Investor Attention (投资者关注度)

A family of cross-sectional alpha factors derived from observable proxies of investor attention — retail (search, social media), institutional (analyst coverage, fund flows), and behavioural (on-site visits). Different proxies have different costs of acquisition and therefore different predictive powers.

Why attention matters

In a CAPM world, attention is free; in reality, attention is finite and costly. Allocators of attention are subject to:

• Cognitive bandwidth limits — cannot process every signal equally.
• Acquisition costs — different signals cost different amounts to obtain.
• Selection effects — agents disproportionately attend to where the expected payoff is high.

The empirical implication is that observable attention measures are informative about expected returns: high attention can mean over-reaction, low attention can mean slow incorporation of public news. Whether attention predicts positive or negative returns depends on who is paying attention and why.

The taxonomy

The hierarchy is informative: free attention signals are weak and often absorbed by other factors; costly attention signals survive and carry alpha. This is the Costly Information Acquisition story.

Institutional vs. retail attention

The literature distinguishes:

• Retail attention — Google searches, retail trading volume, Robinhood holdings. Da, Engelberg & Gao (2011) and Ben-Rephael, Da & Israelsen (2017) show retail attention predicts short-term price increases and long-term reversals — consistent with attention-driven over-reaction by unsophisticated investors.
• Institutional attention — analyst coverage, fund holdings changes, on-site visits. Institutional attention tends to be persistent and value-relevant, predicting future fundamentals (Cheng et al. 2019) or future earnings (So, Wang & Zhang 2024). The mechanism is information acquisition, not over-reaction.

This is the asymmetry in Hong & Stein (2007): bad news travels slowly because few investors pay costly attention to under-covered firms. The consequence is that institutional attention to neglected names is the most reliable source of cross-sectional alpha.

Where A-share attention data lives

• Retail: 东方财富股吧论坛、雪球、百度搜索指数.
• Sell-side / institutional: CSMAR 分析师覆盖、CSMAR 基金持股.
• Buy-side on-site: CSMAR 调研记录 + 互动易 portal (see Shenzhen Stock Exchange Disclosure Regime).
• Fund flows: CSMAR 基金净值与份额.

The unique feature of A-share data is the mandatory on-site visit disclosure, which gives a real-time, point-in-time buy-side attention measure that is hard to replicate in US data.

Implementation: a layered attention model

A practitioner combining multiple attention layers typically:

1. Compute each layer's signal monthly: SVI (retail), A_{TOT} (sell-side), AIV (buy-side).
2. Z-score cross-sectionally within each layer.
3. Either average (equal-weight composite) or stack: long the names with high institutional attention and low retail attention, short the reverse.
4. Conditional on neglect: keep only low-analyst-coverage / low-turnover names in the universe.

Limitations

• Crowding: once an attention signal is published, replication compresses alpha. Most published attention factors are known to decay in the 5–10 years post-publication.
• Free-attention signals are mostly noise; the cost-acquiring sub-sample is where alpha lives, but the cost itself is not always observable.
• Endogeneity: high attention can mean high information or high noise; the two are hard to disentangle in real time.
• Survivorship: attention measures (analyst coverage, mutual-fund holdings) are subject to dropout that itself correlates with returns.

See also

• Abnormal Investor Visits (AIV) — the institutional-attention signal that is the focus of this wiki.
• Investor Corporate Site Visits — the underlying behaviour.
• Costly Information Acquisition — the principle that gives institutional attention its bite.
• Neglected Stocks — where attention-based alpha concentrates.
• So Wang Zhang 2024 Investor Corporate Visits — primary reference.