Abnormal Investor Visits (AIV, 异常机构调研信号)

The cross-sectional residual of institutional on-site visit frequency after stripping out the expected component driven by firm size, liquidity, and past performance. On SZSE-listed A-shares (2012–2019), AIV is a positive cross-sectional alpha proxy that survives standard controls and Fama-French 5-factor adjustment.

Definition

For firm i and calendar month m, let

\text{LNVISIT}_{im} = \log\bigl(1 + \#\text{on-site investor visits in months } m-2, m-1, m\bigr) Run the monthly cross-sectional regression \text{LNVISIT}_{im} = \alpha_m + \beta_1\,\text{SIZE}_{im} + \beta_2\,\text{TURN}_{im} + \beta_3\,\text{MOMEN}_{im} + \beta_4\,\text{ROA}_{im} + \varepsilon_{im}

with

• \text{SIZE}_{im} = log(circulation market cap in CNY millions)
• \text{TURN}_{im} = 12-month average share turnover
• \text{MOMEN}_{im} = trailing 12-month cumulative return (skip month m)
• \text{ROA}_{im} = operating income / total assets

Define

\text{AIV}_{im} \;\equiv\; \hat\varepsilon_{im}.

\text{AIV} > 0 means the firm was visited more often than its size, liquidity, momentum, and profitability would predict. The signal is published monthly; in the original So Wang Zhang 2024 Investor Corporate Visits paper it is winsorised cross-sectionally at 1/99% before use.

Empirical properties (SZSE, 2012-07 to 2019-12)

Lagged AIV predicts returns for 6 (VW) to 10 (EW) months without sign reversal — inconsistent with transient price pressure.

Where the alpha concentrates

The signal is stronger when (all interactions significantly positive):

• The firm is neglected — low analyst coverage, low institutional ownership, recent losses, or small size (Table 5 Panel B).
• Trading volume is low — combining high-AIV / low-abnormal-turnover with low-AIV / high-abnormal-turnover yields ~1.49% / month DGTW-adjusted alpha (Table 6).
• The visiting institution is paying higher costs — non-holder visits, initial visits (no visit in past 6 months), or above-median travel time from Baidu Map API (Table 5 Panel A).
• Public attention is low — interaction terms with analyst revisions and media/analyst attention are significantly negative.

The signal is not subsumed by:

• Standard cross-sectional controls (size, B/M, momentum, profitability, asset growth, turnover, reversal) — Table 3 col 3.
• Abnormal analyst coverage (Lee & So 2017's A_{TOT}) — Table 3 col 3-4.
• Average signed returns around visits (Cheng et al. 2019's AVG_SAR) — Table 3 col 4.
• Lagged / contemporaneous / forward mutual-fund holdings (column 7 even intentionally includes a forward-looking fund-holding control) — Table 3 cols 5-7.
• Sell-side-analyst visits (AIV_NOFUND has no incremental power) — Table 3 col 8.
• Information risk (VPIN, bid-ask spread) — Internet Appendix A1; both proxies are negatively correlated with AIV.

Mechanism

AIV predicts subsequent earnings news (Table 7): positive on standardised unexpected earnings (SUE), earnings-announcement size-adjusted returns (SAR), analyst forecast errors (FE), and analyst revisions (REV). Hedge returns are 3.6–4× larger on the next earnings-announcement day than on non-announcement days. AIV also precedes institutional buying (Table 8): holdings of visiting funds rise with AIV while holdings of non-visiting funds fall — direct evidence of informed trading capitalising the signal into price.

AIV does NOT survive when the visit is moved online: post-COVID online AIV has no return predictability, consistent with the costly-information story.

Relationship to other "abnormal attention" factors

All three follow the same template — strip out the mechanical component driven by size / liquidity / past returns from a flow-based attention measure, take the residual. The deep mechanism differs (analyst coverage is a free public good, AIV is a costly private vetting, hedge-fund info acquisition is a paid service).

Implementation notes

• Use the disclosure date, not the visit date — ~88.9% of visits are disclosed within the mandated 2-day window.
• The signal is published in real time on irm.cninfo.com.cn/szse/ (互动易). A-share data vendors (CSMAR, Wind) carry a cleaned version.
• Combine with a size / liquidity screen: alpha concentrates in small / illiquid names — typical of an attention / neglect story.
• Crowding risk: once the paper is published and the signal is widely replicated, expect the EW alpha to compress. Realistic OOS expectation is roughly half the in-sample.

Limitations and open questions

• SZSE-only; no Shanghai counterpart because the SHSE does not require same-day disclosure.
• No transaction-cost layer; turnover of the quintile-5 minus quintile-1 portfolio and A-share commissions / impact / stamp duty are not modelled.
• The "abnormal" definition is deliberately parsimonious (4 controls); sector, geographic-cluster, or institutional-coverage fixed effects might further reduce the residual variance but should not change the sign.
• Combination with Cross-Sectional Momentum or Time-Series Momentum is unexplored.

See also

• So Wang Zhang 2024 Investor Corporate Visits — source paper.
• Investor Corporate Site Visits — the underlying behaviour.
• Shenzhen Stock Exchange Disclosure Regime — the institutional rule that makes the signal observable.
• Costly Information Acquisition — the general principle.
• Neglected Stocks — where the alpha concentrates.
• Post-Earnings Announcement Drift — the release channel through which the alpha realises.
• Investor Attention — broader family of attention-based alpha factors.