RSI Regular Bullish Divergence Pattern

Key results (1h only)

Results (1h)

Distribution statistics

• Mean return: -0.205%
• Median: 0.032% (P25 -2.908 · P75 2.759)
• Std. deviation: 6.41%
• Win rate (ret > 0): 50.2%
• Mean MFE / MAE: 4.217% / -4.88%
• Median max-up / max-down (×NATR): 1.761 / -2.045
• Sideways rate: 0.53% (pattern-end unresolved; not a zero-move class; see ×NATR histograms)
• Both-cleared (pre-label): 43.82%

Fuel volume & momentum scores

Fuel volume and momentum scores below use only RSI Regular Bullish Divergence snapshots on 1h. Zones match the live Fuel gauge (Dead Water / Organic Flow / The Powder Keg); each zone reports mean max-up / max-down over the next 20 1h candles.

Active timeframe: 1h — switch with the buttons under the pattern title.

Feature distributions vs max-up / max-down

Feature bins below use only RSI Regular Bullish Divergence snapshots on 1h. Each bin reports mean max-up (MFE) and mean max-down (MAE) over the next 20 1h candles.

Active timeframe: 1h — switch with the buttons under the pattern title.

Timeframe comparison

Side-by-side comparison across timeframes for RSI Regular Bullish Divergence. Each column is an independent sample — use it to contrast horizons, not as a pooled average.

Feature correlations

Within RSI Regular Bullish Divergence on 1h, the strongest absolute Spearman correlate of breakout is momentum_score (ρ = 0.041). Ranks use pattern geometry, momentum, and BTC-context features. * p<0.05 · ** p<0.01 · *** p<0.001.

Outcome mix by timeframe

Exclusive breakout / breakdown / sideways rates by candle timeframe (hover for exact values).

Timeframe breakdown

Methods note

Limitations & disclosures

1. This is a historical, in-sample statistical study — not a forward-tested trading strategy. It excludes exchange fees, funding rates, slippage, and execution timing, all of which affect real trading outcomes. A fixed temporal holdout split is reported in methodology for transparency; thresholds are not retuned on the holdout window.
2. The dataset covers Binance USD-M perpetual futures only. Results may not generalize to spot markets, other exchanges, or lower-liquidity assets with different microstructure.
3. Confidence and recency filters were switched off while building the dataset, so figures reflect the detector’s raw geometric/RSI trigger rate — not the filtered, higher-conviction signals surfaced live to subscribers.
4. Exclusive breakout / breakdown / sideways labels use the first ±1.0× NATR threshold (or pattern-specific structural boundary) touched in candle order over the 20-candle forward horizon (target_first_hit / outcome_label). Trendline families evaluate confirmed candle closes against projected support/resistance bands. Double/triple bottoms require breaking the bottom support floor for breakdown. Head & shoulders evaluate from neckline confirmation. Same-bar dual clears resolve via open gap / nearer threshold from open, then candle body direction.
5. Overlapping detections were deduplicated on formation start (longest span kept), but different pattern families can still fire on the same underlying move, so correlated market regimes (e.g. one broad rally) can inflate sample counts across several pattern types simultaneously.
6. Spearman tables use Benjamini–Hochberg FDR q-values; stars also require |ρ| ≥ 0.05. Large samples can still make tiny associations look noteworthy without being economically meaningful. Primary correlations and lifts are within a single timeframe; multi-TF pooled tables are exploratory only.

Glossary

Bull flag / bullish flag

Common search names for the bullish flag chart pattern: a sharp advance (the pole) followed by a downward-sloping or rectangular consolidation. See the Bullish Flag pattern study for breakout statistics.

Head and shoulders pattern

A three-peak reversal pattern (left shoulder, head, right shoulder) with a neckline. Traders search this as “head and shoulders crypto” as well as the full name. See the Head and Shoulders study for exclusive breakdown rates.

Falling wedge bullish

A common query for the falling wedge: converging down-sloping trendlines often treated as a bullish continuation or reversal. See the Falling Wedge study.

Cup and handle pattern

A rounded-bottom continuation pattern with a shallow handle. Also searched as “cup and handle crypto”. See the Cup and Handle study for breakout rates.

Fuel gauge zones

Fixed partitions of Fuel component scores: Dead Water (0.00–0.33), Organic Flow (0.34–0.66), and The Powder Keg (0.67–1.00). Used for Fuel volume/momentum outcome charts on research pages.

NATR(14) & Formation Avg NATR

Normalized Average True Range over 14 periods — average candle range as a percentage of price. In this research, formation average NATR (natr_pattern_avg) is measured across the full pattern span to size breakout/breakdown thresholds (1.0× NATR) relative to market volatility rather than fixed percentages.

\[\mathrm{ATR}_{14}=\mathrm{AverageTrueRange}(H,L,C,14)\]

\[\mathrm{NATR}_{14}=\dfrac{\mathrm{ATR}_{14}}{C_{t}}\times 100\]

\[\mathrm{NATR}_{\mathrm{avg}}=\dfrac{1}{N}\sum_{i=\mathrm{start}}^{\mathrm{end}}\mathrm{NATR}_{14,i}\]

Breakout / breakdown / sideways

Mutually exclusive outcome classes over the 20-candle forward horizon. Headline KPIs, ranks, and pattern-end geometry bins use breakout, breakdown, and sideways (sum to 100%). Breakout means the +1.0× NATR(14) upside threshold (or resistance trendline/neckline) is touched first; breakdown is the downside class; sideways on those samples means the pattern-end candle already tagged a line but neither ±1.0× NATR printed in the next 20 candles. Sideways is not a zero-move class: typical inside-band wick travel is the sideways ×NATR max-up / max-down. Trendline families evaluate confirmed candle closes.

\[\mathrm{Sideways}\iff\text{terminal and no }\pm 1.0\times\mathrm{NATR}\text{ tag within }20\text{ candles}\]

Forming (trendline still inside)

Fourth exclusive class on trendline families (triangles, wedges, channels, flags, pennants) for snapshots that are not the last candle of the formation: price is still between the lines and neither side tagged in the next 20 candles. Age-balanced Fuel / RSI / point-in-time bins report breakout + breakdown + sideways + forming = 100%. Forming-only trees (live still-inside matches) use breakout + breakdown + forming = 100%. Non-trendline patterns have forming_rate 0.

\[\mathrm{Forming}\iff\text{still inside trendlines and no }\pm 1.0\times\mathrm{NATR}\text{ tag within }20\text{ candles}\]

Both-cleared rate

Share of windows where both +1.0× and −1.0× NATR thresholds were cleared within the 20-candle forward horizon before the exclusive label was assigned. High both-cleared rates indicate volatile two-sided windows.

Expected-direction hit rate

For patterns with a directional bias (e.g. bullish patterns expect a breakout, bearish patterns expect a breakdown), this is the exclusive-class rate for that expected side only. Neutral patterns report breakout and breakdown separately.

Spearman ρ (rho)

A rank correlation coefficient between −1 and 1 that measures monotonic association without assuming a linear relationship or normal distribution — appropriate for skewed, fat-tailed return data. Tables emphasize pattern geometry, momentum, and BTC-context features.

p-value, q-value & significance stars

Two-sided Spearman p-values are adjusted with Benjamini–Hochberg FDR (q-values). Stars require q < 0.05 and |ρ| ≥ 0.05 so large-n micro correlations are not marked significant. Always read stars with effect size.

Wilson 95% CI

Outcome rate fields ending in _ci95 are Wilson score intervals for the binomial point estimate. They quantify sampling uncertainty on rates, not trading risk or costs.

Null baseline / temporal holdout

Null baseline shuffles exclusive outcomes within each timeframe to show a chance expected-direction hit band. Temporal holdout compares rates before vs on/after a fixed cutoff (thresholds not retuned on holdout).

MFE / MAE

Maximum Favorable Excursion and Maximum Adverse Excursion — the best and worst price movement recorded within the forward horizon, expressed as a percentage of the snapshot close. MFE÷|MAE| summarizes average upside versus downside excursion magnitude.

\[\mathrm{MFE}=\mathrm{max\_up}_{20c}=\dfrac{\max(H_{t+1},\ldots,H_{t+20})-C_{t}}{C_{t}}\times 100\]

\[\mathrm{MAE}=\mathrm{max\_down}_{20c}=\dfrac{\min(L_{t+1},\ldots,L_{t+20})-C_{t}}{C_{t}}\times 100\]

Timeframe-specific feature bins

On each pattern study, feature distributions are computed separately for 15m, 1h, 4h, and 1d. Use the timeframe control to switch; never treat pooled multi-TF histograms as interchangeable because candle horizons differ.

Feature quartile bins

Numeric features that are not Fuel, RSI, or σ-banded (e.g. slopes, current_scale_number, btc_dist14_mean) are split into quartiles. Each bin reports exclusive breakout, breakdown, and sideways rates plus mean MFE.

Feature typical / elevated bands

geom_width and head-and-shoulders symmetry use mean ± σ bands computed separately per pattern and timeframe: below typical (< μ−σ), typical (μ±σ), elevated (μ+σ to μ+2σ), and extreme (> μ+2σ). Pattern height and extension age use the same labels after a log1p transform so right-tailed size features are not dominated by outliers. Bins are never pooled across timeframes.

Pattern end (snapshot candle)

The single candle where features (RSI, Fuel, BTC context) and the forward 20-candle outcome window are measured. Trendline families use the last close still inside ±0.5× NATR of the projected channel; doubles/triples use a neckline re-touch within ±0.5× NATR; H&S uses the neckline break; cup & handle uses the rim touch after the handle; RSI divergences use the second price extreme. See Data & methodology for the full family list.

BTC regime (btc_dist14_median)

Terciles of BTC last-14 median distance at the live snapshot (btc_weak / btc_mid / btc_strong). Used as a simple proxy for BTC-relative market regime when slicing outcome rates (not in correlation ranks).

volume_score (Fuel volume)

Closed-candle volume vs a 20-period SMA, mapped to 0–1 with a small trend adjustment. Same Fuel Index volume component as the live gauge; open (forming) candle is not used.

\[\mathrm{volume\_ratio}=\dfrac{V_{t}}{\mathrm{SMA}(V,20)}\]

Piecewise base score from volume_ratio:

\[\mathrm{ratio}<0.8:\ \mathrm{base}=\dfrac{\mathrm{ratio}}{0.8}\times 0.25\]

\[0.8\le\mathrm{ratio}\le 1.5:\ \mathrm{base}=0.3+\dfrac{\mathrm{ratio}-0.8}{0.7}\times 0.3\]

\[\mathrm{ratio}>1.5:\ \mathrm{base}=0.6+\min\!\bigl(1,\tfrac{\mathrm{ratio}-1.5}{1.5}\bigr)\times 0.4\]

\[\mathrm{trend\_adj}=\mathrm{clamp}(s_{V}\times 0.15,-0.05,+0.05)\]

\[\mathrm{volume\_score}=\mathrm{clamp}(\mathrm{base}+\mathrm{trend\_adj},\,0,\,1)\]

momentum_score (Fuel momentum)

Direction-agnostic energy score from RSI(14), 3-bar RSI change, and timeframe-specific ROC. Same Fuel Index momentum component as the live gauge.

\[\mathrm{ROC}=\dfrac{C_{t}-C_{t-N}}{C_{t-N}}\]

\[\Delta\mathrm{RSI}_{3}=\mathrm{RSI}_{t}-\mathrm{RSI}_{t-3}\]

ROC lookback N by timeframe: 15m→8, 1h→5, 4h→3, 1d→2.

\[\mathrm{momentum\_score}\in[0,1]\ \text{(RSI/ROC regime map)}\]

rsi_14

14-period Relative Strength Index on the asset close at the live snapshot (forming + terminal; feature_builder: RSIIndicator(close, window=14)).

\[U_{t}=\max(C_{t}-C_{t-1},0),\quad D_{t}=\max(C_{t-1}-C_{t},0)\]

\[\mathrm{RS}=\dfrac{\mathrm{SMA}(U,14)}{\mathrm{SMA}(D,14)}\]

\[\mathrm{RSI}_{14}=100-\dfrac{100}{1+\mathrm{RS}}\]

rsi_div_span

Candle count between the first and second price extremes of an RSI divergence.

\[\mathrm{rsi\_div\_span}=i^{\mathrm{price}}_{2}-i^{\mathrm{price}}_{1}\]

rsi_div_price_divergence

Signed relative price move between extremes (detector confidence input).

\[\mathrm{price\_div}=\mathrm{signed}\,\dfrac{|P_{2}-P_{1}|}{P_{1}}\]

rsi_div_rsi_divergence

Signed RSI move between extremes, scaled by 100 (detector confidence input).

\[\mathrm{rsi\_div}=\mathrm{signed}\,\dfrac{|\mathrm{RSI}_{2}-\mathrm{RSI}_{1}|}{100}\]

btc_dist14_mean

Mean of BTC last-14 percentage distances (feature_builder dist14_mean on BTC).

\[\delta_{i}=\dfrac{C^{\mathrm{BTC}}_{t-i}-C^{\mathrm{BTC}}_{t}}{C^{\mathrm{BTC}}_{t}}\times 100,\quad i=1,\ldots,14\]

\[\mathrm{btc\_dist14\_mean}=\dfrac{1}{14}\sum_{i=1}^{14}\delta_{i}\]

rsi_div_price_delta_pct

Percent price change from first to second extreme

rsi_div_rsi_delta

RSI-point change from first to second extreme