Neil Thalheim has more in-depth experience in the markets than most of us ever will. For the past 20 years he has traded options, futures, and even derivatives for some major players in the markets such as Bankers Trust, AIG Financial Products, and the Union Bank of Switzerland (now part of the Swiss Bank). He managed multi-billion dollar accounts including hedge funds and pension funds.

Neil’s over 20 years of experience was as a “thinker” however. “I got the ideas and turned them over to a “quant” for implementation,” Neil says.

In spring 1998 Neil decided to put his experience to work for himself by opening his own fund, the Thalheim Fund, based out of New York. He is currently focusing only on a few large investors and is not looking to expand at this time.

Neil started using the NeuroShell Trader to build trading systems fairly early on because, after all, he no longer had a large staff of expert “quants” to help him!

Neil started with a few mechanical trading systems and eventually built up a “portfolio” of about 20-30 such systems for different time frames from hours to months out. His portfolio concentrates on stock indexes, primarily the S&P and the NASDAQ. He looks for a 2 to 4% return per month without taking big risks.

Neil was pretty open in his interview with us about how he started building his systems. Initially he was pretty skeptical about the value of neural nets and did not use them at all. His systems identify trends and then take advantage of the trends, using indicators such as the CCI and %R, appropriately “smoothed and filtered.”

We asked Neil about this “filtering” and found out that he actually builds groups of indicators, like MACD, that look at levels of volatility. It takes 2 or 3 out of about seven such indicators to kick in before his filter accepts a trade. (This is much like the panel of experts approach that we often recommend for neural net trading systems.)

There are also momentum-based triggers that Neil’s strategies use once he is in a trend. His systems look for a large move relative to recent history as another signal to get in or out of the market.

That was Phase One for the Thalheim fund. Phase Two began when Neil started using neural nets. Like Newton, he discovered that nets could see better into the future when he let them stand on the shoulders of giants. The giants were in his portfolio of trading systems that were already working well for the fund. Basically, he simply used some of his trading signals (-1, 0, or 1) as inputs into neural nets and found that they did indeed add value!

Of course, the nets needed more than the trading signals. Neil also feeds them extremely simple breakout and momentum indicators, as well as information on global markets, specifically measurements of the institutional flow of money between the equities and fixed income.

There is one more simple but probably very effective input into the Thalheim fund neural nets. Neil and his staff use their own pattern recognition abilities to classify the top 50 stocks in the S&P as bullish, neutral, or bearish. They don't use sophisticated classifiers like the NeuroShell Classifier to do this; they quantify their own eyeballing of the charts!

As for the technical aspects, Neil uses training periods of between 3 and 5 years. He likes between 4 and 6 walk-forward tests of 3 to 6 months each. Generally, he looks for a total two year out-of-sample period. During this period, he looks for a stable, steady profit, not necessarily an optimal one. “I’d rather have a consistent smaller profit than an inconsistent larger one,” Neil explains. “I’m looking to nibble as much off my markets as I can with the least amount of drawdown.” The trading strategy he uses with the nets is just a simple threshold analysis.

The Thalheim funds credits the NeuroShell Trader, and especially the neural nets, with raising its percent of profitable trades from around 55% to about 68 or 69%! It is also credited with reducing drawdown by shutting down systems when questionable trades would have occurred. There are less trades now, but they are substantially better.

An unexpected benefit of the neural nets has recently emerged by indicating better trading systems in the portfolio. If the nets find an unusually high contribution for certain indictors, then those indicators usually make pretty good trading systems. They have worked very well in cyclic markets, and have produced trading systems that the Fund would not have otherwise used.

So what is Phase Three for the Thalheim Fund? The NeuroShell Trader Professional. They already have been utilizing multiple copies to do their work quicker and easier. Finding better indicators will improve both the neural nets and the trading system portfolio.