AlphaFlow Quant Research V2: Low-Volatility Upward Grind → Breakout Continuation (1H)

1. Objective

Design and execute a rigorous quantitative experiment to determine whether a small, low-volatility, gently upward-trending consolidation following a bearish impulse tends to break upward and continue higher, and whether this structure provides predictive information beyond ordinary upward drift.

The original visual observation is a compact, quiet, gently ascending price structure. It often appears after a sharp downward move, with price gradually recovering while volatility remains compressed.

The hypothesis is that this structure is more likely to produce an upward breakout than a downward breakdown.

This is an upward-breakout study. Do not frame breakdown prediction as the primary research question. Downward breakdowns are the negative outcome and remain important for measuring failure rates.

Do not assume the hypothesis is true. The experiment must be capable of rejecting it.

Primary research question

Following a bearish impulse, does a low-volatility upward grind have a higher probability of breaking above its consolidation high before breaking below its consolidation low within the next 12 hours, compared with comparable upward-drifting structures?

Secondary research questions

1. Does the pattern predict positive forward returns after detection?
2. Does volatility compression add information beyond positive drift alone?
3. Does the prior bearish impulse improve the signal?
4. Does the pattern predict continued upward movement after the breakout, or merely an initial boundary touch?
5. Can the effect survive realistic transaction costs when converted into a trading strategy?

2. Inspect the existing AlphaFlow infrastructure

Inspect the prior experiment lowvol_grind_v1 and its report before implementing anything.

Inspect:

• The existing detection logic and frozen parameters.
• The event-level datasets.
• The baseline comparison methodology.
• The treatment of overlapping events.
• The first-touch outcome implementation.
• The boundary-distance calculations.
• The chronological train, validation, and test splits.
• The trading simulation and cost assumptions.
• The existing experiment repository and reporting conventions.

Audit the previous implementation for logical errors, inconsistent denominators, look-ahead bias, and outcome-labeling errors.

Reuse reliable infrastructure and historical data. Do not silently overwrite lowvol_grind_v1. Save the new study as lowvol_grind_v2 so the two experiments remain reproducible and comparable.

If a previous result cannot be reproduced or appears inconsistent with the underlying event data, investigate and document the discrepancy.

3. Dataset

Use the existing AlphaFlow universe:

• BTCUSDT
• ETHUSDT
• SOLUSDT
• ZECUSDT
• NEARUSDT
• XRPUSDT
• OGNUSDT
• HYPEUSDT
• BNBUSDT
• DOGEUSDT

Venue: Binance USD-M perpetual futures.

Timeframe: 1H, using the established 1-minute-to-1-hour resampling pipeline.

Research period: January 2025 through September 2026, subject to actual data availability.

Use 1-minute OHLCV data where necessary to determine intrahour price ordering and simulate realistic execution.

Inspect data coverage, funding availability, timestamps, gaps, and symbol-specific start dates. Respect the shorter HYPEUSDT history.

4. Pattern definition

Start with the original lowvol_grind_v1 detector so that the experiment remains comparable with the first study.

The original primary pattern uses:

• An eight-candle rolling window.
• Positive linear-regression slope of log-price, normalized by trailing 168-hour realized volatility.
• Normalized slope between 0.1 and 0.6.
• Pattern realized volatility no greater than 0.8 times trailing 168-hour realized volatility.
• A prior 12-hour bearish impulse with normalized impulse no greater than -1.5, measured before the consolidation window.
• Frozen pattern high and low based on the detected window.

Reproduce these conditions exactly before making changes.

Then run a predefined sensitivity analysis over:

• Durations of 4, 6, 8, 10, and 12 hourly candles.
• Multiple positive-slope thresholds.
• Multiple volatility-compression thresholds.
• Multiple prior-impulse thresholds.
• Alternative volatility lookbacks already supported by the research framework.

Do not select a configuration using the final out-of-sample results.

Additional structural features

Compute the following as separate candidate features:

• Consecutive higher closes.
• Higher-low frequency.
• Regression slope and slope stability.
• Candle-range compression.
• Candle overlap.
• Maximum pullback relative to the preceding upward drift.
• Wick-to-body ratio.
• Number of failed downward pushes.
• Distance between the detection close and pattern high.
• Distance between the detection close and pattern low.
• Normalized pattern range.
• Distance to the fast EMA and slow EMA.
• Fast EMA ROC and acceleration.
• EMA spread and spread velocity.
• Volume relative to its trailing baseline.
• Time elapsed since the preceding bearish impulse.

Do not require every feature for a pattern to qualify. Evaluate incremental predictive value through ablation tests.

5. Primary outcome: upward breakout first

For every detected event, freeze:

• pattern_high
• pattern_low
• pattern_range
• detection_close
• detection_timestamp

The outcome window starts after the pattern has been fully detected. No future information may influence pattern qualification.

Use forward horizons of 3, 6, 12, 24, and 48 hours. The primary horizon is 12 hours.

Classify the result as:

1. BREAKOUT_FIRST: price crosses above the frozen pattern high before crossing below the frozen pattern low.
2. BREAKDOWN_FIRST: price crosses below the frozen pattern low before crossing above the frozen pattern high.
3. NEITHER: neither boundary is crossed within the horizon.
4. AMBIGUOUS_SAME_BAR: both boundaries are crossed within the same 1-minute candle and their order cannot be determined.
5. DATA_INVALID: the outcome horizon cannot be evaluated reliably.

Use 1-minute data to resolve ordering within an hourly candle.

Report the number of events in every class. Do not force neither, ambiguous, or invalid events into either directional outcome.

Primary metrics

Calculate:

• Upward-breakout-first probability among resolved events.
• Downward-breakdown-first probability among resolved events.
• Neither-boundary probability.
• Ambiguous-event frequency.
• Breakout probability using all valid events as the denominator.
• Breakdown probability using all valid events as the denominator.
• Confidence intervals for both directional outcomes.

The primary statistic is:

breakout_first_rate = breakout_first_count / resolved_event_count

where resolved events include only breakout-first and breakdown-first events.

Also report:

breakout_rate_all_valid = breakout_first_count / all_valid_events

Keep the denominators explicit throughout the report.

6. Critical control: boundary-distance bias

This is a central requirement of the experiment.

A gently upward-sloping pattern will often finish closer to its upper boundary than its lower boundary. The upper boundary may therefore be reached first simply because it is closer to the current price.

A high upward-first rate does not, by itself, prove that the pattern predicts a bullish move.

Evaluate this issue explicitly.

Analysis A: Original boundary race

Measure which frozen pattern boundary is reached first.

This preserves the original visual hypothesis, but treat the result as descriptive until boundary geometry is controlled.

Analysis B: Match boundary geometry

Compare the pattern with controls matched on:

• Symbol.
• Pattern duration.
• Trailing volatility.
• Prior bearish impulse status.
• Distance from the detection price to the upper boundary.
• Distance from the detection price to the lower boundary.
• Normalized pattern range.
• Market period.

Where exact matching is impractical, use predefined distance bins or a documented matching method.

Evaluate whether the pattern has a higher upward-first rate after controlling for boundary geometry.

Use only information available at detection time for matching. Do not match on future outcomes.

Analysis C: Equal-distance directional barriers

Construct a second outcome test that removes the mechanical advantage of the nearer pattern boundary.

Define symmetric upward and downward barriers around an executable reference price:

• Establish the reference price at the first tradable price after signal confirmation.
• Set an equal-distance upward barrier and downward barrier.
• Express barrier distance in ATR or another predefined volatility-normalized unit.
• Test several predefined barrier distances.
• Observe which barrier is reached first within the same forward horizons.
• Apply consistent rules for gaps, ambiguous touches, and unfilled outcomes.

Use the same barrier rules for the pattern and all baseline groups.

Do not use the asymmetric frozen pattern high and low as the only evidence for directional prediction.

This analysis is essential for establishing whether the structure predicts direction

← All reports · source (.md)