UsLargeCapShortTermReversalWeeklyLS

US Large-Cap Cross-Sectional Short-Term Reversal, Dollar-Neutral Long/Short (USEQ, Daily Bars, 5-Day Lookback / 5-Day Hold, Weekly Rebalance, 3-Parameter) — the Commission-Free Fix for the Anomaly Crypto Fees Killed

A DOLLAR-NEUTRAL, CROSS-SECTIONAL LONG-SHORT short-term reversal strategy over a fixed basket of ~16 highly liquid S&P 500 large caps on the USEQ venue, daily bars, weekly (5-trading-day) rebalance. Every rebalance day it ranks each name by its trailing 5-day simple return, goes LONG the recent relative LOSERS (bottom ~30%) and SHORT the recent relative WINNERS (top ~30%), equal-weighted within each leg and sized so long notional == short notional (zero net beta). This is NOT a crypto momentum reformulation and NOT a cointegration pair (L95/L96): it is a decades-documented cross-sectional overreaction anomaly (Jegadeesh 1990, Lehmann 1990) whose ONLY robust historical killer is transaction costs. The factory already ran the identical mechanism on Binance USD-M ('Cross-Sectional 2-Day Reversal, Majors-Only') and it died — crypto's ~0.10-0.20% round-trip ate the ~0.4%/week reversal spread. The USEQ venue removes exactly that killer: zero commissions, ~0.02-0.05% round-trip spread+impact on mega-caps, and DECADES of split-adjusted daily history for a low-overfit test. It simultaneously fills three under-represented buckets: USEQ venue (1.3% vs decades of usable data), long/short direction (14.7% vs >=45% target headroom), and a mechanism family (short-horizon reversion) that is fee-dead everywhere in crypto but structurally alive here.

Implements the hypothesis exactly: Jegadeesh(1990)/Lehmann(1990) short-horizon cross-sectional overreaction, 5-day formation / 5-day hold, weekly rebalance, dollar-neutral long-loser / short-winner over ~16 large caps, on the venue whose cost structure is the whole point (USEQ: zero commission, ~0.02-0.05% round-trip spread+impact, vs the ~0.10-0.20% that killed the identical mechanism on Binance USD-M). Exactly three tunables are exposed - lookback_days, top_frac, gross_exposure - each clamped in code with the identical range declared in _param_bounds, keeping the overfitting surface minimal against decades of daily history. Venue is USEQ MARGIN with leverage 2.0 because the strategy is genuinely long+short and the sizing consumes the multiplier directly (equity * gross_exposure * self.config.leverage). Dollar-neutrality is enforced at the notional level: both legs get identical per-name dollar targets and identical name counts (k = round(top_frac * 16), capped at half the basket so the legs can never overlap), which is the specific defect behind past market-neutral -100% blowups. The rebalance schedule is anchored to a CALENDAR week index derived from the bar's own timestamp (Monday-anchored days-since-epoch arithmetic), never a process-relative bar counter, so a Monday holiday just moves the rebalance to the week's first real session and the schedule is identical across restarts, backtests and paper replay. The basket is 16 currently-listed mega-caps whose 1-DAY bars were confirmed present in the catalog (all back to 2012 or earlier), spread across six sectors so the cross-section is a relative-value bet rather than a hidden sector trade. calculate_signal returns the continuous winner-minus-loser return spread every session (natural return units, ~0.03-0.10 typically) rather than a gated constant; the book is managed there because a 16-leg basket cannot be expressed through the single-instrument should_enter path, which is left inert so the base class never submits a competing order on the primary leg. There is no entry gate on the spread - the hypothesis calls for an unconditional weekly re-rank - so trade frequency is a deterministic ~52 rebalances/year, well clear of the zero-trade sandbox failure mode. Sizing is whole shares (USEQ size precision 0) and warm-up is only 7 bars, a negligible fraction of the sandbox window.

coding_stuck_3_timeouts

Dollar-neutral cross-sectional short-term reversal over 16 liquid US mega-caps (AAPL, MSFT, GOOGL, AMZN, NVDA, META, JPM, BAC, XOM, CVX, JNJ, UNH, PG, KO, WMT, HD) on USEQ 1-DAY session bars. Each session every name's trailing 5-day simple return is measured from a single synchronized price panel; on the first session of each calendar week the cross-section is ranked and the book moves to LONG the bottom 30% (recent relative losers) and SHORT the top 30% (recent relative winners), equal dollar weight within each leg so long notional equals short notional (~zero net beta). Positions are held one week and re-ranked. Rebalancing is executed as netting delta orders in whole shares with a $300 micro-delta filter, so names that stay in the same bucket keep their position and only genuine turnover pays the spread. Gross notional = equity * gross_exposure * config.leverage (0.75 * 2.0 = 1.5x at defaults, inside the USEQ Reg-T 2.0 cap).

Consider triggering the aligned-row build/rebalance from on_extra_bar as well (or from whichever leg completes the panel for a ts) so alignment does not depend on the primary leg being dispatched last within a timestamp.

calculate_signal (and therefore _latest_aligned_row / _rebalance) is only invoked on PRIMARY (AAPL) bars; extra legs route to the default no-op on_extra_bar. The aligned-row logic requires every leg's buf[-1] to carry the SAME ts as the primary bar. Whether this succeeds depends on the engine's within-timestamp delivery order (if AAPL is always dispatched before its 15 peers update, alignment would lag or fail). Empirically it works here (2238 trades over 1649 bars), so this is not blocking, but the correctness of the panel synchronization rests on an engine-ordering assumption rather than an explicit guarantee.

Acceptable at current sizing; if equity/per_name shrinks in optimization, monitor realized net notional to confirm dollar-neutrality holds.

Whole-share truncation (float(int(targets[i]/price))) plus the $300 micro-delta filter means the two legs are only approximately dollar-neutral per name; residual net beta accumulates across 16 names. At defaults (per_name ~$15k) the truncation error is small relative to notional, so net exposure stays near zero, but on lower-priced names or smaller equity the neutrality can drift.

Sandbox is deeply negative (total_return -48.8%, profit_factor 0.90, avg_trade_return_pct -0.075%) and this is the dollar-neutral cross-sectional basket family (L45, 0/84 survivors). A near-identical sibling, UsLargeCapShortTermReversalDollarNeutralLS, was abandoned 2026-09-07 at PF~1.02 with impact_cost_pct 64.5% and capacity_usd only $240k — i.e. the reversal spread on this mega-cap universe is largely consumed by impact even on a zero-commission venue. total_impact_usd 6335 and turnover 80.5 here point the same way. The short leg is the likely drag: shorting recent relative winners in a decades-long up-drifting mega-cap cross-section carries structural cost. Analyst should size deployable notional vs capacity and check whether the WML spread survives spread+impact before optimizing — a PF-1.0 signal 60%+ eaten by impact is not liftable by tuning 3 params.

Backtest gate: - metrics unreliable: Incoherent: Sharpe 0.27 disagrees in sign with total_return -33.21% — the Sharpe is computed from the mtm_equity_curve series while the return comes from the equity curve. Judge on the equity curve.

UsLargeCapShortTermReversalWeeklyLS-cc828d113f

This strategy revisited the short-horizon cross-sectional reversal anomaly (Jegadeesh/Lehmann) that had died on Binance to crypto fees, moving it to the commission-free USEQ venue as a dollar-neutral weekly long/short over 16 S&P 500 mega-caps. It never completed the pipeline: across 2 iterations the developer agent's coding stage timed out three times and the run was abandoned as coding_stuck, so it reached neither a backtest-review nor an analyst decision. The one backtest on record was unpromising — -33.21% total return, a 73.94% drawdown, and just 0.056% average per-trade return against a 0.15% viability floor. The commission-free venue removed the documented fee killer but did not surface a tradeable edge, and the strategy's operational complexity prevented it from ever being properly evaluated.

A 16-instrument synchronized-panel long/short is heavy to code and verify; the coding stage kept timing out, and the one backtest that did run showed the reversal spread (0.056%/trade) still fell short of even USEQ's low costs — the commission-free venue did not resurrect the edge.

The run was abandoned with reason coding_stuck_3_timeouts — after 2 iterations the coding stage timed out three times and never cleared, so no backtest-review or analyst verdict was ever recorded.

A dollar-neutral cross-sectional short-term reversal on 16 liquid US mega-caps (USEQ daily bars): each weekly rebalance goes long the trailing-5-day relative losers and short the winners in equal dollar legs, betting that the classic overreaction anomaly killed by crypto fees survives on a commission-free equity venue.

The available backtest over the full 1980-2026 history was a loss: total return -33.21% (CAGR -2.56%), max drawdown 73.94%, profit factor 0.98 across 4,980 trades at a 53.9% win rate. Average per-trade return was only 0.056% of notional — well below the 0.15% viability floor — and the reported Sharpe (0.27) was flagged incoherent against the negative equity curve.