ETFs and information asymmetry of underlying securities: evidence on the volume-conditioned return autocorrelation

ABSTRACT

This study investigates whether exchange-traded funds (ETFs) increase the information asymmetry of underlying securities by examining the relationship between the US equity ETF ownership or turnover and the volume-conditioned daily return autocorrelation proposed by. Specifically, we find that stocks with higher ETF ownership exhibit weaker return reversals. We confirm that the ETF ownership effect is robust in a variety of settings, such as various volume measures, within industry, and various regression frameworks. We also address the difference between the volume-conditioned autocorrelation and the unconditional autocorrelation. Moreover, we document that higher ETF turnover leads to weaker return reversals, independently of the ETF ownership effect. These findings suggest that higher ETF presence allows more informed traders to engage in trades or more liquidity traders to migrate to ETFs from underlying securities, resulting in higher information asymmetry for underlying securities.

KEYWORDS:

• ETF
• information asymmetry
• ETF turnover
• volume-conditioned autocorrelation

JEL CLASSIFICATION:

• G10
• G14

Disclosure statement

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

Notes

¹ See Easley, O’Hara, and Srinivas (1998) and Cao (1999) for the information effect of stock options and derivatives on underlying securities. See Subramanyam (1991) for stock index futures and Dow (1998) for a financial innovation.

² Cooper (1999) finds that increasing-volume stocks exhibit weaker weekly return reversals, supporting the implication of informed trading for the volume-return dynamics..

³ We also obtain qualitatively the same result in the robustness check when we employ the panel data analysis with time- and firm-fixed effects for yearly variables.

⁴ However, we find no evidence for the effect of ETF shorts on the volume-conditioned autocorrelation.

⁵ We find no pattern between the unconditional autocorrelation and ETF turnover. The result is available upon request.

⁶ Wang (1994) and Lo and Wang (2006) argue that the joint behaviour of price and trade volume reveals more information than do prices alone. See Kyle (1985), Blume, Easley, and O’Hara (1994), Conrad, Hameed, and Niden (1994), and Cooper (1999) for the literature on the volume-stock price behavior.

⁷ Both the noisy rational expectations theory (Grossman and Stiglitz 1980) and the market microstructure study (Kyle 1985) argue that informed traders respond rationally or strategically to a change in liquidity trading, which might lead to no change in the price informativeness of underlying securities. However, this study intends to analyse the impact of ETFs on information asymmetry, not the price informativeness, though they are closely related.

⁸ Conrad, Hameed, and Niden (1994) provide empirical evidence supporting CGW. The return reversals effect is widely used as a measure for stock illiquidity in the literature. (Pastor and Stambaugh 2003; Avramov, Chordia, and Goyal 2006; Nagel 2012).

⁹ We exclude pre-2000 data because there are a limited number of ETFs due to filtering. Filtering with a book asset of $10 million makes no difference. We restrict our sample to the following Lipper Objective Codes for broad-based U.S. equity: CA, EI, G, GI, MC, MR, SG, and SP. We also include sector funds that invest in U.S. companies with codes AU, BM, CG, CMD, CS, FS, H, ID, NR, RE, S, SESE, TK, TL, and UT. Ben-David, Franzoni, and Moussawi (2018) identify 454 ETFs between 1999 and 2015; Da and Shive (2018) identify 549 ETFs between 2006 and 2013; Glosten, Nallareddy, and Zou (2021) identify 447 ETFs between 2004 and 2013; Huang et al. (2021) employ 508 ETFs between 1999 and 2017.

¹⁰ We decide to employ quarterly estimates given the availability of quarterly ETF ownership. In Section IV, we also employ yearly estimates in the robustness check.

¹¹ We employ the zero/20/60-day moving average detrending because our autocorrelation is updated quarterly. LMSW uses the 200-day moving average method because they estimate a single autocorrelation per stock based on a multiyear daily return series. However, we obtain qualitatively the same result for the 200-day moving average detrending.

¹² Although the potential nosiness of our quarterly estimates may prevent us from discovering a desired empirical outcome, we succeed in documenting significant evidence supporting the hypothesis, as addressed in the next sections.

¹³ We confirm a lower turnover for underlying stocks in unreported analysis.

¹⁴ See Table II Panel B of Ben-David, Franzoni, and Moussawi (2018).

¹⁵ ETFTN and ETFSI show a low correlation with ETF because both variables are value-weighted, where the sum of the weights is equal to 1, not equal to total ETF ownership for a stock.

¹⁶ We employ the Fama and MacBeth (1973) quarterly cross-sectional analysis given the availability of quarterly ETF ownership. LMSW employ a single regression analysis for the volume-conditioned autocorrelation estimate based on the 6-year-long daily return series. As our quarterly estimates may be contaminated by noise, our Fama and MacBeth (1973) quarterly cross-sectional analysis may underestimate, not overestimate, the significance of our analysis. In Section IV, we also conduct the panel data analysis for yearly estimates with firm and time fixed effects.

¹⁷ LMSW conduct a simple, not multiple, regression analysis by using only one explanatory variable in each analysis, compared with our multiple regression analysis. However, their autocorrelation estimation based on the 6-year-long daily return series allow LMSW to obtain a higher explanatory power (i.e. adjusted r-squared) than our estimation based on the quarter-long daily return series.

¹⁸ We include the lagged dependent variable for the completeness while we recognize that the lagged dependent variable is a noisy variable estimated by the model.

¹⁹ In the augmented setting, the only noticeable change occurs to stock analyst coverage. Interestingly, ANA shows a significant, positive effect in Column (6) while it has no significant association in the baseline analysis. It might be related to a strong negative association between ANA and AMH with the correlation of −0.742, as shown in Table 2.

²⁰ We obtain qualitatively the same result for the 200-day detrended volume-conditioned autocorrelation in unreported analysis.

²¹ We calculate the monthly turnover of underlying stocks in unreported analysis.

²² Da and Shive (2018) find that stocks with high ETF turnover show the price reversal in the daily return beta analysis. We attribute this difference to our research design based on the volume-price movement analysis. We discuss this issue in Section IV.

²³ LMSW also discuss the behaviour of the unconditional autocorrelation.

²⁴ Unlike the results for Columns (3) to (6) of Table 8, Columns (1) and (2) in Table 8 show the same pattern as the volume-conditioned autocorrelation, as shown in Table 5, but they show weaker results.

²⁵ We drop the lagged dependent variable in the panel data analysis because the lagged dependent variable requires a dynamic panel data model. See Holtz-Eakin, Newey, and Rosen (1988) and Arellano and Bond (1991).