MidCapMomentumRotationAbsGateLong

Mid-Cap Cross-Sectional Momentum Rotation, Long-Only with Absolute-Momentum Cash Gate (BINANCE USD-M, 10 Deep-History LIQUID MID-CAPS — NOT the Deployed Majors — Weekly Hold Top-3 by Relative Strength, Skip to Cash Any Slot Whose Own Momentum Is Negative; Mid-Cap Cross-Sectional Dispersion Is FAR Larger Than Majors, Giving Real Fee Headroom, 2-Parameter)

A REAL, codeable strategy that reuses the strongest LIVE, non-decayed, non-options deployed winner — cross-sectional relative-strength momentum rotation (majors version Sharpe 2.6) — but on a genuinely DIFFERENT, materially-larger-edge OPPORTUNITY SET: deep-history liquid MID-CAP USD-M perps rather than the 8 majors. This is more than an L24 re-weight: mid-cap crypto exhibits substantially STRONGER cross-sectional momentum and much larger return DISPERSION than the majors (winners and losers separate by tens of percent per week vs single digits for majors), so the relative-strength premium per rotation is far larger relative to the ~0.10% USD-M round-trip — the fee headroom the majors version only marginally has. To control the higher mid-cap drawdown/blowup risk that killed naive long-only baskets, it adds an ABSOLUTE-momentum cash gate (a held slot goes to CASH if that name's own trailing momentum is negative), so in broad mid-cap bear phases the book de-risks toward cash rather than holding falling knives, and conservative per-name caps prevent single-name blowups. Universe = 10 liquid, multi-year-history mid-caps (LTC, ETC, ATOM, ALGO, FIL, EOS, XLM, AAVE, SAND, MANA) — deep enough for a valid backtest, liquid enough to trade. This diversifies the portfolio's heavy majors concentration (single/majors dominate) into a distinct symbol set. NOT the deployed majors rotation (different universe with a documented stronger effect), NOT funding/order-flow/options/reversion. Long-only, diversified top-3, weekly, USD-M. 2 core parameters (momentum lookback, number held).

Implements the hypothesis on a distinct, larger-dispersion opportunity set: it reuses the deployed cross-sectional relative-strength momentum-rotation engine (majors version Sharpe 2.6) but swaps the universe to 10 liquid mid-cap USD-M perps, where cross-sectional momentum is stronger and return dispersion is much larger, so each rotation's relative-strength premium is bigger relative to fees. The distinct new control vs the deployed basket is the ABSOLUTE-momentum cash gate in _rebalance: rank by relative strength, take top_k, but only fill a slot if that name's own momentum is positive; negative-momentum names are left in cash, so in broad mid-cap downturns the book de-risks instead of holding losers. UNIVERSE INTEGRITY (critical): the hypothesis listed EOSUSDT, which is delisted (~2025-05) and would return zero/stale bars — and in a multi-instrument basket ONE delisted leg abandons the whole backtest as data_unavailable — so EOS is replaced with NEARUSDT (a liquid, multi-year-history mid-cap) to preserve a valid 10-name set; the other nine are actively-listed deep-history USD-M perps. Purely price-driven (no supplementary or tick data) so it avoids the cross-instrument-supp and tick-timeout pitfalls and is inherently timeout-safe (bounded per-bar work: 10 O(1) momentum reads; rebalance only weekly). All trading across the 10 legs (including the primary) happens in the rebalance routine via _submit_entry_instrument/close_position; the default primary entry path is neutralized (should_enter None, should_exit False, position_size 0) so the base never double-trades the primary. VENUE: long-only/no-leverage nominally suggests spot, but USD-M is chosen deliberately — it matches the deployed momentum winner and USD-M taker fees (0.05%) are HALF spot's (0.10%), and several of these mid-caps have deeper/again-only perp liquidity; leverage stays 1.0. Risk is bounded (long-only, <=3 names x 0.20 = 0.60 gross equity, no leverage, no compounding blowup). 2 core parameters (momentum_lookback, top_k) shared across all names — not per-name tuned — to resist overfit.

Net-losing momentum basket: profit_factor 0.926 (<1.0), total_return -15.2% over ~6.5 years, expectancy -$204/trade — the strategy loses money net, disqualifying regardless of anything else (L8). On top of that, max_drawdown is 73.73% (CI high 96%), a hard abandon (>50% floor, L19), and the absolute-momentum cash gate demonstrably failed to contain it — recent years are catastrophic (2025 -47.8%, 2026 -41.5%). The hypothesis's central claim, that mid-caps offer 'far larger fee headroom' via bigger cross-sectional dispersion, is falsified: the larger dispersion cuts both ways, mid-cap losers crater harder, and the momentum premium net of the deeper drawdowns is negative. Sharpe 0.144 (CI [-0.45, 0.72] straddling zero), alpha -0.006, and information_ratio -0.27 vs its equal-weight-basket benchmark confirm no edge. This is worse than the majors momentum variants reviewed this session (which at least had PF>1 and positive returns) — it is a PF<0.93, negative-return, 74%-DD loser on a structurally more crash-prone universe. Optimizing momentum_lookback/top_k cannot flip a losing edge; it would only overfit the 2021/2024 bull spikes and fail every gate. Not worth 2 hours of optimization.

Long-only weekly cross-sectional momentum rotation over 10 deep-history liquid MID-CAP BINANCE USD-M perps (LTC/ETC/ATOM/ALGO/FIL/NEAR/XLM/AAVE/SAND/MANA, 1-DAY bars). Weekly it ranks the names by trailing relative strength (return over momentum_lookback days) and holds an equal-weight top-3, with an ABSOLUTE-momentum cash gate: only names with POSITIVE own trailing momentum are eligible, so any slot whose name is falling stays in cash (de-risk toward cash in mid-cap bear phases rather than holding falling knives). Conservative per-name caps (0.20 equity each) limit single-name blowups. Mid-caps are chosen for their far larger cross-sectional dispersion vs majors, giving more fee headroom over the ~0.10% round trip. Pure-price, no supp data, no ticks. 2 core parameters: momentum_lookback, top_k.

No code change required. Analyst: note 1 of 10 legs deviates and the dispersion-headroom argument rests on the other 9.

Universe substitution vs the hypothesis text: it names EOSUSDT but the config uses NEARUSDT. Documented and defensible (EOS USD-M perp delisted ~2025-05; one delisted leg would abort the whole multi-instrument backtest as data_unavailable). Not a whole-strategy mislink — 9 of 10 legs match exactly and the mechanism/universe class is preserved. Caveat: NEAR is closer to a major than a mid-cap, slightly diluting the 'mid-cap dispersion > majors' thesis.

Analyst: strong abandon candidate on negative expectancy + ~47% drawdown. Do not iterate the code — the implementation is correct.

PERFORMANCE FLAG FOR BACKTEST_REVIEW (not a QA code defect). Sandbox decisively negative: total_return -42.1%, max_drawdown 46.8% (near the >50% hard-abandon band), profit_factor 0.21, win_rate 0.19, Sharpe -1.50, avg_trade_return_pct -9.72%. These are large-move directional losses, not a fee/measurability artifact (~10% per-trade moves >> 0.10% round-trip). Exactly the naive-long-only mid-cap blowup risk the hypothesis warned of; the absolute-momentum cash gate did not prevent it (kept buying positive-trailing-momentum names that then reversed). An EDGE failure for the analyst to abandon, not a code bug to re-code.

If alignment isn't guaranteed by the base, gate the snapshot on all legs sharing ts_event before computing momenta.

Leg-alignment assumption: calculate_signal snapshots each leg's latest close assuming an alignment barrier buffered all 10 same-tf legs for that timestamp. If the base does not strictly guarantee same-ts_event alignment, a non-primary leg's _latest_bar could be one bar stale, introducing a consistent 1-bar momentum lag. Would not flip polarity or break trading.

Diversifies the portfolio's majors concentration into a distinct mid-cap symbol set; pure-price and timeout-safe

Well-sampled (193 trades) so the negative verdict is decisive

Net loser: profit_factor 0.926 (<1.0), total_return -15.2%, expectancy -$204/trade over ~6.5 years

max_drawdown 73.73% (CI high 96%) — far past the 50% hard-abandon floor; the absolute-momentum cash gate did not contain it

Deeply negative recent years: 2025 -47.8%, 2026 -41.5% — the edge lived only in 2021/2024 bull spikes

The 'mid-caps = larger fee headroom' thesis is falsified: bigger dispersion cuts both ways and losers crater harder, so momentum net of the deeper drawdowns is negative

Sharpe 0.144 (CI straddling zero), alpha -0.006, information_ratio -0.27 vs its benchmark

MidCapMomentumRotationAbsGateLong took the deployed majors momentum rotation to a distinct 10-name mid-cap universe on the thesis that stronger dispersion would give real headroom over the ~0.10% round-trip, adding an absolute-momentum cash gate to de-risk in bear phases. Instead the wider dispersion cut both ways: over ~6.5 years and 193 trades the book was a net loser (profit factor 0.926, total return -15.2%, expectancy -$204/trade) with a 73.73% max drawdown and catastrophic recent years (-47.8% in 2025, -41.5% in 2026). Sharpe of 0.144 straddling zero and an information ratio of -0.27 confirmed no edge over the benchmark, and the cash gate demonstrably failed to contain the tail. The analyst abandoned it at backtest-review as worse than the majors momentum variants — a PF<1, negative-return, 74%-drawdown loser whose central fee-headroom claim was falsified and that optimization could only overfit to the 2021/2024 bull spikes.

Larger cross-sectional dispersion is not free edge — it amplifies losers as much as winners, so moving a long-only momentum rotation to a more volatile mid-cap universe deepens drawdowns and can turn a marginally-positive majors edge into a net-losing one; an own-momentum cash gate does not rescue a structurally more crash-prone book.

It was abandoned at backtest-review (verdict: abandon) as a net-losing basket (PF<1.0, negative return) that also breached the 50% hard-drawdown floor; the absolute-momentum cash gate failed to contain the tail, and the core 'mid-caps offer more fee headroom' thesis was falsified because larger dispersion cuts both ways and mid-cap losers crater harder.

It applied the deployed cross-sectional relative-strength momentum rotation to a genuinely different universe — 10 deep-history liquid mid-cap USD-M perps instead of the 8 majors — betting that mid-caps' far larger return dispersion gives more fee headroom, and added an absolute-momentum cash gate so any slot whose own trailing momentum was negative went to cash rather than holding falling knives.

Over ~6.5 years (2,388 days) it made 193 trades but was a net loser: profit factor 0.926, total return -15.2%, expectancy -$204/trade, Sharpe 0.144 (CI straddling zero), win rate 34%, information ratio -0.27 versus its equal-weight benchmark, and a 73.73% max drawdown (CI high 96%) with recent years deeply negative (2025 -47.8%, 2026 -41.5%).