After screening and choosing his issues appropriately, Scott builds neural models for his entry and exit rules, supplemented by occasional manual intervention. His favorite inputs are Aroon indicators (see our Advanced Indicator Set 1 add-on). He also uses the Kaufman adaptive moving averages, from which he makes his own indicators that are similar to Bollinger Bands. "My outputs are a linear regression of short and long term crossovers of adaptive moving averages, he added, "but in some of my models I just predict the low, having found the high unreliable for predicting. I also use specific models for IPO’s, and companies with strong reactions to earnings announcements. For example, with IPO’s I look for the bottoming, such as MSO last April and May, and then the subsequent $18-$28 swing range established in the last 8 months. This type of stock action makes for an easily patterned trading system. With earnings, it can be as simple as Ford, whose earnings reactions for the last couple of years produced the exact opposite stock price reaction. Or it may be as complex as EBAY’s reactions.”

When we asked him about training set and evaluation set sizes, Scott prefaced his answers by stating that he found that the old rules didn't apply after 1999, and they changed again after March 2000. He now uses around 6 months of daily bars or about two weeks of smaller bars, with about three walk forward tests of similar size. Before March 2000, he used even smaller size sets. He retrains every two to three weeks. “There is no fixed timing,” he added. “If the market is volatile and seemingly unpredictable with your current nets, you have to make adjustments to capture what is happening locally. In a ranging market one can lose sight of whether an issue is being accumulated or sold off. And you can lose sight of the stock’s general price direction. If you have a valid macro view, then it’s only a question of selecting acceptable entry points and establishing reasonable stop losses.”

In addition to his own trading, Scott advises private funds and companies. (Sorry, but he is not taking new business at this time!) His passion, however, is Fundpilot.com, where he will provide free (so far) entry and exit signals for your very own mutual fund. The system is scheduled to go live in a few weeks. These signals are based on "technical indicators extracted from my nets", a proprietary method on which he would not elaborate. He finds his signals work well for about 70% of the funds he's looked at so far.

We asked Scott what advice he had for the rest of us. "Having a powerful tool like the NeuroShell Trader is not the whole solution," he advised. "There is no Holy Grail of trading. You also need a keen awareness of the changes in the market, such as trading volume, sector rotation, and general economic forces. You have to continually tune your tool to be in concert with current conditions.”

“You also need to accept that your models are a reflection of your own perceptions, and your ability to correlate events with opportunity,” he added. “There is nothing wrong with standing back from trading and taking time to reassess what is happening when you suffer losses. The one thing I have discovered I do differently from most is that I accept that whatever model I am using right now has a limited life, and will eventually fail. Fortunately, with all the indicators, tools and capabilities of NeuroShell Trader, there is no limit to what I am able to explore and tune."