In this issue:
I. Opinions of Steve Ward, CEO
II. We get letters
III. Neural network calibration
IV. Christmas and New Years holiday closures
V. Request from our sales department
VI. Make sure you continue to get the news
*******************************************************
I. Opinions of Steve Ward, CEO
Item 1. There’s still time to vote for us if you want to help us out. A vote for us in Technical Analysis of Stocks and Commodities Magazine will go a long way toward convincing new prospective buyers to give us a try. If you are a TASC subscriber, please go their website now and vote before it is too late. Here’s how to vote:
Get your copy of Stocks & Commodities magazine and visit www.Traders.com. Look in the upper left hand corner of the web site for the 2004 Readers’ Choice Awards and click on the Vote Now! link.
On the page that is displayed enter your subscriber ID from your magazine address label (look at their example to see which numbers on the label are the subscriber ID). After you are logged in, look at page 3 of the ballot for the Artificial Intelligence Software (Expert, Neural) category. Be sure to mark you ballot for NeuroShell Trader (Ward Systems Group Inc.) Do the same on page 2 of the ballot for the Standalone Analytical Software, $500 and more category. Look for NeuroShell Trader Professional.
Item 2. This month I have written an article in this issue called “Neural network calibration” which I hope you will read late one night to put you to sleep. I thought of it because I was talking to one of our users about some other model building software – for trading as it happens, but the calibration article isn’t about trading at all.
Anyway, it seems this other software builds many models and then displays the results of them on a subsequent period of time. You are supposed to see which ones work best on that subsequent period of time and then trade those best ones.
Now those of you familiar with www.ward.net will recognize the situation as the one described in the tip called “Major modeling caution”, but that isn’t my point. The point is calibration is an automated way for you to find the model that works best on “out of sample” evaluation data, not the one that worked best on the “in sample” training data. In NeuroShell Trader Pro, we even protect you from making the Major Modeling Caution error by keeping the evaluation data pure. However, since so many other companies don’t even realize what the error is, and so many of their users believe in the wonderful results they see, I’m wondering why we chose to take the high road and look inferior in the process!
Item 3. Next week I’ll be teaching a three day AI Trilogy class in Texas. You didn’t get an invitation because it is a custom class set up for the staff of one company. I’ve done a few of these in my time here at Ward Systems Group, and although they don’t make as much money as classes with open enrollment, they are a little easier for all concerned.
They are easier for me because I use the company’s classroom facilities including their computer and video projector. I don’t have to make hotel arrangements (besides my own room), and I don’t have to worry about how many people are coming to meet hotel quotas.
They are easier for the company, because the employees don’t have to travel, and the class concentrates on areas of interest specific to their company. I work with the company during class hours on real problems the company has in the comfort and privacy of their own classroom. Also, I charge a fixed fee, not a fee based on how many students there are, so it is cheaper to educate up to 20 people.
I’ll consider doing a couple more custom classes in 2004, so if you are interested, call me after the first of the year.
Item 4. The other day a user called and said he saw a company profiled on (he thinks) McNeil Lehrer. He recognized NeuroShell Trader on the company’s computer screens. If this was your company, please tell us – we’d like to brag about you.
Thanks for listening.
*******************************************************
II. We get letters
“So far this year, my combination trades (swing trades) have averaged about 69% per annum. That’s better than even the high-flying NASDAQ, and I am invested in the market only about 33% of the time (that is, the strategy is RISK-AVERSE). I started trading with net $5,000 positions a year ago. Now I am at net $25,000 positions. The goal next year is to be at net $65,000 to $80,000 positions (it might not work, but net $25,000 was my goal for this year, and I hit it). And that’s with no futures or options. NeuroShell Trader is still the backbone of my trading program.”
David Buffalo
*******************************************************
III. Neural network calibration
This subject will be of interest to you no matter which of our neural net products you own: NeuroShell 2, NeuroShell Trader, NeuroShell Predictor, or NeuroShell Classifier. Those of you with NeuroShell 2 are probably already pretty familiar with it by the name “calibration”. If you own NeuroShell Predictor or NeuroShell Classifier, you’ve probably at least seen it under the name “Enhanced Generalization” (we are going to concentrate on the neural method). However, those with NeuroShell Trader probably don’t know much about it at all even if you are aware that it goes by the name “Generalize on optimal walk forwards”.
Here’s the idea of calibration. You can easily over-train on the training set, especially if you use too many inputs, too many hidden neurons, or too few training patterns. The only way to know if you did or not is to apply the nets to “out-of-sample” evaluation data – data that isn’t in the training set. If it does well there you probably didn’t over-fit.
So calibration tries the evaluation set every time training comes up with a new net. The net that is eventually kept is not the one that did best on the training set, but the one that did best on the evaluation set.
In NeuroShell 2 things stop there.
In NeuroShell Predictor and Classifier, things work a little differently. Enhanced Generalization loosens the training after training is finished until the net works best on the evaluation set – a process that amounts to about the same thing.
But here’s the rub – the evaluation set is now no longer “out of sample”, so you aren’t really, really sure whether you calibrated well, or just over-fit on the evaluation set! (For more information on this see the tip on www.ward.net called “Major modeling Caution”.) Therefore, in NeuroShell Trader we do the calibration on yet another evaluation set to preserve the integrity of the real evaluation set.
*******************************************************
IV. Christmas and New Years holiday closures
December 25th is Christmas, so we’ll be closed that day. However, like most companies, we’ll be giving our employees December 26th off as well. Please note that so you don’t get frustrated trying to reach us those days.
Of course, we’ll be closed New Years day as well.
*******************************************************
V. Request from our sales department
As you know we are closed for technical support from 12 to 1PM EST. The reason for that is that even though we go to lunch on shifts, we are still short staffed during the “lunch hour”. If we tie up our personnel doing tech support, we will miss some vital sales calls. Therefore we ask everyone to please respect that and don’t try to get help during that hour. Remember, we pay our support staff from new sales, not by charging for technical support. We hope that you appreciate our free support.
*******************************************************
VI. Make sure you continue to get the news
We are still surprised that, several months after we sent out our June 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.
*******************************************************