By Cree Lawson
There is usually another setting you can change, another period you can try or another filter you can add. At some point, I want to know whether the next test will answer a useful question.
That decision is easier when the reason for testing was written down before the results arrived.
An Expert Advisor might be under review for a particular market, trading pace, sizing method or portfolio role. Those requirements give you something concrete to compare with the observed behavior.
Separate the reasons for stopping
One candidate may fall outside your requirements. Another may be impossible to evaluate under the current setup. A third may complete the test and produce a result you don't want to pursue.
Those outcomes need different notes. A run that stopped because history was unavailable is not a measured trading loss. An EA that uses a sizing mode your procedure doesn't support may need another procedure. A completed test can still be unsuitable for your intended use.
Keeping those distinctions makes later research more useful. It also prevents operational problems from becoming unsupported claims about a product's performance.
Give the next test a question
Suppose two tests disagree. “Does the difference come from execution delay?” is a question you can investigate by controlling the other conditions. “Can I make the result better?” leaves much more room to keep searching until something looks attractive.
Record what you intend to change, why, and how you will interpret the result. This doesn't require a complicated research system. A short note saved with the preset and report is a useful start.
If you change the acceptance criteria after seeing a disappointing result, record that change too. You may have a good reason, but the new decision shouldn't silently replace the original one.
Distinguish a pause from a conclusion
You can pause because a required version, report or setup detail is missing. You can stop because the observed behavior falls outside your requirements. You can continue because a specific uncertainty is worth resolving.
I would also consider the amount of time the next step requires. A technically interesting candidate may still be a low priority if it doesn't serve the purpose of the shortlist.
For broader testing ideas, MetaQuotes' guide to evaluating a robot before buying provides a useful starting point. Choose checks relevant to the EA's documented use and explain any departure from that setup.
My final note on a candidate would state the decision, the evidence behind it and what would justify reopening the investigation. “Unresolved” is a legitimate outcome when the required information isn't available.
The next backtest should earn its place by answering a question you can name.
Educational discussion. This article describes a proposed review practice, not measured pass rates or results from a completed EA study. Trading can result in substantial loss.
