ETFs’ two‐sided trading costs and order imbalances

• Published date: 01 May 2022
• Author: Marie‐Eve Lachance
• Date: 01 May 2022
• DOI: http://doi.org/10.1111/fire.12292

DOI: 10.1111/fire.12292

ORIGINAL ARTICLE

ETFs’ two-sided trading costs and order

imbalances

Marie-Eve Lachance

Finance Department, San Diego State

University, San Diego, California, USA

Correspondence

Marie-EveLachance, Finance Department, San

DiegoState University, 5500 Campanile Drive,

SanDiego, CA 92182—8236, USA.

Email:marie.lachance@sdsu.edu

Abstract

This paper proposes an integrated trading cost measure for

exchange-traded funds (ETFs) that includes premiums and

spreads. The measure is computed separately for buyers and

sellers as the distance between the ETF’s bid/ask prices and

a new measure of intraday intrinsic value constructed with

the ETF’s daily holdings. The two-sided measure shows that

order imbalances increase trading costs in the direction of

the imbalance, bias midpoint quotes, and result in observed

premiums and discounts. It also provides a link between

these higher trading costs and the need for authorized par-

ticipants to pass through the costs associated with primary

market activity.

KEYWORDS

exchange-traded funds, intraday analysis, premiums, transaction

costs

JEL CLASSIFICATION

G14, G23

1INTRODUCTION

The exchange-tradedfunds (ETFs) market has grown exponentially in recent years, with assets tallying $5.4 trillion in

the United States (U.S.)by the end of 2020.1ETFs differ from mutual funds because they can trade at prices away from

their net asset values (NAVs).Premiums over NAVs are effectively an additional transaction cost for the buyerand a

rebate for the seller and vice versa for discounts. Yet,premiums and discounts are more often analyzed in the context

of arbitrage opportunities and are portrayed as mispricings or inefficiencies. Regulation also compartmentalizes the

concepts of premiums and spreads and, until recently, required only the disclosure of premiums. This paper adopts a

more integrated view of costs that combines premiums and spreads.

1InvestmentCompany Institute, 2021 Investment Company Factbook, p. 92.

Financial Review. 2022;57:273–294. wileyonlinelibrary.com/journal/fire ©2022 The Eastern Finance Association 273

274 LACHANCE

The paper extends the literature on ETFs in three directions. First, it introduces an intraday measure of intrinsic

value based on the ETF’s daily holdings and applies it to compute intraday premiums and two-sided trading costs.

Specifically,the cost of buying (selling) an ETF is the distance between the ask (bid) price and the intrinsic value. Sec-

ond, it investigates how buying/sellingtrends, as measured by order imbalances, affect the cost of buying and selling

ETFs. Order imbalances have an asymmetric impact on the two costs, and this uneven response biases the midpoint

quotes of ETFs and thus the returns computed with these prices. Third, the paper links the need to pass through the

transaction costs associated with primary marketactivity with higher trading costs on the side of the imbalance.

ETFs’ premiums have been studied in several works, including Ackert and Tian (2000, 2008), Elton et al. (2002),

Engle and Sarkar (2006), Delcoure and Zhong (2007), and Petajisto (2017). ETFs have a special feature whereby

authorized participants (APs) can create or redeem shares with the fund—that is, in the primary markets—through an

exchangeof the basket of underlying securities. Madhavan and Sobczyk (2016) and others argue that this mechanism

enables arbitrage activity that should keep an ETF’s price in line with its NAV. A common thread in the studies cited

above is that domestic equity funds, indeed, havesmall average premiums, but deviations can be more substantial for

other asset classes such as international equities.

This paper focuses on domestic equity funds and thus avoids the asynchronous measurement issues of interna-

tional funds. Within domestic funds, there is scant information on intraday premiums, as most of the academic liter-

ature relies on end-of-day values provided by fund sponsors. One notable exception is Engle and Sarkar (2006), but

their study covers only a portion of the year 2000 and a few ETFs. The sample for this paper’s investigation includes

409 U.S. equity ETFs for the period 2012−2017. This analysis is facilitated by the ETF Global data, which improveon

previous data sets by providing portfolios at a daily frequency instead of the more common monthly or quarterly fre-

quencies. I use these daily portfolios—along with the midpoint quotes of the underlying stocks—to compute an ETF’s

intraday intrinsic value. Using midpoint quotes instead of last-tradeprices alleviates most of the staleness issues that

are problematic with intraday intrinsic values (IIVs).2Thepaper closest to this one in terms of intraday analysis is that

of Box et al. (2021), who use monthly data on ETFs to infer daily holdings and compare the bid and ask prices of the

ETF and its underlying portfolio. Their focus is different from mine, as they analyze the impact—or,more precisely, the

lack thereof—of ETF trading on their underlying securities.

This paper first provides statistics on intraday premiums and two-sided tradingcosts. The average premium is less

than one basis point, and a two-standard-deviations rangearound that mean is under ±10 basis points (bps). The aver-

age costs of buying and selling ETFs are 6.02 and 5.04 bps, respectively.The analysis of intraday patterns reveals that

these costs can be three to four times higher immediately after the open and that they are much more volatile at that

time. The role of premiums and discounts can be better appreciated by computing the average trading costs in each

scenario. The average cost for buying ETFs that trade at a premium is 8.60 bps, compared to only 2.43 bps for selling

them. The gap is similar for funds that trade at a discount: the averagecost is 2.98 bps for buying them versus 8.14 bps

for selling them.

The formulation of the results in the previous paragraph is perhaps misleading because it suggests that premiums

and discounts make ETFs more or less expensive to trade. Instead, this paper considers the idea that premiums and

discounts can arise because the relative costs of buying and selling are different. The two concepts are connected since

premiums are equivalent to half of the difference between the costs of buying and selling. The notion thattrading costs

can be asymmetric or,equivalently, that bid and ask prices are not necessarily centered on the stock’s intrinsic value, is

present in several marketmicrostructure models featuring inventory costs (see, e.g., Amihud & Mendelson, 1980;Ho

& Stoll, 1981; Stoll, 1978).3For instance, Stoll (1989, p. 116) explains, “If the spread reflects inventoryholding costs,

2IIVsare typically reported every 15 s during the day. Madhavan and Sobczyk (2016) mention that these values are based on possibly stale last-trade prices

andare not useful for APs, which compute their own values using midpoint quotes. The shortcomings of IIVs have led the Securities and Exchange Commission

toremove the dissemination requirement from its new ETF rule.

3Adverse selection issues can also affect how the bid and ask quotes are selected. Fora more complete review of models of the bid-ask spread, see Huang

andStoll (1997).