EthFourHourRsi2OverboughtMomentumContinuationLong

ETH 4H RSI(2) Extreme-Overbought Momentum-Continuation Long with Daily Bull Regime Filter

A long-only single-instrument MOMENTUM-CONTINUATION strategy on ETHUSDT perpetual futures using 4-hour bars and OHLCV-only data, with a daily-bar bull regime filter. This proposal is the EMPIRICALLY-CORRECTED INVERSION of the failed EthFourHourRsi2OversoldMeanRevLong strategy. The analyst's failure analysis of that strategy revealed a critical asymmetry: **RSI(2) mean-reversion is BTC-specific**. On BTC, oversold conditions revert because institutional flows and ETF rebalancing absorb supply at extremes (PF 1.29, Sharpe +1.41). On ETH, the same RSI(2) oversold conditions produce CONTINUATION because: (a) leverage-driven liquidation cascades create runaway selling rather than absorption, (b) staking validators EXIT when ETH drops below cost basis (rather than entering as the BTC institutional buyers do), (c) ETH has higher derivative open interest concentration that amplifies one-way moves. This proposal applies the SYMMETRIC CORRECTION: if ETH oversold = downward continuation, then ETH overbought IN A BULL REGIME = upward continuation. The leverage and staking dynamics that produce runaway selling on oversold ETH should produce runaway buying on overbought ETH in confirmed uptrends: (a) leveraged long entries cascade as RSI(2) extremes attract momentum traders, (b) staking validators ADD to positions during uptrends (rewards-rebalancing flows), (c) ETF inflows accelerate during ETH-cycle momentum phases. The mechanism: 4H RSI(2) >= 95 (extreme overbought — top ~2-3% of historical readings) in bull regime triggers a long entry expecting CONTINUATION over the next 4-12 hours. This is mechanistically distinct from EthDailyGoldenCrossMomentumLong (which fires on RARE 50/200 EMA cross regime-change events ~once per year) and from EthBtcRatioDailyEmaTrendLongEth (which captures slow cross-asset rotation). RSI(2) overbought continuation fires ~30-50× per year, providing intra-trend momentum capture that the regime-change strategies miss. This passes ALL failure-pattern filters: (1) ETH is top-2 (not mid-cap altcoin), (2) 4H is high-frequency (not daily-rare), (3) momentum-continuation (not trend-following with tail dependence), (4) linear futures (framework supported), (5) explicit articulation of WHY this mechanism produces SAME-DIRECTION signal on ETH (the analyst's required justification for cross-asset proposals).

Fixes the Layer-3 'no trades' failure with the smallest viable change. Diagnosis: should_enter never fired because its two-part gate (RSI(2) >= 95 AND bull_regime) requires bull_regime, which in the previous version was set ONLY from a separate daily extra-bar stream (on_extra_bar -> _daily_closes -> _update_regime). RSI(2) >= 95 is a frequent event across 14k volatile 4H bars, so the perpetual blocker was bull_regime staying False — the daily regime stream was not populating _daily_closes (same-instrument daily extra bar not delivered), leaving the regime filter permanently off. The fix removes that external dependency and derives the identical bull-regime condition from the primary 4H bar stream that is verifiably flowing (14016 bars processed): regime = current 4H close > mean of the last regime_period*bars_per_day (=50*6=300) 4H closes, equivalent to a daily SMA(50) regime. self._bars holds up to 500 bars so the 300-bar window fits. min_bars_required is raised to cover the regime window. The daily extra bar type is dropped from the config, eliminating both the failure mode and any daily-data-availability risk. All previously passing logic is untouched: the continuous RSI(2) signal (Layer-2 varied-signal safety), the exit ladder, capital-relative leverage-consuming sizing (0.60*2.0 = 1.2x notional, so leverage is genuinely used and passes the leverage gate), and the clean import set. In a confirmed uptrend RSI(2) routinely spikes >=95 on up bars, so this now yields the hypothesized ~30-50 entries/year.

The strategy has a fundamentally negative edge that no optimization can fix. Over 461 trades (large, representative sample) the ETH 4H RSI(2)>=95 overbought momentum-continuation long returns -50.1% with Sharpe -1.17, Sortino -1.82, profit_factor 0.82, expectancy -$120/trade, and PSR 0.085, losing in 6 of 7 years (2020 -19%, 2022 -26%, 2025 -21%) with a 53.6% max drawdown spanning essentially the entire backtest. The trades correctly implement the hypothesis, so this is not a code bug — it is a refuted premise: ETH extreme-overbought readings do NOT lead to upward continuation, so the 'symmetric inversion' thesis is empirically false. Optimization tunes parameters over an existing edge; it cannot create one, and no RSI-threshold / stop-TP / regime-window sweep will flip a -1.17 Sharpe, 0.82 PF signal positive. This is the second failed inversion in the chain (RSI(2) mean-reversion failed, and this overbought-continuation inversion fails too), confirming RSI(2) extremes carry no exploitable same-direction edge on ETH 4H. Abandon rather than spend 2 hours optimizing a losing signal; further RSI(2)-extreme variants on ETH should not be re-proposed.

Long-only ETH 4H RSI(2) extreme-overbought momentum-continuation strategy on ETHUSDT.BINANCE USD-M futures, gated by a bull-regime filter derived from the primary 4H bars (close > SMA of the last regime_period days, i.e. regime_period*bars_per_day 4H bars — the 4H equivalent of a daily SMA(50)). Enters long when RSI(2) >= 95 in a bull regime, expecting continuation over the next 4-12h. Exits on stop-loss, take-profit, RSI fade below 50, or a 3-bar time stop. Leverage 2.0 (notional ~1.2x equity). Pure OHLCV, no supplementary data, no extra instruments or bar types.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: ETHUSDT.BINANCE-4-HOUR-LAST-EXTERNAL, Bars processed: 14016 Diagnostics: should_enter() returned a side 0 times over 13997 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Clean implementation matching the hypothesis: 461 long-only RSI(2)>=95 bull-regime entries, adequate sample, OHLCV-only, no data-availability issues

Negative edge, not a calibration issue: total_return -50%, Sharpe -1.17, Sortino -1.82, profit_factor 0.82, expectancy -$120/trade, PSR 0.085

Loses in 6 of 7 years (2020 -19%, 2022 -26%, 2025 -21%, 2026 -7.6%); max_drawdown 53.6% over a 2202-day drawdown spanning the whole test

The core thesis is refuted: ETH extreme-overbought does NOT produce upward continuation — buying it has negative expectancy across all regimes

Severe fee drag (commission 18.65% of gross, turnover 640%) compounds an already-negative gross edge

Optimization cannot flip the sign of the edge; parameter tuning over a losing signal wastes 2 hours

EthFourHourRsi2OverboughtMomentumContinuationLong was the second attempt in an RSI(2) chain: after ETH RSI(2) mean-reversion failed, this strategy inverted the logic to buy extreme-overbought readings in a bull regime, expecting leverage-driven continuation. The clean, hypothesis-faithful implementation generated 461 trades but bled steadily — -50.1% total return, Sharpe -1.17, a 53.6% drawdown spanning nearly the entire test, and losses in six of seven years, worsened by heavy fee drag. Because the trades correctly executed the thesis, the analyst judged this a refuted premise rather than a fixable defect and abandoned it before optimization. The chain's two failures together confirm that RSI(2) extremes offer no exploitable directional edge on ETH 4H, and further variants of it were explicitly discouraged.

RSI(2) extremes carry no exploitable same-direction edge on ETH 4H in either direction — both the mean-reversion and the overbought-continuation inversion lost — so 'symmetric inversion' of a failed thesis is not a reliable way to manufacture an edge, and further RSI(2)-extreme variants on ETH should not be re-proposed.

The analyst abandoned it at the pre-optimization BACKTEST_REVIEW gate (verdict: abandon; optimization never ran) because the negative result reflects a refuted premise, not a calibration bug — the trades correctly implement the hypothesis, so no parameter tuning can flip a -1.17 Sharpe / 0.82 PF signal positive.

A long-only ETH 4H momentum-continuation strategy that buys extreme-overbought conditions (RSI(2) ≥ 95) inside a 4H-derived bull regime, built as the 'symmetric inversion' of a failed ETH RSI(2) mean-reversion strategy on the thesis that ETH's leverage cascades and validator/OI dynamics turn extremes into continuation rather than reversion.

Over 461 long trades it lost decisively — total return -50.1%, Sharpe -1.17, Sortino -1.82, profit factor 0.82, expectancy -$120/trade, PSR 0.085, a 41.6% win rate, and a 53.6% max drawdown — losing in 6 of 7 years (2020 -19%, 2022 -26%, 2025 -21%, 2026 -7.6%), with severe fee drag (commissions 18.65% of gross on 640% turnover).

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['rsi_period', 'min_notional', 'position_pct', 'max_hold_bars', 'regime_period', 'rsi_threshold', 'stop_loss_pct', 'rsi_exit_level', 'take_profit_pct'] Check that __init__ sets all attributes from self.parameters.get(). - steady_downtrend: Frozen signal: all 481 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies.