Is it possible to find robust systems?

Hi all,

I wanted to open a thread that is difficult to find on the internet.

I have been studying and programming algorithmic trading for 3 years.
I have solved many of the associated problems: platform, live trading robot, error capture, money management algorithms. I have the ability to operate multi-instrument + multi-timeframe and be able to allocate capital to different systems in real time based on their performance. Likewise, global capital management of accounts with withdrawal policies, MSD control (%), etc.

But ... I have a big problem. I CAN'T FIND SYSTEMS THAT I CONSIDER ROBUST TO BE ABLE TO OPERATE IN REAL.

My initial approach has been mainly through data mining, using programs like StrategyQuantX and BuildAlpha.
My experience is that they are good programs to get the best curve in the past, as they use genetic algorithms to extract that much desired set of rules.
But when I want to analyze the strategies in OOS and with robustness tests (monte carlo, noise test, etc.) 99% fall.

Now I am learning Zorro and I see that it is a good platform to test ideas quickly and to be able to validate them in OOS.

I throw several questions:

1 - Can you get robust automatic systems that work in real life?
2 - Can anyone confirm that you have them and operate with them?
3 - What opinion do you have about builders (data mining) vs trying ideas with logic?
4 - How is your process of finding systems?

I have been very stuck for 6 months and already very frustrated.
I hope someone can help me and give me some real opinion of their systems.

Thank you very much in advance.

1. I'm confident you can, but it can be a long road of testing & evaluating certain concepts, rules, models, etc.
2. Not yet!
3. I believe that machine learning is def the way to go. These algos are way more objective & sophisticated than simple trading rules (please forget technical analysis).
4. Create a large data set, train a model (I use R), while preventing over-fitting as much as possible by applying techniques like cross-validation or the AIC or BIC criterion and out-of-sample test this in Zorro. Store all results in an organized way.

Bonus tip: there are esp 3 important concepts about machine learning that you need to understand: sample size determination, variable-selection methods (i.e. variable importance) and the bias-variance tradeoff.

Grant,

thank you very much for the reply. It's a great help.

And thanks for the machine learning tips, I didn't know about the bias-variance tradeoff (I'm also studying it).