When Intraday Trades are carried overnight
Quantitative research is, at its core, about following rules. As in any other STEM discipline (science, technology, engineering, and mathematics), precise frameworks give research rigor, discipline, and comparability. Yet, because such frameworks often remain unquestioned, challenging one of their constraints on purpose can sometimes be an informative experiment. In intraday trading, the first and perhaps most sacred rule is simple: all positions must be fully closed by the end of the trading session, with no risk carried overnight.
This article challenges that core assumption. We ask whether an intraday trend trade should always be closed no later than the end of the session, or whether value can be extracted by carrying the position overnight and liquidating it at the following morning’s open. The hypothesis is straightforward: if an intraday move reflects persistence in price behavior, a fraction of the edge could still survive after the closing bell.
We first pick a plain-vanilla (gross-of-cost1) intraday-trend framework on SPY, tested from January 2006 to January 2026. Similarly to our work in QuanTips #2 (Improving Performance with Fast Alphas), the signal relies on ATR bands centered around the session open: the model enters long positions when price breaks sufficiently above the opening level, and shorts are triggered when price breaks sufficiently below it, with the session open always behaving as an exit level. The strategy controls exposure through a volatility scaling mechanism, targeting a daily volatility of 2%, with a standard 400% leverage cap.
We then define two strategy variants, keeping entries and sizing unchanged. The baseline version closes all positions at the end of the day. The alternative carries any residual exposure overnight and exits at the next session open. This may look like a minor implementation choice, but it changes the nature of the strategy. Once positions cross the close, the model is no longer purely intraday. It becomes exposed to overnight news flow and gap risk.
In a combined long-short implementation, the comparison between the two variants remains mixed (see Figure 1 for reference). End-of-day liquidation delivers a 14.1% CAGR, a Sharpe ratio of 0.88, and a maximum drawdown of -25.6%. Shifting the exit to the next open lifts CAGR by roughly 1%, but Sharpe falls to 0.75 and maximum drawdown widens to -45.4%. In other words, carrying open trades overnight seems to slightly improve returns, but it does so with materially higher risk.
At this stage, one important factor to ponder is the impact of the overnight risk premium. It has been widely documented that a large part of long-term equity performance has historically been earned outside regular trading hours (Cooper, Cliff, and Gulen, Return Differences between Trading and Non-Trading Hours: Like Night and Day; Berkman, Koch, Tuttle, and Zhang, Paying Attention: Overnight Returns and the Hidden Cost of Buying at the Open). If such an effect is indeed present, one would expect overnight exposure to enhance the long side of our strategy and simultaneously cause the short side to bleed.
Once we split our long-short portfolio into its two sleeves (see Figure 2), that is exactly what we notice. On the long side, holding overnight boosts performance decisively: CAGR rises from 7.3% to 13.4% and Sharpe jumps from 0.76 to almost 0.97. On the short side, we notice the opposite effect, with CAGR falling from 6.3% to 1.5% and Sharpe dropping from 0.52 to nearly 0.17.
Seeing this asymmetry in behavior, we can infer that intraday trends do not seem to persist overnight: if that were true, we should observe somewhat balanced benefits across the long and short legs, but that is simply not what the evidence shows.
In practice, we can say that holding trades overnight changes the nature of the returns we are capturing: by crossing the close, the strategy effectively loads on the positive overnight risk premium present in equity markets. As a result, long positions benefit from that effect, while short positions suffer from it. This also explains why the original long-short implementation does not exhibit a clean performance improvement: the gains collected by overnight longs are largely offset by the consistent headwind introduced by overnight shorts.
This insight can be used to transform the strategy into a hybrid framework. It holds overnight exposure only when the end-of-day position is long. In a sense, we are using the overnight risk premium as a fast alpha signal: something that might struggle to survive transaction costs on its own but can still enhance a pre-existing strategy when used to condition its exit mechanism, without increasing turnover and thus implementation costs.
As we can see in Figure 3, the enhancement from holding long positions overnight is economically meaningful. Relative to the baseline end-of-day exit, CAGR rises from 14.1% to roughly 20% and Sharpe improves from 0.88 to almost 1.07, although maximum drawdown widens from -25.6% to -34.1% (results presented are gross-of-cost).
From a research standpoint, these results highlight why unconventional but controlled experiments are often worthwhile. Breaking a rule, even one as firmly established as end-of-day liquidation in intraday trading strategies, can reveal findings that would otherwise remain hidden. Experiments like this can also be useful reminders of why thinking out of the box matters. Even when results are not striking at first glance, breaking a familiar rule and examining it more closely can reveal exploitable market dynamics.
Are you willing to break the rules?
- Since the objective of this analysis is to assess whether a statistical edge arises from carrying positions overnight, the only modification relative to the baseline intraday strategy is the timing of the exit, not the number of trades. Both implementations, the pure intraday version, where positions are closed at the end of the session, and the overnight version, where positions are liquidated at the following morning’s open, generate the same number of transactions. As a result, including transaction costs would not affect the relative comparison. ↩︎