BTC Funding-Squeeze — 1h Edition

Status: Paper‑trade candidate · Library entry #5 · Validated 2026‑08‑06 Instruments: BTCUSDT perpetual only · Timeframe: 1h · Direction: long only Frequency: ~17 trades/yr · Requires a futures fee tier ≤ 0.05%/side taker

Why this is its own entry

This is entry #1's mechanism — fade crowded shorts when funding goes negative — traded on 1h instead of 4h. It earns a standalone page under the library's narrow‑clearer rule, stated plainly:

• It is BTC‑only. ETH is negative and BNB window‑flips at 1h even at futures fees. The mechanism is not single‑symbol curve‑fit — it passed cross‑asset (BTC + BNB) at 4h — but its 1h expression concentrates where the perp market is deepest and funding says the most.
• It needs real futures fees. At the 0.05%/side taker standard it clears every check below; at 0.1%/side (spot‑level costs) it degrades to marginal and does NOT clear.
• It was the only sub‑daily strategy to survive six research sweeps (research log): every momentum, mean‑reversion, session, breakout, passive‑fill and volume hypothesis tested on low timeframes died on costs or regime instability. The durable sub‑daily edge it found reads market internals, not price shapes. The second sub‑daily entry, #7, reads neither — it reads a settlement calendar.

The setup (compilable spec)

Market: crypto BTCUSDT perpetual, 1h trading timeframe. Longs only. Entry (1h, same bar): funding_rate < 0 AND RSI(14) crosses back above 30. Exit: trailing ATR stop, 2×ATR(14). Fees: 0.05%/side (Binance USDT‑M standard taker) · funding charged/credited exactly.

Identical conditions to entry #1 — only the timeframe and the (real) fee schedule differ.

Validation evidence — full funding history, two independent windows

Data note. The price path is Binance spot klines; the funding series and the fee schedule are USDT-M perpetual. Spot gives the cleaner, longer candle history and the two track closely, but perp wicks, basis and liquidation cascades do not appear in a spot candle. The funding the trigger reads is the real perpetual series.

• Two windows: ✅ both independent multi‑year regimes positive, PF ~1.4 in each.
• Strip‑best‑trade: ✅ in both — 122 trades total, no single trade carries it (~3× the sample of the 4h form).
• Re‑derived from committed code (2026‑10‑09): both rows reproduce to the decimal (scripts/research/replicate-sweep4-corrected-costs.ts; the research log's replication section).
• Cost reality: the table uses the actual futures schedule. At 0.1%/side the same spec is BTC‑marginal (PF ~1.2, a 20.5% DD breach) — check your fee tier before adopting.
• Monte‑Carlo: the W1 shuffled p95 drawdown is 24.5 sum‑points — size the sleeve for that, not the 17% you observed.
• The edge arrives lumpy: the most recent ~24 months inside W2 were roughly flat (about +2% at PF ~1.1) — W2 still clears because the 2022–2024 stretch paid. Expect long flat spells; that is what a PF‑1.4 workhorse looks like up close, and it is exactly what the app's drift alerts watch for once the plan is live.

Null test (2026‑09‑18) — clears it

The vetting bar's sixth check: same window, same 2×ATR exit, same fees, entry signal replaced by 66 random days, 400 times. Real +12.9% against a null median of −4.8%, null 95th percentile +12.5% — the 95.5th percentile. The detail worth noticing is the sign of the null: random 1h entries with this exit lose money, and the funding signal is what turns that into a profit. Run fees‑only on both sides, matching this doc's own tables.

Sweep #12 reference run (2026‑10‑08) — the recent window does not clear the bar

Sweep #12 re-ran this exact config as its reference row — on perpetual candles (the tables above use spot), on that sweep's windows, with real funding charged on both the strategy and its random-entry null:

The first window clears every check, the null included. The second is the flat stretch the evidence section above already describes, now measured against the full bar: it fails strip‑best, and with this exit and real funding it cannot be told apart from random entry timing.

Inside that flat total, though, the edge comes and goes rather than being gone. By half-year of entry, on the same run:

The last twelve months (Oct 2025 → Oct 2026): 33 trades, +8.5%, profit factor 1.57, still positive with the best trade removed, 7.7-point drawdown — at the 93.8th percentile of random entry, just short of the 95th the bar asks for. Over the last 24 months it is +1.9% (76th).

Different windows and candles from the validation tables, so the validation stands as published. What the recent evidence says is that this is a regime-dependent edge: it paid in 2022, went quiet from 2023 to mid-2025, and has been paying again for the past year. Paper-trade it, don't size it on the 2019–2024 numbers, and expect long quiet spells.

Guardrails (read before trading)

• BTC only. Never alts, not even ETH. At 1h the edge does not exist elsewhere.
• Futures taker fees ≤ 0.05%/side or don't trade it. This is a thin, high‑frequency‑ish edge; one fee tier is the difference between PF 1.4 and marginal.
• PF ~1.4 is a workhorse, not a moonshot — ~15%/window with shallow drawdowns. Its virtues are the sample size and the internals‑based mechanism, not the raw return.
• Prefer the 4h original if you want the cleanest per‑trade edge (PF ~2.5) at lower frequency and spot‑conservative costs; use the 1h edition when you want ~3× the sample and can execute on futures fee tiers.
• Same asymmetry as entry #1: long only — shorting crowded longs has no edge.
• 1h is the floor — do not go lower. The same frozen spec at 15m is negative on BTC in both windows even at futures fees (research log, sweep #7). One spec, three timeframes: strong at 4h, BTC‑only at 1h, gone at 15m.

How it was found

Sweep #4's stated market‑structure correction (research log): prior sweeps had priced perp strategies at spot‑default fees (0.1%/side), a 2× overcharge for the instrument actually traded. Re‑validated verbatim at the real schedule, BTC‑1h cleared every per‑asset check across the full funding history. Nothing was tuned — the spec is entry #1's, frozen since 2026‑07‑31.