Stop levels, chosen by grid search.

Two people can trade the exact same signal and land on opposite results, because of one decision neither of them thought much about: where to put the stop.

Why placement matters

The entry gets most of the attention, but the exit does most of the work. A signal with real predictive power can still lose money if the stop and target around it are wrong. The same pattern, traded with two different stop distances, can produce a strategy that works and one that does not.

The stop-loss paradox

A tight stop protects you from a large loss, but it also gets clipped by ordinary noise, turning trades that would have worked into small losses. A wide stop survives that noise, but when it does get hit, it costs more, and a run of those losses is hard to sit through. Neither extreme is correct on its own. The right distance depends on the pattern and the market, and finding it takes data, not a rule of thumb.

Three ways to place a stop

• Fixed. The same distance every time, regardless of what the market is doing.
• ATR-based. A multiple of the market's recent range, so the stop widens in volatile conditions and tightens in calm ones.
• Structure-based. Anchored to a level on the chart, like a recent high or low, rather than a fixed distance or a volatility multiple.

None of the three is automatically right. What matters is testing the choice against the pattern rather than assuming it.

Grid search, with guardrails

Once a pattern has a real edge, grid search tests a range of stop and target combinations against it and reports how each one performed. Instead of picking one distance and hoping, you see the whole landscape: where the combination works, where it does not, and where the result is fragile rather than solid. Every combination tested counts toward the same guardrails that protect any other test: it is tracked, and the bar for calling a result meaningful rises as more combinations are tried.

Edge first, then optimize

Grid search only means something once the underlying pattern has already cleared a real test. Optimizing stop and target levels on a pattern with no edge just finds the combination that happened to look best on that specific history, which is a coincidence dressed up as a result. The order matters: validate the signal with an event study or a backtest first, and only then search for the stop and target that capture it.

Checking it holds

A walk-forward check splits the history into pieces, finds the best stop and target on one piece, and checks whether that same choice still works on the next piece. If the best distance keeps changing from one slice of history to the next, the result is fragile. If a similar distance keeps working across slices, that is a stronger sign the choice will hold going forward rather than just fitting the past.

Common mistakes

The two most common errors both come from skipping a step. One is optimizing before the edge is validated, which produces a number that looks good and means nothing. The other is picking a single best-looking combination from a grid without checking whether nearby combinations also work. A result that only holds at one exact distance, and falls apart a step away in either direction, is not a stop level. It is noise that happened to line up once.

How Validation returns exact levels

Validation runs the grid search only after a pattern has cleared its statistical test, then reports the full set of stop and target combinations tried, with the guardrails applied to every one. When a combination holds up, it returns exact entry, stop, and target levels on current prices, along with the audit trail behind them, so the number you get is traceable back to the test that produced it.

Questions people ask

What is the best way to place a stop loss?

There is no single best way. Fixed, ATR-based, and structure-based stops each behave differently, and the right choice depends on testing the option against the specific pattern rather than assuming one is always correct.

Should I optimize stops before or after finding an edge?

Always after. Optimizing stop and target levels on a pattern that has no real edge just finds a combination that happened to look good on that history, which is a coincidence, not a result.

What is the stop-loss paradox?

A tight stop avoids large losses but gets clipped by normal noise, turning good trades into small losses. A wide stop survives that noise but costs more when it is hit. The right distance is a data question, not a rule of thumb.

What is grid search for stop and target levels?

Grid search tests a range of stop and target combinations against a validated pattern and reports how each one performed, so you see the whole landscape instead of guessing at one distance.

How do I know a stop level will hold up going forward?

Check whether nearby combinations also work, not just the single best one, and check whether the same distance keeps performing across different slices of history through a walk-forward check.