Do Bull Flags Still Work? A Testing Framework

Bull flags have not become automatically valid or obsolete in 2026. Academic studies have reported positive results for particular bull-flag recognition rules in particular index datasets, but those findings do not create a universal win rate for every chart, asset, timeframe, or visual flag. The useful question is not simply “Does the pattern work?” It is: Which exact bull-flag definition, confirmation rule, market universe, holding period, cost model, and failure rule produced the result?

A bull flag can be evaluated as a repeatable hypothesis. First freeze the pattern definition. Then identify every qualifying candidate without knowing the future, apply one confirmation rule, include unresolved and failed examples, measure outcomes over a fixed horizon, subtract realistic costs, and reserve later data for out-of-sample testing. A result without those controls is a chart collection, not evidence of effectiveness.

Educational note: Technical analysis uses historical price and volume data to form hypotheses. It does not guarantee direction or profit. Simulated results can differ materially from live trading because of spreads, slippage, liquidity, execution timing, fees, and human decisions.

Key Takeaways

There is no single, market-wide bull flag success rate.
A broad bull flag and a strict high-and-tight flag are not interchangeable datasets.
Pattern effectiveness depends on how the flagpole, consolidation, boundaries, breakout, invalidation, outcome, and time horizon are defined.
Testing only completed winners creates selection bias; all candidates, failed breakouts, expired patterns, and unresolved cases must remain in the record.
A breakout target being reached is only one outcome measure. Expectancy, adverse movement, time to outcome, costs, and benchmark performance may matter more.
In-sample results are not enough. Rules should be checked on later or otherwise untouched data.
ChartMini can support blind replay and record keeping, but it cannot reproduce full live execution.

Which Bull Flag Guide Should You Use?

ChartMini has several related pattern guides. Their roles should remain separate.

Question	Primary guide
What does a bull flag look like, and how do I identify its parts?	Bull Flag Cheat Sheet
How can I evaluate whether a frozen bull flag rule still has evidence?	This guide
How do flags compare with triangles, wedges, and reversal patterns?	Chart Patterns Cheat Sheet
How should range, triangle, flag, and volatility breakouts be classified?	Breakout Patterns Guide
What is the bearish mirror structure?	Bear Flag Guide
How can I practice pattern classification without seeing future candles?	Chart Replay Pattern-Recognition Guide

This page does not replace the broad Bull Flag Cheat Sheet. It focuses on effectiveness claims, test design, outcome measurement, robustness, and evidence limits.

What Does “Still Effective” Mean?

The phrase sounds precise, but it can describe several different questions.

A person may mean:

Does a completed bull flag breakout move higher more often than lower?
Does a specific entry and exit rule produce positive average returns?
Does it remain profitable after spread, commission, slippage, and failed fills?
Does it outperform buy-and-hold or another trend rule?
Does it work in stocks but not Forex or crypto?
Does it work on daily data but disappear on five-minute data?
Does it work in rising markets but fail during broad declines?
Does the result survive later data and reasonable rule changes?

These are not equivalent.

A pattern can have a high directional hit rate but poor expectancy if losing cases are much larger than winning cases. It can reach a measured-move target often before costs but lose its advantage after realistic execution assumptions. It can appear effective during one strong market regime and fail when the regime changes.

Before collecting examples, write one testable question.

“Among candidates detected by Definition V1 in Universe A between Dates B and C, what percentage of Close-Confirmation breakouts reach Outcome X before Invalidation Y within Horizon Z, and what is the average net return after the stated cost model?”

That sentence is less exciting than “bull flags work 85% of the time,” but it can be audited and repeated.

What Published Research Can—and Cannot—Tell You

Bull flags have been studied in academic pattern-recognition research, but the available evidence is definition-specific.

A 2002 study described a particular bull-flag charting heuristic, tested it on several years of New York Stock Exchange Composite Index history, and reported positive results. A 2024 study applied template matching to the Shanghai Stock Exchange Composite Index from 1991 through 2021 and reported that its bull-flag trading rules predicted direction most of the time and produced significantly positive excess profits under the study design.

Those results matter because they show that bull-flag rules can be expressed computationally and evaluated rather than treated only as subjective drawings. They do not establish that:

every visual bull flag is profitable;
the same thresholds transfer to another country, instrument, or timeframe;
results remain identical after 2021;
a retail trader can obtain the paper’s assumed execution;
all definitions of breakout, target, stop, and holding period are equivalent;
a pattern found after inspecting the completed chart was detectable in real time.

A separate source of widely repeated statistics is Thomas Bulkowski’s work on the high and tight flag. That is a stricter formation with unusual prior appreciation and separate qualification rules. Bulkowski’s own updated tables divide samples by flag depth and duration and warn against relying on small sample groups. Statistics for that pattern cannot be assigned to ordinary bull flags merely because both names contain “flag.”

The evidence-based conclusion is therefore conditional:

Bull-flag rules have shown positive results in some published samples, while the size and persistence of any advantage depend on the exact rule, market, period, execution assumptions, and validation method.

Bull Flag Usage in 2026 Is Not Proof of Effectiveness

Technical analysts still use flag and pennant terminology in current market commentary. For example, a Reuters technical-analysis article published in July 2026 described a bull pennant in the S&P 500 Energy Index, a measured-move hypothesis, an attempted breakout that initially stalled, and a clear downside condition if price moved back below the pattern boundary.

That example demonstrates that the vocabulary remains in active use. It does not establish a general success rate. A live chart interpretation is one conditional scenario, not a controlled sample.

The same distinction applies to search results. Many current bull-flag guides explain the flagpole, consolidation, breakout, volume, stop placement, and projected target. Some present the pattern as reliable. Few state the complete sampling method behind the reliability claim.

A current label does not make old evidence current. A 2026 article must still disclose the dataset and rules behind any statistic.

Freeze the Pattern Definition Before Looking at Outcomes

A visual bull flag can be drawn many ways. Testing begins by turning the drawing into fields and thresholds.

1. Chart identity

Record:

symbol and exact product;
exchange, venue, broker, or data provider;
price type: last, bid, ask, midpoint, mark, index, or settlement;
raw or adjusted prices;
regular, extended, overnight, or continuous session;
timezone;
timeframe or bar-construction method;
volume source;
whether each decision bar was closed.

Different feeds and session settings can change swing points, volume, flag boundaries, and breakout timing.

2. Flagpole start and end

Choose a repeatable endpoint rule.

Possible versions:

lowest confirmed pivot before the impulse to highest pivot before consolidation;
breakout close from a prior range to first confirmed impulse high;
lowest low in a fixed lookback to the highest high before the flag;
first close above a prior boundary to the final close before contraction.

Do not move the endpoints after seeing the breakout target.

Useful measurements include:

Flagpole height = Flagpole end price - Flagpole start price

Flagpole return = (Flagpole end price - Flagpole start price) ÷ Flagpole start price

Flagpole duration = End bar index - Start bar index + 1

3. Consolidation start and end

Define when the flag begins. It might begin on the first bar after the pole high, after a confirmed pivot, or after a volatility-contraction condition.

Before breakout, define:

minimum and maximum number of bars;
whether a horizontal rectangle qualifies;
whether a downward channel qualifies;
whether a narrowing pennant is excluded;
how many boundary contacts are required;
whether wicks or closes define boundaries;
how much boundary tolerance is allowed.

4. Pullback depth

One normalized measure is:

Pullback depth = (Flagpole end price - Lowest flag price) ÷ Flagpole height

This allows comparisons across different price levels. The threshold is a research variable, not a universal law.

5. Flag width and volatility

Record:

Flag duration = Breakout-candidate bar index - Flag-start bar index

Flag range = Highest flag price - Lowest flag price

Normalized flag range = Flag range ÷ Flagpole height

You may also record average true range or another volatility measure, but state the exact formula and lookback.

6. Boundary method

Possible versions include:

parallel trendlines through selected pivot highs and lows;
horizontal upper and lower envelope;
regression channel;
highest close and lowest close inside the flag;
fixed tolerance around selected pivots.

If the boundary is redrawn after each new candle, save each version. Otherwise the final line may use information unavailable at the original decision time.

Separate Ordinary Bull Flags From High-and-Tight Flags

A common statistical error is mixing distinct pattern classes.

An ordinary bull flag may describe a strong advance followed by a relatively contained pause. A high-and-tight flag generally requires a much more extreme prior move and a narrowly defined consolidation. It is rarer and often studied as a separate formation.

Use a classification field such as:

pattern_class = ordinary_bull_flag
pattern_class = high_tight_flag
pattern_class = pennant
pattern_class = rectangle_after_advance
pattern_class = rejected_candidate

Do not pool the classes until you have tested whether their distributions are sufficiently similar. If one rare subclass has stronger historical results, combining it with broad flags can hide the difference. Assigning the rare subclass’s result to every broad flag creates the opposite error.

Also preserve rejected candidates. A steep rise followed by a loose, deep, or irregular pullback may fail the frozen flag definition, but it should remain recorded as a rejected candidate rather than disappear from the research history.

Define the Breakout Confirmation Rule

A visible move above the upper boundary can be classified in several ways.

Rule A: Intrabar break

Confirmation occurs when the high exceeds the boundary.

Advantages:

earlier signal;
larger sample;
captures brief boundary crossings.

Limitations:

includes wick-only breaks;
assumes execution near a threshold that may be crossed briefly;
can increase false-break classification.

Rule B: Completed close above the boundary

Confirmation occurs when a closed candle finishes above the boundary.

Advantages:

avoids decisions based on a changing live candle;
simpler to reproduce from OHLC data.

Limitations:

later signal;
close may occur far above the boundary;
still does not guarantee follow-through.

Rule C: Buffered close

Confirmation requires a close above the boundary plus a predefined absolute, percentage, tick, or volatility buffer.

Advantages:

tests whether a larger break filters marginal crossings.

Limitations:

reduces sample size;
can worsen assumed entry price;
the buffer can be overfit.

Rule D: Break plus follow-through

Confirmation requires the break and an additional event, such as the next bar closing above the boundary or making a higher close.

Advantages:

separates immediate failures from sustained breaks.

Limitations:

introduces another parameter;
delays entry assumptions;
excludes breakouts that continue without the specified bar sequence.

Rule E: Break plus successful retest

Confirmation requires price to break, revisit a boundary zone, and then satisfy a predefined hold rule.

Advantages:

explicitly tests post-breakout acceptance.

Limitations:

many breakouts never retest;
“successful retest” can become subjective;
requires exact zone and timing definitions.

Run these as separate variants. Do not describe one chart as a Close-Confirmation winner and another as an Intrabar winner simply because each version produces a better result.

Define Invalidation, Failure, and Expiry

These terms answer different questions.

Pre-breakout invalidation

The candidate stops qualifying before confirmation.

Possible triggers:

close below the lower flag boundary;
pullback depth exceeds the frozen maximum;
consolidation exceeds maximum duration;
boundaries become too wide or irregular;
a new swing changes the structure classification.

Breakout failure

The candidate confirmed, but the breakout later failed under a predefined rule.

Possible versions:

first close back inside the flag;
close below the breakout boundary by a buffer;
close below the flag low;
predefined loss threshold reached before the outcome target;
no positive follow-through within a fixed number of bars.

Expiry

Neither target nor failure occurs before the observation horizon ends.

An expired case is not automatically a win or a loss. Keep it as a separate status unless the strategy rules specify a time exit.

A useful status sequence is:

candidate
rejected_before_confirmation
confirmed
failed_after_confirmation
outcome_reached
expired
unresolved_at_dataset_end

This prevents survivorship within the pattern sample.

Choose Outcome Measures Before Running the Test

A single “success rate” can hide important information.

Directional outcome

Did the return over a fixed horizon have the expected sign?

Horizon return = (Exit reference price - Entry reference price) ÷ Entry reference price

State whether the reference uses next open, confirmation close, boundary price, or another executable assumption.

Target-before-failure outcome

Did price reach a predefined target before a predefined failure level?

This is common but highly sensitive to target and failure distances. A measured move equal to the flagpole is a hypothesis, not an obligation.

Maximum favorable excursion

MFE = Maximum favorable price movement during the horizon

MFE shows how far the trade moved in the expected direction, even if the final close was lower.

Maximum adverse excursion

MAE = Maximum adverse price movement during the horizon

MAE helps identify the path risk hidden by endpoint returns.

Time to outcome

Record the number of bars from confirmation to target, failure, expiry, and maximum excursion.

Expectancy

For a rule with defined trade returns:

Expectancy = Mean net return across all eligible signals

or, when using win/loss categories:

Expectancy = Win probability × Average win
           - Loss probability × Average loss

Use net values after the stated costs. A high win percentage does not guarantee positive expectancy.

Benchmark-relative result

Compare the test with a stated alternative:

buy-and-hold for the same horizon;
random dates with similar market exposure;
generic momentum rule;
breakout rule without a flag filter;
same rule in non-flag consolidations.

Without a benchmark, you may be measuring the broader uptrend rather than added information from the flag itself.

Include Costs and Executability

The CFTC warns that hypothetical results may not reflect actual market conditions and can omit spread, execution uncertainty, liquidity, fees, and the effect of the trade itself.

A bull-flag study should disclose:

assumed entry time and price;
whether a market, stop, limit, or next-open order is modeled;
bid/ask spread;
commission and exchange fees;
slippage method;
gaps across the trigger;
partial or missed fills;
borrowing or financing costs where relevant;
market impact for the assumed size;
whether extended-hours liquidity is included.

Avoid assuming every historical boundary was filled exactly at the plotted line. A candle high above the boundary proves that the market traded somewhere above it; it does not prove your entire order would have filled at the boundary.

Run sensitivity cases. For example:

Cost model 1: zero-cost diagnostic
Cost model 2: typical liquid-market estimate
Cost model 3: stressed spread and slippage

The zero-cost version can help isolate the pattern signal, but it should not be presented as a live-performance result.

Avoid the Main Sources of False Confidence

Hindsight-defined candidates

Finding flags only after seeing the breakout removes failures and unresolved formations. Candidates must be marked while the future is hidden or detected by rules that use only available bars.

Data snooping

Testing many pole thresholds, flag depths, time limits, volume ratios, buffers, stops, and targets until one combination wins increases the chance of finding a random result.

Keep a rule-development log and limit the number of variants. Treat the best in-sample version skeptically.

Survivorship bias

A current list of active securities excludes delisted, acquired, failed, or illiquid instruments that existed during the test period. Use point-in-time constituents where possible.

Look-ahead bias

Examples include:

using a pivot that requires future bars before the pivot could be known;
using the final regression channel to classify earlier bars;
entering at the breakout price before the closing confirmation exists;
using revised fundamental or index-membership data.

Overlapping signals

One extended trend can create several similar candidates. Decide whether overlapping flags are separate signals, one parent structure, or excluded duplicates.

Regime concentration

A strategy tested mostly during a powerful bull market may be capturing market beta or momentum. Segment results by a predefined regime rule rather than describing the regime after the fact.

Small samples

Rare patterns can produce impressive percentages from few examples. Report sample counts, confidence intervals or uncertainty measures, and the number of independent instruments and periods.

Outcome censoring

Do not delete trades that did not reach either target before the data ended. Label them unresolved or exclude them under a rule established before testing.

Use In-Sample, Validation, and Out-of-Sample Data

A robust workflow separates rule creation from rule evaluation.

Development sample

Use this portion to define candidate detection, thresholds, confirmation, invalidation, and outcome fields.

Validation sample

Use a separate period or instrument set to compare a limited number of versions and check obvious fragility.

Final out-of-sample test

Lock the selected rule before opening the final sample. Do not revise the definition after seeing the final results.

A chronological split is often easier to audit:

Development: earliest period
Validation: middle period
Out of sample: latest untouched period

Walk-forward testing can repeat the process through time, but every retraining and parameter-selection step must be included in the simulation.

The 2024 Shanghai Composite study is useful precisely because it describes a computational recognition and trading-rule framework. Your personal test still needs its own universe, period, definition, costs, and untouched data.

Test Robustness Instead of One Perfect Parameter Set

A genuine effect should not vanish after every small, reasonable rule change.

Compare nearby versions of:

minimum flagpole return;
flag duration;
maximum pullback depth;
boundary tolerance;
intrabar versus close confirmation;
confirmation buffer;
volume filter;
holding horizon;
failure threshold;
cost assumptions.

Create a parameter table rather than reporting only the best cell.

Rule component	Version 1	Version 2	Version 3
Breakout	Intrabar high	Completed close	Buffered close
Flag depth	Narrow limit	Medium limit	Wider limit
Expiry	Short	Medium	Long
Outcome	Fixed horizon	Target before failure	Net expectancy
Costs	Diagnostic zero cost	Base case	Stress case

If one isolated combination performs well while adjacent versions collapse, the result may be overfit or unstable.

Compare Markets and Timeframes Without Pooling Them Blindly

Stocks, futures, Forex, and crypto differ in session structure, price source, gaps, volume, transaction costs, and liquidity.

Record separate groups for:

market and product type;
venue and data source;
regular versus extended hours;
intraday versus daily or weekly bars;
liquid versus less-liquid instruments;
broad market regime;
event and non-event periods.

Do not conclude that a daily-equity result applies to five-minute Forex or perpetual crypto merely because the shape looks similar.

Also distinguish standard exchange volume from broker-specific tick volume or venue-specific crypto volume. A universal “breakout volume must be twice average” rule has no stable meaning until the source, baseline, lookback, and comparison method are defined.

Worked Research Example

Assume a researcher wants to test a daily-stock bull flag rule. This is an illustration of documentation, not a recommended strategy.

Candidate definition

point-in-time liquid-stock universe;
split-adjusted daily OHLCV;
flagpole endpoints selected by a predefined pivot rule;
minimum pole return and maximum pole duration frozen before testing;
flag begins after the confirmed pole high;
consolidation must last within a stated bar range;
pullback depth below a stated maximum;
upper and lower boundaries based on a defined close or wick method.

Confirmation

first completed daily close above the saved upper boundary;
entry reference is next session open;
gaps through the trigger are accepted at next open rather than backfilled at the boundary.

Invalidation and expiry

candidate invalid if it closes below the lower boundary before confirmation;
confirmed signal fails if the predefined loss threshold is reached first;
observation expires after a fixed number of sessions.

Outcomes

Record:

next-open entry price;
net return at fixed horizons;
target-before-failure status;
MFE and MAE;
bars to outcome;
market benchmark return;
cost-model result.

Validation

rules developed on the earliest period;
limited variants compared on a middle period;
final rule tested unchanged on later data;
results reported separately by regime and sample count.

A reader could reproduce or challenge this study because each field is visible. “I reviewed 100 good-looking flags” is not reproducible because “good-looking” changes after the result is known.

A Bull Flag Effectiveness Worksheet

Use one row per candidate.

Chart and sample identity

Candidate ID
Symbol and product
Market and venue
Data provider
Price type
Adjustment method
Session
Timezone
Timeframe
Candidate timestamp
Dataset split: development, validation, or out of sample

Flagpole

Start bar and price
End bar and price
Endpoint rule
Height
Return
Duration
Volume baseline and pole-volume statistic

Flag

Start bar
Upper-boundary method
Lower-boundary method
Boundary version
Pullback depth
Duration
Normalized range
Slope or shape class
Volume statistic
Ordinary or high-and-tight classification

Confirmation

Confirmation rule version
Confirmation timestamp
Boundary value at confirmation
Trigger value
Assumed entry method
Assumed entry price
Gap or missed-fill status

Outcome

Pre-breakout invalidation status
Post-breakout failure status
Expiry status
Fixed-horizon returns
MFE
MAE
Target-before-failure result
Bars to outcome
Gross result
Costs
Net result
Benchmark return

Audit fields

Was the candidate marked before outcome?
Was the latest bar closed?
Was any boundary redrawn?
Was the rule changed after seeing the outcome?
Was the case excluded? If so, under which prewritten rule?
Reviewer notes

How to Use ChartMini for Bull Flag Research Practice

ChartMini can support a manual blind-replay workflow:

Choose a historical chart and record its settings.
Hide future candles.
Advance until a possible flagpole completes under your frozen rule.
Mark the candidate before seeing the breakout.
Save the flag boundaries and definition version.
Record whether the candidate invalidates, confirms, expires, or remains unresolved.
If confirmed, advance through the fixed outcome horizon.
Record every result, including false breakouts and no-action cases.
Review batches rather than changing the definition after each example.
Retest the frozen rule on a different period or group of charts.

Use the pattern-recognition replay guide for the broader practice process.

What ChartMini does not reproduce

ChartMini is best suited for lightweight historical candle replay and observation practice. It does not provide:

live broker order routing;
exchange queue position;
complete bid/ask history;
precise spread and slippage simulation;
guaranteed stop or limit fills;
market impact;
partial-fill modeling;
news latency;
full Level 2 or order-book reconstruction;
the emotional and operational conditions of live risk.

Replay results should therefore be labeled as historical or hypothetical, not live performance.

Common Errors in Bull Flag Effectiveness Claims

Quoting a success rate without a definition

Ask which pattern class, market, timeframe, period, breakout rule, target, stop, and cost model generated the number.

Assigning high-and-tight statistics to ordinary flags

The strict subclass should remain separate unless the study explicitly pools it with ordinary flags.

Selecting only confirmed breakouts

This may answer a conditional breakout question, but it does not measure how often pre-breakout candidates confirm. Report both stages.

Counting wick breaks as executable close-confirmation trades

A wick above the line and a close above the line are different events. Entry assumptions must match the chosen event.

Treating the flagpole measured move as guaranteed

The projected distance is an outcome hypothesis. Record partial moves, failures, and timeouts.

Ignoring the broad market trend

If every test occurs during a rising index, the apparent pattern edge may partly reflect market exposure. Use benchmark-relative and regime-separated results.

Claiming algorithms killed or improved the pattern

That claim requires comparative evidence across periods with the same definitions, data quality, costs, and universe. The presence of algorithmic trading alone does not prove the direction or size of a change.

Using a scanner’s label as ground truth

A scanner implements one vendor’s definition. Save the parameters and verify whether the label matches your test rule.

Reporting only the best settings

Show parameter sensitivity and out-of-sample results, not only the highest in-sample score.

Treating replay as live execution

Historical candles omit important execution details. Use replay to study classification and decision rules, not to claim guaranteed live returns.

Practical Next Step

Write a one-page Bull Flag Rule Card before opening another completed chart.

Include:

chart identity;
flagpole endpoints;
minimum and maximum pole conditions;
flag boundaries;
pullback-depth and duration limits;
ordinary versus high-and-tight class;
breakout confirmation;
invalidation and expiry;
entry and cost assumptions;
outcome horizon and benchmark;
dataset split;
rules for exclusions and unresolved cases.

Then collect a complete candidate set in blind replay. Do not change the card until the planned batch is complete. The result will not prove that every bull flag works, but it can tell you whether one explicit rule showed useful evidence in one defined sample.

Frequently Asked Questions

Do bull flag patterns still work?

Bull flags can produce useful results under some definitions, markets, periods, and confirmation rules, but there is no universal bull flag success rate. The pattern must be defined before testing, evaluated on a complete candidate set, adjusted for costs, and checked on data that was not used to design the rules.

What does bull flag effectiveness mean?

Effectiveness must be tied to a measurable outcome. Examples include the percentage of confirmed breakouts that remain above the boundary, reach a predefined return before invalidation, produce positive expectancy after costs, or outperform a stated benchmark over a fixed horizon. Different outcome definitions can produce different answers.

Is a high and tight flag the same as a normal bull flag?

No. A high and tight flag is a much narrower pattern with unusually strong prior appreciation and its own qualification rules. Statistics published for that strict pattern should not be transferred to every ordinary bull flag, shallow pullback, or breakout consolidation.

How should a bull flag breakout be confirmed?

Choose one confirmation rule before reviewing outcomes. Possible versions include an intrabar break, a completed close above the upper boundary, a buffered close, a break plus follow-through, or a break followed by a successful retest. Each version changes signal timing, sample size, entry assumptions, and failure rates.

What invalidates a bull flag test candidate?

Invalidation depends on the frozen test rules. Examples include a close below the lower flag boundary, a maximum pullback-depth breach, consolidation lasting beyond a predefined duration, or failure to confirm before an expiry point. Invalidation rules must be recorded before the later price path is known.

Can ChartMini prove that bull flags are profitable?

No. ChartMini can help you replay historical candles, hide future bars, classify candidates, and record what happened next. It does not prove universal profitability or reproduce live spread, slippage, queue position, partial fills, market impact, news reaction, or trading psychology.

Sources and Evidence Boundaries

Leigh, Paz, and Purvis, “Market timing: a test of a charting heuristic”, tested one bull-flag heuristic on NYSE Composite history and reported positive results. The finding belongs to that rule and sample.
Lobão, Pacheco, and Fernandes, “Trading rule discovery using technical analysis and a template matching technique for pattern recognition: Evidence from the Chinese stock market”, tested template-matched bull-flag rules on the Shanghai Composite over 1991–2021 and reported positive excess profits under the study design.
Thomas Bulkowski’s High and Tight Flag study reports results for a stricter pattern and separates samples by flag depth and duration. It should not be treated as a universal ordinary-bull-flag statistic.
The CFTC trading-system advisory explains limitations of hypothetical results, including execution, spread, liquidity, fees, and historical selection concerns.
Reuters’ July 2026 energy-index bull pennant analysis shows that flag-family terminology remains in current market analysis, while also presenting confirmation and failure conditions. It is a live example, not performance proof.

Do Bull Flags Still Work? How to Test Pattern Effectiveness