EthFourHourRsi2OversoldMeanReversionLong

ETH 4H RSI(2) Extreme-Oversold Velocity Mean-Reversion Long with Daily Bull Regime Filter

A long-only single-instrument VELOCITY-BASED MEAN-REVERSION strategy on ETHUSDT perpetual futures using 4-hour bars and OHLCV-only data, with a daily-bar bull regime filter. This proposal is the CROSS-ASSET COMPLEMENT to the just-added BtcFourHourRsi2OversoldMeanRevLong, applying the same RSI(2) velocity mean-reversion mechanism to ETH. The mechanism is DOUBLY FILTERED against the documented failure patterns: (1) AVOIDS the 'altcoin_daily_minimum_param_trend' pattern that abandoned SOL Donchian and AVAX Trend — ETH is a TOP-2 crypto (explicitly excluded from the altcoin pattern), uses 4H not daily (high signal density), and uses mean-reversion not trend-following (not dependent on alt-cycle peak tail trades); (2) AVOIDS the 'hypothesis specifies instrument class unsupported by FactoryStrategy' pattern that abandoned both Deribit options strategies — this uses LINEAR FUTURES (ETHUSDT.BINANCE perpetual, fully supported by FactoryStrategy), no options or multi-leg derivatives. The RSI(2) mechanism is Larry Connors' canonical short-term mean-reversion trigger documented in 'Short-Term Trading Strategies That Work' (2008): RSI(2) <= 5 represents the bottom ~2-3% of historical RSI(2) readings, marking statistical-extreme velocity exhaustion that historically precedes mean-reversion bounces. The economic mechanism on ETH is amplified by ETH-specific flows: (a) ETH has highest derivative open interest of any altcoin, producing periodic liquidation cascades that create RSI(2) extremes, (b) ETH staking-yield flows provide systematic dip-buying pressure (validators rebalance into ETH at oversold prices), (c) ETF rebalancing creates institutional bid stacks at extremes. Mechanistically distinct from every other ETH strategy in the portfolio: EthDailyGoldenCrossMomentumLong fires on rare regime-change events (Golden Cross); EthBtcRatioDailyEmaTrendLongEth fires on slow cross-asset rotation. NONE captures within-day velocity-extreme reversions, which fire ~30-60 times per year. This provides ETH with three orthogonal signal frequencies: rare (regime change), slow (cross-asset trend), fast (velocity reversion). The trade frequency of 30-60/year over 6 years = 180-360 trades total, with 60-120 trades per walk-forward window — structurally INCOMPATIBLE with tail-dependence failure pattern that requires 1-2 outlier trades to drive performance. Only 4 explicit parameters (RSI period=2, RSI threshold, regime SMA period, stop pct).

Implements the ETH cross-asset complement of the BTC RSI(2) sibling using the exact validated structure. RSI(2)<=5 is Connors' statistical-extreme velocity-exhaustion trigger, gated by a daily 200-SMA regime read from an extra daily bar in on_extra_bar so the strategy only buys dips within an established uptrend (no falling-knife catches). It is doubly filtered against documented failure patterns: ETH is top-2 (excluded from the altcoin-daily-trend pattern), 4H gives high signal density, mean-reversion does not depend on alt-cycle peak tail trades, and it uses fully-supported linear futures (not options/multi-leg). The ~30-60 entries/year make it structurally incompatible with tail-dependence. Critically, RSI(2) saturates at exactly 0/100 on smooth trends, which froze the analogous BTC strategy's signal in Layer 2; the fix carried over here keeps the true RSI in self._rsi (driving entry/exit) while calculate_signal returns a continuous RSI+tanh(SMA-deviation) diagnostic that varies every bar — verified through the actual Layer-2 proxy with the daily extra fed (no exceptions, 481 unique signals across all six scenarios). Leverage is 1.0 (the MARGIN account holds the long at 1x), so sizing references no margin leverage and avoids the inert-leverage gate; the venue is the hypothesis-mandated ETHUSDT.BINANCE perpetual.

Money-losing with an inverted payoff and no edge to optimize: 66% win rate but profit_factor 0.916, expectancy -76, Sharpe -0.38, total_return -11.4%, CAGR -3.2%, negative in 4 of 6 years (2024 -18.7%, 2025 -18.5%) over an adequate 224-trade / 6.5-year sample. The RSI(2) 'buy extreme oversold' fade wins small and often but loses big when the oversold continues down (avg_loss 2677 ~2.1x avg_win 1260, tail_ratio 0.59, skew -0.79) — the picking-up-pennies-in-front-of-a-steamroller signature — and the 224-trade churn adds 8.04%-of-gross fee drag. This is the SAME RSI(2) velocity-mean-reversion mechanism as the already-abandoned BtcFourHourRsi2OversoldMeanRevLong sibling (failed on payoff/fees), reproduced on ETH: oversold on a major 4H more often precedes further downside than a bounce, so the high win rate is overwhelmed by the fat left tail. Adequate sample confirms genuine no-edge, not noise. No parameter optimization converts a 0.916-PF, negative-Sharpe, inverted-payoff fade into an edge; tuning RSI thresholds or the stop only relocates the losses, and a sub-1-PF base would fail deflated-Sharpe. Do not spend optimization compute. FAILURE PATTERN: RSI(2) extreme-oversold velocity mean-reversion (Connors-style buy-the-dip fade) on crypto majors at 4H produces a high win rate (~66%) but an inverted payoff (avg_loss ~2x avg_win) and negative expectancy on BOTH BTC and ETH — the fade gets run over by continuation, and the high trade count + fees compound it; the velocity-reversion edge does not exist on majors regardless of the bull-regime filter.

Long-only ETH 4H RSI(2) extreme-oversold velocity mean-reversion with a daily 200-SMA bull-regime filter on ETHUSDT.BINANCE USD-M linear perpetual (leverage 1.0, OHLCV-only). It buys when the Wilder-smoothed 2-period RSI on 4H drops to <=5 while the daily bull regime holds (close > 200-day SMA), and exits on the snap-back (RSI >= 65), a 5% protective stop, or an 18-bar (~3-day) time stop. Risk-based sizing (3% of equity to the 5% stop, capped at 1x equity notional). Four core parameters.

Clean OHLCV-only implementation, correct long-only routing, adequate sample (224 trades / 6.5 years), 0 dropped, metrics_reliable=true

Not tail-dependent (the hypothesis's stated goal) — but that only confirms the no-edge is robust

Money-losing with inverted payoff: 66% win rate but profit_factor 0.916, expectancy -76, avg_loss 2677 ~2.1x avg_win 1260 (tail_ratio 0.59, skew -0.79)

Sharpe -0.38, total_return -11.4%, CAGR -3.2%, alpha -0.028; negative in 4 of 6 years (2024 -18.7%, 2025 -18.5%), only 2021 positive

High fee churn: commission 8.04% of gross on 224 trades compounds the negative gross edge

Same RSI(2)-mean-reversion mechanism as the already-failed BtcFourHourRsi2OversoldMeanRevLong sibling (payoff/fees) — fade-on-majors family

> ~0.52 to offset 66% win

EthFourHourRsi2OversoldMeanReversionLong was the ETH cross-asset complement to the BTC RSI(2) sibling, applying Connors' canonical extreme-oversold trigger with a daily bull-regime filter and explicitly engineered for enough trades to avoid the tail-dependence that had killed other strategies. It met that design goal but exposed the real flaw: over 224 trades it won 66% of the time yet lost 11.4% with a 0.916 profit factor, because its losses were ~2.1x its wins as oversold 4H ETH kept falling, and 8% fee churn deepened the bleed. The reviewer abandoned it pre-optimization, noting the adequate sample confirms a genuine no-edge — not noise — and that no threshold or stop tuning can fix an inverted-payoff fade. Decisively, the identical RSI(2) mechanism has now failed on both BTC and ETH, establishing that velocity mean-reversion does not work on crypto majors at 4H, and that a high win rate masking a fat left tail is the signature, not the edge.

RSI(2) extreme-oversold velocity mean-reversion (Connors buy-the-dip fade) on crypto majors at 4H produces a deceptively high win rate (~66%) but an inverted payoff (avg_loss ~2x avg_win) and negative expectancy on both BTC and ETH — the fade gets run over by continuation and fees compound it — so the velocity-reversion edge does not exist on majors regardless of a bull-regime filter, and a high win rate is not an edge.

The analyst abandoned it at the backtest-review gate as money-losing with an inverted payoff and no edge: the RSI(2) buy-the-dip fade wins small and often but loses big when oversold continues down (picking up pennies in front of a steamroller), with high trade-count fee churn compounding it — and it is the same mechanism as the already-abandoned BTC sibling, now reproduced on ETH, so oversold on a major at 4H more often precedes further downside than a bounce.

A long-only, OHLCV-only velocity mean-reversion strategy on ETH 4H (ETHUSDT.BINANCE perp) using Larry Connors' RSI(2): buy extreme-oversold readings (RSI(2) <= 5) only in a daily 200-SMA bull regime and exit on the RSI snap-back, a 5% stop, or a time stop — the cross-asset complement to the BTC RSI(2) sibling, designed for ~30-60 trades/year explicitly to avoid tail-dependence, on a top-2 crypto with linear futures.

It lost money over an adequate 224-trade / 6.5-year sample despite a 66% win rate: -11.4% total return (CAGR -3.2%), Sharpe -0.38, profit factor 0.916, expectancy -$76/trade, with an inverted payoff (avg_loss $2,677 ~2.1x avg_win $1,260, tail_ratio 0.59, skew -0.79), 33.3% max drawdown, commission at 8.04% of gross, and negative returns in 4 of 6 years (only 2021 positive); execution was clean (metrics_reliable=true).