This issue of Futures does not only represent our 500th, it is also the 20-year anniversary of our Technology & Trading feature, almost exclusively written by Murray Ruggiero. Murray has been writing for Futures since 1994. We recently sat down with him to find out what he’s learned from two decades of writing about technology in trading.
Futures Magazine: How did you get involved in trading?
Murray Ruggiero: I have an undergraduate degree in physics astronomy and computer hardware/software. My first job was conducting failure testing of jet engines. I then ended up at Olin Chemical helping out the researchers. That’s when I got my introduction to artificial intelligence (AI). We would use all these databases to try to find relationships among physical properties. We were buying all these products to analyze the databases, and one of the scientists there asked, “Can’t I just plug this into my damn spreadsheet?” That was the genesis of Braincell back in the late 1980s.
Braincell was originally a neural network embedded into a Lotus clone. But when it was reviewed, it was reviewed as a spreadsheet and not as a neural network.
So, we pivoted our concept and made it an add-in to Excel. We were at the Windows 3 rollout in Boston in 1991. It took off and our clients were using it to trade the markets. Because of that, I had to learn the markets to help my clients use Braincell effectively. One of the early uses I found that worked was using neural nets to predict moving averages. I learned early on that you can’t treat the markets as a signal-processing problem. You can’t just data mine relationships and expect to find something that is robust. You must have domain expertise. I committed to learning technical analysis. My goal was to become a technician without using AI.
FM: What was your approach to get that trading education?
MR: A lot of reading. A lot of testing. I made a lot of friends whom I learned from. For example, I got to know George Pruitt over at Futures Truth around this time. Because of Braincell, I had a lot of notoriety—we were in Business Week; we were in The Wall Street Journal two or three times. It was easy to get people in the business to talk to me. Then in December ‘93 or January ‘94 I ended up calling Ginger Szala at Futures magazine and she gave me an assignment, which was basically a smoothing of data using neural nets to compress data, to filter the data to have a zero lag filter.
A lot of the early articles were on neural nets. Most of the people writing about neural nets weren’t combining them with domain expertise. People were expecting too much of them. I knew this approach was doomed to failure and that I would have to incorporate traditional technical analysis to have a long career. I could give it a technology twist, but I knew neural nets on their own didn’t have a long shelf life.
FM: How did you develop more as a technician?
MR: In 1995, Larry Williams hired me as a consultant and I worked for him about three-and-a-half years. Larry would basically have me conduct research projects. Some of it he would keep proprietary and some he would let me disclose. But he was very instrumental in my development. Larry is one of the few people in the business who actually understands the problem solving required to make stuff that works. One of the things that came out of my research for Larry was the adaptive channel breakout concept—setting the channel length to the dominant cycle. Larry told me, “Channel breakout works, but the length has to be right. You figure out how to adapt it. That’s your project.”
That’s how he would leave it. This is one concept that I’ve covered in Futures, and I’ve used it in a couple systems. If you have a 30-day cycle, the market should go up for 15 days and down for 15, ideally speaking. So my underlying concept was you have an n-day high or low where the "n" is approximately the dominant cycle length.
Click on the image above for access to each of the Top 10 Articles.
FM: In terms of analysis, where do you feel you’ve had the most influence?
MR: One of the things I’m most proud about is intermarket analysis. The concept of intermarket divergence is an arbitrage play. The only time you know things are mispriced is when they are moving in the wrong direction. The only time you can tell there is mispricing is when the intermarket and the market you’re trading are not on the same scale. A lot of my early systems are based on this concept. In 1998, I published a system that used utility stocks to trade bonds. You can do it with currencies, you can do it with gold. It’s not just end of day. I have intermarket-based systems that trade intraday. I have a gold system that trade 45-minute bars, but that’s about as short as I can go. For the logistics of my trading and that of my clients, I’m looking for trades that average north of $100 per contract. I’m sure high-frequency traders can make it work on an even shorter time frame.
FM: Can you rely too much on technology?
MR: When I was working for Larry, I used genetic algorithms to create rule templates and evolve trading rules. This was back before genetic algorithms became mainstream. The articles we did for Futures were some of the first on this topic in trading. However, one of the issues with genetic algorithms, which I learned early on, is they can be curve-fitting machines. I had a client who had a system developed by some very high-level people. He called me in early to mid-2002, saying, “I’m down 30%, and I’m going to lose all my clients. Can you look at this system?” He gives me the output from the indicator that he says was developed with “some type of AI modeling.” I dig into it and find that it’s about 90% correlated with a strategy trading today the four previous days. That in and of itself isn’t a bad thing. If you look at 1998, you would have made a fortune trading today minus four days.
So, the system worked [perfect] for two years, but if you analyzed the system, you would have realized that something that correlated was going to be dangerous. Once the market started moving sideways, you would have to pull the plug. One of the big things I learned with all this AI stuff is you don’t want to use AI to make your system. You want to start with a system that works and use AI to improve it. That way, if the AI elements blow up, you still have a tradable system.
FM: Is a high-tech approach right for everybody?
MR: A high-tech approach can give you more juice, but you have to understand it. The biggest problem is not following the system—traders often make the mistake of pulling the plug before the system reaches the max drawdown. Then two months later the system is at a new equity peak, but by then they’re out of the markets. Also, try to develop stuff on your own. I don’t believe in black boxes.
FM: What is your focus now?
MR: I’m moving back toward advanced technologies, particularly what I call “bot technology” in which I create self-adaptive walk-forward trading systems. I have an algorithm that picks the best parameters dynamically in n-dimension space. This isn’t necessarily a new concept, but the hardware—and software specifically designed for that hardware—is now capable of supporting it. Multicore hardware and software is opening up new possibilities. The first time I did similar work was for Larry back in ‘94. I developed an adaptive system that used a lot of walk-forward optimization. There was good news and bad news. The good was it was the best system I had ever seen. The bad news was if he started it at 7 p.m., he got orders by noon the next day. The hardware couldn’t do its job. Those issues have gone away.
For the typical trader, he can now be more effective trading multiple trading systems with a combination of trend following and intermarket analysis. You need to understand the basics to make the technology work, and you have to understand what’s going to make the technology work. There are trading systems that are profitable to marginally profitable now, but most trading systems mask those assumptions. With the relative strength index, for example, we make the assumption that the tops and bottoms are lined up and we’re trading half the dominant cycle. If you’re trading a fixed-length RSI, that’s not true because phases shift.
Now you can plug in cycle analysis and keep a certain percentage of the dominant cycle. That’s going to give you a more robust solution, and it’s adaptive to the market.
Using AI, we can create smart components to manage parts of the system. As computers get faster, this technology can be brought to intraday trading.
Murray A. Ruggiero Jr. is the author of “Cybernetic Trading Strategies” (Wiley). E-mail him at email@example.com.
Top 10 Ruggiero articles
While 1994 doesn’t seem so long ago, the time since encompasses several generations of technology upgrades. Here are the 10 most important articles written by Murray Ruggiero over the years.
- Intermarket analysis is fundamentally sound (April 1998)
- Using correlation analysis to predict trends (February 1996)
- Debunking the drawdown myth (January 2002)
- Nothing like net for intermarket analysis (May 1995)
- Breeding a super trader (January 1997)
- Testing the black box system that feeds you (March 1995)
- Building the wave (April 1996)
- Seasonality trades, a sometime thing (July 1996)
- The money trilogy: Gold, interest rates and the dollar (September 2002)
- Making uncertainty work for you (September 1998)