader Professional Sale
July 15th is the last day you may purchase or upgrade to the NeuroShell Trader Professional at the introductory price of $1295. On July 16th, you will have to pay $100 more. If you were thinking of upgrading or purchasing, now would be the time to do it!
We have further discounts for existing owners of the NeuroShell Trader. If you purchased the NeuroShell Trader on or before April 15, 1999, your discount is $895. If you purchased after April 15, 1999 your discount is $795.

Visit http://www.neuroshell.com for detailed information about what is new in release 2 of NeuroShell Trader as well as the advanced features in NeuroShell Trader Professional.

II. MESA and WAVEFIN Now Available for the NeuroShell Trader.

We are pleased to announce that several “outside” developers are now producing products to run on the NeuroShell Trader. Two highly respected products are already available: MESA and WAVEFIN. You purchase these products directly from the vendor: we do not sell them ourselves. We have included information about both of these fine products below with further information about how to contact the vendors. You will need to be on release 2.1 or higher of the NeuroShell Trader to use MESA and WAVEFIN.


Market cycles are difficult to measure because they come and go. Valid measurements can be made only when the data is stationary over the observation time span. That is, the cycle must have the same amplitude and period. MESA techniques provide good measurements because only a short amount of data is used. The shorter the observation time span, the higher the probability of grabbing stationary data. Therefore, MESA has the ability to make the best measurement of market cycles.

But the real question is what to do with the cycle measurement once it has been made. If we model the market as consisting of a cycle plus a trend, and removes the cycle component, then we are left with an “instantaneous trendline”. We can do this by taking a simple average over the period of the measured cycle. There are as many sample points above the mean as below it over this period, with the result that the dominant cycle is completely removed. Since the dominant cycle period continuously shifts, we are left with the instantaneous trendline as an indicator. In practice, this trendline needs to be further smoothed to remove cyclic components other than the dominant cycle.

The MESA instantaneous trendline is an indicator that has been added to the Trader to assist your analysis of trend mode conditions. The phase of the cycle is another indicator available to you. Now you can tell exactly what in what part of the cycle you are located so that you can better estimate market turning points.

By taking the Sine of the phase angle we can create an oscillator-like indicator for the cycle modes of the market. This has an advantage over a Stochastic or RSI that a leading indicator can be created simply by advancing the phase angle by 45 degrees. The crossings of the two lines are precursors of cycle mode turning points. Advancing the phase does not increase the noise level of the indicator, as is the case with momentum functions. Perhaps one of the most significant benefits of the Sinewave Indicator is that the phase ceases to advance in the trend mode. Therefore, this oscillator does not give false crossing signals when the market is in a trend.

The final indicator in the MESA group shows the mode of the market; whether the mode is a trend mode or a cycle mode. This indicator is based on measuring the rate-change of the phase. Since the phase does not advance in trends, this indicator enables you to select when to use trend mode techniques such as moving averages or when to use cycle mode techniques such as Stochastics, RSI, or the Sinewave Indicator.

The MESA indicators are a valuable addition to the Trader because they give you a completely independent viewpoint of market activity.

For further information about MESA contact:

Mesa Software, Inc.
John Ehlers
PO Box 1801
Goleta, CA 93116
Voice: (805) 969-6478
Fax: (805) 969-1358


Sooner or later every neural network practitioner learns that the choice and preprocessing of input data is the most critical task when noisy data is used. Neural networks tend to perform better when filtered data used as inputs ‘under-reacts’; filters should yield values that are smooth, and consistent when faced with similar patterns. Reliability of filter response is even more important when the training data set size is relatively small since the neural network easily can memorize the historical patterns without retaining an ability to recognize them in real time when the price action will be different than the historical price action. Wavelet filtered variables provide an excellent form of preprocessing for a neural network model in addition to otherwise proper normalization.

Noise Elimination and Feature Detection

Noise is the bane of most trading systems used by off-the-floor traders. One person’s noise is likely to be another person’s information, so it is important to avoid generalizing about precisely what constitutes noise. However, once noise has been defined in the context of an application or trading system, it is relatively easy to design a filter to eliminate it: We simply define a filter that captures the noise, then subtract the captured noise from the original series. What remains is (presumably) the important information.

The complement to noise elimination is feature detection. For information varying at an approximately known rate it is appropriate to apply a filter that is concentrated at the expected rate. The output of this filter will be the information in which we are most interested, with extraneous variation removed.

Feature detection is closely related to noise reduction and the difference is largely a question of perspective. Perhaps we believe that very high frequency information in a price series is largely due to the activity of floor traders or otherwise represents untradeable information in the context of our system. In such a case we will want to remove the filtered information. On the other hand, perhaps when the larger commercial account traders enter a market their presence causes some short-term price fluctuations that our model recognizes as indicative of future price changes. In this case we will want to keep the short-term filtered information.

Sometimes we have no idea where the important information may lie. Worse still, the various periodic features are all jumbled together in an incomprehensible mess in the raw series. Vital information may be hidden under worthless clutter, invisible to the naked eye as well as to most prediction models. This situation is best remedied by using multiple filters to separate the original series into two or more components, each of which can be examined separately without interference from other components. When we are wallowing in ignorance, filtering for information separation can be most useful.

Why Use Wavelets?

Most people would agree that extended repetition in financial series is the exception rather than the rule. Almost any time extended repetition becomes apparent, alert observers capitalize on it and thereby eliminate it. Hence there is a need for filters that extend across relatively short stretches of time.

Probably the best known family of short-term filters is wavelets. The term wavelet comes from the fact that it is a little wave. There is an infinite number of wavelets that can be used for analyzing a series, and nearly this many are in use today. Wavelets have many valuable properties. Some of these properties are the following:

– All of the individual wavelets in a family are derived from the same mother wavelet. Once we know the properties of the mother wavelet, we know the properties of all members of the entire family.

– Wavelets easily lend themselves to shifting along time. If we do things correctly, a feature that appears at some particular time will reveal itself in a wavelet analysis in a consistent way. Its wavelet characteristic will be the same, no matter when the feature occurs. This lets us use history to predict the future.

– Wavelets are inherently based on a meaningful time scale. For skilled users, this is no great advantage because it is not terribly difficult to learn to think in terms of frequencies. But for beginners and casual users, it is much simpler to think in terms of phenomena whose cycle length is a specified number of samples.

For these and other more technical reasons, wavelets make an excellent choice for analyzing financial series.

Why are Morlet Wavelets Valuable?

There is a wide variety of wavelets from which one can choose. For many reasons, the Morlet wavelet is probably the best choice for feature detection in financial series. These reasons include the following:

– Morlet wavelets are naturally robust against shifting a feature in time. Little or no special precautions are needed to ensure that a feature will make itself known in the same way no matter when it occurs. Daubechies wavelets, and in fact all orthogonal wavelets so fond to the signal processing community, present great challenges in ensuring consistency across time.

– A famous mathematical formula called the Heisenberg Uncertainty Principle decrees (roughly) that no filter can, with arbitrary accuracy, simultaneously locate a feature in terms of both its period and its time of appearance. In order to gain more precision in one, the other must be sacrificed. The laws of physics are quite firm here. This principle imposes a bound on how well a wavelet can detect a feature. The Morlet wavelet, for all practical purposes, achieves this bound. In other words, other filters can do no better at simultaneously locating a feature in terms of its period and when it appears. Most other wavelets do worse, and many wavelets and other filters do considerably worse. This is a valuable property.

– Morlet wavelets have a very intuitive nature and definition.

Why use Gabor Filters?

Wavelets are very narrow in their capabilities. They act as bandpass filters only, a given wavelet responds only to periodic variation in the vicinity of its center frequency. Sometimes we want to keep not only the variation around the center frequency, but all variation above or below as well. Lowpass filters pass all information at or below a given cutoff frequency, and highpass filters pass all information at or above a given cutoff frequency. These types of behavior are not available from a strictly implemented wavelet. A good way to generalize wavelet filters to include lowpass and highpass operations is the Morlet’s close relative, Gabor filters.

Gabor filters exactly meet the Heisenberg Uncertainty Limit; there is no other filter that offers a better combination of reliability and the ability to hone in on events in both time and frequency.

For more information about WAVEFIN contact:

Steve Helme
Cornice Research

III. Neural Nets in Astronomy and Spectral Analysis

If you are into astronomy or spectral analysis then you might want to check out the Interesting Reading section of our web page, www.wardsystems.com. There you will find a technical article about how one user used our neural nets to classify stars. This wasn’t just any user; he is the Director of the Monterey Institute for Research in Astronomy, Dr. Wm. Bruce Weaver. In fact, Dr. Weaver previously discovered a new star in the solar system with NeuroShell 2!
IV. Watch What You Say About Neural Nets; They May Be Hiring You For Your
Next Job!

In a previous newsletter, we reported on how two companies are using our software to bring neural nets and genetic algorithms to the domain of Human Resources. Neural nets are actually being used to analyze job applicants! We now have the complete text of the article that was published in PCAI Magazine. Watch for it in the Interesting Reading section of our web page, www.wardsystems.com.

V. Using Neural Nets to Select Stocks for Options Trading

The head of a stock market advisory service a while ago contacted Dr. Andrew Kramer for assistance in quantitative analysis. The service was producing a biweekly report with lists of stocks that were believed to be good candidates for buying naked calls (or writing puts). The list was produced by utilizing only two technical indicators on data from 1200 optionable stocks. The accuracy rate they were achieving was good but not spectacular. Dr. Kramer was asked to improve upon their stock screening process. The signal for each stock was whether that stock had a high probability of at least an 8% move upwards anytime within the succeeding 90 days. Dr. Kramer suggested that 8% moves downwards also be examined, thus creating two signals; bull moves and bear moves.
Daily high, low and closing prices as well as volume were obtained on each stock for the past two years. A group of 16 technical indicators were chosen for testing. These indicators became the input nodes for the neural network analysis that Dr. Kramer was going to conduct. Dr. Kramer now uses the NeuroShell Trader for his personal trading, but at that time he chose the NeuroShell Classifier 2.0 for this particular work. This program’s ease of use and powerful TurboProp 2 algorithm allowed for quick testing and pruning of various input sets. Separate networks were constructed for bull moves and bear moves. The final results on the out-of-sample data (i.e. test data set) are given below:

Bull Move Prediction
Actual Market Bull Move?
Predict Bull? No Yes Total
No 130 164 (62.1%) 294
Yes 119 401 (77.1%) 520
Total 249 565 815

Since the Classifier allows results to be exported, he was able to test whether a filter would have an impact on the results. He chose a priori a filter that selected stocks that had an output probability of 100%. The results were even more impressive:

Bull Move Prediction (filtered stocks)
Actual Market Bull Move?
Predict Bull? No Yes Total
No 16 26 (57.8%) 42
Yes 27 212 (88.7%) 239
Total 43 238 281

Thus someone who used the results of the filtered neural network model to select call option candidates would have had close to a 90% success rate!

The bear move results were similarly striking:

Bear Move Prediction
Actual Market Bear Move?
Predict Bear? No Yes Total
No 248 114 (31.5%) 362
Yes 190 263 (58.1%) 453
Total 438 377 815

There were only two instances where a stock had an output probability of 100%, so he used a threshold of >90%. This resulted in the following:

Bear Move Prediction (filtered stocks)

Actual Market Bear Move?
Predict Bear? No Yes Total
No 51 23 (31.1%) 74
Yes 11 23 (67.6%) 34
Total 62 46 108

Even during a strong bull market someone who selected put option candidates using the final model would have been correct an average of 2 out of 3 times.

Who says you can’t beat the pros!

Dr. Kramer is President of Future Analytics, Inc., a firm specializing in providing data mining and statistical services. He is available for consultation and can be reached at 540-687-3692 or at ramer@futureanalytics.com. Please contact him if you have questions about his techniques.

VI. Warning to Users of Older NeuroShell 2 Releases

There a large number of NeuroShell 2 users who have not upgraded from release 3.0 to release 4.0. We know that release 3.0 was a great release, and we understand your reluctance to pay for the $109 upgrade when everything is working fine. However, we feel that we have to warn you of two things:

1. Release 3.0 was never tested for Y2K compliance, so if your company has Y2K police, this is your perfect excuse to upgrade.

2. Release 3.0 is not a full 32 bit system like release 4.0 is. It has 32 bit training modules, but the other modules are 16 bit modules. Why is this important? Suppose a future version of Microsoft Windows no longer runs 16 bit programs? You’ll just upgrade then you say? Suppose we are no longer selling NeuroShell 2 by then? Sorry, but you’re going to be out of luck! Once we stop selling NeuroShell 2, we will NOT be selling upgrades either. Although we have NO CURRENT PLANS to stop selling this venerable program, its classic technology HAS been improved in our newer programs, and at some point we may have to move on.

The $109 upgrade fee (plus shipping) is the only real reason not to upgrade. The interface is the same on release 4.0, so there is no learning curve. But the $109 is actually a bargain because we have included free all the things that used to be extra cost options! (If your release is older than 3.0, the upgrade prices are a little higher.)

We know that there are many many die-hard NeuroShell 2 users reading this. It is a great product. Nevertheless, as well as upgrading, you may want to consider purchasing the NeuroShell Predictor, NeuroShell Classifier, NeuroShell Trader, or NeuroShell Trader Professional. In fact, if you do this before the end of August, we’ll throw in a free NeuroShell 2 upgrade anyway! Don’t think of it as REPLACING NeuroShell 2; think of it as SUPPLEMENTING NeuroShell 2! The Trader Professional even fires NeuroShell 2 nets.

VII. Secure Order Form Available Online

We now have a secure online order form for your convenience. Just go to www.wardsystems.com or www.neuroshell.com and select Order Info on the side bar. Follow the instructions. Your order will be processed by one business day, and you will be send an emailed confirmation.

This Newsletter was published by Ward Systems Group, Inc.
voice 301-662-7950
fax 301-663-9920

Portions of this newsletter were supplied by MESA, Cornice Research, and Future Analytics and may be covered by copyright of those companies.

Was this article helpful?

Related Articles