Does a portfolio of positioning rules beat holding BTC from 2025?
FRAME
- Subject: BTCUSDT USDT-M perpetual, hourly, Binance Vision (1h klines, fundingRate, metrics).
- Why it matters: LB6's strength is a portfolio of weak, partly independent rules found by a threshold sweep; this applies the same process to public positioning data.
- Prior work: positioning-sweep (FALSIFIED): single best rule, edge from the 2022 crash alone, lost 19.9% from 2025.
- Scope: locked sweep, N = 29,160; dev 2020-09-02 to 2024-12-31; out of sample 2025-01-01 to 2026-10-31 (a screen: used once before). Locked in
docs/bets/2026-11-18-positioning-sweep-lb6.md, commit193c659. - Terms: 10 features funding_rate, oi_chg_24h, global_ls, taker_buy_share, ret_24h, vol_24h, funding_chg_24h, oi_vs_7d, taker_share_24h, global_ls_chg_24h; rule = f1 op q1 and f2 op q2, one side, held 24 h after the last true hour, fee 0.05%/side, funding included; plateau = median Calmar of the 3x3 threshold neighbourhood; portfolio = up to 6 rules by plateau, one per feature pair, daily correlation < 0.5 on dev, equal weight.
BET
A portfolio of the best positioning rules is more efficient than holding BTC out of sample.
- Adjective: more efficient
- why: rules that trade part of the time must earn more per unit of risk than holding.
- Measured by: portfolio Calmar out of sample; higher wins.
- why: same measure as LB6 and every report here.
- Because: LB6's strength is a portfolio of weak, partly independent edges.
- Falsified if: portfolio Calmar out of sample <= max(holding BTC, 0).
KILL
| test | falsified if | result | status |
|---|---|---|---|
| portfolio Calmar out of sample | <= 0.00 | -0.05 | FAIL |
Portfolio: dev total +204.6%, max drawdown 14.9, Calmar 13.76; out of sample total -0.8%, max drawdown 16.0, Calmar -0.05. Holding BTC: dev +292.1% / 125.9 / 2.32; out of sample -1.5% / 69.3 / -0.02.
Rules (top5_share = share of the rule's dev result from its 5 best trades):
| rule | dev_trades | dev_plateau | dev_calmar | top5_share | oos_trades | oos_total | oos_calmar |
|---|---|---|---|---|---|---|---|
| long: global_ls <= p50 and taker_share_24h >= p90 | 155 | 11.83 | 15.82 | 23% | 96 | -2.6 | -0.09 |
| long: global_ls <= p60 and ret_24h <= p60 | 254 | 8.96 | 9.86 | 20% | 100 | -10.1 | -0.21 |
| long: funding_rate <= p20 and global_ls <= p30 | 95 | 8.83 | 10.88 | 38% | 52 | +3.3 | 0.12 |
| long: global_ls <= p30 and oi_vs_7d <= p50 | 121 | 7.35 | 6.62 | 53% | 57 | +7.3 | 0.32 |
| long: global_ls <= p20 and global_ls_chg_24h <= p10 | 113 | 7.02 | 8.54 | 38% | 56 | -3.4 | -0.14 |
| long: global_ls <= p50 and funding_chg_24h <= p10 | 108 | 6.81 | 4.34 | 58% | 8 | +0.6 | 0.10 |
By year (% of capital, summed):
| year | portfolio_pct | hold_btc_pct |
|---|---|---|
| 2020 | +35.1 | +94.5 |
| 2021 | +56.8 | +83.7 |
| 2022 | +29.6 | -82.5 |
| 2023 | +38.4 | +103.1 |
| 2024 | +44.9 | +93.2 |
| 2025 | -0.6 | +3.3 |
| 2026 | -0.2 | -4.8 |
Placebo, reported not deciding: the whole pipeline on 10 shifted datasets gave portfolio dev Calmar median 8.02, max 11.34; real 13.76.
UPDATE
- Status: FALSIFIED.
- Next bets:
positioning-sweep-diverse(Does a positioning portfolio forced to mix ideas (at most two rules per measure, long and short) hold up from 2025?)