The Effect of Institutional Ownership on the Timing of Earnings Announcements: Evidence from the Russell Index Inclusion

Abstract

Using the annual Russell 1000/2000 index reconstitution as an exogenous shock to institutional ownership (IO), I examine the effect of IO on firms’ decisions regarding the time of day to announce earnings. I argue that firms with high IO strategically choose to announce earnings after hours to facilitate post-announcement price discovery and reduce volatility because the after-hours period is largely dominated by sophisticated institutional investors. I find that firms with greater IO are more likely to announce earnings during after-market sessions (i.e., after hours after the market closes), but not during premarket sessions (i.e., after hours before the market opens). My analysis further shows that transient IO has a stronger influence on the likelihood of after-market announcements relative to quasi-indexer or dedicated institutional holdings and that firms with high IO are even more likely to announce during after-market sessions when firms have bad earnings news or when earnings include large transitory items. Lastly, I find that announcing earnings during after-market sessions indeed facilitates post-announcement price discovery and reduces volatility for firms with high IO. Collectively, my findings suggest that IO is a significant factor in firms’ disclosure timing decisions and that the timing of disclosures affects price discovery and volatility.

Keywords:

• Institutional investors
• Earnings announcements
• After-hours trading
• Price discovery
• Price volatility

JEL classification:

• G23
• G14
• M41

Acknowledgements

I appreciate helpful comments and suggestions from my dissertation committee (Charles Wasley, Joanna Wu, Sudarshan Jayaraman, and Jaewoo Kim), Gerald Lobo (editor), two anonymous referees, Jonathan Miller, Clark Wheatley, and workshop participants at the University of Rochester and Florida International University. I thank Jason Chao of Russell Investments for providing the index membership data and Brian Bushee for sharing the data on institutional investor classification. All errors and omissions are my responsibility.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplemental Data and Research Materials

Supplemental data for this article can be accessed on the Taylor & Francis website.

Appendix OA.A1: Discussion of the choice of empirical method

Appendix OA.A2: Discussion of control variables for the main test

Appendix OA.A3: Discussion of control variables for the price discovery and volatility test

Appendix OA.A4: Discussion of potential reasons for the different results between premarket and after-market announcements

Table A1: The effect of institutional ownership on the likelihood of after-market announcements within a narrower bandwidth

Table A2: Descriptive statistics for trades during the 24 hours following earnings announcements

Table A3: The effect of institutional ownership and the timing of earnings announcements on price discovery during the 4 hours following earnings announcements

Notes

1 Doyle and Magilke (2009) argue that announcements after the market closes are associated with greater trading volume because such announcements are successful at disseminating the news to a broader audience.

2 In my sample, while approximately 9% of earnings announcements happen during regular trading hours, 59% of announcements are made after hours. More specifically, 21% (38%) of announcements are released during the after-hours period before (after) the market opens (closes). Lastly, the remaining 32% of announcements happen during non-trading hours.

3 I discuss these studies in more detail in the literature review section.

4 After hours, also called extended-trading hours, include a premarket session and an after-market session, which run from 4 am to 9:30 am Eastern time (ET) and 4 and 8 pm ET, respectively, by the end of my sample period.

5 There has been a considerable debate about which empirical specification is correct to use for the Russell index setting in the literature. I discuss this issue and my choice of empirical design in section A1 of the Online Appendix.

6 I discuss potential reasons for different results between premarket and after-market announcements in the Online Appendix.

7 Although transient institutions are not required to mechanically change their portfolio allocation following the reconstitution, they hold firms with relative index weights to the extent that they benchmark to the index.

8 During my sample period, after hours vary by time period and exchange (see Appendix A). In empirical tests, I define a firm's after hours based on its exchange listing and the specific time period.

9 I examine trade size during different types of trading hours in my sample and find evidence consistent with Barclay and Hendershott (2003) and Li et al. (2015). Prior studies find that IIs typically place large orders (Lee & Radhakrishna, 2000; Ali et al., 2008). Results in the Online Appendix show that trade size during after-hours periods is significantly larger relative to during regular trading hours, which suggests that after-hours trades are largely executed by IIs.

10 Float adjustment may slightly alter the ranks of firms based on the end-of-May market capitalization. However, the float adjustment itself does not change the index assignment.

11 One might wonder whether post-announcement noise trading would be a concern for firms with very high IO (e.g., 95% of shareholdings). Even firms with very high IO could be concerned about noise trading during regular trading hours as long as their IIs trade based on earnings information. Suppose the firm is 95% owned by IIs (e.g., 20% owned by transient IIs and 75% owned by quasi-indexer and dedicated IIs) prior to an earnings announcement and the firm missed the earnings target. Transient IIs may sell shares. If the earnings announcement is made during regular trading hours, these shares sold by IIs could easily be bought by algorithmic day traders and be sold to other day traders, who are likely to sell again based on intraday price movements. This chain of noise trading can exacerbate price volatility and drift prices away from fundamental values. On the other hand, if the announcement is made during after hours, both sides of post-announcement trades are likely to be from informed institutional traders because trading of noise traders is limited during such hours (Barclay & Hendershott, 2003; Gregoriou, 2015).

12 The banding policy allows firms that would otherwise be excluded from the indexes to retain index membership if their market capitalization is close to the index rank threshold. For example, if the market capitalization of the firm ranked 1,000th drops to the market capitalization of the firm ranked 1,010th, but its market capitalization is still within 2.5% of the new 1,000th firm's market capitalization, the firm retains its membership in the Russell 1000 (see Boone & White, 2015). I drop all observations after 2006 because the banding policy reduces the local discontinuity of firm assignment around the threshold.

13 Following Boone and White (2015), I leave a two-month gap between the index recomposition (the end of June) and the beginning of the earnings announcement period to allow enough time for the index assignments to influence IO. However, my results are unchanged when a one-month or a three-month gap is used.

14 See http://acct.wharton.upenn.edu/faculty/bushee/IIclass.html. I thank Brian Bushee for sharing the data on institutional investor classification.

15 For trading that occurs during trading hours, I only keep trades that meet the following criteria: (1) the trade occurs on the NYSE, AMEX, or NASDAQ; (2) the trade was made under regular market conditions (i.e., COND codes ∗, @, E, F, and blank); (3) trades without subsequent cancellations; and (4) the transaction price and the number of shares traded were both positive. For after-hours trades, I include trades with TAQ ‘COND’ codes that include T or F, which represent the bulk of all after-hour trades. For TAQ ‘EX’ codes, I exclude ‘after-hour trades’ on the NYSE and AMEX as they likely represent regular session closing transactions reported after 4:00 pm ET.

16 I use +/-100 firms around the index threshold because the difference in IO dissipates as the bandwidth becomes wider than +/-100. This is apparent in Panel A of Figure 2. Using a narrower bandwidth allows more reliable inferences but reduces the power of the test. I repeat the main analysis with ±50 firms and report a result in the Online Appendix. The result is qualitative similar.

17 Prior research that uses the Russell index inclusion as a setting implemented several different empirical approaches (Appel et al., 2016; Bird & Karolyi, 2016; Bird & Karolyi, 2017; Boone & White, 2015; Chang et al., 2015; Chen et al., 2019; Crane et al., 2016; Khan et al., 2017; Schmidt & Fahlenbrach, 2017; Young, 2018). I discuss why I adopt this empirical specification in more detail in the Online Appendix.

18 I subtract 1,000 from the raw ranking so that Rank indicates the location relative to the index threshold. Rank is zero for the smallest firm in the Russell 1000, negative for the other Russell 1000 firms, and positive for the Russell 2000 firms.

19 The definitions of these variables are provided in Appendix B.

20 I estimate ordinary least squares regressions rather than logit or probit models to avoid an incidental parameter problem with fixed effects (see Wooldridge, 2010). As a robustness check, I estimate probit models with fixed effects which yield qualitatively similar results.

21 I discuss why these variables are expected to influence the dependent variables (i.e., PreMarketEA and AfterMarketEA) in the Online Appendix.

22 I also use the tercile and quintile of SUE to identify good and bad news samples. I find that my empirical results are not sensitive to different ways of defining good and bad news samples.

23 I examine price discovery for the first few hours (i.e., 1–4 h) because Jiang et al. (2012) suggest that a significant portion of price discovery occurs within the first few hours following earnings announcements. Consistent with Jiang et al. (2012), I find that on average 48.7% of price discovery occurs during the first 4 h following earnings announcements in my sample.

24 I examine price volatility for a relatively longer horizon (compared to the time horizon for price discovery) because prior research (e.g., Landsman & Maydew, 2002) finds that heightened price volatility around earnings announcements typically lasts 2–3 days following the announcements.

25 I discuss why these variables are expected to influence the dependent variables (i.e., PreMarketEA and AfterMarketEA) in the Online Appendix.

26 The F-statistic for dedicated IO is 0.147 which is well below the threshold (i.e., 10). However, this is not surprising because dedicated institutional holdings are unlikely to be affected by the Russell index assignment (e.g., Boone & White, 2015; Crane et al., 2016).

27 I discuss potential reasons why I find different results for premarket and after-market announcements in the Online Appendix.

28 Results for price discovery throughout the first 4 hours are qualitatively similar and reported in the Online Appendix. I do not examine price discovery for a longer time horizon because I find that a significant portion (i.e., on average 48.7%) of price discovery occurs during the first 4 hours in my sample.

29 I use the 3rd quartile IO rather than the median or mean because the effect of announcement timing on price discovery is conditioned on firms’ level of IO.

Additional information

Funding

This work was supported by Florida International University.