MultiSymbolMajorCryptoFourHourDonchianBreakoutPortfolioLong

Multi-Symbol Major Crypto 4H Donchian Breakout Portfolio Long

A long-only multi-instrument portfolio breakout strategy on a basket of 5 major Binance USD-M perpetuals (BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, AVAXUSDT) using 4-HOUR bars. Each symbol is monitored independently for a Donchian channel breakout: 4H close > prior 30-bar high (~5 days lookback) triggers a long entry on that single symbol. Each leg risks 4% of equity (notional ~25% per leg given the 8% stop), with portfolio cap at 3 concurrent positions max (12% aggregate risk). Exits: trailing stop at prior 15-bar low (~2.5 days), 8% hard stop, or 20-bar time stop. NO volume confirmation filter (the 4H volume breakout single-instrument template is already in pipeline for these symbols — this strategy deliberately uses pure-price Donchian to test whether the aggregator structure alone supplies the edge). NO regime/trend gate (200-SMA filters caused the DOGE Bollinger failure by firing at trend changes rather than dips). The aggregator multiplies trade frequency: each symbol produces ~40-60 4H breakouts/year, so the 5-symbol basket generates ~200-300 trades/year — well above the walk-forward statistical threshold and avoiding the rare-event concentration that empirically falsified XRP/ADA/DOGE/LINK daily breakouts at N=4. Per-symbol drawdowns are imperfectly correlated (BTC vs SOL vs AVAX have different regime sensitivities), so equity-curve smoothing comes from intra-basket diversification, not from a single symbol's hot streak. Risk profile: 4% per-trade risk, max 12% portfolio exposure, hard 8% stop per position, no leverage stacking.

Implements the hypothesis as a centralized multi-leg portfolio using the proven AbsoluteMomentumLSBasketDaily pattern: all order management for the 5 legs lives in calculate_signal, which the base fires only after the cross-leg alignment barrier (_sync_extra_iids) confirms every same-4H-timeframe leg has a bar at the primary timestamp, so each pass reads contemporaneous prices and computes each leg's Donchian channels on synchronized data. should_enter/should_exit/position_size are inert because the base entry path manages only the primary instrument. Two-pass per bar: pass 1 increments per-leg held-bar counters and processes exits (trailing Donchian-low, 8% hard stop, 20-bar time stop) to free slots first; pass 2 enters fresh breakouts (close > prior 30-bar high) while respecting the max-3-concurrent cap, skipping legs exited this bar to avoid same-bar churn. Sizing is risk-based (4% risk / 8% stop) but capped at 25% notional per leg so 3 legs keep gross <=75% within the leverage-1.0 futures margin ceiling, matching the stated ~25%/leg and ~12% aggregate-risk figures and clearing the BTCUSDT $100 min-notional by a wide margin (each leg ~25k). A continuous per-bar proxy (primary pct change) is returned so Layer-2 frozen-signal detection passes even though entries are event-gated. min_bars_required = max(donchian windows)+2 guarantees valid channels before any trade; pure OHLCV, no supplementary data, no leverage stacking.

Failed deflated Sharpe: DSR=0.2512 (vs 0.95 bar), expected-max=1.577 over 225 trials — the optimized Sharpe 1.23 is below the best-of-N luck bar, indistinguishable from selection noise (is_significant=false). The decisive forward gates also fail: the HOLDOUT FAILED with a NEGATIVE Sharpe -2.088 (walk-forward OOS feeding it also negative, -0.299), and walk-forward is_overfitted=TRUE with avg IS 1.66 collapsing to avg OOS -0.299 and ALL THREE OOS windows negative ([-0.62, -0.07, -0.21]) — uniformly negative, not propped by a single window — with PBO 0.5463. The clean sensitivity grid (0 cliffs) and positive sharpe_ci_low (0.156) are whole-sample/resampled-selection artifacts (the bootstrap and grid both draw from the bull-dominated 2020-2024 history) while the time-ordered holdout exposes the decay: the edge is concentrated in 2021 (+99%) and bleeds thereafter (2022 -11%, 2026 -12%, rolling Sharpe deeply negative through 2026). The PSR 0.9905 / DSR 0.2512 split is the tell of best-of-225 selection inflation. Not iterate (attempt 1 of 2): with 0 sensitivity cliffs the optimizer is already on its robust plateau, so there is no under-explored region to tune toward; the failure is uniform OOS negativity plus a negative holdout from bull-regime dependence, and a second sweep re-selects the same bull-fit config. Not revise_hypothesis: the Donchian breakout mechanism is already validated and landed on these very majors as single-instrument strategies — the aggregator structure does not add a forward edge, it merely diversifies the same decaying long-breakout exposure, and the hypothesis's own question ('does the aggregator structure ALONE supply the edge?') is answered negatively. FAILURE PATTERN: a multi-symbol pure-price Donchian breakout portfolio on the 5 majors optimizes to a respectable Sharpe (1.23) with a clean robustness surface (0 cliffs, positive CI lower bound, PSR 0.99), yet fails deflated Sharpe (0.25, below a 1.58 luck bar), with a NEGATIVE holdout (-2.09) and ALL three OOS windows negative — the identical bull-concentrated, holdout-failing signature as the single-instrument volume-breakout siblings (DOGE DSR 0.0006/holdout -0.41, AVAX holdout -3.78). Removing the volume/regime filters to 'test the aggregator alone' produces undifferentiated long bull-beta that has no forward edge; intra-basket diversification smooths in-sample variance but does not create alpha, and the time-ordered holdout exposes the regime decay the resampled metrics mask.

Long-only multi-instrument 4H Donchian-breakout portfolio across 5 major Binance USD-M perpetuals (BTC, ETH, SOL, BNB, AVAX). Each symbol independently enters long when its 4H close breaks above the prior 30-bar high (pure price, no volume filter, no trend gate). Each leg exits on a trailing 15-bar Donchian-low break, an 8% hard stop, or a 20-bar time stop. Sizing risks 4% of equity per leg (risk_budget/stop_distance) capped at 25% notional per leg, with at most 3 concurrent positions so gross stays ~75% and aggregate risk ~12%. The 5-symbol aggregator multiplies trade frequency to ~200-300/year and smooths the equity curve via imperfectly-correlated intra-basket diversification. Leverage 1.0.

Strong, distributed sample: 760 trades across 5 symbols and 6 years, tail_ratio 1.38 — the aggregator delivered diversified frequency, avoiding the single-outlier concentration that killed the single-symbol daily breakouts

Well-formed and deployable: no correctness bug, capacity_usd $6.2M with impact only 12.7%, edge spread across multiple bull years (2021/2023/2024)

Sharpe 0.75, PF 1.17, recovery_factor 13.1 — a modest but real headline edge worth stress-testing

Severe recent decay: 2025 -12.9% and 2026 -29.0%, rolling Sharpe down to -10 by early 2026 — the chronological holdout is very likely to fail

Thin/uncertain edge: sharpe_ci_low -0.059 (CI straddles 0), PF 1.17 — deflated Sharpe will be tight

max_DD 40.3% contradicts the hypothesis's intra-basket-diversification smoothing claim (the 5 majors are correlated, benchmark_correlation 0.61)

Deliberate no-regime/no-volume pure-price design is what bleeds in the 2025-2026 choppy/declining regime

Strong, distributed in-sample sample (760 base / 366 optimized trades), clean sensitivity (0 cliffs), positive sharpe_ci_low (0.156), and deployable capacity ($50M, impact 4.5%)

Optimization improved the in-sample risk profile (max_DD 40% -> 15.6%, Sharpe 0.75 -> 1.23)

HOLDOUT FAILED with a NEGATIVE Sharpe -2.088 (walk-forward OOS feeding it also negative, -0.299)

Walk-forward is_overfitted=TRUE with ALL THREE OOS windows negative [-0.62, -0.07, -0.21] — uniform, not propped by one window

Failed deflated Sharpe: DSR 0.2512, is_significant=FALSE; optimized Sharpe 1.23 below the expected-max luck bar 1.577 over 225 trials

PBO 0.5463 (>0.5); bull-concentration and decay confirmed (2021 +99%, 2022 -11%, 2026 -12%, rolling Sharpe deeply negative through 2026)

PSR 0.9905 vs DSR 0.2512 split confirms the headline is best-of-225 selection inflation, not a generalizing edge

MultiSymbolMajorCryptoFourHourDonchianBreakoutPortfolioLong aggregated pure-price 4H Donchian breakouts across five majors to test whether the portfolio structure alone — without volume or regime filters — could supply an edge, and it produced an encouragingly distributed 760-trade, 784% base backtest that earned an 'optimize' verdict. Optimization sharpened the in-sample profile to Sharpe 1.23 with zero sensitivity cliffs and a positive CI low, but every forward gate broke: a negative holdout (-2.088), an overfit walk-forward with all three OOS windows negative, and a deflated Sharpe of 0.25 below a 1.577 luck bar. The PSR 0.99 / DSR 0.25 split confirmed best-of-225 selection inflation over a 2021-concentrated history that decays badly through 2026. On its first iteration the analyst abandoned it, concluding the aggregator merely diversifies the same decaying long-breakout exposure already landed as single-instrument strategies on these majors — the identical holdout-failing signature as the DOGE and AVAX volume-breakout siblings.

Intra-basket diversification smooths in-sample variance but does not create alpha — removing the volume and regime filters to 'test the aggregator alone' produced undifferentiated long bull-beta with no forward edge, answering the hypothesis's own question negatively, and a clean sensitivity grid cannot rescue a breakout portfolio whose time-ordered holdout loses money.

Despite passing sensitivity (0 cliffs) with a positive optimized CI low (0.156), the analyst abandoned it post-optimization because the deflated Sharpe was 0.25 (vs the 0.95 bar) with the optimized Sharpe 1.23 below the 1.577 luck bar, the holdout FAILED with a negative Sharpe of -2.088, and walk-forward was_overfitted=TRUE with all three OOS windows negative ([-0.62, -0.07, -0.21]) and PBO 0.55.

A long-only, OHLCV-only multi-instrument 4H Donchian breakout portfolio across 5 Binance majors (BTC/ETH/SOL/BNB/AVAX) — entering each symbol independently on a 4H close above its prior 30-bar high (no volume or regime filter, to test whether the aggregator structure alone supplies the edge), with per-leg trailing/hard/time stops, 4% per-leg risk, and at most 3 concurrent positions.

The aggregator delivered a strong distributed sample — 760 trades over 6 years for a 784.5% return (CAGR 42.4%), profit factor 1.17, but Sharpe only 0.75 with a CI low of -0.059 and a 40.3% max drawdown, with severe recent decay (2025 -12.9%, 2026 -29.0%). Optimization improved the in-sample profile (Sharpe 1.23, PF 1.32, max DD 15.6% over 366 trades) on a clean sensitivity surface, but the forward windows turned negative.