Are retail traders compensated for providing liquidity?

Jean-Noel Barrot† Ron Kaniel‡ David Sraer§

Abstract

This paper examines the extent to which individual investors provide liquidity to the

stock market, and whether they are compensated for doing so. We show that the ability

of aggregate retail order imbalances, contrarian in nature, to predict short-term future

returns is significantly enhanced during times of market stress, when market liquidity

provisions decline. While a weekly rebalanced portfolio long in stocks purchased and

short in stocks sold by retail investors delivers 19% annualized excess returns over a

four factor model from 2002 to 2010, in periods of high uncertainty it delivers up to

40% annualized returns. Despite this high aggregate performance, individual investors

do not reap the rewards from liquidity provision because (i) they experience a negative

return on the day of their trade, and (ii) they reverse their trades long after the excess

returns from liquidity provision are dissipated. Finally, we show that experienced

traders tend to do better on both dimensions.

∗
For helpful comments and suggestions, we thank Remy Chicheportiche, Christian Gollier, Augustin
Landier, Tarun Ramadorai, Elias Rantapuska (our HFS discussant), Patrick Roger, David Thesmar, and par-
ticipants of the Helsinki Finance Summit. We acknowledge the support from the Observatoire de l’Epargne
Europeenne.
†
MIT Sloan (email: jnbarrot@mit.edu)
‡
Simon School of Business at University of Rochester and CEPR (email: ron.kaniel@simon.rochester.edu)
§
Princeton University and NBER (e-mail: dsraer@princeton.edu)

1
1. Introduction

What is the contribution of individual investors to the formation of prices and liquidity in
financial markets? A longstanding literature has considered them as “noise” traders, in the
sense of Black (1986), and Shleifer and Summers (1990), who push prices away from fun-
damentals and destabilize markets. In contrast to this literature, recent empirical evidence
suggests that individual investors’ trades provide liquidity to meet the demand for immedi-
acy of other market participants (Kaniel et al., 2008, 2012; Kelley and Tetlock, 2013). While
retail investors may be less sophisticated than their institutional counterparts, they also face
lower agency costs and liquidity constraints relative to institutional investors such as mutual
funds (Chevalier and Ellison, 1999; Coval and Stafford, 2007). Retail traders could thus have
some ability to act as market makers, especially when institutional liquidity dries up, as was
the case during the recent financial crisis.
This paper examines the extent to which individual investors provide liquidity to the stock
market, and whether or not they are compensated for doing so. We use a unique dataset
obtained from a leading European online broker in personal investing and online trading.
This dataset allows us to track the orders of a large sample of individuals from January
2002 to December 2010. In particular, the data covers the 2008-2009 financial crisis, when
the liquidity-provision capacity of traditional market makers was plausibly reduced (Nagel,
2012). We uncover three main findings.
First, individuals provide liquidity especially at times when conventional liquidity providers
are constrained. We begin by showing that in our sample, consistent with recent literature,
aggregate retail buy-sell imbalances are contrarian and positively predict the cross-section of
stock returns at a horizon of a couple of weeks. A one standard deviation increase in daily
order imbalances is associated with an increase in return over two trading weeks of about 15
additional basis points (4% increased annualized return). We then test whether this increase
in returns earned by retail investors corresponds to compensation for liquidity provision. To
do so, we first construct a weekly rebalanced portfolio that goes long in stocks purchased and

2
short in stocks sold by retail investors (the “retail” portfolio). We then compare the returns
on this portfolio with time-series variation in the supply of liquidity provided by institutional
investors. Recent work, including Adrian et al. (2012), Ang et al. (2011), and Ben-David et
al. (2012) suggests that intermediaries are especially constrained in their ability to provide
liquidity for high values of the VIX index.1 We thus simply split our sample into periods of
high and low VIX, when the VIX is higher or lower than 20, its 2002-10 median, and contrast
the returns on the “retail” portfolio in these two subsamples. We find robust evidence that
the rewards to liquidity provision increase sharply in times of high uncertainty. While the
“retail” portfolio earns 19% annualized excess returns over a four factor model from 2002
to 2010, it earns up to 40% annualized returns when traded over the weeks where the VIX
is above its sample median. These results indicate that rather than merely adding noise to
the market, retail traders do indeed provide liquidity to the stock market, especially when
institutional liquidity dries up.
Second, we show that retail investors fail to reap the actual returns from liquidity provi-
sion, and provide two explanations for this result. The first has to do with the price at which
retail orders are executed on the day of trading. To benefit from the predictable short-term
returns that follow a day of intense imbalances, individual investors need to avoid being
picked-off on day 0. To understand why, suppose that institutions holding stock S are hit
with liquidity shocks and need to fire sell their shares of S. The price of S will plummet on
day 0 and recover in the short-term thereafter. Individuals buying stock S at its lowest on
day 0 will fully benefit from the price reversal in the subsequent days. However, those who
purchased S before it reaches its lowest price experience a negative intra-day return on day
0, which may more than offset the gain from price reversal. Our analysis of order-level data
indicates that in our sample, retail investors do indeed get picked-off on day 0. The average
retail order experiences large and negative returns on this day, so much so that returns on
1
In support of this hypothesis, Nagel (2012) uses the returns to short-term reversal strategies as a proxy
for the returns to liquidity provision and finds that they are almost perfectly correlated with the level of the
VIX.

3
day 0 more than offset the rewards from liquidity provision that could arise subsequently.
The second reason for the low performance of individual investors in our sample has to do
with the speed at which they reverse their trades. Individuals cannot benefit from liquidity
provision unless they reverse their trades quickly enough thereafter, before the benefits are
dissipated. This is exactly what retail investors in our sample fail to do. The average holding
period among retail investors in our sample is above 300 days, while most of the returns from
liquidity provision are dissipated on average after 20 days. Thus, surprisingly, low trading
frequency and specifically slow reversal of trades is one of the reasons why individual in-
vestors in our sample underperform. This result contrasts with Odean (1998), or Barber and
Odean (2000), who argue that over-trading is responsible for the low performance of retail
traders.
Finally, we take advantage of the richness of our data to document cross-sectional het-
erogeneity in the returns to liquidity provision. We first sort orders based on the experience
of the individuals placing them. We find that highly experienced individuals are much less
prone to the picking-off effect. In addition, they also flip their orders much more quickly.
These two components explain a significant share of their outperformance relative to less ex-
perienced traders. We also sort orders based on the average speed at which the individuals
who place them usually reverse their trades. We find that “fast” traders are less prone to
the picking-off effect, and thus experience higher returns relative to slower traders.
This paper adds to the ongoing debate on the contribution of retail trades to stock market
efficiency. A number of papers have found that individual trades positively predict short-
term returns. A first body of work has interpreted this as evidence of noise trading pushing
prices away from fundamentals. Barber et al. (2009) find that stocks that individual investors
are buying (selling) during one week have positive (negative) abnormal returns that week
and in the subsequent two weeks. These returns then reverse over the next several months.
Although Barber et al. (2009) interpret their results as evidence of noise trading, they are also
consistent with individual investors providing liquidity to institutional investors. Hvidkjaer

4
(2008) finds that stocks with a high level of sell-initiated small-trade volume, measured
over the prior several months, outperform stocks with a high level of buy-initiated small-
trade volume at horizons of two years.2 Another body of work has associated the short-term
predictability of retail trades with liquidity provision. Kaniel et al. (2008) identify individual
investor trades using the NYSEs Consolidated Audit Trail Data files, which contain detailed
information on all orders executed on the exchange, including a field that identifies whether
the order comes from an individual investor. They show that the top decile of stocks heavily
bought by individuals outperform those heavily sold by individuals, a result again consistent
with retail traders providing liquidity to institutions that require immediacy. Dorn et al.
(2008) show that correlated limit orders predict subsequent returns in a manner consistent
with executed limit orders receiving compensation for accommodating liquidity demands.
Kaniel et al. (2012) also find evidence that stocks purchased by individual investors prior to
earnings announcement outperform those that they sell and that compensation for risk-averse
liquidity provision accounts for approximately half of this over-performance. Finally, Kelley
and Tetlock (2013) argue that retail traders provide liquidity to the market and benefit from
the reversal of transitory price movements. Our contribution to this body of work is two
fold. First, we show that the predictability of individual trades increases when the rewards to
liquidity provision are high; consistent with them providing liquidity. Second, utilizing our
order level data we demonstrate that individuals fail to benefit from their liquidity provision
role.
Because we focus on returns at the order level, our results also relate to the literature on
individual investors’ performance.3 The average household trades in excess of what liquidity
and hedging motives would command and loses money in the process (Odean, 1998; Barber
and Odean, 2000; Barber et al., 2006; Grinblatt and Keloharju, 2000) especially when going
2
A significant part of small trades are likely due to institutions splitting orders, especially in the later
part of the Hvidkjaer (2008) sample. Campbell et al. (2009) find that trades below $2,000 are more likely
to come from institutions than from individuals.
3
For an extensive review of the performance of individual investors’ behavior and performance, see Barber
and Odean (2011).

5
online (Barber and Odean, 2002). This is generally attributed to behavioral biases such
as overconfidence or gambling (Statman et al., 2006; Glaser and Weber, 2007; Grinblatt
and Keloharju, 2009; French, 2008). A small select group of retail traders however manage
to generate absolute performance (Barber et al., 2014), with some persistence (Coval et al.,
2005). Linnainmaa (2010) finds losses on limit orders and gains on market orders in Finland,
for portfolios long (short) in stocks that individuals on aggregate net bought (sold). We add
to this body of work by showing that individual investors’ returns are low because (i) they
get picked-off and (ii) they fail to reverse their trades soon enough. Retail investors do not
trade fast enough to collect the benefits from their liquidity provision.
Our results, which indicate that experienced investors trade at a better price on a given
day and reverse their positions more quickly, contributes to the recent and growing literature
on learning dynamics in finance. The fact that investors’ own experience shapes their future
decisions has been shown in the context of IPOs (Kaustia et al., 2008; Chiang et al., 2011),
retirement savings decisions (Choi et al., 2009), and mutual funds management (Greenwood
and Nagel, 2009). Learning may occur in a variety of ways. Investors may gradually dis-
cover their true type by rationally updating their priors in a Bayesian way after each action
(Mahani and Bernhardt, 2007; Linnainmaa, 2011). Investors may otherwise update their
beliefs in a non-Bayesian way as in Gervais and Odean (2001). The performance of retail
investors might increase through time due to learning-by-doing (Nicolosi et al., 2009; Seru
et al., 2009). Seru et al. (2009) show that individual investors’ behavior is consistent with
both learning about one’s type and learning by doing but that the former is quantitatively
more significant than the latter. Finally, List (2003), Agarwal et al. (2008), and Kaustia et
al. (2008) show in various frameworks that with experience, investment behaviors tend to
get closer to what full rationality would command.
The rest of the paper is organized as follows. We present the data in Section 2. Section 3
presents the results, and Section 4 concludes.

6
2. Data

We consider a large sample of French retail investors trading between January 2002 and
December 2010, provided by a leading European broker in personal investing and online
trading. In the past twelve years, this broker accounted for an average 15 percent of online
brokers stock trades on Euronext Paris, which collectively represented 14 percent of all
trades in the market.4 This sample is thus fairly representative of the behavior of individual
investors directly investing in the French stock market. This data was also used in Foucault
et al. (2011), who study the effect of retail investors on the volatility of stock returns.
There are 91,647 investors placing approximately 4.6 million orders in 730 stocks in our
sample. For each order, we track the trading exchange identifier (ISIN), the trading date,
the quantity, and amount traded in euros. Given that we do not have the exact timing of
the order within the day, we aggregate trades by individual × stock × day. The average
trade size in our sample is 7,741 euros. We obtain daily stock returns from EUROFIDAI.5
Our sample stands out in a number of ways. First, it includes information at the order
level, which allows us to perform detailed analyses of retail trading. Second, it spans a long
time period which includes episodes of market stress, such as the recent financial crisis. This
makes it possible to contrast the behavior of a large number of retail investors at different
points in time, when the returns to liquidity provision vary. Recent work, including Adrian
et al. (2012), Ang et al. (2011), and Ben-David et al. (2012) suggest that intermediaries are
especially constrained for high levels of the VIX index of implied volatilities of S&P index
options. Nagel (2012) uses the returns to short-term reversal strategies as proxies for the
returns to liquidity provision. He finds that they are almost perfectly correlated with the
level of the VIX. We thus simply construct a dummy High VIX equal to one if the level of
the VIX is above its 2002-10 median.6 In addition, we also define a Crisis dummy equal to
4
According to “Acsel”, the association of French online brokers (see
http://www.associationeconomienumerique.fr/) which collects monthly data on online trading.
5
EUROFIDAI is a research institute funded by the CNRS (French National Center for Scientific Research)
whose mission is to develop European stock exchange databases for academic research.
6
We check and find that all results are unchanged when we use the VSTOXX, the European volatility

7
one in the seven months from September 2008 to April 2009.
In some of the analyses, we adjust returns and cumulative returns for systematic risk. To
do so, we estimate the exposure of each of the 730 stocks in the sample to systematic risk
factors (Market, Small-minus-big, High-minus-low, and Momentum) over the sample period
(2002-2010) at the weekly level. All risk factors are obtained from EUROFIDAI.7 We run
the following OLS model for each stock in the sample:

Ret[t] − rf = a + b.M ktt + c.SM Bt + d.HM Lt + e.M OMt + it , (1)

where Ret[t] is a given stock’s return in week t, and M ktt , SM Bt , HM Lt and M OMt
are respectively the returns of the Market factor, Small-minus-big, High-minus-low, and
ˆ and ê are then used to define the risk adjusted
Momentum. The estimated coefficients b̂, ĉ, d,
return on any given stock i in any given period t, AdjRet[t], as the difference between the
realized return Ret[t] and its predicted value:

AdjRet[t]i = Ret[t]i − (b̂i .M ktt + ĉi .SM Bt + dˆi .HM Lt + êi .M OMt + rf ) (2)

3. Results

3.1. Evidence on liquidity provision

This paper studies the relationship between retail orders and future short-term returns both
at the stock-day level and the order level. We start by aggregating individual orders at the
stock-day level. Our main measure of imbalances, Imb[0], is similar to the one used in Kelley
and Tetlock (2013). It is computed daily as the number of shares bought by retail investors
minus the number of shares sold by retail investors divided by shares bought plus shares
index, instead of the VIX.
7
The procedure used by EUROFIDAI to compute factor returns for the French market is similar to the
one used by Kenneth French for the US market.

8
sold.8 As in Kelley and Tetlock (2013), we exclude from the sample stock × days with less
than five orders. We are left with 91,647 individuals trading 730 stocks from 2002 to 2010,
leaving us with 217,511 stock-days. We control for the size of firms with the log of their
market capitalization (Size). We denote as Ret[x, y] the holding period return between day
x and day y.
Panel A of Table 1 provides summary statistics. The average and median of Imb[0]
are very close to zero. The average aggregate volume traded in a stock on a given day is
just over 165,000 euros, which represents an average of 2.4% of the total daily volume for
these stocks.9 We first estimate whether retail order imbalance dynamics are consistent with
liquidity provision, i.e., if they seem to respond to past liquidity shocks. More specifically, we
want to measure the sensitivity of retail imbalances to past returns, controlling for market
conditions and stock invariant characteristics. We do so by running the following linear
regression:

Imb[0]it = α0 + α1 .Ret[−5, −1]it + α2 .Ret[−26, −6]it + α3 .Sizeit + πt + ηi + it , (3)

where Imb[0]it is the imbalance of stock i on day t, and Ret[−5, −1]it and Ret[−26, −6]it
are the cumulative returns over the past week and the prior month on stock i. We control
for the size of firms with the log of their market capitalization (Sizeit ), which is a known
predictor of returns. πt and ηi are respectively day and stock fixed effects. Standard errors
are clustered at the stock level. Results are presented in Table 2. The first specification
includes only day fixed effects, while the second adds stock fixed effects. Consistent with
evidence in Kelley and Tetlock (2013), and Kaniel et al. (2008), we find that retail imbalances
react strongly to past returns. The estimates are highly statistically significant, and the
coefficients are economically large. A one standard deviation decrease in the past week’s
returns, Ret[−5, −1]it , leads to an increase of about 7 percentage points in Imb[0], which
8
We check below that our results are robust when we use alternative measures.
9
Our coverage is comparable to Kelley and Tetlock (2013) where retail trades account for 2.3% of total
listed (NYSE/Amex/NASDAQ) volume, over a period of 5 years.

9
represents 12% of the sample standard deviation of Imb[0]. The estimate is unaffected by
the inclusion of stock fixed effects, suggesting that time invariant stock-level characteristics
are not responsible for the cross-sectional correlation between retail imbalances and past
returns. This pattern of buying and selling in reversal strategies resembles the trading of a
market maker who takes opposite positions to the rest of the market, and is overall consistent
with the idea that retail trades provide liquidity.
We then turn to the analysis of the returns to liquidity provision. Our regression model
for predicting cumulative holding period returns from day x to y is:

Ret[x, y]it = β0 +β1 .Imb[0]+β2 .Ret[−5, −1]it +β3 .Ret[−26, −6]it +β4 .Sizeit +πt +ηi +it , (4)

where Imb[0]it is the retail imbalance in stock i in day t, and Ret[−5, −1]it and Ret[−26, −6]it
are the cumulative returns over the past week and the past month on stock i. The coefficient
of interest is β1 , which measures the sensitivity of future returns to current imbalances from
retail orders.
We first run separate regressions for the cumulative returns from day x = 1 to day y
where y takes values from 1 to 100. We plot the coefficient along with 95% confidence
intervals in Figure 1. The graph shows that stocks heavily purchased by retail investors
outperform those that are heavily sold by a significant 25 basis points over the first couple
of weeks. This outperformance then gradually dissipates over the subsequent 85 days. We
obtain identical results when we perform the same analysis with risk adjusted cumulative
returns. We formalize this result in a regression setting by estimating equation 4. The results
are presented in Table 3. Columns 1 and 2 present the estimates of specifications including
day fixed effects, while Columns 3 and 4 present the results of the same model augmented
with stock fixed effects. Columns 1 and 3 use x=1 and y=16 days, i.e. these columns look at
returns over the next couple weeks following the initial imbalance. Columns 2 and 4 use x=17
and y=100. The main finding is that retail imbalances positively predict cumulative returns

10
from day 1 to day 16 (Columns 1 and 3). A one standard deviation increase in Imb[0] leads
to a 15 basis points increase in cumulative returns over the following couple of weeks. These
estimates are comparable in magnitudes to those obtained by Kelley and Tetlock (2013).
Columns 2 and 4 show that the effect is short-lived, since it is nearly fully reversed after 100
days. As we already noted in Table 2, the estimates are virtually unaffected by the inclusion
of stock fixed effects. Taken together, Table 2 and Table 3 suggest that individual investors
provide liquidity by placing contrarian trades and receive a significant compensation in the
form of high returns over the couple of weeks following their trades. We perform a number
of robustness tests to ensure that these results are robust to the proxy we use for imbalances
created by retail investors’ trades. In particular, we show in Table 4 that the results hold
(i) when we standardize Imb[0] (by subtracting its within-stock mean and scaling it by its
within-stock standard deviation), (ii) when we use terciles of Imb[0], or (iii) when we define
imbalances on stock i as the ratio of buy minus sell orders by retail investors on this stock
normalized by the market-wide volume on the stock.

3.2. Retail Investors’ Performance and the Demand for Liquidity

As we already emphasized, a natural interpretation of our results in Section 3.1 is that

retail investors provide liquidity to the market and receive in exchange higher returns on
their trades. An alternative interpretation of the results in Section 3.1 is that the predictive
power of retail imbalances results from noisy correlated trading pressure from individuals.
To disentangle these two alternative hypotheses, we simply look at how the sensitivity of
retail imbalances to past returns, as well as the predictive power of these imbalances on
future returns, vary with general liquidity conditions in the market. If the predictability of
individual imbalances increases when liquidity dries up elsewhere in financial markets, then
it is likely that these trades are indeed providing liquidity, rather than pushing prices away
from fundamentals.
Recent work, including Adrian et al. (2012), Ang et al. (2011), and Ben-David et al.

11
(2012) suggest that intermediaries are especially constrained when the VIX index of implied
volatilities of S&P index options reaches high levels. Nagel (2012) uses the returns to short-
term reversal strategies as proxies for the returns to liquidity provision. He finds that they
are almost perfectly correlated with the level of the VIX. We thus use the level of the VIX
as a proxy for the returns to liquidity provision and split our sample into weeks of high and
low VIX, where high VIX is defined as a VIX level higher than 20, its 2002-10 median. We
simply re-run the analysis of Section 3.1 for each of these two subsamples.
We first estimate whether the sensitivity of retail imbalances to past returns is signifi-
cantly different in periods of high and low uncertainty, controlling for market conditions and
stock invariant characteristics. To do so, we start with model 3 and interact all terms with
a dummy equal to 1 on days when the VIX is higher than 20, its 2002-10 median, and zero
otherwise. The results are presented in Table 5, Columns 1 and 2. The interaction term is
positive, small, and insignificant. This suggests that the response of individual investors to
past returns does not vary with the level of VIX. While this is not proof that retail investors
are the residual investors in times of high uncertainty, this suggests that they are probably
less constrained in their liquidity provisions than other market participants identified in the
literature, since their liquidity provision does not seem to decrease on days where the VIX
is high.
As a robustness check, we also use the crisis period as another period in our sample where
the returns to liquidity provision are likely to be very high. We thus estimate an augmented
version of equation (3), where all the variables are interacted with a Crisis dummy equal to 1
in the seven month period from September 2008 to April 2009, i.e. the peak of the financial
crisis. The results are presented in Table 5, Columns 3 and 4. The interaction of the Crisis
dummy and past returns is now positive and significant. This indicates that the sensitivity
of retail imbalance to past returns did in fact decrease during the crisis relative to the rest of
the sample period, suggesting some decline in the liquidity provision capacity of individual
investors: while a 1% return in the prior 5 days leads to a 1.1 percentage point decrease in

12
imbalances at date 0 outside the crisis, it leads only to a decrease of .66 percentage point
during the crisis. Note, however, that the sensitivity of retail imbalance to past returns
remains negative and statistically significant during the crisis period, albeit smaller. When
we throw both interactions (Crisis and VIX) in the regression (Columns 5 and 6), the results
remain very similar.
We then investigate how the returns following large retail imbalances varies with our two
proxies for the returns to liquidity provision. We start with a graphical analysis, based on
equation 4. We split the sample into low and high VIX days, where high VIX days occur
when the VIX is higher than 20, its 2002-10 median. In each of these subsamples, we run
separate regressions of the cumulative returns from day x = 1 to day y where y take values
1 to 100 on day 0 retail imbalance, controlling for past weekly and monthly returns, as well
as the log of the market capitalization of the stock. We plot the estimated coefficients for
Imb[0] at each horizon (1 to 100), along with their 95% confidence intervals in Figure 2.
Panel A (resp. Panel B) corresponds to days of high VIX (resp. low VIX). Two interesting
facts emerge from Figure 2. First, when the VIX is high, retail imbalances are followed
by a much larger price increase over the subsequent couple of weeks than when the VIX
is low: stocks heavily purchased by individuals reach 30 basis points in cumulative returns
over the subsequent 16 days, while they reach about 15 basis points in cumulative returns
on low VIX days. Second, the reversal is much more pronounced on low VIX days than it
is on high VIX days. A potential interpretation of these results is that when uncertainty is
high, retail trades provide liquidity and are significantly compensated for this. Conversely,
in times of low uncertainty, retail trades are more likely to be picked-off by informed traders
and eventually generate negative cumulative returns.
We confirm these results in formal regression tests. We interact all terms in equation 4
with a dummy equal to 1 on days when the VIX is higher than its sample median between
2002 and 2010, and zero otherwise. The results are presented in Columns 1 and 2 of Table 6.
Column 1 includes only day fixed effects, while Column 2 adds stock fixed effects. Consistent

13
with the intuition obtained from Figure 2, the short-term rewards to individual investors’
liquidity provision are two times greater on high VIX days relative to low VIX days. A
one standard deviation increase in date-0 imbalances leads to a 10 basis points increase in
cumulative returns over the subsequent couple of weeks on low VIX days while it leads to
a 20 basis points increase when the VIX is above its sample median.10 As we did earlier,
we also investigate how the returns earned by stocks experiencing high retail imbalances
changed during the Crisis period (Columns 3 and 4 of Table 6). We find that the short-term
returns to individuals’ liquidity provision almost triples in value during the financial crisis:
while the sensitivity of three-week cumulative returns to retail imbalances is .0024 outside
the financial crisis, it is .0075 during the financial crisis.11
Altogether these results provide compelling evidence that the outperformance of stocks
heavily purchased by individuals over those heavily sold by individuals amounts to compen-
sation for liquidity provision. Of course, one objection to our interpretation is that during
high VIX periods, limits to arbitrage increase so that correlated trading by individual in-
vestors would be less likely to be arbitraged away by constrained sophisticated arbitrageurs.
As a result, the increased predictability during high VIX periods would simply be the result
of noisy price pressure from individuals and be unrelated to the returns to liquidity provi-
sion. However, we find in unreported tests that the autocorrelation of our measure of retail
imbalances is significantly lower on high VIX days, making this alternative interpretation
much less compelling than our hypothesis based on liquidity provision.
These results also emphasize the value of our dataset: because we track individuals for
a long period of time, we are able to reconcile apparently contradictory findings from the
literature with respect to the reversal of short-term returns. Indeed, while Hvidkjaer (2008),
10
We check and find that all results are unchanged when we use the VSTOXX, the European volatility
index, instead of the VIX.
11
In a robustness test presented in Table 14, we interact the Crisis and the High VIX dummies with a
dummy called SBF120, which takes the value of one for stocks included in the SBF120 index, and zero for
the 120 largest stocks not included in the SBF120. We find that most of the additional returns to liquidity
provision obtained during the crisis or in times of high volatility are found in stocks included in the SBF120
index. Given that institutions are likely to track the SBF120 index, this is consistent with the idea that
individual investors provide liquidity to meet institutional demand for immediacy.

14
and Barber et al. (2009) found evidence of reversal following short-term returns, Kaniel et
al. (2008), and Kelley and Tetlock (2013) found none. Our results suggest that different
samples and time periods might explain these differences.

3.3. Portfolio analysis

In this section, we show that portfolios mimicking the trades of individual investors generate
significant positive abnormal returns at the weekly horizon. There are several reasons for
this. First, we want to check that the effects we documented in the previous Section are
economically meaningful, and in particular, that they are not driven by the smallest trades
observed in our sample of individual investors. The portfolio we build aggregates trades
across individuals and assets and thus gives a greater weight to stocks heavily traded. Second,
we want to make sure that the results are not driven by a particular feature of Imb[0], our
measure of retail imbalances. In the analysis that follows, we pool all stocks into a long and
a short portfolios, and we therefore abstract from any measure of imbalances. Finally, we
want to analyze how the returns of the short-term reversal strategies of individual investors
load on systematic risk factors.
We proceed as follows. Each week over the sample period spanning 2002 to 2010, we
aggregate individual trades at the stock level. We sort stocks based on their net retail
aggregate position into two subsets of stocks sold and stocks purchased. We form a long and
a short portfolios by value-weighting the stocks in each of these two subsets. Hence a stock
enters in respectively the long and the short portfolio with a weight that reflects the size of
the aggregate individual imbalance in that stock in a given week. We rebalance each of the
two portfolios at the end of each week. We consider the returns on the long-short portfolio
(the “retail” portfolio). More precisely, we regress the returns on the long-short portfolio
on a model similar to that of equation 2. The coefficient of interest here is a, which is an
estimate of the weekly returns on the “retail” portfolio, adjusted for exposure to systematic
risk. We introduce the risk factors one by one in the model to assess their effects on the

15
estimate of excess returns. The results of the main specification are presented in Panel A of
Table 7. The unadjusted return is 26 weekly basis points, which amounts to 15% annualized
returns. A CAPM market model (Column 2) generates an alpha of 23 basis points per week,
which amounts to an annualized risk adjusted return of 13%. Moving to a Fama French risk
model (Column 4) increases the estimates to 33 basis points, an annualized return of 19%.
Introducing the Momentum factor (Column 5) in the model does not affect the estimate of
alpha.
If these excess returns represent compensation for liquidity provision, then we should
find, consistent with the analysis presented above, that these excess returns increase in
times of high uncertainty, i.e. when the rewards to liquidity provision increase. We thus
split the sample based on the level of the VIX in the last day of the portfolio formation
week. We then regress the excess returns on the zero cost “retail” portfolio on the same risk
factors as those used in equation 2, but using only those weeks where the level of the VIX is
above its sample median. The results are presented in Panel B of Table 7. The unadjusted
returns are 52 basis points weekly, 30% in annualized terms. Adjusting for the exposure to
the market increases the estimates to 57 basis points, which amounts to an annualized risk
adjusted performance of 34%. Introducing the three other risk factors pushes the returns
even further to an impressive annualized return of nearly 40%. In other words, irrespective
of the particular risk-adjustment, the excess returns earned by the “retail” portfolio are
two times greater in high VIX weeks relative to low VIX weeks. These impressive returns
strengthen the conclusion that retail trades provide liquidity to the markets.
Finally, we check whether the effects are stronger if we restrict the sample to stock-weeks
when retail imbalances are extreme. We define an imbalance as extreme if the stock-week
lies in the top or the bottom terciles of the distribution of the stock’s retail imbalances
between 2002 and 2010. We then compute the long and short portfolios as described above.
Given that the returns to liquidity provision are higher for large retail imbalances, we expect
the abnormal returns on this portfolio to be higher than what we obtained using the whole

16
sample. We run the same model and present the results in Panel C of Table 7. Unsurprisingly,
the results are larger than those obtained in Panel A. A CAPM market model delivers a
return of 34 basis points weekly, which amounts to an annualized return of 19%. A four-
factor model increases the alpha to 47 basis points weekly, or about 28% annually.

3.4. Order level analysis

The results aggregated at the stock level seem at odds with the results commonly found
in the literature that individual investors lose money on average (Odean, 1998; Barber and
Odean, 2000; Barber et al., 2006; Grinblatt and Keloharju, 2000), either because they trade
too much, or because they pick the losing stocks. In this Section, we attempt to reconcile
these results with our findings. We first notice that the fact that the “retail” portfolio
earns positive and significant excess returns does not necessarily mean that retail traders
earn significant trading profits. In fact, for individual investors to collect the returns from
liquidity provision, it needs to be the case that (i) the return on the day of the trade (day
0) should not be lower than the subsequent excess returns, and that (ii) their trades are
reversed before the rewards from liquidity provision are dissipated. We exploit the richness
of our data to explore the behavior of individual investors along these two dimensions.
For each of the approximately 5 million orders in the sample, we construct the following
variables. First, we define the return on day 0, Ret[0] as the difference between the closing
price at the end of the day when the order was placed, and the price at which the order was
executed during the day. We also define days to reversal as the number of days between the
date of an order, and the earliest date at which the order was at least partially reversed.12
We measure the holding period return as the cumulative return from the time of execution to
the close of the earliest date at which the order is partially reversed.13 For ease of comparison
12
We would obtain similar results by considering the number of days until the position is fully reversed,
however our measure is more conservative for the purpose of this study.
13
For simplicity, we cap the holding period to 500 days. When an order is never reversed in the sample,
we cap its holding period to the earliest of (i) the last day of trading in the sample and (ii) the last quotation
day of the stock if it delisted.

17
across holding periods, we also compute the internal rate of return of each trade. For most
of the analysis, we adjust both the holding period returns and the internal rate of returns
for exposure to systematic risk, following the procedure described in Section 2. Again, we
call Ret[x, y] the cumulative returns obtained from day x to day y. In addition, since we
are interested in the heterogeneity in experience across individual investors, we define the
cumulative number of orders for a trader i as the total number of orders placed prior to
placing a given order.
Summary statistics for the sample at the order level are presented in Table 1. The average
order is worth 7,741 euros. There are slightly more purchases than sales in the sample, but
purchases are slightly smaller (7,186 euros) than sales (8,342 euros). Turning to our variables
of interest, the average holding period is 310 days and the median is 40. This is much longer
than the average time at which the returns to liquidity provision are dissipated on average:
as is apparent from Figure 1, the cumulative returns following retail order imbalances are
only 10 basis points after 30 days and are 0 after 80 days. Additionally, the average return
on day zero, Ret[0], is -90 basis points. This is much larger (in absolute value) than the
average estimated rewards from liquidity provision (which is at best 25 basis points). Hence
the average trade in the sample does not reap the returns to liquidity provision because (i)
it is picked-off on day 0 and (ii) it is not reversed quickly enough.
Table 1 indicates that the average holding period return is -2.7%, and the average inter-
nal return is 4 basis points per day. On a risk adjusted basis, these numbers are respectively
-90 basis points, and 3 basis points.1415 Table 11 presents the correlations between these
variables. The number of days to reversal is negatively related to Ret[0] and the internal
rate of return, which are positively correlated. Quickly reversed trades are picked-off less,
on average. We decompose the holding period return in Table 8. We present the decompo-
14
The discrepancy between holding period returns and internal rates of returns comes from the fact that
losing positions tend to be held longer, so that their holding period returns are larger in absolute value than
the holding period returns of winning positions that are reversed quickly. The internal rate of return rescales
returns to the daily horizon, therefore correcting this bias.
15
The adjusted return on day 0 is computed as the difference between Ret[0] and the stock return predicted
by the four factor model presented in equation 2.

18
sition on both unadjusted and risk-adjusted terms. The loss on day 0, Ret[0], accounts for
approximately one third of the negative holding period returns, while the rest comes from
Ret]16, R], the returns from day 16 to the reversal of the trade. On a risk adjusted basis, the
results are very similar. One of the striking implications of these results is that individuals
seem to be losing money because they do not reverse their trades soon enough, i.e., because
they do not trade enough.

3.5. Heterogeneous effects

In this Section, we exploit the richness of our dataset to document the heterogeneity in the
behavior of individual investors. In particular, we show that certain individual characteristics
seem to be associated with a better ability to capture the returns to liquidity provision. The
first characteristic we consider is a trader’s experience. We sort orders in our sample based
on the experience of the trader, measured by the total number of prior orders placed. We
expect that experienced individual investors should be less “picked-off” and should reverse
their trades quicker. We thus simply compute the average of all our return variables across all
deciles of experience. The results are presented in Panel A of Table 9. The average cumulative
number of orders placed in the first decile of experience is 12 vs. 3,323 in the highest decile
of experience. Experienced traders flip their trades much faster than inexperienced ones.
Interestingly, experienced traders also get less picked-off. The difference between the first
and the tenth deciles of experience in Ret[0] is an impressive 100 basis points, which is
more than half their difference in holding period returns in risk-adjusted terms. Experienced
traders have slightly better returns between day 1 and day 16. Their risk adjusted holding
period return and internal rate of return are respectively 20 and 10 basis points larger than
those of inexperienced investors.
The second characteristic we consider is a traders’ average holding period. More precisely,
we sort orders in the sample based on the propensity of each trader to quickly reverse its
trades. We expect that individuals who have shorter holding periods should be more capable

19
of seizing the returns from liquidity provision. To check whether this is the case, we compute
the average holding period of the trader over the sample. We then compute the average of all
our return variables across the ten deciles of the distribution of holding periods. The results
are presented in Panel B of Table 9. The average cumulative number of orders placed in the
first decile of speed to reversals is 33, versus 1,065 in the highest decile of reversals. Traders
who usually reverse their positions faster also get less picked-off. The difference between
decile 1 and 10 in the average date-0 return, Ret[0], is 35 basis points. The risk adjusted
holding period returns and internal rates of returns of traders quickly reversing their trades
are respectively 4.5% and 10 basis points higher than those of traders in the bottom decile,
and decline in internal rates of return is monotonic.
Taken together, these results do suggest that experience and the average holding period
are two characteristics associated with higher reward to liquidity provision. In principle,
this result could emanate from two different channels. First, it could be that the worst
performing types (low experience, long holding periods) exit the sample more frequently.
Second, in the case of experience, it could be that individual investors indeed experience
some form of learning-by-doing. To get a quantitative sense of these two channels, we run
simple regressions of Ret[0] and of the log of the number of days to reversal on vectors of
time varying trader characteristics (including the log of the cumulative number of past orders
and its square, the log of the size of the account and the log monthly volume traded) and
day and stock × day fixed effects. Most importantly, we add individual fixed effects. If
there is any learning-by-doing, the outcome variables should be associated with our measure
of experience, the log of the cumulative number of past orders. We present the results in
Table 10. The results confirm that experience is strongly positively related to Ret[0], and
negatively related to the number of days to reversal. Traders with a larger number of past
orders are less picked-off and reverse their trades faster. However, the coefficients decrease
substantially when we introduce individual fixed effects. This suggests that an important
part of the learning occurs via the attrition of our sample, i.e., the survival of the trades

20
which are less picked-off and which reverse their trades more quickly.16 The results are
similar when we proxy for experience with the cumulative volume traded by a given retail
investors, as evidenced in Table 15.

3.6. Discussion

The results presented in this paper indicate that individual investors provide liquidity to
stock markets, and that some of them are compensated for it, especially in periods of high
uncertainty such as the 2008-09 financial crisis, when institutional liquidity providers were
most constrained. This seems at odds with the view according to which retail investors flee
to liquidity during times of financial market stress and thereby amplify the initial stress.
Financial newspapers, both in Europe and in the U.S., reported a massive exodus of small
retail investors from the stock market following the financial crisis, with potentially worrisome
consequences. According to the Wall Street Journal,17 in the U.S., “from 2007 through 2009,
[retail investors] withdrew money [from mutual funds that invest in U.S. stocks] for three
consecutive years”, which “marked the first three-year period of withdrawals since 1979-
1981”.
Using our data, we find that on aggregate, individual investors decreased their exposure
to mutual funds. However, they also significantly increased their exposure to equities. As
evidenced from Figure 3, the net outflows of individual investors in our sample from equity
mutual funds reached 150 million euros from mid-2007 to the first quarter of 2009. In the
meantime, inflows into stocks amount to approximately 100 million over the same period.
The results presented in this paper offer a new perspective on this somewhat surprising
finding: in the aggregate, individual investors acted as liquidity providers for the rest of the
market.
16
Interestingly, larger orders get picked-off less and start to get reversed faster.
17
See “Small Investors Flee Stocks, Changing Market Dynamics”, Wall Street Journal, June 2010.

21
4. Conclusion

This paper examines the extent to which individual investors provide liquidity to the stock
market, and whether or not they are compensated for doing so. We start by confirming with
our data that aggregate retail buy-sell imbalances are contrarian and positively predict the
cross-section of stock returns at a horizon of couple weeks. We then uncover three main
findings. First, rewards to liquidity provision increase sharply during the financial crisis of
2008-09, or more generally in times of high uncertainty. Second, individual investors fail to
reap the benefits from liquidity provision and this for two reasons: (i) they get picked-off
on the day of trading (ii) they do not reverse their trades quickly enough so that when they
close their trades, the returns to liquidity provision are dissipated. Third, we take advantage
of the richness of our data to document heterogeneity in the returns to liquidity provision
across individuals. We show that experience traders get less picked-off and reverse their
trade much faster than less-experienced traders. Overall, these two components explain a
significant share of the outperformance of experienced traders relative to less experienced
traders.
At least for investors in our sample, it is procrastination that leads to under-performance,
not too frequent trading. Finally, our data suggests that during the financial crisis retail
investors on aggregate fled from delegation, yet at the same time stepped up to the plate,
increased stock holdings and provided liquidity.

22
References

Adrian, Tobias, Erkko Etula, and Tyler Muir, “Financial intermediaries and the cross-
section of asset returns,” Staff Reports, 2012, 464.

Agarwal, Sumit, John C. Driscoll, Xavier Gabaix, and David Laibson, “Learning
in the Credit Card Market,” NBER Working Paper, February 2008, (13822).

Ang, Andrew, Sergiy Gorovyy, and Gregory B Van Inwegen, “Hedge fund leverage,”
Journal of Financial Economics, 2011, 102 (1), 102–126.

Barber, Brad M and Terrance Odean, “Trading Is Hazardous to Your Wealth: The
Common Stock Investment Performance of Individual Investors,” The Journal of Finance,
2000, 55 (2), 773–806.

and , “Online Investors: Do the Slow Die First?,” Review of Financial Studies, 2002,
15 (2), 455–487.

and , “The behavior of individual investors,” Handbook of the Economics of Finance,

2011, 2, 1533–1570.

, , and Ning Zhu, “Do retail trades move markets?,” Review of Financial Studies,
2009, 22 (1), 151–186.

, Yi-Tsung Lee, Yu-Jane Liu, and Terrance Odean, “Just How Much Do Individual
Investors Lose by Trading?,” Review of Financial Studies, 2006, 22 (2), 609–632.

Barber, Brad, Yi-Tsung Lee, Yu-Jane Liu, and Terrance Odean, “The cross-section
of speculator skill: Evidence from daytrading,” Journal of Financial Markets, 2014, 18,
1–24.

Ben-David, Itzhak, Francesco Franzoni, and Rabih Moussawi, “Hedge fund stock
trading in the financial crisis of 2007–2009,” Review of Financial Studies, 2012, 25 (1),
1–54.

23
Black, Fischer, “Noise,” The Journal of finance, 1986, 41 (3), 529–543.

Campbell, John Y., Tarun Ramadorai, and Allie Schwartz, “Caught on tape: In-
stitutional order flow, stock returns and earnings announcements,” Journal of Financial
Economics, 2009, 92 (1), 66–91.

Chevalier, Judith and Glenn Ellison, “Career Concerns of Mutual Fund Managers,”
Quarterly Journal of Economics, 1999, 114 (2), 389–432.

Chiang, Y M, D Hirshleifer, Y Qian, and A E Sherman, “Do Investors Learn from

Experience? Evidence from Frequent IPO Investors,” Review of Financial Studies, 2011,
24 (5), 1560–1589.

Choi, James J, David Laibson, Brigitte C Madrian, and Andrew Metrick, “Rein-
forcement Learning and Savings Behavior.,” The Journal of finance, 2009, 64 (6), 2515–
2534.

Coval, Joshua and Erik Stafford, “Asset fire sales (and purchases) in equity markets,”
Journal of Financial Economics, 2007, 86 (2), 479–512.

Coval, Joshua D, David A Hirshleifer, and Tyler Shumway, “Can Individual In-
vestors Beat the Market ? Can Individual Investors Beat the Market ?,” Social Science
Research, 2005, 1001 (04), 1–37.

Dorn, Daniel, Gur Huberman, and Paul Sengmueller, “Correlated trading and re-
turns,” The Journal of Finance, 2008, 63 (2), 885–920.

Foucault, Thierry, David Sraer, and David J Thesmar, “Individual Investors and
Volatility,” Journal of Finance, 2011, LXVI (4), 1369–1406.

French, Kenneth R, “Presidential Address: The Cost of Active Investing,” The Journal
of Finance, 2008, 63 (4), 1537–1573.

24
Gervais, Simon and Terrance Odean, “Learning to be overconfident,” Review of Finan-
cial Studies, 2001, 14 (1), 1–27.

Glaser, Markus and Martin Weber, “Overconfidence and trading volume,” The Geneva
Risk and Insurance Review, 2007, 32 (1), 1–36.

Greenwood, Robin and Stefan Nagel, “Inexperienced investors and bubbles,” Journal
of Financial Economics, 2009, 93 (2), 239–258.

Grinblatt, Mark and Matti Keloharju, “The investment behavior and performance
of various investor types: a study of Finland’s unique data set,” Journal of Financial
Economics, 2000, 55 (1), 43–67.

and , “Sensation Seeking, Overconfidence, and Trading Activity,” The Journal of

Finance, 2009, 64 (2), 549–578.

Hvidkjaer, Soeren, “Small trades and the cross-section of stock returns,” Review of Fi-
nancial Studies, 2008, 21 (3), 1123–1151.

Kaniel, Ron, Gideon Saar, and Sheridan Titman, “Individual Investor Trading and
Stock Returns,” Journal of Finance, 2008, 63 (1), 273–310.

, Shuming Liu, Gideon Saar, and Sheridan Titman, “Individual investor trading
and return patterns around earnings announcements,” The Journal of Finance, 2012, 67
(2), 639–680.

Kaustia, Markku, Eeva Alho, and Vesa Puttonen, “How Much Does Expertise Reduce
Behavioral Biases? The Case of Anchoring Effects in Stock Return Estimates,” Financial
Management, 2008, 37 (3), 391–412.

Kelley, Eric K and Paul C Tetlock, “How wise are crowds? Insights from retail orders
and stock returns,” The Journal of Finance, 2013, 68 (3), 1229–1265.

25
Linnainmaa, J T, “Why Do (Some) Households Trade So Much?,” Review of Financial
Studies, 2011, 24 (5), 1630–1666.

Linnainmaa, Juhani T, “Do limit orders alter inferences about investor performance and
behavior?,” The Journal of Finance, 2010, 65 (4), 1473–1506.

List, John A, “Does Market Experience Eliminate Market Anomalies?,” Quarterly Journal
of Economics, 2003, 118 (1), 41–71.

Mahani, Reza and Dan Bernhardt, “Financial Speculators’ Underperformance: Learn-

ing, Self-Selection, and Endogenous Liquidity,” The Journal of Finance, 2007, 62 (3),
1313–1340.

Nagel, Stefan, “Evaporating Liquidity,” Review of Financial Studies, 2012, 25 (7), 2005–
2039.

Nicolosi, Gina, Liang Peng, and Ning Zhu, “Do individual investors learn from their
trading experience?,” Journal of Financial Markets, 2009, 12 (2), 317–336.

Odean, Terrance, “Do Investors Trade Too Much?,” SSRN Electronic Journal, 1998, 89
(5), 1279–1298.

Seru, A, T Shumway, and N Stoffman, “Learning by Trading,” Review of Financial

Studies, 2009, 23 (2), 705–739.

Shleifer, Andrei and Lawrence H Summers, “The noise trader approach to finance,”
Journal of Economic perspectives, 1990, 4 (2), 19–33.

Statman, M, S Thorley, and K Vorkink, “Investor Overconfidence and Trading Vol-

ume,” Review of Financial Studies, 2006, 19 (4), 1531–1565.

26
5. Graphs

.001 .002 .003 .004 .005 .006

Cumulative returns
0
-.005 -.004 -.003 -.002 -.001

0 20 40 60 80 100
Days

Figure 1: Predicting returns using retail order imbalances. This graph plots the
coefficient on aggregate retail imbalances, Imb[0], in stock × day level regressions where the dependent
variable is R[1, x], the cumulative return from day 1 to day x (from 1 to 100) following the trading day,
and controls include past weekly and monthly returns, as well as market equity (see equation 4). Imb[0] is
defined as shares bought minus shares sold divided by shares bought plus shares sold. There are 730 distinct
stocks traded between 2002-2010.

27
 .001 .002 .003 .004 .005 .006
Cumulative returns
-.005 -.004 -.003 -.002 -.001 0

0 20 40 60 80 100
Days

Panel A: High VIX days

.001 .002 .003 .004 .005 .006
Cumulative returns
0
-.005 -.004 -.003 -.002 -.001

0 20 40 60 80 100
Days

Panel B: Low VIX days

Figure 2: Predicting returns using retail order imbalances, high vs. low VIX days.
This graph plots the coefficient on aggregate retail imbalances, Imb[0], in stock × day level regressions where
the dependent variable is R[1, x], the cumulative return from day 1 to day x (from 1 to 100) following the
trading day, and controls include past weekly and monthly returns, as well as market equity. Imb[0] is
measured using shares bought minus shares sold divided by shares bought plus shares sold. There are 730
distinct stocks traded between 2002-2010. Panel A and B present the estimates on the subsample of days
28 median.
when the VIX is respectively above and below its 2002-10
 50
Aggregate imbalance (EURMln)
-100 -50-150 0

2006q1 2007q1 2008q1 2009q1 2010q1 2011q1

Quarter

Stocks Equity Mutual Funds

Figure 3: Equity investment by individual investors during the crisis. This graph
plots the cumulative aggregate net flows into stocks and equity mutual funds from 2006 to 2010, in million
euros. The sample includes the trades of the 81,946 investors in our sample who traded during this period.

29
6. Tables

30
 Table 1: SUMMARY STATISTICS
This table presents summary statistics for the stock × day level sample (Panel A) and the order level
sample (Panel B). There are a total of 91,647 traders placing approximately 4.6 million orders in 730 stocks
from 2002 to 2010 in our sample, which leaves us with 217,511 stock-days. Imb[0] is our measure of retail
imbalances, defined for a stock-day as the number of shares purchased minus sold over the number of shares
purchased plus sold. Market equity is the log of the market capitalization of the stock. Retail volume is the
absolute value, in euros, of trades in the stock originating from traders in our sample. Share of retail volume
is the ratio of the number of shares of the stock traded in our sample divided by the market-wide number
of shares traded. Ret[x, y] is the cumulative holding period return from day x to day y. Days to reversal is
the number of days from the day the order was placed until the earliest date at which the order is at least
partially reversed. Crisis dummy is a dummy taking the value of one in the seven months from September
2008 to April 2009. High VIX is a dummy taking the value of one if the VIX is above its 2002-10 median.

Obs. Mean Std. dev. Median

Panel A: Stock×day level statistics

Imb[0] 217511 -0.0034 0.5381 0.0014

Market equity 217511 21.6121 2.0717 21.8274
Retail Volume 217511 165161 362247 54940
Share of retail volume 217480 0.0241 0.0388 0.0073
Ret[−26, −6] 217511 0.0116 0.1239 0.0108
Ret[−5, −1] 217511 0.0050 0.0646 0.0031
Ret[1, 16] 217511 0.0035 0.1001 0.0044
Ret[17, 100] 217511 0.0198 0.2336 0.0218
Ret[0] 217511 -0.0043 0.0137 -0.0037
Days to reversal 217511 223.8163 381.9827 62.0000
Crisis dummy 217511 0.0618 — —
High VIX 217511 0.4457 — —

Panel B: Order level statistics

Ret 4603607 -0.0269 0.2974 0.0045

Ret[0] 4639850 -0.0090 0.0251 -0.0066
AdjRet 4637552 -0.0093 0.2339 0.0034
AdjRet[0] 4637552 -0.0066 0.0269 -0.0051
IRR 4534305 0.0004 0.0101 0.0001
IRR[0] 4639850 -0.0090 0.0251 -0.0066
AdjIRR 4622152 0.0003 0.0084 0.0001
AdjIRR[0] 4637552 -0.0066 0.0269 -0.0051
Log cum. nb. of orders 4639850 5.3528 1.6332 5.4889
Squared log cum. nb. of orders 4639850 31.3194 16.7155 30.1284
Log days to reversal 4639101 3.8015 2.3198 3.6889
Purchase 4639850 0.5196 0.4996 1.0000
Log order size 4639850 7.8978 1.3445 7.8215
Market Equity 4639850 22.3344 1.9291 22.7396
Log mthly volume traded 4639850 2.9508 2.1494 2.9247
Log Account size 4639850 2.5069 1.7928 2.7193
Log nb of months since inception 4637671 4.1298 0.9851 4.3820

31
 Table 2: LIQUIDITY PROVISION
This table presents the results of stock×day level OLS regressions of retail order imbalances on past returns,
controls, and day and stock fixed effects. There are 730 distinct stocks traded between 2002-2010. Imb[0]
is our measure of retail imbalances, defined for a stock-day as the number of shares purchased minus sold
over the number of shares purchased plus sold. Market equity is the log of the market capitalization of the
stock. Ret[x, y] is the cumulative holding period return from day x to day y. Standard errors are corrected
for clustering at the stock level and are presented in parenthesis. ***,**, and * indicate significance at the
1, 5, and 10% respectively.

Dependent variable: Imb[0]

Ret[−5, −1] -1.03∗∗∗ -1.05∗∗∗

(0.05) (0.05)
Ret[−26, −6] -0.26∗∗∗ -0.26∗∗∗
(0.02) (0.02)
Market equity -0.02∗∗∗ -0.02∗∗∗
(0.00) (0.01)

Day FE Yes Yes

Stock FE No Yes

Observations 217511 217511

R2 0.075 0.090

32
 Table 3: RETURNS TO LIQUIDITY PROVISION
This table presents the results of stock×day level OLS regressions of future returns on retail order imbal-
ances, past returns, controls, and day and stock fixed effects. There are 730 distinct stocks traded between
2002-2010. Imb[0] is our measure of retail imbalances, defined for a stock-day as the number of shares
purchased minus sold over the number of shares purchased plus sold. Market equity is the log of the market
capitalization of the stock. Ret[x, y] is the cumulative holding period return from day x to day y. Standard
errors are corrected for clustering at the stock level and are presented in parenthesis. ***,**, and * indicate
significance at the 1, 5, and 10% respectively.

Ret[1, 16] Ret[17, 100] Ret[1, 16] Ret[17, 100]

Imb[0] 0.0027∗∗∗ -0.0028∗∗ 0.0027∗∗∗ -0.0021∗∗

(0.0004) (0.0011) (0.0004) (0.0010)
Ret[−5, −1] -0.0383∗∗∗ 0.0410∗∗ -0.0418∗∗∗ 0.0380∗∗
(0.0080) (0.0183) (0.0077) (0.0175)
Ret[−26, −6] -0.0002 0.0480∗∗∗ -0.0059 0.0332∗
(0.0069) (0.0182) (0.0065) (0.0183)
Market equity 0.0012∗∗∗ 0.0040∗∗ -0.0253∗∗∗ -0.1202∗∗∗
(0.0004) (0.0016) (0.0026) (0.0129)

Day FE Yes Yes Yes Yes

Stock FE No No Yes Yes

Observations 217511 217511 217511 217511

R2 0.362 0.406 0.388 0.496

33
 Table 4: RETURNS TO LIQUIDITY PROVISION - ALTERNATIVE PROXIES
This table presents the results of stock×day level OLS regressions of future returns on retail order imbalances,
past returns, controls, and day and stock fixed effects.There are 730 distinct stocks traded between 2002-2010.
Imb[0] is our measure of retail imbalances, defined for a stock-day as the number of shares purchased minus
sold over the number of shares purchased plus sold. Market equity is the log of the market capitalization
of the stock. Ret[x, y] is the cumulative holding period return from day x to day y. Standard errors are
corrected for clustering at the stock level and are presented in parenthesis. ***,**, and * indicate significance
at the 1, 5, and 10% respectively.

Ret[1, 16] Ret[17, 100] Ret[1, 16] Ret[17, 100]

Panel A: Imb[0], normalized

Imb[0] 0.0017∗∗∗ 0.0004 0.0014∗∗∗ -0.0011∗∗

(0.0002) (0.0006) (0.0002) (0.0006)
Ret[−5, −1] -0.0379∗∗∗ 0.0447∗∗ -0.0420∗∗∗ 0.0381∗∗
(0.0080) (0.0182) (0.0077) (0.0174)
Ret[−26, −6] -0.0002 0.0489∗∗∗ -0.0059 0.0332∗
(0.0069) (0.0181) (0.0065) (0.0183)
Market equity 0.0011∗∗∗ 0.0041∗∗∗ -0.0253∗∗∗ -0.1202∗∗∗
(0.0004) (0.0016) (0.0026) (0.0129)
Day FE Yes Yes Yes Yes
Stock FE No No Yes Yes
Observations 217483 217483 217483 217483
R2 0.362 0.406 0.388 0.496

Panel B: Buy-Sell over market wide volume, normalized

Normalized imbalances 0.0010∗∗∗ 0.0003 0.0008∗∗∗ -0.0008

(0.0002) (0.0005) (0.0002) (0.0005)
Ret[−5, −1] -0.0401∗∗∗ 0.0442∗∗ -0.0438∗∗∗ 0.0392∗∗
(0.0081) (0.0181) (0.0077) (0.0174)
Ret[−26, −6] -0.0007 0.0488∗∗∗ -0.0064 0.0336∗
(0.0069) (0.0181) (0.0065) (0.0182)
Market equity 0.0011∗∗∗ 0.0041∗∗∗ -0.0253∗∗∗ -0.1202∗∗∗
(0.0004) (0.0016) (0.0026) (0.0129)
Day FE Yes Yes Yes Yes
Stock FE No No Yes Yes
Observations 217458 217458 217458 217458
R2 0.362 0.406 0.388 0.495

Panel C: Tertiles of within stock Imb[0]

Medium Imb[0] 0.0017∗∗∗ -0.0012 0.0014∗∗∗ -0.0024∗

(0.0005) (0.0014) (0.0005) (0.0014)
Large Imb[0] 0.0037∗∗∗ 0.0008 0.0030∗∗∗ -0.0024∗
(0.0006) (0.0013) (0.0005) (0.0012)
Ret[−5, −1] -0.0384∗∗∗ 0.0447∗∗ -0.0424∗∗∗ 0.0387∗∗
(0.0080) (0.0182) (0.0076) (0.0174)
Ret[−26, −6] -0.0003 0.0489∗∗∗ -0.0061 0.0334∗
(0.0069) (0.0181) (0.0065) (0.0183)
Market equity 0.0011∗∗∗ 0.0041∗∗∗ -0.0253∗∗∗ -0.1202∗∗∗
(0.0004) (0.0016) (0.0026) (0.0129)
Day FE Yes Yes Yes Yes
Stock FE No 34 No Yes Yes
Observations 217511 217511 217511 217511
R2 0.362 0.406 0.388 0.496
 Table 5: LIQUIDITY PROVISION AND THE CRISIS
This table presents the results of stock×day level OLS regressions of retail order imbalances on past returns,
controls, and day and stock fixed effects. There are 730 stocks traded from 2002 to 2010 in our sample.
Imb[0] is our measure of retail imbalances, defined for a stock-day as the number of shares purchased minus
sold over the number of shares purchased plus sold. Market equity is the log of the market capitalization of
the stock. Ret[x, y] is the cumulative holding period return from day x to day y. Crisis dummy is a dummy
taking the value of one in the seven months from September 2008 to April 2009. High VIX is a dummy
taking the value of one if the VIX is above its 2002-10 median. Standard errors are corrected for clustering
at the stock level and are presented in parenthesis. ***,**, and * indicate significance at the 1, 5, and 10%
respectively.

Dependent variable: Imb[0]

High VIX × Ret[−5, −1] 0.04 0.09 -0.05 0.00

(0.08) (0.08) (0.08) (0.08)
High VIX × Ret[−26, −6] -0.02 -0.01 -0.05 -0.04
(0.03) (0.03) (0.03) (0.03)
Crisis × Ret[−5, −1] 0.41∗∗∗ 0.44∗∗∗ 0.45∗∗∗ 0.44∗∗∗
(0.09) (0.09) (0.09) (0.09)
Crisis × Ret[−26, −6] 0.13∗∗∗ 0.14∗∗∗ 0.16∗∗∗ 0.17∗∗∗
(0.05) (0.04) (0.05) (0.04)
Ret[−5, −1] -1.06∗∗∗ -1.11∗∗∗ -1.08∗∗∗ -1.10∗∗∗ -1.06∗∗∗ -1.11∗∗∗
(0.08) (0.08) (0.06) (0.06) (0.08) (0.08)
Ret[−26, −6] -0.25∗∗∗ -0.26∗∗∗ -0.27∗∗∗ -0.27∗∗∗ -0.25∗∗∗ -0.26∗∗∗
(0.03) (0.03) (0.02) (0.02) (0.03) (0.03)
High VIX × Market equity 0.01∗∗∗ 0.01∗∗∗ 0.01∗∗∗ 0.01∗∗∗
(0.00) (0.00) (0.00) (0.00)
Crisis × Market equity 0.01∗∗∗ 0.01∗∗∗ 0.01∗ 0.01∗
(0.00) (0.00) (0.00) (0.00)
Market equity -0.02∗∗∗ -0.03∗∗∗ -0.02∗∗∗ -0.02∗∗∗ -0.02∗∗∗ -0.03∗∗∗
(0.00) (0.01) (0.00) (0.01) (0.00) (0.01)

Day FE Yes Yes Yes Yes Yes Yes

Stock FE No Yes No Yes No Yes

Observations 217511 217511 217511 217511 217511 217511

R2 0.076 0.090 0.076 0.090 0.076 0.090

35
 Table 6: RETURNS TO LIQUIDITY PROVISION AND THE CRISIS
This table presents the results of stock×day level OLS regressions of future returns on retail order imbalances,
past returns, controls, and day and stock fixed effects.There are 730 stocks traded from 2002 to 2010 in
our sample. Imb[0] is our measure of retail imbalances, defined for a stock-day as the number of shares
purchased minus sold over the number of shares purchased plus sold. Market equity is the log of the market
capitalization of the stock. Ret[x, y] is the cumulative holding period return from day x to day y. Crisis
dummy is a dummy taking the value of one in the seven months from September 2008 to April 2009. High
VIX is a dummy taking the value of one if the VIX is above its 2002-10 median. Standard errors are corrected
for clustering at the stock level and are presented in parenthesis. ***,**, and * indicate significance at the
1, 5, and 10% respectively

Dependent variable: Ret[1, 16]

High VIX × Imb[0] 0.0020∗∗ 0.0016∗∗ 0.0013 0.0010

(0.0008) (0.0008) (0.0008) (0.0008)
Crisis × Imb[0] 0.0062∗∗∗ 0.0051∗∗ 0.0055∗∗ 0.0046∗∗
(0.0021) (0.0021) (0.0022) (0.0021)
Imb[0] 0.0018∗∗∗ 0.0019∗∗∗ 0.0023∗∗∗ 0.0024∗∗∗ 0.0018∗∗∗ 0.0019∗∗∗
(0.0005) (0.0005) (0.0004) (0.0004) (0.0005) (0.0005)
High VIX × Ret[−5, −1] -0.0521∗∗∗ -0.0394∗∗∗ -0.0511∗∗∗ -0.0413∗∗∗
(0.0146) (0.0143) (0.0151) (0.0150)
High VIX × Ret[−26, −6] -0.0391∗∗∗ -0.0318∗∗∗ -0.0401∗∗∗ -0.0366∗∗∗
(0.0121) (0.0117) (0.0122) (0.0118)
Crisis × Ret[−5, −1] -0.0302 -0.0125 -0.0042 0.0085
(0.0283) (0.0279) (0.0295) (0.0293)
Crisis × Ret[−26, −6] -0.0142 0.0065 0.0063 0.0251
(0.0277) (0.0280) (0.0283) (0.0287)
Ret[−5, −1] -0.0095 -0.0198∗∗ -0.0347∗∗∗ -0.0404∗∗∗ -0.0095 -0.0199∗∗
(0.0107) (0.0098) (0.0084) (0.0080) (0.0107) (0.0098)
Ret[−26, −6] 0.0207∗∗ 0.0113 0.0012 -0.0068 0.0207∗∗ 0.0113
(0.0092) (0.0084) (0.0071) (0.0065) (0.0092) (0.0083)
High VIX × Market equity 0.0010∗ 0.0022∗∗∗ 0.0011∗ 0.0020∗∗∗
(0.0005) (0.0006) (0.0005) (0.0006)
Crisis × Market equity 0.0000 0.0023 -0.0007 0.0012
(0.0015) (0.0015) (0.0015) (0.0015)
Market equity 0.0008∗ -0.0266∗∗∗ 0.0012∗∗∗ -0.0256∗∗∗ 0.0008∗ -0.0267∗∗∗
(0.0005) (0.0026) (0.0004) (0.0025) (0.0005) (0.0026)
Day FE Yes Yes Yes Yes Yes Yes
Stock FE No Yes No Yes No Yes
Observations 217511 217511 217511 217511 217511 217511
R2 0.363 0.389 0.362 0.389 0.363 0.389

36
 Table 7: WEEKLY REBALANCED PORTFOLIO RETURNS
This table presents the excess returns (Alpha) on a weekly rebalanced portfolio long in the (value-weighted)
stocks purchased and short in the (value-weighted) stocks sold by retail investors. Each week over the sample
period running from 2002 to 2010, we aggregate individual trades at the stock level. We sort stocks based
on their net retail aggregate position into two subsets of stocks sold and stocks purchased. We form a long
and a short portfolios by value-weighting the stocks in each of these two subsets. We rebalance each of
the two portfolios at the end of each week. The weekly returns on the long-short portfolio are regressed
on the weekly returns on the Market, Small-minus-Big, High-minus-Low and Momentum factors. Panel A
presents the results for the whole sample; Panel B restricts the sample to weeks when the VIX is above its
2002-10; Panel C restricts the sample to stock-weeks with extreme retail imbalances. ***,**, and * indicate
significance at the 1, 5, and 10% respectively.

Panel A: All stock-weeks, all orders

Alpha 0.0026∗∗ 0.0023∗∗ 0.0029∗∗∗ 0.0033∗∗∗ 0.0033∗∗∗

(0.0011) (0.0011) (0.0011) (0.0011) (0.0011)
Market 0.3711∗∗∗ 0.2673∗∗∗ 0.2400∗∗∗ 0.2276∗∗∗
(0.0656) (0.0691) (0.0740) (0.0784)
SMB -0.2008∗∗∗ -0.1925∗∗∗ -0.1893∗∗∗
(0.0480) (0.0487) (0.0492)
HML -0.0754 -0.0618
(0.0731) (0.0784)
Momentum -0.0280
(0.0579)
Observations 467 467 467 467 467
R2 0.000 0.064 0.098 0.100 0.101

Panel B: High VIX weeks, all orders

Alpha 0.0052∗∗∗ 0.0058∗∗∗ 0.0059∗∗∗ 0.0064∗∗∗ 0.0064∗∗∗

(0.0019) (0.0018) (0.0018) (0.0018) (0.0018)
Market 0.4197∗∗∗ 0.2670∗∗∗ 0.2130∗ 0.1962
(0.0909) (0.0984) (0.1102) (0.1223)
SMB -0.2378∗∗∗ -0.2291∗∗∗ -0.2263∗∗∗
(0.0665) (0.0669) (0.0676)
HML -0.1102 -0.0978
(0.1013) (0.1086)
Momentum -0.0265
(0.0830)
Observations 234 234 234 234 234
R2 0.000 0.084 0.132 0.137 0.137

Panel C: All stock-weeks, extreme imbalances

Alpha 0.0037∗∗∗ 0.0034∗∗∗ 0.0042∗∗∗ 0.0047∗∗∗ 0.0047∗∗∗

(0.0012) (0.0012) (0.0012) (0.0012) (0.0012)
Market 0.3819∗∗∗ 0.2488∗∗∗ 0.2073∗∗ 0.1925∗∗
(0.0731) (0.0764) (0.0818) (0.0867)
SMB -0.2576∗∗∗ -0.2449∗∗∗ -0.2412∗∗∗
(0.0531) (0.0538) (0.0543)
HML -0.1145 -0.0984
(0.0808) (0.0867)
Momentum -0.0333
(0.0639)
Observations 467 467 37 467 467 467
R2 0.000 0.056 0.101 0.105 0.105
 Table 8: DECOMPOSITION OF RETURNS
This table presents the decomposition of retail traders’ returns between the return on day 0, the returns
between days 1 and 16 and the returns between day 16 after the date of the order and the date R or reversal.
Ret[x, y] and AdjRet[x, y] are respectively the raw and the risk adjusted (four factor model) cumulative
returns from day x to day y. IRR[x, y] and AdjIRR[x, y] are respectively the raw and the risk adjusted
(four factor model) internal rate of return from day x to day y. Days to reversal is the number of days from
the day the order was placed until the earliest date at which the order is at least partially reversed.

Mean p25 p50 p75

Days to reversal 310 7 40 346

Raw holding period returns

Ret -0.0269 -0.0951 0.0045 0.0775
Ret[0] -0.0090 -0.0166 -0.0066 0.0011
Ret[0, 16] -0.0094 -0.0651 -0.0082 0.0472
Ret[1, 16] -0.0004 -0.0546 -0.0000 0.0539
Ret[1, R] -0.0171 -0.0838 0.0093 0.0841
Ret]16, R] -0.0268 -0.1663 0.0067 0.1359

Risk adjusted holding period returns (four factor model)

AdjRet -0.0093 -0.0666 0.0034 0.0660
AdjRet[0] -0.0066 -0.0169 -0.0051 0.0059
AdjRet[0, 16] -0.0062 -0.0502 -0.0052 0.0388
AdjRet[1, 16] 0.0004 -0.0415 0.0005 0.0424
AdjRet[1, R] -0.0026 -0.0565 0.0062 0.0680
AdjRet]16, R] -0.0053 -0.1067 0.0083 0.1162

Raw internal rates of returns

IRR 0.0004 -0.0017 0.0001 0.0021
IRR[0] -0.0090 -0.0166 -0.0066 0.0011
IRR[0, 16] -0.0009 -0.0039 -0.0005 0.0027
IRR[1, 16] -0.0003 -0.0035 -0.0000 0.0033
IRR[1, R] 0.0014 -0.0008 -0.0000 0.0013
IRR]16, R] 0.0003 -0.0011 0.0001 0.0011

Risk adjusted internal rates of returns (four factor model)

AdjIRR 0.0003 -0.0014 0.0001 0.0015
AdjIRR[0] -0.0066 -0.0169 -0.0051 0.0059
AdjIRR[0, 16] -0.0006 -0.0030 -0.0003 0.0022
AdjIRR[1, 16] -0.0001 -0.0026 0.0000 0.0026
AdjIRR[1, R] 0.0008 -0.0007 0.0000 0.0009
AdjIRR]16, R] 0.0003 -0.0008 0.0001 0.0009

38
 Table 9: HETEROGENEITY IN RETURNS
This table presents the heterogeneity in returns across investor types. To construct Panel A, we sort orders in our sample based on the experience
of the trader, measured with the total number of prior orders placed.We compute the average of all our return variables for all deciles of experience.
In Panel B, we sort orders in the sample based on the propensity of each trader to quickly flip their positions. We compute the average holding
period of the trader. We then compute the average of all our return variables for all deciles of average days to reversal. Ret[x, y] and AdjRet[x, y] are
respectively the raw and the risk adjusted (four factor model) cumulative returns from day x to day y. IRR[x, y] and AdjIRR[x, y] are respectively
the raw and the risk adjusted (four factor model) internal rate of return from day x to day y. Days to reversal is the number of days from the day
the order was placed until the earliest date at which the order is at least partially reversed. ***,**, and * indicate that differences are significant at
the 1, 5, and 10% respectively.

PANEL A: SORTED BY EXPERIENCE

Decile of cum. Nb. of Cum. nb. Days to Raw Risk adjusted
nb. of orders orders of orders reversal Ret[0] Ret[1, 16] Ret IRR AdjRet[0] AdjRet[1, 16] AdjRet AdjIRR
1 462019 12 652 -0.0169 -0.0011 -0.0436 0.0002 -0.0150 0.0007 -0.0094 0.0000
2 468814 41 515 -0.0114 -0.0010 -0.0327 0.0003 -0.0090 0.0002 -0.0097 0.0001
3 452279 79 429 -0.0097 -0.0009 -0.0307 0.0003 -0.0074 0.0003 -0.0112 0.0001
4 460676 129 364 -0.0089 -0.0004 -0.0299 0.0004 -0.0066 0.0003 -0.0104 0.0002
5 458042 198 308 -0.0083 -0.0004 -0.0280 0.0004 -0.0060 0.0003 -0.0091 0.0002
6 461210 294 259 -0.0078 -0.0005 -0.0256 0.0004 -0.0055 0.0004 -0.0097 0.0003
7 460452 436 213 -0.0074 -0.0002 -0.0240 0.0005 -0.0051 0.0004 -0.0091 0.0004
8 459786 665 167 -0.0069 0.0000 -0.0216 0.0005 -0.0044 0.0004 -0.0086 0.0005

39
9 460308 1102 126 -0.0064 0.0001 -0.0175 0.0005 -0.0038 0.0004 -0.0083 0.0005
10 460021 3323 79 -0.0066 0.0006 -0.0151 0.0006 -0.0036 0.0009 -0.0070 0.0007
ALL 4639850 632 310 -0.0090 -0.0004 -0.0269 0.0004 -0.0066 0.0004 -0.0093 0.0003
TEST 9+10 vs 1+2 2185.606*** -480.647*** 0.008*** 0.001*** 0.022*** 0.000*** 0.008*** 0.000* 0.002*** 0.001***
TEST 10 vs 1 3310.952*** -572.647*** 0.010*** 0.002*** 0.028*** 0.000*** 0.011*** 0.000 0.002*** 0.001***

PANEL B: SORTED BY DAYS TO REVERSAL

Decile of days Nb. of Nb. of Cum. nb. Days to Raw Risk adjusted
to reversal traders orders of orders reversal Ret[0] Ret[1, 16] Ret IRR AdjRet[0] AdjRet[1, 16] AdjRet AdjIRR
10 9129 18573 33 1986 -0.0419 -0.0048 -0.1946 -0.0005 -0.0400 -0.0023 -0.0522 -0.0004
9 9189 33847 36 1539 -0.0309 -0.0049 -0.2247 -0.0007 -0.0288 0.0024 -0.0078 -0.0003
8 9165 53599 54 1241 -0.0230 -0.0058 -0.1074 -0.0003 -0.0209 0.0016 -0.0108 -0.0002
7 9175 81995 67 1028 -0.0181 -0.0054 -0.0398 -0.0001 -0.0157 0.0007 -0.0126 -0.0002
6 9165 134651 94 858 -0.0148 -0.0030 -0.0287 0.0000 -0.0125 -0.0001 -0.0095 -0.0001
5 9164 213725 96 706 -0.0129 -0.0012 -0.0323 0.0001 -0.0109 0.0004 -0.0122 -0.0001
4 9165 326482 135 565 -0.0112 -0.0010 -0.0298 0.0002 -0.0092 0.0003 -0.0094 0.0000
3 9165 523789 189 427 -0.0101 -0.0004 -0.0310 0.0003 -0.0082 0.0004 -0.0125 0.0000
2 9165 960769 315 280 -0.0086 -0.0001 -0.0282 0.0004 -0.0066 0.0004 -0.0109 0.0002
1 9165 2292420 1065 111 -0.0066 0.0001 -0.0177 0.0006 -0.0039 0.0004 -0.0071 0.0005
ALL 91647 4639850 632 310 -0.0090 -0.0004 -0.0269 0.0004 -0.0066 0.0004 -0.0093 0.0003
TEST 1 vs 10 1032.256*** -1874.871*** 0.035*** 0.005*** 0.177*** 0.001*** 0.036*** 0.003*** 0.045*** 0.001***
TEST 1+2 vs 9+10 808.473*** -1536.694*** 0.028*** 0.005*** 0.193*** 0.001*** 0.028*** -0.000 0.015*** 0.001***
Table 10: PICKING-OFF EFFECT AND TIME TO REVERSAL WITHIN INDIVIDUAL
This table presents the results of order level OLS regressions of the return on day 0, Ret[0] and the number
of days to reversal on vectors of time varying order characteristics, trader characteristics, and day, trader
and stock × day fixed effects. The dependent variable, Ret[0], is computed as the percentage change from
the execution price to the closing price on the day the order is placed. Days to reversal is the number of days
from the day the order was placed until the earliest date at which the order is at least partially reversed.
Purchase is a dummy equal to one if the order is a purchase and zero if it is a sale. Standard errors are
corrected for clustering at the stock level and are presented in parenthesis. ***,**, and * indicate significance
at the 1, 5, and 10% respectively.

Panel A: Log nb. of days to reversal

Log cum. nb. of orders -0.1093∗∗∗ -0.0852∗∗∗ -0.0405∗∗ -0.0418∗∗

(0.021) (0.022) (0.017) (0.016)
Squared log cum. nb. of orders -0.0184∗∗∗ -0.0208∗∗∗ 0.0232∗∗∗ 0.0236∗∗∗
(0.002) (0.002) (0.002) (0.002)
Log mthly volume traded -0.2521∗∗∗ -0.2458∗∗∗ -0.1006∗∗∗ -0.0991∗∗∗
(0.004) (0.004) (0.002) (0.002)
Log Account size 0.2174∗∗∗ 0.1847∗∗∗ 0.0252∗∗∗ 0.0245∗∗∗
(0.007) (0.005) (0.002) (0.002)
Purchase -1.1840∗∗∗ -1.1799∗∗∗ -1.2068∗∗∗ -1.2098∗∗∗
(0.041) (0.043) (0.041) (0.044)
Log order size -0.2466∗∗∗ -0.2177∗∗∗ -0.2923∗∗∗ -0.2805∗∗∗
(0.012) (0.010) (0.008) (0.005)
Day FE Yes No Yes No
Stock × Day FE No Yes No Yes
Trader FE No No Yes Yes

Observations 4639101 4639101 4639101 4639101

R2 0.331 0.404 0.461 0.516

Panel B: Ret[0]

Log cum. nb. of orders 0.0035∗∗∗ 0.0037∗∗∗ 0.0010∗∗∗ 0.0009∗∗∗

(0.000) (0.000) (0.000) (0.000)
Squared log cum. nb. of orders -0.0002∗∗∗ -0.0002∗∗∗ -0.0001∗∗∗ -0.0001∗∗∗
(0.000) (0.000) (0.000) (0.000)
Log mthly volume traded -0.0006∗∗∗ -0.0007∗∗∗ -0.0002∗∗∗ -0.0002∗∗∗
(0.000) (0.000) (0.000) (0.000)
Log Account size 0.0004∗∗∗ 0.0003∗∗∗ 0.0001∗∗∗ 0.0001∗∗∗
(0.000) (0.000) (0.000) (0.000)
Purchase 0.0024∗∗∗ 0.0024∗∗∗ 0.0015∗∗∗ 0.0016∗∗∗
(0.000) (0.000) (0.000) (0.000)
Log order size 0.0050∗∗∗ 0.0052∗∗∗ 0.0050∗∗∗ 0.0054∗∗∗
(0.000) (0.000) (0.000) (0.000)
Day FE Yes No Yes No
Stock × Day FE No Yes No Yes
Trader FE No No Yes Yes

Observations 4639850 4639850 4639850 4639850

R2 0.096 0.206 0.182 0.285

40
7. Robustness tables

41
 Table 11: CORRELATIONS
This table presents the correlations between variables in the order level sample. There are
a total of 91,647 traders placing approximately 4.6 million orders in 730 stocks from 2002
to 2010 in our sample, which leaves us with 217,511 stock-days. HPR is the holding period
return of any given order. HP R[0] is the difference between the closing price on the day of
the order and the execution price. Days to reversal is the number of days from the day the
order was placed until the earliest date at which the order is at least partially reversed. Panel
A presents the correlations based on raw holding period returns, while Panel B presents those
correlations based on risk adjusted holding period returns.

Panel A: Raw holding period returns

Ret Ret[0] Days to reversal

Ret 1
Ret[0] .103*** 1
Days to reversal -.197*** -.117*** 1

Panel B: Risk adjusted holding periodreturns

All orders

AdjHPR HP R[0] Days to reversal

AdjHPR 1
AdjHP R[0] .101*** 1
Days to reversal -.0322*** -.118*** 1

42
 Table 12: RETURNS TO LIQUIDITY PROVISION - ALTERNATIVE CLUSTERING
This table presents the results of stock×day level OLS regressions of future returns on retail order imbalances,
past returns, controls, and day and stock fixed effects.There are 730 distinct stocks traded between 2002-2010.
Imb[0] is our measure of retail imbalances, defined for a stock-day as the number of shares purchased minus
sold over the number of shares purchased plus sold. Market equity is the log of the market capitalization
of the stock. Ret[x, y] is the cumulative holding period return from day x to day y. In Columns 1 and 2,
standard errors are clustered by day. In Columns 3 and 4, they are clustered two ways, by stock and day.
***,**, and * indicate significance at the 1, 5, and 10% respectively.

Ret[1, 16] Ret[17, 100] Ret[1, 16] Ret[17, 100]

Imb[0] 0.0027∗∗∗ -0.0028∗∗∗ 0.0027∗∗∗ -0.0028∗∗

(0.0004) (0.0008) (0.0005) (0.0011)
Ret[−5, −1] -0.0383∗∗∗ 0.0410∗∗∗ -0.0383∗∗∗ 0.0410∗∗
(0.0070) (0.0144) (0.0095) (0.0208)
Ret[−26, −6] -0.0002 0.0480∗∗∗ -0.0002 0.0480∗∗
(0.0044) (0.0085) (0.0078) (0.0192)
Market equity 0.0012∗∗∗ 0.0040∗∗∗ 0.0012∗∗∗ 0.0040∗∗
(0.0002) (0.0004) (0.0004) (0.0016)

Cluster Day Day Day and Stock Day and Stock

Day FE Yes Yes Yes Yes

Observations 217511 217511 217511 217511

R2 0.362 0.406 0.002 0.003

43
 Table 13: RETURNS TO LIQUIDITY PROVISION AND TRADING VOLUME
This table presents the results of stock×day level OLS regressions of future returns on retail order imbal-
ances, past returns, controls, and day and stock fixed effects. There are 730 stocks traded from 2002 to 2010
in our sample. Imb[0] is our measure of retail imbalances, defined for a stock-day as the number of shares
purchased minus sold over the number of shares purchased plus sold. Market equity is the log of the market
capitalization of the stock. Ret[x, y] is the cumulative holding period return from day x to day y. Days to
reversal is the number of days from the day the order was placed until the earliest date at which the order
is at least partially reversed. Crisis dummy is a dummy taking the value of one in the seven months from
September 2008 to April 2009. High VIX is a dummy taking the value of one if the VIX is above its 2002-10
median. Standard errors are corrected for clustering at the stock level and are presented in parenthesis.
***,**, and * indicate significance at the 1, 5, and 10% respectively

Ret[1, 16] Ret[17, 100] Ret[1, 16] Ret[17, 100]

Panel A: high vs. low market wide volume

Imb[0] 0.0019∗∗∗ -0.0052∗∗∗ 0.0020∗∗∗ -0.0034∗∗
(0.0006) (0.0015) (0.0006) (0.0013)
High vol × Imb[0] 0.0019∗∗ 0.0055∗∗ 0.0016∗ 0.0029
(0.0009) (0.0023) (0.0009) (0.0021)
High vol -0.0026∗ -0.0158∗∗ 0.0010 -0.0112
(0.0014) (0.0063) (0.0016) (0.0081)
Ret[−5, −1] -0.0414∗∗∗ 0.0417∗∗ -0.0455∗∗∗ 0.0380∗∗
(0.0079) (0.0184) (0.0077) (0.0176)
Ret[−26, −6] 0.0002 0.0471∗∗∗ -0.0056 0.0325∗
(0.0068) (0.0178) (0.0065) (0.0181)
Market equity 0.0017∗∗∗ 0.0069∗∗∗ -0.0252∗∗∗ -0.1185∗∗∗
(0.0005) (0.0020) (0.0026) (0.0129)

Day FE Yes Yes Yes Yes

Stock FE No No Yes Yes

Observations 217511 217511 217511 217511

R2 0.363 0.410 0.388 0.496

Panel B: high vs. low within firm market wide volume

Imb[0] 0.0029∗∗∗ -0.0019 0.0026∗∗∗ -0.0023∗

(0.0006) (0.0014) (0.0006) (0.0013)
High vol × Imb[0] -0.0004 -0.0016 0.0001 0.0005
(0.0008) (0.0019) (0.0008) (0.0017)
High vol -0.0057∗∗∗ -0.0305∗∗∗ 0.0003 -0.0028
(0.0009) (0.0040) (0.0009) (0.0033)
Ret[−5, −1] -0.0356∗∗∗ 0.0559∗∗∗ -0.0419∗∗∗ 0.0392∗∗
(0.0081) (0.0188) (0.0076) (0.0175)
Ret[−26, −6] 0.0006 0.0525∗∗∗ -0.0059 0.0334∗
(0.0070) (0.0181) (0.0065) (0.0183)
Market equity 0.0013∗∗∗ 0.0047∗∗∗ -0.0254∗∗∗ -0.1193∗∗∗
(0.0004) (0.0016) (0.0026) (0.0133)

Day FE Yes Yes Yes Yes

Stock FE No No Yes Yes

Observations 217511 217511 217511 217511

R2 0.363 0.410 0.388 0.496
44
Table 14: RETURNS TO LIQUIDITY PROVISION AND THE CRISIS, CONDITIONAL
ON INDEX INCLUSION
This table presents the results of a specification similar to the one whose results are presented in Table 6,
with an additional term, SBF120. SBF120 takes the value of one for stocks included in the SBF120 index,
and zero for the 120 largest stocks not included in the SBF120. Imb[0] is our measure of retail imbalances,
defined for a stock-day as the number of shares purchased minus sold over the number of shares purchased
plus sold. Market equity is the log of the market capitalization of the stock. Ret[x, y] is the cumulative
holding period return from day x to day y. Crisis dummy is a dummy taking the value of one in the seven
months from September 2008 to April 2009. High VIX is a dummy taking the value of one if the VIX is
above its 2002-10 median. Standard errors are corrected for clustering at the stock level and are presented
in parenthesis. ***,**, and * indicate significance at the 1, 5, and 10% respectively

Dependent variable: Ret[1, 16]

SBF120 × High VIX × Imb[0] 0.0028∗ 0.0029∗

(0.0017) (0.0017)
High VIX × Imb[0] 0.0008 0.0007
(0.0013) (0.0013)
SBF120 × Crisis × Imb[0] 0.0088∗ 0.0092∗∗
(0.0046) (0.0045)
Crisis × Imb[0] 0.0020 0.0010
(0.0040) (0.0038)
SBF120 × Imb[0] -0.0000 -0.0001 0.0008 0.0007
(0.0010) (0.0010) (0.0010) (0.0009)
Imb[0] 0.0013∗ 0.0015∗ 0.0015∗∗ 0.0017∗∗
(0.0008) (0.0008) (0.0007) (0.0007)
SBF120 0.0218 0.0487 0.0264 0.0522
(0.0436) (0.1252) (0.0439) (0.1253)

Controls Yes Yes Yes Yes

Day FE Yes Yes Yes Yes
Stock FE No Yes No Yes

Observations 168305 168305 168305 168305

R2 0.419 0.437 0.419 0.437

45
Table 15: PICKING-OFF EFFECT AND TIME TO REVERSAL WITHIN INDIVIDUAL,
ALTERNATIVE PROXY FOR EXPERIENCE
This table presents the results of order level OLS regressions of the return on day 0, Ret[0] and the number
of days to reversal on vectors of time varying order characteristics, trader characteristics, and day, trader
and stock × day fixed effects. The dependent variable, Ret[0], is computed as the percentage change from
the execution price to the closing price on the day the order is placed. Days to reversal is the number of days
from the day the order was placed until the earliest date at which the order is at least partially reversed.
Purchase is a dummy equal to one if the order is a purchase and zero if it is a sale. Standard errors are
corrected for clustering at the stock level and are presented in parenthesis. ***,**, and * indicate significance
at the 1, 5, and 10% respectively.

Panel A: Log nb. of days to reversal

Log cum. volume traded -0.3911∗∗∗ -0.3680∗∗∗ -0.0502∗∗∗ -0.0441∗∗∗

(0.023) (0.024) (0.012) (0.012)
Squared log cum. volume traded 0.0139∗∗∗ 0.0122∗∗∗ 0.0144∗∗∗ 0.0139∗∗∗
(0.002) (0.002) (0.001) (0.001)
Log mthly volume traded -0.2746∗∗∗ -0.2712∗∗∗ -0.1037∗∗∗ -0.1020∗∗∗
(0.005) (0.004) (0.002) (0.002)
Log Account size 0.2139∗∗∗ 0.1821∗∗∗ 0.0278∗∗∗ 0.0268∗∗∗
(0.007) (0.005) (0.002) (0.002)
Purchase -1.1744∗∗∗ -1.1706∗∗∗ -1.2074∗∗∗ -1.2102∗∗∗
(0.041) (0.044) (0.041) (0.044)
Log order size -0.1193∗∗∗ -0.0836∗∗∗ -0.3002∗∗∗ -0.2893∗∗∗
(0.013) (0.011) (0.008) (0.005)
Day FE Yes No Yes No
Stock × Day FE No Yes No Yes
Trader FE No No Yes Yes

Observations 4639101 4639101 4639101 4639101

R2 0.324 0.396 0.461 0.516

Panel B: Ret[0]

Log cum. volume traded 0.0055∗∗∗ 0.0058∗∗∗ 0.0019∗∗∗ 0.0018∗∗∗

(0.000) (0.000) (0.000) (0.000)
Squared log cum. volume traded -0.0004∗∗∗ -0.0004∗∗∗ -0.0002∗∗∗ -0.0002∗∗∗
(0.000) (0.000) (0.000) (0.000)
Log mthly volume traded -0.0000 -0.0001∗∗ -0.0001∗∗∗ -0.0002∗∗∗
(0.000) (0.000) (0.000) (0.000)
Log Account size 0.0003∗∗∗ 0.0002∗∗∗ 0.0001∗∗∗ 0.0001∗∗∗
(0.000) (0.000) (0.000) (0.000)
Purchase 0.0022∗∗∗ 0.0023∗∗∗ 0.0015∗∗∗ 0.0016∗∗∗
(0.000) (0.000) (0.000) (0.000)
Log order size 0.0047∗∗∗ 0.0048∗∗∗ 0.0050∗∗∗ 0.0054∗∗∗
(0.000) (0.000) (0.000) (0.000)
Day FE Yes No Yes No
Stock × Day FE No Yes No Yes
Trader FE No No Yes Yes

Observations 4639850 4639850 4639850 4639850

R2 0.102 0.212 0.182 0.285

46