UseqWeeklyCrossSectionalReversalTwentyLargeCapDollarNeutralLS

US Large-Cap Short-Term Reversal, Dollar-Neutral Cross-Sectional Long/Short (Buy Past-Week Losers, Short Past-Week Winners Across ~20 Liquid S&P Names, Weekly Rebalance on Daily Bars, Commission-Free USEQ, 3-Parameter)

A cross-sectional, dollar-neutral equity long/short that monetizes the classic short-term reversal anomaly: over a ~1-week horizon, names that fell most tend to bounce and names that rose most give back. Deliberately a different family from the crypto graveyard and from the SOL funding-carry it replaces — not directional, not crypto mean-reversion, not a cointegration pair, not a funding-carry. It clears fees where crypto reversion cannot ONLY because USEQ is commission-free (~0.02-0.05% RT), making a high-turnover weekly reversion book viable, with decades of daily history for robust walk-forward.

Implements the hypothesis exactly: buy past-week losers / short past-week winners across ~20 liquid S&P names, weekly rebalance on daily bars, dollar-neutral, commission-free USEQ, 3 searched parameters (formation_days, top_k, gross_exposure) with clamps declared in _param_bounds. Cross-sectional demeaning (rather than an absolute return threshold) keeps signal frequency scale-invariant across vol regimes, and equal name counts + equal notional per leg makes the book dollar-neutral by construction so it can never carry the net directional exposure that has blown up past neutral books. leverage=2.0 is the USEQ Reg-T cap and is genuinely consumed in sizing (gross = equity * gross_exposure * self.config.leverage), giving 1.2x gross at defaults. Rebalance timing is anchored to the CALENDAR week ((days_since_epoch + 3) // 7, Monday-anchored) computed from the bar's own timestamp, never a process-relative bar counter, so the rotation schedule is identical across restarts, backtests and paper replays. Universe members are all listed since the 1990s so the multi-leg alignment barrier is satisfied from the first bar of the window; ~6-session warmup means the sandbox window produces many weekly rotations. New class name, distinct from the previously submitted UsLargeCapShortTermReversalWeeklyLS which belongs to another hypothesis.

negative_expectancy: net profit factor is 1.014 (break-even) over a full 36-year, 11.9k-trade sample and impact consumes 67.2% of gross PnL, leaving capacity of only ~$222k. The gross short-term-reversal signal is too thin to survive spread+impact, and it has decayed structurally — five of the last six calendar years are deeply negative (2022 -22.9%, 2023 -21.0%, 2024 -11.7%, 2026 -37.1%). Optimizing three parameters (formation_days/top_k/gross_exposure) cannot overcome a 67% cost drag on a break-even anomaly; this is the known-dead USEQ commission-free reversion class (PF<1.2 => anomaly absent, per prior outcomes) and the 0/84 dollar-neutral multi-instrument L/S class. Not worth 2 hours of optimization.

Dollar-neutral cross-sectional short-term reversal across 20 continuously-listed liquid S&P 500 large caps (AAPL primary + 19 extras) on USEQ 1-DAY session bars. Every session it computes each name's trailing formation_days (~1 week) return from a single synchronized close buffer and scores it as the NEGATED cross-sectional z-score of that return (positive = relative loser, negative = relative winner). On the first session of each calendar week it longs the top_k highest scores (past-week losers) and shorts the top_k lowest (past-week winners), equal-dollar weighted, gross = equity * gross_exposure * leverage split into two identical-notional legs. Rebalancing trades whole-share netting deltas so unchanged names hold, keeping turnover to a few round trips per week.

Consider matching each leg's most recent bar with ts <= primary ts (nearest-at-or-before) rather than exact ts equality, so a same-week rotation is not skipped when one leg's session bar lands in a different delivery batch.

_aligned_row requires every one of the 20 legs to have its latest buffered bar stamped at exactly the primary bar's ts; when any same-session bar has not yet been delivered the method returns None and calculate_signal falls back to the inert _proxy (no book update / no rebalance for that session). Empirically alignment succeeds often enough to produce 2049 trades, but on any session where the venue delivers legs across the timestamp boundary unevenly the weekly rotation could silently skip. Not a correctness defect (fails safe to no-trade), but worth confirming the weekly rebalance actually fires on the intended first-session-of-week across the full history rather than slipping a session.

Sandbox smoke test is net-losing: total_return -35.66%, profit_factor 0.933 (<1, gross-negative territory), avg_trade_return_pct -0.0478% over 2049 trades. On zero-commission USEQ the entire drag is spread+impact, and a per-trade edge that sits BELOW the ~0.05% round-trip spread floor is exactly the failure mode for this family. This is the dollar-neutral cross-sectional large-cap basket family whose sibling UsLargeCapShortTermReversalDollarNeutralLS was abandoned at PF~1.03 with impact 64.5% of gross and capacity ~$240k, and the broader multi-instrument dollar-neutral L/S class shows 0/84 survivors. Regime mix is inverted from the thesis (loses in calm -19% and normal -53%, only profitable in the stressed tercile +35.8%), so the reversal 'edge' here is a stressed-vol tail bet, not a diversified market-neutral premium. Analyst should check on full history: (1) is the gross signal positive before impact, and (2) impact_cost_pct / capacity at deployable size — the sandbox left impact_cost_pct null but total_impact_usd $8.1k on a small book already erodes the thin per-trade return. One unoptimized parameter draw, so not blocking, but the structural odds for this family are poor.

Trades implement the hypothesis: ~balanced long (5953) vs short (5960) legs, dollar-neutral, weekly rotation, per-trade return 0.179% > the USEQ ~0.05% spread floor.

Deep, clean data (36 years, 20 continuously-listed large caps); the multi-leg alignment barrier is satisfied from the start so the sample is real (metrics_reliable=true, 11,913 trades).

Net profit factor is 1.014 — essentially break-even. Over 36 years the net edge is indistinguishable from zero once costs are folded in (impact_folded=true).

impact_cost_pct = 67.2% ($550k of gross): spread+impact consume two-thirds of the gross edge. capacity_usd is only ~$222k — the residual edge exists only at toy scale.

Max drawdown 76.99% with a 2,115-day (5.8yr) recovery; recovery_factor carried entirely by the 1990s-2000s.

The anomaly has decayed: annual returns 2021 -17.5%, 2022 -22.9%, 2023 -21.0%, 2024 -11.7%, 2026 -37.1% (five of the last six years negative). This is a dead-in-recent-regime short-term-reversal book, not a live edge.

Matches two known dead classes: USEQ commission-free reversion with PF<1.2 (anomaly absent, not fee-eaten) and dollar-neutral multi-instrument L/S (0/84 survivors).

UseqWeeklyCrossSectionalReversalTwen-3b899c0733

This strategy tried to monetize the classic short-term reversal anomaly with a dollar-neutral weekly long/short across 20 large-cap US names, on the thesis that commission-free USEQ trading would let a high-turnover book clear costs that would sink a crypto version. The 36-year, 11,913-trade backtest was clean and genuinely dollar-neutral, but produced only a 1.014 profit factor and 0.67 Sharpe against a 77% max drawdown, with impact costs devouring 67% of gross PnL and capacity capped near $222k. The analyst abandoned it at the pre-optimization backtest review, judging the edge break-even and structurally decayed — five of the last six years were deeply negative — and matching two known-dead classes (USEQ commission-free reversion with PF<1.2 and 0/84 dollar-neutral multi-instrument L/S). Optimization and all later stages were never reached; three parameters could not overcome a 67% cost drag on a break-even signal.

Commission-free venue access does not rescue a high-turnover reversion edge when spread and market impact still eat two-thirds of gross PnL — a break-even profit factor over a clean 36-year sample signals an absent, not a fee-eaten, anomaly.

The analyst issued an 'abandon' verdict at the pre-optimization backtest review for negative/break-even expectancy: the reversal signal was too thin to survive spread and impact (67% cost drag), capacity was tiny, and the anomaly had decayed with five of the last six calendar years deeply negative (2022 -22.9%, 2023 -21.0%, 2024 -11.7%, 2026 -37.1%).

A dollar-neutral, cross-sectional short-term reversal book that buys the past-week relative losers and shorts the past-week relative winners across 20 liquid S&P large caps, rebalanced weekly on USEQ daily bars, betting that commission-free US equities make a high-turnover reversion strategy viable where crypto fees would not.

Over a 36-year backtest (11,913 trades, win rate 52.8%) it returned 244.8% total with a Sharpe of 0.67, but net profit factor was just 1.014 and per-trade return 0.179%, alongside a 76.99% max drawdown. Impact costs consumed 67.2% of gross PnL and capacity was only ~$222k, leaving expectancy essentially at break-even (-22.98 per trade).