Strategies

ETH Daily Three-Day Down-Close Reversal Long in Confirmed Uptrend

Outcome: Abandoned

EthDailyThreeDownCloseReversalUptrendLong

Outcome Summary

This strategy applied the Connors 3-down-close-in-uptrend reversal pattern to ETH daily, buying oversold dips in a confirmed uptrend on the thesis they snap back within days, and it was implemented and tested fairly. But it lost money — profit factor 0.74, Sharpe -0.70, -17.6%, negative in four of five years — despite a 51.9% win rate, because of a structural asymmetry: the many small +4% snap-back wins were erased by the few crash days it bought into, where the 5% daily-close stop gapped through to -25%/-30% (largest loss ~30% of equity). The analyst abandoned it at the backtest-review gate on the first iteration, judging the capped-upside/uncapped-gap-downside tail inherent to daily bars and sizing-invariant — with only 27 trades too thin to optimize — so it never advanced to optimization or risk review.

Strategy report

Hypothesis

A LONG-ONLY short-term reversal strategy on ETHUSDT.BINANCE perpetual futures (Binance USD-M) using DAILY bars. Entry trigger: ETH closes DOWN for 3 consecutive days while still in a confirmed long-term uptrend (price > 100-day SMA, 100-day SMA > 200-day SMA, 100-day SMA rising). Exit within 5 days on profit target, stop, or time. This is the Connors-Sen 'Short-Term Trading Strategies That Work' 3-day-down-in-uptrend pattern — a well-documented short-term reversal setup with reported 65-75% win rates and 2-3 day holding periods across stocks, ETFs, futures, and crypto. Mechanism: in a healthy uptrend, 3 consecutive down days represents a normal oversold pullback (not trend breakdown) and is statistically followed by a snap-back rally within 1-5 days. The uptrend filter (>100-SMA, 100>200, 100 rising) ensures we only trade reversals during the regime where short-term reversal works; in a downtrend, 3 consecutive down days are just the trend continuing and the strategy would lose. ETH is the chosen instrument because: (1) the current pipeline has only 1 ETH strategy (Golden Cross Momentum Long — a slow momentum entry); a short-term reversal pattern is structurally orthogonal and adds within-ETH diversification; (2) ETH has 5+ years of clean Binance USD-M data, no deprecation risk, in the precision map; (3) ETH has historically shown cleaner short-term reversal patterns than BTC (BTC's higher institutional ownership dampens the retail-panic-then-snapback dynamic that the strategy exploits). Structurally orthogonal to every prior session failure and every pipeline strategy: it is NOT a breakout (entry on weakness, not strength), NOT a moving-average pullback (uses consecutive-day-count not MA distance — different from XRP pullback already in pipeline), NOT sub-daily mean reversion (daily bars, low frequency), NOT cross-asset, NOT supp-data dependent, NOT accumulation, NOT options, NOT cross-venue, NOT calendar, NOT a death-cross short. Trade frequency naturally falls in the 30-50/year range (3-down-day events are common but uptrend filter prunes many), holding period 1-5 days produces many discrete closed trades that the factory framework can evaluate via walk-forward.

Implements the Connors-Sen 'Short-Term Trading Strategies That Work' 3-day-down-in-uptrend pattern. calculate_signal returns a continuous, bar-varying conviction score (streak + pullback depth) rather than gating to 0.0, while the three uptrend conditions are stored as self._regime_ok and ANDed in should_enter against the entry_streak threshold (default 3.0) — so the strategy only buys oversold dips during the regime where short-term reversal works, and avoids buying into genuine downtrends. min_bars_required clears 200+slope so SMA200 uses a full window. Exit blends a profit target (snap-back), a stop (failed reversal), a 5-day-SMA recovery exit (Connors-style), and a 5-day time stop to keep holds in the documented 1-5 day band. 4% target comfortably clears the ~0.10% round-trip futures taker cost. Venue: BINANCE USD-M futures at leverage 1.0 — long-only with no margin needed, and futures RT taker (~0.10%) is half of spot (~0.20%), making 1x futures the cheapest viable route while matching the hypothesis instrument; leverage is left at 1.0 (not referenced in sizing) to avoid the leverage_set_but_unused gate.

Fair test of the Connors 3-down-in-uptrend premise that fails on negative-EV economics, not a coding/sizing artifact. The strategy is correctly implemented (win_rate 51.9%, near the claimed setup), sizing is ~as intended (avg_position_pct 98.4%, no blow-up), and there is no return artifact — so the negative result is real. It loses (profit_factor 0.74, Sharpe -0.70, expectancy -$815/trade, negative in 4 of 5 years) because of a STRUCTURAL asymmetry: winners are capped at the +4% profit target while losers gap through the 5% stop to -25%/-30% on daily bars (largest_loss -$30,800 ≈ 30% of equity, a 6x stop overshoot; avg_loss 1.45x avg_win; skew -2.49). This is inherent to buying 3 consecutive down days — occasionally that is the first leg of a real crash (May 2021 -25%, Mar 2024 -11%), and a stop checked only at daily close cannot contain an overnight gap. Not iterate: smaller position sizing reduces drawdown magnitude but does NOT change the sign of expectancy or PF 0.74 (sizing-invariant), and no daily-bar stop/target tuning escapes the gap-through tail; with only 27 trades the sample is also too thin to optimize. FAILURE PATTERN: long-only short-term reversal that buys consecutive down-closes in an uptrend on a daily-bar crypto major has capped upside (fixed profit target) versus uncapped gap-through downside, producing negative expectancy (PF 0.74) despite a ~52% win rate — the few crash days the strategy buys into (caught at daily resolution where a 5% stop becomes a -25% to -30% loss) erase the many small snap-back wins, and no parameter tuning can contain an overnight gap on daily bars.

Implementation

Long-only short-term reversal on ETHUSDT.BINANCE USD-M perpetual futures using DAILY bars. Enters long after >=3 consecutive down closes while a confirmed long-term uptrend is in place (price > 100-day SMA, 100-day SMA > 200-day SMA, 100-day SMA rising over the last 10 days). The signal is continuous (consecutive down-day count + fractional pullback depth from the recent 10-day high) and recomputed every bar; the uptrend regime is a separate boolean gate. Exits within 5 days on a 4% profit target, 5% stop-loss, recovery above the 5-day SMA, or a time stop. Position sized at 95% of equity notional, leverage 1.0, never shorts.

Backtest Review

Sharpe: -0.70
Total return: -17.61%
Max drawdown: 35.25%
Trades: 27
Win rate: 51.9%
Profit factor: 0.74

Correctly implements the Connors 3-down-in-uptrend pattern; win_rate 51.9% is roughly in line with the documented setup

Fair, interpretable test: avg_position_pct 98.4% matches intended sizing (no blow-up), no return artifact, clean entry gate (27 signaled = 27 submitted)

Low exposure (3.88%) — the uptrend filter does prune as intended

Negative edge: profit_factor 0.74, Sharpe -0.70, expectancy -$815/trade, losing in 4 of 5 years

Structural capped-upside / uncapped-downside asymmetry: winners capped at the +4% target while losers gap through the 5% stop to -25%/-30% (largest_loss -$30,800 = ~30% vs a 5% stop), avg_loss 1.45x avg_win, skew -2.49

Unfixable on daily bars: buying 3 down days occasionally buys the start of a crash (May 2021 -25%, Mar 2024 -11%), and a daily-close stop cannot contain an overnight gap — inherent left-tail risk

Reckless 0.95 near-all-in sizing amplifies the gap losses (35% max drawdown)

Sparse: only 27 trades over 6 years — thin for walk-forward even if the edge were positive

Outcome Summary

A documented ~52% win-rate reversal setup can still be negative-EV on a daily-bar crypto major — capping upside at a fixed target while leaving downside to gap through the stop (a 5% stop becoming a -25%/-30% loss on the crash days you buy into) inverts the payoff, and no daily-bar stop/target tuning can contain an overnight gap.

The backtest-review analyst issued an 'abandon' verdict: the pattern was correctly implemented and fairly tested (no artifact, sizing as intended), so the negative result is real — a structural capped-upside/uncapped-downside asymmetry where winners cap at +4% while losers gap through the 5% stop to -25%/-30% because buying 3 down days occasionally buys the first leg of a crash and a daily-close stop cannot contain an overnight gap; not fixable by sizing or tuning, and only 27 trades anyway.

A long-only short-term reversal strategy on ETHUSDT daily Binance futures (~0.95 position fraction) implementing the Connors 3-day-down-in-uptrend pattern — buying after 3 consecutive down closes while a confirmed uptrend held (price > 100-SMA, 100-SMA > 200-SMA, 100-SMA rising) and exiting within 5 days on a +4% target, 5% stop, MA recovery, or time stop.

Across 27 trades it returned -17.6% with profit factor 0.74, Sharpe -0.70, expectancy -$815/trade and a 35.2% max drawdown, losing in 4 of 5 years — and despite a 51.9% win rate roughly in line with the documented setup, the payoff was deeply asymmetric (avg loss 1.45x avg win, skew -2.49, largest loss -$30,800 ≈ 30% of equity, a 6x overshoot of the 5% stop).

Backtest and paper results are hypothetical. Trading involves risk of loss.