Creating robust trading strategies can be a difficult task, sometimes taking months or even years to generate something you find acceptable.
Even then, once you start trading it live there is no guarantee it’ll work in the future.
With strategy creation being such an involved process at times, how would you like it if you could just tell the computer the results you wanted and let it figure out the trading rules? Is it actually possible to create robust trading strategies that way?
In this episode Michael Bryant from Adaptrade talks to us about automatic code generation, methods to exploit trade dependency and techniques to trade the equity curve.
Michael has been trading the markets since 1994, providing trading systems for the futures markets and even managing money as a CTA.
He is founder of Adaptrade, a company which provides innovative software tools for individual and professional traders.
In this episode we discuss
- The traditional approach to creating trading systems and issues caused by this approach
- Potential areas of improvement in traditional approaches to system development
- Evolution of the strategy creation process
- Genetic programming and optimisation and it’s use in trading strategy creation
- The advantages of automatic code generation
- Measuring and reducing over-fitting when using genetic optimisation techniques
- Addressing concerns with removing human logic from the strategy development process
- Degrees of freedom and the impact if can have on strategy results
- Trade dependency, how to detect it and methods to exploit it
- Trading the equity curve based on trade dependency and trading style
- Which stage of the strategy creation process to include position sizing
- Common position sizing mistakes traders make
Show your support
While you’re listening, it would be great if you could show your support by leaving a brief review of the podcast in iTunes. This really helps to spread the word and reach even more listeners!
Resources mentioned in this episode
- Mike can be followed on his website Adaptrade.com, where you can download free trials of his Builder and Market System Analyser software.
- If you’d like to know more about Trade Dependency, check out this article, it explains how to measure it and possible ways to exploit it.
- Here is an article that explains trading the Equity curve
- Recommended books:
 |
 |
 |
Quotes
Top tips from this episode
Here are few of my favourite takeaways from the chat with Mike this week:
- Trade dependency – This is a concept that some may not consider when analysing their strategy results but Michael explained some possible methods of exploiting it, so if you’d like to know more information, including how to measure it and several ways it could be exploited at significant levels, check the resources section just above, I have a link to more information there.
- Equity curve trading – We’ve covered that a little in previous podcast episodes but it’s always interesting to hear some of the different ways this can be accomplished, especially when it’s based on trade dependency, if you’d like to know more about it I’ve included a link to more information in the resources section above.
Got A Question, Topic or Guest you want to see on the Podcast?
Do you have a specific question, topic or guest you’d like to see on a future podcast episode?
Click here and submit your suggestion for a chance to have it featured on an upcoming podcast episode.
Subscribe to Better System Trader and never miss another episode!
 |
 |
 |
 |
Please support the podcast by giving an honest Rating/Review for the show on iTunes!
Episode Released:
13 December 2015
110 – George Costanza Tip for Traders
005 – Kevin Davey
045 – Entry Techniques with Andrea Unger
062 – Mean Reversion strategies with PJ Sutherland
Are your backtest results fooling you?
Hello there – I used that adaptrade software for a year and got very frustrated with. System after system failed in forward trading even after passing out of sample tests. Most systems produced by the software had weird logic and seemed over-fitted to data. I have contacted some other users and most had same experience. I have concluded that this method of producing strategies is problematic because of the problems inherent in testing thousands of different variants. One of the quants in Quantocracy has a good article about this over here: http://eranraviv.com/sample-data-snooping/
According to the analysis by Eran the data-snooping problem is real and cannot be solved by cross-validation because it is inherent in the process. Traders ought to be extremely careful with any genetic programming hype and related products. Could be that one has higher odds winning the casino jackpot than having one of those softwares produce a profitable system forward.
Hi Bill, thanks for the feedback, appreciate you taking the time to post a comment.
I passed this onto Mike and he responded with the following:
———————————-
Adaptrade Builder is a tool for building trading strategies. Like any tool or any software, there’s a learning curve involved, and the software will only generate strategies as good as the specifications that go into them. Just like with manual methods of generating strategies, it’s possible to generate over-fit strategies if the user doesn’t take full advantage of the methods built into the software to prevent that. However, unlike manual methods of strategy development, Builder incorporates specific techniques and approaches that help to minimize over-fit strategies and produce very high quality strategies. The software is designed to not only accelerate the strategy building process but improve it by incorporating these technique directly into the process. Over-fitting is rarely a problem if the recently added rules to monitor the build process are employed. Builder also includes a significance test that takes the data mining bias into account to help determine if the selected strategy benefitted unduly from good luck. Unless the user tries very hard to “cheat”, the program does not data snoop.
Genetic programming is simply the optimization method used for the build process in Builder. There’s nothing inherent in genetic programming that leads to over-fit strategies. Its primary benefit is that it generates the strategy code. However, other optimization methods could potentially be used to achieve similar results.
While it’s always disappointing to hear that a customer did not have success with the program, I have many customers who have been using Builder successfully for years, including professional traders and money managers. Nonetheless, I welcome all feedback about the program and try to incorporate as much as possible into future versions. Like many software programs, Builder is under continuous development, and each release adds further functionality and improvements.
– Michael Bryant
———————————-
All the best,
Andrew.
It is difficult or even impossible to find a profitable system with something like adaptrade. I actually tried and I could not. I tried Monte Carlo, WFA, validating on unseen data. Result was random. Same random results with a similar system from StrategyQuant. Spent time I will never recover.
Hi sal, thanks for the comment. I’ve passed it onto Mike for a response, here it is:
“When sufficient data is supplied, metrics relating to strategy robustness are employed, and the features of the program are utilized to avoid over-fitting, the results with Adaptrade Builder are anything but random. Of course, as with any strategy building approach, not all strategies will pass the validation tests. One of the benefits of a quantitative approach, as in Builder, is that it allows you to quickly weed out the strategies that are not worth pursuing and avoid trading strategies with a low likelihood of success.”
Cheers,
Andrew.
I think that was one of the weakest episodes of this show. In general I love this show, and you’re doing a great work, but this was kind of a waste of time.
Hey Michalis, what type of stuff are you looking to hear?
Cheers,
Andrew.
To be honest, as a newbie I’m not totally sure. To give you an idea though, I loved these episodes : Ernest Chan, Kevin Davey, Michael Himmel, Larry Williams.
In any case, keep it up, I love this podcast.
Thanks Michalis, good to know.
Cheers,
Andrew.
The question I would have loved to hear would have been, ‘so, now you have this software which is capable of generating profitable trading systems, why don’t you just trade the systems and make money? Instead of dealing with all the headaches that go with selling software?’.
Hey craig, great question.
Mike did mention somewhere around the start of the interview why he now focusses on software, I think it may have been in the first 5 or 10 minutes.
Cheers,
Andrew.
I’ve listened again, he describes a gradual move from trading to software development without offering any more reason than software is more satisfying. The reason, one has to infer, that trading was unsatisfying is that he likely had no edge. People who trade using these techniques keep their cards very close to their chest, nobody wants to away the goose which lays the golden eggs. That is not to say there is no value in the ideas of GP etc, but like all things it’s going to require some current domain traction to get meaningful results. Looking at the screenshots of his product, it seems to be fitting various TA indicators and their look-back windows, which is probably not going to cut it these days.
Hey Craig, thanks for the response.
I can’t really comment any further on the transition from trading to software as all I know was in the interview but I do understand your point.
I guess it’s like anything in trading, you need to do you own due diligence to determine if something is right for you.
Cheers,
Andrew.