Known levels · matched regimes · equal costs

NQ Pullbacks vs Breakouts: A Fair Forward-Only Test

By Kyle ParrottUpdated August 26, 2026

A chart makes the best pullback and the true breakout obvious after the move is complete. At the decision time, neither label may exist yet. A fair NQ comparison must define both entries from levels known before the order, match the market states they face and charge the same execution reality.

Define entries Write retirement

Levels: Known before trigger
Samples: Regime matched
Costs: Same standard
Winner in advance: None

Define the signal at its true availability time

Create Complete Pullback and Breakout Rules

Both rules need the same data clock, order clock and evidence standard. Use completed observations through t to create a candidate. Apply computation, decision and submission latency before determining the first price the strategy could receive.

Pullback candidate

Counter-move inside a prior state

Define an up or down state from trailing, completed data.
Measure a counter-state retracement inside fixed depth and duration bounds.
Require a resumption trigger available now; do not name it a successful pullback yet.
Enter under the declared order rule or record no fill.

Breakout candidate

Move beyond a prior-known boundary

Define the boundary using data completed before the trigger interval.
Require a fixed executable extension, dwell or confirmation rule.
Apply the same source, computation and order latency.
Enter under the declared order rule or record no fill.

Field	Pullback rule	Breakout rule	Fairness requirement
Reference	Prior state and retracement anchor	Prior-known high/low or range	Both available before trigger
Trigger	Resumption evidence	Extension/dwell evidence	Machine-readable and timestamped
Order	Side, type, quantity, time-in-force	Same fields	No optimistic fill convention for either
Invalidation	Price/state/time rule	Price/state/time rule	Fixed before outcome
No-trade	Missing state, cost or risk gate	Missing level, cost or risk gate	Count every rejection

Centered pivots cannot be live references.

If a swing high or low requires later bars for confirmation, its label-availability time is later than the pivot. Use it only as a retrospective outcome or delay the decision until confirmation exists.

Strategy labels can select different markets

Freeze Comparable Pre-Entry State Variables

A breakout rule may fire more often in high-volatility expansion, while a pullback rule may select established directional states. Comparing their raw averages can compare regimes rather than entry logic. Capture pre-entry covariates for matching and stratification.

Path

Prior direction and efficiency

Trailing return, range, directional efficiency, age of state and distance from prior-known extremes.

Volatility

Scale and jump state

Pre-entry realized variation, range percentile and scheduled-event flag. Never use full-day data at an intraday decision.

Liquidity

Spread and capacity

Decision-time spread, depth for the order size, replenishment state and recent impact.

Calendar

Session and contract

UTC window, cash state, macro event, holiday, dated contract and roll migration.

Use the current CME schedule and official BLS, BEA and Federal Reserve calendars. These sources establish timing and market availability, not a strategy outcome.

Compare overlap, then disclose what does not overlap

Build Matched Samples Without Erasing Strategy Identity

Estimate the probability of each signal from pre-entry variables on the training period, or use exact/coarsened matching on prespecified state bins. Compare only observations with credible overlap. Report unmatched signals separately because excluding them changes the population being studied.

Define unit

Choose independent decision opportunities and a rule for overlapping signals.

Record state

Freeze all matching variables before either order can arrive.

Check overlap

Reject comparisons where one strategy has no comparable state in the other sample.

Match/weight

Apply the frozen training method without outcome-aware edits.

Report coverage

Show eligible, matched, unmatched, rejected and missing observations.

Two estimands, two honest answers

Strategy-population result: each rule trades every eligible signal under its own definition. This answers how the full frozen systems performed, but regime mix can differ. Overlap result: compare rules only in states supported by both samples. This improves like-for-like interpretation but applies to a narrower population. Report both rather than letting one replace the other.

Do not match on variables measured after the trigger, on realized fill quality or on future path. Matching can reduce observed imbalance; it does not prove that all confounding is removed.

Charge both strategies the same execution reality

Use a Shared Cost and Risk Model

Apply identical conventions for source latency, decision delay, order acknowledgement, bid/ask side, visible depth, partial fills, slippage stress and dated fees. Equal contract count is not equal risk if stop distances differ.

Comparable net result per risk unitR_net = (fill-to-fill P&L − reconciled actual transaction-fee ledger counted once) / authorized dollars at riskActual fills already contain realized slippage. Reconcile broker commission, exchange, clearing, regulatory and other charges into one non-overlapping ledger. Or use benchmark P&L minus a modeled execution package; never subtract modeled slippage from a fill-based result.

Cost/risk item	Shared standard	Bias if omitted
Marketable entry	Decision book plus latency and size-conditioned sweep	Rewards fast breakout triggers at unavailable prices
Passive entry	Queue/fill model plus adverse selection	Counts touches as free fills
Stop	Executable trigger behavior and gap-through	Caps losses unrealistically
Time in market	Report duration and overnight exposure	Hides unequal risk occupation
Capacity	Same order-size tiers and depth rules	Compares small-order fiction with scalable execution

Lock every choice before the deciding sample

Use a Chronological Holdout and Full Scorecard

Training data may define states, thresholds, matching and costs. Freeze the systems, then run one later period without retuning. The scorecard should show coverage, fill outcomes, net economics, uncertainty and tail behavior.

Score family	Required outputs	Reason
Coverage	Eligible, signaled, matched, unmatched, rejected, missing	Shows what the result represents
Execution	Fill rate, partials, delay, shortfall, capacity by size	Tests attainability
Economics	Gross/net P&L, risk units, drawdown, tail loss	Separates pattern from cost
Stability	Subperiod, session, event, volatility and roll strata	Exposes conditional dependence
Uncertainty	Counts and intervals under declared dependence treatment	Prevents false precision

A comparison must be allowed to end with neither

Precommit Retirement Rules

Retire a rule if its labels leak future information, its sample lacks overlap, its economics fail after costs, its forward calibration decays or current states leave training support. Do not automatically switch to the other strategy; it must pass independently.

Definition failure

Entry needs a future-confirmed swing, day type or path label.

Invalidate all affected results

Comparison failure

Matched overlap is too small or unstable for the planned claim.

Report incomparable

Economic failure

Neither frozen rule clears the forward net threshold.

Retire both

All features and levels are available before the simulated order.
Both strategies use the same data, latency and fee evidence.
Full-population and overlap results are separately reported.
Parameter searches and losing variants remain in the research ledger.
The untouched holdout is not reused for repair.
Neither strategy is presumed superior from its name or chart appearance.

Research status as of August 26, 2026

No original NQ pullback or breakout win rate, profitability estimate or superiority finding is reported. This article defines a fair comparative protocol and retirement policy.

Sources, methods and editorial disclosure — reviewed August 28, 2026

CME E-mini Nasdaq-100 product page for NQ contract context.
CME futures and options data catalog for trade, top-of-book, depth and market-by-order evidence layers.
CME Equity Execution Statistics FAQ for width, depth, fill and slippage measurement categories.
Nasdaq-100 Index Methodology for the underlying index identity.
CME current trading-hours and holiday calendar for session controls.

Sources were reviewed August 28, 2026. Entry and matching rules define how a future comparison should be run. No pullback or breakout advantage is reported here.