Attention Drift: �Salience Bias, Social Networks, and Predictable Returns

​

Alfred Qi FAN

HKUST Business School

Motivation: Networks in Asset Pricing

• Literature shows predictability of signals using past returns of a firm’s peers with economic links:
• Let’s use Tesla as a focal firm, e.g.
• Industry momentum: a stock tend to outperform if its industry peers have higher past returns.
• Similar idea applies to peers with other economic links
• Technological links, demand-supply relations, analyst coverage
• These are known as cross-stock momentum.

Attention Drift

Page 2

October 20, 2026

• Typically, a signal for return prediction is from a firm’s own characteristic
• For Tesla, we collect its size, B/M ratio, past returns,…

How cross-stock momentum can be explained?

Page 3

October 20, 2026

Attention Drift

What do we mean by salience?

Page 4

October 20, 2026

Attention Drift

An example for salience in stock market

• A screenshot on November 15, 2024
• J.P. Morgan is quite different from overall market
• Or has a salient performance
• Salient thinking: what stands out may affect the decision maker’s judgement
• attention is attracted to salient feature
• Investors with salient thinking tend to overweight this (Bordalo et al, 2013 AER)
• When evaluating past returns for JPM, this day is important

Page 5

October 20, 2026

Attention Drift

Hypothesis Development

• hypothesis 1: future outperformance when peers had attention-grabbing outperformance
• Peng and Xiong (2006): attention efficiently allocated to more precise signals, e.g. industry level
• attention on focal firm is distracted, which is allocated to facilitate buying decision on peers
• less attention leads to relatively lower price pressure and underpricing
• while not as pronounced when the peers has attention-grabbing underperformance
• Barber and Odean (2008 RFS): attention is more important for buying decisions

​

Page 6

October 20, 2026

• hypothesis 2: amplified effect among peers with stronger social ties with focal firm
• Han et al. (2022): investors tend to be attracted by salient right-skewed (extreme positive returns) assets by social interactions

Attention Drift

Hypothesis Development

Page 7

October 20, 2026

Barber and Odean (2008 RFS): attention is more important for buying decisions

focal

peer

Attention

Price Pressure

optimistic

belief

tend to

buy peers

distracts

underpricing

negative

belief

less intent

to buy peers

not necessarily

distract that much

peer outperformance

peer underperformance

moderate

overpricing

Attention Drift

Literature Review

• social connectedness in financial markets
• market outcome: improve market efficiency (Hirshleifer et al., 2024); engender anomalies by provoking behavioral bias (Bali et al., 2021)
• investment decisions: individual (Hu, 2022) and institutional (Kuchler et al., 2022)
• salience in asset pricing
• negative predictability of (focal) salience bias: across various asset classes
• largely attributed to short-term reversals (overreaction)
• This paper: positive predictability of peer salience bias
• lead-lag effects in financial markets
• past stock returns can predicts return of linked firms (return-to-return)
• e.g. industry momentum (Moskowitz and Grinblatt, 1999), customer momentum (Cohen and Frazzini, 2008), connected-firm momentum (Ali and Hirshleifer, 2020)
• This paper: lead-lag effects from behavioral bias to returns (microfoundation of investor behavior)

Page 8

October 20, 2026

Attention Drift

Main Variable

Page 9

October 20, 2026

FF48 industry peer firms for focal firm i

spST example

ST example

Attention Drift

Time-Series Variation in spST

Page 10

October 20, 2026

Attention Drift

Main Result: Univariate Analysis

• main exercise: sort stocks based on social peer salience bias spST into ten groups
• where spST is computed using data in previous month
• compute average returns in current month within each decile
• time-series averages of spST-sorted portfolio returns are reported above
• sample: CRSP common stocks, 1963–2022; penny stocks excluded

​

Page 11

October 20, 2026

Attention Drift

Main Result: Further Evidence

• robustness of main results
1. not explained by prevailing risk and behavioral factor models
2. significant in both subsample periods: 1963–1992, 1993–2022
3. not dominated by microcaps (firms below 20th NYSE size breakpoints)
4. insensitive to industry classifications methods
5. robust to alternative formation of focal firm level salience bias ST results
• model parameters, market benchmark, mental window of past daily returns
• presence in international markets
• headquarter location mapped to GADM level-2 regions (equiv. to U.S. county).
• out of 15 markets, long-short strategy is postive in 14 markets (8 significant)
• long-term performance results
• return continuation in 12 months after portfolio formations
• no significant reversal pattern in the following two years (12 – 36 months)

​

Page 12

October 20, 2026

Adjusted return

Robustness results

Robustness in forming ST

Global evidence

Long-run results

Attention Drift

Main Result: Fama-MacBeth Regressions

Page 13

October 20, 2026

Controls ✓ ✓ ✓ ✓ ✓ ✓

incremental power of social networks

Attention Drift

Is spST subsumed by cross-stock momentum?

• Racing with leading cross-stock signals
• Outraces social peer returns (spRET), SCI-weighted peer return (Peng et al., 2023)
• Burt and Hrdlicka (2021): need to use idiosyncratic returns of peer firms to avoid the correlated alphas of economically-linked firms

Page 14

October 20, 2026

Results

Attention Drift

Mechanism Test: Attention Distraction

Page 15

October 20, 2026

peer-level effects: persistent distraction

focal-level effects: temporary attraction

Attention Drift

Inspecting Mechanisms

• Cross-Sectional Heterogeneity
• main results more pronounced for
• firms with less investor attention/higher arbitrage costs
• high IVOL, low inst. ownership, small size, high illiquidity, less coverage
• firms whose HQs are in the counties with high centrality in FB social networks
• time periods following higher investor sentiment
• Difference primarily due to overpricing of short legs in high sentiment (Stambaugh-Yu-Yuan)
• Decompose information contained in peer and focal firms
• salience ST, and return: positive v.s. negative
• earning surprise: no announcement (no news), positive, negative

Page 16

October 20, 2026

Attention Drift

Micro-Level Evidence

• employ Odean’s discount brokerage data
• randomly select 10,000 accounts for computational efficiency
• apply data filter as in Ben-David and Hirshleifer (2012, RFS)
• include: available CUSIP, common shares
• exclude: negative commissions, inactive stocks, negative cum. share positions
• two piece of evidence, pointing to corroborate hypothesis
• spST (attention-grabbing peer outperformance) drives investor trading
• contagion in buying/selling activity: more pronounced among socially connected investors
• inspired by Bali et al. (2025) – “social interactions and lottery stock mania”

​

​

Page 17

October 20, 2026

Attention Drift

spST drives individual investor trading

Page 18

October 20, 2026

controls include absolute return, return volatility, log market cap

dummy: investor k act on (focal) stock i on day t

Attention Drift

Contagion across peer investors

Page 19

October 20, 2026

Attention Drift

Conclusion

• a new perspective to examine lead-lag effects in asset pricing from microfoundation of investor behavior
• salience bias among peer firms positively predicts focal firm’s future return
• in contrast to the short-lived, negative predictability of focal firm’s own salience bias documented in literature
• social networks do play a role to corroborate this effect
• consistent to the argument that investor attention is asymmetrically distracted when peer firms are outperforming saliently, leading to the underpricing

​

Page 20

October 20, 2026

Attention Drift

Thank you!

Look forward to questions and comments!

Appendix: Simple Example for ST

Back

Appendix: Simple Example for spST

Back

Appendix: visualization of social connectedness

Back

Manhattan, NY

LeHigh, PA

Appendix: Adjusted Return of spST portfolios

Back

Appendix: Robustness, Part I

Back

Appendix: Robustness, Part II

Back

Appendix: International Evidence

Back

Appendix: Long-run performance

Back

Appendix: Racing with other cross-stock signals

Back

Appendix: cross-sectional heterogeneity

Back

Appendix: different information types

Back

construct spST by decomposing peers

conditional on focal firm’s info