High Probability Trade Setups: Definitions, Filters and Testing

Before describing a setup as high probability, ask what event is supposed to be likely. A target being reached before a stop is different from a trade closing above break-even, and both are different from a strategy earning money over a period. The exit definition determines the outcome being counted.

A reward-to-risk ratio is a payoff relationship, not a probability. A planned 2R target says how large the intended reward is relative to the planned risk. It does not say how often the target will be reached or what the actual average fill will be.

Write the event, the rule and the data source before looking for confirmation. If your sample contains missed orders, flat exits or unknown outcomes, explain how they are classified. Counting only completed winners and losers can conceal opportunities the strategy could not execute.

A moving average, a trend label and a breakout can all be transformations of the same price history. Three bullish descriptions do not automatically supply three independent pieces of evidence. Adding filters also changes when you enter and how many opportunities remain.

Use a baseline setup and one precisely defined proposed filter. Log every baseline opportunity, the filter state known before entry and the eventual result under the declared execution rules. This lets you compare rejected and accepted candidates instead of building a story around the trades you remember.

These are definitions for research, not optimized settings. Keep the strategy specification alongside the sample so a later reviewer can reproduce the classification.

Suppose an invented record contains 100 eligible opportunities, with winners of +2R, losers of −1R and an assumed 0.25R round-trip cost per trade. A filter defined before evaluation accepts 20 of those opportunities. The counts below are invented solely to show the arithmetic; they do not come from a tested market strategy.

The baseline has 40 × 2R − 60 × 1R = 20R before costs. Costs total 25R, leaving −5R. The subset has 14 × 2R − 6 × 1R = 22R before its 5R of costs, leaving 17R. Arithmetic alone does not tell us whether a real filter would repeat that difference.

Use actual average outcomes, not the target printed on a plan. For a sample with winners and losers, gross expectancy in R is win frequency × average winning R minus loss frequency × average losing R. Deduct costs only if those outcomes are not already net; otherwise you count the charges twice.

For an illustrative opening-range test, first name the instrument, official session clock, data feed and a 15-minute opening window. Mark the high and low from that window only. This is one candidate specification, not a claim that 15 minutes is the best duration.

1. After the opening window ends, identify the first completed bar closing above its high. State the bar interval in advance.
2. Model the entry at the next executable bar open, with a declared spread/slippage model. Do not award yourself the earlier breakout price.
3. Place the illustrative invalidation below the opening low plus a declared buffer. Check whether order increments and the monetary risk budget permit the trade.
4. Define the target or time exit before evaluation. When a bar touches both stop and target and event order is unknown, use higher-resolution evidence or mark the result unresolved.
5. Allow only the named number of attempts and the specified entry window. Record every cancellation, missed fill and rejected candidate.

The long example is not a complete short-side or multi-market system. Changing the session, target or entry window creates another version to log. Price leaving a range does not prove that an institution has committed to a direction.

Size from the declared invalidation with the Trading Calculator, while checking the broker's actual contract values and order policy. Then evaluate the hypothesis through the validation workflow; do not use the example as an entry instruction.

SETUP VALIDATION — DEFINE BEFORE TESTING
Instrument, venue, feed, price type:
Timezone, session and bar interval:
Candidate setup and exact level definition:
Eligibility known before entry:
Trigger, order, cancellation and invalidation:
Exit and maximum holding time:
Baseline version:
One proposed filter and its threshold:
Every opportunity logged, including rejected candidates:
Historical development period:
Untouched evaluation period:
Number of alternatives tested:
Fees, spread, financing and slippage model:
Outcomes: wins / losses / flat / missed fills / unknown.
Measures: net R, expectancy, drawdown, frequency and uncertainty.
What would reject the hypothesis:
Forward-test plan:

Record the number of variants tried, not only the survivor. Compare net expectancy, drawdown, opportunity count and practical execution across periods. A filter that removes almost every trade can produce an appealing average while leaving too little evidence to interpret.

In forward simulation, preserve the pre-entry observation and the intended order. Use the journal to distinguish a missing condition from a correctly executed trade that lost. Do not relabel a losing candidate after the chart is complete.

• Bailey et al.: The Probability of Backtest Overfitting

  Primary research on selecting apparent winners from repeated strategy tests; it does not validate confluence or the invented sample.

• SEC Investor.gov: Stop, Stop-Limit and Trailing Stop Orders

  Why an order trigger and the resulting execution price need separate treatment.

No tested win rate is claimed for a chart pattern, filter or opening-range rule. The worked counts and parameters illustrate a research process, not evidence of future returns.