BtcFundingExtremeMomentumContinuationLS

BTC Perp 4H Funding-Extreme Momentum Continuation (Single-Instrument USD-M, Long-Short, Major-Asset Generalization Test of Validated Edge)

A long-SHORT, SINGLE-INSTRUMENT, single-venue MOMENTUM-CONTINUATION strategy on BTCUSDT.BINANCE USD-M perpetual futures using 4-HOUR bars, applying the validated funding-LEVEL-extreme momentum-follow template (SOL version advanced on analyst decision) to a MAJOR asset to answer the open portfolio-construction question: does the funding-momentum edge generalize to the most efficient, least-funding-extreme market, or is it structurally alt-only? Funding is a SIGNAL only (never credited as a cash flow; P&L is fully price-based and creditable). Thesis: extreme funding marks strong, persistent leveraged-positioning demand that CONTINUES in price (the session proved FADING it loses across SOL/ETH, and FOLLOWING the LEVEL extreme works — while funding-DERIVATIVE momentum failed on AVAX, PF 0.78, confirming the edge is in the funding LEVEL, not its rate-of-change). BTC is the deliberate hard test: its funding is the least extreme of any liquid perp (most efficient market), so this is the strongest generalization test of the edge — per-asset z-score normalization fires on BTC's RELATIVE funding extremes, keeping trade count adequate at 4H over the full multi-year history. If it works on BTC the edge is robust and market-wide (build the book everywhere); if not, the funding-momentum book should stay alt-focused. Either outcome is decision-relevant, distinguishing this from a redundant 4th-alt copy. It stays on the only confirmed-testable substrate (OHLCV + funding, both dense across full history) on the clean single-instrument USD-M MARGIN path, avoiding every structural blocker: NOT supplementary-data-gated (OI/liquidations/options-IV recent-only, untestable), NOT BINANCE_SPOT/CASH (metric/fill bug), NOT COIN-M (not backtestable), NOT cross-venue/Hyperliquid (basis artifact), NOT a multi-instrument pair (extra-leg feed bug), NOT funding-carry (no uncreditable cash flow), NOT a breakout (overfits), NOT a funding-veto overlay (proven inert), NOT a funding-derivative (AVAX-acceleration anti-predictive), NOT a slow regime signal (untestable trade count). It is distinct from both pending BTC strategies: the price-based time-series momentum (1D) and the FAILED daily smoothed-funding-sign regime (low-turnover) — this is 4H funding-LEVEL-extreme, the validated SOL construction. Kept to ~4 parameters.

The Layer-3 failure was 'No trades / should_enter returned a side 0 times over 13919 evaluated bars', and should_enter requires len(self._level_series) > z_lookback. The root cause was _build_funding's key filter `if not isinstance(k, int): continue`: supplementary_data['funding_rates'] is a flat {ns: rate} map whose keys arrive as STRINGS after the backtest subprocess's JSON serialization (the data spec documents the dual int+str keying), so isinstance(k, int) was false for every key, _fund_ts/_fund_rate stayed empty, _prevailing_funding returned None on every bar, _level_series never grew, the z-score was always 0, and the entry threshold (|z| >= 1.5) could never be met. The minimal fix replaces the isinstance check with a try/int(k) parse (accepting both int and string keys) and de-duplicates timestamps (the dual keys map to the same ts), exactly the robust pattern used by the landed funding strategies. With the funding series now populated, the 120-bar z-score reaches +/-1.5 regularly over BTC's multi-year history, producing the expected trade count. Every other component -- the prevailing-funding causal lookup, the rolling z-score, the momentum-follow entry gate, the normalize/stop/time exits, the continuous z+price blended signal, and the precision-aware sizing -- is byte-identical, so Layers 1-2 stay green and the 'fail-loud, no price-only fallback' design is preserved (if funding is genuinely empty the series stays empty and the strategy stands aside). leverage stays 1.0 and is not referenced as an amplifier, so the unused-leverage gate does not apply. This restores the generalization test of whether the validated funding-level-extreme momentum edge (advanced on SOL) holds on BTC, the least funding-extreme major.

The BTC generalization test of funding-extreme momentum-follow returned a decisive negative answer: the edge does NOT generalize to BTC. Over a dense 482-trade sample it loses -17.9% (CAGR -3.5%) with Sharpe -1.69 and the ENTIRE confidence interval below zero (-2.85 to -0.66), probabilistic_sharpe 0.0016, PF 0.65, sortino -2.09, and negative payoff geometry (avg_win $182 < avg_loss $214) — and it is negative in ALL 7 years (2020-2026). This is not noise, sparsity, or regime-dependence; following funding-level z-extremes on the most efficient major is systematically anti-predictive (the efficient market mean-reverts after a funding extreme rather than continuing), the opposite of the thesis. The hypothesis pre-committed to the decision rule that a BTC failure means the funding-momentum book stays alt-only, and that is the meta-finding here. No optimization can help — no entry_z/exit_z setting turns a robustly negative, all-years-losing signal positive, and a best-of-225 sweep would fit noise against an 11.3%-of-gross fee drag. FAILURE PATTERN: funding-as-directional-signal strategies have shown no exploitable forward-price edge on single instruments all session (SOL/DOGE momentum-follow were mania-carried, AVAX acceleration anti-predictive, ETH settlement/regime null); the BTC level-extreme test confirms the family does not generalize to efficient majors and is decisively negative there. The genuinely promising funding construct remains the negative-funding-CARRY basket, where funding is the credited cash flow rather than a price-direction predictor.

Long-short single-instrument funding-LEVEL-extreme momentum-continuation on BTCUSDT.BINANCE USD-M perpetual futures, 4-hour bars. It z-scores the prevailing 8h funding level over a rolling 120-bar window and follows the extreme: long when funding z >= +1.5 (persistent crowded-long demand), short when z <= -1.5; it exits when the z-extreme normalizes (|z| < 0.5 for the held side), on a 5% stop, or after a 30-bar max hold. Funding is a signal only (never credited as a cash flow); P&L is fully price-based. Iteration 2 fixes the zero-trades bug: the funding series builder now accepts string-keyed timestamps (the subprocess JSON round-trip stringifies the funding dict keys), so the funding level series populates and the z-score gate can fire. Four core parameters.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: BTCUSDT.BINANCE-4-HOUR-LAST-EXTERNAL, Bars processed: 14040 Diagnostics: should_enter() returned a side 0 times over 13919 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Clean, decisive experiment: dense 482-trade sample, single-instrument USD-M, funding-as-signal-only, no exposure bug (avg_position 10.3%), metrics_reliable — the generalization test was well-constructed and gives an unambiguous answer

Per-asset z-normalization correctly produced adequate trade count on BTC as intended

Robustly negative: Sharpe -1.69 with the ENTIRE CI below zero (-2.85 to -0.66), probabilistic_sharpe 0.0016, PF 0.65, sortino -2.09, expectancy -$42/trade

Negative in ALL 7 years (2020-2026) — a systematic losing edge, not regime-dependent or sparse

Anti-predictive on BTC: following funding-level extremes loses consistently, implying the efficient major mean-reverts after funding extremes (opposite of the continuation thesis); avg_win $182 < avg_loss $214

High fee drag (commission 11.3% of gross) compounds the negative signal

>0 to be viable

negative 7/7 years

majority positive

BtcFundingExtremeMomentumContinuationLS was a deliberately decision-relevant generalization test: take the funding-level-extreme momentum-follow construction and, via per-asset z-normalization, apply it to BTC — the hard case, the least funding-extreme major — to learn whether the edge is market-wide or structurally alt-only. The experiment was clean and dense (482 trades, no exposure bug) and returned an unambiguous answer: a -17.9% return with Sharpe -1.69, the entire confidence interval below zero, a profit factor of 0.65, and losses in every one of seven years. The analyst abandoned it at the backtest-review gate on its second iteration, concluding the edge is systematically anti-predictive on the efficient major and invoking the hypothesis's own pre-committed rule that a BTC failure keeps the funding-momentum book alt-only. It joined the session's broader finding that single-instrument funding-as-direction has no exploitable edge, pointing instead to the negative-funding-carry basket as the promising path.

Following funding-level extremes on the most efficient major (BTC) is systematically anti-predictive, confirming the funding-as-directional-signal family does not generalize to efficient majors and is decisively negative there — per the hypothesis's own pre-committed decision rule, the funding-momentum book should stay alt-only, while the genuinely promising construct remains the negative-funding-CARRY basket where funding is the credited cash flow, not a price predictor.

On its second and final iteration the analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate, so optimization, analysis, and risk review never ran: the experiment cleanly answered its own question — the funding-momentum edge does NOT generalize to BTC, it is systematically anti-predictive (the efficient major mean-reverts after a funding extreme rather than continuing), and no entry/exit-z setting can turn an all-years-losing signal positive.

A long-short, single-instrument momentum-continuation strategy on BTCUSDT.BINANCE USD-M 4H perpetual futures that took a per-asset rolling z-score of the prevailing funding LEVEL (signal only, never carry) and traded WITH the extreme — long on z >= +entry_z, short on z <= -entry_z, exiting on z-normalization — explicitly a generalization test of whether the validated alt funding-level momentum edge extends to BTC, the least funding-extreme (most efficient) major.

On a dense 482-trade sample the answer was decisively negative: total return -17.9%, CAGR -3.5%, Sharpe -1.69 with the ENTIRE CI below zero (-2.85 to -0.66), probabilistic Sharpe 0.0016, profit factor 0.65, Sortino -2.09, expectancy -$42/trade with negative payoff geometry (avg win $182 < avg loss $214), and negative in ALL 7 years (2020-2026), compounded by 11.3%-of-gross fee drag.