EthSpotCapitulationAccumulationRunWinner

ETH Spot Deep-Drawdown Capitulation Accumulation, Long-Only with Stabilization-Confirmed Entry and Run-the-Winner Recovery Exit (Single-Instrument BINANCE_SPOT, Daily Bars, Buy Confirmed Rebounds From >=30% Drawdowns, No Leverage, Tail-Safe, Favorable Payoff Geometry, Low-Parameter)

A LONG-ONLY, SINGLE-INSTRUMENT, no-leverage accumulation strategy on ETHUSDT.BINANCE_SPOT (daily bars) that buys ETH only after a DEEP drawdown (capitulation) once the decline has demonstrably STABILIZED, then holds for the recovery. It is deliberately constructed to avoid every recurring failure mode in the factory: (1) it is NOT a single-name ALT trend/breakout book (the BNB/LTC/LINK/AVAX family that produced no-edge and account ruin) — it is a behavioral mean-reversion-to-trend on a top major, long-only; (2) it is NOT an intraday small-TP/large-stop fade (the SOL VWAP / ETH taker / cross-sectional reversal family that wiped out via INVERTED payoff geometry) — it operates at a multi-WEEK scale with explicitly FAVORABLE payoff (enter from an already-depressed level after stabilization, then RUN THE WINNER on the recovery, so avg_win >> avg_loss by construction); (3) it is TAIL-SAFE — long-only, no leverage, on spot CASH, entering only AFTER a large decline has paused (not catching a falling knife), so it structurally cannot blow through zero like the leveraged LINK book; (4) it is SINGLE-INSTRUMENT spot (ONE position) so it cannot trigger the multi-instrument equity/return aggregation engine bug that corrupted the spot-sleeves / rotation / pair / cash-carry runs. It fills the under-represented BINANCE_SPOT venue (9.5% vs 15%) with a non-BTC name, is the CORRECT venue per the routing rule (long-only no-leverage belongs on spot), uses pure OHLCV (no funding/OI/options/liquidations data dependency), and has a low parameter count to resist overfitting. The edge is the documented post-capitulation rebound in structurally-appreciating crypto majors, gated by a stabilization confirmation so it does not buy into ongoing collapse.

Implements the hypothesis directly. Venue: long-only no-leverage belongs on BINANCE_SPOT (CASH), which the routing rule confirms and which under-represented venue this fills with a non-BTC major. Tail-safe: spot, no leverage, entering only after a large decline has paused, so it cannot blow through zero. Favorable payoff geometry: entry from an already-depressed, stabilized level plus run-the-winner trailing exit makes avg_win >> avg_loss by construction, the opposite of the inverted small-TP/large-stop fades that wiped out. Single-instrument spot (one position) avoids the multi-instrument equity-aggregation engine bug. Pure OHLCV, no funding/OI/options dependency. Low parameter count (drawdown depth, stabilization MA, trailing/stop/take-profit) resists overfitting. Spot long-only is respected: should_enter only ever returns BUY; SELL occurs solely on exit. Stabilization gating is the explicit safeguard against buying into ongoing collapse, capturing the documented post-capitulation rebound in structurally-appreciating majors.

Failed deflated Sharpe: DSR=0.0679, expected-max=10.46 over 225 trials — the optimized Sharpe 5.90 is BELOW the best-of-N luck bar and statistically indistinguishable from selection noise (is_significant=false, PBO=0.7209>0.5, sharpe_ci_low -0.41 straddles 0). Every decisive generalization gate fails in concert: walk-forward is_overfitted=TRUE with avg IS 10.30 collapsing to avg OOS 0.0 across ALL three windows [0.0, 0.0, 0.0]; the holdout failed (holdout_sharpe 0.0, ratio 0); and sensitivity failed with 4 cliffs (trail_pct, dd_lookback, deploy_frac, dd_threshold) — the deploy_frac grid drops to Sharpe 0.0 the instant it exceeds 1.0 and dd_lookback degrades monotonically. Root cause is structural sparsity, not a tunable region: the >=30% capitulation + stabilization entry is a rare event firing only ~33 times (20 after optimization) over 8+ years, so walk-forward OOS windows and the 15-day holdout contain essentially zero qualifying entries and the out-of-sample edge is unmeasurable. The flashy in-sample Sharpe is a handful of big bull-recovery winners (avg_win ~$99k vs avg_loss ~$44k, win rate 33-45%, tail_ratio 4.47) fit to noise. The 'tail-safe' premise is also contradicted by max_drawdown 79.5% (initial)/48% (optimized): a 95%-deployed long-only spot book absorbs ETH's full drawdowns despite the 15% per-trade stop. Not iterate (attempt 1 of 2): there is no robust parameter region to tune toward when all OOS windows and the holdout are zero, DSR is 0.07, PBO 0.72, and the core edge parameters are all cliffs — parameter changes cannot manufacture out-of-sample trades from a structurally rare signal. Not revise_hypothesis: this is overfitting + event sparsity, not a promoted mechanism stranded on a dead target; the post-capitulation-rebound idea has no promoted sibling and reframing onto another major would re-hit the same rare-event sparsity. FAILURE PATTERN: deep-drawdown (>=30%) capitulation-accumulation long-only spot entries are too rare (~33 entries / 8yr) to validate out-of-sample — the optimizer over-selects a few bull-recovery winners into a 5-10 IS Sharpe that fails DSR (0.07), is_overfitted=TRUE with all-zero walk-forward OOS, a zero holdout, and a 4-cliff sensitivity; rare-event mean-reversion-to-trend on a single major does not survive walk-forward.

Long-only, no-leverage, single-instrument accumulation strategy on ETHUSDT.BINANCE_SPOT (daily bars). It buys ETH only after a deep capitulation drawdown (>=30% off a 90-day rolling high) AND a stabilization confirmation (a 10-day MA turning up with price reclaiming it), so it does not catch a falling knife. It then runs the winner on the recovery: exits are a trailing stop measured from the in-trade peak (15% give-back, armed only after the trade is up >8%), a 60% take-profit, and a 15% hard stop for tail safety. calculate_signal returns the continuously-varying drawdown depth each bar; the entry decision (deep-drawdown threshold + stabilization gate) lives in should_enter. Sizing deploys ~95% of cash equity into the single spot position.

Trades the intended mechanism: long-only (0 shorts), 33 entries on genuine capitulation episodes across 2017-2026, buying confirmed rebounds and running winners

Favorable payoff geometry as designed: avg_win $41.6k vs avg_loss $16.7k (2.5x), profit_factor 1.25, omega 1.66, tail_ratio 3.08, positive expectancy — big recovery captures vs capped stops

Genuinely tail-safe: single-instrument CASH spot, no leverage, liquidated=false — structurally cannot blow through zero like the abandoned leveraged alt books

Correct venue/routing (long-only no-leverage on BINANCE_SPOT), pure OHLCV, low parameter count, fills the under-represented spot bucket with a non-BTC major

Thin sample: 33 trades over 9 years -> walk-forward windows ~11 trades each, overfitting risk; Sharpe 2.90 has a very wide CI (sharpe_ci_low -3.22 to 7.28), not yet robustly significant

Large 79.5% max drawdown: the stabilization gate gave false 'stabilized' signals through the prolonged 2018->2019 grind (annual -35% then -46%), buying rebounds that resumed falling

Negative information_ratio (-0.45) vs buy-hold and only marginal alpha (+0.014) — at 31% time-in-market it underperforms simply holding ETH; promotion must justify the trade-off

Very high reported annualized_volatility/downside_deviation reflect the lumpy 33-point daily-return series — supporting reason to treat the headline Sharpe cautiously

Sound venue routing (long-only no-leverage on BINANCE_SPOT), single-instrument so no multi-leg aggregation artifact, pure-OHLCV low data dependency

Favorable payoff geometry by construction (avg_win ~$99k vs avg_loss ~$44k; tail_ratio 4.47, omega 2.83) — when it wins, it wins big

Failed deflated Sharpe: DSR=0.0679 (<<0.95), is_significant=false, sharpe_ci_low=-0.41 straddles 0, PBO=0.72>0.5 — the optimized Sharpe 5.90 is BELOW the expected-max luck bar (10.46) over 225 trials

Walk-forward is_overfitted=TRUE: avg IS Sharpe 10.30 collapses to avg OOS 0.0 with ALL three OOS windows = [0.0, 0.0, 0.0]

Holdout failed: holdout_sharpe 0.0, ratio 0, passed=false

Sensitivity failed with 4 cliffs (trail_pct, dd_lookback, deploy_frac, dd_threshold); deploy_frac>1.0 and longer dd_lookback drive Sharpe to 0

Structural sparsity: only 33 (initial)/20 (optimized) entries over 8+ years — the >=30% capitulation trigger is too rare to populate OOS/holdout windows, so out-of-sample edge is unmeasurable and the IS Sharpe is a few bull-recovery winners fit to noise

'Tail-safe' claim contradicted: max_drawdown 79.5% (initial)/48% (optimized) — a 95%-deployed long-only book absorbs ETH's full drawdowns despite the 15% per-trade stop

EthSpotCapitulationAccumulationRunWinner aimed to exploit post-capitulation rebounds in ETH spot, buying confirmed stabilizations after >=30% drawdowns and holding for the recovery with a deliberately favorable payoff structure (avg_win roughly 2.5x avg_loss). The initial backtest looked promising enough — 91.9% return, Sharpe 2.90, positive expectancy — that the reviewer waved it through to optimization, though it flagged the thin 33-trade sample and an 79.5% drawdown. Optimization then exposed the strategy as overfit: deflated Sharpe of 0.068 below the best-of-N luck bar, PBO 0.72, all three walk-forward OOS windows at zero, a failed holdout, and four sensitivity cliffs. The root cause was structural sparsity rather than a tunable region — the capitulation trigger is simply too rare to populate out-of-sample windows — so the analyst abandoned it on attempt 1 of 2 and logged the failure pattern that rare-event mean-reversion-to-trend on a single major does not survive walk-forward.

A rare-event entry (~33 trades over 8+ years) cannot be validated out-of-sample — walk-forward and holdout windows hold too few qualifying trades, so an impressive in-sample Sharpe is selection noise, not edge.

The analyst abandoned it after optimization (iteration 1 of 2): deflated Sharpe was 0.068 vs an expected-max-luck bar of 10.46 over 225 trials (is_significant=false, PBO 0.72, sharpe_ci_low -0.41), walk-forward was overfitted with all three OOS windows at 0.0, the holdout scored 0.0, and sensitivity showed 4 cliffs.

A long-only, no-leverage, single-instrument ETHUSDT spot strategy that buys deep (>=30%) drawdowns only after the decline stabilizes, then runs the winner on the recovery for a favorable avg_win >> avg_loss payoff.

The initial backtest returned 91.9% total (Sharpe 2.90, 33 trades, 33% win rate, avg_win $41.6k vs avg_loss $16.7k) but with a 79.5% max drawdown; optimization lifted the headline Sharpe to 5.90 and total return to 419% (20 trades) yet collapsed every generalization gate.