The Striker Report

John F. Gallwas: How do you define "Artificial Intelligence" as it is currently being used in the financial markets?

David Register: To me, artificial intelligence means software that can adapt and change its decision-making based on the data that it is processing. This is in contrast to algorithmic and formulaic approaches that always process the incoming data in the same manner. Some of the major ways to implement artificial intelligence are with genetic algorithms, neural networks, machine vision, and expert systems. In my work, I use my training in expert systems and advanced machine classification techniques to develop trading systems.

John Gallwas: As "Artificial Intelligence" is the backbone of your current trading systems, tell us your background and how it led you to become a system developer?

David Register: My training and experience in statistics, artificial intelligence, and technical analysis of the financial markets has uniquely prepared me for trading systems development. I obtained a Bachelors Degree in Physics in 1969 and a Masters in Statistics in 1974 before starting work as a statistician with a NASA contractor, Lockheed Engineering and Management Services Company.

I first started trading futures in 1978 and learned a number of strategies that did not work in the markets. In 1985, I started my work with artificial intelligence and Expert Systems. (An Expert System is a software program that mimics the decision making of an expert or a group of experts.) In 1991, I left Lockheed to work full-time for Robert Prechter’s financial advisory company, Elliott Wave International (EWI). While there, I added features, upgrades and improvements to the software that automates the Elliott Wave Principle.

In 1999, I left EWI for a few years to research and develop a variety of successful indicators and strategies specifically related to futures trading. These tools and methods helped to form the basis of our artificially intelligent, mechanical expert trading system, ExitPoints.

John Gallwas: How do your "Epic", "Score", and "Starter" trading systems differ?

David Register: All three day trading systems are more similar than different. For each system, entry orders are placed each evening to take advantage of overnight price moves. All trades are exited at the end of the session so no positions are held over. This means smaller day trading margins apply. All entries are based on volatility breakouts.

The SCORE and STARTER differ only in the number of markets traded. The SCORE system trades twenty of the more popular futures markets. The twenty are from seven different market sectors and offer considerable diversification. The STARTER system trades eleven of the same twenty markets that SCORE trades. Other than the fewer number of markets traded, STARTER orders are identical to SCORE.

The EPIC system trades fewer markets, different indicators and uses a smaller volatility measure for determining breakouts and entries. This results in the EPIC system trading much more frequently than STARTER or SCORE.

It trades eight highly liquid futures markets. Each market was selected because it typically trades more than 100,000 contracts per day.

John Gallwas: I understand you have developed a new and different approach to back testing your systems. How does it differ?

David Register: In developing a forecasting or trading system, the traditional approach is to "train" a model by optimizing a set of parameters to best fit the model to the price data. This "In-Sample" data set will naturally have an artificially high performance that cannot be used to estimate future market performance since results will be better than can be expected.

At ExitPoints, we do not approach model development in this manner. Instead of optimizing a set of parameters for a model, we use a trade memory approach. From a training data set we build a database (warehouse) of trades. Simulations on this training data set would also appear better than expected for future years, except we simulate "Out-of Sample" forward testing by programming the simulation to temporarily "forget" any future trades for the markets.

Using this technique, ExitPoints can simulate Out-of-Sample forward testing on large datasets of dozens of futures markets over decades of data and provide a more realistic expectation of future trading system performance than traditional approaches.

John Gallwas: Do you personally trade your trading systems?

David Register: We always trade our systems before and after we release them to the public. At the moment, I am only trading one of our unreleased products, but, collectively, the principals at ExitPoints are trading all of our systems.

John Gallwas: Do your systems require updating and if so is it scheduled or event driven?

David Register: No external updates are ever performed on our system. It is self-learning.

John Gallwas: Is there anything in "what's new" you can share with us now?

David Register: Most of our efforts are focused on plans for a swing and position system at some point in the future.