When to stop optimizing?

March 30, 2025 by
When to stop optimizing?
Quantalpha Algorithms
| No comments yet

Question:
The current combination of my optimization in MT6 strategy tester is around 10,000+. It has already 300+ optimization results and all of those results are losing and not giving good performance. Can I conclude that all the remaining in the 10,000+ is not good as well?

Answer:
Not necessarily. Just because the first 300+ optimization results are bad doesn't mean all 10,000+ combinations will also be bad. Here’s why:

1. The Optimization Process is Non-Linear

  • The Fast Genetic Algorithm (GA) in MT5 does not test all 10,000+ combinations. Instead, it uses evolutionary selection—eliminating weak settings and focusing on better-performing ones.
  • If the first 300+ results are bad, it might mean the algorithm is still exploring bad areas of the parameter space, but it could still find good ones later.

2. The Search Space Might Be Too Large

  • If you have too many parameters with wide ranges, the strategy tester could be testing a lot of weak combinations first.
  • A large number of bad results at the start might mean that good ones are in unexplored regions.

3. Early Stopping Can Be Misleading

  • If you stop optimization too early, you might miss out on better configurations that would appear later in the optimization cycle.


What You Can Do Instead:

Check If Some Parameters Are Causing the Losses

  • Look at the worst-performing input values and check if they are consistently bad.
  • If certain parameter values (e.g., stop loss too tight, wrong RSI settings) are always producing losses, consider adjusting their range or removing them.

Use a More Focused Optimization

  • Instead of full-range optimization, try narrowing down the ranges to more realistic values based on your trading logic.

Run a Small Sample First

  • Instead of optimizing all 10,000+, try testing a smaller subset (e.g., 1,000) with different settings.
  • If a smaller batch still shows no good results, it may indicate a deeper problem with your EA logic.

Check Strategy Logic & Market Conditions

  • If all results are negative, it might indicate your EA logic needs adjusting (e.g., poor entry signals, bad risk management).
  • Consider testing your EA on different market conditions to see if it performs better in a specific environment.

Conclusion:

No, you cannot conclude yet that all remaining 10,000+ are bad. But if after a significant portion (e.g., 1,000+ tests) all results are still bad, it’s a red flag that something is wrong with your strategy rather than just the parameters.

Would you like help refining your optimization ranges or improving your EA logic? 🚀


Share this post
Sign in to leave a comment