EthFundingExtremeMeanReversionCarryLS8H

ETHUSDT Funding Rate Extreme Mean-Reversion Carry — Contrarian Position on 95th-Percentile Funding

A bidirectional carry strategy on ETHUSDT perpetual futures (Binance USD-M) that takes contrarian positions when the 8-hour funding rate reaches the 95th percentile (positive or negative) of its trailing 90-day distribution. When funding is extremely positive (longs over-leveraged, paying shorts heavily), the strategy goes SHORT to collect the funding payment AND benefit from the documented post-extreme-funding price mean reversion. When funding is extremely negative (shorts over-leveraged), the strategy goes LONG for the same dual edge. Holding period is naturally short (1-7 days) because funding rates mean-revert within 24-72 hours after extreme readings. This is structurally orthogonal to all 10 mechanism classes falsified in this session: it is not trend-following, not price mean-reversion, not cross-asset, not cross-sectional, not calendar-based, not sub-daily intraday, not accumulation, not cross-venue arbitrage, not channel breakout, and not oscillator-based. The signal is funding rate (a supplementary data feed with multi-year Binance coverage), and the edge is the well-documented carry+mean-reversion combination that has produced Sharpe ~0.8-1.2 in academic crypto funding studies (Schmeling-Brueggemann-Krempel 2022; Naranjo 2019; Lo-Sornette 2024 update).

Implements the documented carry + post-extreme mean-reversion combination as a bidirectional strategy, structurally distinct from the falsified mechanism classes (it is funding-percentile-driven, not price/trend/cross-asset/calendar/oscillator based). 'Extreme' is a self-calibrating percentile of the current funding rate within its trailing 90-day window (a single entry_pct knob sets the short tail at >=0.95 and, by symmetry, the long tail at <=0.05), so the trigger adapts as ETH's baseline funding regime drifts; the percentile uses only funding events at/before the current bar (bisect window), so there is no lookahead. Funding is parsed from the documented flat funding_rates {ns: rate} dict (dual int/str keys deduped, ns-normalized) and there is NO price-only fallback edge — if funding is missing the strategy stays flat. Critically, the backtest engine does not credit funding cash flows, so the exit targets funding NORMALIZATION (percentile back inside the [0.4,0.6] band) — the trade is held exactly while the dislocation persists, capturing the realized price mean reversion rather than relying on uncredited carry; a 21-bar (~7 day) time stop and 12% hard stop bound risk. calculate_signal returns (percentile - 0.5) so the signal is continuous and varies every bar (positive at the short tail, negative at the long tail), with a tiny ROC fallback before the window fills; should_enter is gated on the cached percentile and is long+short. Long+short requires a margin venue, so it routes to BINANCE USD-M futures; leverage=1.0 is referenced in sizing so config and the engine margin model agree. ETHUSDT 8H is a deeply liquid major with multi-year Binance funding coverage, avoiding data_unavailable; multi-period holds give per-trade price moves above the ~0.10% round-trip taker cost.

Not worth optimizing — the strategy loses even with funding credited. Per current ground truth the engine credits funding cash flows (since 2026-06-10), so this backtest's -30% INCLUDES the carry; the code comment claiming otherwise is stale. (I am explicitly NOT abandoning on any 'funding-uncreditable' basis.) The result is decisively negative: profit_factor 0.88, expectancy -$66/trade, Sharpe -0.29, total return -30% over 6 years, max_drawdown 40%, negative in 4 of the last 5 years (311-trade sample). Root cause: this is a SINGLE-LEG DIRECTIONAL contrarian bet (short the perp on high funding / long on low funding), NOT a delta-neutral carry — there is no spot hedge, so it carries full price risk. Shorting crowded-long ETH in bull regimes loses on price direction even while collecting funding, and the directional losses overwhelm the carry; this is the same failure as the SOL funding-contrarian short. The premise that a single-leg funding fade delivers a 'dual edge' is refuted — the price risk dominates. NOTE for research: a genuine funding-carry edge must be delta-neutral (long spot + short perp, or cross-venue), which the engine now credits correctly — that is a different strategy, not an iteration of this directional one. Stop proposing single-leg directional funding-fade strategies on trending majors; they fight the trend and lose.

Bidirectional funding-extreme carry + mean-reversion on ETHUSDT.BINANCE USD-M perpetual futures, 8-HOUR bars. It ranks the current Binance funding rate within its trailing 90-day distribution and takes contrarian positions at the tails: SHORT when funding is in the >= 95th percentile (longs over-leveraged), LONG when in the <= 5th percentile (shorts over-leveraged). It exits when funding normalizes (percentile back inside [0.4, 0.6]), after a ~7-day (21-bar) time stop, or on a 12% hard stop. ~25% notional per leg, leverage 1.0, funding-driven signal on OHLCV price execution.

Mechanism correctly implemented: no-lookahead funding percentile, bidirectional, clean diagnostics; funding data available; good sample (311 trades).

Genuinely orthogonal funding-driven mechanism; worked in the 2022 bear (+15.9%).

Loses net of credited funding: profit_factor 0.88, expectancy -$66/trade, Sharpe -0.29, total return -30% over 6 years, max_drawdown 40%.

Negative in 4 of the last 5 years (2024 -4.8%, 2025 -5.6%, 2026 -4.7%); only the 2022 bear was strongly positive.

Single-leg DIRECTIONAL bet, not a delta-neutral carry (no spot hedge) — full price risk: shorting crowded-long ETH in bull regimes loses on price despite collecting funding.

Same failure mode as the SOL funding-contrarian short reviewed earlier; commissions 4.25% of gross.

EthFundingExtremeMeanReversionCarryLS8H aimed for a documented carry-plus-mean-reversion edge by fading 95th-percentile funding extremes on ETH perps, going short into over-leveraged longs and long into over-leveraged shorts. The funding signal was implemented cleanly over a healthy 311-trade sample, but the strategy lost -30% over six years (profit factor 0.88, Sharpe -0.29) with losers outsizing winners and only the 2022 bear paying off. The analyst abandoned it at backtest review after one iteration, noting the loss persists even with funding credited because it's a single-leg directional bet that fights ETH's uptrend — exactly the SOL funding-short failure mode. The recorded guidance is to stop proposing single-leg directional funding fades and instead build a genuinely delta-neutral carry.

A single-leg funding fade is not a carry trade — it carries full price risk and fights a trending major's direction, so the 'dual edge' premise is refuted; a genuine funding-carry edge must be delta-neutral (long spot + short perp, or cross-venue), which is a different strategy rather than an iteration.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: with the engine now crediting funding cash flows, the -30% result already includes the carry, so the loss is real — the root cause being that this is a single-leg directional bet, not a delta-neutral carry, so shorting crowded-long ETH in bull regimes loses on price direction faster than funding can compensate (the same failure as the earlier SOL funding-contrarian short).

A bidirectional funding-rate carry/mean-reversion strategy on ETHUSDT perpetual futures (8H bars) that took contrarian positions at funding-rate extremes — shorting when the rate hit the 95th percentile of its trailing 90-day distribution (over-leveraged longs) and going long at the bottom tail — to capture both funding payments and post-extreme price mean reversion, exiting on funding normalization, a ~7-day time-stop, or a 12% hard stop.

Over 6 years and 311 trades it lost: profit factor 0.88, expectancy -$66/trade, Sharpe -0.29, -30% total return, and a 40% max drawdown, negative in 4 of the last 5 years with only the 2022 bear strongly positive (+15.9%). The win rate was 51% but average losses ($1,081) exceeded average wins ($916), and commissions added 4.25% of gross drag.