BtcDailyFundingExtremeNegativeSqueezeLong

BTC Daily Funding-Rate Extreme-Negative Short-Squeeze Long with 200-SMA Bull Regime Filter

A long-only single-instrument FUNDING-RATE-BASED CONTRARIAN strategy on BTCUSDT perpetual futures using daily bars + funding-rate supplementary data, with a 200-SMA bull regime filter. This proposal is GENUINELY NOVEL within the portfolio: every existing strategy uses price-action signals (OHLCV-only). NONE use derivative-microstructure signals from funding rates. The accumulated failure pattern this session — three consecutive mechanism-transfer abandonments (ETH RSI(2) MR, BTC BB Lower MR, SOL RSI(2) Overbought) — has documented that mean-reversion and momentum-continuation edges are NOT transferable. Rather than attempt another transfer, this proposal introduces a FUNDAMENTALLY DIFFERENT SIGNAL CLASS that is self-contained and economically grounded. The economic mechanism: BTC perpetual funding rates measure the cost imbalance between long and short positions. When funding goes extremely negative (e.g., bottom 10% of trailing 30-day distribution), shorts are paying longs heavily — this signals over-leveraged short positioning. Three forces then converge to drive a long-side squeeze: (a) carry-cost pressure forces shorts to close positions (they're paying funding to longs every 8 hours), (b) arbitrageurs systematically take the long side to collect the funding premium, (c) any catalyst that triggers short-covering creates a cascade because shorts are crowded. This is documented behavior in crypto perp markets (per BitMEX/Binance research on funding-rate anomalies, 2019-present). Crucially, this is NOT a mean-reversion or momentum-continuation transfer from another strategy — it's a contrarian-on-derivative-positioning thesis with its own economic foundation. BTC is chosen because: (a) BTC has the most reliable daily-bar data path (validated by multiple in-pipeline strategies), (b) BTC has the deepest perpetual market with the most reliable funding-rate data, (c) BTC's institutional flow magnifies the squeeze effect (ETF rebalances respond to derivative positioning), (d) BTC's funding-rate history extends back 5+ years on Binance, providing sufficient backtest data. The 200-SMA filter eliminates the dominant failure mode (extreme negative funding during structural bear markets, where shorts are RIGHT and squeezes don't materialize). Only 4 explicit parameters (funding lookback window, funding percentile threshold, regime SMA period, stop pct).

Iteration 3 fix for Layer-3 'no trades'. Root cause: _ensure_funding filtered funding-dict keys with `if not isinstance(k, int): continue`, intending to skip the dual string keys. But when the sandbox's funding_rates dict arrives with string keys (JSON round-trip) or numpy.int64 keys (isinstance(numpy.int64, int) is False), that filter dropped EVERY entry, leaving the funding series empty so the percentile gate always returned None and should_enter never fired. Fix: coerce any numeric-looking key via int(k) inside a try/except, accepting int, numpy-int, float, and numeric-string keys; dual keys for the same timestamp still dedupe safely because entries collapse into pairs[ts]=rate keyed by parsed nanosecond timestamp. All Layer-1/Layer-2 logic (regime band, percentile computation, exit rules, sizing) is unchanged, so earlier passing layers are preserved. No price-only fallback added — funding remains the sole entry edge.

Not worth the 2-hour optimization budget: the funding-squeeze signal has essentially no edge on BTC daily and optimization cannot manufacture one from a near-zero base. Sharpe 0.102 with sharpe_ci_low -1.1688 (CI straddles zero), probabilistic_sharpe 0.5626, total_return 2.81% over 6.5 years (~0.4%/yr), profit_factor 1.042. The per-trade edge (~$14.74, ~0.07% on a 20% position) is below trading costs (commission 5.49% of gross), and the strategy carries negative alpha (-0.0058) and negative information ratio (-0.7316) vs holding BTC. Nearly the entire tiny return is a single year (2023 +6.6%), outlier-shaped (kurtosis 27.7), with toy capacity ($1.89M) and 23% impact drag. The strategy is functional and the signal class is genuinely novel and well-reasoned, but 'runs correctly' is not 'has an edge' -- starting from Sharpe 0.10 with a CI straddling zero, any Sharpe an optimizer finds over 225 trials would be curve-fit noise that fails the deflated-Sharpe and holdout gates. FAILURE PATTERN: a novel, economically-grounded signal (extreme-negative funding contrarian long) with correct plumbing still fails the review gate when the base backtest shows a near-zero, sub-fee, single-year-carried edge with a Sharpe CI straddling zero -- novelty and a working implementation do not substitute for a measurable edge net of costs.

Long-only BTCUSDT.BINANCE daily-bar strategy that buys when the perpetual funding rate falls into the bottom 25th percentile of its trailing 30-day distribution (crowded/over-leveraged shorts) while price is within a 5% band of, or above, its 200-day SMA (not a structural bear). Exits on funding normalization above the 50th percentile, an 8% stop, a 15% take-profit, or a 20-day time stop. Funding rate is a derivative-microstructure signal, not price action.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: BTCUSDT.BINANCE-1-DAY-LAST-EXTERNAL, Bars processed: 2369 Diagnostics: should_enter() returned a side 0 times over 2168 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Genuinely novel, economically-grounded signal class (perp funding as a directional contrarian signal) — distinct from the portfolio's price-action strategies

Functional and on-mechanism: funding plumbing fixed, 109 trades signaled -> 109 submitted, correct long-only direction, adequate trade count (~17/yr)

Low drawdown (11.97%) and low market beta (0.026)

No exploitable edge: Sharpe 0.102 with sharpe_ci_low -1.1688 / ci_high 1.3048 (CI straddles zero); probabilistic_sharpe 0.5626. Total return 2.81% over 6.5 years (~0.4%/yr), profit_factor 1.042.

Per-trade edge is below trading costs: avg_trade_return ~$14.74 (~0.07% on a 20% position) vs commission_pct_of_gross 5.49% — fees consume nearly all the gross edge.

Negative information_ratio (-0.7316) and negative alpha (-0.0058) vs holding BTC (benchmark_meaningful=true).

Return is a single-year artifact: nearly all of the tiny total came from 2023 (+6.6%); every other year flat-to-negative, kurtosis 27.7 (outlier-shaped).

Toy capacity ($1.89M) with 23.0% market-impact drag.

BtcDailyFundingExtremeNegativeSqueezeLong broke from the session's pattern of failed mechanism transfers by introducing a genuinely new signal class: contrarian long entries triggered by extreme-negative BTC perpetual funding (crowded shorts), gated by a 200-SMA bull-regime band. The implementation was correct and on-mechanism — funding plumbing fixed, 109 long trades, low drawdown — but the edge simply wasn't there: Sharpe 0.102 with a CI straddling zero, a +2.81% total return over 6.5 years that was almost entirely a 2023 artifact, per-trade profit below fees, and negative alpha and information ratio versus holding BTC. Judging that optimization could only manufacture curve-fit noise from a near-zero base, the reviewer abandoned it at the backtest-review gate after 3 iterations, recording that a novel, well-reasoned, working signal still cannot substitute for a measurable edge net of costs.

Novelty and correct plumbing do not substitute for a measurable edge — an economically grounded, self-contained signal class still fails when its base backtest is near-zero-Sharpe, sub-fee per trade, single-year-carried, and negative on information ratio versus buy-hold.

The backtest-review gate returned 'abandon' before optimization ran (optimization_report is null): starting from Sharpe 0.10 with a CI straddling zero and a sub-fee, single-year-carried edge, any Sharpe found over 225 optimization trials would be curve-fit noise that fails the deflated-Sharpe and holdout gates — a working implementation of a novel signal is not the same as a measurable edge net of costs.

A long-only BTC daily strategy using a genuinely novel derivative-microstructure signal — entering long when the perpetual funding rate fell into the bottom 25th percentile of its trailing 30-day distribution (crowded, over-leveraged shorts) while a 200-SMA regime band confirmed the market was not in a structural bear, exiting on funding normalization, a stop, take-profit, or 20-day time stop.

The 6.5-year backtest was functional and on-mechanism (109 clean long trades, ~17/yr, low 11.97% drawdown, low beta 0.026) but showed essentially no edge: Sharpe 0.102 with a bootstrap CI floor of -1.1688 straddling zero, profit factor 1.042, total return just +2.81% (~0.4%/yr), a negative information ratio (-0.7316) and negative alpha versus holding BTC. The per-trade edge (~$14.74, ~0.07%) sat below trading costs (commission 5.49% of gross), nearly all the tiny return came from 2023 (+6.6%), and capacity was toy-scale ($1.89M) with 23% impact drag.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: BTCUSDT.BINANCE-1-DAY-LAST-EXTERNAL, Bars processed: 2335 Diagnostics: should_enter() returned a side 0 times over 2134 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.