EthSpotCapitulationBuyTheDipDaily

ETH Spot Capitulation Buy-the-Dip on Daily Bars

A simple long-only mean-reversion strategy on ETHUSDT spot that accumulates ETH after sharp drawdowns from a recent peak. The strategy exploits a well-documented pattern in major crypto assets: deep, fast drawdowns of 20%+ from recent highs typically mark short-term capitulation lows that mean-revert toward the prior trend within 1-4 weeks. The strategy uses only 4 parameters (drawdown threshold, RSI oversold, exit RSI, ATR stop multiple) to minimize overfitting. Operating on daily bars and spot venue means fees (0.20% RT) are trivial relative to expected per-trade returns of 8-15%. By avoiding leverage, funding rates, pairs trades, calendar seasonality, and multi-instrument complexity, this hypothesis sidesteps every recurring failure pattern in the recent batch.

Implements the hypothesis as a clean single-position mean-reversion that fits the base template natively — should_enter opens the long on the cached capitulation flag, should_exit closes on RSI recovery or the ATR stop — so each dip is a discrete round-trip, which both matches the proven sibling pattern (AAVE/LINK/DOT spot drawdown-accumulation strategies that landed) and naturally clears the factory's >= 5-round-trip walk-forward reliability floor. Venue is BINANCE_SPOT (CASH, leverage 1.0): the strategy is strictly long-only, unleveraged, with no funding/pairs/multi-instrument complexity, so the spot fee model (~0.20% round-trip) is correct and, per the hypothesis, trivial against expected 8-15% per-trade reversion moves. Per the capitulation thesis I intentionally omit a trend filter (the sibling's close>SMA100 guard) — the whole point is to catch the falling knife at oversold RSI — and instead bound downside with the ATR stop captured at entry (a fixed level anchored to entry-time risk, so it doesn't loosen as volatility expands), keeping the parameter count at the stated 4. ETH has experienced many 20%+ daily-scale drawdowns (2018 bear, 2021-22 unwinds, 2024-25 corrections), so real-data entries are plentiful across regimes. Verified locally: Layer 1 static (clean), Layer 2 synthetic (all 6 scenarios pass with a non-frozen SMA-deviation signal), and an 800-bar behavioral simulation with periodic 25-35% crashes confirming 20 complete entry/exit round-trips (entries on drawdown+oversold, exits on RSI recovery or ATR stop) with no stuck positions — comfortably above the trade-count floor for Layer 3 and walk-forward.

The ETH spot capitulation buy-the-dip has essentially no edge at baseline: profit factor 0.97 (below 1.0, i.e. net losing), Sharpe 0.16, negative expectancy (-$61/trade), and total return -2.3% over ~8 years (CAGR ≈ 0). The 58% win rate is undone by avg_loss ($4,328) > avg_win ($3,057) — the ATR stop cuts the dip-bounce winners short while losers run. The sample is thin and gappy (45 trades over 8 years, ~5.6/yr, with NO trades in 2023), so a 3-window walk-forward risks empty out-of-sample windows. The premise is unreliable in crypto: deep 20%+ drawdowns frequently continue (2018/2020/2022 bears continued down well past the entry), so buying them on RSI-oversold does not reliably mean-revert — annual returns are mixed (2018 -14%, 2020 -15%, 2026 -13% negative) with 2021 (+30%) the only strong year. Optimizing the four parameters (drawdown_threshold, rsi_oversold, exit_rsi, atr_stop_mult) over 225 trials on a 45-trade, no-baseline-edge series is best-of-N over noise and would fail DSR/holdout, especially given the gappy sample. This is the 'results so poor no optimization can help' case, in the same mean-reversion-fade family that has failed repeatedly this session. Not revise_hypothesis: the dip-buy mechanism has no proven sibling and the failure is absence of edge plus an unreliable premise (crypto deep-drawdown continuation), not a structurally-dead target for a working mechanism. FAILURE PATTERN: long-only daily capitulation buy-the-dip on a single crypto major (ETH spot) has ~zero/negative edge — deep drawdowns often continue rather than mean-revert, the ATR stop produces an avg_loss > avg_win profile, and the thin/gappy trade count (no 2023 trades) cannot support robust optimization.

Long-only capitulation buy-the-dip mean-reversion on ETHUSDT spot, daily bars. Each day it tracks the rolling peak (high over peak_lookback=30 bars), the drawdown from that peak, RSI(14), and ATR(14). It enters LONG (while flat) when the drawdown from the recent peak is at least drawdown_threshold (20%) AND RSI is at or below rsi_oversold (30) — i.e., a deep, fast decline into oversold territory that marks short-term capitulation. It exits when RSI recovers to exit_rsi (55), capturing the mean reversion, or when price falls to an ATR stop set at entry_price - atr_stop_mult (2.5) x ATR-at-entry. Position size is equity-relative (30% of capital per trade) with a min-notional guard. calculate_signal returns a continuous SMA-deviation diagnostic (non-frozen) and caches the discrete entry decision; should_enter/should_exit/position_size implement the round-trip. Long-only SPOT (CASH): only BUY opens and the only SELL is the exit. 4 core parameters, leverage 1.0, zero supplementary data.

Clean, simple long-only spot design (no leverage/funding/pairs); realized metrics, metrics_reliable=true

Reasonable max drawdown (33%) and low exposure (21%)

Essentially no edge: profit factor 0.97 (<1.0), Sharpe 0.16, negative expectancy (-$61/trade), total return -2.3% over ~8 years

avg_loss > avg_win at 58% win rate — the ATR stop cuts dip-bounce winners while losers run

Thin, gappy sample: 45 trades over 8 years with NO trades in 2023 — walk-forward windows risk being empty

Premise unreliable: deep 20%+ crypto drawdowns often continue (2018/2020/2022 bears), so dip-buying doesn't reliably mean-revert

EthSpotCapitulationBuyTheDipDaily was a deliberately simple, four-parameter long-only spot strategy designed to buy oversold capitulation dips and ride the bounce, explicitly avoiding leverage, funding, and pairs complexity. In ~8 years of testing the premise did not hold: deep crypto drawdowns often kept falling (2018/2020/2022 bears), the ATR stop cut dip-bounce winners while losers ran, and the result was a sub-1.0 profit factor, 0.16 Sharpe, negative expectancy, and -2.3% total return across only 45 trades with none in 2023. The analyst classified it as the 'results so poor no optimization can help' case in the repeatedly-failed mean-reversion-fade family and abandoned it before optimization. It ended after one iteration as abandoned, never advancing to optimization or risk review.

Long-only daily capitulation dip-buying on a single crypto major has near-zero/negative edge because deep 20%+ drawdowns frequently continue rather than mean-revert, and an ATR stop that cuts winners while letting losers run yields avg_loss > avg_win even at a >50% win rate — a thin, gappy sample (no 2023 trades) also cannot support robust walk-forward optimization.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) as a no-baseline-edge, thin/gappy mean-reversion-fade that optimizing four parameters over a 45-trade series could not rescue, so optimization and all later stages were never reached.

A long-only ETH spot daily mean-reversion strategy that bought capitulation dips — entering when drawdown from a 30-day rolling peak reached >=20% and RSI(14) was oversold (<=30), then exiting on RSI recovery to >=55 or an ATR-multiple stop — using just four core parameters and no leverage, funding, or pairs.

Over ~8 years and 45 trades it had essentially no edge: total return -2.3% (CAGR ~0), Sharpe 0.16 (CI [-4.31, 5.57]), profit factor 0.97, negative expectancy of -$61/trade, and a 58% win rate undone by an avg loss ($4,328) larger than avg win ($3,057), with a 33% max drawdown at low 21% exposure and no trades at all in 2023.