Strategies

BTC Extreme Funding Rate Contrarian Short (8h)

Outcome: Abandoned

BtcExtremeFundingContrarianShort1H

Outcome Summary

BtcExtremeFundingContrarianShort1H faded crowded longs by shorting BTC on extreme positive 8h funding, a clean three-parameter mechanism orthogonal to every OHLCV strategy. It was the family's rare positive case — profit factor 1.30, Sharpe 1.09, no losing years — and notably worked on BTC where the SOL/ETH funding shorts had failed. But optimization exposed the catch: the edge lived almost entirely in 2020-2021 high-funding regimes, so the holdout had zero trades, the deflated Sharpe fell to 0.16, and the WF-OOS rested on one outlier window. The analyst abandoned it after one iteration, concluding the trigger has structurally decayed and can't be tuned back into existence.

Strategy report

Hypothesis

A short-only single-instrument MEAN-REVERSION strategy on BTCUSDT perpetual futures (Binance USD-M) triggered exclusively by extreme positive 8-hour funding rates. When the most recent funding rate settles above a threshold indicating crowded long positioning, the strategy opens a short and holds until funding normalizes or a price-based exit fires. The strategy uses NO chart-based signals (no SMA, EMA, candlestick patterns, breakouts, or trend filters) — entirely funding-rate-driven, which makes it structurally orthogonal to every OHLCV-based strategy currently in the pipeline and to all 10 recent failure patterns. Only 3 parameters total (entry threshold, exit threshold, time-stop hours), keeping it robust against overfitting. Data requirements are minimal: 1H bars (for execution/exit checks) plus funding_rate_updates from the supplementary data store.

Implements the hypothesis exactly with three free parameters and zero chart signals, making it structurally orthogonal to every OHLCV-based pipeline strategy. Entry/exit use absolute funding thresholds (matching the documented Binance funding magnitude range: extremes ~0.0003, normalization ~0.0001) rather than a percentile, keeping the design minimal and robust to overfitting. Funding is parsed from the documented flat funding_rates {ns: rate} dict (dual int/str keys deduped, timestamps normalized to ns) and the current rate is the most-recent settlement at/before the bar via bisect (no lookahead); 1H bars let price-stop and time-stop checks run hourly between 8h funding settlements, with a 16h nearest-past lag tolerance so each bar still sees the latest funding. Per the supplementary-data rules there is NO price-only fallback edge — if funding is missing the strategy stays flat (funding is the signal, not a creditable cash flow; the realized edge is the price mean-reversion after crowded longs deleverage). Exits encode the hypothesis's 'funding normalizes OR price-based exit': funding<=exit_threshold (primary), a 6% hard stop, and a 48h time stop. calculate_signal returns the funding rate in natural units so the signal is continuous and varies as funding settles (with a tiny ROC fallback before funding is available), and should_enter is SELL-only gated on the cached trigger. Short-only 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. BTCUSDT 1H is the most reliable data path with full funding coverage, avoiding data_unavailable; the multi-hour hold and 6% stop give per-trade moves above the ~0.10% round-trip taker cost.

Failed deflated Sharpe: DSR=0.16, expected-max=2.17 over 225 trials — the optimized Sharpe of 1.22 is below the best-of-N luck bar and statistically indistinguishable from noise. Compounded by a failed holdout (ratio 0.0, holdout_sharpe 0.0 — zero qualifying trades in the final window), PBO 0.68 (>0.5), is_significant=false, and a Sharpe CI [-0.72, 3.12] straddling 0. The walk-forward avg-OOS of 2.80 is a mirage from a single 9.02 outlier window (the other two OOS windows are 0.0 and -0.61), and although WF flags is_overfitted=false, the robustness block (DSR/PBO) shows the 225-trial selection overfit a sparse, outlier-heavy sample. Root cause is structural decay of the trigger: extreme positive 8h funding on BTC was a 2020-2021 phenomenon (returns concentrated in 2020 +12.2% / 2021 +5.0%, near-zero trade frequency afterward, zero trades in 2025-2026), so the holdout window is empty. The absolute edge is also marginal (total 8.5% over ~6yr, CAGR 3.95%). Not iterable: sensitivity is already cliff-free and the optimizer searched the whole space — no parameter change can create extreme-funding events that no longer occur on BTC. Not a confident revise either: although fading crowded-long funding is economically plausible and would need a venue/instrument with persistent positive-funding skew, there is no statistical evidence (DSR 0.16, Sharpe below the luck bar) that the mechanism produced a real edge even on BTC, so reframing would most likely re-overfit a similarly thin sample. FAILURE PATTERN: single-instrument extreme-funding contrarian fades on BTC have a 2020-2021-concentrated edge that decays to zero trade frequency in recent regimes — passes sensitivity but fails DSR/holdout with an outlier-driven WF-OOS.

Implementation

Short-only funding-rate-driven mean-reversion on BTCUSDT.BINANCE USD-M perpetual futures, 1-HOUR execution bars. It opens a SHORT when the most recent 8h funding rate settles at/above 0.03% (entry_threshold) — crowded long positioning paying shorts — and holds until funding normalizes to <= 0.01% (exit_threshold), a 6% price stop fires, or a 48-hour time stop elapses. Uses NO chart-based signals; entirely funding-driven. Three parameters (entry threshold, exit threshold, time-stop hours); 25% notional per short, leverage 1.0.

Backtest Review

Max drawdown: 0.00%
Win rate: 0.0%
Profit factor: 0.00

Genuinely positive edge: profit_factor 1.30, Sharpe 1.09, expectancy +$118/trade, win_rate 53%, positive in every year it traded (no losing years), good capacity ($21.6M).

Funding-extreme reversion works on BTC specifically (unlike the abandoned SOL/ETH funding shorts that bled against stronger alt uptrends); funding now credited adds carry to the short.

Headline understates realized edge: end_unrealized_pct -12.4% means a losing open short at backtest end, so realized return (~+21%) exceeds the 8.5% headline.

Trades match the hypothesis (177 short, clean diagnostics); minimal 3-parameter design.

Trades cluster in 2020-2021 and the strategy is dormant in 2022/2025/2026 (funding below threshold) — uneven distribution risks a degenerate walk-forward/holdout window with few/no recent trades.

Sharpe CI [-0.48, 2.83] is wide; commission 5.11% of gross.

No recent (2025/2026) validation — the edge is established mostly in the 2020-2021 high-funding regime.

Analysis

Clean sensitivity: 0 cliff parameters; min_notional and funding_lag are inert, per_leg/stop_loss robust.

Mechanism is orthogonal to OHLCV strategies and trades match the hypothesis (177 short-only fades on extreme positive funding); low costs (impact 5-7%, capacity $21-38M); not liquidated, modest max_drawdown 13-16%.

Failed Deflated Sharpe: DSR 0.16 with expected-max 2.17 over 225 trials — optimized Sharpe 1.22 is BELOW the best-of-N luck bar (best-of-N noise).

Holdout failed: ratio 0.0, holdout_sharpe 0.0 — zero qualifying trades in the final window (BTC extreme positive funding has structurally decayed).

PBO 0.68 (>0.5), is_significant=false, Sharpe CI [-0.72, 3.12] straddles 0.

WF avg-OOS 2.80 is a single-outlier mirage: OOS windows [0.0, -0.61, 9.02]; two of three non-positive.

Marginal absolute edge: total return 8.5% over ~6 years (CAGR 3.95%), concentrated in 2020-2021; near-zero trade frequency post-2021, zero in 2025-2026.

Outcome Summary

Single-instrument extreme-funding contrarian fades on BTC have a 2020-2021-concentrated edge that decays to zero trade frequency in recent regimes — a strategy can pass sensitivity yet fail deflated-Sharpe and holdout because its qualifying events have structurally disappeared, and uneven trade distribution is itself a fatal validation risk.

The analyst issued an 'abandon' verdict at the post-optimization ANALYZING stage: the trigger has structurally decayed — extreme positive 8h BTC funding was a 2020-2021 phenomenon, with near-zero trade frequency afterward and zero trades in 2025-2026, leaving the holdout window empty — and the optimizer overfit a sparse, outlier-heavy sample, so no parameter change can create funding events that no longer occur.

A short-only, single-instrument funding-rate mean-reversion strategy on BTCUSDT perpetual futures that used no chart signals at all: when the most recent 8h funding rate settled above an extreme-positive threshold (crowded longs), it opened a short on 1H execution bars and held until funding normalized below an exit threshold, a 6% stop, or a 48h time-stop — just three parameters.

The default backtest was the rare positive one in this family: 177 short trades, profit factor 1.30, Sharpe 1.09, expectancy +$118/trade, 53% win rate, no losing years, and good $21.6M capacity (with realized return ~+21% understated by a losing open short). But it cleared the gate and then failed validation: deflated Sharpe 0.16 (optimized Sharpe 1.22 below the luck bar of 2.17 over 225 trials), PBO 0.68, a holdout with zero qualifying trades (ratio 0.0), and a WF-OOS average of 2.80 that was a single-outlier mirage (windows [0.0, -0.61, 9.02]).

Backtest and paper results are hypothetical. Trading involves risk of loss.