Strategies

AVAX Spot Capitulation-Exhaustion Accumulation, Long-Only (BINANCE_SPOT, 4h Bars, Buy When Perp AGGRESSIVE-SELL Order-Flow Spikes to Panic Extremes and Then Recedes — Sellers Exhausting Into a Drawdown; Non-Price Taker-Flow Feed, 3-Parameter)

Outcome: Abandoned

AvaxSpotCapitulationExhaustionAccumulationLong

Outcome Summary

AvaxSpotCapitulationExhaustionAccumulationLong tried to convert the taker-flow feed into a long-only spot accumulation edge, buying AVAX when aggressive perp selling spiked to panic extremes and then receded during a drawdown — deliberately structured to differ from the liquidated SolTakerFlow and impact-dead EthRetail siblings by being spot, unlevered, and exhaustion-of-selling. It succeeded on plumbing (262 signals, no data wall, no liquidation risk) but not on edge: it bought into continued declines for a -43.4% return, PF 0.911, near-zero Sharpe, and a 58.57% drawdown past the floor, losing money in most years. The reviewer abandoned it as a genuine negative-expectancy verdict, noting it was the latest death in the single-name order-flow-timed family (SolTakerFlow liquidated, EthRetail impact-dead at $1M capacity, SolOI-flush -13%) and that optimization could only overfit the lone 2023 up-window.

Strategy report

Hypothesis

A LONG-ONLY, cash-account (no leverage, no liquidation) accumulation strategy that trades AVAXUSDT on BINANCE_SPOT, timed by a genuine NON-PRICE order-flow feed: the aggressive taker SELL fraction from the AVAXUSDT.BINANCE perp metrics (Binance VISION daily 'metrics' files, sum_taker_long_short_vol_ratio column — MULTI-YEAR back to ~2020 at 5-min granularity, NOT the 30-day API). The insight is capitulation-exhaustion: a spot drawdown accompanied by a spike in aggressive market-SELLING that then RECEDES marks forced/panic selling drying up, and spot mean-reverts up. This deliberately fills the under-represented BINANCE_SPOT bucket (11.5% vs ≥15% target, and the highest-survival venue in the portfolio) with a mechanism NOT already deployed. It is distinct from every recent failure and pending item: NOT pure OHLCV (L30 — the taker-sell fraction is a separate order-flow series), NOT funding-direction (L13) or carry (L28 — spot earns no funding, this is pure directional accumulation), NOT the SolTakerFlow perp z-divergence sibling (this is SPOT, LONG-ONLY, exhaustion-of-selling — opposite construction), NOT positioning-fade (EthRetail uses the L/S account ratio, not taker volume), NOT OI-flush (the SolOI-flush reversal that lost -13% on SOL — different feed, different venue, long-only recovery hold vs symmetric perp fade), NOT liquidations (L3 broken feed), NOT a basket (L12), NOT options (L29), NOT HL (L11). 3 parameters to resist overfit; targets 50-120 trades over the AVAX spot history.

A capitulation-exhaustion accumulation strategy on a NON-PRICE order-flow feed, deliberately filling the under-represented BINANCE_SPOT bucket (the highest-survival, unliquidatable venue). Data wiring verified against the loader: 'metrics' is loaded by SYMBOL (AVAXUSDT), venue-agnostic, so a SPOT primary still loads the AVAXUSDT PERP taker-flow metrics — no cross-instrument-supp leg needed and no venue/equity mismatch (single instrument, single spot account). The taker ratio is read via the exact emitted key (the Vision parser outputs 'taker_long_short_vol_ratio', which the parse matches). Mechanism: sell_frac = 1/(R+1) (R = taker buy/sell ratio) as-of the bar timestamp (O(log n) searchsorted over arrays precomputed once -> timeout-safe); its rolling percentile (O(window) boolean-mean) detects a panic spike, and the exhaustion trigger requires (a) a recent panic peak >= panic_pct, (b) the current percentile receding >= recede_drop off that peak, and (c) a concurrent price drawdown — sellers drying up into a drop, which accumulates for the mean-reversion bounce. Distinct from every sibling: NOT pure OHLCV (the taker-sell fraction is a separate order-flow series), NOT funding/carry (spot earns no funding — pure directional accumulation), NOT the SolTakerFlow perp z-divergence (this is SPOT, LONG-ONLY, exhaustion-of-selling — opposite construction), NOT positioning-fade (EthRetail uses the account L/S ratio, a different metrics column), NOT OI-flush/liquidations/basket/options/HL. 3 core parameters (sell_window, panic_pct, hold_bars) to resist overfit. VENUE/COMPLIANCE: BINANCE_SPOT CASH account -> structurally unlevered (leverage pinned 1.0), long-only (only BUY entries; exits close longs, never short — the base's spot-short guard also enforces this), so it cannot be liquidated. If the metrics feed were unavailable the percentile/exhaustion signal cannot form and it does not trade — no fallback to a price-only proxy (the edge is the order-flow exhaustion, not price). Trade frequency: panic (percentile>=0.90) + recede + drawdown is a rare confluence, targeting ~50-120 accumulation trades over multi-year AVAX 4h spot history, clearing the measurability floor without high-turnover fee bleed (4h bars, multi-day holds). Risk is bounded: long-only cash, <=0.50 of equity per position, no leverage, no compounding-liquidation path; an 8% stop caps the downside of a capitulation that keeps extending. AVAXUSDT is a deep-history liquid BINANCE_SPOT symbol with a multi-year perp metrics feed, so data is available (no data_unavailable).

Net-losing single-name accumulation: profit_factor 0.911 (<1.0), total_return -43.4%, expectancy -$165.87 over 262 trades, max_drawdown 58.57% (CI to 97.4%, past the 50% floor / L19), Sharpe 0.149 with CI [-0.54, 0.82] straddling zero, and information_ratio -0.59 vs its meaningful buy-hold benchmark. The strategy is correctly unlevered long-only spot (no liquidation) and the taker-flow metrics feed is present (262 signals fired, no data wall), so this is a genuine strategy verdict — the capitulation-exhaustion timing (buy when aggressive-sell flow spikes then recedes into a drawdown) simply has no fee-clearing edge on AVAX spot: it bought into continued declines and lost money in most years (2022 -37%, 2025 -44%), with only 2023 (+52%) and 2024 (+12%) positive. This is the same single-name order-flow-timed fade/accumulation family that has died repeatedly this session (SolTakerFlow perp z-divergence liquidated, EthRetail positioning-fade impact-dead at $1M capacity, SolOI-flush -13%). Optimizing sell_window/panic_pct/hold_bars cannot flip a PF-0.91, negative-expectancy, 58%-DD loser — the 225-trial optimizer would overfit the single 2023 up-window and fail the deflated-Sharpe gate. Not worth 2 hours of optimization.

Implementation

Long-only, unlevered cash-account accumulation on AVAXUSDT.BINANCE_SPOT (4h bars), timed by a non-price perp order-flow feed: the aggressive taker SELL fraction derived from the AVAXUSDT 'metrics' taker buy/sell volume ratio (Binance VISION, multi-year). Each 4h bar it computes the sell fraction = 1/(R+1) and its rolling percentile over sell_window bars. It ACCUMULATES (buys spot) when the sell fraction has recently spiked to a panic extreme (percentile >= panic_pct) and is now receding off that peak (dropped >= recede_drop in percentile) WHILE price is in a drawdown — i.e. forced/panic selling is exhausting into a drop, which tends to mean-revert up. Exits on the recovery bounce (take-profit), a protective stop, or a time stop. Long-only spot: only buys, never shorts. 3 core parameters: sell_window, panic_pct, hold_bars.

Verification Results

Analyst should weigh the negative dollar expectancy against the positive per-trade % and decide iterate/abandon.

Sandbox is a net loser despite clearing the nominal per-trade floor: total_return -4.21%, profit_factor 0.957, expectancy -$78.9/trade (NEGATIVE dollar PnL), Sharpe 0.028 with CI [-1.44, +1.64] straddling zero, max_drawdown 35%. Note the divergence — avg_trade_return_pct 0.389% (above the 0.25% spot floor) yet avg_trade_pnl_usd is negative — the simple-average percent edge is thinner than fees+impact on a dollar-weighted basis. Performance concern for the analyst, not a code defect.

Optimizer/analyst should reduce position_frac to a size the book can absorb.

Severe capacity/impact red flag: capacity_usd only $1,663 with impact_cost_pct 775% on ~$50k AVAX spot positions (position_frac 0.50). The impact model implies the strategy cannot execute at this size. Aligned with the hypothesis's stated <=0.50 sizing so not a code defect, but far above realistic AVAX-spot capacity.

Optionally promote 'taker_long_short_vol_ratio' to primary to reflect the actual emitted schema.

The metrics primary key 'sum_taker_long_short_vol_ratio' does NOT match the Vision parser output ('taker_long_short_vol_ratio', col 7). Correctness is preserved only via the first fallback key — verified by the sandbox producing 54 trades. The primary key is effectively dead.

Backtest Review

Sharpe: 0.15
Total return: -43.42%
Max drawdown: 58.57%
Trades: 262
Win rate: 44.3%
Profit factor: 0.91

Correctly unlevered long-only spot (no liquidation risk, avg_position 55.6%); fills the under-represented BINANCE_SPOT venue

Uses a genuine multi-year non-price taker-flow feed (262 signals fired, no data wall); well-sampled

Net loser: profit_factor 0.911 (<1.0), total_return -43.4%, expectancy -$165.87/trade

max_drawdown 58.57% (CI to 97.4%) — past the 50% hard-abandon floor; calmar -0.74

Sharpe 0.149 with CI [-0.54, 0.82] straddling zero; information_ratio -0.59 vs benchmark

Negative most years (2022 -37%, 2025 -44%) — the exhaustion-accumulation signal bought falling knives; same dead single-name order-flow-timed family (SolTakerFlow liquidated, EthRetail impact-dead, SolOI-flush -13%)

Outcome Summary

A 'capitulation-exhaustion' order-flow trigger does not reliably mark bottoms — buying spot into receding panic-sell spikes still catches falling knives across bear years, and being correctly unlevered on spot removes liquidation risk but neither creates an edge nor caps a 58% drawdown; the single-name order-flow-timed fade/accumulation family has no viable edge this session.

It was abandoned at backtest-review (verdict: abandon) as a net-losing single-name accumulation with PF<1.0, negative expectancy, and a 58.57% drawdown past the 50% hard-abandon floor — the exhaustion-accumulation timing simply bought falling knives with no fee-clearing edge, in the same repeatedly-dead single-name order-flow-timed family.

It was a long-only, unlevered accumulation strategy on AVAXUSDT BINANCE_SPOT (4h bars) timed by a non-price order-flow feed — the aggressive taker-sell fraction from the AVAX perp Vision 'metrics' — buying capitulation-exhaustion setups where panic selling spiked to a high rolling percentile and then receded while price was in a drawdown, aiming to catch mean-reversion bounces and fill the under-represented spot venue.

It fired 262 well-sampled trades but was a net loser: total return -43.4%, profit factor 0.911, expectancy -$165.87/trade, Sharpe 0.149 (CI [-0.54, 0.82] straddling zero), win rate 44%, max drawdown 58.57% (CI to 97.4%), and information ratio -0.59 vs its buy-hold benchmark, negative in most years (2022 -37%, 2025 -44%) with only 2023 (+52%) and 2024 (+12%) positive.

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