Strategy Card

Hybrid Overnight Momentum with Intraday Extension Fade — strategy card

Spec ID: spec-hybrid-overnight-intraday-us-stocks-1782341488 · Generated: 2026-06-24 23:37 UTC

Cluster: Momentum · Sub Cluster: Overnight Momentum Intraday Fade

One-line description

Daily long-short strategy on liquid U.S. stocks that buys names with persistent positive overnight behavior and shorts names with weak overnight behavior plus stretched intraday moves. Positions are rebalanced at the close and held close-to-close with volatility-adjusted equal weighting.

Why this trade exists

Overnight returns may contain slow-moving information diffusion, retail/institutional order imbalance, and risk premia associated with bearing gap risk outside regular trading hours. Securities with repeated positive overnight returns and high overnight hit rates may continue to attract demand into the next session.

The intraday extension penalty/fade attempts to avoid buying names whose same-day regular-hours move is already statistically stretched, and to identify short candidates where positive intraday extension is not confirmed by overnight strength. Structurally, this combines an overnight continuation effect with a short-horizon intraday mean-reversion filter.

Algorithm

[code omitted from public view]

Parameters

Param	Value	Notes
Universe	Top 100 liquid U.S. stocks	Sorted by 1-year dollar volume; stock-only universe.
Backtest window	2018-01-01 to 2023-01-01	Daily bars.
Rebalance	Daily at close	Holding period is close-to-close.
Long book	50% gross target	Top 20 eligible names by composite score.
Short book	50% gross target	Bottom 20 eligible names; shorts enabled.
Max leverage	4.0x	Gross leverage limit from rule parameters.
Max position size	10%	Per-symbol cap before/after volatility adjustment as implemented.
Liquidity filter	20d ADV >= $10mm	Uses volume_in_dollars; skip if filter fails.
Overnight windows	5d and 20d	Used in composite score; 20d also used for long/short filters.
Overnight hit-rate window	20d	Long threshold >= 0.55; short threshold <= 0.45.
Intraday z-score window	20d	Penalizes positive intraday extension; long max 1.25, short min 1.5.
Risk filter	20d realized vol <= 60%	Positions volatility-adjusted using 20d window.
Long exits	Rank < 40; intraday z >= 2.0; 20d overnight return <= 0	Exit rules from spec.
Short exits	Rank > 60; intraday z <= 0.25; overnight return >= 0	Spec mixes 20d/5d rolling overnight references; implementation should be verified.
Costs	Not specified	Treat reported results as pre-cost unless a platform default was applied.

Look-ahead audit

#	Concern	Status
1	Same-day close used for features and close rebalance	Needs confirmation that signals computed at close are executable only after the close or with realistic closing auction assumptions.
2	Overnight/intraday decomposition	Formulas use open, close, and prior close; no future bars are required if day-t close is known before trading decision.
3	Rolling feature windows	Should be lagged consistently when generating orders; verify no inclusion of post-trade returns in the signal.
4	Universe selection	Top 100 by 1-year dollar volume may introduce survivorship or future-liquidity bias unless point-in-time membership is used.
5	Corporate actions and delistings	Requires split/dividend-adjusted OHLC and delisting handling to avoid overstated returns.
6	Transaction costs and borrow	No explicit costs supplied; short borrow fees, bid-ask spread, and close auction slippage remain unaudited.

Caveats / known limitations

• Backtest covers only 2018-2023, a short sample with unusual volatility regimes including COVID and the 2022 rate shock.
• Explicit transaction costs are absent; 28,181 trades imply that spread/slippage can materially reduce realized Sharpe.
• Close-to-close execution assumes reliable participation at or near the close, which may be optimistic for high turnover.
• Universe appears restricted to large liquid names but may still suffer survivorship bias if constituents are not point-in-time.
• Short-side implementation requires borrow availability and borrow-cost modeling.
• The strategy is not sector- or beta-neutral by construction, so residual factor exposures may explain part of the return.
• Parameter set is fairly detailed and may be overfit without walk-forward or out-of-sample validation.

Results

The backtest produced a 38.07% total return with Sharpe 0.86, Sortino 1.18, and max drawdown of -11.35% over the 2018-2023 window. The profile is moderate rather than exceptional: win rate is 55.88% and profit factor is 1.15, suggesting a small edge applied frequently. Given the high trade count and missing explicit cost assumptions, post-cost robustness is the key unresolved question.

Backtest metrics snapshot

Metric	Value
Total Return	38.07%
Sharpe	0.86
Sortino	1.18
Calmar	0.59
Max Drawdown	-11.35%
Volatility	8.83%
Win Rate	55.88%
Profit Factor	1.15
Total Trades	28181
Symbols	100 (AAPL, ABBV, ABT, ACN, ADBE, ADI, AMAT, AMD, AMGN, AMZN, +90 more)