BTC Funding‑Squeeze Long

Status: Paper‑trade candidate · Library entry #1 · Found 2026‑07‑31 Instruments: BTCUSDT perp (primary), BNBUSDT perp (secondary) · majors only Timeframe: 4h · Direction: long only · Frequency: ~6 trades/yr/asset (low)

The edge

Fade crowded shorts. When the perpetual funding rate goes negative, shorts are paying longs — positioning is one‑sided short, and the market is primed for a short squeeze. Buy that fear as price bounces from oversold, and ride the squeeze out with a trailing stop.

This is a liquidity/positioning read, not a price‑pattern read — which is exactly why it survives when simple price strategies don't. Funding is a genuine, persistent inefficiency on the deepest perps, where funding actually reflects real positioning.

The setup (compilable spec)

Market: crypto BTCUSDT perpetual, 4h trading timeframe. Longs only. Entry (4h, same bar): funding_rate < 0 AND RSI(14) crosses back above 30. Stop: entry − 2×ATR(14). Exit: trailing ATR stop (~2×ATR), ride the squeeze. Direction: long only.

Compiled shape: entryConditions = funding_rate < 0 AND rsi(14) crosses_above 30; exitStrategy = atr_trailing, atrMultiplier ≈ 2, atrPeriod = 14; fees on.

Validation evidence

Held to the library vetting bar. Backtested via the same engine the app uses, with the warm‑up run‑in so indicators are valid from bar one.

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

Note (2026‑08‑01): these tables are effectively fees‑only — the engine's perp funding cost model (includeFunding) was silently a no‑op on intraday timeframes when this entry was validated (fixed in 616ca9f). Re‑run with funding correctly credited, results improve slightly (longs entered during negative funding receive it): BTC recent‑24mo PF 2.72 → 2.74, BNB prior‑24mo −8.7% → −6.7%. No conclusion changes.

• Two windows: ✅ BTC is net positive in both independent 2‑year windows.
• Strip‑best‑trade: ✅ on BTC (recent) and BNB — the result is not one lucky trade.
• Cross‑asset: ✅ BTC + BNB (and AVAX passes on the baseline, at higher DD). ETH marginal.
• Drawdown: ✅ 1–5% on the majors — very low.

Timeframe & direction (explored, net of fees)

The edge is real across timeframes, with a clear sweet spot — and it is long‑only, asymmetric:

1h update (2026‑08‑05, revised 2026‑08‑06 — research log): the row above is the recent 24‑month window only, and this doc's tables price fees at 0.1%/side — the SPOT‑taker default, deliberately conservative. The strategy trades perpetual futures, whose standard schedule is 0.05% taker. Re‑validated across the full funding history at the real futures schedule, BTC‑1h clears every per‑asset check: W1 (2019‑09→2022‑08) +18.2%, PF 1.39, max DD 17.0%, survives strip‑best; W2 (2022‑08→2026‑08) +13.1%, PF 1.36, DD 11.6%, survives strip‑best — 122 trades (~17/yr, ≈3× the 4h sample). BTC‑only: ETH is negative in W1 and BNB in W2 even at futures fees, so the 1h variant never gains a second asset. Guardrails: futures fee tier ≤ 0.05%/side required (at 0.1%/side it degrades to marginal — the 2026‑08‑05 note), BTC only, and the Monte‑Carlo p95 shuffled drawdown in W1 is 24.5 sum‑points — size for that, not for the 17% you observed. 4h remains the primary form; 1h is the larger‑sample BTC option for futures‑fee traders.

• Direction is asymmetric. Long crowded shorts works; the mirror (short crowded longs on extreme positive funding) has no edge — ~0 signals and net negative. Crypto's upward drift and short squeezes make fading the crowd a one‑sided (long) trade. So: long only.
• Trade it on 4h; BTC‑only 1h is legitimate on futures fee tiers. 4h works even at this doc's conservative spot‑level costs. The 1h variant clears the per‑asset bar only at real futures fees (≤ 0.05%/side) and only on BTC — see the 1h update above; not below 1h.

Null test (2026‑09‑18) — clears the recent window, marginal in the older one

The vetting bar's sixth check — the entry signal replaced by the same number of random days, 400 draws, everything else held fixed:

Random 4h entries with this trailing exit have a negative median return; the funding signal is what makes the difference. The older window's 90th percentile is short of the 95th the bar asks for — on a ten‑trade sample, which is the entry's known weakness and already the reason it ships as paper‑trade‑first rather than adopt‑ready. Both rows are run fees‑only, matching this doc's tables.

Guardrails (read before trading)

• Majors only. It fails, sometimes catastrophically, on volatile alts — SOL −51% (55% DD), ADA −22%, XRP/DOGE negative. On a dying alt, negative funding just means it's collapsing, and "buy the bounce" catches a falling knife. Never run this across the alt universe.
• Low frequency / modest return. ~6 trades/yr, ~13% over 24mo. It is not a moonshot — it's a small, low‑risk, defensible edge.
• Small samples. 9–12 trades per window; the prior BTC window leans on one trade. Treat it as a candidate: paper‑trade forward to confirm it holds live before real size.
• Attempts to broaden it failed honestly: a 4h‑200‑EMA filter kills the setup; a daily‑trend filter cuts the alt blow‑ups but thins the signal so much only BTC stays robust. The edge is genuinely a BTC/BNB phenomenon — refining further would be curve‑fitting.

How it was found

Ruled out first (none robust): mean‑reversion fake‑breakdowns (no stop rescues them), and trend‑following (4h loses; daily BTC looked great at +46% but was one trade — 0% without it, and a diversified basket lost −29%). The funding‑squeeze was the first idea to survive the strip‑best‑trade test on more than one asset and window. That's the whole point of the vetting bar: it killed the flattering results and kept the real one.