EthSpotDailyDrawdownAccumulationLong

ETH BINANCE_SPOT Daily Drawdown-Accumulation Long-Only (Spot Venue Quota Fill, Single-Instrument, Defined Recovery Exit, 3-Parameter, Anti-Buy-Hold Guards)

A LONG-ONLY, single-instrument, DAILY strategy on ETHUSDT.BINANCE_SPOT that buys deep drawdowns from a rolling high and exits on recovery. This is a direct implementation of the analyst's explicit recommendation after the XRP-spot VWAP abandon: 'the BINANCE_SPOT venue quota remains unfilled, but VWAP-MR is not the vehicle — a spot long-only edge needs genuine UPSIDE-ASYMMETRY (e.g., the proven drawdown-accumulation BTC-daily class on a spot major), not dip-fading.' It targets the BINANCE_SPOT venue bucket (11.1% actual vs 15% target) — the only under-represented venue that is reliably testable (HYPERLIQUID data-blocked, COIN-M engine-blocked, options/carry/cross-venue tooling-blocked, funding data terminates ~Jan-2023). It uses the PROVEN drawdown-accumulation mechanism class (the BTC-daily sibling reached PAPER at Sharpe 3.33 with adequate trade count) on a FRESH spot major (ETH, not BTC; single asset, not a basket), and it explicitly engineers around every documented failure mode of this class this session: (1) DAILY bars, NOT 4H — the SOL-4H and basket variants failed because at 4H the 15-21% drawdown events were too rare to fill walk-forward; the proven sibling is DAILY and so is this. (2) SINGLE asset — avoids the basket-correlation fallacy (BTC/ETH/SOL drawdowns are simultaneous, so a 'diversified' basket is exposure-stacking, which sank the 3-majors basket). (3) DEFINED RECOVERY EXIT + moderate threshold — prevents the buy-and-hold degeneration where the optimizer chose drawdown_threshold=21%/3-trades/693-day-holds (metrics_reliable=false, single-position end-of-data MTM artifacts). A modest threshold plus a concrete recovery exit forces MULTIPLE completed round-trips across ETH's 2018/2020/2021/2022/2024/2025 drawdown cycles, so positions actually CLOSE (no unrealized-PnL spike dominating the final window). (4) Genuine upside-asymmetry: ETH is a secular-recovery major (every >30% drawdown in its history has eventually recovered), so buying deep discounts is structurally different from intraday dip-fading on a mean-declining alt (the falling-knife trap that inverted the XRP/SOL VWAP payoffs). Exactly 3 tunables (dd_threshold, high_lookback_days, recovery_exit_pct); all infra params (min trades floor, min_notional, max concurrent scale-ins) FIXED and bounded — not in the Optuna search space.

Direct implementation of the analyst's post-XRP-VWAP recommendation: fill the under-represented BINANCE_SPOT venue quota with a genuine upside-asymmetry edge (the proven drawdown-accumulation class whose BTC-daily sibling reached PAPER at Sharpe 3.33), NOT dip-fading. Venue routing is correct: long-only, no leverage, no funding edge -> BINANCE_SPOT (CASH account, leverage clamped to 1, only OrderSide.BUY ever returned), which also has multi-year continuous ETH daily history so there is no data-insufficiency or data-blocked failure mode (unlike HYPERLIQUID / COIN-M / options / cross-venue). It engineers around every documented failure of this class: (1) DAILY bars (not 4H) so 20%+ drawdown events are frequent enough to fill walk-forward; (2) SINGLE asset (not a basket) to avoid correlated-drawdown exposure-stacking; (3) a moderate dd_threshold (20%) plus a concrete recovery exit (+15%, far above the ~0.2% spot round-trip fee) that forces multiple completed round-trips, and a re-arm latch that prevents both immediate re-entry-after-exit-while-still-in-the-drawdown-zone and scale-in stacking -- directly closing the buy-and-hold-degeneration / single-position-end-of-data-MTM hole (verified by simulation: two separate drawdown cycles produce two clean buy->exit round-trips, no churn); (4) ETH's secular-recovery profile gives true upside-asymmetry rather than the falling-knife trap of mean-declining alts. The returned drawdown signal varies every bar (non-frozen on real data; computable on partial windows so Layer-2 short synthetic scenarios still exercise it, while should_enter requires the full high_lookback_days before any real entry). Exactly three tunable parameters (dd_threshold, high_lookback_days, recovery_exit_pct); POSITION_PCT, MAX_SCALE_INS=1 and min_notional are fixed bounded constants kept out of the Optuna search space.

Too thin to optimize, with distorted artifact metrics and outlier concentration. The anti-buy-hold guards (re-arm latch + single tranche + 20% dd_threshold + 180-day lookback) over-restricted this to only 10 trades over ~8.7 years — far below any walk-forward floor; a 3-window split gives ~1 OOS trade/window (degenerate), and avg_holding_period is 279 days. The headline Sharpe 9.62 is a per-period-attribution artifact (annualized_volatility 161% from booking ~9-month-hold PnL onto single exit days; sharpe_ci [-0.45, 141.7], sortino 0.0), not a real edge. The result is 2021-bull-recovery-outlier-concentrated (2021 +87% carries it) with a single 2026-05-18 trade at -22.3% as the lone loser (avg_loss $49.8k >> avg_win $13.7k, return_skew -2.55), and exposure 142% on a CASH spot account is anomalous. Two compounding reasons not to optimize: (1) 10 trades cannot support meaningful optimization (the optimizer fits noise, walk-forward degenerate); (2) the drawdown-DCA-on-alt family has failed post-optimization on every port this session — SOL (DSR 0.0008, zero OOS trades, negative holdout), BNB (zero OOS trades), LINK (is_overfitted, zero-OOS, artifact Sharpe 9.33) — only the BTC sibling reached paper. Loosening the threshold to manufacture more trades would just produce another 2021-bull-recovery-outlier sample that fails DSR the same way, so this is abandon rather than iterate (no 3-parameter drawdown-DCA config turns ETH's handful of deep-drawdown recoveries into a multiple-testing-significant edge). FAILURE PATTERN: single-asset daily drawdown-accumulation on an alt major (ETH spot) with deep-drawdown guards fires too rarely (10 trades / 8.7 years) and concentrates in 1-2 bull-recovery years, producing an artifact Sharpe (vol 161% from long-hold per-period attribution) on an untestably thin, outlier-driven sample; the mechanism only cleared the gates on BTC.

Long-only single-instrument daily drawdown-accumulation on ETHUSDT.BINANCE_SPOT. Each day it computes the rolling high over high_lookback_days (max of daily highs) and the drawdown of the close from that high. It BUYS when the drawdown reaches dd_threshold (default 20% below the rolling high) and is 'armed', deploying a fixed 50% of equity (CASH, 1x, long-only). It EXITS on a defined recovery: when the close reaches entry_price x (1 + recovery_exit_pct) (default +15%). A re-arm latch disarms the strategy on entry and only re-arms once the drawdown recovers back below dd_threshold, so it takes exactly one tranche per drawdown cycle (no scale-in stacking) and produces clean, completed round-trips across ETH's repeated drawdown cycles rather than one perpetual position. The drawdown value is returned as the continuous signal every bar.

Clean execution (10 signaled/submitted, 0 dropped); sibling mechanism (BTC spot drawdown-DCA) reached paper at Sharpe 3.33; anti-buy-hold guards prevented the single-perpetual-position degeneration

Genuine upside-asymmetry premise (ETH secular recovery) and healthy capacity

Far too thin: only 10 trades over ~8.7 years — a 3-window walk-forward gets ~1 OOS trade/window (degenerate); avg holding 279 days

Distorted artifact metrics: Sharpe 9.62 with annualized_vol 161% (per-period attribution on ~9-month holds), sharpe_ci [-0.45, 141.7], sortino 0.0

Outlier-concentrated: 2021 (+87%) carries it; single 2026 -22.3% loser (avg_loss $49.8k >> avg_win $13.7k, skew -2.55); exposure 142% anomalous on CASH spot

Drawdown-DCA family failed post-opt on every alt port this session (SOL, BNB, LINK) — only the BTC sibling worked

EthSpotDailyDrawdownAccumulationLong implemented the analyst's exact recommendation after the XRP-spot VWAP abandon — a genuine upside-asymmetry spot edge via the proven BTC-daily drawdown-accumulation class on ETH, carefully guarded against the buy-and-hold and basket failure modes that sank prior variants. The guards worked as designed but throttled the strategy to only 10 trades over nearly nine years, yielding a dazzling-but-meaningless Sharpe 9.62 driven by per-period attribution on ~9-month holds and carried almost entirely by the 2021 recovery. The analyst ruled it untestably thin and outlier-concentrated — and noted the drawdown-DCA-on-alt family had failed on SOL, BNB, and LINK while only the BTC sibling reached paper — and abandoned it before optimization. It ended after one iteration as abandoned, never advancing to optimization or risk review.

Single-asset daily drawdown-accumulation with deep-drawdown guards fires too rarely on an alt major (10 trades / 8.7 years) and concentrates in 1-2 bull-recovery years, producing an artifact Sharpe from long-hold per-period attribution on an untestably thin, outlier-driven sample — the mechanism cleared the gates only on BTC, and loosening the threshold to manufacture trades would just re-find a bull-outlier sample that fails deflation the same way.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) because the anti-buy-hold guards over-restricted it to just 10 trades — far below any walk-forward floor (~1 OOS trade/window) — with an artifact Sharpe and bull-recovery-outlier concentration, and the drawdown-DCA-on-alt family had already failed post-optimization on every port this session (SOL, BNB, LINK) with only the BTC sibling working, so optimization and later stages were never reached.

A long-only single-instrument daily drawdown-accumulation strategy on ETHUSDT.BINANCE_SPOT (OHLCV-only) that bought when the close fell >=20% from its rolling 180-day high and exited on a defined 15% recovery above entry, using a re-arm latch and single-tranche cap to force clean round-trips — a port of the paper-stage BTC-daily drawdown-accumulation sibling (Sharpe 3.33) to a fresh spot major, explicitly engineered against the buy-and-hold and falling-knife failure modes.

It produced eye-catching but artifact metrics — Sharpe 9.62, 90% win rate, +73.6% return, profit factor 2.47 — over only 10 trades in ~8.7 years with a 279-day average hold; the headline was distorted by per-period attribution (annualized vol 161%, Sharpe CI [-0.45, 141.7], Sortino 0.0), outlier-concentrated in 2021 (+87%), with a single 2026 -22.3% loser (avg loss $49.8k >> avg win $13.7k, skew -2.55) and anomalous 142% exposure on a CASH account.