Flat

https://doi.org/10.
1017/S0022109021000387 Published online by Cambridge University Press

JOURNAL OF FINANCIAL AND QUANTITATIVE ANALYSIS Vol. 56, No. 7, Nov. 2021, pp. 2356–2388
© THE AUTHOR(S), 2021. PUBLISHED BY CAMBRIDGE UNIVERSITY PRESS ON BEHALF OF THE MICHAEL G. FOSTER
SCHOOL OF BUSINESS, UNIVERSITY OF WASHINGTON
doi:10.1017/S0022109021000387
Flattening the Illiquidity Curve: Retail Trading

During the COVID-19 Lockdown
Gideon Ozik
EDHEC Business School
[email protected]
Ronnie Sadka
Boston College Carroll School of Management
[email protected] (corresponding author)
Siyi Shen
Chinese University of Hong Kong, Shenzhen School of Management and Economics
[email protected]
Abstract
This article studies the impact of retail investors on stock liquidity during the COVID-19
pandemic lockdown in spring 2020. Retail trading exhibits a sharp increase, especially
among stocks with high COVID-19–related media coverage. Retail trading attenuated
the rise in illiquidity by roughly 40% but less so for high-media-attention stocks. Causality
is addressed using the staggered implementation of the stay-at-home advisory across
U.S. states. The results highlight that ample free time and access to financial markets
facilitated by fintech innovations to trading platforms are significant determinants of retail-
investor stock market participation.
“New investors […] sense a generational-buying moment […] We have

heard anecdotally about younger individuals with less market experience
viewing the March plunge as a unique time to start portfolios.”
—Citi chief U.S. equity strategist Tobias Levkovich in a note to clients in
May, reported by CNBC (June 9, 2020, “Robinhood Traders Cash in on
the Market Comeback That Billionaire Investors Missed”)
Ozik and Sadka are affiliated with MKT MediaStats, LLC. For helpful comments and suggestions,
we thank Philip Bond, Ran Duchin (editor), Slava Fos, Thierry Foucault, Jarrad Harford (editor), Zhiguo
He, David Hirshleifer, Mark Kamstra, Stefan Nagel, Lin Peng, Alessandro Rebucci, Alexi Savov, Amit
Seru, Kelly Shue, Phil Strahan, Dick Syron, Pierre-Olivier Weill, Wei Xiong, and Liyan Yang, as well as
seminar and conference participants at Hebrew University, Stockholm Business School, Tel Aviv
University, the University of South Carolina, the U.S. Securities and Exchange Commission, the
2020 Journal of Finance/Fama–Miller Center Conference on the Financial Consequences of the
COVID-19 Pandemic, the 2021 Australasian Finance and Banking Conference, State Street Associates,
the Goldman Sachs QES Academic Research Club, the 2021 Midwest Finance Association (MFA)
Annual Conference, and the 2021 Journal of Financial and Quantitative Analysis COVID-19 Sympo-
sium. The views expressed are solely those of the authors.
2356
https://doi.org/10.1017/S0022109021000387 Published online by Cambridge University Press
Ozik, Sadka, and Shen 2357
I. Introduction
The COVID-19 pandemic forced unprecedented challenges in all aspects of
our lives. What is the aftermath of this pandemic for the stock market? Investors
cumulatively pulled more than $150 billion from U.S. domestic equity funds over
the first 5 months of 2020, based on estimated flow reports from the Investment
Company Institute (ICI).1 Although institutions posted record capital outflows and
the Federal Reserve had not directly injected liquidity to stock markets, retail
trading took off amid the COVID-19 pandemic and major brokerage firms saw
record new accounts in the first half of 2020. A fintech trading app, Robinhood, the
trading accounts of which underline the main conclusions of this article, saw a
record 3 million new accounts open within the first quarter of the year, with 3 times
its average trading volume compared to 2019.2
The surge in retail trading was largely made possible due to the recent wave of
fintech innovations in the retail-brokerage space. In the past year, to compete with
fintech trading apps like Robinhood, which provide low-cost stock trading, legacy
brokerage houses, such as Charles Schwab, Fidelity, TD Ameritrade, and E-Trade
Financial, started to offer low commissions and one-stop-shop financial apps
accessible on investors’ smartphones. As a result, E-Trade reported growth of
roughly 900,000 net new self-directed accounts from the second to the fourth
quarter of 2020, and Charles Schwab counted nearly 30 million active brokerage
accounts at the end of 2020.3 Using SimilarWeb, a popular online marketing
analysis tool, we assess the growth in online activity of retail brokers. Focusing
on total daily visits (mobile and desktop) for the 5 aforementioned retail brokers, we
find that in 2020:Q2, Robinhood recorded a 115% quarter-over-quarter increase in
average daily visits. The 4 legacy brokers, TD Ameritrade, E-Trade, Fidelity, and
Schwab, also recorded sharp increases of 52%, 46%, 38%, and 32%, respectively.
Market makers have stood to benefit from the surging volume in retail trading. For
example, Bloomberg reported that Citadel Securities estimates that retail trades
accounted for approximately 25% of the stock transactions on the most active days
during the pandemic and that they have handled approximately 40% of equity retail
trades.4 Although retail trading activity has clearly represented a growing portion of
stock transactions in the recent period, the implications of such activity to the stock
market amid the COVID-19 pandemic are yet unknown.
Motivated by the aforementioned observations, this article studies the trading
behavior of retail investors and its implications during the pandemic. Here is the
story in a nutshell: With high volatility and low liquidity, financial markets entered a
panic mode in Mar. 2020. Then, a lockdown advisory was put in place across most
of the United States (mobility indicators provided by Apple and Google confirm
1
For more details, see https://www.ici.org/research/stats/flows.
2
For more details, see https://www.cnbc.com/2020/06/17/robinhood-drives-retail-trading-
renaissance-during-markets-wild-ride.html and https://www.wsj.com/articles/everyones-a-day-
trader-now-11595649609.
3
For more details, see https://www.wsj.com/articles/trading-surge-strains-online-brokerages-
11611692363.
4
For more details, see https://www.bloomberg.com/news/articles/2020-07-09/citadel-securities-says-
retail-is-25-of-the-market-during-peaks.
2358 Journal of Financial and Quantitative Analysis
a significant drop in mobility starting around Mar. 15). With much of the country
(and the world) under stay-at-home-advisory mandates and live sporting broadcasts
and entertainment events canceled, many people were confined at home with an
abundance of free time. How did they respond? By directing their attention to the
alarming statistics of COVID-19 infections, hospitalizations, and deaths and to the
stock market. Increased savings (Li, Strahan, and Zhang (2020)) and the availability
of fintech trading apps, conveniently accessed through mobile devices, led to a
significant increase in retail stock market participation and trading activity through-
out the lockdown period.
We find that while overall liquidity deteriorated during lockdown, the increase
in retail trading activity improved it, lowering stock bid–ask spreads and the price
impact of trades. The difference in average effective spread between the low and high
deciles of stocks sorted by retail trading activity (23 basis points (bps)) is roughly 40%
of the average level of effective spread during lockdown (60 bps). These results are
consistent with prior evidence, for example, utilizing data on French retail investors’
trading. Barrot, Kaniel, and Sraer (2016) show that individual investors tend to supply
liquidity when institutional liquidity dries up, as during the financial crisis of 2008–
2009 (using the same data, Foucault, Sraer, and Thesmar (2011) show that individual
investors tend to decrease stock volatility and the price impact of trades). However,
the results in this article suggest a far larger impact of retail trading during the recent
pandemic. Additionally, and in contrast to the general behavior, we find that retail
trading seems to have a significantly lower impact on high-media-attention stocks,
which we further discuss later in the article. When states started to reopen in early
May, resulting in an increase in mobility, the rate of increase in retail trades attenuated
and, in turn, their liquidity provision.
Time-series plots of equity price levels and aggregate illiquidity during the
2008 financial crisis and the recent pandemic provide further motivation. Graph A
of Figure 1 plots the cumulative returns for the Standard & Poor’s (S&P) 500 index
(SPY) and the average effective spread. The average effective spread displays
elevated levels for the period of mid-September to mid-December 2008, with
multiple spikes over that period (the largest on Sept. 19, 2008). Notice that the
most significant drop in liquidity is observed before the strong declines in asset
prices in early Oct. 2008. In contrast, market illiquidity exhibits a deterioration
during Mar. 2020, with a single significant jump on Mar. 20, 2020 (following
market decline since prior high on Feb. 19, 2020).
Four additional series are displayed in Figure 2 (not available for the financial
crisis period): Apple’s U.S. driving mobility trend index;5 the intensity of COVID-19
coverage, estimated as the average fraction of COVID-19–related media articles to
all media articles per stock; the average number of Robinhood trading accounts
per stock (in hundreds); and monthly estimated U.S. domestic equity fund net
flow (in $billions) from the ICI.6 Although elevated levels of effective bid–ask
spread are noticeable since the end of Feb. 2020, illiquidity peaks with a single spike
on Mar. 20, 2020, well after the market dropped by more than 25%. Significant
declines in U.S. driving mobility occurred on Mar. 15, to a score of 76.16 (from the
5
For more details, see https://www.apple.com/COVID19/mobility.
6
For more details, see https://www.ici.org/.
FIGURE 1
Price and Illiquidity
Figure 1 shows the U.S. stock prices (scaled to start at 100) and average daily effective spreads surrounding the 2008
financial crisis (Graph A) and 2020 COVID-19 pandemic (Graph B).
Graph A. 2008 Financial Crisis

120 3
100 2.5
Effective Spread (%)

80 2
Price
60 1.5
40 1
20 0.5
0 0
8/1/08
8/8/08
8/15/08
8/22/08
8/29/08
9/8/08
9/15/08
9/22/08
9/29/08
10/6/08
10/13/08
10/20/08
10/27/08
11/3/08
11/10/08
11/17/08
11/24/08
12/2/08
12/9/08
12/16/08
12/23/08
12/31/08
1/8/09
1/15/09
1/23/09
1/30/09
2/6/09
2/13/09
2/23/09
3/2/09
3/9/09
3/16/09
3/23/09
3/30/09
Effective Spread (%) Price (scaled)
Graph B. 2020 COVID-19 Pandemic

120 3
100 2.5
Effective Spread (%)

80 2
Price
60 1.5
40 1
20 0.5
0 0
1/21/20
1/28/20
2/4/20
2/11/20
2/19/20
2/26/20
3/4/20
3/11/20
3/18/20
3/25/20
4/1/20
4/8/20
4/16/20
4/23/20
4/30/20
5/7/20
5/18/20
5/26/20
6/2/20
6/9/20
Effective Spread (%) Price (scaled)
previous day’s score of 102.87), recovering over 90% of its pre-COVID-19 level
only by May 8, 2020. (Therefore, we identify the lockdown period as Mar.
16 through May 7, 2020.) Retail trading accounts on Robinhood display an increas-
ing time trend since Jan. 2020, with an accelerated rate since early March. The
average COVID-19 media coverage rate per stock increased from 29% to 72% over
the period from mid-February to the end of March.
The liquidity shock in 2008 lasted for several months, whereas the one during
the recent pandemic seems more short-lived. Although one may postulate that
FIGURE 2
Mobility, COVID-19–Related Media Coverage, and Retail and
Institutional Flows During the 2020 COVID-19 Pandemic
Graph A of Figure 2 reports the daily U.S. driving mobility index published by Apple (https://www.apple.com/COVID19/
mobility) and 7-day moving average of the daily fraction of COVID-19–related articles to total media coverage per stock (in %).
Graph B reports the daily average number of Robinhood trading accounts and monthly estimated U.S. domestic equity fund
cumulative net flow (in $billions) from the Investment Company Institute (https://www.ici.org/).
Graph A. Mobility and COVID-19 Media Coverage

150
Mobility / Med Coverage (%) [7-day MA]
125
100
75
50
25
0
1/21/20
1/24/20
1/29/20
2/3/20
2/6/20
2/11/20
2/14/20
2/20/20
2/25/20
2/28/20
3/4/20
3/9/20
3/12/20
3/17/20
3/20/20
3/25/20
3/30/20
4/2/20
4/7/20
4/13/20
4/16/20
4/21/20
4/24/20
4/29/20
5/4/20
5/7/20
5/14/20
5/19/20
5/22/20
5/28/20
6/2/20
6/5/20
6/10/20
Mobility Lockdown Mobility COVID-19 Media Coverage
Graph B. Retail User Counts and Equity Fund

9,000 0
US Equity Fund Flow (cumulative net; $B)

8,000 –40
7,000 –80
Retail Accounts
6,000 –120
5,000 –160
4,000 –200
3,000 –240
1/21/20
1/24/20
1/29/20
2/3/20
2/6/20
2/11/20
2/14/20
2/20/20
2/25/20
2/28/20
3/4/20
3/9/20
3/12/20
3/17/20
3/20/20
3/25/20
3/30/20
4/2/20
4/7/20
4/13/20
4/16/20
4/21/20
4/24/20
4/29/20
5/4/20
5/7/20
5/14/20
5/19/20
5/22/20
5/28/20
6/2/20
6/5/20
6/10/20
US Equity Fund Flow ($B) Average Retail Accounts Per Stock
the Federal Reserve’s liquidity-injection programs indirectly transmitted to equity

markets,7 we argue that it is the significant increase in retail trading activity and the
decrease in mobility during the lockdown period that have contributed to “flattening
7
These include the unscheduled Federal Open Market Committee (FOMC) meetings followed by
rate-cut announcements on Mar. 3 and Mar. 15, 2020, as well as the Federal Reserve announcement to
buy corporate bonds on Mar. 23, 2020.
the illiquidity curve.”8 We therefore advance that recent fintech innovations to

trading platforms ease retail traders’ access to equity markets, allowing them to
provide liquidity in times of stress while reducing the need for further government
intervention.
We further study the role of the media insofar as explaining retail trading
activity. Given the evidence in Barber and Odean (2008) that retail investors tend
to trade attention-grabbing stocks, we also focus on stocks mentioned by the
media, specifically in the context of COVID-19. We find that during the pan-
demic, retail investors tend to trade these stocks above average and that this
“media-attention-driven” trading results in less increase in liquidity than aver-
age. That is, although retail investors tend to act as liquidity providers overall
during the pandemic, they seem to do so less when their trading activity is moti-
vated by chasing firms under the spotlight in the context of COVID-19. Consistent
with our finding, Eaton, Green, Roseman, and Wu (2021) use Reddit/WallStreet-
Bets mentions as a proxy for stocks receiving high attention from retail traders, and
they find that Robinhood app outages are associated with improved liquidity for
those stocks. They further show that such attention-driven retail trades are more
likely to herd and persist; thus, market makers may find it more difficult to unload
inventory risk. In addition, high-frequency traders (HFTs) may obtain information
about retail order flow and, in turn, increase adverse selection for other (unin-
formed) market makers. Also, along the lines of the evidence provided by von
Beschwitz, Keim, and Massa (2020), news analytics of media coverage ignite
algorithmic trading, and although they tend to speed up stock price and trading
volume in response to articles, they also reduce liquidity.9 When states began to
reopen, this media-driven liquidity demand by retail investors decreased.
Common time-series trends may be subject to endogeneity concerns. For
example, the Federal Reserve announcements of a rate cut and a liquidity-injection
program for the bond market coincide with the time of significant drops in the
mobility index. These announcements may simultaneously increase retail trading
and liquidity or even lead to a reverse causality during lockdown (i.e., improved
stock market liquidity leading to more retail trading).
To verify our findings are indeed causal, we use an identification strategy that
utilizes the staggered implementations of stay-at-home advisory across U.S. states,
which is likely independent of financial market conditions. Ideally, in a perfect
setting, the stay-at-home mandates would serve as a shock to retail investors’
mobility based on their geographic location, but investor location data are unavail-
able to us. To overcome this caveat, we provide two types of tests. First, we rely on
the well-documented home bias in stock investment (e.g., Coval and Moskowitz
8
Unreported results show that daily changes in stock market effective bid–ask spreads regressed on
lagged changes in the TED spread, lagged changes in credit spread, and lagged changes in the number of
Robinhood users over the period of Jan. 21 through May 7, 2020, produce a significant coefficient only
for the latter variable (negative).
9
This evidence complements that of Peress and Schmidt (2020), who show that market liquidity
drops when retail investors are distracted by non–stock-market-related news. That is, stocks that are
mentioned in the context of a major event, such as COVID-19, may experience a drop in liquidity as well.
Also related is an article by Lou (2014), who documents that increased firm advertising spending is
associated with a rise in retail trading (see also Fang, Madsen, and Shao (2020)).
(1999), Ivković and Weisbenner (2005)). Specifically, we use a firm’s headquarters

location as a crude proxy for household location. Despite being a noisy proxy, any
finding based on it can be viewed as a lower bound of the true effect. Our difference-
in-differences (DID) analysis confirms that as a result of the mobility shock, the
(negative) effect of (attention-driven) retail trading on liquidity provision is signif-
icantly larger for treated firms relative to control firms. Second, we proxy for the
retail-investor daily activity of a given stock based on the stock’s Google Trends
search-volume index (Da, Engelberg, and Gao (2011), Ben-Rephael, Da, and
Israelsen (2017)) measured at the state level. We show that the stock ticker searches
of a given firm during a state lockdown are associated with improved stock
liquidity, and more so for searches originating from the state in which that firm’s
headquarters is located.
Why does retail trading improve stock liquidity? Decomposing effective
spread into a (variable) price-impact component and a (fixed) realized-spread
component, we find that although retail trading improves both components, the
relative impact on the price impact is higher. Given that the price-impact component
is inversely related to noise-trading activity (e.g., Kyle (1985)), it follows that retail
traders improve stock liquidity because they act as noise traders rather than
informed investors. We also find significant insider-trading activity for stocks with
elevated levels of retail trading during lockdown, consistent with Collin-Dufresne
and Fos (2015), (2016), who suggest that corporate insiders advantageously time
liquidity in the presence of uninformed retail trading.
Expanding the analysis to stock returns, we find that although retail investors
act as short-term momentum traders, who, on average, tend to chase stocks that
perform well over the prior week during lockdown, their activity does not seem to
significantly affect contemporaneous stock returns. However, retail trading of poor-
performing stocks over the prior week is consistent with a demand for liquidity
relative to well-performing stocks. Finally, we demonstrate that our main results
remain largely unchanged under robustness tests using alternative liquidity proxies,
choices of reopening date, and model specifications.
The main contributions of this article are summarized as follows: First, we
demonstrate that retail trading has a first-order effect on financial markets, as
investors step in and act as liquidity providers during the pandemic lockdown,
potentially alleviating the need for further government intervention. The article
highlights the key role of recent fintech innovations to trading platforms, less
prevalent during the financial crisis of 2008, in weathering illiquidity shocks.
In particular, the ease with which users can access the stock market via trading
platforms that feature low commissions and trading costs has allowed for a signif-
icant increase in stock market participation by retail investors. Recent data extracted
from SimilarWeb suggest that the level of online activity on the 5 aforementioned
retail brokers’ websites (TD Ameritrade, E-Trade, Fidelity, Charles Schwab, and
Robinhood) during the first 2 months of 2021 increased by over 80% compared to
its level during 2020:Q4. We therefore believe retail trading will continue to exhibit
a significant impact on financial markets moving forward.
Second, the article contributes to the larger literature that studies retail trading.
Although some articles shed light on stock characteristics that may drive retail
trading, such as glamour stocks, momentum stocks, and high-media-coverage
stocks, the relatively low market participation rate of individual investors has
remained a puzzle. Some articles point to fixed participation costs as a possible
explanation (e.g., Vissing-Jørgensen (2003), Campbell (2006)), where investor
cognitive skills (Grinblatt, Keloharju, and Linnainmaa (2011)), financial literacy
(Van Rooij, Lusardi, and Alessie (2011)), and risk aversion (Haliassos and Bertaut
(1995)) are offered as the main factors that determine the magnitude of such
participation costs. This article offers yet another explanation for the low partici-
pation rate: the lack of free time. During lockdown, with ample free time on their
hands, retail investors significantly increased their stock market participation.
Although we utilize unique data from the Robinhood trading platform to demon-
strate the patterns in retail trading over the pandemic, we view our results as the
lower bounds to more general behavior. Indeed, during lockdown, we find a
significant increase in Google searches of other popular retail trading platforms
(e.g., TD Ameritrade, E-Trade, Fidelity, and Charles Schwab).
The rest of the article is organized as follows: Section II describes the data and
the construction of the main variables. In Section III, we examine the trading
behavior of retail investors throughout the pandemic. Section IV presents an
analysis of the relation between retail trading and liquidity, as well as additional
evidence on the role of media. Section V provides additional tests and robustness
checks. A discussion of the importance of retail trading is offered in Section VI.
Section VII concludes.
II. Data and Sample

This section describes the data and sample, defines the main variables, and
provides descriptive statistics for the sample.
A. Retail User Accounts

To measure retail trading activity for a given stock, we use hourly snapshots of
Robinhood popularity metrics, which represent the number of unique Robinhood
user accounts holding at least one share of the stock. We are grateful to Robinhood-
Track.net, a website that downloads hourly snapshots from Robinhood through an
application programming interface (API) and makes all historical snapshots avail-
able for download on the website. To align with the frequency of the other variables
in our article, we use the data snapshot of the last available hour in a given trading
day as the number of unique Robinhood user accounts holding each stock each day.
B. COVID-19–Related Media Coverage
To estimate firm-level COVID-19 media-coverage intensity, we rely on data

provided by MKT MediaStats, an alternative data company that maintains multiple
information reservoirs, including media coverage pertaining to companies. MKT
MediaStats collects information from roughly 100,000 distinct U.S. and interna-
tional media sources, amounting to approximately 1.5 million articles per week
across these reservoirs. COVID-19 media intensity for a given firm is measured as
the fraction of media articles that mention COVID-19 relative to the total number of
media articles mentioning the firm. The media data cover the largest 3,000
U.S. stocks included in the Russell 3000 index.
C. Mobility Trends in the United States

Since Jan. 13, 2020, Apple has started publishing daily mobility trends by
counting the number of requests made to Apple Maps for directions in each location
for U.S. states and major cities. We rely on the daily U.S. driving mobility index to
identify the effective dates of lockdown and economic reopening amid the
COVID-19 pandemic.
Graph A of Figure 2 shows significant declines in U.S. driving mobility on
Mar. 15, to a score of 76.16 (from the previous day’s score of 102.87), and a further
drop to its lowest level, at 37.42, on Apr. 12.10 The mobility index recovered to over
90% of its pre–COVID-19 level only by May 8. Based on the mobility pattern, we
identify the lockdown period as ranging between Mar. 16 and May 7 and the reopen
period as beginning May 8.11
D. Liquidity Measures
We obtain daily liquidity measures from Wharton Research Data Services
(WRDS) Intraday Indicators constructed by using the daily Trade and Quote
database (DTAQ), which utilizes intradaily data of trades and quotes, signs trades
using Lee and Ready (1991), and applies the filters and adjustments described by
Holden and Jacobsen (2014).12
Quoted and effective spread are the two main measures of stock liquidity
employed in this study. The daily average quoted spread for each stock i on day t is
calculated as follows:
1X T
Ai,s Bi,s
QSPREADi,t ¼ ,
T s¼1 M i,s
where Ai,s is the National Best Ask, Bi,s is the National Best Bid, and M i,s is the
midpoint (i.e., the average of Ai,s and Bi,s ) assigned to time interval s for firm i. For a
given stock i, the daily average percent effective spread is defined as follows:
1X N
2Di,k ðPi,k M i,k Þ
ESPREADi,t ¼ ,
N k¼1 M i,k
where Di,k is equal to +1 for buyer-initiated trades and –1 for seller-initiated trades
using the Lee and Ready (1991) algorithm, Pi,k is the price of the kth trade, and M i,k
10
The baseline pre–COVID-19 mobility level equals 100 as of Jan. 13, when Apple started pub-
lishing the mobility index.
11
In robustness tests, we show that overall findings remain largely unchanged if we identify the
reopen date as May 1 or May 15, when the U.S. mobility score recovered to its 80% or 100% pre–
COVID-19 level, respectively.
12
The code for making these adjustments is available on Craig Holden’s Web page (http://kelley.iu.edu/
cholden/).
is the midpoint of the National Best Bid and Offer (NBBO) quotes assigned to the
kth trade.
In an extended analysis, we further examine the 2 components of effective
spread: price impact and realized spread. For a given stock i, the daily average
percent price impact is computed as follows:
1X N
2Di,k ðM i,kþ5 M i,k Þ
PIMPACTi,t ¼ ,
N k¼1 M i,k
where M i,k is the midpoint of the NBBO quotes assigned to the kth trade, and M i,kþ5
is the midpoint of the NBBO prevailing 5 minutes after the M i,k . For a given stock i,
the daily average percent realized spread is computed as follows:
1X N
2Di,k ðPi,k M i,kþ5 Þ
RSPREADi,t ¼ :
N k¼1 M i,k
The price impact can be viewed as a permanent component of the effective

spread, whereas the realized spread is a measure of revenue to market makers that
nets out losses to better-informed traders; thus, it is a temporary component of the
effective spread. Furthermore, volatility for stock i on day t is calculated as follows:
2
X T
RETi,j RETi,j
VOLATILITYi,t ¼ :
j¼1
T 1
E. Other Data and Summary Statistics
Throughout our analyses, we focus on common stocks (share codes 10 or 11)

listed on the New York Stock Exchange (NYSE), American Stock Exchange (AMEX),
or NASDAQ (exchange codes 1, 2, or 3), and exclude small stocks (closing price ≤ $5
as of Dec. 31, 2019). We obtain daily stock returns from Thomson-Reuters for the
period from Jan. 21, 2020, through June 11, 2020. We retrieve institutional holding
information from the U.S. Securities and Exchange Commission (SEC) 13F filings
compiled by Thomson-Reuters. Since 1978, all institutional investment managers that
have investment discretion of over $100 million or more in Section 13(f) securities
(mostly publicly traded equity) are required to disclose their quarter-end holdings in
these securities. This filing requirement applies to equity positions of greater than
10,000 shares or with a fair market value of at least $200,000. For each stock, we
calculate firm size and the level of institutional ownership at the end of year 2019.
After merging the data from all sources, our final sample consists of 100 trad-
ing days with 2,265 unique stocks for the period from Jan. 21 through June 11,
2020. Table 1 provides summary statistics of the main variables. Our sample
contains the largest U.S. stocks, with a market capitalization of $13.1 billion on
average. In addition, the average daily quoted and effective spread are 0.553%
and 0.258%, respectively. The average price impact is 0.157%, indicating that
the permanent component of the efficient spread is more than 50% larger than
its temporary component (i.e., realized spread) at 0.097%. On a given day, a firm
on average is held by 5,145 unique Robinhood trading accounts. The mean
TABLE 1
Summary Statistics
Table 1 reports the summary statistics of the main variables. Firm-level variables include log firm size measured at the end of
year 2019, daily return (RET), past-week stock returns (PRET), daily time-weighted percent quoted spread (QSPREAD), daily
average percent effective spread (ESPREAD), price impact (PIMPACT), and realized spread (RSPREAD) based on the Lee
and Ready (1991) trade classification; daily stock VOLATILITY; daily log number of Robinhood trading accounts for each
stock (RETAIL); and the fraction of a firm’s daily COVID-19–related media coverage to the firm’s total daily media coverage for
each firm (COVERAGE). Daily liquidity measures are from Wharton Research Data Services (WRDS) Intraday Indicators using
the daily Trade and Quote database (DTAQ). The sample is from Jan. 21, 2020, through June 11, 2020.
Variable No. of Obs. Mean Std. Dev. P25 P50 P75
SIZE (log) 2,265 21.574 1.688 20.287 21.421 22.572

RET (%) 226,014 0.096 5.424 2.729 0.098 2.393
PRET (%) 226,014 0.246 11.535 5.778 0.141 5.110
QSPREAD (%) 225,948 0.553 0.868 0.125 0.261 0.568
ESPREAD (%) 225,910 0.258 0.390 0.065 0.123 0.260
PIMPACT (%) 225,726 0.157 0.194 0.044 0.088 0.186
RSPREAD (%) 225,755 0.097 0.261 0.005 0.026 0.075
VOLATILITY (106) 225,962 8.675 27.868 0.278 0.885 3.459
RETAIL (log) 226,015 6.191 1.889 4.820 6.059 7.385
COVERAGE 226,015 0.173 0.378 0.000 0.000 0.000
COVID-19–related media coverage ratio is approximately 17.3%, but almost 90%

of firm-day observations do not have any COVID-19–related media coverage.
III. Retail Trading During the COVID-19 Pandemic

To motivate our study, we start by examining patterns of retail investors’
trading activity during the COVID-19 pandemic. Graph B of Figure 2 plots the
average number of Robinhood trading accounts per stock each day. The graph
displays an overall increasing interest in directly participating in the stock market
from retail investors since Jan. 2020, with an accelerated rate since early March.
A firm on average was held by 3,060 unique accounts on Jan. 21, and this number
rose to 3,708 around Mar. 15. Moreover, the first week of lockdown experienced
a 14.3% increase in retail trading, reaching an average of 4,280 trading accounts
per stock. Despite being at a lower speed, the stock market participation from the
retail investors continued to soar. In contrast, as estimated by the ICI, the monthly
U.S. equity funds experienced historical capital outflows, amounting to a cumula-
tive loss of $150 billion over the 5-month period.
A. Retail Trading and COVID-19–Related Media Coverage

Extending the patterns depicted in Figures 1 and 2, we now study the behavior
of retail trading during the pandemic using regression analysis. We first examine
whether retail trading strongly responds to COVID-19–related media coverage
using the following OLS model:
(1) RETAILi,t ¼ αi þ β COVERAGEi,t þ γ CONTROLS þ ϵi,t ,
where RETAILi,t is the log number of unique Robinhood accounts holding stock i
at day t, and COVERAGEi,t is a dummy variable equal to 1 if firm i’s COVID-19–
related media-coverage ratio at day t is greater than 0, and 0 otherwise. We include
the past-week returns of firm i to control for the tendency of retailers to buy stocks
exhibiting extreme returns, as documented by Odean (1999) and Barber and Odean
(2008). In all regressions henceforth, unless otherwise specified, we add firm fixed
effects to control for firm-level heterogeneity and cluster standard errors by firm and
by trading days.13
Panel A of Table 2 reports the regression results of equation (1) over the entire
sample period. The coefficient estimate on COVERAGEi,t is positive and signif-
icant at the 1% level. In terms of economic significance, a stock with COVID-19–
related coverage is associated with a 3.07% increase in the log number of retail
accounts (relative to the sample mean of 6.19). These findings suggest that retail
investors tend to trade attention-grabbing stocks, which is consistent with Welch
(2021) and Barber, Huang, Odean, and Schwarz (2021), who document that Robin-
hood investors increase their holdings on stocks experiencing large price changes.
Prior evidence shows that retail investors’ attention can be caught by news (Barber
and Odean (2008)), by media coverage (Engelberg and Parsons (2011)), and by
corporate advertisements (Fang et al. (2020)). It is worth noting that the coefficient
on past-week returns is positive and significant at the 1% level, suggesting that retail
traders tend to chase stocks that have performed well over the prior week.
B. Attention-Driven Retail Trading During Lockdown
The findings shown in Section III.A indicate that retail trading significantly
corresponds to media coverage during the sample period. In addition, we conjecture
that attention-driven trading from retail investors will be more pervasive during
lockdown because COVID-19–related media coverage that attracts investors’
attention increased substantially since early March.
To explore this conjecture, we divide the sample into 3 phases. Phase 1 is the
normal period from Jan. 21 to Mar. 13, phase 2 is the lockdown period from
Mar. 16 to May 7, and phase 3 is the reopen period from May 8 onward. Note that
we utilize a pairwise-phase-comparison framework throughout this article because
it allows us to clearly identify the transition of retail trading and liquidity evolve-
ment between consecutive phases. Specifically, we modify the baseline model in
equation (1) to run the following OLS models:
(2) RETAILi,t ¼ αi þ β1 COVERAGEi,t þ β2 LOCKDOWNt

þ β3 COVERAGEi,t LOCKDOWNt
þ γ CONTROLS þ ϵi,t ,
(3) RETAILi,t ¼ αi þ β1 COVERAGEi,t þ β2 REOPENt

þ β3 COVERAGEi,t REOPENt
þ γ CONTROLS þ ϵi,t ,
13
In robustness tests, we show that adding day fixed effects does not change the main results of the
article, except that the lockdown and reopen dummies are subsumed. We discuss this further in
Section IV.D (and Table IA.7 in the Supplementary Material).
TABLE 2
Retail Investors During the COVID-19 Pandemic
Table 2 reports the OLS regression results of the log number of retail trading accounts on the contemporaneous ratio of
COVID-19–related media coverage for the sample from Jan. 21, 2020, through June 11, 2020. The dependent variable is the
daily log number of Robinhood trading accounts for each firm. Results based on the entire sample period, normal and
lockdown periods, and lockdown and reopen periods are reported in Panels A, B, and C, respectively. LOCKDOWN is a
dummy variable equal to 1 between Mar. 16 and May 7. REOPEN is a dummy variable equal to 1 since May 8. Lockdown and
reopening dates are identified based on the U.S. driving mobility index published by Apple (https://www.apple.com/
COVID19/mobility). COVERAGE is a dummy variable equal to 1 if the fraction of a firm’s daily COVID-19–related articles to
its total daily media coverage is greater than 0, and 0 otherwise. All regression models include past-week returns (PRET) and
firm fixed effects (FE). The t-statistics reported in parentheses are based on standard errors clustered at the firm and day
levels. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. Bold coefficients highlight the
results of interest.
Dependent Variable = ln(NO_OF_USER_ACCOUNTS) 1
Panel A. Entire Period
COVERAGE 0.190***
(9.44)
PRET 0.924***
(7.35)
Firm FE Yes
N 226,014
Adj. R2 0.954
1 2
Panel B. Normal Versus Lockdown
COVERAGE 0.015
(1.06)
LOCKDOWN 0.371*** 0.347***
(14.98) (14.50)
COVERAGE LOCKDOWN 0.103***
(5.04)
PRET 0.329*** 0.332***
(3.20) (3.24)
Firm FE Yes Yes
N 171,831 171,831
Adj. R2 0.975 0.975
Panel C. Lockdown Versus Reopen
COVERAGE 0.031***
(4.21)
REOPEN 0.279*** 0.283***
(12.70) (13.20)
COVERAGE REOPEN 0.018*
(1.84)
PRET 0.396*** 0.396***
(3.65) (3.65)
Firm FE Yes Yes
N 140,066 140,066
Adj. R2 0.985 0.985
where LOCKDOWNt is a dummy variable equal to 1 in the lockdown period and

0 in the normal period, and REOPENt is a dummy variable equal to 1 in the reopen
period and 0 in the lockdown period.
Column 2 of Panel B in Table 2 reports the regression results of equation (2).
The coefficient estimate on LOCKDOWNt is positive and significant, indicating
that the log number of Robinhood trading accounts is 34.7% larger during lock-
down than during the normal period. The variable COVERAGEi,t carries an insig-
nificant coefficient estimate, suggesting that COVID-19–related media coverage
does not stimulate retail trading during the normal period. In contrast, the interac-
tion term COVERAGEi,t LOCKDOWNt has a positive and significant coeffi-
cient, confirming that attention-driven retail trading is prevalent during lockdown.
As for economic significance, stocks with COVID-19–related media coverage are
associated with 0.103 more retail trading during lockdown, which translates into
an increase of 10.8% in the number of Robinhood trading accounts. Equation (3)
examines the retail trading activities during the reopen period. As reported in
column 1 of Panel C in Table 2, the coefficient estimate of 0.279 on REOPENt
indicates that retail trading is roughly 32% higher compared to that during lock-
down. However, the coefficient of COVERAGEi,t REOPENt is negative, sug-
gesting that when mobility increased as most states started to reopen in early May,
the increase in attention-driven retail trading was significantly attenuated.
The collective evidence reported in Table 2 indicates that although retail
trading keeps surging over the entire sample period, the attention-driven (as proxied
by the intensity of COVID-19–related media coverage) stock trading is largely
pronounced only during lockdown. In Section IV, we examine the effect of
(attention-driven) retail trading on weathering stock liquidity shocks.
IV. Retail Trading and Stock Liquidity

Prior literature documents that individual investors tend to supply liquidity
when institutional liquidity dries up, as during the financial crisis of 2008–2009,
and tend to decrease stock volatility and the price impact of trades (e.g., Foucault
et al. (2011), Barrot et al. (2016)).
As shown in Figure 1, although the uncertainty was reflected in financial
markets, with a sharp increase in volatility for the recent COVID-19 pandemic,
the elevated levels of effective spread are noticeable since the end of Feb. 2020, and
illiquidity peaks with a single spike on Mar. 20, 2020, well after the market dropped
by more than 25%. Given significant increases in retail-investor trading activity, we
hypothesize that retail trading significantly contributes to dampening illiquidity
during the pandemic.
A. Overall Retail Trading: Baseline Analysis
To test our hypothesis, we study the effect of retail trading on stock liquidity by
estimating the following model:
(4) SPREADi,t ¼ αi þ β1 RETAILi,t þ β2 LOCKDOWNt
þ β3 RETAILi,t LOCKDOWNt þ γ CONTROLS þ ϵi,t ,
where RETAILi,t is the log number of unique Robinhood trading accounts for stock
i at day t, and LOCKDOWNt is a dummy variable equal to 1 during lockdown and
0 in the normal period. The dependent variable SPREADi,t is either the quoted
spread or the effective spread. For brevity, throughout the article, we predominantly
discuss results using the effective spread as the outcome variable, but all the
findings hold when using the quoted spread.
Panel B of Table 3 reports the results of estimating equation (4). The coef-
ficient estimate on LOCKDOWNt is positive and significant at the 1% level
TABLE 3
Retail Investors and Illiquidity During Lockdown
Table 3 reports the OLS regression results of illiquidity measures on the number of retail trading accounts for the sample from
Jan. 21, 2020, through June 11, 2020. The dependent variables are the daily time-weighted percent quoted spread
(QSPREAD) and daily average percent effective spread (ESPREAD) based on the Lee and Ready (1991) trade
classification. Results based on the entire sample period, normal and lockdown periods, and lockdown and reopen
periods are reported in Panels A, B, and C, respectively. LOCKDOWN is a dummy variable equal to 1 between Mar. 16
and May 7. REOPEN is a dummy variable equal to 1 since May 8. Lockdown and reopening dates are identified based on the
U.S. driving mobility index published by Apple (https://www.apple.com/COVID19/mobility). RETAIL is the daily log number of
Robinhood trading accounts for each firm. All regression models include past-week returns (PRET) and firm fixed effects (FE).
The t-statistics reported in parentheses are based on standard errors clustered at the firm and day levels. *, **, and *** indicate
statistical significance at the 10%, 5%, and 1% levels, respectively. Bold coefficients highlight the results of interest.
QSPREAD ESPREAD
Dependent Variable (%) 1 2
Panel A. Entire Period
RETAIL 0.039 0.024**

(1.63) (2.40)
PRET 0.359** 0.142**
(2.39) (2.44)
Firm FE Yes Yes
N 225,948 225,910
Adj. R2 0.757 0.799
Panel B. Normal Versus Lockdown
RETAIL 0.041* 0.001

(1.95) (0.16)
LOCKDOWN 1.094*** 0.395***
(10.49) (9.33)
RETAIL LOCKDOWN 0.117*** 0.041***
(9.82) (8.43)
PRET 0.473*** 0.185***
(4.17) (4.26)
Firm FE Yes Yes
N 171,779 171,747
Adj. R2 0.791 0.827
Panel C. Lockdown Versus Reopen
RETAIL 0.318*** 0.123***

(10.23) (8.36)
REOPEN 0.659*** 0.239***
(8.68) (6.75)
RETAIL REOPEN 0.077*** 0.027***
(8.84) (6.46)
PRET 0.193** 0.076**
(2.47) (2.50)
Firm FE Yes Yes
N 140,023 139,999
Adj. R2 0.849 0.862
in both columns, confirming a worsened stock-liquidity condition during the

pandemic. Regarding economic magnitude, the coefficient on LOCKDOWNt is
0.395%, indicating that the effective spread during lockdown is almost 200%
larger than that during the normal period (the average effective spread is 0.202%
during the normal period). Further, consistent with our hypothesis, the signifi-
cant and negative coefficient on RETAILi,t LOCKDOWNt indicates that the
increase in retail trading activity contributed to lowering spreads of trades during
the pandemic, thus “flattening the illiquidity curve.” In terms of economic sig-
nificance, a 1-standard-deviation increase in retail trading (1.876) is associated
with an absolute 7.7-bps drop in the effective spread in lockdown. Given that
during lockdown (and 0 retail trading), the average effective spread is approxi-
mately 59.7 bps (= 0.202% + 0.395%), the top–bottom decile spread of retail
trading (approximately 3 times the standard deviation) is roughly 23.1 bps (0.041
1.876 3), or 38.7% (= 23.1/59.7) of the average effective spread during
lockdown. That is, moving from the bottom to the top decile of stocks sorted on
their retail trading, there is a drop of 38.7% in the effective spread.
Further, when mobility increases and the economic uncertainty is gradually
resolved as the country starts to reopen in early May, aggregate illiquidity is
attenuated. As shown in Figure 2, although stock market participation from retail
investors continues to soar after reopening, the speed of the increased number of
accounts holding per stock slows down. For example, the average number of
trading accounts per stock increases over 42% in the first month of lockdown,
whereas the average number of trading accounts per stock increases approximately
12% in the first month of reopening. Taken together, we expect that the impact of
retail trading on liquidity will be smaller after reopening. To validate this hypoth-
esis, we test the following model:
(5) SPREADi,t ¼ αi þ β1 RETAILi,t þ β2 REOPENt

þ β3 RETAILi,t REOPENt þ γ CONTROLS þ ϵi,t ,
where REOPENt is a dummy variable equal to 1 since May 8, 2020, and 0 during
lockdown. The dependent variable SPREADi,t is again either the quoted spread or
the effective spread.
Panel C of Table 3 reports the results of estimating equation (5). Consist
with the notion that increased mobility improves liquidity, the coefficient estimate
on REOPENt in column 2 is significant, –0.239%, indicating a roughly 72% drop
in the effective spread from its lockdown average of 0.332%. The net effect of
retail trading on liquidity (i.e., summing up the coefficients of RETAILi,t and
RETAILi,t REOPENt ) remains positive and significant. In addition, the positive
coefficient estimate on RETAILi,t REOPENt confirms our conjecture that the
impact of retail trading on liquidity provision after reopening is smaller than that
during lockdown.
Taken together, our results advance that retail trading helped attenuate the rise
in illiquidity over the crisis on average. In addition, when mobility increased as
most states started to reopen in early May, the increase in retail trades lessened and,
in turn, their liquidity provision.
B. Attention-Driven Retail Trading

Evidence documented in Table 2 shows that retail investors are particularly
attracted by attention-grabbing stocks, especially so during lockdown, when people
pay full attention to financial markets. However, whether attention-driven retail
trading will provide or demand liquidity is yet unknown.
To examine this question, we extend equation (4) by interacting with
COVID-19–related media coverage, as follows:
(6) SPREADi,t ¼ αi þ β1 COVERAGEi,t þ β2 RETAILi,t COVERAGEi,t

þ β3 RETAILi,t LOCKDOWNt
þ β4 RETAILi,t COVERAGEi,t LOCKDOWNt
þ β5 Xi,t þ ϵi,t ,
where COVERAGEi,t is a dummy variable equal to 1 if the fraction of firm i’s

COVID-19–related articles to its overall media coverage at day t is greater than
0, and 0 otherwise. The dependent variable SPREADi,t is again either the quoted
spread or the effective spread. RETAILi,t , LOCKDOWNt , and COVERAGEi,t
LOCKDOWNt are included in the model but are packed into Xi,t for brevity.
Panel A of Table 4 reports the results of estimating equation (6). Consistent
with our prior finding, the coefficient estimate on RETAILi,t LOCKDOWNt
remains negative, –0.048, and significant at the 1% level. However, the coefficient
of RETAILi,t COVERAGEi,t LOCKDOWNt is positive and significant at the
1% level, suggesting that although retail investors tend to act as liquidity providers
overall during the pandemic, they seem to do so significantly less when their trading
activity is motivated by chasing firms under the spotlight in the context of COVID-
19. Consistent with our finding, Eaton et al. (2021) use Reddit WallStreetBets
mentions as a proxy for stocks receiving high attention from retail traders and find
that Robinhood app outages are associated with improved liquidity for those stocks.
They further show that such attention-driven retail trades are more likely to herd and
persist; thus, market makers may find it more difficult to unload inventory risk. In
addition, high-frequency traders (HFTs) can become informed by paying to observe
the (autocorrelated) retail order flow and, in turn, can increase adverse selection for
other (uninformed) market makers. Also, along the lines of the evidence provided
by von Beschwitz et al. (2020), news analytics of media coverage ignite algorithmic
trading, and although they tend to speed up stock price and trading volume in
response to articles, they also reduce liquidity.
Although attention-driven retail trading seems to provide less liquidity for
high-attention stocks relative to non–media-driven trading, the net effect of
attention-driven retail trading on liquidity provision (i.e., summing up the
coefficients of RETAILi,t LOCKDOWNt and RETAILi,t COVERAGEi,t
LOCKDOWNt ) is still moderately positive (i.e., a negative net impact on illiquid-
ity). In addition, we confirm a negative relation between attention-driven retail
trading and the bid–ask spread during the normal period (i.e., the coefficient
estimate on RETAILi,t COVERAGEi,t ) and over the entire sample period
(in untabulated analysis). Furthermore, when most states begin to reopen, we expect
that the impact of attention-driven retail trading would attenuate because retailers
are more likely to be distracted. To test this hypothesis, we modify equation (6)
by replacing LOCKDOWNt with REOPENt and report the results in Panel B of
Table 4. Indeed, the negative and significant coefficient estimate on RETAILi,t
COVERAGEi,t REOPENt confirms our hypothesis that media-driven liquidity
demand by retail investors is attenuated in the reopen period compared with that
during lockdown.
TABLE 4
Retail Investors and COVID-19–Related Media Coverage
Table 4 reports the OLS regression results of illiquidity measures on the number of retail trading accounts and COVID-19–related
media coverage for the sample from Jan. 21, 2020, through June 11, 2020. The dependent variables are the daily time-weighted
percent quoted spread (QSPREAD) and daily average percent effective spread (ESPREAD) based on the Lee and Ready (1991)
trade classification. Results based on the normal and lockdown periods and the lockdown and reopen periods are reported in
Panels A and B, respectively. LOCKDOWN is a dummy variable equal to 1 between Mar. 15 and May 7. REOPEN is a dummy
variable equal to 1 since May 8t. Lockdown and reopening dates are identified based on the U.S. driving mobility index published
by Apple (https://www.apple.com/COVID19/mobility). COVERAGE is a dummy variable equal to 1 if the ratio of a firm’s daily
COVID-19–related articles to its total daily media coverage is greater than 0, and 0 otherwise. RETAIL is the daily log number of
Robinhood trading accounts for each firm. All regression models include past week returns (PRET) and firm fixed effects (FE). The
t-statistics reported in parentheses are based on standard errors clustered at the firm and day levels. *, **, and *** indicate
statistical significance at the 10%, 5%, and 1% levels, respectively. Bold coefficients highlight the results of interest.
QSPREAD ESPREAD
Panel A. Normal Versus Lockdown
RETAIL 0.033 0.001

(1.60) (0.11)
COVERAGE 0.473*** 0.173***
(10.36) (10.55)
COVERAGE RETAIL 0.059*** 0.021***
(9.94) (9.49)
LOCKDOWN 1.224*** 0.437***
(10.60) (9.12)
RETAIL LOCKDOWN 0.140*** 0.048***
(9.93) (8.18)
COVERAGE LOCKDOWN 0.821*** 0.286***
(11.55) (9.06)
COVERAGE RETAIL LOCKDOWN 0.109*** 0.037***
(11.09) (8.40)
PRET 0.464*** 0.182***
(4.12) (4.20)
Firm FE Yes Yes
N 171,779 171,747
Adj. R2 0.793 0.828
Panel B. Lockdown Versus Reopen
RETAIL 0.325*** 0.125***

(10.50) (8.49)
COVERAGE 0.186*** 0.062***
(3.97) (2.90)
COVERAGE RETAIL 0.026*** 0.009***
(4.26) (3.14)
REOPEN 0.740*** 0.264***
(8.66) (6.39)
RETAIL REOPEN 0.090*** 0.031***
(8.70) (5.98)
COVERAGE REOPEN 0.336*** 0.109***
(5.64) (3.51)
COVERAGE RETAIL REOPEN 0.046*** 0.015***
(5.94) (3.58)
PRET 0.192** 0.075**
(2.46) (2.50)
Firm FE Yes Yes
N 140,023 139,999
Adj. R2 0.849 0.862
C. Identification
Our results so far can be interpreted as an association rather than a causal
relation between retail trading and stock liquidity. Common time-series trends
may be subject to endogeneity concerns. For example, the Federal Reserve
FIGURE 3
Implementation Dates for State Stay-at-Home Order
Figure 3 reports the staggered implementation dates of the stay-at-home order issued by each U.S. state, district, and region.
March 2020 April 2020
State Firms Lockdown from 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 1 2 3 4 5 6 7 8 9 10 11 12 13

Puerto Rico 5 3/15/2020
California 369 3/19/2020
Illinois 104 3/21/2020
New Jersey 72 3/21/2020
New York 176 3/22/2020
Connecticut 44 3/23/2020
Louisiana 13 3/23/2020
New Mexico 1 3/23/2020
Ohio 78 3/23/2020
Oregon 12 3/23/2020
Washington 45 3/23/2020
Delaware 10 3/24/2020
Indiana 37 3/24/2020
Massachusetts 143 3/24/2020
Michigan 41 3/24/2020
West Virginia 7 3/24/2020
Hawaii 9 3/25/2020
Idaho 6 3/25/2020
Vermont 2 3/25/2020
Wisconsin 42 3/25/2020
Colorado 50 3/26/2020
Kentucky 12 3/26/2020
Minnesota 44 3/27/2020
New Hampshire 7 3/27/2020
Alaska 1 3/28/2020
Montana 2 3/28/2020
Rhode Island 8 3/28/2020
Kansas 12 3/30/2020
Maryland 36 3/30/2020
North Carolina 13 3/30/2020
Virginia 70 3/30/2020
Arizona 36 3/31/2020
Tennessee 39 3/31/2020
District of Columbia 7 4/1/2020
Florida 84 4/1/2020
Nevada 23 4/1/2020
Pennsylvania 102 4/1/2020
Maine 6 4/2/2020
Texas 206 4/2/2020
Georgia 65 4/3/2020
Mississippi 7 4/3/2020
Alabama 10 4/4/2020
Missouri 30 4/6/2020
South Carolina 13 4/7/2020
Arkansas 12 N/A
Iowa 16 N/A
Nebraska 10 N/A
North Dakota 3 N/A
Oklahoma 15 N/A
South Dakota 6 N/A
Utah 21 N/A
announcements of a rate cut and a liquidity-injection program for the bond market
coincide with the time of significant drops in the mobility index. These announce-
ments may simultaneously increase retail trading and liquidity or even lead to a
reverse causality during lockdown (i.e., improved stock market liquidity leading to
more retail trading). Thus, in this section, to verify that our findings are indeed
causal, we use an identification strategy that utilizes the staggered implementation
of the stay-at-home advisory across U.S. states, which is likely independent of
financial market conditions.
Most U.S. states implemented stay-at-home orders during the pandemic. As
depicted in Figure 3, Puerto Rico is the first U.S. territory to shut down on Mar.
15, whereas a few states, such as Arkansas, have never officially issued such an
order statewide.14 Once the stay-at-home order is implemented, people in affected
14
Table IA.1 in the Supplementary Material details the stay-at-home order implementation date for
each state.
states will be forced to stay at home most of the time, and in turn, their attention to
and participation in stock markets are expected to be significantly higher. At the
same time, retail investors in other states will not be affected because the mandates
in their states are not yet in place. Ideally, in a perfect setting, the staggered
implementation of the stay-at-home order across states provides a shock to retail
investors’ mobility based on their locations, but investor location data are unavail-
able to us.
To overcome this caveat, we provide 2 types of tests. First, we rely on the well-
documented “home-bias” phenomenon. Specifically, Coval and Moskowitz (1999)
show that the preference for investing close to home applies to portfolios of domestic
stocks. Ivković and Weisbenner (2005) further document that households exhibit a
strong preference for local investments. Therefore, we use a firm’s headquarters
location as a coarse proxy for household location. Despite being a noisy proxy,
any findings based on it can be viewed as a lower bound of the true effect. We
conduct the DID analysis using a 15-trading-day window surrounding the imple-
mentation date of the stay-at-home order, and we divide it into five 3-day horizons
(day –1 to +1 as the event period). Except for firms in a few states that never
implemented the stay-at-home order, most firms will be treatment firms at some
point during the lockdown. For each treatment firm, we find a control firm, whose
headquarters states are not yet affected, based on a one-to-one nearest-neighbor
propensity-score matching. Variables used in the propensity-score matching include
bid–ask spread, firm size, past-week returns, log number of retail trading accounts,
and COVID-19–related media-coverage ratio at the beginning of the event window,
with replacement. The final DID sample contains 2,213 treatment firms and
995 unique control firms. Panel A of Table 5 reports the quality of the matching.
We show that the characteristics of the treated group are not statistically different from
those of the control group. We then run the following DID regression:
(7) SPREADi,t ¼ αi þ β1 RETAILi,t TREATi POSTt

þ β2 RETAILi,t COVERAGEi,t TREATi POSTt
þ β3 Xi,t þ ϵi,t ,
where the dependent variable SPREADi,t is the averaged quoted or effective spread
for each 3-day window. COVERAGEi,t is a dummy variable equal to 1 if the 3-day
average fraction of firm i’s COVID-19–related media coverage to its overall media
coverage is greater than 0, and 0 otherwise; RETAILi,t is the 3-day average log
number of unique Robinhood trading accounts for stock i; POSTt is a dummy
variable equal to 1 after the stay-at-home order is implemented based on each
treatment firm, and 0 for all the days before; and TREATi is a dummy variable
equal to 1 for treatment firms and 0 for firms in the control group. All other relevant
variables (direct effects and double-interaction terms) are included in the model but
packed into Xi,t for brevity.
We present the DID results in Panel B of Table 5. The negative coefficient
of RETAILi,t TREATi POSTt suggests that retail trading provides more
liquidity for treatment firms relative to firms in the control group after the stay-
at-home order is implemented. Furthermore, the significant and positive coefficient
TABLE 5
State-Level Stay-at-Home Advisory as a Shock
Table 5 reports the results of the difference-in-differences (DID) test that examines how exogenous changes in retail trading
due to the stay-at-home advisory affect stock liquidity. We match firms using one-to-one nearest-neighbor propensity-score
matching, with replacement. Panel A compares the average values of the variables used to estimate propensity scores for
firms in the treatment and control groups. The dependent variable, TREAT, is equal to 1 if the firm-day belongs to the treatment
group, and 0 otherwise. Panel B provides the results of variables of interest in the DID test. The dependent variables in Panel B
are the time-weighted percent quoted spread (QSPREAD) and average percent effective spread (ESPREAD) based on the
Lee and Ready (1991) trade classification. POST is a dummy variable equal to 1 for firm-day observations after the stay-at-
home order is in place in the firm’s headquarters state. The sample uses 15 trading days surrounding the effective date of the
state stay-at-home mandate, we and divide the 15 days into five 3-day windows. COVERAGE is the 3-day average ratio of a
firm’s daily COVID-19–related articles to its total daily media coverage. RETAIL is the 3-day average log number of daily
Robinhood trading accounts for each firm. All regression models include firm fixed effects (FE), and other variables are
omitted for brevity. The t-statistics reported in parentheses are based on standard errors clustered at the firm level. *, **, and ***
indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
Treatment Group Control Group Difference (t-stat.)
Panel A. Post-Match Differences

Panel A.1. QSpread Matched Sample
QSPREAD (%) 0.911 0.903 (0.20)
SIZE 21.581 21.596 (0.28)
PRET (%) 0.095 0.096 (0.21)
RETAIL 6.059 6.122 (1.11)
COVERAGE 0.149 0.154 (0.78)
Panel A.2. ESpread Matched Sample

ESPREAD 0.390 0.372 (1.07)
SIZE 21.581 21.536 (0.91)
PRET 0.095 0.097 (0.29)
RETAIL 6.059 6.128 (1.23)
COVERAGE 0.149 0.148 (0.11)
Panel B. DID Test
QSPREAD ESPREAD
RETAIL POST TREAT 0.022*** 0.010***

(6.54) (9.01)
COVERAGE RETAIL POST TREAT 0.071*** 0.015**
(3.36) (2.01)
Firm FE Yes Yes
N 16,399 16,465
Adj. R2 0.860 0.931
on RETAILi,t COVERAGEi,t TREATi POSTt confirms that attention-driven

retail trading tends to demand liquidity during lockdown. The results documented
in the DID analysis verify, albeit imperfectly, that our findings on the relation between
retail trading and liquidity are likely causal rather than a simple correlation.
Second, we improve our identification of the geographic location of retail
trading by relying on the Google search-volume index.15 Da et al. (2011) and
Ben-Rephael et al. (2017) show that the Google stock ticker search volume is a novel
and direct measure of investor attention on stock, and such searches are positively
related to retail trading.16 For our analysis, we download the daily search-volume
index for individual stocks at the state level (SSVI) and proxy for the retail-investor
activity of a given stock. Similar to the pattern of Robinhood trading accounts,
Figure 4 shows that the average daily Google search volume per stock ticker peaks
15
Search terms are publicly available on Google Trends (http://www.google.com/trends).
16
We verify that the correlation of the daily log Google stock ticker search volume and the daily log
number of Robinhood accounts per stock over our sample is 0.46 (statistically significant).
FIGURE 4
Google Search Volume of Stock Ticker Symbols During the COVID-19 Pandemic Outbreak
Figure 4 reports the daily time-series average of the Google search-volume index across U.S. states and stock tickers, as well
as the corresponding 7-day moving average. The Google search volume (normalized by its time-series average) is obtained
via Google Trends (http://www.google.com/trends).
3.0
2.5
2.0
1.5
1.0
0.5
0.0
1/21/20
1/24/20
1/29/20
2/3/20
2/6/20
2/11/20
2/14/20
2/20/20
2/25/20
2/28/20
3/4/20
3/9/20
3/12/20
3/17/20
3/20/20
3/25/20
3/30/20
4/2/20
4/7/20
4/13/20
4/16/20
4/21/20
4/24/20
4/29/20
5/4/20
5/7/20
5/14/20
5/19/20
5/22/20
5/28/20
6/2/20
6/5/20
6/10/20
Google Ticker Search Volume (Normalized) 7-day Moving Average
at the beginning of lockdown and remains at a high level afterward. The results in
Panel A of Table 6 further confirm a significant increase in ticker search volume
among states in which stay-at-home orders are implemented. Next, to examine the
impact of retail trading on stock liquidity, we conduct the following OLS regres-
sions using the stock-state-day Google search-volume data:
(8) SPREADi,t ¼ αi þ β3 SSVIi,j,t LOCKDOWN j,t

þ β4 SSVIi,j,t LOCAL_COVERAGEi,j,t LOCKDOWN j,t
þ β5 Xi,j,t þ ϵi,j,t ,
where SSVIi,j,t is the daily number of searches for firm i by individuals in state j at
day t, scaled by its time-series average. LOCAL_COVERAGEi,j,t is a dummy
variable equal to 1 if firm i is covered by local media distributed in state j at day
t, and 0 otherwise. LOCKDOWN j,t is a dummy variable equal to 1 after the state j
(where the ticker search takes place) starts to issue the stay-at-home order, and
0 otherwise. The dependent variable SPREADi,t is again either the quoted spread or
the effective spread. LOCAL_COVERAGEi,j,t , SSVIi,j,t LOCAL_COVERAGEi,t ,
SSVIi,j,t , LOCKDOWN j,t , and LOCAL_COVERAGEi,j,t LOCKDOWN j,t are
included in the model but are packed into Xi,j,t for brevity. Panels B and C of
Table 6 report the results of estimating equation (8). In line with the main findings of
the article, the negative coefficient on SSVIi,j,t LOCKDOWN j,t suggests that
searches during state lockdown are associated with improved stock liquidity.
Furthermore, we utilize the state-level Google search-volume index to verify
the home-bias assumption employed in our DID test. To do so, we create a dummy
variable, HOME_STATE, which is equal to 1 if searches of a given firm originating

from the state in which the firm headquarters is located, and 0 otherwise. As
reported in Panel A of Table 7, HOME_STATEi,j is positive and significant,
confirming the existence of home bias in our sample. The insignificant coefficient
estimate of HOME_STATEi,j LOCKDOWN j,t indicates that the magnitude of
home bias is largely unchanged during lockdown.
TABLE 6
Google Search Volume, Local Media Coverage, and Illiquidity During Lockdown
Table 6 reports the OLS regression results of illiquidity on an alternative retail trading measure proxied by the Google search
volume on stock ticker symbols for the sample from Jan. 21, 2020, through May 7, 2020. For each state on a specific date, SSVI
is the daily log number of searches by users in that U.S. state (district) on search terms (i.e., stock ticker symbols), scaled by its
time-series average, obtained via Google Trends (http://www.google.com/trends). The dependent variables are SSVI in Panel
A and the daily time-weighted percent quoted spread (QSPREAD) and daily average percent effective spread (ESPREAD)
based on the Lee and Ready (1991) trade classification in Panels B and C, respectively. LOCKDOWN is a dummy variable
equal to 1 since the implementation date of the U.S. state (district) stay-at-home order (Table A1 provides stay-at-home
implementation dates for each state (district)). LOCAL_COVERAGE is a dummy variable equal to 1 if, on a day, the stock is
covered by local media distributed in the state where the ticker search happens, and 0 otherwise. All regression models
include past-week returns (PRET) and firm and day fixed effects (FE). The t-statistics reported in parentheses are based on
standard errors clustered at the firm and day levels. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels,
respectively. Bold coefficients highlight the results of interest.
Google Search
Dependent Variable 1 2
Panel A. SSVI
LOCAL_COVERAGE 0.519***
(12.88)
LOCKDOWN 0.239*** 0.239***
(8.58) (8.57)
LOCAL_COVERAGE LOCKDOWN 0.021
(0.48)
PRET 0.035*** 0.035***
(2.97) (2.97)
Firm and day FE Yes Yes
N 8,765,276 8,765,276
Adj. R2 0.185 0.185
Illiquidity
1 2
Panel B. QSPREAD (%)
LOCAL_COVERAGE 0.018***
(2.74)
SSVI 0.001 0.000
(0.32) (0.07)
LOCAL_COVERAGE SSVI 0.016***
(5.56)
LOCKDOWN 0.005*** 0.007***
(4.02) (4.19)
SSVI LOCKDOWN 0.014** 0.015***
(2.57) (2.85)
LOCAL_COVERAGE LOCKDOWN 0.065***
(3.86)
LOCAL_COVERAGE SSVI LOCKDOWN 0.019***
(5.83)
PRET 0.037 0.037
(0.87) (0.87)
N 8,762,572 8,762,572
Adj. R2 0.799 0.799
(continued on next page)

TABLE 6 (continued)
Google Search Volume, Local Media Coverage, and Illiquidity During Lockdown
Illiquidity
Dependent Variable 1 2
Panel C. ESPREAD (%)
LOCAL_COVERAGE 0.007*
(1.77)
SSVI 0. 001 0.000
(0.49) (0.19)
LOCAL_COVERAGE SSVI 0.006***
(4.54)
LOCKDOWN 0.002*** 0.002***
(2.99) (2.79)
SSVI LOCKDOWN 0.004 0.005*
(1.48) (1.67)
LOCAL_COVERAGE LOCKDOWN 0.019**
(2.22)
LOCAL_COVERAGE SSVI LOCKDOWN 0.006***
(3.32)
PRET 0.038* 0.038*
(1.68) (1.68)
N 8,760,908 8,760,908
Adj. R2 0.731 0.731
Next, in Panels B and C of Table 7, we examine the impact of such home-

state stock attention (home-bias–induced trading) on the liquidity provision during
the pandemic. The negative coefficient of HOME_STATEi,j SSVIi,j,t
LOCKDOWN j,t suggests that searches on home-state stocks during lockdown
are associated with even more liquidity provision above and beyond the average effect.
D. Retail Trading and Stock Returns
The results so far demonstrate that the increased trading activity by retail
investors significantly contributes to dampening illiquidity during lockdown,
whereas their trading activities motivated by COVID-19–related media coverage
result in less liquidity provision for those stocks. Thus, what is the impact of retail
trading on contemporaneous stock returns?
To check this question, we rerun the model specified in equation (6) by
replacing daily stock returns as the dependent variable and report the results in
Panel A of Table 8. The coefficient estimate on RETAILi,t LOCKDOWNt is
insignificant, suggesting that retail trading during lockdown does not have a sig-
nificant impact on contemporaneous stock returns. However, the coefficient esti-
mate on RETAILi,t COVERAGEi,t LOCKDOWNt is negative and significant
at the 1% level, indicating that retail trading on stocks covered by COVID-19–
related media on average incurs a loss on the day of trading. One possible expla-
nation is that the negative coefficient simply reflects that those stocks selected by
the media overall perform poorly during the lockdown. Contrary to this conjecture,
we find that the coefficient estimate on COVERAGEi,t LOCKDOWNt is positive
and significant at the 1% level.
TABLE 7
Google Search Volume, Home Bias, and Illiquidity
Table 7 reports the OLS regression results of home bias and its impact on illiquidity using the Google search volume on stock
ticker symbols for the sample from Jan. 21, 2020, through May 7, 2020. For each state on a specific date, SSVI is the daily log
number of searches by users in that U.S. state (district) on search terms (i.e., stock ticker symbols), scaled by its time-series
average, obtained via Google Trends (http://www.google.com/trends). The dependent variables are the SSVI in Panel A and
the daily time-weighted percent quoted spread (QSPREAD) and daily average percent effective spread (ESPREAD) based on
the Lee and Ready (1991) trade classification in Panels B and C, respectively. LOCKDOWN is a dummy variable equal to 1
since the implementation date of the U.S. state (district) stay-at-home order (Table A1 provides stay-at-home implementation
dates for each state (district)). HOME_STATE is a dummy variable equal to 1 if the searches of a given firm originate from the
state in which the firm headquarters is located, and 0 otherwise. All regression models include past-week returns (PRET) and
firm and day fixed effects (FE). The t-statistics reported in parentheses are based on standard errors clustered at the firm and
day levels. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. Bold coefficients highlight
the results of interest.
Dependent Variable Google Search
Panel A. SSVI
HOME_STATE 0.320***
(20.23)
LOCKDOWN 0.236***
(8.38)
HOME_STATE LOCKDOWN 0.005
(0.37)
PRET 0.035***
(2.97)
Firm and day FE Yes
N 8,765,276
Adj. R2 0.190
Panel B. QSPREAD (%) Illiquidity
HOME_STATE 0.000
(0.03)
SSVI 0.002
(0.46)
HOME_STATE SSVI 0.006***
(4.23)
LOCKDOWN 0.004***
(4.03)
SSVI LOCKDOWN 0.013**
(2.40)
(0.21)
HOME_STATE SSVI LOCKDOWN 0.016***
(5.20)
PRET 0.037
(0.87)
Firm and day FE Yes
N 8,762,572
Adj. R2 0.799
Panel C. ESPREAD (%) Illiquidity
HOME_STATE 0.000
(0.46)
SSVI 0.001
(0.62)
HOME_STATE SSVI 0.003***
(3.88)
LOCKDOWN 0.001***
(2.99)
SSVI LOCKDOWN 0.004
(1.31)
(0.99)
HOME_STATE SSVI LOCKDOWN 0.007***
(4.70)
PRET 0.038*
(1.68)
Firm and day FE Yes
N 8,760,908
Adj. R2 0.731
TABLE 8
Retail Investors and Stock Returns
Table 8 reports the OLS regression results of stock returns on the number of retail trading accounts and COVID-19–related
media coverage for the sample from Jan. 21, 2020, through June 11, 2020. The dependent variables are the
contemporaneous daily stock returns. Results based on the normal and lockdown periods and the lockdown and reopen
periods are reported in Panels A and B, respectively. LOCKDOWN is a dummy variable equal to 1 between Mar. 16 and May 7.
REOPEN is a dummy variable equal to 1 since May 8. Lockdown and reopening dates are identified based on the U.S. driving
mobility index published by Apple (https://www.apple.com/COVID19/mobility). COVERAGE is a dummy variable equal to 1 if
the ratio of a firm’s daily COVID-19–related articles to its total daily media coverage is greater than 0, and 0 otherwise. RETAIL
is the daily log number of Robinhood trading accounts for each firm. All regression models include past-week returns (PRET)
and firm fixed effects (FE). The t-statistics reported in parentheses are based on standard errors clustered at the firm and day
levels. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. Bold coefficients highlight the
results of interest.
Dependent Variable = RET (%) 1 2
Panel A. Normal Versus Lockdown
RETAIL 0.594 0.598

(0.85) (0.87)
COVERAGE 2.051***
(2.73)
COVERAGE RETAIL 0.175**
(2.52)
LOCKDOWN 0.677 0.678
(0.54) (0.49)
RETAIL LOCKDOWN 0.101 0.081
(1.37) (0.94)
(2.94)
COVERAGE RETAIL LOCKDOWN 0.249**
(2.63)
PRET 7.083 7.112
(1.51) (1.52)
Firm FE Yes Yes
N 135,649 135,649
Adj. R2 0.025 0.026
Panel B. Lockdown Versus Reopen
RETAIL 1.317 1.282

(1.03) (0.99)
COVERAGE 0.269
(0.39)
COVERAGE RETAIL 0.026
(0.30)
REOPEN 0.280 0.233
(0.24) (0.19)
RETAIL REOPEN 0.072 0.080
(0.78) (0.70)
COVERAGE REOPEN 0.015
(0.01)
COVERAGE RETAIL REOPEN 0.013
(0.11)
PRET 10.956** 10.943**
(2.61) (2.60)
Firm FE Yes Yes
N 110,690 110,690
Adj. R2 0.045 0.046
E. Are Retail Investors Short-Run Contrarian Traders?
As documented in Table 2, over the entire sample period, retail investors are, on
average, momentum investors who chase recent performers. However, news articles
frequently report that retail investors tend to be short-term contrarian traders during the
pandemic, in the hope of a quick recovery of the economy, resulting in a “flight to crap.”
To examine this argument, we further interact key variables with past-week

stock returns. Panel A of Table 9 reports the results using the log number of unique
Robinhood trading accounts as the dependent variable. The coefficient on
PRETi,t LOCKDOWNt is statistically positive at the 10% level, confirming the
results, reported in Table 2, that retail investors are, on average, short-run momen-
tum traders.17 The insignificant coefficient estimate on COVERAGEi,t PRETi,t
LOCKDOWNt rejects the conjecture that retails investors flow into stocks that have
performed poorly over the past week and are covered by the media during lock-
down. Furthermore, Panel B of Table 9 reports results using the effective spread as
the dependent variable. Both coefficients of PRETi,t RETAILi,t LOCKDOWNt
and PRETi,t RETAILi,t COVERAGEi,t LOCKDOWNt are negative and sig-
nificant, suggesting that their trading activity on stocks that have performed poorly
over the past week is more like to demand liquidity compared with their trading on
stocks performing well over the past week. Last, as reported in Panel C of Table 9,
when using return as the dependent variable, we find that retail trading does not lead
to a significant return reversal.
V. Further Analysis
A. Sample Splits by Institutional Ownership
It is well documented that the impact of “noise” trading is more pronounced

among stocks with a smaller size and a lower level of institutional ownership.
For example, Peress and Schmidt (2020) show that the effect of sensational
news distraction on lowering liquidity is strongest for small stocks and/or
stocks with a low fraction of institutional ownership. Hence, one may wonder
if our results pertain only to a subsample of firms predominantly held by retail
investors.
To test this possibility, we sort stocks into 2 groups based on the fraction
of institutional ownership (IO) measured at the end of year 201918 and rerun
equation (6). As reported in Panel A of Table IA.2 in the Supplementary Material,
the coefficient estimate on RETAILi,t COVERAGEi,t LOCKDOWNt is posi-
tive and significant at the 1% level for both the low- and high-IO subsamples,
suggesting that our findings hold for stocks with different levels of IO. However, the
economic magnitude of our finding is significantly larger among low-IO stocks.
The collective evidence thus suggests that our findings are stronger for (but are not
limited to) smaller stocks or stocks primarily held by retail investors.
B. Liquidity Timing of Insider Trading
Another question arising is as follows: Who may benefit from the increased
retail trading activities amid the pandemic? Prior literature (e.g., Collin-Dufresne
and Fos (2015), (2016)) and, recently, Cookson, Fos, and Niessner (2021)
17
In untabulated analysis using 52-week returns as a benchmark for past performance, we find that
retail investors are contrarian investors over the long run.
18
Untabulated analysis shows that the results are qualitatively similar if the sample is partitioned
based on market capitalization at the end of year 2019.
TABLE 9
Retail Investors’ Response to Short-Run Performance
Table 9 reports the OLS regression results of the log number of retail trading accounts (Panel A), effective spread (Panel B,
ESPREAD), daily stock return (Panel C, RET) on the retail trading, COVID-19–related media coverage, and past-week returns
for the sample from Jan. 21, 2020, through June 11, 2020. LOCKDOWN is a dummy variable equal to 1 between Mar. 16 and
May 7. Lockdowns dates are identified based on the U.S. driving mobility index published by Apple (https://www.apple.com/
COVID19/mobility). COVERAGE is a dummy variable equal to 1 if the ratio of a firm’s daily COVID-19–related articles to its total
daily media coverage is greater than 0, and 0 otherwise. RETAIL is the daily log number of Robinhood trading accounts for
each firm. PRET is the past-week stock returns. All regression models include firm fixed effects (FE). The t-statistics reported in
parentheses are based on standard errors clustered at firm and day levels. *, **, and *** indicate statistical significance at the
10%, 5%, and 1% levels, respectively. Bold coefficients highlight the results of interest.
Dependent Variable
Panel A. ln(NO_OF_USER_ACCOUNTS)
COVERAGE 0.014
(1.11)
LOCKDOWN 0.354***
(15.59)
(4.98)
PRET 0.108
(1.09)
PRET LOCKDOWN 0.247*
(1.72)
PRET COVERAGE 0.087
(1.27)
PRET COVERAGE LOCKDOWN 0.000
(0.00)
Firm FE Yes
N 171,831
Adj. R2 0.975
Panel B. ESPREAD (%)
RETAIL 0.004
(0.43)
COVERAGE 0.143***
(6.74)
COVERAGE RETAIL 0.017***
(5.96)
LOCKDOWN 0.483***
(11.78)
RETAIL LOCKDOWN 0.055***
(11.14)
(7.82)
PRET 1.555***
(6.80)
PRET LOCKDOWN 1.040***
(3.75)
PRET RETAIL 0.192***
(6.99)
PRET COVERAGE 0.059**
(2.30)
COVERAGE RETAIL LOCKDOWN 0.034***
(7.41)
PRET COVERAGE LOCKDOWN 0.367***
(3.45)
PRET RETAIL LOCKDOWN 0.130***
(3.92)
PRET RETAIL COVERAGE LOCKDOWN 0.044***
(3.24)
Firm FE Yes
N 171,747
Adj. R2 0.832
(continued on next page)
TABLE 9 (continued)
Retail Investors’ Response to Short-Run Performance
Dependent Variable
Panel C. RET (%)
RETAIL 1.215
(1.50)
COVERAGE 2.055***
(2.85)
COVERAGE RETAIL 0.182*
(1.96)
LOCKDOWN 0.119
(0.10)
RETAIL LOCKDOWN 0.096
(1.04)
(3.67)
PRET 3.182
(0.20)
PRET LOCKDOWN 2.743
(0.15)
PRET RETAIL 0.438
(0.41)
PRET COVERAGE 1.550
(0.30)
COVERAGE RETAIL LOCKDOWN 0.271***
(2.86)
PRET COVERAGE LOCKDOWN 2.244
(0.36)
PRET RETAIL LOCKDOWN 0.019
(0.01)
PRET RETAIL COVERAGE LOCKDOWN 0.083
(0.19)
Firm FE Yes
N 171,821
Adj. R2 0.009
demonstrate, both theoretically and empirically, that informed traders strategically

choose to trade more when noise-trading activity is high. Indeed, as reported by the
Wall Street Journal, top executives at U.S.-traded companies sold a total of roughly
$9.2 billion between Feb. 1 and Mar. 20, possibility to unload uncertainty regarding
COVID-19.19 If that is the case, we expect that insiders are more likely to sell their
stocks when retail trading is more active but less likely to do so when their firms are
attracted by a lot of attention-driven retail trading.
As reported in column 2 of Panel A in Table IA.3 of the Supplementary
Material, the significant and positive coefficient estimate on LOCKDOWNt indi-
cates that insider sales in general are less likely during lockdown as a result of the
severely deteriorated market condition. Consistent with our hypothesis, the
coefficient estimate on RETAILi,t LOCKDOWNt is positive and significant at
the 1% level, suggesting that insider sales time the liquidity provided by retail
investors during lockdown. Furthermore, the negative coefficient of RETAILi,t
COVERAGEi,t LOCKDOWNt demonstrates that insiders are less likely to sell
19
For more details, see https://www.wsj.com/articles/bezos-other-corporate-executives-sold-shares-
just-in-time-11585042204.
their shares when retail trading is likely to be motivated by the media coverage.
However, we do not observe similar liquidity-timing strategies for insider pur-
chases.
C. How Does Retail Trading Improve Stock Liquidity?

In this section, we examine whether retail investors are likely to be noise or
informed investors. To check this, we decompose the effective spread into two
components: price impact and realized spared.
First, we rerun the analysis specified in equation (6) with 2 alternative measures
of liquidity. In addition, we examine the effect of retail trading on stock price volatility.
The results reported in Table IA.4 of the Supplementary Material are qualitatively
similar to the baseline findings reported in Table 4. The coefficient estimate on
RETAILi,t LOCKDOWNt is negative and significant at the 1% level for all 3 alter-
native measures, indicating that overall, retail investors tend to act as liquidity
providers, and their trading attenuates stock-return volatility during the lockdown.
The coefficient estimate on RETAILi,t COVERAGEi,t LOCKDOWNt is posi-
tive and significant at the 1% level across all 3 measures, again confirming that
attention-driven trading by retail investors tends to demand liquidity and induce
more price volatility compared with non–attention-driven trades. Thus, the conclu-
sion from this table is that on an absolute basis, retail investors help reduce both
asymmetric information (price impact) and inventory risk (realized spread).
Furthermore, we examine this question using relative-based illiquidity mea-
sures, that is, the percentage ratio of price impact to effective spread and the
percentage ratio of realized spread to effective spread. The results reported in
Table IA.5 of the Supplementary Material show that the contribution of retail
trading on liquidity is mostly driven by reducing information asymmetry. Spe-
cifically, the coefficient estimate on RETAILi,t LOCKDOWNt is negative and
significant at the 1% level for the percentage ratio of price impact to effective
spread and is significantly positive for the percentage ratio of realized spread to
effective spread. The overall evidence suggests that retail investors act as noise
traders rather than informed investors, given that the price-impact component is
inversely related to noise-trading activity (e.g., Kyle (1985)).
D. Robustness: Reopen Dates and Model Specifications
In this section, we check the robustness of our results to alternative choices of

reopening dates and model specifications.
First, we check whether our results are sensitive to the choice of reopening
dates because people may have different views regarding the actual reopening dates
in the United States. We rerun our baseline model using either May 1 or May 15 as
the reopening date on which the corresponding mobility score reached 80% or
100% of the pre–COVID-19 level, respectively. The results reported in Table IA.6
are qualitatively similar to our baseline findings, demonstrating that our finding is
not driven by the specific choice of reopening date.
Finally, to control for common time trends (e.g., market-wide funding liquid-
ity shock and subsequent Federal Reserve liquidity-injection program), we add day
fixed effects to the baseline model. The results reported in Table IA.7 show that
adding day fixed effects does not change the overall finding documented in the
baseline model. In addition, we examine whether our results vary if we combine the
three phases (normal, lockdown, and reopen) into a framework of one regression.
To do so, the LOCKDOWN dummy turns on since Mar. 16, and the REOPEN
dummy turns on since May 8. The results reported in column 3 of Table IA.7 of the
Supplementary Material indicate that our finding is not sensitive to this alternative
model specification.
VI. Is Retail Trading Important?

This article demonstrates that retail trading provided liquidity during the pan-
demic, possibly preventing a severe liquidity crunch in the stock market. Can one
quantify the importance of retail trading? We can point to some anecdotal evidence
gathered over the course of writing this article. Here are a few examples collected
from various media publications over 2020:Q2 and 2020:Q3: i) Morgan Stanley
acquired E-Trade for $13 billion, ii) Charles Schwab set to close the acquisition of TD
Ameritrade for $26 billion, iii) Morgan Stanley announced it would buy Eaton Vance
for $7 billion, and iv) Fidelity Investments hired 2,000 mostly customer-facing staff
through June 2020 to meet client interest amid COVID-19.
Indeed, it seems that the asset-management industry is undergoing significant
changes that promote the importance of retail-investor flow. One reason presum-
ably is the profits generated by trading commissions with retail investors, who
are now more engaged in direct investment in financial markets, but such trading
commissions have been dropping at an increasing rate over recent years.
Yet, we postulate that the main reason that retail-investor flow is important is that
it tends to be predictable, making prior access to such information quite valuable. An
interesting anecdote in this context was provided by a Q2 SEC filing by Robinhood
(first cited by The Block), which revealed that Citadel Securities and a handful of
other firms paid Robinhood nearly $100 million in 2020:Q1 for its information about
the retail trading accounts on its platform. This suggests that not only the direct
trading commissions on behalf of retail investors may be a significant source of
revenue but also trading with or ahead of them. For example, Yang and Zhu (2020)
suggest that payment for (retail) order flow is a common practice in U.S. equity
markets; nevertheless, it is a largely overlooked source of institutional investors’
profit. The numerical solutions of their model point out that institutional investors’
profits are on the order of 70–90 bps per retail dollar volume. In other words, the
impact of retail trading on stock liquidity is a source of alpha for such fund managers.
Although this article utilizes the unique data from the Robinhood platform,
whose availability at the daily frequency allows us to better identify the impact of
retail trading on asset liquidity, we view our results as the lower bounds to more
general behavior. Indeed, the results shown in Table IA.8 of the Supplementary
Material indicate that the Google search volume of all popular retail trading plat-
forms (e.g., TD Ameritrade, E-Trade, Fidelity, and Charles Schwab) experienced a
significant increase during state lockdown. Given the evidence presented previ-
ously, the patterns unveiled using Robinhood data are likely part of a general trend
exhibited by retail investors recently and especially during the pandemic.
Our results also hint at the potential weaknesses of such easy access to
financial markets by retail investors. We often think of large financial institutions,
such as banks and large asset-management firms, as presenting systemic risk, yet
under a new regime of significant retail trading, retailers as a group might present
similar risks. If they suddenly decide to buy or sell certain assets, they might
significantly affect prices (e.g., Hertz during lockdown and, more recently, Game-
Stop) and generate a liquidity spiral (e.g., Brunnermeier and Pedersen (2008)).
Although institutional capital flows are at the very least monitored and are subject to
constraints, retail trading is not. Over time, this group of retail traders in aggregate,
with direct access to the market, may emerge as a significant driver of asset prices.
Therefore, although innovations in financial technology are welcome and generally
viewed as positive disruptions, we should also beware of some perhaps unintended
risks and consequences.
VII. Conclusion
This article shows that retail trading activity played a significant role in damp-
ening market illiquidity during the recent COVID-19 pandemic. With the country
under lockdown since mid-March 2020, individual investors turned their focus to the
stock market. We find that this increase in retail trading, benefiting from the easy
access of trading platforms particularly tailored to individual investors, contributes
to lowering spreads and the price impact of trades during lockdown. When mobility
increased as most states started to reopen in early May, the increase in retail trades
was significantly attenuated and, in turn, their liquidity provision. Although retail
investors tend to act as liquidity providers overall during the pandemic, they seem to
do so significantly less when their trading activities are motivated by media cover-
age. Using identification strategies that utilize the staggered implementation of the
stay-at-home advisory across states, we verify that retail trading provides more
liquidity for treatment firms relative to firms in the control group.
Overall, the findings highlight that advances in fintech in recent years, particu-
larly the availability of trading platforms to retail investors with low commissions and
trading costs, have disrupted the industry and have allowed retail investors easy, direct
access to financial markets. Recent data extracted from SimilarWeb suggest that
the level of online activity on the 5 aforementioned retail brokers’ websites
(TD Ameritrade, E-Trade, Fidelity, Charles Schwab, and Robinhood) during the first
2 months of 2021 increased by over 80% compared with its level during 2020:Q4. We
therefore believe retail trading will continue to exhibit a significant impact on financial
markets moving forward. The unusual circumstances presented during the pandemic
lockdown provide a fruitful testing ground to demonstrate the important role of retail
investors. Armed with direct market access and an abundance of free time, retail
investors emerged as a major force that contributed to attenuating or “flattening” the
rise in stock market illiquidity during the early months of the pandemic.
Supplementary Material
To view supplementary material for this article, please visit http://dx.doi.org/
10.1017/S0022109021000387.
References
Barber, B. M.; X. Huang; T. Odean; and C. Schwarz. “Attention-Induced Trading and Returns: Evidence
from Robinhood Users.” Working Paper, University of California at Davis (2021).
Barber, B. M., and T. Odean. “All That Glitters: The Effect of Attention and News on the Buying
Behavior of Individual and Institutional Investors.” Review of Financial Studies, 21 (2008),
785–818.
Barrot, J.; R. Kaniel; and D. Sraer. “Are Retail Traders Compensated for Providing Liquidity?” Journal
of Financial Economics, 120 (2016), 146–168.
Ben-Rephael, A.; Z. Da; and R. D. Israelsen. “It Depends on Where You Search: Institutional Investor
Attention and Underreaction to News.” Review of Financial Studies, 30 (2017), 3009–3047.
Brunnermeier, M. K., and L. H. Pedersen. “Market Liquidity and Funding Liquidity.” Review of
Financial Studies, 22 (2008), 2201–2238.
Campbell, J. “Household Finance.” Journal of Finance, 61 (2006), 1553–1604.
Collin-Dufresne, P., and V. Fos. “Do Prices Reveal the Presence of Informed Trading?” Journal of
Finance, 70 (2015), 1555–1582.
Collin-Dufresne, P., and V. Fos. “Insider Trading, Stochastic Liquidity, and Equilibrium Prices.”
Econometrica, 84 (2016), 1441–1475.
Cookson, J. A.; V. Fos; and M. Niessner. “Does Disagreement Facilitate Informed Trading? Evidence
from Activist Investors.” Working Paper, University of Colorado (2021).
Coval, J. D., and T. J. Moskowitz. “Home Bias at Home: Local Equity Preference in Domestic
Portfolios.” Journal of Finance, 54 (1999), 2045–2073.
Da, Z.; J. Engelberg; and P. Gao. “In Search of Attention.” Journal of Finance, 66 (2011), 1461–1499.
Eaton, G. W.; T. C. Green; B. Roseman; and Y. Wu. “Zero-Commission Individual Investors, High
Frequency Traders, and Stock Market Quality.” Working Paper, Oklahoma State University (2021).
Engelberg, J. E., and C. A. Parsons. “The Causal Impact of Media in Financial Markets.” Journal of
Finance, 66 (2011), 67–97.
Fang, V. W.; J. Madsen; and X. Shao. “Noise Trading: An Ad-Based Measure.” Working Paper,
University of Minnesota (2020).
Foucault, T.; D. Sraer; and D. J. Thesmar. “Individual Investors and Volatility.” Journal of Finance, 66
(2011), 1369–1406.
Grinblatt, M.; M. Keloharju; and J. Linnainmaa. “IQ and Stock Market Participation.” Journal of
Finance, 66 (2011), 2121–2164.
Haliassos, M., and C. Bertaut. “Why Do So Few Hold Stocks?” Economic Journal, 105 (1995),
1110–1129.
Holden, C. W., and S. Jacobsen. “Liquidity Measurement Problems in Fast, Competitive Markets:
Expensive and Cheap Solutions.” Journal of Finance, 69 (2014), 1747–1785.
Ivković, Z., and S. Weisbenner. “Local Does as Local Is: Information Content of the Geography of
Individual Investors’ Common Stock Investments.” Journal of Finance, 60 (2005), 267–306.
Kyle, A. S. “Continuous Auctions and Insider Trading.” Econometrica, 53 (1985), 1315–1335.
Lee, C. M. C., and M. J. Ready. “Inferring Trade Direction from Intraday Data.” Journal of Finance,
46 (1991), 733–746.
Li, L.; P. E. Strahan; and S. Zhang. “Banks as Lenders of First Resort: Evidence from the COVID-19
Crisis.” Review of Corporate Finance Studies, 9 (2020), 472–500.
Lou, D. “Attracting Investor Attention Through Advertising.” Review of Financial Studies, 27 (2014),
1797–1829.
Odean, T. “Do Investors Trade Too Much?” American Economic Review, 89 (1999), 1279–1298.
Peress, J., and D. Schmidt. “Glued to the TV: Distracted Noise Traders and Stock Market Liquidity.”
Journal of Finance, 75 (2020), 1083–1133.
Van Rooij, M.; A. Lusardi; and R. Alessie. “Financial Literacy and Stock Market Participation.” Journal
of Financial Economics, 101 (2011), 449–472.
Vissing-Jorgensen, A. “Perspectives on Behavioral Finance: Does ‘Irrationality’ Disappear with
Wealth? Evidence from Expectations and Actions.” NBER Macroeconomics Annual, 18 (2003),
139–194.
von Beschwitz, B.; D. B. Keim; and M. Massa. “First to ‘Read’ the News: News Analytics and
Algorithmic Trading.” Review of Asset Pricing Studies, 10 (2020), 122–178.
Welch, I. “Wisdom of the Robinhood Crowd.” Working Paper, University of California, Los Angeles
(2021).
Yang, L., and H. Zhu. “Back-Running: Seeking and Hiding Fundamental Information in Order Flows.”
Review of Financial Studies, 33 (2020), 1484–1533.

Flat

Uploaded by

Copyright:

Available Formats

Flat

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Flat

Uploaded by

Copyright:

Available Formats

https://doi.org/10.

1017/S0022109021000387 Published online by Cambridge University Press

Flattening the Illiquidity Curve: Retail Trading

“New investors […] sense a generational-buying moment […] We have

Graph A. 2008 Financial Crisis

Effective Spread (%)

Graph B. 2020 COVID-19 Pandemic

Effective Spread (%)

Effective Spread (%) Price (scaled)

Graph A. Mobility and COVID-19 Media Coverage

Graph B. Retail User Counts and Equity Fund

US Equity Fund Flow (cumulative net; $B)

US Equity Fund Flow ($B) Average Retail Accounts Per Stock

the Federal Reserve’s liquidity-injection programs indirectly transmitted to equity

the illiquidity curve.”8 We therefore advance that recent fintech innovations to

(1999), Ivković and Weisbenner (2005)). Specifically, we use a firm’s headquarters

II. Data and Sample

A. Retail User Accounts

B. COVID-19–Related Media Coverage

To estimate firm-level COVID-19 media-coverage intensity, we rely on data

C. Mobility Trends in the United States

The price impact can be viewed as a permanent component of the effective

E. Other Data and Summary Statistics

Throughout our analyses, we focus on common stocks (share codes 10 or 11)

SIZE (log) 2,265 21.574 1.688 20.287 21.421 22.572

COVID-19–related media coverage ratio is approximately 17.3%, but almost 90%

III. Retail Trading During the COVID-19 Pandemic

A. Retail Trading and COVID-19–Related Media Coverage

B. Attention-Driven Retail Trading During Lockdown

(2) RETAILi,t ¼ αi þ β1 COVERAGEi,t þ β2 LOCKDOWNt

(3) RETAILi,t ¼ αi þ β1 COVERAGEi,t þ β2 REOPENt

Panel A. Entire Period

Panel B. Normal Versus Lockdown

Panel C. Lockdown Versus Reopen

where LOCKDOWNt is a dummy variable equal to 1 in the lockdown period and

IV. Retail Trading and Stock Liquidity

A. Overall Retail Trading: Baseline Analysis

Panel A. Entire Period

RETAIL 0.039 0.024**

Panel B. Normal Versus Lockdown

RETAIL 0.041* 0.001

Panel C. Lockdown Versus Reopen

RETAIL 0.318*** 0.123***

in both columns, confirming a worsened stock-liquidity condition during the

(5) SPREADi,t ¼ αi þ β1 RETAILi,t þ β2 REOPENt

B. Attention-Driven Retail Trading

(6) SPREADi,t ¼ αi þ β1 COVERAGEi,t þ β2 RETAILi,t COVERAGEi,t

where COVERAGEi,t is a dummy variable equal to 1 if the fraction of firm i’s

Panel A. Normal Versus Lockdown

RETAIL 0.033 0.001

Panel B. Lockdown Versus Reopen

RETAIL 0.325*** 0.125***

March 2020 April 2020

State Firms Lockdown from 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 1 2 3 4 5 6 7 8 9 10 11 12 13

(7) SPREADi,t ¼ αi þ β1 RETAILi,t TREATi POSTt

Panel A. Post-Match Differences

Panel A.2. ESpread Matched Sample

RETAIL POST TREAT 0.022*** 0.010***

on RETAILi,t COVERAGEi,t TREATi POSTt confirms that attention-driven

RETAIL 0.318* 0.123*

RETAIL 0.325* 0.125*

RETAIL POST TREAT 0.022* 0.010*