AdaFourHourDownsideMomentumShortDailyBearTrend

ADA 4H Downside-Momentum Short-Only with Daily Bearish-Trend Confirmation on ADAUSDT.BINANCE

A SHORT-ONLY, single-instrument, multi-timeframe defensive crash-protection strategy on ADAUSDT.BINANCE — confirmed testable per the recent successful ADA-based pipeline additions (ADAKeltnerRangeTradingLongShort, ADAVolContractionSqueezeLong, ADALINKCointegrationPairsReversion60Day4H). Structurally distinct from EVERY current portfolio strategy: the promoted ETH+SOL multi-timeframe breakout is long-only; my DOGEMultiTimeframeDonchianBreakoutLong is long-only; all current pairs/cointegration strategies are dollar-neutral; the Keltner range strategy is symmetric. The portfolio has ZERO dedicated SHORT-ONLY DEFENSIVE strategies. This fills a critical role: drawdown protection during regime breaks. The mechanism: when the DAILY trend on ADA is confirmed bearish (close < 50-day SMA AND SMA slope NEGATIVE), 4H breakdowns below the 20-bar Donchian low represent confirmed continuation entries SHORT. The DAILY bearish-trend filter prevents the strategy from fading countertrend bounces during bull markets (where pure short-only is structurally money-losing in trending crypto). DOWNSIDE momentum in crypto is documented to be FASTER than upside momentum (the 'crashes are faster than rallies' asymmetry — Liu-Tsyvinski-Wu 2022): when sentiment breaks, retail panic-selling cascades through alts more aggressively than bull-cycle momentum. ADA specifically has higher retail-flow concentration than majors, making crash regimes more pronounced. The strategy activates ONLY during ~10-20% of the time when the daily trend is genuinely down — providing rare but high-conviction short exposure that diversifies the portfolio's structural long bias. Uses ONLY ADAUSDT.BINANCE 4H + Daily bars — confirmed testable, no cross-asset auto-collection issues. Fills critical gaps: (1) FIRST SHORT-ONLY DEFENSIVE strategy in portfolio (84.8% long_only → reduces toward 55% target by adding short exposure), (2) CRASH-PROTECTION mechanism class (entirely absent from current portfolio), (3) SINGLE-INSTRUMENT scope (avoids multi-instrument audit_stale traps), (4) confirmed-testable instrument. Position sizing: 18% per trade. Risk per trade: 1.5% capped by ATR stop.

Iteration-2 minimal fix for the Layer-3 'No trades produced' failure. Root cause: the bearish-regime gate (self._bear) read from self._daily_close, which is populated only by on_extra_bar from a separate DAILY extra-bar feed; that feed never populated in the sandbox, so _daily_bearish() was permanently False and should_enter (which requires self._bear AND self._breakdown) never returned a side over all 13,771 bars. The Donchian-breakdown half already worked, so I changed ONLY the regime computation: _trend_bearish() now computes the identical 50-day trend filter from the existing 4H close buffer (sma_period*bars_per_day = 300-bar SMA, slope over slope_lookback*bars_per_day = 30 bars), removing the cross-feed dependency. Config drops extra_instruments/extra_bar_types (single feed). min_bars_required and the buffer cap were raised to guarantee the 330-bar regime window is available before trading. Imports, calculate_signal's continuous breakdown-margin signal, entry/exit/sizing logic, and the short-only margin-venue routing (BINANCE futures, leverage 1.0) are unchanged, so Layers 1 and 2 stay green while the strategy now produces short entries whenever ADA is in a confirmed 4H-derived downtrend and breaks the 20-bar low.

Clean-looking base that fails every generalization gate under 3-phase validation — the base Sharpe 0.79 was a best-of-N point estimate that does not deflate. deflated_sharpe is 0.0018 (vs 0.95 bar) with the Sharpe BELOW the expected-max luck bar 2.045 over 225 trials (is_significant=false, sharpe_ci_low -1.73 straddles 0, PBO 0.5635). Walk-forward is_overfitted=true: avg IS 2.76 collapses to avg OOS -1.25 with OOS windows [2.85, -6.98, 0.38] (one catastrophically negative), and the holdout 'pass' is degenerate (ratio 0 against a NEGATIVE walk-forward-OOS denominator of -1.25; holdout_sharpe only +0.35). Decisively, OPTIMIZATION MADE IT WORSE: the selected config posts a NEGATIVE Sharpe -0.118 (total_return -1.70%, PF 0.96, expectancy -$25/trade, max_consecutive_losses 16) versus the base +18.4%/PF 1.31 — when tuning produces a worse full-period result, the walk-forward selection is fitting in-sample noise. Sensitivity also FAILED with 2 cliffs on the core regime params (sma_period, bars_per_day). The optimized annual returns expose the cause: the short-only edge earns 2022 +15.3% (the bear) but bleeds 2024 -8.1%, 2025 -12.6%, 2026 -1.3% (rolling Sharpe ~-9 in 2025) — it only works in a sustained bear and loses through the 2023-2025 recovery, exactly the 'pure short-only is structurally money-losing in trending crypto' risk the hypothesis itself named. Not iterate (attempt 1 of 2): the optimizer already drove it negative, OOS collapses to -1.25 with a -6.98 window, DSR is 0.0018, and the edge is structurally concentrated in one bear year — a second sweep re-finds another 2022-fit config that fails the recovery. Not revise_hypothesis: short-only momentum on a crypto alt is structurally negative outside sustained bears, the edge has no promoted sibling, and no different target rescues a mechanism that profits ~15% of the time and bleeds the rest. FAILURE PATTERN: single-asset short-only 'crash-protection' momentum on a crypto alt produces a clean-looking, properly-sized base (Sharpe 0.79, PF 1.31) whose return is concentrated in the 2022 bear; after best-of-225 selection it fails deflated Sharpe (0.0018, below the 2.045 luck bar), is_overfitted=true with avg OOS -1.25 and a -6.98 window, and the optimizer drives it to a NEGATIVE Sharpe because short-only bleeds through every non-bear regime. A diversifier that passes the pre-optimization gate on negative correlation but concentrates its edge in a single bear year does not survive multiple-testing deflation or walk-forward.

Short-only single-instrument crash-protection strategy on ADAUSDT.BINANCE USD-M futures, 4-HOUR entry bars gated by a bearish 50-day trend filter that is now derived from the SAME 4H buffer (50 days = 300-bar SMA, 5-day slope = 30 bars) instead of a separate daily feed. Enters SHORT on a 20-bar 4H Donchian breakdown only when the long-term trend is bearish (close < 300-bar SMA and SMA falling). Exits on a 2x ATR adverse stop, regime flip, structure reclaim above the prior 10-bar high, or a 30-bar time stop. Sized at 18% equity notional, capped so ATR-stop risk <= 1.5% of equity. leverage 1.0.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: ADAUSDT.BINANCE-4-HOUR-LAST-EXTERNAL, Bars processed: 13792 Diagnostics: should_enter() returned a side 0 times over 13771 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Properly sized (avg_position_pct 17.8%, respects the 0.18 cap with risk-based reduction) — controlled 11.1% vol and 6.7% max drawdown, unlike the reckless-0.95 strategies

Positive net-of-fee edge: profit_factor 1.31, Sharpe 0.79, expectancy +$152/trade (~0.84%), positive skew +3.73, avg_win/avg_loss 1.74x

Correctly short-only as specified (121 shorts, 0 longs); iter-2 single-feed trend filter fires cleanly (no zero-trades)

Genuine diversifier / crash protection: negative beta (-0.02) and benchmark_correlation (-0.35), with best years in down regimes (2022 +12.8%, 2025 +9.8%) — fills a real portfolio gap (first short-only defensive)

Edge spread across multiple years, healthy sample (121 trades), decent capacity ($27.5M)

Modest standalone return (CAGR 2.96%) — value is the negative correlation, not absolute return

Slight bleed in up years (2024 -0.96%, 2026 -3.9%) — the bear filter is imperfect; regime-gate robustness needs confirmation in walk-forward

Short-only in trending crypto is structurally hard; durability of the bear-filter edge is the key open question for holdout

Correctly implemented and properly sized (avg_position_pct ~18-22%, respects cap); genuine short-only diversifier (negative beta/benchmark_correlation) — earned the optimization spend

Base config trades as specified (121 shorts, 0 longs) and profits in the 2022 bear, confirming the mechanism does what it claims in its target regime

Deflated Sharpe 0.0018 (vs 0.95 bar), is_significant=false: the Sharpe is below the expected-max luck bar (2.045) over 225 trials — best-of-N noise; sharpe_ci_low -1.73 straddles 0, PBO 0.5635

Walk-forward is_overfitted=true: avg IS 2.76 collapses to avg OOS -1.25 with OOS windows [2.85, -6.98, 0.38] (one catastrophically negative)

Holdout 'passed=true' is a degenerate artifact: ratio 0 against a NEGATIVE walk-forward-OOS denominator (-1.25); weakly-positive holdout_sharpe 0.35 does not establish robustness

Optimization made it WORSE: selected config posts Sharpe -0.118, total_return -1.70%, PF 0.96, expectancy -$25/trade, max_consecutive_losses 16 — strictly below the base

Sensitivity FAILED with 2 cliffs on the core regime params (sma_period, bars_per_day)

Single-bear-year concentration: optimized config earns 2022 +15.3% but bleeds 2024 -8.1%, 2025 -12.6%, 2026 -1.3% (rolling Sharpe ~-9 in 2025) — short-only loses through the recovery, the premise's own acknowledged structural risk

This strategy filled a real portfolio gap — the first short-only crash-protection diversifier — shorting ADA Donchian breakdowns only in a confirmed bearish 4H-derived trend, and its base run was clean, well-sized, and genuinely anti-correlated (Sharpe 0.79, PF 1.31, +18.4%, 6.7% drawdown), earning an 'optimize' verdict. But 3-phase validation exposed the edge as a 2022-bear artifact: deflated Sharpe 0.0018, walk-forward overfitted with OOS collapsing to -1.25 (one -6.98 window), two sensitivity cliffs, and an optimized config that turned negative (-0.118 Sharpe). The analyst abandoned it at the analyzing stage on its second iteration, concluding short-only momentum is structurally money-losing outside sustained bears — concentrating its edge in one year and bleeding through 2023-2025 — with no robust region to tune and no sibling to redirect to, so it stopped before risk review.

A clean, negative-correlation diversifier that passes the pre-optimization gate can still fail under deflation when its edge is concentrated in a single bear year — short-only momentum on a crypto alt earns in sustained bears (2022) but bleeds through every recovery (its own acknowledged structural risk), and when optimization makes the full-period result worse, the walk-forward is fitting in-sample noise.

It cleared the backtest-review gate with an 'optimize' verdict, but the post-optimization analyst abandoned it: deflated Sharpe 0.0018 (<0.95) below the 2.045 expected-max over 225 trials, is_significant false, a CI straddling zero, PBO 0.56, walk-forward overfitted (avg IS 2.76 → avg OOS -1.25 with a -6.98 window), a degenerate holdout pass, sensitivity failing with 2 cliffs on the core regime params — and decisively, optimization drove the full-period result negative.

A short-only, single-instrument crash-protection strategy on ADAUSDT 4H Binance futures (1.0x leverage, 0.18 position cap with ATR-risk reduction) — shorting 20-bar Donchian breakdowns only when a 4H-derived 50-day trend filter was confirmed bearish (price below a falling SMA) — explicitly built as the portfolio's first short-only defensive diversifier against regime breaks.

The base backtest looked clean and well-sized — 121 shorts (0 longs), +18.4% total, Sharpe 0.79, profit factor 1.31, expectancy +$152/trade, just 6.7% max drawdown — and was a genuine diversifier (negative beta -0.02, benchmark correlation -0.35, best years in down regimes 2022 +12.8% / 2025 +9.8%); but the optimized config was strictly worse, posting Sharpe -0.118, -1.70%, PF 0.96.