jgrusewski
1cf38232e8
fix(surfer): full-16y MFT settle — regime-adaptive surfer significantly negative
...
Pulled 16y ES ohlcv-1m continuous (year-chunked, $20 credits, no 504) + light
to_ndarray loader (to_df OOM'd at 20GB). On 1.1M 5-min bars the 1.3y +0.50/t=0.67
top-5% hint collapsed: regime-adaptive top-5% = -0.52 ticks/trade t=-2.51
(significantly negative); ALL cells/signals/convictions significantly negative.
Decisive: no capturable intraday directional edge for crossing/non-colocated setup.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com >
2026-06-06 12:15:37 +02:00
jgrusewski
fa700c7b5f
feat(surfer): MFT regime-adaptive quality test (intraday, GPU, 4GB-safe)
...
Tests the actual surfer thesis: intraday 5-min decisions/15-min holds, flat-by-close,
regime-adaptive (ride trend / fade range / stand aside), quality-over-quantity
(top-conviction only). Clean ES OHLCV-1m -> 5-min bars, per-trade net edge in ticks,
IS/OOS + t-stat. Result: regime-adaptive top-5% = +0.50 ticks/trade OOS, sign-consistent,
but t=0.67 (not significant); all else significantly negative. Adaptive+selective is the
only non-losing structure (thesis directionally right, within noise). 4GB-safe: cumsum
rolling-std (no unfold) + alloc cap.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com >
2026-06-06 10:08:16 +02:00
jgrusewski
2bb2b50fd1
fix(surfer): correct roll handling (held-contract return stitching) + cost stress + dual gates
...
The +0.32 'edge' was a roll-zeroing artifact (discarded held-contract roll-day P&L).
Correct roll stitching + 2bp cost: floor Sharpe ~0.05 (sign) / 0.06 (cont) / 0.14
(cont+band), all NEGATIVE at 2x cost. Marginal, implementation-dominated edge.
Adds FLOOR_SIGNAL mode (sign|cont), no-trade band, 2x-cost stress, develop/deploy gates.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com >
2026-06-06 01:43:06 +02:00
jgrusewski
8df1c7eea2
feat(surfer): Phase 0 GPU floor + validation harness — first real (weak) OOS trend edge
...
PyTorch-GPU diversified-trend floor (TSMOM 1/3/12mo + inverse-vol + 10% vol-target)
+ CPCV/Deflated-Sharpe validation, over 22 CME futures x 19.7y (Databento GLBX
ohlcv-1d via budget-capped fetcher, ~$32 credits). Verdict: Sharpe +0.32, CPCV
median +0.30, IS +0.36/OOS +0.08 (sign-consistent), Deflated Sharpe ~0.92 at honest
n_trials. Real but weak edge; fails deploy-grade gates (correct), passes edge-exists.
Includes roll-neutralization fix (max-vol outright + zero roll-day returns) that
eliminated 988%/day continuous-contract artifacts (RB vol 550%->31%). Plus
install_torch_gpu.sh. Data gitignored.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com >
2026-06-06 01:31:37 +02:00
