Cross-Sectional Short-Term Reversal, Dollar-Neutral Long/Short Basket of Liquid US Large-Caps (USEQ, Daily Bars, 3-5 Day Holding, Commission-Free)
Outcome: Abandoned
UsLargeCapShortTermReversalDollarNeutralLS
Outcome Summary
The strategy bet USEQ's zero commissions would let the small 3-5 day cross-sectional reversal spread survive. On a decisive 36-year sample it produced a positive Sharpe but a profit factor of 1.0156, with market impact swallowing 64.5% of gross PnL, ~$240k capacity, and a 71.4% drawdown concentrated in the stressed vol tercile — a leverage-amplified tail bet, not a neutral edge. The analyst abandoned it before optimization, noting a sibling in the same 0/84 dollar-neutral basket family had just been abandoned at PF~1.03.
A DOLLAR-NEUTRAL, MULTI-INSTRUMENT, cross-sectional short-term REVERSAL stat-arb on a basket of ~16 liquid US large-caps (USEQ). Every N daily bars, rank the universe by trailing 3-5 day return; go LONG the worst-performing quintile and SHORT the best-performing quintile, weighting each side to equal gross so the book is ~market-neutral. This is deliberately NOT the momentum-confluence family (the LTC/BTC/ADA dual-timeframe trend recipe that keeps dying or is already promoted) -- it trades the OPPOSITE sign at a……Show moreShow less
A DOLLAR-NEUTRAL, MULTI-INSTRUMENT, cross-sectional short-term REVERSAL stat-arb on a basket of ~16 liquid US large-caps (USEQ). Every N daily bars, rank the universe by trailing 3-5 day return; go LONG the worst-performing quintile and SHORT the best-performing quintile, weighting each side to equal gross so the book is ~market-neutral. This is deliberately NOT the momentum-confluence family (the LTC/BTC/ADA dual-timeframe trend recipe that keeps dying or is already promoted) -- it trades the OPPOSITE sign at a shorter horizon, on a different asset class, market-neutral rather than directional. It exploits the Lehmann/Lo-MacKinlay short-term reversal anomaly: over 3-5 days, cross-sectional winners give back and losers bounce, driven by liquidity-provision compensation and non-information price pressure. The mechanism is fee-DEAD on crypto (0.10-0.20% RT eats the 0.2-0.4% reversal spread) but VIABLE on USEQ, where commissions are zero and cost is only ~0.02-0.05% RT spread+impact on these names. Distinct from the abandoned USEQ classical anomalies (L74: dual-TF trend, calendar seasonality, single-name/sector-ETF reversion, staples cointegration pairs) -- this is a broad cross-sectional decile spread portfolio, the canonical form of the anomaly, whose edge is much stronger and more diversified than a single-pair or sector-ETF reversion, and it is realized ACROSS ~16 names so no single name carries the book.
Implements the Lehmann/Lo-MacKinlay short-term reversal anomaly in its canonical form: a cross-sectional decile(quintile)-spread portfolio rather than a single-name or single-pair reversion. calculate_signal returns a CONTINUOUS reversal score every session — the NEGATED cross-sectional z-score of the primary name's trailing formation-window return (positive = it underperformed the basket and is expected to bounce) — while the book decision fires on the calendar-anchored rebalance clock. The negation is what……Show moreShow less
Implements the Lehmann/Lo-MacKinlay short-term reversal anomaly in its canonical form: a cross-sectional decile(quintile)-spread portfolio rather than a single-name or single-pair reversion. calculate_signal returns a CONTINUOUS reversal score every session — the NEGATED cross-sectional z-score of the primary name's trailing formation-window return (positive = it underperformed the basket and is expected to bounce) — while the book decision fires on the calendar-anchored rebalance clock. The negation is what makes this reversal, the opposite sign of the momentum family the hypothesis explicitly distinguishes itself from. Dollar-neutrality is enforced structurally (both sides sized from the same per_name notional and the same k), so the book cannot drift into a levered directional bet — the failure mode behind past -100% 'market-neutral' books. leverage=2.0 is genuinely consumed: gross notional multiplies by self.config.leverage. Venue is USEQ because the mechanism is fee-dead on crypto (0.10-0.20% RT vs a 0.2-0.4% reversal spread) and only clears costs where commissions are zero; long+short on 2x Reg-T margin requires the MARGIN equity venue, not a CASH one. All 16 legs are currently-listed mega-caps with full daily history in the catalog, so the multi-leg alignment barrier is satisfied from the start of the window and ~73 rebalances/year give a large trade sample.
Breakeven/drag-band expectancy with no deployable edge on a decisive 11,753-trade / 36-year sample. profit_factor 1.0156 sits in the [0.85,1.10] drag band and below the L75 USEQ PF<1.2 line — on a zero-commission venue that means the reversal spread itself is absent, not fee-eaten. The dominant cost is IMPACT: impact_cost_pct 64.5% consumes two-thirds of gross PnL and capacity_usd is only $240k, so the nominal net 0.109% per-trade return survives only at toy scale and vanishes at any deployable size (QA's……Show moreShow less
Breakeven/drag-band expectancy with no deployable edge on a decisive 11,753-trade / 36-year sample. profit_factor 1.0156 sits in the [0.85,1.10] drag band and below the L75 USEQ PF<1.2 line — on a zero-commission venue that means the reversal spread itself is absent, not fee-eaten. The dominant cost is IMPACT: impact_cost_pct 64.5% consumes two-thirds of gross PnL and capacity_usd is only $240k, so the nominal net 0.109% per-trade return survives only at toy scale and vanishes at any deployable size (QA's smoke-test concern of 31.4% impact is confirmed and worse on full history). The book is not the diversified market-neutral edge it claims: max_drawdown 71.4% and return concentrated entirely in the stressed vol tercile (calm -100%, normal -89%, stressed +2807%) = a leverage-amplified tail bet. This is the dollar-neutral cross-sectional basket family (L45, 0/84) whose sibling UsSectorEtfCrossSectionalMomentumDollarNeutralLS was just abandoned at PF~1.03; two formation/rebalance parameters cannot lift a PF-1.02 signal already 65% consumed by impact above breakeven. Backtest is not numerically viable (PF<1.05, DD>60%). abandon_class=negative_expectancy.
Implementation
Cross-sectional short-term REVERSAL stat-arb across 16 liquid US large-caps on USEQ daily session bars (AAPL MSFT JNJ JPM XOM PG KO PFE WMT CVX MRK PEP CSCO INTC VZ T). Every 5-calendar-day bucket (~3-4 sessions, anchored to the bar's own timestamp so the schedule is restart-invariant) the basket is ranked by its trailing 4-session return: the 3 WORST performers are bought and the 3 BEST are shorted, equal dollar per name and identical notional per side, so the book is dollar-neutral by construction. Gross = equity x 0.60 x leverage 2.0 = 1.2x equity (0.6x long + 0.6x short) under USEQ Reg-T margin. Rebalancing runs as NETTING delta orders in whole shares, so names that stay in the same bucket keep their position and deltas below $500 are skipped, holding turnover down against USEQ's ~0.02-0.05% round-trip spread+impact.
Verification Results
Treat backtest short-side PnL as slightly optimistic; not a blocker for these highly liquid names.
Short legs (3 names per rebalance) incur no borrow cost in the backtest (USEQ borrow is not modeled). On these mega-caps borrow is cheap and the book is dollar-neutral (long+short balanced), so the omission is small, but it modestly overstates realized net edge.
No change required; noted for transparency.
min_bars_required overrides the base framework method (static analysis base_shadow warning). It is deliberate and documented (the real warm-up is the aligned price buffer), and returns a sane 25, so it is functionally correct.
Marginal economics on the smoke test: avg_trade_return_pct 0.1107% sits just above the ~0.05% USEQ spread+impact floor, profit_factor 1.046 is in the drag band, Sharpe 0.428 (CI [-0.37,1.20] straddling zero), and impact_cost_pct 31.4% already consumes a third of gross at only ~$1M capacity. The cross-sectional dollar-neutral USEQ basket family has been abandoned recently at breakeven expectancy (UsSectorEtfCrossSectionalMomentumDollarNeutralLS, PF ~1.03). This is a broader 16-name reversal (opposite sign, canonical anomaly form) so it is a fairer test, but the analyst should confirm the reversal spread net of impact actually clears costs on full history before optimizing. Advisory only.
Backtest Review
Sharpe
0.64
Total return
137.71%
Max drawdown
71.45%
Trades
11753
Win rate
52.0%
Profit factor
1.02
Large, reliable sample (11,753 trades, 36-year span) and a genuinely distinct mechanism vs the recently-dead momentum-confluence family
Sharpe CI low 0.54 does not straddle zero on the MTM equity curve
Net per-trade return 0.109% nominally clears the USEQ prefer-threshold
profit_factor 1.0156 sits in the [0.85,1.10] drag band and below the L75 USEQ PF<1.2 line (anomaly absent, not fee-eaten) on a zero-commission venue
impact_cost_pct 64.5% consumes two-thirds of gross PnL; capacity_usd only $240k means the edge is real only at toy scale
max_drawdown 71.4% on a supposedly dollar-neutral book; return entirely concentrated in the stressed vol regime (calm -100%, normal -89%, stressed +2807%) = leverage-amplified tail bet, not a diversified reversal edge
Dollar-neutral cross-sectional basket family (L45 multi_instrument 0/84); sibling UsSectorEtfCrossSectionalMomentumDollarNeutralLS abandoned days ago at PF~1.03
Outcome Summary
A reversal basket only pays if the raw spread clears impact at deployable size; PF ~1.02 with two-thirds of gross eaten by impact and $240k capacity can't be tuned into an edge.
Abandoned at backtest-review as negative-expectancy — PF in the drag band below the USEQ 1.2 line, returns concentrated entirely in the stressed vol regime (calm -100%, normal -89%, stressed +2807%).
A dollar-neutral cross-sectional short-term reversal stat-arb on 16 liquid US large-caps (USEQ, daily bars), long the worst / short the best 3-5 day performers to harvest the reversal anomaly on a zero-commission venue.
36-year, 11,753-trade backtest: 137.7% total return (CAGR 2.37%), Sharpe 0.64, but PF 1.0156, 71.4% max drawdown, net 0.109% per trade, impact costs 64.5% of gross, capacity only ~$240k.
Backtest and paper results are hypothetical. Trading involves risk of loss.