AbsoluteTrendRegimeRotationQqqTltGldUseq

Absolute-Trend Regime Rotation Across US Index/Defensive ETFs (Hold QQQ in Risk-On; Rotate to TLT+GLD in Risk-Off; SPY 200-Day Regime Gate with Slope Confirmation, Daily Bars, Decades of History, 3-Parameter)

A LONG-HORIZON, multi-instrument REGIME-ROTATION strategy on US equity/defensive ETFs — a mechanism family with essentially zero history in this corpus and nothing to do with the dying crypto families (directional single-name momentum overfit, ratio-spread mean-reversion L96, cross-venue funding carry, naked-strangle VRP). The regime is defined on the broad market (SPY.USEQ) by a long absolute-trend filter; capital rotates between a risk-ON sleeve (QQQ, higher-beta equity) and a risk-OFF sleeve (TLT long-duration Treasuries + GLD gold, the classic uncorrelated defensives). This is NOT cross-sectional ETF ranking (that sibling, SectorEtfCrossSectionalMomentumRotation, is already in the pipeline) — it is an ABSOLUTE trend-regime switch, a structurally different signal. Why it is robust where the crypto clones die: (1) USEQ is COMMISSION-FREE (cost = ~1bp half-spread + impact, ~0.02-0.05% round-trip on these mega-liquid ETFs), so the fee graveyard that killed the funding carries does not apply. (2) SPY/QQQ/TLT/GLD all have DECADES of split-adjusted daily history, so walk-forward windows and the last-20% holdout are densely populated across multiple full bull/bear cycles (2000 dot-com, 2008 GFC, 2020 COVID, 2022 drawdown) — the exact regime diversity these strategies need. (3) Holds last weeks-to-months with only a handful of regime flips per year, so per-trade capture is multiple percent against a sub-0.05% cost — >20x the cost floor, the opposite of the sub-fee microstructure trap. The edge is a well-documented behavioral/structural one (trend persistence + flight-to-quality correlation flip) that commission-free daily execution finally makes harvestable without friction killing it. FEE ARITHMETIC: a typical risk-on QQQ hold spans ~2-6 months and captures on the order of +8-15% in an uptrend leg; a risk-off TLT/GLD hold preserves capital / earns +2-5% when equities fall. Even a conservative average realized per-switch capture of ~2-4% net of the ~0.04% round-trip leaves avg_trade_return_pct comfortably above 1.5%. Low/whipsaw regimes produce a few extra round-trips per year — bounded by the slope-confirmation buffer (hysteresis) so whipsaw churn is minimal.

Implements the hypothesis literally: an ABSOLUTE trend-regime switch (not cross-sectional ranking) defined on the broad market via a 200-day SMA gate with slope confirmation and a hysteresis buffer, rotating one fully-invested sleeve between the risk-on equity proxy (QQQ) and the classic defensives (TLT + GLD). Exactly three searchable parameters (sma_days, slope_days, hysteresis_pct) over decades of split-adjusted daily USEQ history; the remaining knobs (notional_pct, risk_off_tlt_weight) are structural risk policy with declared bounds. Venue is USEQ per the hypothesis: commission-free, long-only, leverage 1.0 (no margin used), whole-share sizing off get_account_equity(). Holds last weeks-to-months so per-switch capture is multiple percent against ~0.02-0.05% round-trip cost. One caveat, implemented as asked: the risk-off sleeve is unconditional (no own-trend gate on TLT/GLD), so a 2022-style joint stock/bond drawdown is only cushioned by the GLD half.

failed deflated Sharpe: DSR=0.0, expected-max=1.53 over 225 trials (32 effective) — the selected Sharpe (0.57) is indistinguishable from best-of-N noise, is_significant=false, and it does not survive programme-level FDR (keeps 4 of 279). Two HARD gates also fail and cannot be waived: PBO 0.85 >> 0.5 (selection is overfit) and holdout only 4 trades < 10. The core defect is structural and not tunable: at ~34 trades over 33 years the strategy is un-validatable — no history slice is both short enough to leave training data and long enough to populate a meaningful vault (~16 expected trades vs 100 needed), so the thin holdout and inflated PBO cannot be fixed by any parameter change without destroying the low-turnover hysteresis premise that defines the idea. This is not a fee problem (per-trade edge 7.85% is 150x the USEQ floor) — it is overfitting + structural slowness. Drawdown 36.7% is also 2x the pre-registered 18% cap. Reframing the same regime-rotation mechanism onto another USEQ ETF basket has already been tried and abandoned 5 times in the failure store (SPY/TLT, QQQ/GLD/TLT, SPY/QQQ/GLD, XLK/XLU, SPY/TLT/GLD macro — all overfit or risk_reject), so this is not a salvageable revise_hypothesis target either.

Long-only absolute-trend regime rotation on US ETFs (USEQ, daily bars). SPY is a signal-only gate: risk-ON requires SPY's close to be more than hysteresis_pct above its sma_days SMA AND that SMA's slope over slope_days to be positive; risk-OFF triggers when the close is more than hysteresis_pct below the SMA; inside the band the prior state is held (hysteresis => a handful of flips per year). Risk-ON holds a single 95%-notional QQQ sleeve; risk-OFF rotates that same sleeve into TLT + GLD (50/50 by default). All three traded legs are managed manually with whole-share sizing; a rotation flattens the outgoing legs on the flip session and enters the incoming legs on a later session so gross notional never exceeds equity at 1x.

Consider (at the analyst/iteration stage, not now) an own-trend filter on TLT or a cash/T-bill fallback when the defensive sleeve is itself in a downtrend. Not required for this pass.

The risk-OFF sleeve buys TLT+GLD unconditionally whenever SPY's absolute trend is down, with no trend/own-momentum gate on the Treasuries leg. In a regime where both equities and long-duration Treasuries fall together (2022), this does not preserve capital — it substitutes one losing book for another (sandbox rate_shock_2022 -41.8%). This faithfully implements the hypothesis as written (the developer flagged the same caveat in the config rationale), so it is not a defect, but it is the strategy's primary structural risk and should be surfaced to the analyst.

The risk-OFF sleeve (TLT+GLD, 50/50) carries NO own-trend gate, so a joint stock/bond drawdown is only cushioned by the GLD half. The sandbox bears this out: the 2022 rate-shock window returned -41.8% (DD 43.1%) and the stressed vol tercile was -6.4% (Sharpe -0.14); overall sandbox max_drawdown 43.3% with Sharpe CI [-0.21, 1.24] straddling zero. The mechanism is sound and commission-free daily execution makes the per-switch capture (~1.9% sandbox avg) far exceed the ~0.05% USEQ cost floor, but the headline edge depends heavily on the 2022-style joint-decline behavior the unhedged defensive sleeve produces. Worth the analyst checking drawdown and TLT-leg attribution on full multi-cycle history before optimizing.

Edge far exceeds venue costs: avg_trade_return_pct = 5.38% of notional vs the ~0.05% USEQ floor (>100x), profit_factor 4.26, positive Sharpe 0.60 with CI low 0.25 (>0) and PSR 0.997 — the opposite of the L75 'USEQ anomaly absent' signature (PF<1.2).

Distinct, under-represented mechanism family: absolute-trend regime rotation (QQQ risk-on / TLT+GLD risk-off) on decades of split-adjusted daily history spanning 2000/2008/2020/2022 cycles — unrelated to the dying crypto momentum/carry/VRP clones.

Trades implement the stated hypothesis: 58 trades, 0 shorts (long-only correct), ~few regime flips/year, hysteresis + slope confirmation working as designed; headline is mostly realized (end_unrealized only ~16% of total return).

Positive alpha (+2.15%) and beta 0.59 vs the benchmark; adequate sample (58 trades, daily-or-slower bars need 30) and data span for meaningful optimization.

max_drawdown 41.8% (2022 rate-shock: -42% window, DD 43.3%) driven by the unhedged risk-off TLT+GLD sleeve during the joint stock/bond decline — well above the equity venue default and a promotion-stage quality concern.

Information ratio ~0 (0.0009) — the active edge over the basket benchmark is thin; stressed vol tercile Sharpe only 0.31.

Risk-off sleeve carries no own-trend gate, so a joint drawdown is cushioned only by GLD — the primary lever optimization must address.

Per-trade edge is genuine and far above the USEQ cost floor: avg_trade_return_pct 7.85% of notional vs ~0.05% round-trip (>150x), profit_factor 4.62, PSR 0.996 — fees were never the failure mode here.

Walk-forward is_overfitted=false with avg OOS Sharpe 0.94 > 0.5, and the sensitivity grid is smooth with zero cliff parameters — the signal is not brittle to parameter perturbation.

Distinct, under-represented mechanism (absolute-trend regime switch on US ETFs) and correctly commission-free venue selection.

TWO HARD gate failures, non-waivable: PBO 0.85 >> 0.5 (parameter selection is more likely than not overfit) and holdout only 4 trades < 10 minimum.

Robustness collapse: deflated_sharpe 0.0, is_significant=false, expected-max Sharpe 1.53 exceeds the selected 0.57, and the candidate does NOT survive programme-level Benjamini-Hochberg FDR (keeps 4 of 279). Selected Sharpe indistinguishable from best-of-N noise over 225 trials (32 effective).

Structurally un-validatable: ~34 trades over 33 years; provenance note states no history slice is both short enough to leave training data and long enough to validate a strategy this slow (~16 vault trades vs 100 needed). Low turnover is intrinsic to the hysteresis premise — no parameter change raises trade count without manufacturing whipsaw.

Max drawdown 36.7% is 2x the hypothesis's own pre-registered 18% cap; rate_shock_2022 window -36.9% / 38.2% DD; stressed-regime Sharpe only 0.12.

Pre-registered prediction missed 2 of 5 (trades 34 vs 60 declared; DD 36.7% vs 18% declared).

The 5 most-similar prior candidates ALL abandoned (SPY/TLT regime rotation -> overfit, QQQ/GLD/TLT dual-momentum -> risk_reject 44% DD, SPY/QQQ/GLD 12-1 -> overfit, XLK/XLU beta-regime -> overfit over 3 iters, SPY/TLT/GLD macro -> overfit). This USEQ ETF regime-rotation family is a repeated graveyard.

Benjamini-Hochberg at q=0.10 over 279 programme candidates keeps 4. A candidate that does not survive here is not distinguishable from the programme's own noise, however good its individual statistics look.

AbsoluteTrendRegimeRotationQqqTltGld-fdcb3098bf

Very low-turnover regime-switch strategies can't be statistically validated no matter how strong the per-trade edge, and this USEQ ETF family had already been abandoned five times.

Analyst abandon after optimization: DSR 0.0 (not significant), PBO 0.85 and 4-trade holdout both breach non-waivable hard gates, drawdown ~2x the pre-registered 18% cap — and structurally un-validatable at ~34 trades over 33 years.

An absolute-trend regime-rotation strategy on commission-free US ETFs that holds QQQ when SPY's 200-day trend and slope are up (risk-on) and rotates to a TLT+GLD defensive sleeve otherwise (risk-off), gated by hysteresis to limit whipsaw.

914% total return (CAGR 11.1%), Sharpe 0.60, PF 4.26, avg per-trade return 5.38% of notional over 58 trades, but 41.8% max drawdown; optimized to 34 trades, Sharpe 0.57, 7.85% per-trade, 36.7% drawdown.