Abstract
We re-examine a published funding-rate mean-reversion strategy on BTC-USDT perpetual futures under a frozen specification. The core effect replicates and survives benchmark controls and out-of-sample testing. Parameter stability is weak — the edge concentrates in a narrow entry-threshold band — and execution costs are material at realistic participation. We classify the examination as keep-researching: the mechanism is real, the tradability is unproven.
§1 · Specification
The specification froze before the first run: Enter short perp exposure when 8h funding ≥ +0.10% (annualized ≥ 109.5%); exit at funding ≤ +0.03%. Mirror for negative extremes. 2020-09 → 2025-06 · out-of-sample boundary frozen at 2024-03 Failure criterion: net-of-cost sharpe must exceed 0.8 oos with max drawdown under 25% to count as reproduced.
§2 · Data check
Two conflicts were found and resolved: Nov 2022 — divergent open-interest snapshot during venue outage window. Segment flagged and excluded from sensitivity runs. sha256-pinned · point-in-time verified · no survivorship issue (single instrument)
§3 · Validation results
- Replication
- PASS
core effect reproduces: funding-capture PnL positive in 7 of 8 subperiods
- Controls
- PASS
clears buy-and-hold and rule-matched random entry on the same window
- Out-of-sample
- PASS
holds beyond the frozen boundary; decay present but net-positive
- Parameter stability
- WEAK
entry threshold sensitive: effect halves outside the 0.08–0.13% band. Flagged as the main open risk
- Execution costs
- MATERIAL
taker fees + slippage consume 38% of gross capture in the base case; break-even at 9 bps per side
§4 · Stress tests
Funding regime shift
2021-style sustained positive funding: strategy flips to persistent short-carry — PnL holds, turnover rises 2.4×
Venue outage
Exchange halt during held position: mark-to-market path shows 11% adverse excursion before exit
Crowding
Simulated 3× participation: capture per trade decays ~55%; capacity flagged as limited
Monte Carlo
Block bootstrap of funding paths: 5th-percentile outcome is breakeven, not ruin — no tail blowup found
§5 · Verdict
The core funding-reversion effect replicates, but parameter stability is weak and execution costs are material. Promising, not yet classifiable.
Figure 1 — Examination pipeline
- Idea
- Specification
- Data Check
- Validation
- Verdict
Figure 1. Every examination follows the same five-stage path. The specification freezes before any code runs; the data check precedes all validation; the verdict derives from mandatory findings.
Provenance
- Data snapshot
- Frozen, checksummed (sha256). Adjustment and reconciliation methods recorded.
- Code version
- Pinned at run time. Deterministic re-run verified.
- Checker
- Independent reviewer with falsification mandate. Findings signed.
- What this report does not establish
- Future funding regimes, live execution quality, capacity at size, or suitability for any account.
Inspect this examination in the interactive workspace on the front page.