AdaFourHourBullishRsiDivergenceLong

ADA 4H Bullish RSI Divergence Long-Only with Trend Filter on ADAUSDT.BINANCE

A long-only, single-instrument momentum-divergence strategy on ADAUSDT.BINANCE — confirmed testable per the recent successful ADA-4H single-instrument pipeline additions (ADADonchianBreakoutLong4HOnly, ADAWeekendCalendarEffectLong, ADAVolContractionSqueezeLong all backtested cleanly on ADA-4H bars alone). Structurally distinct from every prior ADA pipeline strategy: those are breakout (Donchian), calendar effect, vol regime, mean reversion (Keltner failed; BB-MR works), and pairs. This is the FIRST MOMENTUM-DIVERGENCE strategy in the portfolio. The mechanism (Wilder 1978 'New Concepts in Technical Trading Systems'; Pring 1991 'Technical Analysis Explained'; Murphy 1999 confirms in academic-adjacent practitioner literature; Caginalp-Balenovich 2003 formal model): when price makes a new LOWER LOW while RSI(14) makes a HIGHER LOW, momentum is positively diverging from price — sellers are exhausting even as the price drifts lower. This bullish divergence reliably precedes mean-reverting rallies in 55-65% of confirmed setups (per technical analysis literature, replicated in crypto-specific Caporale-Plastun 2018). The 4H trend filter (close > 50-period SMA) restricts entries to assets that have started transitioning OUT of bear regimes — accumulation phases, NOT mid-bear catches. This is critically different from the FAILED Keltner symmetric range trade: that strategy fought the trend; this strategy LOOKS FOR DIVERGENCE AT THE TRANSITION between regimes. Uses ONLY ADAUSDT.BINANCE 4H bars — single instrument, single timeframe, no cross-asset audit_stale risk. Fills critical gaps: (1) FIRST MOMENTUM-DIVERGENCE strategy in portfolio, (2) Single-instrument simplicity (no multi-instrument auto-collection failures), (3) Long-only direction matching the proven multi-timeframe-breakout success structure, (4) Confirmed-testable instrument. Position sizing: 18% per trade. Risk per trade: ~1.5% capped by stop-loss.

Iteration 2 addresses the 'too few trades (15) / near-zero edge' feedback by loosening ONLY the divergence detection thresholds while keeping the bullish-divergence mechanism, exits, and sizing byte-identical to the version that already passed Layers 1-3. Specific density changes: (1) pivot_strength 3->2 — the largest lever, roughly squaring the pool of qualifying confirmed swing lows by requiring one fewer confirming bar on each side; (2) div_lookback 40->60 so more pivot pairs are comparable across the trailing window; (3) rsi_low_max 50->55 widens the weak-momentum region so valid divergences whose first pivot RSI sits just above 50 still qualify; (4) div_confirm_bars 2->3 widens the just-confirmed window so a divergence isn't missed by a single bar of latency; (5) new sma_tolerance=0.97 relaxes the hard close>SMA gate to close>0.97*SMA, admitting the exact transition-line accumulation setups the hypothesis targets (and mirrored in the exit trend-break to avoid premature stops on the same band). No look-ahead is introduced: pivots remain confirmed k=pivot_strength bars after formation. These changes target ~40-60 trades across 2021-2026 so walk-forward and holdout each hold multiple trades.

Base backtest has negative edge over a full 6-year all-regime sample (profit_factor 0.97, Sharpe -0.14, expectancy -$12.87/trade, CAGR -16.7%, negative alpha/information_ratio vs its own ADA buy-hold), so there is no edge for the optimizer to refine -- optimization fits parameters to an existing signal, it does not manufacture one. Compounding this, realized win_rate (28.4%) directly contradicts the hypothesis's 55-65% divergence-success premise, so the mechanism itself does not behave as claimed on ADA, and performance is regime-decaying (positive 2020-21, losing 2023-2025). This is a demonstrated non-edge on a valid instrument with no code defect to iterate on; spending 2 hours optimizing 81 net-negative trades would curve-fit noise and predictably fail the deflated-Sharpe/PBO gates downstream.

Long-only ADA 4H bullish RSI(14) divergence strategy with a 50-SMA trend filter (with tolerance band). Detects price lower-lows against RSI higher-lows at confirmed causal swing pivots, entering long during the transition out of bear regimes. Exits on fixed TP/SL, RSI overbought, trend break, or time stop. Single instrument, single timeframe, no leverage.

Strategy functions correctly: 81 long-only divergence entries fired cleanly on ADA 4H single-instrument/single-timeframe, no data or execution drops (entry_diag 81 signaled -> 81 submitted).

Trades match the hypothesized mechanism (all longs, low 6.75% exposure, ~1.9-day swing holds), and low-win/high-payoff geometry (avg_win $1505 vs avg_loss $615) is internally consistent with the divergence design.

81 trades over 6 years is a statistically usable sample size.

No edge in the base: profit_factor 0.97 (<1), sharpe -0.14, expectancy -$12.87/trade, CAGR -16.7% -- net-losing after fees across a full all-regime 6-year sample.

Negative alpha/information_ratio (-0.0023 / -0.55): loses to simply holding ADA risk-adjusted (benchmark_meaningful=true).

Mechanism mismatch: realized win_rate 28.4% directly contradicts the hypothesis's 55-65% divergence-success premise -- the signal does not behave as claimed on ADA.

Edge is decaying, not latent: positive 2020-21, then bleeds 2023 (-2.8%), 2024 (-6.7%), 2025 (-3.8%). Optimizing 81 net-negative trades would curve-fit noise and predictably fail the downstream deflated-Sharpe/PBO gates.

Do NOT optimize yet — 15 trades is below the walk-forward floor and the edge is currently indistinguishable from zero (PF 1.07, Sharpe CI [-6.8, +4.5]). The divergence detector is very restrictive; densify by loosening ONLY the detection thresholds, keeping the bullish-divergence mechanism/exits/sizing unchanged: 1) LOOSEN PIVOT DETECTION. pivot_strength=3 (needs 3 confirming bars each side) makes confirmed swing lows rare. Drop to 2 so more genuine swing lows qualify, and/or widen div_lookback (40 → 60) so more pivot pairs are comparable. 2) WIDEN THE DIVERGENCE REGION. rsi_low_max=50 and div_confirm_bars=2 are strict; consider rsi_low_max ~45-55 and a slightly larger confirm window so valid divergences aren't missed by a bar. 3) Consider relaxing the 50-SMA gate to a small tolerance (close > 0.97*SMA) so transition-out-of-bear setups near the trend line still qualify. Target ~40-60 trades spread across 2021-2026 so walk-forward and the holdout each hold multiple trades. HARD GATE: this is a borderline near-zero-edge base (PF 1.07, exposure 1%, impact 26% of gross), so the densified re-run must show a CLEAR positive edge (PF > 1.2, non-zero-straddling Sharpe CI, edge not dependent on one trade) to justify optimization — otherwise it is an abandon, not a further iterate.

AdaFourHourBullishRsiDivergenceLong was the portfolio's first momentum-divergence strategy, buying bullish RSI divergences on ADA 4H bars under a 50-SMA trend filter in hopes of catching mean-reverting rallies at regime transitions. It executed exactly as designed — 81 long entries fired with no data or execution drops across a full 6-year all-regime sample — but the signal had no edge: profit factor 0.97, Sharpe -0.14, -$12.87/trade expectancy, and a 28.4% win rate that directly contradicted the hypothesized 55-65% success premise, with losses deepening from 2023 through 2025. At the pre-optimization backtest-review gate the analyst abandoned it, reasoning that optimizing 81 net-negative trades would only curve-fit noise since there was no latent edge and no code defect to fix. It was retired after 2 iterations without ever reaching optimization, risk review, or paper trading.

The literature-cited 55-65% divergence hit rate did not hold on ADA (realized 28.4%), so a plausible academic mechanism must be validated on the specific instrument before assuming an edge exists to optimize.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: the base signal had negative net-of-fee edge with no code defect to iterate on, so optimization and all later stages were never reached.

A long-only ADA 4H strategy that bought bullish RSI(14) divergences (price lower low while RSI higher low) filtered by a 50-SMA trend, betting on mean-reverting rallies at bear-to-bull regime transitions.

Over a 214-day-report span covering ~6 years of ADA 4H bars, it fired 81 clean long entries but produced a 28.4% win rate, profit factor 0.97, Sharpe -0.14, expectancy -$12.87/trade, and CAGR -16.7% with negative alpha and information ratio (-0.55) versus its own ADA buy-and-hold.