Bitcoin after Weekly RSI crosses 50
How Bitcoin performed after weekly momentum crossed into positive territory.
median forward return
Context before conviction.
After removing overlapping forward windows, N=11 independent observations remain from 16 raw events. Their 26W median return was +22.14% with a 54.5% win rate and -17.39% median adverse excursion.
Outcome by horizon
Primary non-overlapping results are shown beside the complete raw event population.
| Horizon | Population | Sample | Median | Mean | Win rate | 25th percentile | 75th percentile | Median MAE |
|---|---|---|---|---|---|---|---|---|
| 4W | Independent | N=17 | -3.4% | -4.6% | 47.1% | -21.8% | +10.0% | -17.7% |
| 4W | Raw events | N=17 | -3.4% | -4.6% | 47.1% | -21.8% | +10.0% | -17.7% |
| 13W | Independent | N=14 | -2.2% | +12.0% | 50% | -20.3% | +40.8% | -23.8% |
| 13W | Raw events | N=17 | -7.8% | +11.0% | 47.1% | -18.0% | +45.2% | -28.2% |
| 26W | Independent | N=11 | +22.1% | +20.9% | 54.5% | -25.3% | +52.9% | -17.4% |
| 26W | Raw events | N=16 | +0.5% | +7.8% | 50% | -33.2% | +46.5% | -28.9% |
| 52W | Independent | N=6 | +63.2% | +109.2% | 66.7% | -23.5% | +98.8% | -22.5% |
| 52W | Raw events | N=15 | +30.5% | +71.1% | 60% | -40.9% | +97.8% | -55.6% |
Distribution and drawdown
Inspect the stored event-level outcomes behind the aggregate table.
Temporal validation
Chronological 70/30 holdouts and expanding-window folds use only independent observations. Later outcomes never influence event selection or thresholds.
1 of 3 expanding folds retained the training median direction.
| Horizon | Independent N | Train median | Test median | Direction | Stable folds | Assessment |
|---|---|---|---|---|---|---|
| 4W | N=17 | -15.8% | +2.6% | Changed | 1/3 · 33.3% | Unstable |
| 13W | N=14 | -7.8% | +3.4% | Changed | 1/3 · 33.3% | Unstable |
| 26W | N=11 | -4.9% | +38.8% | Changed | 1/3 · 33.3% | Unstable |
| 52W | N=6 | — | — | — | — | Insufficient Data |
| Fold | Train through | Future test window | Train / test N | Train median | Test median | Direction |
|---|---|---|---|---|---|---|
| 1 | 2021-08-02 | 2022-03-28 → 2023-01-16 | 5 / 2 | -4.9% | -7.5% | Stable |
| 2 | 2023-01-16 | 2023-10-02 → 2024-07-15 | 7 / 2 | -4.9% | +105.0% | Changed |
| 3 | 2024-07-15 | 2025-04-28 → 2025-10-27 | 9 / 2 | +44.9% | -4.6% | Changed |
Sensitivity analysis
One detector setting changes at a time around the published configuration. Returns never determine which configuration is canonical.
100% of 4 comparable alternatives retained the canonical median direction.
| Configuration | Parameters | Raw N | Independent N | Median | Win rate | Direction |
|---|---|---|---|---|---|---|
| RSI threshold 45 | Cooldown 8 · Threshold 45 | N=13 | N=10 | +39.7% | 70% | Stable |
| Canonical · RSI 50 / 8W cooldownPublished | Cooldown 8 · Threshold 50 | N=16 | N=11 | +22.1% | 54.5% | Baseline |
| RSI threshold 55 | Cooldown 8 · Threshold 55 | N=20 | N=12 | +44.8% | 83.3% | Stable |
| Cooldown 6 weeks | Cooldown 6 · Threshold 50 | N=16 | N=11 | +22.1% | 54.5% | Stable |
| Cooldown 12 weeks | Cooldown 12 · Threshold 50 | N=14 | N=11 | +22.1% | 54.5% | Stable |
This is a robustness check, not model selection. Each row changes one assumption while every other detector rule remains fixed.
Threshold and regime comparison
Compare every predeclared weekly RSI threshold using raw events and horizon-specific non-overlapping observations, segmented by the regime known at each signal close.
| Threshold | Raw N | Independent N | Median | Win rate | P25 / P75 | Median MAE | Confidence | Interpretation |
|---|---|---|---|---|---|---|---|---|
| RSI 45 | N=13 | N=10 | +39.7% | 70% | -2.9% / +49.7% | -8.1% | Low | Eligible |
| RSI 50Published | N=16 | N=11 | +22.1% | 54.5% | -25.3% / +52.9% | -17.4% | Low | Eligible |
| RSI 55 | N=20 | N=12 | +44.8% | 83.3% | +21.3% / +110.9% | -10.5% | Low | Eligible |
| RSI 60 | N=20 | N=11 | +48.4% | 90.9% | +11.6% / +100.0% | -12.8% | Low | Eligible |
Non-overlap is selected globally per threshold and horizon before regime rows are formed. Threshold variants may share source events, so rows are descriptive comparisons—not independent experiments. The interpretation gate is N=8; smaller samples remain visible and explicitly marked insufficient.
Cross-exchange benchmark
The published detector is replayed independently on each venue over the shared completed-candle window. Exchange populations are never mixed.
Both eligible venue samples retain the same independent median direction.
Coinbase remains the canonical research source. Bitstamp is an external robustness check; its candles, events, and forward-return population remain separate.
Weekly RSI crosses above 50 · 8-week cooldown
The signal is evaluated only with information available at the event close. Future observations are used solely to calculate outcomes.
RSI uses Wilder smoothing over 14 completed weekly candles.
The event occurs only when RSI crosses from below to at least 50.
An eight-week cooldown limits threshold clusters.
Forward returns begin at the completed weekly close.
Primary confidence and headline statistics use the earliest chronological non-overlapping sample for each horizon; raw events remain reported for audit.
Temporal validation uses the independent sample only: a chronological 70/30 holdout plus expanding-window folds with at least two later observations per fold and no threshold tuning.
Sensitivity analysis changes one detector parameter at a time around the published canonical configuration, compares independent 26W outcomes, requires at least 8 observations per comparable variant, and never selects a parameter from performance.
Cross-exchange robustness replays the canonical detector independently on Coinbase BTC-USD and Bitstamp BTC/USD over their shared completed-daily history. Each source keeps its own signal and non-overlapping outcome population, with at least 8 observations required per benchmark.
RSI threshold research is predeclared at 45, 50, 55, 60 and reports every configured forward horizon without selecting a threshold from performance.
Each RSI threshold first selects its chronological non-overlapping outcome population globally for a horizon, then partitions raw and independent observations by the exact point-in-time market regime stored at the signal close. Missing historical regimes remain explicitly unclassified.
Threshold and regime cells require at least 8 independent observations for interpretation. Smaller cells remain visible as descriptive insufficient-data results, and threshold variants can share source events so they are comparisons rather than independent experiments.
Research, not a recommendation. Live market data can be delayed or temporarily unavailable. Historical results are displayed only when they have been computed and stored by the database research pipeline.