How Do You Handle Strategy Optimization Without Overfitting?
This is what I do ...
I design the strategy to be dynamic and self-adaptive as best it can, so as not to require optimisations. That includes the EA analysing the market behaviour (and contract specifications) and adapting accordingly.
And when some optimisation is still required (which is rare), I build the logic into the EA code itself so that it self-optimises incrementally in real-time.
The most I use the Strategy Tester's optimisation functionality is for statistical analysis of ideas I want to test, or debugging code.
To be honest, I dislike the time and effort required to constantly keep optimising EAs for live trading. I did that in the beginning of my trading journey, and I hated it. It made be nervous, always doubting if the chosen parameters would be good enough (or not) for the live trading. And when it did not go well, I never knew if it was "normal" or if I had chosen incorrectly.
So, I started investing in more mathematical and statistical approaches that could self-adapt and for which I could run in the tester and see if it was being successful or not in different conditions and symbols. This gave me the confidence to let the EA run live and not to keep constantly worrying about it.
Instead of investing time and effort in optimisations, I decided to invest that time and effort in designing better statistically based strategies.
Your topic has been moved to the section: Expert Advisors and Automated Trading
Please consider which section is most appropriate — https://www.mql5.com/en/forum/172166/page6#comment_49114893
Do you rely more on walk-forward analysis[?]
No.
Monte Carlo testing[?]
No.
[L]imit the number of parameters you tweak[?]
Yes.
The most important factors in a backtest for me are the quality of the historic data and the amount of the historic data. Coming up profitable on a 35 year backtest with 99 to 100% quality, and only one input optimization, is generally my goal.
I should note that I don't use AI, machine learning, nor adaptive components. I also do intensive manual study to estimate potential profitability of logic in advance of committing to coding a new EA.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
Hey everyone,
I’ve been running multiple backtests lately and noticed how easy it is to get great historical results that completely fall apart in live trading. Curious to know — how do you balance optimization versus keeping your strategy general enough to perform well in real conditions?
Do you rely more on walk-forward analysis, Monte Carlo testing, or just limit the number of parameters you tweak? Would love to hear what’s worked best for you.