AvaxFourHourRsi2PullbackUptrendLong

AVAX 4H RSI(2) Pullback in Uptrend

A long-only single-instrument mean-reversion strategy on AVAXUSDT perpetual futures using 4-hour bars. Implements the classic Connors RSI(2) extreme-oversold pullback signal, gated by a long-term trend filter (50-bar SMA). The hypothesis: in established crypto uptrends, sharp short-term pullbacks (RSI(2) below 10) represent profit-taking and weak-hand capitulation rather than trend reversal, and revert to the mean within 1-2 trading days. The 50-SMA trend filter eliminates signals during structural downtrends where mean reversion fails. This is a deliberately minimal strategy — only OHLCV data, only 3 indicators (RSI, SMA, ATR), only 4 parameters — designed to pass verification quickly and be easy to validate. Average expected trade: +2.5% over 8-32 hours, well above the 0.15% fee threshold.

Implements the classic Connors RSI(2) extreme-oversold pullback with a 50-SMA trend filter, a deliberately minimal long-only mean-reversion strategy (OHLCV-only, 3 indicators: RSI, SMA, ATR). calculate_signal returns the continuous RSI(2) value, which varies smoothly every bar, so the verifier sees real signal variation without any synthetic decoration; should_enter thresholds it (signal < entry_rsi) and the uptrend (close > 50-SMA) is the separate boolean gate, avoiding the gated-to-0.0 anti-pattern. RSI is computed with Wilder smoothing seeded by an SMA over all available closes for stability. Exits combine an RSI mean-reversion target, a Connors-style short-SMA reclaim, a wide ATR stop (mean-reversion needs room to avoid being shaken out), and a short time stop matching the 8-32h expected hold. Venue AVAXUSDT.BINANCE USD-M 4H matches the proven AVAX 4H family and futures RT taker (~0.10%) is cheaper than spot; long-only with leverage left at 1.0 (no amplification) so the leverage_set_but_unused gate does not trigger, and the ~2.5% expected per-trade move clears the ~0.10% round-trip fee comfortably.

The RSI(2) pullback-in-uptrend premise has no edge on AVAX: profit_factor 0.900, Sharpe -0.61 (CI -2.28 to 1.37), total_return -3.6% over 5.7 years, negative expectancy (-$24.6/trade), omega 0.87, PSR 0.249. The 58.4% win rate is deceptive — avg_loss $591 exceeds avg_win $379 with return_skew -1.56 and tail_ratio 0.60, the classic picking-up-pennies mean-reversion profile of many small wins and occasional larger losses. The edge is marginal-to-none and decayed: tiny-positive in 2021-2024 (+0.5 to +2.2%/yr) then 2025 -7.0% with rolling Sharpe to -4 by 2026, and commission is 7.51% of gross (fees eat the thin per-trade edge on ~17-hour trades). The hypothesis projected +2.5%/trade; actual is ~-0.13%. The code is correct and trades exactly as specified (185 entries, 0 dropped), so this is a refuted premise, not a bug: RSI(2) oversold washouts in AVAX uptrends continue (the pullback extends) more often than they revert profitably. Not optimize: no RSI/SMA threshold tuning flips a systematically-losing fade with negative Sharpe/expectancy and asymmetric tails into a winner. Not iterate: the code correctly implements the premise; the only 'fix' is a different mechanism. Not worth 2 hours. FAILURE PATTERN: Connors RSI(2) oversold mean-reversion on a high-beta crypto major produces a net-losing, negatively-skewed result (PF 0.90, negative Sharpe/expectancy, avg_loss > avg_win) because short-term washouts on alts are continuation signals that extend rather than revert -- the same sign-wrong reversion failure as the abandoned BNB RSI, BTC VWAP-distance, and BTC vol-spike reversion strategies; oscillator-based mean reversion does not generate a fee-surviving edge on majors, and the high win rate masks fat-tailed losses.

Long-only Connors RSI(2) oversold-pullback mean reversion on AVAXUSDT.BINANCE USD-M futures, 4H bars, pure OHLCV single feed. Enters LONG when RSI(2) < 10 while price is above the 50-bar SMA (uptrend filter that screens out structural downtrends). Exits when RSI(2) reverts >= 70, when close reclaims the 5-bar SMA, on a 2.5x ATR stop, or after a 12-bar (~2 day) time stop. The signal is the continuous RSI(2) value (0-100, varies every bar); should_enter thresholds it and the uptrend boolean gates it. Sized at 20% equity notional, capped so ATR-stop risk <= 1.5% of equity. leverage 1.0.

Implemented correctly and trades as specified: 185 RSI(2) pullback entries, 0 dropped, all long; healthy sample (~32/yr), minimal design, low max_drawdown 8.8%

Net losing: profit_factor 0.900, Sharpe -0.61 (CI -2.28 to 1.37), total_return -3.6%, expectancy -$24.6/trade, omega 0.87, Sortino -0.63, PSR 0.249

High win rate (58%) is deceptive: avg_loss $591 > avg_win $379, return_skew -1.56, tail_ratio 0.60 — picking-up-pennies mean-reversion with fat left tails

Marginal/decayed edge: tiny-positive 2021-2024 (+0.5 to +2.2%/yr) then 2025 -7.0%, rolling Sharpe to -4 by 2026; the +2.5%/trade thesis is refuted (actual ~-0.13%)

Fee-fragile: commission 7.51% of gross on ~17-hour reversion trades — fees consume the thin per-trade edge

This deliberately minimal strategy bought RSI(2) oversold pullbacks in AVAX uptrends, expecting a quick mean-reversion bounce of +2.5% per trade. The implementation was clean (185 entries, none dropped) and the win rate looked attractive at 58%, but the payoff geometry was inverted — average losses ran larger than average wins with a heavy left tail — yielding profit factor 0.90, Sharpe -0.61, -3.6% total, and an actual per-trade return near -0.13%, with the edge decaying sharply into 2025/2026. The analyst abandoned it before optimization, concluding the premise is backwards: short-term washouts on a high-beta alt extend rather than revert, fees eat the razor-thin edge, and no threshold tuning rescues a sign-wrong fade. It was recorded alongside the other oscillator-reversion failures on majors this run.

Oscillator-based mean reversion (Connors RSI-2) does not generate a fee-surviving edge on a high-beta crypto major — short-term washouts on alts are continuation signals that extend rather than revert, and a high win rate paired with avg loss > avg win and negative skew is the picking-up-pennies profile, the same sign-wrong reversion failure as the abandoned BNB RSI, BTC VWAP-distance, and BTC vol-spike strategies.

The analyst abandoned it at backtest review as a refuted premise rather than a bug — the code traded exactly as specified, but RSI(2) oversold washouts in AVAX uptrends continue (the pullback extends) more often than they revert profitably, fees consume the thin per-trade edge on ~17-hour trades, and no RSI/SMA tuning flips a negative-expectancy fade with asymmetric tails into a winner.

A long-only Connors RSI(2) mean-reversion on AVAXUSDT 4H perpetual futures (pure OHLCV), buying extreme short-term oversold washouts (RSI(2) < 10) only within a confirmed uptrend (close > 50-SMA), on the thesis that sharp pullbacks in uptrends are profit-taking that reverts within 1-2 days, exiting on RSI reversion ≥70, a 5-SMA reclaim, a wide ATR stop, or a time stop.

Over a 185-trade sample (~32/year, all long, none dropped) it was net-losing: profit factor 0.900, Sharpe -0.61 (CI -2.28 to 1.37), total return -3.6%, expectancy -$24.6/trade (~-0.13%, versus the hypothesized +2.5%/trade), low 8.8% max drawdown; the 58.4% win rate was deceptive because average loss ($591) exceeded average win ($379) with return skew -1.56, and the edge decayed (tiny-positive 2021-2024, then 2025 -7.0% with rolling Sharpe to -4 by 2026), with commission at 7.51% of gross.