AvaxFundingAccelerationMomentumLS

AVAX Perp Funding-Acceleration Momentum (Single-Instrument USD-M, Daily, Long-Short, Building-Positioning Leading Signal)

A long-SHORT, SINGLE-INSTRUMENT, single-venue directional strategy on AVAXUSDT.BINANCE USD-M perpetual futures using DAILY bars, where the FIRST DIFFERENCE of the funding rate (its acceleration / rate-of-change) is the signal — funding used purely as a signal, never credited as a cash flow (P&L is fully price-based and creditable). Distinct from every pending funding strategy, which keys off the funding LEVEL extreme (momentum-follow), smoothed SIGN (regime-persistence), price-funding CONFLUENCE, the settlement CLOCK, or breakout VETO: this trades the CHANGE in funding. Thesis: when funding is rising rapidly (becoming more positive day over day), leveraged long demand is actively BUILDING — an accelerating-demand signal that leads price during the build-up phase, EARLIER than a level-extreme which only triggers at peak crowding. Trade in the direction funding is accelerating: long when funding is rising fast, short when funding is falling fast; exit when the acceleration fades. This aligns with the one positive signal this session (the analyst's note that FOLLOWING funding — not fading it — carries edge), but expressed through funding's derivative for earlier, distinct timing. It uses ONLY the two confirmed-full-history feeds (OHLCV + funding) on the clean single-instrument USD-M MARGIN path, sidestepping the recurring supplementary-data-coverage wall (OI/liquidations/options-IV are all recent-only) that rendered three strategies untestable this session, and avoids every other structural blocker: 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 pure price-pattern (the dead families). AVAX is chosen for its volatile funding (a strong, frequent acceleration signal), deep multi-year USD-M funding history, and zero overlap with the BTC/ETH/SOL/XRP instruments in the pending funding set — diversifying instrument while staying on the confirmed-testable substrate. Kept to ~3 parameters to resist overfitting.

The Layer-3 failure was 'No trades produced / should_enter returned a side 0 times over 2059 evaluated bars'. The signal logic ran fine; the root cause was in _build_funding, which filtered funding keys with `if not isinstance(k, int): continue`. supplementary_data is delivered to the backtest subprocess as JSON, where object keys are strings, so every funding timestamp key is a str -> the isinstance(int) test rejected all of them -> _fund_ts/_fund_rate stayed empty -> _daily_funding never accumulated -> the `len(self._daily_funding) <= accel_lookback` guard in should_enter was always true -> zero entries. The minimal fix parses each key with int(k) under try/except (accepting both int and str forms, the portable pattern every working funding strategy this session uses) and dedupes into a {ts: rate} dict so the feed's dual int/str keys never double-count a day's funding mean. With the series now loading, the daily funding accumulates, the accel = funding[-1] - funding[-1-accel_lookback] becomes non-zero, and entries trigger when |accel| >= entry_accel_threshold (0.00005, reachable for AVAX's volatile daily funding deltas). Only _build_funding changed; the daily-funding windowing, acceleration computation, continuous accel+price signal, directional entry gate, fade/stop/time exits, and capital-relative sizing are byte-identical, so Layers 1-2 stay green. Single-instrument OHLCV + funding on the clean USD-M MARGIN path, funding as signal only (no uncreditable cash flow), leverage 1.0 used as a non-amplifying notional fraction so no unused-leverage gate applies.

The funding-acceleration (first-difference) signal has no directional edge on AVAX — in fact a negative one. Over a dense, clean 511-trade sample (5.7 years), the strategy loses -25.3% (CAGR -6.1%) with Sharpe -0.61 (CI -1.47 to +0.20, mostly negative), PF 0.795 (<1), expectancy -$59/trade, and negative payoff geometry (win rate 47.4% with avg_win $485 < avg_loss $548). It loses in 5 of 7 years (2020/2021/2022/2023/2025) with a 27% max drawdown and recovery factor -0.98. This is a dense sample, not a sparsity artifact, so the negative result is a real verdict on the mechanism: the rate-of-change of funding does not lead AVAX forward price. It is a premise failure, not a parameter problem — no accel_lookback / entry-threshold setting manufactures a directional edge that is absent, and a best-of-225 sweep would fit noise; inverting the signal won't rescue it either, because fees are paid in both directions and the exit structure (accel-fade / stop / time-cap) is not a clean sign-flip (the same fallacy that sank the SOL funding-momentum flip). FAILURE PATTERN: funding-derived directional signals on a single crypto perp — whether the level extreme (momentum-follow), smoothed sign (regime-persistence), settlement clock (reversion), or now the first difference (acceleration) — repeatedly show NO exploitable forward-price edge and typically a negative one after fees; the entire single-instrument funding-as-direction family is dead on this substrate, distinct from the genuinely promising negative-funding-CARRY basket (where funding is the credited cash flow, not a price-direction predictor).

Long-short single-instrument AVAXUSDT.BINANCE USD-M daily strategy that trades the ACCELERATION (first difference) of the funding rate. Funding is used as a signal only (never credited as cash flow); P&L is price-based. Each daily bar it computes a daily funding value (mean of the day's settlements, carried forward on empty days) and the change in that value over accel_lookback days; it goes long when funding is accelerating up (building long demand) and short when accelerating down, exits when the acceleration fades/reverses, and uses a 6% protective stop and a 15-bar max-hold backstop. Iteration 2 fixes the funding-key parsing so the funding series actually loads (the prior code rejected the string keys that the subprocess JSON round-trip produces, leaving the series empty and producing zero trades). ~3 tunable parameters, leverage 1.0.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: AVAXUSDT.BINANCE-1-DAY-LAST-EXTERNAL, Bars processed: 2063 Diagnostics: should_enter() returned a side 0 times over 2059 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Clean, dense execution: 511 trades over 5.7 years, single-instrument USD-M, funding-as-signal-only, no exposure bug (avg_position_pct 10.7%), low fee drag (3.6% of gross)

Balanced long/short (255 long / 256 short) — the symmetric design worked mechanically

No directional edge — it is negative: total return -25.3%, CAGR -6.1%, Sharpe -0.61 (CI -1.47 to +0.20), PF 0.795 (<1), expectancy -$59/trade

Negative payoff geometry: win rate 47.4% AND avg_win $485 < avg_loss $548 — both work against it

Loses in 5 of 7 years (2020/2021/2022/2023/2025), max DD 27%, recovery factor -0.98, sortino -0.70

Dense 511-trade sample means the negative edge is a real verdict, not noise; return_kurtosis 68.4 shows it loses even with outlier days included

negative 5/7 years

majority positive

AvaxFundingAccelerationMomentumLS tried a novel angle on the funding signal — trading its derivative (day-over-day acceleration) rather than its level, on the thesis that building leveraged demand leads price earlier than peak crowding — on AVAX for its volatile funding and clean USD-M testability. The dense 511-trade backtest delivered an unambiguous negative verdict: a -25.3% return, Sharpe -0.61, profit factor 0.795, negative expectancy, and losses in five of seven years, with both win rate and win/loss size working against it. The analyst abandoned it at the backtest-review gate on its second iteration, judging it a premise failure rather than a tuning problem and noting that inverting the signal would not help because of fees and an asymmetric exit. It was filed as the fourth confirmation that the single-instrument funding-as-direction family — level, sign, clock, and acceleration alike — is dead on this substrate, while the credited funding-carry basket remains the genuinely promising path.

Funding-derived directional signals on a single crypto perp — level extreme, smoothed sign, settlement clock, and now first-difference acceleration — repeatedly show no exploitable forward-price edge and typically a negative one after fees; the entire single-instrument funding-as-direction family is dead on this substrate, distinct from the genuinely promising negative-funding-CARRY basket where funding is the credited cash flow rather than a price-direction 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 dense sample makes the negative result a real verdict, not noise — the rate-of-change of funding does not lead AVAX forward price — and it is a premise failure no parameter or signal-inversion can rescue (fees paid both ways, exit structure not a clean sign-flip).

A long-short, single-instrument directional strategy on AVAXUSDT.BINANCE USD-M daily perpetual futures that used the FIRST DIFFERENCE (acceleration/rate-of-change) of the funding rate as the signal — funding as signal only, never carry — going long when funding was rising fast (long demand building) and short when it was falling fast, exiting when acceleration faded, on the thesis that accelerating funding leads price earlier than a level extreme.

On a dense, clean 511-trade sample over 5.7 years the signal had a clearly negative edge: total return -25.3%, CAGR -6.1%, Sharpe -0.61 (CI -1.47 to +0.20), profit factor 0.795, expectancy -$59/trade, with negative payoff geometry (47.4% win rate and avg win $485 < avg loss $548), losing in 5 of 7 years, 27% max drawdown, and recovery factor -0.98 (kurtosis 68.4 showing it loses even with outlier days included).