Does adding a positioning portfolio to LB6 make it more efficient?
FRAME
- Subject: LB6 Config B (pak Yon's frozen six-engine configuration as replayed by the trading-infra oracle; simulated tradelist, hourly mark-to-market, 1/6 capital per engine) and a positioning portfolio from Binance Vision BTCUSDT perp data.
- Why it matters: a second strategy on different data is worth adding only if it lowers risk more than return.
- Prior work: positioning-sweep-diverse (SURVIVED, modestly): a 3 long + 3 short positioning portfolio found on 2020-2024 made +6.6% in 2025-26 with a worst fall of 12.9.
- Scope: in-sample, 2025-04-23 10:00 to 2026-08-24 02:00 UTC. Positioning search N = 29,160 on this window, thresholds = percentiles of this window. Locked in
docs/bets/2026-11-18-lb6-plus-positioning.md, commit7cc0b7b. - Terms: plateau = median Calmar of the 3x3 threshold neighbourhood, centre >= 30 trades; portfolio = 3 long + 3 short by plateau, one per feature pair, each feature in at most 2 rules, daily correlation < 0.5; combination = half capital each.
BET
LB6 plus a positioning portfolio is more efficient than LB6 alone.
- Adjective: more efficient
- why: a second source only helps if it lowers risk more than it lowers return.
- Measured by: Calmar on hourly mark-to-market; higher wins.
- why: the measure of every LB6 report here.
- Because: different data; if bad stretches differ, a mix falls less deep.
- Falsified if: Calmar(combination) <= Calmar(LB6 alone).
KILL
| test | falsified if | result | status |
|---|---|---|---|
| Calmar, half LB6 + half positioning | <= 27.99 | 18.90 | FAIL |
| variant | total_pct | max_dd_pts | calmar |
|---|---|---|---|
| LB6 alone | +39.9 | 1.43 | 27.99 |
| positioning portfolio alone | +28.8 | 4.70 | 6.12 |
| half LB6 + half positioning | +34.4 | 1.82 | 18.90 |
| holding BTC | -7.9 | 69.35 | -0.11 |
Daily correlation, LB6 vs positioning portfolio: +0.28. Positioning portfolio over the last 10 days: +4.8.
Also observed, not deciding: smaller positioning shares.
| positioning share | total_pct | max_dd_pts | calmar |
|---|---|---|---|
| 0% | +39.9 | 1.43 | 27.99 |
| 10% | +38.8 | 1.32 | 29.33 |
| 20% | +37.7 | 1.27 | 29.63 |
| 30% | +36.6 | 1.41 | 26.03 |
| 50% | +34.4 | 1.82 | 18.90 |
Positioning rules (in-sample):
| rule | trades | total_pct | max_dd_pts | calmar |
|---|---|---|---|---|
| short: funding_rate >= p80 and taker_share_24h <= p30 | 59 | +41.3 | 10.84 | 3.81 |
| long: oi_vs_7d <= p30 and global_ls_chg_24h <= p20 | 47 | +36.6 | 16.11 | 2.27 |
| long: global_ls <= p40 and funding_chg_24h >= p90 | 43 | +24.9 | 9.50 | 2.62 |
| long: funding_rate <= p20 and funding_chg_24h >= p50 | 38 | +27.0 | 9.88 | 2.74 |
| short: vol_24h >= p60 and oi_vs_7d >= p90 | 36 | +25.6 | 11.49 | 2.23 |
| short: vol_24h <= p80 and taker_share_24h <= p30 | 111 | +17.4 | 20.72 | 0.84 |
Placebo (note, not deciding): the positioning search on 10 shifted datasets gave portfolio Calmar min 7.08, median 10.48, max 14.41; real 6.12.
UPDATE
- Status: FALSIFIED.