Pacific-Basin Finance Journal 79 (2023) 102044

Contents lists available at ScienceDirect

Pacific-Basin Finance Journal

journal homepage: www.elsevier.com/locate/pacfin

COVID caused a negative bubble. Who profited? Who lost? How

stock markets changed?
Numan Ülkü a, Fahad Ali b, *, Saidgozi Saydumarov c, Deniz İkizlerli d
a
Institute of Economic Studies, Faculty of Social Sciences, Charles University, Opletalova 26, Prague 11000, Czechia
b
School of Finance, Zhejiang University of Finance and Economics, Xiasha Higher Education Zone, Hangzhou 310018, China
c
Westminster International University in Tashkent, Istiqbol str. 12, Tashkent, Uzbekistan
d
Istanbul Bilgi University, Kazım Karabekir Cad, no:2/13, Eyüpsultan İstanbul 34060, Turkiye

A R T I C L E I N F O A B S T R A C T

JEL classification: Compiling a unique, worldwide collection of trading data, we analyze investor types’ aggregate
G11 trading in stock markets throughout the COVID-19 episode, to assess investor types’ role in a
G41 worldwide negative bubble and their degree of sophistication in responding to it. Individual in­
G15
vestors were the main buyers and consequently the winners during the rebound. Foreign insti­
Keywords: tutional investors exited host markets; some domestic institutions exploited the negative bubble
Investor types’ trading behavior
by well-timed buying. In US index futures, asset managers heavily sold into crash; dealers profited
COVID-19
from the rebound. Individual investors’ buying was driven by their contrarian behavioral traits
Negative bubble
Retail investors and a unique positive shock to retail investor demand for self-serviced investing in stocks, driven
by work-from-home practices and unprecedented stimulus. This shock has changed the partici­
pant composition of world stock markets. Overall, the COVID-19 episode has many unique aspects
that cannot be accounted for under existing theories.

1. Introduction

Despite a vast literature on and renewed interest in stock price bubbles (e.g., Greenwood et al., 2019), little is known regarding
investor types’ trading behavior in actual bubbles. Existing literature on investor behavior in bubbles relies on experimental evidence
(e.g., Weitzel et al., 2020); the only study of investor types’ trading in an actual marketwide bubble is Griffin et al. (2011) on the
2000–2001 tech bubble.1 Negative bubbles, i.e., substantial dips followed by full rebounds (see Goetzman and Kim, 2018),2 constitute
a subset of the bubble phenomenon, with even more scarce evidence regarding which investor types foment and which ones exploit
them.3 Typically driven by overreaction, negative bubbles lead to significant wealth transfers and offer natural experiments to assess
investor sophistication.

* Corresponding author.
E-mail addresses: [email protected] (N. Ülkü), [email protected] (F. Ali), [email protected] (S. Saydumarov), deniz.ikizlerli@bilgi.
edu.tr (D. İkizlerli).
1
A limited literature using broker-level data and focusing on narrower bubbles is also emerging: Gong et al. (2017), Li et al. (2021).
2
The term ‘negative bubble’ was first used by Shiller (2000). He interprets 1929 and 1987-Black Monday crashes as negative bubbles.
3
Hoopes et al. (2016) use tax return data to identify who sold during the 2008–09 crisis. There are also studies examining investor response to
terror events and natural disasters, which are potential triggers of negative bubbles (e.g., Wang and Young, 2020; Bourdeau-Brien and Kryzanowski,
2020).

https://doi.org/10.1016/j.pacfin.2023.102044
Received 13 August 2022; Received in revised form 25 April 2023; Accepted 29 April 2023
Available online 3 May 2023
0927-538X/© 2023 Elsevier B.V. All rights reserved.
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

With the outbreak of the COVID-19 pandemic (COVID hereafter), world stock markets formed a negative bubble, which, as we
show below, led to substantial wealth transfers among investor types. This episode presents an interesting opportunity to identify
investor types’ roles during a remarkable worldwide negative bubble. While the return behavior of financial instruments during the
COVID period has been extensively investigated, research on investor types’ trading behavior is quite limited4; a broad worldwide
picture of investor types’ aggregate trading during the COVID episode is still missing.
This article provides the first comprehensive documenting of investor types’ trading in an actual worldwide negative bubble. As a
counterpart of Griffin et al. (2011), we follow a similar exposition, starting with presenting descriptive evidence to answer the
following questions: who sold into the COVID crash in March–April 2020; who bought the dip; who contributed to and benefited from
the subsequent rebound; how investor types’ aggregate trading affected their wealth? This is followed by an inquiry into what factors
drove their reaction and the implications regarding investor types’ sophistication. We also document a change in the investor-type
structure of stock markets, a substantial increase in individual investor presence, brought about by COVID. Finally, we discover a
new pattern, asymmetric negative feedback trading by individual investors, whose increased role in stock market dynamics is likely to
endure going forward.
Our primary contribution is compiling a unique collection of worldwide trading data disaggregated by investor type, to enable a
generalizable analysis of investor trading around a worldwide negative bubble. Our dataset comprises a combination of precise daily or
weekly data from all major international (mostly East Asian) stock exchanges that provide such data and available proxies from the US.
The latter includes weekly trader positions in equity index futures and Robinhood client numbers for each stock.
Available data point to a worldwide pattern: Individual investors were the main buyers during the COVID episode in every country
analyzed. Foreign institutional investors heavily sold and indiscriminately exited host markets, in line with the ‘when in peril,
retrench’ phenomenon suggested by Broner et al. (2006). Some domestic institutional investors exploited the negative bubble with
well-timed buying. In the US index futures, asset managers heavily sold into the crash in March 2020. Their net selling was absorbed by
individual investors and dealers. Dealers successfully captured the rebound with sophisticated timing.
Retail individual investor interest, as proxied by the number of Robinhood clients holding S&P500 stocks and further corroborated
by individual investor statistics from many countries, displayed a sharp acceleration, literally an explosion, during the COVID period.
Timing of this retail investor demand shock coincided with, and enabled them to capitalize on, the rebound and the subsequent
prolonged uptrend.
Our estimates of the incremental effect of investor types’ trading on their wealth during the COVID episode indicate significant
wealth transfers. In the US index futures, dealers profited from sophisticated timing at the expense of asset managers. In international
markets, individuals and domestic institutions were the main winners at the expense of foreign institutions. The aggregate individual
investor’s winner status on a global basis in this episode is in stark contrast to previous characterizations of individual investors as
noise traders (Schmeling, 2007; Barber et al., 2009; Chang et al., 2015; Choi et al., 2020).
Griffin et al. (2011) find that it was institutional investors who both drove and burst the tech bubble (against media reports pointing
to individual investors) and the burst was associated with a coordinated pivoting by institutions while individuals initially continued
buying. The COVID episode is similar in that institutions drove the negative bubble, but differs in that the aggregate individual investor
was the main profiter, and the correction was not associated with a coordinated pivoting by institutions. The latter suggests that the
COVID negative bubble cannot be accounted for by theories of delayed arbitrage due to arbitrageurs’ synchronization problem (Abreu
and Brunnermeier, 2002, 2003).
Next, we explore drivers of investor types’ trading during the COVİD episode, in particular, how individual investors’ net buying
led to superior performance and what drove foreign institutional investors’ sustained selling. Individual investors’ buying during the
crash is consistent with their previously-documented contrarian trading behavior (Griffin et al., 2003; Kaniel et al., 2008),5 but
contradicts with increased risk aversion following catastrophic events documented in numerous studies (Wang and Young, 2020;
Bourdeau-Brien and Kryzanowski, 2020; Hasso et al., 2020) and findings of a high (low) level of sophistication of institutional (in­
dividual) investors in reacting to transient negative news (Carpentier and Suret, 2021). We find that individual investors’ buying was
driven by a sequence of their usual contrarian trading behavior followed by a unique positive shock to retail investor demand for
equities. Contrarian behavior is further confirmed in stock selection: individual investor buying was concentrated in COVID-vulnerable
stocks, as opposed to COVID-beneficiary stocks. The positive shock was probably due to a unique combination of excessive worldwide
monetary easing, fiscal stimulus and ample free time to trade stocks due to lockdowns and a new work-from-home lifestyle. The lucky
coincidence of this sequence with a V-shaped rebound led to a peculiar outcome and helped individual investors exploit the specific
term structure of risk premia documented by Berkman and Malloch (2023).
Japanese data reveal that cash versus margin trading was also a driver of individual investor behavior: margin traders failed to buy
the dip during the COVID crash, but later benefited from the prolonged rebound, while cash traders successfully picked the bottom but

4
Bing and Ma (2021) and Zhang et al. (2021) study foreign and domestic institutional investors’ trading in China using limited proxies (Chinese
stock markets do not provide exact daily trading data). Djalilov and Ülkü (2021) cover individual investors in Russia, while Naik et al., 2022 cover
institutional investors in India.
5
What drives their contrarian trait? One argument is that individual investors tend to place ‘out-of-money’ limit orders, which leads to
endogenous contrarian trading elicited by institutional liquidity consumption (Linnainmaa, 2010). This could explain specifically the contempo­
raneous negative relationship between their net trading and returns. Behavioral drivers such as the disposition effect, belief in mean reversion (e.g.,
gambler’s fallacy) can also drive the negative relationship with past returns. Onishchenko and Ülkü (2022) summarize all these effects under a
single concept: “uninformed attempt to buy low and sell high using recent prices as heuristic reference points”.

2
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

exited early.
Further analysis to identify skill in exploiting a negative bubble (and to distinguish it from an exogenous positive shock to indi­
vidual investor demand for equities) suggests that some domestic institutions may have acted specifically to exploit the negative
bubble in March–April 2020. There is some weak evidence that individual investors also exploited the negative bubble. Yet, a positive
shock to individual investor demand is much more significant statistically, implying that a lucky coincidence was a main driver of
individual investors’ success in this episode.
Foreign institutions’ exit from host markets and asset managers’ heavy selling of US equity index futures into the crash in March
2020 were, at least partly, driven by client outflows. Significant outflows from equity funds clustered in March 2020. The aggregate
household, who rushed into equities, sold equity funds; implying a sharp divergence between fund clients’ and self-serviced individual
investors’ behavior. We conjecture that excess free time due to lockdowns and a new work-from-home lifestyle induced individual
investors to shift from delegated investment management toward self-servicing. Data from subsequent periods shows that fund clients
returned back later after the full rebound when markets headed to new highs. Thus, aggregate fund client flows displayed poor timing,
consistent with the ‘dumb money’ tag proposed by Akbaş et al. (2015).
The notable increase of individual investor presence in stock markets during the COVID period has prompted renewed interest in
individual investors’ aggregate trading behavior.6 While individual investors have been associated with buying frenzies, our results
confirm that contrarian trading is the dominant behavior of the aggregate individual. Such contrarian trading yielded profits by
counteracting institutional selling pressure during the COVID crash, as it did during the 2008–09 crisis (Barrot et al., 2016). The
contrarianism of the aggregate individual investor once again served as a stabilizer.
Last but not least, we uncover a pervasive asymmetry in individual investor reaction to positive and negative returns: Contem­
poraneously, they are less contrarian to negative returns than to positive ones. With respect to previous days’ returns, however, they
are more contrarian to negative returns. In other words, contrarian dip-buying intensifies with a lag. This asymmetry is present in the
full sample, but dominates market behavior under easy money conditions with an influx of individual investor money and charac­
terizes rebounds seen following every sell-off during the post-COVID period.
This article is structured to provide a complete picture of trading behavior during the COVID episode, to serve as a reference point
for future deeper analysis of the specifics. Section 2 describes trading data brought together in this study. Section 3 provides a graphical
documentary of investor types’ aggregate trading during the COVID episode, and estimates the incremental effect of their trading on
their wealth. Section 4 analyzes drivers of investor types’ trading. Section 5 summarizes the conclusions.

2. Investor trading data

We use all available trading data disaggregated by investor type from around the world excluding small frontier markets. In the
existing literature, lack of data has constrained the task of deriving a clear picture of investor types’ role in actual bubbles. To identify
investor types’ aggregate role in a bubble, complete and exact market-wide data from centralized electronic records of all transactions
are needed, as opposed to limited samples from one brokerage house or proxies derived under assumptions.
Such exact and complete data are provided by only few stock exchanges. Among them, Korea (KRX) offers the world’s most
comprehensive daily trading dataset by investor type, along with a breakdown of domestic institutional investors, which helps
distinguish motivations behind institutional investors’ trading. Taiwan (TWSE) and Thailand (SET) also publish precise daily data of
investor types’ trading. Japan (JPX) offers weekly data from electronic surveys, with a breakdown of domestic institutions.
As cultural factors might influence investor behavior, evidence from outside East Asia is needed for generalizing results. In
particular, east Asian individuals are reported to have a stronger belief in mean reversion (Shu et al., 2005; Arkes et al., 2010), which
may have driven findings of contrarian trading by individuals in the literature. Serving this purpose, we use novel data from Borsa
Istanbul, Turkiye (BIST) and Pakistan Stock Exchange (PSX), both for the first time in the literature. BIST is highly correlated with, and
share similar participants as, the European stock markets. We also obtain weekly data of individual investor trading from Russia
(MOEX).7 Our sample countries represent a range of Hofstede individualism scores,8 ranging from as low as 14 for Pakistan,18 for S.
Korea and 20 for Thailand to 37 for Turkiye, 39 for Russia, 46 for Japan and 91 for the US.
In these datasets, three main investor types are identified: individuals, domestic institutions and foreign investors (predominantly
institutional),9 which we denote as Indiv, Inst and For, respectively. Our data comprises daily values (in home currency) of each
investor types’ marketwide-aggregated purchases and sales. We compute the daily net trading flow of each investor type (Fm t ) as
purchases minus sales by investor type m on day t, normalized by current free-float market capitalization. Thus, Fm t represents the size
of net trading as a proportion of the value of outstanding shares.
In the US, no complete records of investor types’ trading are kept, and the fragmented structure of the US stock market precludes
compiling such data. We utilize two data sources to obtain at least a partial picture of investor types’ trading in the US: i) Weekly long/
short positions in equity index futures from the weekly Traders in Financial Futures survey by the US Commodity Futures Trading

6
Several price events, such as short-squeezes in meme-stocks, have been attributed to individual investor crowds.
7
Mirror image of individual investors’ trading can be taken to represent domestic and foreign institutions combined. Russia data starts from
January 2018.
8
Individualism has been suggested as a driver of financial behavior; e.g., Foley et al. (2022).
9
Wherever a break-down of foreign individual and institutional investors is available, we leave out foreign individuals which constitute a
negligible proportion of total traded value (< 0.0008 in Japan, < 0.002 in Turkiye, < 0.007 in Pakistan).

3
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Commission (CFTC), ii) Investor positions from retail broker Robinhood tracker data. CFTC data classifies open interest positions held
by five categories: asset managers (AM), leveraged funds (i.e., hedge funds, HF), dealer/intermediary (DL), other reportables (OR) and
non-reportables (NR). NR is a residual category that contains all smaller non-reporting traders, presumably dominated by individual
investors.10 Robinhood data reports the number of clients holding each particular stock. As such, it is more an indicator of the popular
interest of retail investors,11 while NR in CFTC data quantifies relatively more-experienced and large-sized individuals’ trading.
Consistent with this, we find that Robinhood flows are positively correlated with the individual investor sentiment reported by AAII
(American Association of Individual Investors) while NR is uncorrelated (see in Section 4).
Proxies for Fm
t in the US data are as follows: For the CFTC data, weekly net position changes, computed as the number of S&P500
index (ES) futures contracts normalized by open interest. For Robinhood data, daily changes in the number of clients holding S&P500
index constituent stocks (dRH); we also replicate for SPY and VOO (most popular S&P500 index-tracking ETFs) and for portfolios of
COVID-beneficiary and COVID-vulnerable stocks.
Robinhood data are available per stock. We also employ stock-level data from Korea and Turkiye. Stock-level data enables us to
perform a difference-in-difference analysis of stock selection behavior in response to the COVID news.
Our sample period ends in mid-November 2020; thus, it covers the period in which the rebound was completed without the actual
reopening of the economy (i.e., the negative bubble was corrected solely on expectations) and excludes the effect of the Biden-era extra
stimulus, so that the analysis of trading in relation to the negative bubble is not distorted by subsequent news. We employ a pre-COVID
period, defined as 4.January.2010–21.February.2020, as a benchmark of regular trading behavior. 24.February.2020–20.
November.2020 is designated as the COVID period. (24.February is when selling in stock markets began upon early signs of COVID.
Even though COVID attracted public attention in March, it is imperative to monitor trading from when smart money may have started
to act). Robinhood data are available from 3.May.2018 to 13.August.2020. A brief description of the data and summary statistics are in
Appendix Table A.1. A practical glimpse of investor types’ trading is provided in Table 1 with statistics using net trading normalized by
total traded value.
Auxiliary data: To address specific questions about drivers of investor behavior, we also employ the following sets of data:

i) Individual investor statistics: based on a worldwide search, we employ all centralized/nationwide statistics on individual
investor account openings and demographics available as of the time of writing this paper. For the US, we employ statistics from
a leading online discount broker.
ii) Weekly data on client flows toward mutual funds, reported by three different sources: EPFR, Refinitiv-Lipper and Investment
Company Institute (ICI, monthly).
iii) Weekly (American Association of Individual Investors) AAII individual investor sentiment survey.

3. Investor types’ aggregate trading during the COVID-19 episode

3.1. A graphical documentary of cumulative trading around the COVID period

In this section, we address the following questions, which naturally involve descriptive evidence: Who sold into the COVID negative
bubble? Who bought the dip in late March – early April 2020? Who drove and/or benefited from the subsequent rebound? These
questions are collectively and concisely answered by graphs in Fig. 1, which plot the stock market index levels along with cumulative
net trading by main investor types (i.e., accumulated Fm t ) from the end of 2019 to November 2020 in KRX (Korea), TWSE (Taiwan), SET
(Thailand), BIST (Turkiye), JPX (Japan), MOEX (Russia) and PSX (Pakistan), respectively. These graphs provide a complete broad
glimpse of how investor types responded to the COVID negative bubble.
First, identify the negative bubble in main stock indices (the black line) in all graphs. As Gürkaynak (2008) concludes, “econometric
detection of asset price bubbles cannot be achieved with a satisfactory degree of certainty”. Two approaches to identifying bubbles are
based either solely on price pattern or on deviation from fundamentals. In the former approach, full rebound following a substantial
crash, especially if the rebound unfolded in the absence of news surprises, is taken as ex-post evidence of a negative bubble. Based on
this definition, the black lines show that every stock market in our sample had a negative bubble.12
Regarding investor behavior with respect to the negative bubble, in all countries a common pattern emerges: Individual investors’
cumulative net buying started to rise with the outbreak of COVID, accelerated during the crash in March–April 2020 and kept rising
during the rebound. Foreign investors started to sell with the outbreak and continued selling during the rebound. Their selling was
indiscriminate: for example, they sold Taiwan stocks despite the country’s success in containing the pandemic, although the US, the
main source of equity portfolio flows, suffered the most. Domestic institutions were net buyers, notably in Thailand, Turkiye, Pakistan
and Japan, albeit in relatively small sizes; the main offset to foreign selling was provided by individual investors.
Panel A of Fig. 2 plots cumulative net trading by subcategories of domestic institutions in Korea. Korean pension funds successfully

10
See https://www.cftc.gov/MarketReports/CommitmentsofTraders/index.htm for further information. Our data is retrieved from the ‘Traders in
Financial Futures’ section.
11
See Welch (2020) and Pagano et al. (2021) for concurrent studies of Robinhood data.
12
While one may argue that the strong response of the monetary policy was a surprise, strong policy support in response to natural disasters,
especially in the case of a global pandemic, was a sure thing. Below we also provide an assessment regarding the second approach (deviation from
fundamentals) looking at the price behavior of COVID-beneficiary stocks.

4
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Table 1
Descriptive statistics of net trading by investor types as a proportion of trading volume.
Pre-COVID period: COVID period:

Country Variable Mean Median St.Dev. Mean Median St.Dev.

4 Jan 2010–21 Feb 2020 24 Feb - 20 Nov 2020

Indiv − 0.0041 − 0.0015 0.0447 0.0169 0.0164 0.0454
Korea
Inst 0.0002 − 0.0014 0.0420 − 0.0050 − 0.0074 0.0300
For 0.0034 0.0031 0.0455 − 0.0130 − 0.0064 0.0340
4 Jan 2010–21 Feb 2020 24 Feb - 20 Nov 2020
Indiv − 0.0017 − 0.0078 0.0760 0.0165 0.0111 0.0733
Taiwan
Inst − 0.0012 − 0.0012 0.0062 0.0011 0.0012 0.0050
For 0.0041 0.0090 0.0691 − 0.0138 − 0.0072 0.0641
4 Jan 2010–21 Feb 2020 24 Feb - 20 Nov 2020
Indiv 0.0008 0.0008 0.0591 0.0178 0.0193 0.0478
Thailand
Inst 0.0052 0.0046 0.0398 0.0044 0.0051 0.0393
For − 0.0061 − 0.0056 0.0494 − 0.0228 − 0.0219 0.0324
4 Jan 2010–21 Feb 2020 24 Feb - 30 Oct 2020
Indiv − 0.0004 0.0001 0.0202 0.0042 0.0035 0.0111
Turkiye
Inst 0.0000 0.0004 0.0103 0.0015 0.0006 0.0089
For 0.0004 − 0.0004 0.0200 − 0.0057 − 0.0044 0.0079
24 Jan - 21 Feb 2020 24 Feb - 20 Nov 2020
Indiv 0.0030 0.0101 0.0318 0.0100 0.0088 0.0429
Pakistan
Inst 0.0437 0.0432 0.0452 0.0288 0.0258 0.0499
For − 0.0453 − 0.0370 0.0474 − 0.0395 − 0.0308 0.0403
3 Jan 2011–21 Feb 2020 24 Feb - 20 Nov 2020
Indiv − 0.0023 − 0.0028 0.0081 − 0.0003 0.0009 0.0076
Japan
Inst 0.0037 0.0031 0.0089 0.0042 0.0036 0.0078
For − 0.0013 − 0.0021 0.0108 − 0.0034 − 0.0031 0.0096
2 Jan 2018–21 Feb 2020 24 Feb - 20 Nov 2020
Russia
Indiv 0.0037 0.0031 0.0252 0.0160 0.0221 0.0420
3 May 2018–21 Feb 2020 24 Feb - 13 Aug 2020
US Robinhood
dRH 0.0000014 0.0000013 0.0000021 0.0000141 0.0000130 0.0000098
4 Jan 2010–18 Feb 2020 19 Feb - 17 Nov 2020
NR 0.0003 0.0004 0.0082 − 0.0007 0.0007 0.0122
US CFTC AM 0.0001 0.0004 0.0098 − 0.0004 0.0000 0.0091
DL − 0.0002 − 0.0005 0.0088 0.0019 − 0.0006 0.0131
HF − 0.0003 − 0.0001 0.0104 − 0.0003 − 0.0008 0.0074

Indiv, Inst and For stand for net trading of individual, domestic institutional and foreign investors, respectively, as a proportion of total traded value of
the equity market. For the US Robinhood data, net trading is approximated by the daily change in the number of investors holding S&P500 stocks
(dRH). For the US CFTC data, net trading is approximated by the net change in futures position (NR = Nonreportables, comprising mainly individual
investors, AM = Asset Managers, DL = Dealers, HF=Hedge Funds. CFTC weeks start on Wednesday and end on Tuesday).

executed a well-timed contrarian strategy and took profits after the rebound. Other domestic institution subcategories’ trading during
the COVID is trivial.13 Pension funds, which have sufficiently long horizons free of the threat of client mass-redemptions, were able to
implement a successfully-timed strategy to exploit the negative bubble. One possibility is that pension fund clients switched to equity
funds and drove this shift. (No data are available to examine this possibility). However, mutual funds and ETFs (‘Investment Trusts’)
were rather net sellers. Redeemability seems to account for the difference between pension funds’ and investment trusts’ trading, as
nonredeemable pension fund clients may have switched to equity funds, while mutual fund clients took their money to self-invest.
Trading behavior of hedge funds, often called ‘smart money’, has been a focus of interest in the literature; for example, Akbaş et al.
(2015) find that hedge fund flows tend to attenuate anomalies. Korean hedge funds, however, played little role during the COVID
episode, and rather, by selling into rebound they counteracted the correction from the negative bubble. (Below, we revisit this issue
when discussing US hedge funds’ index futures trading).
Panel B of Fig. 2 provides a selected breakdown from Japan and yields the following observations:

i) Japanese trust banks (a special type of investment banks providing custody services for clients) started net-buying around the
bottom of the negative-bubble and fully exploited the rebound, indicative of a sophisticated strategy.
ii) Foreign individuals’ trading was totally different from that of foreign institutions: foreign individuals exploited the negative
bubble just like domestic individuals. This result is further reinforced on the other markets that provide a breakdown of foreign
individuals versus institutions: Fig. OA.2 in Supplementary Material 2 shows that in Turkiye foreign individuals were net buyers
in sharp contrast to foreign institutions.

13
Brokers/dealers (named ‘Financial Investment’ by KRX) engaged in heavy selling throughout 2020, which appears to have a component un­
related to the COVID crisis. This heavy selling offsets pension funds’ well-timed buying in the aggregated domestic institution trading.

5
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Fig. 1. Main investor types’ cumulative net trading during the COVID episode.
Investor types’ cumulative net trading as a proportion of market capitalization are on the left-vertical axis (set to zero at the beginning of 2020).
Indiv: domestic individuals, For: foreign investors, Inst: domestic institutions. In all graphs below consistently: orange represents individuals; yellow
represents domestic institutions; grey represents foreign investors. An increase (decrease) implies net buying (selling) on the day it occurs. Market-
index values (black curve) are on the right-vertical axis.

6
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Fig. 1. (continued).

7
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Fig. 1. (continued).

8
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Fig. 1. (continued).

iii) A breakdown of cash versus margin trading of individual investors indicates sharp differences in trading behavior: Cash in­
vestors pursued a contrarian strategy; were significant net buyers at the bottom of the negative bubble; and then sold during the
rebound. They did not fully exploit the rebound and their cumulative net trading turned into deeply negative when the rebound
was completed. On the other hand, margin traders initially bought at the beginning of the sell-off, which they had to reverse
around the bottom and the first leg of the rebound14; then during the recovery they gradually built-up net purchases again.
Thus, cash investors’ early exit during the rebound offset their profits from successfully picking the bottom of the negative
bubble, while margin investors missed out on picking the bottom and rather joined at the later stages of the rebound.

Existing literature on cash versus margin traders’ behavior is limited. Shu et al. (2005) find that margin traders in Taiwan display
muted disposition effect, consistent with our observation on Japan. We also observe that margin traders’ forced selling during a crash
can reduce the measured disposition effect.
As to the US, cumulative S&P500 futures position changes in Fig. 3 indicate that asset managers sold into the crash in March 2020
and did not contribute to the early stages of the rebound, which they joined at much later stages. Nonreportables, which contain
individual investors, were the most net buyers in the early weeks of the crash in March; they reduced part of their long positions with a
poor timing at the bottom of the crash (likely, another case of forced selling due to futures leverage), and kept remaining long positions
through the first leg of the rebound. Dealers were the most net buyers at the bottom of the negative bubble as well as during early stages
of the rebound. They started reducing long positions only after fully exploiting the rebound. Hedge funds were minor net buyers during
the crash, and minor net sellers (buyers) in early (late) stages of the rebound. Similar to Korean hedge funds, US hedge funds’ index
futures trading during the COVID episode is relatively insignificant.
Results for Nasdaq and Russell-2000 index futures (NQ and RTY, respectively) are broadly similar: Dealers’ sophisticated
contrarian buying is particularly notable in RTY (presented in Supplementary Material 1, Fig. A1), which contains smaller stocks
vulnerable to the pandemic. As the literature on CFTC investor types’ ability to exploit market dislocations is scarce, our finding opens
up a potential pattern, Dealers’ sophisticated timing, for further investigation.
Aggregated number of Robinhood clients holding S&P500 index-constituent stocks in Fig. 4 displayed a sharp acceleration starting
with the crash in March. Graphs of the number of Robinhood clients holding SPY and VOO ETFs (Fig. A.2 in Supplementary Material 1)
are similar. Notice that the previous market dip in December 2018 did not trigger a similar acceleration in the number of Robinhood
clients, suggesting that the unique jump in the COVID period cannot be explained by contrarian trading alone. This provides the first
evidence that the lockdowns combined with monetary easing had a unique effect on individual investor participation, which in Section
4 we identify as a unique exogenous positive shock to retail investor demand for equities.

14
Foley et al. (2021) identify raised margin requirements as a potential driver of forced selling.

9
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

(caption on next page)

10
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Fig. 2. Cumulative net trading by subcategories of domestic institutions during the COVID episode.

To summarize, individual investors were net buyers during the COVID episode and rode the rebound. Asset managers in the US sold
the dip. Foreign institutional investors in host markets not only sold the dip but also continued selling during the rebound. Some
domestic institutions and Dealers in the US futures undertook well-timed buying to exploit the negative bubble.

3.2. The effect of investor types’ trading during the COVID period on their wealth

The accumulated net trading of investor types at each point in time, the variable in the above graphs, can be used to obtain an
estimate of the monetary value of the incremental effect of investor types’ trading on their wealth from that point onwards. This
incremental effect is measured in comparison with the status quo alternative where they did not trade. Based on this, we construct a
performance metric (PM):
( )
∑T ∑
t− 1
PM = Ft Rt (1)
t=1 t=1

where Rt is the return of the market index, T is the final day in the sample period and 24 February 2020 is set as t = 1. The performance
metric in Eq.(1) is a rough approximation as it assumes that investor types’ net trading is executed in the same composition as the index
portfolio (this assumption is certainly true for the S&P500 index futures) and at the closing index level (this effect becomes significant
whenever weekly data is used, in which case we make an adjustment by using average daily data).
The resulting estimates, converted into US dollars at the average exchange rate of the period, are reported in Table 2. The main
message is that individual investors’ trading during the COVID period had an economically significant positive incremental effect on
their wealth, at the expense of foreign investors. In many countries, domestic institutional investors also made significant incremental
gains as a result of their trading. Subcategories, for which we detected interesting trading patterns above, end up with notable per­
formance outcomes (not in the table): In Korea, pension funds had a significant gain of $ 1.457 bln. In Japan, trust banks gained $ 2.61
bln.
The breakdown of Japanese individuals as cash versus margin traders (not in the table) provides an interesting observation: Cash
traders successfully exploiting the negative bubble had a peak profit at $599 million in June 2020 when margin traders, forced to sell
during the crash, were losing money. Margin traders’ loss hit a maximum in August 2020 at − $639 million and then started to recover.
By the third week of November, margin traders reached a profit of $220 million while cash traders moved into a loss of $ − 406 million
due to early exiting, missing a large portion of the rebound. (Table 2 reports cumulative profits/losses as of the end of our sample
period. Figures depicting the evolution of these profits over time are available upon request.) Japanese aggregate individual investor’s

Fig. 3. Investor types’ positioning in S&P500 E-mini index futures during the COVID episode.
Investor categories’ cumulative net position changes as a proportion of total open interest are on the left-vertical axis. S&P500 index values are on
the right-vertical axis. DL: Dealers, AM: Asset managers, HF: Hedge funds, NR: Nonreportables (dominated by individual investors).

11
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Fig. 4. Number of Robinhood clients holding S&P500 stocks during the COVID episode.
Aggregated number (in millions) of Robinhood clients holding S&P500 index constituent stocks is on the left-orange axis. S&P500 index levels are
on the right axis. A bullishness indicator derived from AAII individual investor sentiment surveys (% bullish minus % bearish) is on the left-green
axis (referred to in Section 4.2). To illustrate the pre-COVID period trend, this graph covers all available Robinhood data. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)

Table 2
Incremental effect of investor types’ trading on their wealth during the COVID period.
Indiv Inst For Prop.

Korea 8.934 − 1.191 − 8.470 –

Taiwan 7.53 0.538 − 6.856 − 0.451
Thailand 0.694 0.302 − 1.000 0.005
Turkey 0.653 0.168 − 0.833 –
Pakistan 0.030 0.067 − 0.100 –
Japan − 0.186 7.870 − 6.672 5.944
Russia 0.550 – – –
DL AM HF NR
US CFTC 30.242 − 18.731 − 0.803 − 2.065

Estimates are obtained using the performance metric in Eq.(1). Reported figures are in billion US dollars, computed in local currency and converted to
US$ at the period-end exchange rate.

cumulative incremental profit thus peaked earlier at $ 0.297 billion before falling to $ − 0.186 billion reported in Table 2.
In the US index futures, the clear winners are Dealers with an incremental profit above $30 billion. Individual investors, as rep­
resented by NR, did not manage to profit from their contrarian trading: they initially suffered sharp losses buying at the beginning of
the crash, which they just recovered during the first wave of the rebound; but they closed long positions in June to miss the rest of the
rebound. Main losers were asset managers, who sold into the crash and bought back only in late stages of the rebound.

3.3. Has the investor-type structure of stock markets changed during the COVID episode?

In Table 3, we compare investor types’ share in total traded value as of the last month in our sample period with the same month of
the previous year; i.e., an October 2019 versus October.15

15
Comparisons using earlier months (September, August) yield similar outcomes.

12
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Table 3
Investor types’ share in total traded value.
Oct - 2019 Oct - 2020

Market Indiv Inst For Indiv Inst For t (dif)

Korea 0.470 0.222 0.297 0.643 0.179 0.166 7.24

Taiwan 0.664 0.060 0.276 0.665 0.048 0.287 0.13
Thailand 0.320 0.273 0.406 0.410 0.222 0.368 6.33
Turkey 0.645 0.067 0.286 0.746 0.047 0.206 9.83
Japan 0.134 0.252 0.614 0.175 0.211 0.614 4.04
Russia 0.360 0.149 0.491 0.417 0.114 0.469 3.26

Oct - 2019 Oct - 2020

DL AM HF NR DL AM HF NR t (dif)

US - CFTC 0.185 0.458 0.121 0.138 0.184 0.425 0.135 0.147 1.04

This table compares proportions of traded value accounted for by each main investor type, derived from our complete trading flows data, in
October.2019 versus October.2020. (For the US CFTC, share of each investor type in total non-spreading long and short positions). t (dif) is a t-statistic
that tests the null-hypothesis that individual investors’ share in October 2020 was equal to that in October 2019.

In all countries with two exceptions, significant increases in individual investors’ share in traded value are observed.16 Thus, the
participant composition of world equity markets has changed over the COVID period such that the share of individual investors has
increased notably, consistent with anecdotal reports.

3.4. Investor trading in COVID-vulnerable versus COVID-beneficiary stocks

Stock-level data, available from the US Robinhood and two countries in our sample (Korea and Turkiye), allow us to contrast
investor types’ trading behavior in COVID-vulnerable and COVID-beneficiary stocks, in order to gain insight as to the underlying
motivations of individual investors’ trading. For this purpose, we form equal-weighted portfolios of COVID-vulnerable stocks (airlines,
hotels, cruise lines, restaurant chains, oil drillers, refineries, etc.) and COVID-beneficiary stocks (online retailers, telecommunication,
remote-work technologies, health services, pharma, some IT services and home entertainment businesses). Where possible, our se­
lection of COVID-vulnerable and COVID-beneficiary stocks is based on direct knowledge of the company’s products (e.g., Zoom,
Moderna); where such information is ambiguous, we have sought additional confirmation from price action.17 This procedure yields
the following numbers of COVID-vulnerable versus COVID-beneficiary stocks, respectively, in each market: 20 versus 55 stocks for the
US, 15 versus 27 for Korea, and 8 versus 7 for Turkiye.
Panel A of Fig. 5 plots accumulated changes in the number of Robinhood clients holding stocks included in these portfolios against
the cumulative returns of these portfolios. The message from Panel A is twofold: First, during the crash in March 2020, even COVID-
beneficiary stocks experienced significant initial price declines, in negative relation with their future fundamentals. This validates
arguments for a negative bubble based on deviations from fundamentals (the second approach to bubble detection).
Second, retail investor buying interest in COVID-vulnerable stocks with negative fundamentals was much stronger than that in
COVID-beneficiary stocks during the March–June period. The graphs clearly indicate that the divergence was triggered by the COVID
outbreak. This difference can be explained by behavioral biases of individual investors: contrarian trading, belief in mean reversion,
and attention-driven buying (Barber and Odean, 2008). In particular, drastic events make the most-severely affected stocks attention-
grabbing, attracting individual investor attention and triggering buying behavior, providing fresh case-based evidence consistent with
Barber and Odean (2008) theory. COVID-vulnerable stocks’ continued underperformance during the next 3–6 months supports a
behavioral bias interpretation, given that retail investor buying in COVID-vulnerable stocks flattened before a relative rebound. Yet, a
longer-horizon play for the anticipated eventual full recovery cannot be ruled out.
Panel B of Fig. 5 plots individual investors’ cumulative net trading in COVID-vulnerable and COVID-beneficiary stocks in Korea and
Turkiye. The outcome is similar: Consistent with our first conclusion on the US stock market, COVID-beneficiary stocks also experi­
enced significant initial price declines, validating a negative bubble interpretation. Consistent with our second conclusion based on
Robinhood investors, net individual investor buying in COVID-vulnerable stocks was much stronger than in COVID-beneficiary stocks

16
The two exceptions are Taiwan, where individual investors were already dominant with the number of investment accounts at around 40% of
the population, and the US index futures where the NR category representing large, experienced individual futures traders was barely affected by the
influx of the COVID-generation.
17
To be included in the COVID-vulnerable (− beneficiary) portfolio, a stock must have significant negative (positive) market-adjusted return
during the 24.February – 31.March 2020 period. We also restrict the selection to the constituents of the S&P500, KOSPI-200 and BIST-100 index,
respectively, unless news about the firm’s product made the stock of special interest for traders during the COVİD period.

13
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Fig. 5. Investor trading in COVID-vulnerable and COVID-beneficiary stocks.

In each market, equal-weighted portfolios of COVID-vulnerable and COVID-beneficiary stocks are constructed and their return indexes are set to 100
at the beginning of 2020 (red and green dashed lines, respectively; left-vertical axis).

14
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Fig. 5. (continued).

in both countries.
It is interesting to note that the divergence in individual investors’ net trading in the two portfolios started early in both Korea and
Turkiye.18 In both markets as well as the US, the divergence in cumulative returns started during the second half of January, but
remained minor until the last week of February. As a result, a notable part of individual investor net buying in COVID-vulnerable stocks
was performed before the global crash in March. It seems that small price concessions lured individual investors before the crash. This
particular point lends additional support to behavioral drivers of individual investors’ trading.
The cumulative net trading of domestic and foreign institutions (not included in order to keep the graphs tractable; their sum is the
mirror image of individual investors) indicates strong net selling of COVID-vulnerable stocks and flat to mild selling of COVID-
beneficiary stocks in both Korea and Turkiye. This suggests that institutional investors responded to information besides the pres­
sure of outflows.

4. Understanding drivers of investor trading behavior during the COVID period

We now investigate drivers of investor types’ significant trading that resulted in wealth transfers portrayed above. Candidate
explanations are differences in sophistication (i.e., successfully exploiting versus joining a negative bubble), a coincidence of the price
dynamics of the episode with behavioral (e.g., feedback trading) traits, or a unique shock to investor types’ demand for equities. In this
section, we distinguish potential explanations using available data and tools by utilizing an empirical setup that accommodates well-
known time-series properties of trading flows (i.e., persistence) and their interaction with returns (i.e., feedback trading).
Within a model that fully accounts for these patterns, nonlinear responses to negative bubble levels or exogenous shocks to investor
types’ demand for equities can be identified by differential intercept terms. For this purpose, we create two dummy variables: i) an ex-
post identified COVID negative bubble dummy, Bt, which takes the value of 1 when the market index is a certain percentage below its
average value during the two months before COVID hit and 0 otherwise. ii) a COVID-period dummy, Ct, which takes the value of 1 after
the first lockdown took effect in the respective country and 0 before, to capture the period in which individuals started to adapt to a
new lifestyle with free time to monitor and trade stock markets.

18
It is possible to attribute the early divergence in Korea to its geographical proximity to China which experienced an initial wave ahead of the rest
of the world, while in Turkiye domestic individual investor buying seems to be driven by the domestic low interest rate policy coincident with
foreign investor selling.

15
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

4.1. Any skills of identifying a negative bubble?

While aforementioned studies of investor behavior following transient catastrophic events find de-risking and a reduction in
trading activities, negative bubbles discussed by Shiller (2000) and Goetzman and Kim (2018) imply that such events offer highly
profitable opportunities in excess of rational risk premiums. Hence, transient catastrophic events offer unique assessments of investor
sophistication (studies linking investor response during such events to the degree of sophistication are common, e.g., Carpentier and
Suret, 2021).
To find out whether the negative-bubble levels hit during the COVID crash in March 2020 had a nonlinear association with investor
types’ net trading, we test the significance of the negative-bubble dummy Bt in each investor type’s augmented net trading (Fm t )
equation.
∑k ∑k
(2)
′
Ftm = ω + ω Bt + ρ F m + λRt +
i=1 i t− i i=1
θi Rt− i + ε1,t

where Rt is the market index return and k = 5 with daily data and k = 1 with weekly data (further lags turn out insignificant). Eq.(2)
controls for autocorrelation in net trading (captured by ρ coefficients) and net trading’s response to contemporaneous and past returns
(captured by λ and θ coefficients, respectively). Where an investor types’ net trading is highly positively-correlated with contempo­
raneous returns (which is the case for foreign investors and in some markets for domestic institutions) raising the possibility of price
impact driving returns, we also estimate an alternative version of Eq.(2) excluding the λRt term, to address the simultaneity between
returns and trading flows. This provides a view on the possibility that inappropriately controlling for returns driven by the price impact
of an investor type’s trading interferes with the measured significance of the other terms in Eq.(2).
A significantly positive ω’ coefficient in Eq.(2) would imply skill in identifying and exploiting the COVID negative bubble (i.e.,
significant net buying at negative-bubble levels after controlling for net trading’s regular interaction with returns). A significantly
negative ω’, on the other hand, would imply fomenting (i.e., selling into) a negative bubble. Hereby, we introduce an asymmetric
version of a negative-bubble dummy, taking into account a potential asymmetry in perceived risk: In a V-pattern, the perceived risk can
be higher on the left wing during the sell-off than at the same level on the right wing during the rebound. Hence, perceived-risk-
adjusted expected returns by which a rational investor would identify a negative bubble level might differ at the same index level
on the left and right wings of the V-pattern. To our knowledge, this study is the first to incorporate such asymmetry. The plain version
of Bt takes the value of 1 when the index is 20σ below its 23.Dec.2019–21.Feb.2020 average value, where σ is the pre-COVID standard
deviation of daily returns.19 The asymmetric version uses a threshold of 22σ on the left wing and 18σ on the right wing (reasonable
variation of these parameters yields similar results).
Table 4 reports results from estimating Eq.(2) using the asymmetric version of Bt.20 The first B column on the left reports the
estimated coefficient of the COVID negative-bubble dummy before controlling for F’s interaction with returns; i.e., excluding the terms
with λ and θ in Eq.(2). If this B-coefficient loses its significance in the augmented Eq.(2) on the right block, it would imply that the
investor type’s response to the negative bubble can in fact be accounted for by its regular feedback trading traits.
The most-left column suggests that B-coefficients before controlling for the interaction with returns are significantly positive for
domestic individuals and domestic institutions, and significantly negative for foreign investors in most cases. In others, they are
marginally insignificant with similar signs. (Weaker result for NR in US CFTC data can be due to forced selling of long positions in
margined futures accounts during the sharp sell-off). Thus, before controlling for trading styles, domestic investors appear to have
exploited/counteracted the COVID negative bubble while foreign/institutional investors fomented it.
In the center-right large block, B-coefficients are significantly or marginally-insignificantly positive for domestic individuals and
domestic institutions and negative for foreign investors. Significance levels are reduced compared with the first column for individuals
in Thailand and Turkiye, for domestic institutions in Taiwan, for pension funds in Korea and for DL and NR in the US equity futures,
indicating that regular contrarian traits helped these investor types exploit the negative bubble. For domestic institutions in Turkiye,
who have no significant regular contemporaneous contrarian trading behavior, the B-coefficient rather gains significance after con­
trolling for interaction with returns, indicating that their trading may have particularly aimed at exploiting the negative bubble. In the
last column, excluding the contemporaneous return (i.e., the λRt term) in the F equation strengthens the significance of foreign in­
vestors’ selling into the negative bubble in smaller markets such as Thailand and Turkiye where foreign investors’ price impact is
expected to drive returns.
In summary, the above results provide moderate evidence that domestic individual and institutional investors exploited the
negative bubble in March–April 2020 and foreign institutional investors were sellers at negative-bubble levels. They also suggest that
regular contrarian trading traits helped some investor types to exploit the negative bubble. As a side note, significant F-lag coefficients
confirm the well-known persistence in investor types’ trading, while significant R and R-lag coefficients confirm positive feedback
trading by foreign investors and negative feedback trading by individuals.21

19
20 is a parameter set such that a sufficiently long time-span during which a substantially cheaper index level could be exploited for buying.
There is no existing guidance regarding what this number should be; Goetzman and Kim (2018) experiment with a range of threshold levels on
historical data to yield a desirable sample size of negative bubble cases.
20
Results using the plain version are similar with weaker significance levels in several cases.
21
Results remain robust using a generalized method of moments (GMM) estimation which offers alternative ways of dealing with the simultaneity
between flows and returns, and heteroskedasticity.

16
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Table 4
Significance of the COVID negative bubble in driving investor types’ trading.
Market Investor type B R R lag F lag B B (alt.)

Domestic Individual 0.650 -0.625 0.037 0.242 0.972

(2.44)* (− 13.16) (1.04) (8.26) (2.46)*
Domestic Institutions 0.008 0.222 -0.207 0.253 -0.200 -0.085
Korea (0.04) (6.60) (-7.07) (9.56) (-0.50) (− 0.18)
Foreign -0.287 0.458 0.140 0.351 -0.712 -0.452
(-1.85) (12.00) (4.49) (13.40) (-1.41) (− 1.42)

Domestic Individual 0.235 -0.680 -0.016 0.391 0.422

(1.42) (− 18.83) (− 0.64) (22.03) (1.69)
Domestic Institutions 0.686 0.392 -0.097 0.353 0.538 0.612
Taiwan (3.65)** (15.12) (− 3.62) (13.09) (2.61)* (2.35)*
Foreign -0.174 0.621 0.029 0.423 -0.348 -0.214
(− 1.05) (18.14) (1.09) (25.56) − 1.49 (− 1.52)

Domestic Individual 0.160 -0.614 -0.019 0.198 0.189

(1.16) (− 14.93) (− 0.49) (4.80) (0.52)
Domestic Institutions 0.713 0.497 -0.262 0.228 0.451 0.476
Thailand (2.98)* (17.93) (− 8.81) (11.14) (2.05)* (2.07)*
Foreign -0.400 0.253 0.122 0.362 -0.445 -0.428
(− 2.13)* (6.86) (4.03) (9.67) (− 1.68) (− 2.47)*

Domestic Individual 0.519 -0.675 -0.014 0.131 0.602

(4.20)** (− 17.64) (− 0.72) (5.05) (2.18)*
Domestic Institutions 0.393 -0.032 -0.163 0.108 0.361
Turkiye (1.68) (− 1.17) (− 6.76) (4.96) (2.82)**
Foreign Inst. -0.525 0.648 0.038 0.196 -0.628 -0.552
(− 3.10)** (18.46) (1.28) (8.44) (− 2.99)** (− 3.23)**

Domestic Individual 0.249 -0.689 -0.083 0.142 0.286

(3.10)** (− 9.65) (− 1.30) (2.06) (0.76)
Domestic Institutions 0.519 0.19 -0.10 0.49 0.49 0.489
Japan (1.40) (2.73) (-2.04) (10.17) (1.34) (1.38)
Foreign Inst. -0.483 0.409 -0.020 0.398 -0.498 -0.473
(− 1.94) (5.26) (− 0.33) (6.38) (− 1.75) (-1.83)

Domestic Individual 0.527 -0.586 -0.027 0.118 0.408

Russia (1.61) (− 3.89) (-0.27) (1.08) (2.14)*

NR 0.082 -0.149 -0.012 -0.053 -0.001

(0.46) (-3.10) (-1.42) (-2.63) (-0.08)
AM 0.081 0.613 0.08 -0.04 0.433 0.106
(0.00) (11.13) (0.92) (-0.91) (0.25) (0.13)
US - CFTC HF -0.23 -0.297 0.041 -0.158 -0.357
(-1.35) (− 5.60) (1.32) (-3.99) (-2.07)*
DL 0.505 -0.086 -0.167 0.055 0.384
(2.46)* (-1.42) (-2.63) (0.32) (1.73)

Robinhood client # 0.548 -0.016 -0.086 0.808 0.439

US (1.34) (− 0.31) (− 1.51) (16.05) (1.29)

The dependent variable is Fmt . All variables (except the dummy B) are standardized to have mean = 0 and st.dev. = 1, such that estimated coefficients
are comparable across investor types and countries. The first column on the left reports B’s coefficient in a univariate regression controlling for only
autocorrelation in F. The main middle block reports full estimation of Eq.(2). The most-right column reports estimation of B in an alternative version
of Eq.(2) which excludes the contemporaneous return R (only for those investor types whose net trading is significantly positively correlated with
∑
contemporaneous returns). To keep reporting concise, sum of lag coefficients are reported in single statistics: R-lag = ki=1θi is an indicator of
∑k
feedback trading and F-lag = i=1ρi is an indicator of flow persistence. Newey-West t-statistics are in parentheses. *(**) denotes statistical signifi­
cance of the B-coefficient at the 5%(1%) level.

Any coincidence of aggregate trading with the direction of the market return may potentially drive the above inference regarding
sophistication based on a single event. In such cases, additional granularity from the cross-section of net-trading and returns provides a
robustness check. Robinhood data enable a cross-sectional analysis of whether retail investors’ stock selection when buying during the
negative bubble had any predictive value for the cross-section of stock returns during the rebound. For this purpose, we define the
following variables for all S&P500 index constituent stocks:
F0i = %change in the number of Robinhood investors holding stock i during the negative − bubble period

R0i = return of stock i over the negative − bubble period.

R1i = return of stock i over the rebound period

17
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

where the negative-bubble period is defined as 11.March–17.April 2020 and the rebound period as 17.April–13.August 2020 (13.
August is the last day Robinhood data are available). To assess predictive ability, we run the following cross-sectional regression:
R1i = α + ψ F0i + φR0i + εi (3)

Results are reported in Table 5.

The univariate regression in the first row suggests that retail investors’ stock selection during the negative-bubble period has
statistically-significant predictive ability for the cross-section of stock returns during the rebound period. The second row, controlling
for returns during the negative-bubble period, suggests that part of this predictive ability might be owing to contrarian trading: bubble-
period returns negatively predict rebound-period returns and this control wipes out a large part of the statistical significance of retail
investors’ predictive ability. In the third row, we run a regression of F0 on R0 to characterize retail investors’ stock selection. The result
confirms significant contrarianism, where the negative cross-sectional relation with returns explains 19.5% of retail investors’ stock
selection. In sum, the regular contrarian behavior of retail investors helped them pick stocks with larger selling pressure and benefit
from their relatively stronger rebound.

4.2. Understanding individual investors’ success in the COVID episode

Individual investors’ aggregate trading during the COVID episode contrasts with existing views of their sophistication. We now
focus on identifying drivers of individual investors’ trading in this episode. The COVID period had two unique aspects which may have
potentially influenced individual investors’ aggregate trading: i) an unprecedented amount of monetary and fiscal policy support, ii)
lockdowns and a new work-from-home lifestyle that provided individuals excess free-time to monitor and trade financial markets. We
conjecture that the combination of these two effects created a unique positive shock to retail individual investor demand for equities.
An exogeneous shock to individual investor demand for equities can be captured by the COVID-period dummy Ct in a model that
controls for other known patterns in individual investor trading. This involves adding Ct into Eq.(2).
We also uncover an asymmetry in individual investors’ response to positive and negative returns. Investor response to positive and
negative returns may differ as they may have asymmetric effects on risk tolerance. From the econometric viewpoint, omission of a
potential nonlinearity may amount to misspecification and lead to noninformative coefficient estimates. Hence, we estimate a
nonlinear version of Eq.(2). For this purpose, we create a negative-return dummy variable, Nt, that takes the value of 1 when Rt < 0 and
0 otherwise, and estimate:
∑p ∑p ∑p
(4)
′ ′
Ft = ω + ω C t + ω ′ B t + ρ Ft− i + λRt + λ′ Rt Nt +
i=1 i i=1
θi Rt− i + θ′ Nt− i Rt− i + ε1,t
i=1 i

In Table 6, significantly positive λ’ coefficients indicate that individual investors are less contrarian to positive contemporaneous
returns than to negative ones. In other words, selling into price increases is more immediate compared with buying dips. With respect
to past weeks’ returns, however, significantly negative θ’ coefficients indicate that individual investors are more contrarian to negative
past returns than to positive ones. Hence, the immediate effect of negative returns on risk tolerance dissipates quickly, and motivation
to buy the dip starts to dominate. Thus, we document significant asymmetries in individual investors’ response to the sign of
contemporaneous and past returns.
Returning to the main question, in all countries the coefficient of the COVID-period dummy Ct is significantly positive. On the other
hand, the coefficient of the negative-bubble dummy Bt loses its significance in most cases. Thus, controlling for individual investors’
asymmetric response to returns, a positive shock to individual investor demand for equities in the COVID environment is much more
significant than a potential motive of exploiting a negative bubble in driving individual investor trading. These results are robust to
EGARCH estimation which yields qualitatively identical results with sharper significance levels in several cases (available upon
request). The conclusion is that a lucky coincidence of the shock to their demand for equities played a larger role than skill in ac­
counting for the above documented success of individual investors in this episode.
We check robustness of our interpretation of a unique shock to retail investor demand for equities as follows. First, a potential
driver of individual investor trading is sentiment. The historical relationship between AAII sentiment changes (d.S where S is defined as
% bullish minus % bearish) and change in the number of Robinhood clients holding S&P500 stocks (d.RH) is significantly positive (even

Table 5
The cross-section of stock returns and Robinhood investors’ positioning.
Regressors

Dependent Intercept F0 R0 R-squared

0.138 0.059
R1 0.024
(13.08)** (3.55)**
0.146 0.038 − 0.145
R1 0.036
(12.34)** (1.88) (− 2.22)*
0.438 − 1.402
F0 0.195
(24.71)** (− 7.85)**

Each regression in one row. The sample comprises S&P500 index constituent stocks. Heteroscedasticity-adjusted t-statistics are in parentheses. *(**)
denotes statistical significance at the 5%(1%) level.

18
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Table 6
Drivers of individual investors’ trading.
Korea Taiwan Thailand Turkiye Japan Russia CFTC - NR Robinhood

ω 0.000007 − 0.000019 0.000024 0.000017 − 0.000089 0.000160 − 0.001154 − 1171.3

(0.76) (− 1.85) (3.19)** (1.74) (− 2.05)* (2.31)* (− 1.10) (− 0.69)
C 0.000109 0.000061 0.000040 0.000104 0.0002081 0.000142 − 0.001332 20,783.0
(4.78)** (3.39)** (3.10)** (3.90)** (2.21)* (3.73)** (− 0.35) (2.56)*
B 0.000224 0.000026 0.000076 0.000081 0.0001854 0.000234 − 0.00286 31,651.6
(2.86)* (0.30) (0.93) (1.03) (0.89) (1.63) (− 0.38) (1.65)
ρ 0.114 0.157 0.101 0.063 0.1669 0.015 − 0.0552 0.191
(8.25)** (18.97)** (5.19)** (4.00)** (2.34)* (0.44) (− 1.81) (4.22)**
λ − 0.0197 − 0.0254 − 0.0171 − 0.0172 − 0.0299 − 0.0129 − 0.0258 − 87,364.8
(− 13.51)** (− 13.96)** (− 12.96)** (− 18.88)** (− 7.94)** (− 4.95)** (− 0.36) (− 0.27)
λ’ 0.0105 0.0037 0.0058 0.0075 0.0157 0.0117 − 0.1474 196,687.7
(5.13)** (1.74) (3.26)** (5.33)** (3.48)** (3.77)** (− 1.26) (0.34)
θ 0.0009 0.0012 − 0.0008 0.0003 0.0014 − 0.0007 0.0283 201,832.3
(1.66) (2.41)* (− 1.15) (0.75) (0.36) (− 0.57) − 0.53 (1.59)
θ‘ − 0.0019 − 0.0035 0.0000 − 0.0015 − 0.0074 0.0014 − 0.0485 − 451,388.1
(− 2.34)* (− 3.55)** (0.03) (− 3.04)** (− 1.54) (0.72) (− 0.55) (− 2.17)*

This table reports results from Eq.(4) estimated for individual investors’ trading Ft (for Robinhood data, d.RH: change in the number of Robinhood
∑
clients holding S&P500 stocks). To keep reporting concise, sum of lag coefficients are reported in single statistics where ρ = ki=1ρi is an indicator of
∑k
persistence (equivalent to ‘F-lag’ in Table 4), θ = i=1θi is an indicator of feedback trading with respect to positive past returns (R-lag in Table 4) and θ
∑
‘= ki=1θ′ i is an indicator of differential feedback trading with respect to negative past returns. B is an asymmetric negative bubble dummy and C is a
dummy that captures COVID period effects (lockdowns and work-from-home lifestyle). The data are not standardized; hence coefficients reflect
relative scales of raw variables. Newey-West t-statistics are in parentheses. *(**) denotes statistical significance at the 5%(1%) level.

controlling for returns)22: d.RHt = c + d.RHt− 1 + λ1Rt + λ2Rt− 1 + γd.St + γd.St− 1 + εt.
Estimated Newey-West t-statistics: (2.42) (21.58) (− 2.63) (− 1.87) (1.77) (2.66).
Yet, Fig. 4 makes clear that Robinhood investors stepped up their buying although AAII individual investor sentiment remained at
depressed levels. This makes Robinhood clients’ accelerated net-buying more remarkable as it cannot be explained by- and would
actually become more significant after controlling for sentiment.
Second, available statistics from around the world, presented in detail in Supplementary Material 1, clearly document an un­
precedented increase in new account openings. They further show that the influx of new entrants, mostly of younger and working age,
has altered the demographic composition of individual investor population.

4.3. What drove US asset managers’ and foreign investors’ exits?

Significant losses of US asset managers and foreign institutional investors due to selling during the COVID episode begs the question
of whether this was fund managers’ choice or dictated by client outflows. To investigate this, we look at flows toward global equity
funds, compiled and reported by three different sources: EPFR Global, Refinitiv-Lipper and ICI. These three sources slightly differ in their
coverage; we have cross-checked and confirmed the same pattern in each of them. Reported results are based on the weekly Refinitiv-
Lipper data (retrieved from Reuters Eikon).23
Earlier studies find that flows toward funds is a driver of funds’ trading and price impact (Edelen and Warner, 2001). Consistent
with this, US asset managers’ and foreign institutional investors’ net trading are highly positively correlated with flows toward
respective equity funds in our data (correlations on weekly data range between +0.49 and + 0.71). Furthermore, Fig. A.4 in Sup­
plementary Material 1 clearly indicates that cumulative flows toward global equity funds go hand-in-hand with cumulative net trading
by foreign institutions shown in Fig. 1.
To formally test, we run the following panel regression at the weekly frequency:
∑0
Fi,t = ci + β FFt− k + γi Xi,t + εi,t
k=− 1 k
(5)

where Fi,t is foreign institutional investors’ net trading in country i in week t, FFt is net client flows toward global equity funds,24 and Xi,t
is a vector of control variables deemed to drive foreign investors’ trading and flows toward global equity funds, specifically: lagged

22
Qualitatively identical results are obtained using Robinhood client counts in S&P500 ETFs SPY and VOO.
23
Capturing a significant trend-shift from conventional funds into ETFs that has been in place in recent years, we also decompose conventional
mutual funds versus ETFs in further analysis presented in Supplementary Material 2.
24
We confirmed the robustness of our results under several alternative definitions of flows toward equity funds. Reported results pertain to flows
global equity funds including ETFs domiciled in the US.

19
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Table 7
Drivers of foreign / institutional investors’ trading during the COVID episode.
Dependent c FF FF(− 1) R(− 1) Covid cases Covid deaths TEU n

0.016 0.538 0.203 − 0.070 0.017 − 0.096 − 0.228 184

Foreign investors (panel of host countries)
(0.25) (6.47)* (3.06)* (− 1.27) (− 0.07) (− 0.39) (− 3.45)*
0.019 0.594 0.027 − 0.255 0.035 0.274 − 0.308 47
Asset managers (US index futures)
(0.16) (3.72)* (0.25) (− 1.18) (0.29) (2.17) (− 3.08)*

This first line reports the results from estimating Eq.(5) on a panel of host markets. The second line replicates the estimation for asset managers in US
index futures. All variables are standardized to have zero mean and unit standard deviation at the country level, such that coefficients are comparable
across variables and regressions. Panel Newey-West t-statistics are in parentheses.

host market return, current week’s COVID cases and deaths in the host country and weekly average of daily Twitter-based economic
uncertainty (TEU) index (see www.policyuncertainty.com). Estimation results over the January–November 2020 period are in
Table 7.25
In the first line, FF is significant and positive both contemporaneously and with a lag, indicating that flows toward funds is a
significant driver of foreign institutional investors’ trading in host markets (alone, they can explain 41.7%; not in the table). In the
second line, the estimation is replicated for US asset managers in S&P500 index futures: contemporaneous flows toward equity funds is
a significant driver (can alone explain 39.5%) of US asset managers’ futures trading.26 Our result is consistent with Glossner et al.
(2022) finding using F-13 filings data.
Lagged returns, COVID cases and deaths are insignificant, suggesting that institutional investors’ trading cannot be explained by
their reaction to COVID statistics. Thus, US asset managers’ and foreign institutional investors’ role in fomenting the negative bubble
and not contributing to the rebound was largely driven by client flows toward funds at the global level.
TEU enters with significant negative signs, indicating that institutional investor net selling is associated with higher levels of the
uncertainty index. As an interesting note, when TEU is removed from the right-hand side of the regressions, intercept terms become
significantly negative (not in the table). This implies that part of the persistent selling by institutional investors can be related to the
above-average overall economic uncertainty associated with the COVID period.
A striking note is that fund clients exited equity funds while our previous results above indicated that retail investors trading on
their own ramped up their buying of equities. Thus, COVID-period evidence suggests that fund clients behave differently from in­
dividuals trading on their own. Furthermore, Fig. A.4 in Supplementary Material 1 also reveals that investors shifted from conventional
mutual funds toward ETFs during the COVID period (an acceleration of the existing trend). Taken together, we interpret this as a
COVID-induced trend toward self-servicing.
This shift toward self-servicing led to a temporary increase in home bias, as clients’ exiting international funds were replaced by
new entrants’ direct purchases of home-country stocks and ETFs, possibly as individuals with limited expertise to trade foreign equities
focused on home-country equities. Fig. OA4 in Supplementary Material 2 shows that this home bias was rectified later after October
2020.
An extended period of analysis in Fig. OA.4 reveals that fund clients returned back (mostly in the form of ETFs) between late
November 2020 and May 2021 well after the rebound was completed and US indices headed to new highs. Hence, equity fund clients’
aggregate trading during the COVID episode displayed particularly poor timing, consistent with Akbaş et al. (2015) ‘dumb money’
characterization.

5. Conclusion

During the COVID period, individual investors emerged with significant net buying in world stock markets, which can be explained
by a combination of their regular contrarian behavior and a unique positive shock to working-age retail investor demand for equities.
This shock was likely driven by excessive monetary and fiscal stimulus and free time to access stock markets thanks to lockdowns and
work-from-home practices. This lucky sequence made them winners of this episode, as a peculiar outcome. Foreign institutional in­
vestors and US asset managers sold into the negative bubble, at least partly driven by client outflows. Unlike the tech bubble, the
rebound was not associated with their pivoting. Dealers in the US equity futures and some domestic institutions in emerging markets
harvested significant profits from their sophisticated timing exploiting a negative bubble.
The COVID circumstances accelerated the trend from delegated investment management toward self-serviced investing. As a result,
the participant structure of world stock markets has changed. While a component of individual investor influx was due to ultra-
expansionary policies and transient, another component due to a new work-from-home lifestyle is likely enduring. This adds
increased relevance to individual investor behavior as a factor affecting the dynamics of stock markets, on which we document the
following fresh findings: i) characterization of individual investors as noise traders in the existing literature needs caution, ii) the

25
Results employing random effects estimator (accepted by Hausman test) with clustered robust standard errors and those using a two-stage GMM
method (to address potential heterogeneity between institutional investors’ trading and flows toward equity funds) are similar (available upon
request).
26
The correlation between weekly S&P500 returns and aggregate flows toward US equity funds is +0.53 when ETFs are included and + 0.50 ETFs
excluded.

20
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

timing of individual investors’ response to positive and negative returns is asymmetric, iii) margin traders and cash traders behave
differently, where the latter display more contrarian trading and signs of the disposition effect.
This paper’s primary contribution is to provide case-based evidence to answer the question ‘who are agents of negative bubbles?’.
Using Korean data, but based only on the sign of feedback trading with no attempt to identify bubbles, Choi et al. (2015) suggest that
institutional investors, rather than individuals, are agents of bubbles. Using worldwide data on a well-identified negative bubble, we
provide direct evidence that bubble-formations develop with institutional investors responding to significant new information (where
response may be reinforced due to client flows), whereas self-trading individuals tend to counteract thanks to their contrarian
behavioral traits.

CRediT authorship contribution statement

Numan Ülkü: Conceptualization, Data curation, Formal analysis, Methodology, Project administration, Validation, Writing –
original draft, Writing – review & editing. Fahad Ali: Data curation, Validation, Investigation, Resources, Investigation, Writing -
review & editing. Saidgozi Saydumarov: Data curation, Formal analysis, Resources, Investigation. Deniz İkizlerli: Data curation,
Resources, Investigation.

Acknowledgements

We gratefully acknowledge grants of non-public data from Borsa Istanbul, Pakistan Stock Exchange, and National Clearing
Company of Pakistan Limited. This work was supported by the research startup grant from Zhejiang University of Finance and Eco­
nomics, and the Cooperation Program at Charles University, research area Economics. We also thank an anonymous reviewer for
helpful suggestions.

Appendix A. Appendix

Table A.1
Data description and summary statistics.

Pre-COVID period: COVID period:

Country Nature of the data Variable Mean St.Dev. Skew. Kurt. Mean St.Dev. Skew. Kurt.

Daily 4 Jan 2010–21 Feb 2020 24 Feb - 20 Nov 2020

All stocks Indiv − 0.0000162 0.0002128 − 0.41 8.11 0.0001417 0.0003814 0.05 3.64
Precise, compiled Inst 0.0000004 0.0001863 0.45 7.10 − 0.0000491 0.0002607 0.23 3.79
Korea
from electronic For 0.0000163 0.0002179 0.07 8.55 − 0.0001012 0.0002804 − 0.60 4.81
records by the
R 0.000111 0.0093 − 0.46 7.03 0.000808 0.0195 − 0.24 7.46
Exchange
Daily 4 Jan 2010–21 Feb 2020 24 Feb - 20 Nov 2020
All stocks Indiv − 0.0000070 0.0003184 0.36 5.61 0.0001001 0.0004479 0.95 4.56
Precise, compiled Inst − 0.0000046 0.0000251 − 0.05 5.54 0.0000087 0.0000266 0.38 3.44
Taiwan
from electronic For 0.0000154 0.0002876 − 0.41 6.14 − 0.0000867 0.0003927 − 1.07 4.82
records by the
R 0.000293 0.0092 − 0.71 7.51 0.001039 0.0141 − 0.40 6.88
Exchange
Daily 4 Jan 2010–21 Feb 2020 24 Feb - 20 Nov 2020
All stocks Indiv − 0.0000008 0.0002413 − 0.96 21.40 0.0000670 0.0002626 − 0.71 10.38
Precise, compiled Inst 0.0000123 0.0001429 0.19 8.46 0.0000239 0.0002046 0.14 6.80
Thailand
from electronic For − 0.0000127 0.0001998 1.20 28.73 − 0.0000985 0.0001901 1.74 18.40
records by the
R 0.000286 0.0095 − 0.29 7.58 − 0.000387 0.0207 − 1.58 12.39
Exchange
Daily 4 Jan 2010–21 Feb 2020 24 Feb - 30 Oct 2020
All stocks Indiv − 0.0000034 0.0002673 − 0.14 4.40 0.0001358 0.0002478 − 0.23 5.47
Precise, compiled Inst 0.0000039 0.0000931 − 0.10 6.26 0.0000239 0.0001058 0.31 4.84
Turkiye
from daily For 0.0000006 0.0002879 0.08 4.53 − 0.0001663 0.0002335 − 0.37 3.77
settlements by the
R 0.000352 0.0136 − 0.58 7.10 0.000228 0.0184 − 1.42 8.19
Central Registry
Daily 24 Jan - 21 Feb 2020 24 Feb - 31 Dec 2020
All stocks Indiv 0.0000164 0.0001073 − 0.61 3.45 0.0000304 0.0002887 − 0.57 4.95
Precise, compiled Inst 0.0001410 0.0001620 0.38 2.91 0.0001844 0.0003250 0.25 3.95
Pakistan
from daily For − 0.0001530 0.0001610 − 1.20 4.07 − 0.0002188 0.0002345 − 0.96 6.53
settlements by the
R − 0.002729 0.0113 − 0.31 3.59 0.000390 0.0159 − 1.25 8.13
Clearing House
Weekly 3 Jan 2011–21 Feb 2020 24 Feb - 28 Nov 2020
All stocks Indiv − 0.0002381 0.0008587 − 0.58 5.39 − 0.0000663 0.0006867 − 0.33 3.50
Japan
Compiled from Inst 0.0001165 0.0007782 − 0.31 7.76 0.0004258 0.0007938 1.80 8.74
electronic surveys For 0.0001334 0.0010973 0.83 8.26 − 0.0003042 0.0009172 − 0.64 3.45
(continued on next page)

21
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Table A.1 (continued )

Pre-COVID period: COVID period:

Country Nature of the data Variable Mean St.Dev. Skew. Kurt. Mean St.Dev. Skew. Kurt.

R 0.001328 0.0257 − 0.59 5.16 0.001442 0.0465 − 0.77 5.99

Weekly, only Indiv 2 Jan 2018–21 Feb 2020 24 Feb - 28 Nov 2020
Russia All stocks, Precise, Indiv 0.0000230 0.0001361 0.06 2.87 0.0001478 0.0004486 − 2.13 9.85
Compiled by MOEX R 0.003453 0.0179 − 0.28 3.00 0.001273 0.0410 − 1.74 8.04
Daily 3 May 2018–21 Feb 2020 24 Feb - 13 Aug 2020
number of retail dRH 5361.4 10,024 7.67 94.59 84,920.6 69,066 0.83 3.43
US Robinhood
clients holding
R 0.000538 0.0089 − 0.38 6.92 0.000088 0.0298 − 0.64 6.92
S&P500 stocks
Weekly 4 Jan 2010–18 Feb 2020 19 Feb - 20 Nov 2020
Index futures NR 0.0000 0.0151 − 1.26 14.43 − 0.0002 0.0202 − 3.43 17.87
Compiled from AM − 0.0000 0.0176 − 0.40 4.59 − 0.0027 0.0206 − 0.89 3.84
US CFTC
surveys by CFTC DL − 0.0000 0.0158 0.85 6.99 0.0023 0.0232 1.25 7.61
Proportion of open HF 0.0001 0.0178 − 0.77 8.09 0.0009 0.0141 0.16 2.09
interest R 0.002072 0.0191 − 0.88 7.01 0.001334 0.0388 − 1.11 4.99
Indiv, Inst and For stand for net trading flows as a proportion of free-float market capitalization (i.e., Fm
t ) of individual, domestic institutional and
foreign investors, respectively. For the US Robinhood data, the variable is the daily change in the number of investors holding S&P500 stocks (dRH).
For the US CFTC data, the variable is the net change in futures position as a proportion of open interest (NR = Nonreportables, comprising mainly
individual investors, AM = Asset Managers, DL = Dealers, HF=Hedge Funds. CFTC weeks start on Wednesday and end on Tuesday). R is the return
(first difference of the log) of the respective market index.

Appendix B. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pacfin.2023.102044.

References

Abreu, D., Brunnermeier, M., 2002. Synchronization risk and delayed arbitrage. J. Financ. Econ. 66, 341–360.
Abreu, D., Brunnermeier, M., 2003. Bubbles and crashes. Econometrica 71, 173–204.
Akbaş, F., Armstrong, W., Sorescu, S., Subrahmanyam, A., 2015. Smart money, dumb money, and capital market anomalies. J. Financ. Econ. 118, 355–382.
Arkes, H., Hirshleifer, D., Jiang, D., Lim, S., 2010. A cross-cultural study of reference point adaptation: evidence from China, Korea, and the US. Organ. Behav. Hum.
Decis. Process. 112, 99–111.
Barber, B., Odean, T., 2008. All that glitters: the effect of attention and news on the buying behavior of individual and institutional investors. Rev. Financ. Stud. 21,
785–818.
Barber, B., Lee, Y.T., Liu, Y.J., Odean, T., 2009. Just how much do individual investors lose by trading. Rev. Financ. Stud. 22, 609–632.
Barrot, J., Kaniel, R., Sraer, D., 2016. Are retail traders compensated for providing liquidity? J. Financ. Econ. 120, 146–168.
Berkman, H., Malloch, H., 2023. Stock valuation during the COVID-19 pandemic: an explanation using option-based discount rates. J. Bank. Financ. 147, 106386.
Bing, T., Ma, H., 2021. COVID-19 pandemic effect on trading and returns: evidence from the Chinese stock market. Econom. Analy. Policy 71, 384–396.
Bourdeau-Brien, M., Kryzanowski, L., 2020. Natural disasters and risk aversion. J. Econ. Behav. Organ. 177, 818–835.
Broner, F., Gelos, G., Reinhart, C., 2006. When in peril, retrench: testing the portfolio channel of contagion. J. Int. Econ. 69, 203–230.
Carpentier, C., Suret, J., 2021. On the rationality of institutional investors: the case of major industrial accidents. J. Behav. Financ. 22, 289–305.
Chang, C., Hsieh, P., Wang, Y., 2015. Sophistication, sentiment, and misreaction. J. Financ. Quant. Anal. 50 (4), 903–928.
Choi, J.J., Kedar-Levy, H., Yoo, S.S., 2015. Are individual or institutional investors the agents of bubbles? J. Int. Money Financ. 59, 1–22.
Choi, P.B.S., Choi, J.H., Chuing, C.Y., 2020. Do individual traders undermine firm valuation? Financ. Res. Lett. 36, 101567.
Djalilov, A., Ülkü, N., 2021. Individual investors’ trading behavior in Moscow exchange and the COVID crisis. J. Behav. Exp. Financ. 31, 100549.
Edelen, R., Warner, J., 2001. Aggregate price effects of institutional trading: a study of mutual fund flow and market returns. J. Financ. Econ. 59, 195–220.
Foley, S., Kwan, A., Philip, R., Ødegaard, B.A., 2021. Contagious margin calls: how COVID-19 threatened global stock market volatility. J. Financ. Mark. 59, 100689.
Foley, S., Frijins, B., Garel, A., Roh, T.Y., 2022. Who buys bitcoin? The cultural determinants of bitcoin activity. Int. Rev. Financ. Anal. 84, 102385.
Glossner, S., Matos, P., Ramelli, S., Wagner, A., 2022. Do Institutional Investors Stabilize Equity Markets in Crisis Periods? Evidence from COVID-19. ECGI Working
Paper.
Goetzman, W., Kim, D., 2018. Negative bubbles: what happens after a crash. Eur. Financ. Manag. 24, 171–191.
Gong, B., Pan, D., Shi, D., 2017. New investors and bubbles: an analysis of the Baosteel call warrant bubble. Manag. Sci. 63, 2397.
Greenwood, R., Shleifer, A., You, Y., 2019. Bubbles for Fama. J. Financ. Econ. 131, 20–43.
Griffin, J., Harris, J., Shu, T., Topaloğlu, S., 2011. Who drove and burst the tech bubble? J. Financ. 66, 1251–1290.
Griffin, J., Harris, J., Topaloğlu, S., 2003. The dynamics of institutional and individual trading. J. Financ. 48, 2285–2320.
Gürkaynak, R., 2008. Econometric tests of asset price bubbles: taking stock. J. Econ. Surv. 22, 166–186.
Hasso, T., Pelster, M., Breitmayer, B., 2020. Terror attacks and individual investor behavior: evidence from the 2015–2017 European terror attacks. J. Behav. Exp.
Financ. 28, 100397.
Hoopes, J., Langetieg, P., Nagel, S., Reck, D., Slemrod, J., Stuart, B., 2016. Who Sold During the Crash of 2008-9? Evidence from Tax-Return Data on Daily Sales of
Stock. NBER Working Paper 22209.
Kaniel, R., Saar, G., Titman, S., 2008. Individual investor trading and stock returns. J. Financ. 63, 273–310.
Li, X., Subrahmanyam, A., Yang, X., 2021. Winners, losers, and regulators in a derivatives market bubble. Rev. Financ. Stud. 34, 313–350.
Linnainmaa, J., 2010. Do limit orders alter inferences about investor performance and behavior? J. Financ. 65, 1473–1506.
Naik, P.K., Shaikh, I., Huynh, T.L.D., 2022. Institutional investment activities and stock market volatility amid COVID-19 in India. Econ. Res. 35, 1542–1560.
Onishchenko, O., Ülkü, N., 2022. Investor types’ trading around the short-term reversal pattern. Int. J. Financ. Econ. 27, 2627–2647.
Pagano, M., Sedunov, J., Velthuis, R., 2021. How did retail investors respond to the Covid-19 pandemic? The effect of Robinhood brokerage customers on Market
Quality. Financ. Res. Lett. 43, 101946.
Schmeling, M., 2007. Institutional and individual sentiment: smart money and noise trader risk. Int. J. Forecast. 23, 127–145.

22
N. Ülkü et al. Pacific-Basin Finance Journal 79 (2023) 102044

Shiller, R.J., 2000. Irrational Exuberance. Princeton University Press, Princeton, NJ.
Shu, P., Yeh, Y., Chiu, S., Chen, H., 2005. Are Taiwanese individual investors reluctant to realize their losses? Pac. Basin Financ. J. 13, 201–223.
Wang, A., Young, M., 2020. Terrorist attacks and investor risk preference: evidence from mutual fund flows. J. Financ. Econ. 137, 491–514.
Weitzel, U., Huber, C., Huber, J., Kirchler, M., Lindner, F., Rose, J., 2020. Bubbles and financial professionals. Rev. Financ. Stud. 33, 2659–2696.
Welch, I., 2020. Retail Raw: Wisdom of the Robinhood Crowd and the COVID Crisis. NBER Working. Paper 27866.
Zhang, J., Mao, R., Wang, J., Xing, M., 2021. The way back home: trading behaviours of foreign institutional investors in China amid the COVID-19 pandemic. Pac.
Basin Financ. J. 68, 101618.

23