September 2004 Newsletter

In this issue:

I. NeuroShell Trader steals technology from NeuroShell 2, NeuroShell Predictor and Classifier

II. How to Integrate Daily Indicators Into Intraday Charts

III. We won’t delete your email if it includes the product name and serial number!!!

IV. A new twist on Fibonacci and Gann Projections

V. AI College back in session

VI. Make sure you continue to get the news

*******************************************************

I. NeuroShell Trader steals technology from NeuroShell 2, NeuroShell Predictor and Classifier

In NeuroShell 2 the algorithm is called “Calibration”. In the NeuroShell Predictor and Classifier (neural method only) it is called “Enhanced Generalization”. In the NeuroShell Trader and DayTrader Professional it is called “Optimal Walk Forwards”. This algorithm does what most of you probably do anyway. It peeks at some evaluation sets as the optimizer is building different nets, then it keeps the one that worked best “out-of-sample” so to speak. That’s what you do if you build a net, then see how it did on new (out-of-sample) data, then build another, see how it did, etc., until you finally keep the one that did the best on the new out-of-sample data.

When we introduce NeuroShell Trader release 5.0 you will find that the optimal walkforwards concept will be more turnkey and easier to use. For now, we know the concept is complex, but what follows is an attempt to explain this concept as it exists today in NeuroShell Trader Professional and DayTrader Professional.

Now the evaluation sets it is peeking at are not the regular walk forward evaluation sets; they are older ones added especially for this purpose. We call them Optimal walk forwards. So that way the usual walk forwards are still really out of sample. The down side is that the addition of the optimal walk forwards means your training data is a lot older. To bypass that protection and use more recent data (and to make the process really what you are probably doing already) you can set no walk forwards, as we described in the tip called “Using Generalize on optimal walk forwards”.

Here’s how the algorithm works. The GA is finding input combinations. Each such combination is used on the training set to train the net. Then that net gets applied to the first optimal wf evaluation, then the net is retrained on the same input combination and applied to the next optimal wf evaluation, etc, until all optimal wf evaluations are finished. The input combination which produced the most profit (or whatever the objective function is) on all the optimal walkforwards as a complete trading sequence is remembered. Then after the GA is totally done that remembered input combination is used to retrain the nets for the regular walkforward evaluations.

Now the difference between “Generalize” on optimal walkforwards and the “normal” use of optimal walkforwards is subtle, because in both cases the inputs that worked best on the optimal walkforwards are the ones that are kept for later. The difference is in what the GA uses as a fitness function to improve itself. In “normal” use the GA is seeking the best performance on optimal walkforwards. In “Generalize” the GA is seeking the best performance on the original first training set. Therefore, the Generalize version is only rated on how well it works out of sample. The normal version is evolved to work well out of sample. In both cases, the out of sample set is compromised, but the normal version compromises it more. But since both compromise it, that is why we added the additional optimal walkforwards and kept the real walkforwards pure.

*******************************************************

II. How to Integrate Daily Indicators Into Intraday Charts

Often our users have created indicators, predictions, etc, in a daily chart that they would like to have access to in an intraday chart. That isn’t hard to do – basically all you have to do is save the daily data as a text file, and then load it into the intraday chart as Other Instrument Data. NeuroShell shows the daily value on every bar of the intraday chart. The following example will illustrate this in more detail for those not familiar with exporting text files.

Let’s suppose that you feel that the 5 day moving average of the Dell daily close is an important input to a Dell model you’d like to build in a 5 minute chart. You realize that your daily moving average will have the SAME value all day long in the intraday chart. Here are the steps you’ll need to take:

1. Create your daily chart and insert the 5 day moving average

2. Rename the 5 day moving average to some simple name that does not contain any commas. This is important because we are going to save the indicator in a comma separated .csv file.

3. Now save the indicator as a .csv file. Go to the Tools menu and select “Export chart/data”. Follow the wizard. Note that we do not export open, high, low, close, and volume – only the indicator we created. Note also that we do not name the file dell.csv, which is the default. Store the files in the folder called “Some folder”.

4. Now create a DELL 5 minute chart.

5. Go to the Tools, “Data Sources” menu and select the “Directories/Files” tab. Use the “Add Directory” button to make the folder called “Some folder” known to NeuroShell.

6. Go to the insert menu and insert “Other Instrument data”. The text file you created will probably be in the category “Miscellaneous Instruments” unless you chose another category when you added the directory called “Some folder”. Select your text file and the Avg 5 day indicator.

The tip above is also on www.ward.net with graphics that go with the example.

*******************************************************

III. We won’t delete your email if it includes the product name and serial number!!!

Like any e-mail address that is published on the web, i.e., support@wardsystems.com, we receive hundreds of junk e-mails every day. To make sure you receive an answer as soon as possible, make sure the subject line of your email includes the product name and product serial number, e.g., NeuroShell DayTrader D45000000. That will get our attention and there’ll be less chance we delete it thinking it is spam!

*******************************************************

IV. A new twist on Fibonacci and Gann Projections

Dennis Peterson’s Fibonacci and Gann Projections article in the October 2004 issue of “Technical Analysis of Stocks and Commodities” magazine describes how to use those indicators to predict future turning points. However NeuroShell Trader Professional has an inexpensive add-on ready to do just that. The Turning Points Add-on includes peak and valley probability indicators.

The peak probability and valley probability indicators use the mean and standard deviation of the percent change in value from peak to valley or from valley to peak. Once these two characteristics of the percent change are known in the lookback window, it is possible to compute the probability that the most recent bar is at a new peak or at a new valley. The Turning Points Add-on algorithm does not just blindly search for the first consecutive valley after a peak (or a first consecutive peak after a valley). It tries to optimize on turning points and locates such peaks and valleys that a human expert would choose by looking at the chart. If you want to learn more, call us at (301) 662-7950.

*******************************************************

V. AI College back in session

If you are a user of NeuroShell Trader Professional or NeuroShell DayTrader Professional, you will now be able to get some advanced training on your own time in your own home or office. Our online course is $670 + shipping of the AI College CD and is taught by our CEO Steve Ward. Visit www.aicollege.com for more details.

*******************************************************

VI. Make sure you continue to get the news

We are still surprised that, many months after we sent out our June 2003 newsletter announcing release 4.0, users are still calling or emailing to ask when release 4.0 will be available. Seems they all changed their email addresses and never notified us of that. They could have been making money with the new features all of this time. There are many others, no doubt, who still don’t know about release 4.0.

After all these years we are also still getting tech support calls from NeuroShell 2 users who don’t know about the AI Trilogy and its components. So they’re stuck trying to impress their bosses with older technology.

Don’t be left behind – tell us when you change email addresses.

*******************************************************

Was this article helpful?

Related Articles