Does Exchange-traded Fund Ownership Affect a Firm’s Expected Crash Risk?

Abstract

We examine the effect of equity ownership by exchange-traded funds (ETFs) on the expected (ex ante) crash risk of the underlying securities. We observe a positive relationship between ETF ownership and the firm’s expected crash risk. Our findings suggest that ETFs increase information opacity, leading managers to withhold negative news, which in turn amplifies the anticipated crash risk. We demonstrate a positive causal relationship between ETFs and expected crash risk by using the Russell 1000/2000 index reconstitution as an instrument for ETF ownership and ETF initiation as staggered exogenous shocks on ETF ownership. Moreover, this association becomes more noticeable when ETF ownership is broader, the ETF is larger, and the company’s information environment is more opaque.

Acknowledgements

We thank Jeffrey Ng (the editor), Robert Durand, Robert Faff, Michael Keefe, and two anonymous reviewers for constructive comments. Disclosure statement: The authors report there are no competing interests to declare.

Disclosure statement

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

Notes

1 In 2001, global ETF assets were $105 billion; in 2006, they were $580 billion; in 2011, $1,355 billion; and $3,423 billion in 2016. This figure is expected to continue to climb further (Ernst and Young, 2017; Ben-David et al., 2018; Statista Research Department, 2022). These data are obtained from https://www.statista.com/statistics/224579/worldwide-etf-assets-under-management-since-1997.

2 The total AUM is calculated as the product of the market price per share and the number of shares outstanding. We obtain data from: https://www.ssga.com/us/en/intermediary/etfs/funds/spdr-sp-500-etf-trust-spy.

3 Such migration happens due to noise traders’ efforts to lower costs. Specifically, informationally disadvantaged noise traders can avoid the high information-processing costs of researching individual stocks while passively investing in those stocks through an ETF’s diversified portfolio. Thus, noise traders’ losses are presumably lower in ETF markets compared to the market for the individual stocks (Israeli et al., 2017). Therefore, noise traders find higher utility in trading ETFs than the individual securities included in the ETF basket.

4 Although these shares are available for trading as part of a basket transaction at the ETF level, they are no longer available to traders who wish to transact on firm-specific information (Israeli et al., 2017).

5 For example, the crash risk is higher for firms with more accruals-based earnings management (Hutton et al., 2009), more real earnings management (Khurana et al., 2018), less readable and more ambiguous annual reports (Ertugrul et al., 2017; Kim et al., 2019a), a higher degree of earnings smoothing (Chen et al., 2017), and less voluntary corporate social responsibility-related disclosure (Kim et al., 2014; Lee, 2016; Zhang et al., 2016).

6 These statistics are similar to those of Israeli et al. (2017), who identify 443 unique ETFs offered in the US between 2000 and 2014. Our ETF ownership measures are similar to those in Ben-David et al. (2018) (e.g., 2.6%).

7 See, for example, Dennis and Mayhew (2002), Bradshaw et al. (2010), Van Buskirk (2011), Kim and Zhang (2014), and Kim et al. (2019b).

8 For example, the average Market value and MTB are 7.561 and 0.037, similar to the study by Kim et al. (2016b), which finds the average market value and MTB to be 7.582 and 0.037, respectively. In addition, SIR is 5.8%, which is consistent with the 3.71% documented in Callen and Fang (2015). We also find that the average CSR is −0.139, which is consistent with the −0.165 that Lins et al. (2017) document. Overall, the descriptive statistics for our sample firms are consistent with the prior literature.

9 The calculation of economic significance is as follows: 13.9% = 0.311*0.0225/0.0505.

10 Similarly, we also check for robustness by removing the variables that significantly reduce the number of observations. Both the first- and second-stage results remain similar. For instance, after we exclude SIR, CSR, and illiquidity, we find that when the estimated ETF increases by 1 unit, the expected crash risk increases by 2.27–2.48 for the upper band and 2.14–2.36 for the lower band.

11 We exclude analyst following, accrual, SIR, CSR, and illiquidity so that our staggered DiD test will still have a sufficient number of observations to demonstrate statistical power. However, in an unreported result, when we add back those control variables, we continue to find similar results but with significantly fewer observations.

12 Our untabulated summary statistics show that 310 initiations (19.71%) occurred in the year 2000 and 692 initiations (43.99%) in 2001. ETF initiations in other years are evenly distributed (2.42% on average).