AvaxBinanceOneHourDailyVwapExcursionMeanReversionLongShort

AVAX BINANCE USD-M Futures 1H Daily-Anchored VWAP Excursion LONG/SHORT Mean-Reversion (Direction-Quota Fill, Single-Instrument, 3-Parameter, ATR-Relative Stop, Bounded Infra Params)

A symmetric LONG-AND-SHORT, single-instrument intraday mean-reversion strategy on AVAXUSDT.BINANCE perpetual futures using 1-HOUR bars and OHLCV-only data. This targets the long_short DIRECTION bucket (12.6% actual vs the 55% long-only ceiling — i.e. long/short is the most under-represented direction), which can ONLY be filled on a futures venue, and which this session's evidence says must be filled on a RELIABLY-TESTABLE venue: HYPERLIQUID is data-span-blocked (two consecutive HL transplant abandons — ETHUSD-HL 1H had ~7 months, SOL-HL 15m only 38 days), COIN-M is engine-blocked (InstrumentId '_' misclassification), options/carry/cross-venue are tooling-blocked (no funding-cash-flow accounting, no multi-leg booking, no cross-venue MTM netting), and the supplementary-signal families (premium-index, L/S-ratio) have failed repeatedly (daily under-sampling, frozen-signal bugs, catastrophic holdouts). BINANCE USD-M is therefore the correct venue: low ~0.10% round-trip cost (best fee headroom for an intraday MR), shorting allowed, and multi-year continuous 1H history for AVAX (>4 years, >35,000 bars — no data-insufficiency risk). It uses the ONE mechanism class that cleared every optimization gate this session — daily-anchored VWAP excursion mean-reversion (the ETH-1H Binance sibling was PROMOTED; range-MR, funding-MR, premium-MR, and the options/carry families all failed) — applied symmetrically and to a FRESH asset outside the BTC/ETH/SOL/XRP VWAP cluster already in the pipeline. AVAX is chosen for clean liquid-alt behavior: high enough intraday volatility that excursion reversions clear fees with margin, but without DOGE-style meme-squeeze tail risk on the short side. Lessons explicitly applied: (1) FEE VIABILITY — entry z-threshold is set HIGH (>=2.5) so the expected reversion (~0.4-0.9% of price on AVAX 1H) is several multiples of the 0.10% round-trip, clearing the >0.15% futures floor and avoiding the HFT fee-drag anti-pattern. (2) WIN/LOSS ASYMMETRY — the stop is ATR-relative (scales with the volatility that produced the excursion), NOT a fixed % tighter than the reversion target, fixing the inverted avg_loss/avg_win (2.05x) that sank the SOL-HL predecessor. (3) OPTIMIZER-ESCAPE — exactly 3 tunables (z_entry, vwap_std_lookback, stop_atr_mult); ALL infra params (min_session_bars, time-stop bars, min_notional) are FIXED at construction and bounded to physically valid ranges, and MUST NOT enter the Optuna search space (closing the eod_exit_hour=30 escape that voided the ETH-HL hypothesis).

Applies the one mechanism class that cleared every optimization gate this session (daily-anchored-VWAP excursion MR, the promoted ETH-1H Binance sibling) symmetrically to a fresh asset (AVAX, outside the existing BTC/ETH/SOL/XRP VWAP cluster) on the reliably-testable BINANCE USD-M venue, filling the under-represented long_short direction bucket where HYPERLIQUID (data-span-blocked), COIN-M (engine-blocked), and options/carry/cross-venue (tooling-blocked) all fail. Implemented with the standard FactoryStrategy hooks (single instrument). The UTC-day VWAP anchor resets via integer ns math (day = ts_ns // NS_PER_DAY) and the time gates use hour = (ts_ns // NS_PER_HOUR) % 24, so no datetime import is needed. calculate_signal returns the continuous VWAP z-score (varies every bar -> Layer-2 frozen-signal detection passes), while the symmetric discrete gates live in should_enter: z >= z_entry -> SELL (short the upside excursion), z <= -z_entry -> LONG, subject to MIN_SESSION_BARS=4 warmup so the rolling std is trustworthy and a 22:00 entry cutoff so the position has time to revert. Three lessons are explicitly applied: (1) FEE VIABILITY -- z_entry defaults high (2.5) so the ~0.4-0.9% AVAX 1H reversion is several multiples of the ~0.10% round-trip; (2) WIN/LOSS SYMMETRY -- the stop is ATR-relative (stop_atr_mult*ATR captured at entry, applied directionally) so avg_loss scales with the same volatility as the reversion target, fixing the inverted avg_loss/avg_win that sank the tighter-stopped predecessor; (3) OPTIMIZER-ESCAPE -- exactly three tunables (z_entry, vwap_std_lookback, stop_atr_mult), and every infra constant (position_pct, EOD/entry-cutoff hours, MIN_SESSION_BARS, ATR period, min_notional) is fixed AND clamped to a physically valid range in __init__, so a stray optimizer perturbation (e.g. eod_exit_hour=30) is neutralized. Robustness: zero-volume day falls back to mean-of-closes so VWAP stays defined, std<=0 yields z=0 (no entry), and position side/entry context is reconstructed from avg_px_open if lost. The forced 23:00 UTC flat ensures no position is held across the midnight anchor reset, keeping this an intraday overshoot bet (not a multi-day directional short that accrues funding). Sizing is capital-relative (10% of equity) clearing the AVAX $5 min-notional; leverage stays 1.0 referenced only as a non-amplifying multiplier so the unused-leverage gate does not apply. AVAX has >4 years / >35,000 1H bars on Binance, eliminating the data-insufficiency risk that killed the HL transplants.

AVAX 1H daily-anchored VWAP-excursion MR (symmetric long+short) has a NEGATIVE, inverted edge — not a weak one: profit_factor 0.63, Sharpe -2.81 with the entire bootstrap CI negative (ci_low -4.41, ci_high -1.16), probabilistic_sharpe 0.0004, expectancy -$42/trade, negative in 6 of 7 years over a healthy 359-trade / 4+ year sample with finally-adequate data. A PF of 0.63 means the strategy loses ~37% more than it wins systematically — the mechanism is structurally wrong for crypto intraday: VWAP excursions tend to CONTINUE (intraday momentum) rather than revert, so fading them (the entire MR premise) loses by construction. The careful fee-viability and win/loss-asymmetry fixes the hypothesis applied (z_entry>=2.5, ATR-relative stop) did not help because the directional premise, not the sizing, is the loser. Not optimize: no parameter turns a -2.81 Sharpe / 0.63 PF positive — you cannot tune your way out of an inverted payoff; optimizing would just curve-fit noise. Not iterate: the asymmetry fixes were already applied and the mechanism itself is wrong; the only 'fix' (invert to momentum / trade WITH the excursion) reverses the whole premise and is a Research-Lead reframe, not a developer parameter tweak. FAILURE PATTERN: this is the 8th consecutive failure of the daily-anchored VWAP-excursion MR transplant class (0-for-7 prior), and the first with conclusive multi-year data — confirming the 'inverted-payoff signature': symmetric VWAP-excursion mean-reversion on crypto intraday bars systematically loses (PF < 1, negative Sharpe with a fully-negative CI) because excursions continue rather than revert. The promoted ETH-1H sibling appears to be regime/asset-specific luck, not a generalizable mechanism. The Research Lead should retire the VWAP-excursion-MR transplant lane entirely (or test the inverse momentum framing), not propose further single-asset variants.

Symmetric long+short intraday daily-anchored-VWAP excursion mean-reversion on AVAXUSDT.BINANCE USD-M perpetual futures, 1-hour bars, OHLCV-only. Each 1H bar it builds the UTC-day-anchored VWAP (volume-weighted, reset at midnight, mean-of-closes fallback when volume is absent), the rolling std of (close - VWAP) over the last vwap_std_lookback day bars, and z = (close - VWAP)/std. It shorts when z >= +2.5 (price far above fair value) and longs when z <= -2.5 (far below), targeting reversion to VWAP within the session. It exits at the earliest of reversion to VWAP (long z>=0, short z<=0), an ATR-relative stop (2.5*ATR captured at entry), or a forced flat at 23:00 UTC before the anchor resets. Exactly three tunables (z_entry, vwap_std_lookback, stop_atr_mult); all infra params fixed and bounded at construction. Position 10% of equity, leverage 1.0.

Clean, correct implementation with finally-adequate data (>4 years, 150k bars, 359 trades) — no data-insufficiency caveat

Genuinely market-neutral (beta -0.003); disciplined infra-param freezing as designed

Inverted/negative edge: profit_factor 0.63, Sharpe -2.81 with the ENTIRE CI negative (ci_high -1.16), PSR 0.0004, expectancy -$42/trade

Negative in 6 of 7 years over a large sample — systematic, not noise; the MR bet fades intraday excursions that actually continue

8th consecutive failure of the VWAP-excursion-MR transplant class (the hypothesis's own '0-for-7 inverted-payoff' signature), now confirmed with good data

Fee drag heavy (commission 12.97% of gross); avg_loss $218 > avg_win $154 despite the ATR-stop fix

AvaxBinanceOneHourDailyVwapExcursionMeanReversionLongShort was chosen by elimination — Hyperliquid data-blocked, COIN-M engine-blocked, options/carry tooling-blocked — to fill the long/short quota on Binance USD-M with the one VWAP-excursion-MR mechanism whose ETH-1H sibling had been promoted, carefully applying fee-viability (z ≥ 2.5) and ATR-relative-stop fixes. But with finally-conclusive data it revealed an inverted edge: a 0.63 profit factor, a Sharpe of -2.81 with a fully-negative CI, and losses in 6 of 7 years, because intraday excursions continue rather than revert. The careful asymmetry fixes could not help because the directional premise itself was the loser. As the 8th straight failure of the transplant class, the analyst abandoned it at the backtest-review gate and recommended retiring the VWAP-excursion-MR lane entirely (or testing the inverse momentum framing), treating the promoted ETH sibling as luck rather than a generalizable edge.

Symmetric VWAP-excursion mean-reversion on crypto intraday bars systematically loses because excursions continue rather than revert — this was the 8th consecutive failure of the transplant class (0-for-7 prior) and the first with conclusive multi-year data, indicating the one promoted ETH-1H sibling was regime/asset-specific luck rather than a generalizable mechanism.

The analyst abandoned it at the backtest-review gate before optimization because the edge was inverted, not merely weak — a 0.63 profit factor and a fully-negative Sharpe CI mean the mechanism loses by construction: crypto intraday VWAP excursions tend to continue (intraday momentum) rather than revert, so fading them loses systematically, and no parameter tuning or sizing fix can turn an inverted payoff positive.

A symmetric long+short, OHLCV-only intraday mean-reversion on AVAXUSDT 1H Binance futures — anchoring a session VWAP at each UTC midnight, z-scoring the close's excursion from it, and fading extremes (short when z ≥ 2.5, long when z ≤ -2.5) for reversion to fair value, exiting at VWAP, an ATR-relative stop, or a forced end-of-day flat, with 3 tunables and all infra params frozen to fill the under-represented long/short bucket on a reliably-testable venue.

With finally-adequate data (>4 years, 150k bars, 359 trades), it had a clearly negative, inverted edge: -15.3% total return (CAGR -2.8%), Sharpe -2.81 with the entire bootstrap CI negative (-4.41 to -1.16), profit factor 0.63, PSR 0.0004, per-trade expectancy -$42, an avg loss ($218) larger than avg win ($154) despite the ATR-stop fix, commission at 12.97% of gross, and negative returns in 6 of 7 years.