Sleep Disruption and Retail Investor Trading

Han · Hirshleifer · Sheng · Sun — NBER WP 33477 (2025, rev. 2026)

Thesis

Sleep disruption — even mild, environmental, invisible-to-you sleep disruption — measurably degrades how retail investors process information and trade. Later sunsets (by geography or season) and daylight-saving transitions reduce the abnormal returns on retail trades. The market consequences of a bad night are systematic, not noise.

Method

• Terry Odean's classic brokerage dataset: 10,000+ retail accounts, 1991–1996, matched to ZIP-code sunset times and time-zone borders.
• Four identification strategies, each attacking different confounds:

1. Within-ZIP panel: does performance fall when local sunset gets later? (ZIP + year-month fixed effects)

2. Regression discontinuity at time-zone borders: similar households an hour apart by the clock, same real daylight.

3. North-vs-south latitude × season reversal (later sunsets up north in summer).

4. Difference-in-differences around spring DST transitions (observing vs non-observing states, incl. CA-vs-AZ).

• Performance measured as subsequent Fama-French 3-factor abnormal returns on trade-weighted portfolios.

Key findings

• One hour later sunset → −29.3 bps lower monthly trading performance (t = −2.50). Robust to controlling for the standard seasonal-affective-disorder measure.
• Time-zone border RDD: −40.4 bps/month for households on the eastern (later-clock-sunset) side (p = 0.074); the border effect is −0.94 pp/month larger in summer than winter (p = 0.049) — evening light closer to bedtime hurts more.
• Latitude test: North × Summer interaction −64.4 bps/month (t = −2.36) — exactly the seasonal pattern sleep disruption predicts and SAD does not.
• Attention channel confirmed: later sunset → lower probability of trading a held stock during its earnings-announcement window, weaker response to earnings news, and stronger contrarian tilts against earnings surprises — the sleep-deprived miss and misprice announcements.
• DST result: performance declines after the spring clock change in observing locations, though this estimate is more sensitive to portfolio weighting than the monthly results.

Limitations

• Brokerage data is 1991–1996: pre-commission-free, pre-app trading. Effects in today's always-on retail environment could be larger — or different.
• Sunset/DST variation is correlational at the margin; the RDD and latitude designs do the causal work, but the DST DiD is the weakest leg (authors flag the weighting sensitivity themselves).
• Effect sizes are monthly abnormal returns on trades, not annual portfolio drag — the cumulative wealth cost depends on turnover.
• No direct sleep measurement (actigraphy/surveys) — the mechanism is inferred from external variation plus the announcement-attention patterns.

So what for Nick

• Don't make money decisions in the evening dark months of your worst-sleep seasons. The data says impaired processing shows up specifically in reactions to news — if you must react to an earnings print or a market dump, sleep first, act in the morning.
• The "I work fine on 6 hours" feeling is exactly what the sleep-deprived subjects believed too; the returns disagree.
• Practical: put rebalancing or sell decisions on a morning schedule; treat late-night portfolio tinkering as a red flag with a measurable price tag (~30–60 bps/month on affected trades here).