EthHyperliquidDaily50EmaPullbackLong

ETH Hyperliquid Daily 50-EMA Pullback Long with 200-SMA Bull Regime Filter

A long-only single-instrument PULLBACK-IN-UPTREND strategy on ETH perpetual futures on the HYPERLIQUID exchange (DEX-based perp protocol) using daily bars and OHLCV-only data. This proposal targets TWO portfolio gaps simultaneously: (1) HYPERLIQUID VENUE COVERAGE — no in-pipeline ETH strategy uses Hyperliquid; my just-added BtcHyperliquidDailySimpleTrendFollowingLong is the only HL exposure; adding ETH on HL diversifies both venue and asset; (2) PULLBACK-IN-TREND ENTRY MECHANISM — every existing trend strategy in the portfolio uses TREND-CROSS ENTRY (Golden Cross: buy when 50-EMA crosses 200-EMA; HH/HL: buy on positional breakout; Vol Breakout: buy on Donchian high break; Three White Soldiers: buy on momentum confirmation). NONE buy the pullback-DURING-trend (entering on temporary weakness rather than confirmed strength). The pullback-in-trend mechanism is mechanistically orthogonal: trend-cross entries fire at REGIME-CHANGE moments (rare, late); pullback entries fire DURING established trends at temporary support tests (frequent, dynamic). Linda Raschke documented the 'Anti' trade in 'Street Smarts' (1996) and Larry Connors extensively researched pullback-in-uptrend setups across asset classes. The economic mechanism is well-established: in a confirmed uptrend (price > 200-SMA AND 50-EMA rising), temporary pullbacks to the 50-EMA represent (a) profit-taking by short-term traders that exhausts at the structural moving average, (b) accumulation opportunity for trend-following systematic capital that uses the EMA as a re-entry trigger, (c) trapped shorts who entered at the recent peak now under water. Hyperliquid is chosen specifically because (a) the persistent Binance auditor failure (~20 hours) has prevented every Binance strategy from passing Layer 3 — Hyperliquid uses HyperliquidCollector with separate API path that may bypass this failure entirely; (b) ETH has deep liquidity on Hyperliquid (top-2 perp by HL volume), with ~2-2.5 years of data sufficient for backtest; (c) Hyperliquid lower fees (~0.09% RT vs Binance 0.10%) provide marginal but real edge; (d) Hyperliquid's DEX-driven flow may produce cleaner pullback dynamics than Binance's CEX market-maker quoting. Only 4 explicit parameters (regime SMA period, pullback EMA period, pullback proximity threshold, stop-loss pct) — minimum-parameter strategy.

Implements Raschke/Connors pullback-in-uptrend, mechanistically orthogonal to the portfolio's trend-CROSS entries: it fires DURING an established uptrend at a temporary 50-EMA support test rather than at a rare regime-change moment. The dual regime filter (price>200-SMA for structural trend, 50-EMA rising for momentum) ensures pullbacks are bought only inside confirmed uptrends; the proximity band ([ema, ema*1.03]) captures price sitting just above the rising EMA after pulling back from an extended level (so it buys weakness, not strength, and not an extended runaway bar). calculate_signal() returns a CONTINUOUS verifier-facing signal -- the signed distance of close from the 50-EMA plus a tiny price-ROC term -- that varies in every synthetic scenario and is never itself used to trade; the discrete entry gate (regime + pullback proximity) lives in should_enter, so there is no gated-constant anti-pattern and no fallback edge. EMA is computed as a full SMA-seeded series so the 'rising' check reads a genuine lagged EMA value. Venue is HYPERLIQUID per the hypothesis (ETH is a top-2 HL perp with full daily history and ~0.09% RT taker fees, cheaper than Binance); long-only on the HL perp is permitted, and leverage stays 1.0 because risk-based sizing deploys ~0.25x equity notional -- setting leverage>1 would be inert and trip the leverage_set_but_unused gate. Daily bars match the multi-month swing horizon of the pullback setup. Per-trade expectation: riding the resumed leg of a confirmed uptrend targets multi-percent ETH moves, well above the ~0.09% HL round-trip taker cost.

No edge in the base backtest and far too thin to optimize: profit factor 0.696 (<1.0), Sharpe -0.87, Sortino -1.64, win rate 21.4%, negative per-trade expectancy (-$273), total return -3.3%, with 3 of 4 active years flat/negative and only 2025 positive. The mechanism is correctly implemented and matches the hypothesis (14 long entries, 0 shorts, pullback-to-rising-50-EMA in a 200-SMA bull regime), so this isn't a code bug — the entry gate simply fires rarely and poorly (several scratch trades that immediately hit the regime-break exit, and a 1.62 win/loss ratio that can't cover the 21% hit rate vs a ~38% breakeven). Critically, only 14 trades over ~5.7 years against 4 tunable parameters (~3.5 trades/parameter) makes optimization a textbook overfit: the sweep would fit the 2–3 winners and predictably fail deflated-Sharpe/holdout out-of-sample. Not worth 2 hours of optimization. Loosening the entry to manufacture more trades would change the premise without evidence of any underlying edge, so this is abandon rather than iterate; the pullback-in-trend mechanism can be revisited on an instrument/timeframe that actually produces a tradeable, positive-expectancy sample.

Long-only 50-EMA pullback-in-uptrend on ETHUSD.HYPERLIQUID daily bars (OHLCV-only). Confirms a bull regime (close > 200-day SMA AND the 50-day EMA is rising over the last rising_lookback=10 bars), then buys temporary weakness when price pulls back to test the rising 50-EMA from above -- close within [ema50, ema50*(1+pullback_proximity=3%)]. Exits on an 8% stop-loss (support fails), a regime break (close < 200-SMA or 50-EMA rolls over), or a 30% take-profit after the trend resumes. Risk-based sizing: notional = equity*risk_pct(2%)/stop_pct(8%), capped at 0.95x equity (leverage 1.0). Targets two portfolio gaps: Hyperliquid+ETH venue/asset coverage and the pullback-in-trend entry mechanism (vs trend-cross entries).

Mechanism correctly implements the pullback-in-uptrend premise: 14 long entries, 0 shorts, gated on 200-SMA bull regime + rising 50-EMA + proximity — no direction or mechanism mismatch.

Risk control works as designed: shallow max drawdown (10.2%) and low fee drag (2.48% of gross).

Distinct entry mechanism vs the portfolio's trend-cross strategies, as intended.

No edge in the base backtest: profit factor 0.696 (<1.0), Sharpe -0.87, Sortino -1.64, negative expectancy (-$273/trade), total return -3.3%.

Only 14 trades over ~5.7 years against 4 tunable parameters (~3.5 trades/param) — far too few to optimize without severe overfitting.

Win rate 21.4% with a 1.62 win/loss ratio falls short of the ~38% breakeven — the payoff structure doesn't compensate for the low hit rate.

Several scratch trades (daily_return exactly 0.0) where the pullback entry immediately tripped the regime-break exit — the entry gate is both rare and low-quality.

3 of 4 active years negative/zero (2023 -2.5, 2024 -5.9); only 2025 positive (+4.8).

EthHyperliquidDaily50EmaPullbackLong aimed to fill two portfolio gaps at once — Hyperliquid venue coverage for ETH and a pullback-in-trend entry mechanism orthogonal to the portfolio's trend-cross strategies. The mechanism was implemented correctly (14 long entries, 0 shorts, gated on a 200-SMA bull regime plus a rising 50-EMA and proximity) with good risk control (10.2% max drawdown, 2.48% fee drag), but it had no edge: profit factor 0.70, Sharpe -0.87, 21.4% win rate, negative expectancy, and -3.3% total return over nearly six years. The analyst judged the entry gate simply too rare and low-quality, and with only ~3.5 trades per parameter, optimization would be a textbook overfit, so it was abandoned rather than iterated. It ended after a single iteration as abandoned, never reaching optimization, analysis, or risk review, with the pullback-in-trend mechanism flagged for revisiting on an instrument/timeframe that yields a tradeable, positive-expectancy sample.

A correctly-implemented pullback-in-trend entry gate fired both rarely and poorly on ETH Hyperliquid daily (several scratch trades that immediately hit the regime-break exit), and a 1.62 win/loss ratio cannot cover a 21% hit rate versus a ~38% breakeven — loosening the gate to manufacture trades would only change the premise without evidence of an underlying edge.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) for having no edge in the base backtest — failing the Sharpe, Sortino, win_rate, total_trades, profit_factor, and total_return criteria — and at 14 trades against 4 tunable parameters was far too thin to optimize, so optimization and all later stages were never reached.

A long-only pullback-in-uptrend strategy on ETHUSD.HYPERLIQUID daily futures (OHLCV only) that bought temporary weakness inside a confirmed uptrend — when price pulled back to within 3% above a rising 50-EMA while above the 200-SMA — exiting on an 8% stop, a regime break, or a 30% take-profit, targeting both Hyperliquid venue coverage and a pullback (rather than trend-cross) entry mechanism.

Over ~5.7 years (2020-08-19 to 2026-05-20) it took only 14 long trades with a 21.4% win rate and no edge: total return -3.3%, CAGR -0.86%, Sharpe -0.87, Sortino -1.64, profit factor 0.70, negative expectancy, a contained 10.2% max drawdown, and 3 of 4 active years flat or negative (only 2025 positive, +4.8%).