The Question
A major earnings beat drops at 6pm, after markets close. By 9:30am the next day, has that information been priced in? Partially. Different investor segments—retail traders checking morning news, algorithmic systems scraping overnight data, institutional analysts reading detailed reports—absorb and react to information on different timescales.
Information diffusion is not instantaneous. It's a time-dependent process, and that temporal structure is exploitable.
The critical question: When does news arrive relative to trading hours, and how should you weight overnight versus intraday sentiment to maximize predictive power?
The Approach
We divide each trading day into two information regimes:
- Non-trading hours (market close 3:00pm to next open 9:30am): Overnight news, after-hours announcements, off-market commentary
- Trading hours (9:30am to 3:00pm): Intraday news flow, real-time updates, market reactions
Using SmarTag News with precise timestamps, we construct two sentiment factors:
- Open-plan factor: Aggregates news from prior day's 9:25am to current day's 9:25am (captures overnight + early morning sentiment for open-position strategies)
- Close-plan factor: Aggregates news from prior day's 2:55pm to current day's 2:55pm (captures intraday + overnight sentiment for close-position strategies)
For each plan, we test differential weighting schemes: Does overnight news predict better? Or does intraday news matter more?
The hypothesis: Overnight news contains unpriced information because fewer participants react to it in real time, while intraday news gets absorbed faster through continuous trading.
The Finding
For open-plan strategies, higher weights on non-trading-time news (overnight) performed better. This makes intuitive sense: at market open, the overnight information hasn't been fully digested yet—retail investors are just reading headlines, and algorithms are still processing.
For close-plan strategies, higher weights on trading-time news performed better. By market close, intraday news has been continuously incorporated, while overnight news from the previous session is stale.
The effect was most pronounced in small-cap stocks (CSI 1000) with short holding periods. The close-weight factor achieved 311% annualized return with a Sharpe ratio of 7.18 during high-volatility periods—an extraordinary but inherently capacity-constrained opportunity. Medium and long holding periods (10-20 days) favored open-plan approaches, suggesting different information decay patterns for different time horizons.
Combining overnight sentiment with momentum factors amplified returns. Stocks with strong overnight negative news + recent price momentum generated the highest-Sharpe short-side returns, as the market initially underreacted to bad news on trending names, then corrected sharply.
Try It Yourself
Timing-based sentiment strategies require precise data infrastructure: intraday timestamps, fast factor calculation, and low-latency execution for short holding periods.
Practical applications:
- Intraday quant strategies: Use overnight sentiment to generate opening signals, then fade those signals by midday as information gets priced in
- Momentum overlays: Filter momentum strategies by overnight sentiment—negative news on uptrending stocks flags exhaustion, positive news on downtrends signals potential reversals
- Capacity allocation: Deploy size in longer-horizon (5-10 day) strategies where timing effects are smoother and less crowded
Interested in building a timing-aware sentiment pipeline? Book a call to discuss data delivery, factor construction, and execution infrastructure.
Source: 招商证券《“蓝海启航”系列研究之二：新闻情绪数据在中低频Alpha中的应用》 (2020-12-16).