I’ve been designing, testing and publishing trading systems since I left the floor of the Chicago Mercantile Exchange (CME) more than 15 years ago. I had my first program, DCB-Bonds published in Futures Truth in 1999. This system is still actively traded and bounces in and out of their top 25 systems and is currently sitting at 26th. My philosophy with system design has always been to start with a fundamental premise and begin testing from there. Occasionally, circumstances change and can render the fundamental premise moot. This was the case for a suite of mechanical programs I published in December of ’05 in Futures Truth.
First of all, the argument between developing a program around a premise versus evolving a trading system to a set of performance metrics through the use of neural nets is beyond the scope of this week’s piece. This week, we’re focusing on the evolution of market access, the effect it has had on program development and finally, understanding the effects of data selection.
I was fortunate enough as a young trader to be involved in the evolution of the electronic market access devices on the floor of the CME. There was little question that the human beings in the trading pits were able to process more trading volume relative to their error percentage than the computers of the day could handle. This eliminated the big guys from beta participation but allowed myself a firsthand look into the future of trading. It became very clear that once the front-end appliances gained efficiencies that the trading floors and pit traders were quickly becoming the richest passengers on a sinking ship.
Trading volume has since shifted to electronic dominance. Currently, more than 98% of all futures volume is traded electronically. More critically to this piece, many markets now trade nearly around the clock. The transition from pit trading to around the clock electronic trading quickly invalidated the premise of the entire DCB Swing Trading suite of systems. The issue was simple. The programs were designed to capture the flush that accompanies a run in the markets. The flush is usually characterized by a build up of orders on one side of the market near the close which would signal quick acceleration the next morning as these trades are offset. The key way to look at this is to think of a market closing through an important level. Traders see this through their nightly work, remember that this is the late 90’s, and plan their strategies for the next day. In this case it would involve exiting a market that has run out of gas.
The premise was based on entering the market just as the support or resistance threshold is violated thus, creating a position inline with the expected market flush. For example, the DCB Swing system would enter a short trade as a market pulled back or, fell through an important technical level near the market’s close. This creates a short position for the trading program as traders offset their long positions. The hiccup came as volume migrated from the pits to the screens. The exit trigger for this system was based on the, “first profitable open.” Therefore, the following day when the markets would open and the rest of the traders offset their long positions, the additional selling pressure pushed the market lower. Our first profitable open would be triggered by the additional selling allowing us to cover the new short position at a small profit with a high winning percentage. The evolution of around the clock trading eliminated the back pressure build up of traders trying to get out as best they can on the following morning’s open.
Futures Truth still tracks this program. I could have it pulled and would have every right in doing so since the fundamental premise it was based on had gone the way of the dinosaur.
Trading is nothing if not, evolutionary. Interestingly, three out of the seven markets submitted in 2005 are still in the top 100 systems tracked. Perhaps even more interesting is the fact that the lean hog market, which is one of the markets tracked and the reason I dusted this off is not one of them.
See our Mechanical Trading Programs
The CME recently changed the trading hours of the meats and eliminated the overnight markets. This has brought the fundamental premise of the DCB Swing program back into focus, at least in this sector. The chart below plots the program’s trades across three types of data. The first is back adjusted day only data. This is the top data stream and you can see that there are random price gaps due to the the evening session not being included. The middle data stream is non-adjusted, all sessions data. This data is the actual contract month’s open, high, low and close. The problem here is with the giant roll gaps that you can see between the July and August contracts as well as the October to December roll. These gaps distort any technical indicator and will wreak havoc with a profit and loss statement on both the positive and negative sides. Finally, we come to back-adjusted all sessions data. This is the data set used by most developers for end of day trading programs. It eliminates the roll gaps and accounts for the market’s full volatility rather than just a few hour’s worth during the day.
Data comparison and its effect on trading system performance.
Now take a look at the program’s equity curves using each of the different data sets.
Day session only, back adjusted futures contract DCB Lean Hog equity curve.
All sessions, unadjusted DCB Lean Hog futures equity curve.
Back adjusted all sessions DCB Lean Hogs equity curve.
Compile your own equity curve with our mechanical trading program.
These charts make the difference tangibly clear. Therefore, the program’s premise must align correctly with the data. First of all, the designer has to know what’s being measured. Secondly, the designer needs to know what to look for in terms of data anomalies. for example, if the program’s average win is $300 per contract but the performance report shows a biggest win of $3,000 it could be that the trade took place during rollover and the offset was executed in a different contract month after a large roll gap has emerged. The same logic works in reverse as well. If the program’s average loss is $300 but, it’s largest loss is $3,000 it may very well have exited in the new contract month and been unfairly penalized by the price gap between contracts.
For our purpose going forward in the meat markets, we’ll have to focus on the new data set that will be created by the new Monday–Friday day only format. the new hours in the meat markets have brought us full circle with this trading program. It was developed based on past data. It was traded through the transition to electronic markets. It was then rendered obsolete by the market’s new hours. Finally, the fundamental premise upon which it was built has returned. This will be fun to watch going forward. we’ll see if we can’t capitalize on the buildup of pressure in the meat markets due to the new day only trading restrictions.