Investigating the Nature of Retail
Investor Activity
A thesis presented for the degree of
Bachelor of Financial Economics
Special thanks to Marcin Zamojski for the tireless guidance and support throughout the process of
writing this thesis. We are very happy to have had you as our supervisor!
s
Authors: Johan Birgersson
Samuel Carle´n
Supervisor: Marcin Zamojski
Submission Date: June 7, 2021
Contents
1 Introduction 1
2 Literature Review 5
2.1 Trading Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Retrieving information from trade data . . . . . . . . . . . . . . . . . . . 6
2.3 Retail Investor Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3 Method 10
3.1 Data collection and processing . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 The simplified model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4 Results and Analysis 17
4.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3 Results from the full model on short period . . . . . . . . . . . . . . . . . 20
4.4 Decomposition results for short period . . . . . . . . . . . . . . . . . . . 22
1
CONTENTS 2
4.5 Results from the simplified model on long period . . . . . . . . . . . . . . 22
4.6 Decomposition results for long period . . . . . . . . . . . . . . . . . . . . 23
4.7 Differences between the models and time samples . . . . . . . . . . . . . 24
4.8 Reporting days . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.9 The meme stocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.10 Covid effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5 Discussion 28
5.1 Results compared to previous research . . . . . . . . . . . . . . . . . . . 28
5.2 The trend of increased volume of retail trades
A need for better financial literacy? . . . . . . . . . . . . . . . . . . . . . . 31
6 Conclusion 33
A 38
Abstract
We study the nature of retail investor activity and how this differs when looking at pe-
riods in connection with quarterly reports and when looking at a special period on the
market like Covid-19. Further, we investigate how this differs between S&P500 stocks
and meme stocks. We find that retail order imbalances can predict positive future returns
for S&P500 stocks and find suggestive evidence that this doesn’t differ significantly for
periods in connection with reports. We find that this doesn’t differ significantly during
the initial impact of Covid-19. We do however find evidence that retail order imbal-
ances predict negative returns for meme stocks, contrary to the behavior exhibited in the
S&P500 stocks and in previous literature.
1Introduction
We expand on the literature surrounding the nature of retail investor activity by looking
at patterns of retail trade between 2016 to 2021. We look at how the predictability of
returns for retail order imbalances differs in in periods in connection with quarterly reports
and in connection with the initial impact of Covid-19. We also investigate how this differs
between S&P500 stocks and ”meme stocks” like GameStop or AMC Entertainment.
Retail trading has been widely debated among academics and regulators for a long time,
but even more so recently as an emergence of e-trading platforms like Robinhood or
Fidelity have enabled easier access to capital markets for retail traders (Martin and
Wigglesworth 2021). Low interest rates, high volatility creating potential for large gains
in a short time and ”lock-down boredom”, are all factors fueling an increase in retail
investor activity (Shalchi 2021). As retail investor activity has been increasing, we have
witnessed several cases of unusual activity in so-called ”meme stocks”. For example,
in 2020 Tesla rose 700% and Gamestop rose from $18.84 at the end of 2020 to $325
on Jan. 29 2021 to then decline to $90 within the first two trading days in February
(Lazzara 2021).1 According to Gobler (2021) meme stocks often become overvalued and
see drastic price increases in a short amount of time. The price behaviour witnessed in
1A meme stock is a stock that sees an increase in volume due to the hype on social media and online
forums like Reddit rather than due to the company’s performance. They are generally popular among
retail traders.
1
1. INTRODUCTION 2
these stocks aligns with previous findings on retail trading from Barber & Odean (2000),
Barber, Odean and Zhu (2008) and Han and Kumar (2013) who show retail trading to
be speculative by nature and that stocks with high retail proportion are more likely to
have ”lottery features”, attracting retail traders with a propensity for gambling. This
behaviour can be seen frequently on online forums such as Wall-Street Bets, where users
post their speculative investments while encouraging others to do so as well through
”YOLOs”. YOLO stands for you only live once and in this case signals gambling big by
for example going all in on one stock. (Keshner 2021)
Some literature also claims that retail trading is contrarian by nature (Boehmer et al.
2020) and that it leads to retail investors unwillingly providing liquidity and therefore
losing money (Kaniel, Saar, and Titman 1999). However, they also find that retail
activity can accurately predict future returns, further validated by findings from both
Kelley and Tetlock (2013), and Boehmer, Jones, Zhang and Zhang (2020)2. BJZZ also
find evidence that stocks with a high retail order imbalance 3 outperform stocks with
negative imbalances over the following week. Moreover, they find suggestive evidence
that retail marketable orders contain firm-level information that is not yet included in
the stock price, implying that retail trades are informed or that our definition of market
efficiency is incorrect.
Contradictions in previous literature coupled with recent developments on the market
in for example meme stocks makes retail investor activity an important and interesting
subject to investigate with many unanswered questions.
The research area on understanding of how retail traders interact with and cause events
on the market is of interest from a range of standpoints and for multiple applications.
From an economic sustainability standpoint, it provides a basis for decisions on whether
retail investors need to be further regulated for their own protection or, on the contrary,
if a democratization of the capital markets is purely positive as it can contribute to
2Henceforth referenced as BJZZ
3A stock ending a given day with retail buys exceeding retail sells.
1. INTRODUCTION 3
increased equality. Market participants may be interested in finding out if patterns in
retail trading can help predict market movements and if this should be accounted for in
their financial models.
To contribute to this area of research, this study aims to investigate whether the pre-
dictability of returns through retail order imbalances differs between large caps and meme
stocks. Furthermore, it will look at how the effect differs for periods in connection with
quarterly reports compared to non quarterly report periods as well for the period during
the initial impact of Covid-19 compared to non-Covid-19 periods. Thus, the following
hypotheses will be investigated:
1. Retail order imbalances can predict future returns.
2. When decomposing order imbalance into the three components OIB contrarian,
OIB persistence and Other, these will all contribute to the prediction of returns
but to a varying degree.
3. The predictability of returns will differ when looking at:
(a) S&P500 large caps compared to meme stocks
(b) Trading days in connection with reports compared to non-report trading days
(c) Trading days in connection with the initial impact of Covid-19 compared to
other trading days
Previous research both on order imbalances in general (Chordia, Roll, and Avanidhar
Subrahmanyam 2000) and retail order imbalances (Boehmer et al. 2020) & (Farrell et al.
2020) indicates that retail order imbalances can be used to predict future returns and
thus we hypothesize that this will be the case in our paper as well. (Boehmer et al. 2020)
find evidence that both retail order imbalance persistence and retail order imbalance
other contribute to the predictive power of order imbalances while not being able confirm
whether contrarian trading contributes significantly. We believe that we will find similar
results when decomposing the order imbalance into the three components.
1. INTRODUCTION 4
Further, we expect differing results when accounting for the three given circumstances
above. We expect that the patterns of retail trading and the ensuing predictability of
future returns will be significantly different for meme stocks when compared to S&P500
large caps due to the gambling like features features they possess. We expect different
results in connection with reports as previous research finds that the informativeness
of retail order flow increases in connection with the release of new information (Farrell
et al. 2020). Finally, we hypothesize that the initial impact of Covid-19 and the ensuing
fear and panic will have shifted the patterns of retail traders to such a degree that the
predictability of returns will also have become significantly different.
Our thesis uses retail order imbalances to look at the nature of retail investor activity
and how it can be used to predict future returns. Our findings suggest that retail order
imbalances can be used to predict future returns, although this differs when accounting
for particular circumstances. Specifically, when we look at S&P500 large caps compared
to meme stocks, we find that the large caps exhibit similar characteristics as seen in
previous research while the meme stocks predict returns a week ahead negatively. This
is interesting as it indicates that when retail traders buy meme stocks, they are generally
losing money at a one week-ahead period. These findings could support claims of a need
for better financial literacy for retail investors. Though we do not rule it out, we find
no significant evidence supporting that either Covid-19 or periods in connection with
quarterly reports affect the informativeness of retail order imbalances on future returns
any differently than ”normal” periods.
The paper is structured as follows. We look at previous literature related to our subject
in section 2. In section 3 we specify our method of data collection, our retail order
imbalance identification and data analysis. In section 4 we show and analyse our results.
In 5 we discuss our results further while comparing them to previous research and discuss
what they indicate. We conclude the paper in section 6 and in section A we include an
appendix of additional regressions and results.
2Literature Review
2.1 Trading Strategies
Contrarian and momentum trading are two common strategies employed by investors,
founded on the premise of large under- or overreactions to news or research due to behav-
ioral biases (Daniel, Hirshleifer, and Avindhar Subrahmanyam 2002; Barberis, Shleifer,
and Vishny 1999).
Contrarian trading can be explained by the overreaction hypothesis; the hypothesis states
that investors tend to react excessively to information and cause disproportionate stock
movements. The contrarian trader tries to capture the reversed movement as the over-
reaction is corrected. The profitability of contrarian trading strategies has been subject
to debate among academics; (Lehmann 1990) finds that in the short term, contrarian
strategies yield significant positive returns even when correcting for transaction costs
while Conrad, Gultekin, and Kaul (1997) contest this and claim that the positive returns
are rather caused by bid-ask bounce in transaction prices. When accounting for bid-ask
bounce, they find that all profits from price reversals are eliminated for NASDAQ-NMS
stocks and most of the profits for NYSE/AMEX stocks are eliminated.
Momentum trading on the other hand, can be explained by the under-reaction hypothesis
5
2. LITERATURE REVIEW 6
meaning that investors react to information to a lesser degree than they should, leading
to prices not adjusting enough. The momentum trader tries to capture the continuous
positive or negative effect, the momentum, following news announcements or the releases
of research. It is widely agreed upon within the literature that momentum trading is a
profitable trading method but the reason for this is more up for debate (Galariotis 2013).
2.2 Retrieving information from trade data
A common issue when working with trade data is identifying the origin of a trade as
this can’t be retrieved directly from the data. Therefore, researchers looking at patterns
of a specific group of traders need to use estimations or proxies for these instead. For
example when identifying retail order flow, researchers commonly use the trade size as a
proxy. However, with the increased prevalence of algorithmic trading and with institu-
tional investors, to a larger degree, placing more frequent but smaller orders, this proxy
becomes ineffective. (BJZZ) In order to counter these issues and to get a more accurate
measurement of retail order flow, BJZZ propose using retail order imbalances. An order
imbalance occurs when the distribution of buy and sell orders is slanted heavily in one
direction. As shown by Chordia, Roll and Subrahmanyam (2000), order imbalances can,
when aggregated on a daily basis, be used as a means to explain future returns. They find
that order imbalances increase following market declines and reverse following a market
increase, indicating that traders are on average contrarian in their trading behaviour.
Further, they find that order imbalances in either direction, excess buy or sell orders,
reduce liquidity.
In order to calculate order imbalances, researchers must first determine whether a trade
is a market buy or sell. Similarly to the origin of the trade, this can not be identified
directly from the trade data. Two of the most common methods to solve this issue are
called the Tick Rule (TR) and the Lee-Ready algorithm (LR) (Chakrabarty, Pascual,
and Shkilko 2015). TR classifies a trade as buyer-initiated when the trade price is above
2. LITERATURE REVIEW 7
the previously traded price and as seller-initiated when the traded price is below the
previously trade price. If the trade price is the same as the previous trade price, the rule
looks for the closest previous price difference from the current trade price and classifies it
based on the same premise. LR uses both quotes and price changes to sign trades.1 Both
TR and LR often require proprietary data as well as being very time consuming for the
researcher.2. A more recent method called Bulk Volume Classification (BVC) by Easley,
Prado, and O’Hara (2012) claims to provide similar accuracy while solving these issues.
Chakrabarty, Pascual, and Shkilko (2015) test the validity of these claims by comparing
the accuracy of BVC with that of both TR and LR.
In doing so, they find that both TR and LR are significantly more accurate in trade clas-
sification for equities than BVC for most of the common research issues. For identifying
order imbalances, accuracy using TR/LR ranges from 52% to 92.3% while BVC ranges
from 13.9% to 76.6%.
2.3 Retail Investor Activity
Boehmer et al. (2020) propose a method for identifying trades initiated by retail investors
and their direction using US transaction data. In this method, they take advantage of the
fact that most marketable retail order flows in U.S. equity markets are internalized or sold
to wholesalers, and are per SEC 612 regulation subjected to price improvement, making
them easily identifiable. This price improvement is observable as a small fraction on the
penny of a share. Institutional orders almost never receive this kind of price improvement.
Thus it becomes possible to use sub-penny trade prices to identify marketable retail order
flows.
The trading direction is similarly easily identified. Purchases can be distinguished by
1Our algorithm uses the quote rule to sign trades that occur anywhere other than the midpoint, and
the TR to sign trades that occur at the midpoint. The quote rule is applied as follows: if a trade occurs
above (below) the midpoint, it is classified as buyer-(seller-) initiated.
2This paper has access to a limited set of proprietary data and can thus only use this trade signing
algorithm for 2020-01-02 to 2021-04-27
2. LITERATURE REVIEW 8
having prices below a round penny, while sales have prices just above a round penny.
By cross-validation with NASDAQ TRF DATA, BJZZ show that the sub-penny approach
matches the NASDAQ TRF’s correct buy/sell sign 98.2% of the time. The standard Lee
and Ready (1991) trade-signing algorithm has a similar, but lower, accuracy rate of
96.7%.3
BJZZ examine the return predictability using marketable retail order imbalances by a
decomposition the measure to represent price pressure, liquidity provision and, informed
trading. BJZZ find that all three dimensions of OI are highly significant and that mar-
ketable retail order flows are contrarian for horizons ranging between one week and six
months. The control variables indicate that investors tend to buy more aggressively in
relatively larger firms, growth firms, and firms with higher turnover, as well as firms with
higher volatility. They also find that if retail investors buy more than they sell in a given
week, the return on that stock in the following week is significantly higher than days with
similar buy and sell volumes.
Finally, their decomposition shows that approximately half of the predictive power of
the marketable retail order imbalance comes from the persistence of the order imbal-
ance measures. Majority of the remainder comes from the residual component after they
eliminate order persistence and the contrarian trading pattern. As this residual compo-
nent turns out to predict future stock returns, it is consistent with the hypothesis that
marketable retail investor trading contains valuable information about future stock price
movements.
Farrell et al. (2020) further the research conducted by BJZZ by investigating the effect
of social media and specifically that of SeekingAlpha on retail investor activity.
Their results indicate that there is a causal relation between Seeking Alpha articles and
3These accuracy rates are true for stocks with a share price below $100.
2. LITERATURE REVIEW 9
retail investor trading and that the articles have a strong influence on both the intensity
and direction of retail trading. According to their results, retail trading in the half-hour
after the publication of a Seeking Alpha article is 7.73% higher than in the half-hour
before.
Furthermore, they also investigate whether the ability to predict future returns based on
retail order imbalances differs on days with Seeking Alpha articles, media articles, or firm
news compared to normal days without such activity. Their findings align with those
of BJZZ in that retail order imbalances can be informative about returns a week ahead.
However, they also find that the informativeness doubles on days with Seeking Alpha
research.
3Method
3.1 Data collection and processing
We randomly select 11 stocks that are included in the S&P500 as well as GameStop,
AMC Entertainment and The Eastman Kodak Company to perform our analysis on.1
We limit the number of firms due to time constraints.2 For these stocks we extract
quarterly data on market capitalisation, EBITDA, monthly turnover, operating income
before depreciation, total outstanding shares, and earning announcement dates for the
years 2016-2021 from Compustat.
We use data retrieved from Finnhub.io that includes the national best bid offers of the
individual stocks as well as transaction data. Each set of firm data is retrieved at a mil-
lisecond interval using an API, Application Programming interface, which is a software
intermediary that allows two applications to talk to each other and exchange data. How-
ever the data retrieval rate is limited, leading to a smaller selection of stocks is compared
to that of previous research.
We use the method developed by BJZZ to identify trades initiated by retail investors. The
1The randomly selected stocks where extracted from the population of S&P500 stocks using the
RAND function in excel.
2Each ticker takes approximately 12 hours to retrieve, process and manually prepare for Stata
10
3. METHOD 11
method uses the fact that retail order flow can receive minor price improvements down
to a cent per share, contrary to institutional order flow. Most of the orders that execute
with price-improvements down to a cent a share take place off-exchange and are then
reported to a Trade Reporting Facility (TRF)3. Data from a group of wholesale brokers
in January 2020 shows that 85% of all market orders are price-improved (Mackintosh
2020). With access to this data, it is possible to identify both retail sales and buys.
Retail market sales are assumed where the transaction price is slightly above the round
penny, while retail buys are assumed by the transaction price being slightly below the
round penny.
In the initial step of the processing retail trades are identified in three steps: (i) Trades
executed through FINRA which are labeled as “D” in the tick data are extracted. (ii)
We utilize the sub-penny rule described in the previous section to identify retail trades.
We apply the following condition to each price datapoint:
[Pricei,τ × 100] mod(1)6=0, (1)
where i denotes firm and τ denotes timestamp. If this condition is fulfilled, a trade is
categorized as a retail trade. (iii) We identify whether the retail trade is to be identified
as a buy or a sell on the trades that are assumed to be retail initiated
If the following condition is fulfilled, the trade is identified as aggressive buy:
[Pricei,τ × 100] mod(1) > 0.5 (2)
Similarly, if the following condition is fulfilled, the trade is identified as an aggressive sell:
[Pricei,τ × 100] mod(1) < 0.5 (3)
3This is because orders executed by wholesalers or through internalisation must be publicly reported;
they are usually reported to a FINRA Trade Reporting Facility (TRF), which provides broker-dealers
with a mechanism through which to report transactions that take place off-exchange. These TRF execu-
tions subsequently included in the TAQ ”consolidated tape” of all reported transactions with exchange
code ”D”
3. METHOD 12
The result of this process is a dataset with marketable retail trades and their correspond-
ing trade direction. This then enables us to extract firm specific metrics from this data
as well. We determine the signed volume of individual trades in the following manner:
Pricei,τ × ShareV olumei,τ × TradeDirectioni,τ = SignedV olume (4)
Since our method relies on the measurements of order imbalance shown in the next section
we also summarize each day’s volume, irrespective of the trade direction. This produces
the individual stock order imbalance on the given day.
Institutional signed volume is calculated in the same manner, but uses the LR trade
signing algorithm to determine trade direction (BJZZ).
Next, we extract the realized intraday volatility. Firstly we re-sample the data at a
frequency of 5 minutes in order to eliminate the effect of bid-ask bounce, which is a
situation where the price bounces within the very limited range between the bid price
and ask price but no real movement in price occur. We then proceed to evaluate the log
returns.
log(return)i,τ = log(pricei,τ )− log(pricei,τ+1) (5)
Next each data point of difference is squared and the data is summarized as following
equation shows:
RealizedV olatilityi, t =
N∑
i=1
√
log(return)2i,τ (6)
Where i denotes firm, t denotes day and τ denotes intraday. This produces the realized
volatility of the stock for the given day. Similarly to BJZZ, with access to this data, we
can now compute and measure the daily retail order imbalances, and thus retail investors’
3. METHOD 13
directional trades using:
RetailOIBvoli,t =
RetailBvol(i, t)−RetailSvol
RetailBvol(i, t) +RetailSvol(i, t)
(7)
where Retail denotes marketable retail trade, OIB denotes order imbalance, vol stands
for share volume, B and S denote buy and sell respectively.
3.2 The model
We examine the informativeness of retail order imbalances at the one week, two weeks
and 4 weeks in the future by estimating:
Yi(w+1,w+2,w+4) = α + β1Retail OIBi,w + β3Retail OIBi,w × Log(Size)i,q
+ β4Inst OIBi,w + β5Inst OIBi,w × Log(Size)i,q
+ β7meme×Retail OIBi,w + β8report 5×Retail OIBi,w
+ β9covid×Retail OIBi,w + β10Chari,t +Monthi + Firmi + i,t,
(8)
where Yi(w+1,w+2,w+4) is stock i’s return from the close of day compounded to the close
of day compounded into weekly observations in three intervals one, two, and, four weeks
in the future and q denotes quarterly observations. Retail OIBi,w is the daily retail
buy volume minus daily retail sell volume scaled by daily retail trading volume and
compounded to weekly observations. We include Retail OIB × Size as an interaction
term between profitability of retail trading and firm size. Inst OIBi,w is the total non-
retail buy volume minus total non-retail sell volume, scaled by total non-retail trading
volume and Inst OIBi,w × Log(Size)i,q is an interaction term between firm size and
institutional order imbalance. Chari,q is a vector of firm characteristics and includes past
3. METHOD 14
returns estimated over the prior week, market capitalization, monthly turnover, volatility
of daily returns, and operating income before depreciation. report 5 × Retail OIBi,w is
an interaction term between a dummy variable for 5 days around reporting and retail
order imbalance. covid×Retail OIBi,w is an interaction term between a dummy variable
for the Covid-19 for the effect on March and April of 2020, i.e the months where most
governments enforced lockdowns (Sandford 2020) and retail order imbalance. Finally,
meme×Retail OIBi,w is an interaction term between a dummy variable for meme stocks
and retail order imbalance.
We use returns that are similar to BJZZ’s ”CRSP returns” as they are computed based
on close price, that is latest close in trade. This can trigger an upward bias due to bid-ask
bounce according to Blume and Stambaugh (1983). For this reason BJZZ report most
of their results and returns in the average bid-ask price at the end of the day. When
comparing BJZZ’s results for CRSP returns and closing bid-ask price, we note that the
differences are marginal. Furthermore, we mainly compute our results on S&P500 stocks,
meaning that the bid-ask spread should be relatively small and therefore the bid-ask
bounce should only have a marginal effect on our results. However, for the meme-stocks
we admit that there may be a larger spread and that it has a larger significance. For these
reasons, we opt to use the latest closing price for our returns but in case of additional
time at hand we would include this in our report as well.
We then investigate whether the correlation between retail imbalances and future returns
depends on price pressure, liquidity provision or informed trading. We use the daily return
decomposition of BJZZ; we decompose the order imbalance into three components: OIB
persistence representing price pressure, OIB contrarian to represent liquidity provision
and OIB Other to represent informed trading.4 The three components are estimated
using the following panel regression:
Retail OIBw = α + β1Retail OIBi,w−1 + β2Reti,w−1 + i,t, (9)
4This is not included in our regression
3. METHOD 15
where OIB persistence is β1Retail OIBi,w−1, OIB contrarian is β2Reti,w−1and OIB other
is ˆi,t.
Through this decomposition we can differentiate between the different effects’ contribu-
tion towards the predictive power. This is done through a regression with return as the
dependent variable and the decomposed effects as independent variables along with the
residual from the first stage regression as an independent variable.
Returnw = α + β1Retail OIBi,w−1 + β2Reti,w−1 + µ̂i, t+ i, t (10)
In our full-scale model above, we utilize the Lee-Ready trade signing algorithm on data
on national best bid and ask prices to also be able to determine the trade direction of
institutional trade direction and the resulting institutional order imbalance. However, we
rely on finnhub.io for our data collection and they currently only have national best bid
and ask price data dating back to 2020-01-02. Due to limited time and literature showing
signs of low accuracy in using Bulk-Value Classification for calculating order imbalances
(Chakrabarty, Pascual, and Shkilko 2015), we instead opt to use a simplified version of
our model excluding institutional order imbalance for the full period between 2016-05-27
and 2021-04-27.
3.3 The simplified model
Our simplified model is used in the same way as the full-scale model, with the exception
that we omit Inst OIBi,w and Inst OIB × Log(Size)i,q. As the model is simplified, the
results may carry omitted variable bias. However we use it as a robustness check to our
full scale model to control for a longer time period.
3. METHOD 16
Yi(w+1,w+2,w+4) = α + β1Retail OIBi,w + β3Retail OIBi,w × Log(Size)i,q
+ β4meme×Retail OIBi,w + β5report 5×Retail OIBi,w
+ β6covid×Retail OIBi,w + β7Chari,t +Monthi + Firmi + i,t,
(11)
The simplified model is applied on the longer time scope and we then investigate whether
the correlation between retail imbalances and future returns depends on price pressure,
liquidity provision or informed trading in the same manner as for the full-scale model.
4Results and Analysis
4.1 Summary Statistics
Our sample, with summary statistics shown in Table 4.1 through 4.5, ranges back from
25-05-2016 up until 25-03-2021 with statistics computed on 358 million unique trades. We
consider 11 randomly selected S&P500 stocks and an additional two manually selected
meme stocks: GameStop and AMC Entertainment.
The stocks in our sample have a mean market value of USD 44.9 billion and a mean
quarterly turnover of USD 10,6 billion. The mean market size and quarterly turnover
together with the fact that we only use S&P500 constituents, means that our sample is
tilted towards towards larger firms. As such we see lower spreads due to higher trade
volumes (Palmer 2021), which can also be seen in Table 4.1 as the mean realized spread
is zero. Larger firms will also mean that retail order flow is slightly more persistent
according to previous literature (Boehmer et al. 2020). Moreover, in Table 4.5 we can see
that there is a notable general uptrend in the proportion of trade volume being executed
by retail traders. This is especially true for two of the smaller meme stocks, i.e, AMC,
GME.
The mean weekly aggregated Retail OIB ratio is 1.9 through 2016-2021 and 2.1 through
17
4. RESULTS AND ANALYSIS 18
2020 to mid 2021 indicating that, within our dataset, retail traders have been net buy-
ers. The mean weekly aggregated Institutional OIB ratio is -0.012 for the period 2020
to mid 2021, indicating that institutional traders on average buy and sell the stocks ap-
proximately equally much in our sample. When grouping the meme stocks we find a
significantly lower weekly order imbalance in all instances, 1.07 for the period 2016-2021,
1.7 for 2020 through mid 2021. We also see a lower mean institutional order imbalance
ratio of -0.003 indicating that on average institutional investors on average buy and sell
GME, AMC Entertainment approximately equally during the period 2020 to mid 2021.
The mean one week return is 0.4% which is 23.4% aggregated for a full year or 186%
for the full sample period while the mean 5 year return for the sample period is 127%.
These can be compared to the 10% average return of the S&P500 since its inception or
the 101% return of S&P500 between May 2016 and April 2021 (representing a 15.2%
annualized return). This can be interpreted as our sample being relatively representative
of the returns achieved by the S&P500 during the sample period.
Table 4.1: Summary statistics for S&P500 stocks, 2020-2021
Sum Mean SD Min Max N
Returnw 21.15 0.006 0.085 -0.71 1.35 3,564.00
Retail OIBw 7,553.30 2.106 0.823 -1.14 4.40 3,586.00
Inst OIBw -44.68 -0.012 0.828 -3.41 2.94 3,586.00
RealizedVold 55.67 0.015 0.012 0.00 0.20 3,630.00
RealizedSpread 0.13 0.000 0.000 -0.01 0.00 3,630.00
VolShareOfRetail 563.10 0.155 0.064 0.02 0.55 3,630.00
Retailtrades 7,901,239.00 2,176.650 4,509.589 68.00 50,788.00 3,630.00
Market Value - Total 167,110,694.98 46,036.004 60,211.803 1,120.36 283,508.00 3,630.00
Operating Income Before Depreciation - Quarterly 7,007,823.69 1,930.530 3,912.925 -1,581.00 15,079.00 3,630.00
Sales/Turnover (Net) 38,575,943.73 10,626.982 14,991.408 126.47 46,821.00 3,630.00
Table 4.2: Summary statistics for S&P500 stocks, 2016-2021
Sum Mean SD Min Max N
Returnw 51.89 0.004 0.054 -0.71 1.35 13,529.00
Retail OIBw 25,464.72 1.879 0.921 -1.29 4.77 13,550.00
RealizedVold 146.98 0.011 0.009 0.00 0.20 13,595.00
VolShareOfRetail 2,195.53 0.161 0.075 0.01 0.72 13,595.00
Retailtrades 21,044,769.00 1,547.979 3,356.410 13.00 78,501.00 13,595.00
Market Value - Total 610,225,059.75 44,885.992 63,533.882 537.03 283,508.00 13,595.00
Operating Income Before Depreciation - Quarterly 24,211,587.64 1,959.025 3,927.186 -2,018.00 15,871.00 12,359.00
4. RESULTS AND ANALYSIS 19
Table 4.3: Summary statistics for meme stocks, 2020-2021
Sum Mean SD Min Max N
Returnw 58.97 0.091 0.549 -0.80 7.88 648.00
Retail OIBw 696.33 1.068 0.503 0.05 2.68 652.00
Inst OIBw -1.65 -0.003 0.385 -1.10 2.12 652.00
RealizedVold 43.05 0.065 0.064 0.01 0.91 660.00
RealizedSpread 0.31 0.000 0.001 -0.01 0.01 660.00
VolShareOfRetail 204.79 0.310 0.080 0.07 0.68 660.00
Retailtrades 17,129,881.00 25,954.365 66,424.659 258.00 644,976.00 660.00
Market Value - Total 828,034.46 1,254.598 3,887.877 246.91 21,222.50 660.00
Operating Income Before Depreciation - Quarterly -47,592.80 -72.110 198.211 -355.00 244.90 660.00
Table 4.4: Summary statistics for meme stocks, 2016-2021
Sum Mean SD Min Max N
Returnw 48.20 0.020 0.290 -0.80 7.88 2,460.00
Retail OIBw 4,221.49 1.715 1.018 -0.97 4.75 2,462.00
RealizedVold 74.68 0.030 0.040 0.00 0.91 2,472.00
VolShareOfRetail 559.06 0.226 0.095 0.01 0.68 2,472.00
Retailtrades 18,321,634.00 7,411.664 36,092.413 7.00 644,976.00 2,472.00
Market Value - Total 4,582,106.33 1,853.603 2,173.999 246.91 21,222.50 2,472.00
Operating Income Before Depreciation - Quarterly 210,947.26 85.335 155.174 -355.00 342.70 2,472.00
Table 4.5: Mean retail volume share of trades in percentages
2016 2017 2018 2019 2020 2021 Total
AFL 14.5 12.8 13.1 12.0 13.7 10.3 12.9
BBY 17.4 13.0 16.9 17.3 16.2 12.5 15.8
CI 18.7 12.7 14.4 22.5 14.1 9.7 15.9
CTSH 12.8 15.1 14.7 16.3 13.7 15.8 14.7
CZR 20.0 24.7 23.2 18.4 16.7 17.6 20.5
ECL 12.1 15.6 18.6 13.7 12.5 11.1 14.5
ED 15.7 11.5 13.9 12.3 15.9 19.0 14.0
HES 10.1 11.6 12.3 14.6 14.2 15.6 13.0
OMC 12.2 14.3 18.8 18.4 14.2 17.1 16.0
PTC 15.0 12.7 14.4 16.6 14.8 18.7 14.9
T 23.3 25.2 27.2 27.0 24.7 23.3 25.5
AMC 14.1 20.2 23.7 24.3 34.1 34.3 24.7
GME 15.1 16.6 17.5 20.9 28.2 27.0 20.5
All 15.5 15.8 17.6 18.0 17.9 17.8 17.1
4. RESULTS AND ANALYSIS 20
4.2 Results
Our main variables of interest are marketable retail order imbalance and return on indi-
vidual stocks, denoted Retail OIBw and Returnw+t respectively in our results. In our
regressions we use weekly data on both returns and order imbalance.
We start out by looking at the results from the sample of 11 S&P500 stocks with our full
model regressing over the shorter time period followed by the results from our simplified
model on the longer time period. We then repeat this process but for the sample of meme
stocks.
We run our simplified model on the full time period but as we also have access to addi-
tional data allowing us to use our full scale model for the time period 2020-2021, we run
these regressions for this period separately as a form of robustness check.
We note that the return and marketable retail order imbalance residual is likely to be
correlated within firms, so we cluster standard errors by firm in all regressions. Both time
periods have been subjected to month and firm controls individually and are presented
in the appendix Tables A.2 through A.5.
4.3 Results from the full model on short period
We find no statistical significance when regressing without the interaction between order
imbalance and firm size. However, when the interaction term is included we find a slightly
positive coefficient of 0.028 for retail order imbalance when predicting returns one week
ahead. An interpretation of this is that we cannot predict future returns in large firms.
This finding is in line with the findings of BJZZ and Farells et Al, for the one week period
they also find a slight positive coefficient. We use the same calculations for our proxy of
order imbalance, based on share volume, however the returns are calculated as the com-
pounded 5 day difference between close price. A method of calculation, shown to produce
4. RESULTS AND ANALYSIS 21
slightly lower coefficients, albeit with stronger predictive power in BJZZs, findings. Our
findings suggest that if retail investors buy more than they sell (in aggregate) of a stock
in a given week, the returns tend to increase, in effect, retail investors tend to sell losers
and buy winners in the short horizon of one week.
Table 4.6: Predicting future return short period 2020-2021 - full model
Returnw+1 Returnw+1 Returnw+1 Returnw+1 Returnw+2 Returnw+4
Retail OIBw −0.024 −0.019 −0.001 0.028
∗∗∗ −0.004 −0.020
(0.015) (0.014) (0.002) (0.008) (0.020) (0.022)
Inst OIBw 0.010
∗ 0.010∗ 0.010 0.127 0.009 −0.053
(0.006) (0.005) (0.006) (0.102) (0.085) (0.064)
Returnw −0.035 −0.027 −0.027 −0.034 0.058
(0.028) (0.041) (0.041) (0.024) (0.063)
OIBDqq −0.000 −0.000
∗∗ −0.000∗ −0.000∗ −0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Sales/Turnoverq 0.000 0.000 0.000
∗∗∗ 0.000 −0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
RealizedVol 0.553 −0.090 −0.136 0.145 0.631∗∗
(0.319) (0.449) (0.466) (0.240) (0.244)
report 5=1 −0.004 −0.003 −0.032 −0.042
(0.052) (0.051) (0.044) (0.047)
meme=1 0.268∗∗∗ 0.279∗∗∗ 0.215∗∗∗ 0.132
(0.056) (0.061) (0.044) (0.120)
report 5=1 × Retail OIBw −0.003 −0.003 0.013 0.026
(0.021) (0.021) (0.019) (0.022)
meme=1 × Retail OIBw −0.167
∗∗∗ −0.184∗∗∗ −0.131∗∗∗ −0.072
(0.015) (0.020) (0.023) (0.080)
Retail OIBw X logSize −0.003
∗∗∗ −0.000 0.001
(0.001) (0.002) (0.002)
Inst OIBw X logSize −0.011 −0.001 0.005
(0.010) (0.008) (0.006)
Constant 0.066∗ 0.047 0.012 0.010 0.021 0.010
(0.037) (0.038) (0.010) (0.009) (0.015) (0.014)
Observations 4212 4147 4147 4147 4082 3952
Standard errors in parentheses
Dependent variables: Future return, in week 1,2 and,4.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
For the control variables, we observe negative coefficients on the previous week’s re-
turn, which indicates weekly return reversals, size, operating income before depreciation,
turnover, and realized volatility all carry the expected signs, and most are not statisti-
cally significant. This result also confirms that the predictability we find is not due to
size, turnover, or volatility. The average adjusted R2’s from the first stage cross-sectional
estimation are mostly around 4.4%, apart from the first regression without control vari-
ables which has an R2 of 0.18%. This shows that the amount of variance the realized
volatility and previous 5 day return and other controls impact the model.
4. RESULTS AND ANALYSIS 22
4.4 Decomposition results for short period
In table 4.7 we find that the coefficient is significant at an alpha of 0.1 on our proxy
for persistence, i.e marketable retail order imbalance with a five day lag. The positive
coefficient of 0.73 on persistence is in line with previous research by BJZZ and Farell et al.
and shows a high degree of persistence on the weekly level. Contrarian trading is found
to be significant, at an alpha of 0.01 which is in line with previous research. In the second
Table 4.7: Decomposition short period 2020-2021
Retail OIBw Returnw
Retail OIBw−1 0.730∗∗∗ −0.027∗
(0.031) (0.014)
Returnw−1 −0.125∗∗∗ 0.006
(0.033) (0.018)
µ̂i, t −0.009
(0.011)
Constant 0.526∗∗∗ 0.071∗
(0.086) (0.036)
Observations 4160 4147
Standard errors in parentheses
Dependent variables: Retail order imbalance and return.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
step we do not find we find suggestive evidence that persistence is explanatory of return
in the short period. We do not find the proxy for contrarian trading to be significant,
which is in line with previous research by BJZZ. Contrary to previous research we do not
find the residual, i.e the proxy for informational trading to be significant.
4.5 Results from the simplified model on long period
Similarly to the shorter period we do not find statistical significance that retail order
imbalance is a predictor of future return without controlling for firm size. The longer
time period, reported in table 4.8, supports the findings of the full featured model, with
a slightly positive predictive coefficient of 0.008 on marketable retail order imbalance, at
4. RESULTS AND ANALYSIS 23
an alpha of 0.01.
This alludes to the importance of the features in the full model and this simplified albeit
longer time period model not being able to find any coefficients to be statistically signif-
icant. It could also be the the fact the time periods are very different, as is evident when
looking at Figure 4.1 and 4.2.
In the longer period we find that realized volatility is a highly significant contributor
to future return. This finding is different to that of the shorter period where realized
volatility was shown to be largely insignificant apart from future returns four weeks
ahead. This suggests that the risk in stocks from our sample becomes less important in
explaining the future return.
Table 4.8: Predicting future return long period 2016-2021 - full model
Returnw+1 Returnw+1 Returnw+1 Returnw+1 Returnw+2 Returnw+4
Retail OIBw −0.004 −0.004 0.001 0.008
∗∗∗ 0.004 0.001
(0.004) (0.003) (0.001) (0.003) (0.005) (0.008)
Returnw −0.034
∗ −0.033 −0.033 −0.036 0.050
(0.018) (0.020) (0.020) (0.029) (0.058)
OIBDqq −0.000
∗ −0.000 −0.000∗ −0.000∗ −0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Sales/Turnoverq 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
RealizedVol 0.688∗∗∗ 0.566∗∗∗ 0.553∗∗∗ 0.674∗ 0.956∗∗∗
(0.210) (0.181) (0.177) (0.340) (0.164)
report 5=1 −0.004 −0.004 −0.012 −0.016
(0.017) (0.017) (0.013) (0.016)
meme=1 0.044∗∗ 0.043∗ 0.031∗ 0.013
(0.020) (0.020) (0.016) (0.022)
report 5=1 × Retail OIBw −0.003 −0.003 0.004 0.008
(0.007) (0.007) (0.006) (0.007)
meme=1 × Retail OIBw −0.023
∗∗ −0.024∗∗ −0.018∗ −0.010
(0.010) (0.010) (0.009) (0.010)
Retail OIBw X logSize −0.001
∗∗ −0.001 −0.000
(0.000) (0.001) (0.001)
Constant 0.014 0.005 −0.003 −0.002 0.000 −0.003
(0.009) (0.007) (0.002) (0.002) (0.004) (0.002)
Observations 15989 14699 14699 14699 14638 14518
Standard errors in parentheses
Dependent variables: Future return, in week 1,2 and,4.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
4.6 Decomposition results for long period
Results are reported in table 4.9. Similarly to the shorter time period we find a statistical
significance at an alpha of 0.01 for persistence, with a coefficient of 0.7. The insignificant
4. RESULTS AND ANALYSIS 24
contrarian trading is in line with previous research.
From the second step of the decomposition we find that the persistence, contrarian trading
as well as informational trading is insignificant.
Table 4.9: Decomposition short period 2020-2021
Retail OIBw Returnw
Retail OIBw−1 0.704∗∗∗ −0.005
(0.018) (0.003)
Returnw−1 −0.176∗∗ 0.013
(0.059) (0.020)
µ̂i, t −0.001
(0.003)
Constant 0.551∗∗∗ 0.015∗
(0.044) (0.008)
Observations 15937 15924
Standard errors in parentheses
Dependent variables: Retail order imbalance and return.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
4.7 Differences between the models and time sam-
ples
When comparing the decomposition findings in the two data sets, we notice that persis-
tence is statistically significant at an alpha of 0.01 in both regressions, though we find a
higher coefficient in the period 2020-2021.
Disregarding the predictive power, it is interesting that the coefficient for marketable
retail order imbalances increases in magnitude in the period 2020 to mid 2021, compared
to the period 2016-2021. As shown by Figure 4.1 and 4.2 the nature of the samples
looks very different in 2020-2021 compared to 2016-2021 while marketable retail order
flow remains at a similar level.
4. RESULTS AND ANALYSIS 25
Figure 4.1: Weekly return and retail order imbalance for S&P500 sample over the period
mid 2016 to mid 2021
−
.
5
0
.
5
1
1.5
retu
rn_
w
2016 2017 2018 2019 2020 2021
−
2
0
2
4
6
Re
tail
_O
IB_
w
2016 2017 2018 2019 2020 2021
Figure 4.2: Weekly return and retail order imbalance for meme stock sample over the
period mid 2016 to mid 2021
0
2
4
6
8
retu
rn_
w
2016 2017 2018 2019 2020 2021
−
2
0
2
4
6
Re
tail
_O
IB_
w
2016 2017 2018 2019 2020 2021
4.8 Reporting days
We report our findings for reporting dates throughout all time periods and sub-groups of
all 11 stocks and only meme stocks. We do not find any significance for 5-days around
reporting, which might not be surprising considering the size of the data becoming so
much smaller when only accounting for days in connection with reports. The insignificant
coefficients all share the same signs as the overall marketable retail order imbalance
coefficients, though they are, in all cases, close to zero. On this basis, we cannot reject
the null hypothesis in this case. Though it could be the case that there is no contribution,
it seems unlikely since there is an overall correlation between return and marketable retail
order flow in previous literature as well as in most of our findings.
4. RESULTS AND ANALYSIS 26
4.9 The meme stocks
When regressing on only meme stocks in the shorter period we find that on average retail
investor order imbalance tend to negatively correlate with returns one and two weeks
ahead, with a coefficient of -0.18 and -0.13, as seen in table 4.6.
When regressing on the full period with the meme stocks which can be seen in table
4.8, we also find negative coefficients for all time horizons although none of them are
statistically significant. For the full period however, the coefficients are lower than for
the shorter time period. For example, coefficient for one week returns is -0.023 in the
longer period compared -0.18 for the shorter period. This indicates that the recent surge
in interest in these specific stocks may be the cause for our findings.
We also note that the returns for meme stocks are much higher than for the rest of
the sample. Despite this, we find that when retail investors are buying meme stocks to a
higher degree, the following weekly returns are negative. This implies that retail investors
are generally losing money in these stocks.
4.10 Covid effect
When looking at the shorter time period, we note differing results compared to those
observed for the longer time period. A natural explanation for this is Covid-19 starting
to affect the financial markets in the beginning of 2020. To see if there is a significant
difference in the retail order flow and the predictability of returns specifically around the
outbreak of Covid-19 and the resulting intense lockdowns, we introduce a Covid-dummy
variable for the months of March and April in 2020 when lockdowns were at the highest
level. As can be seen in table 4.10, we do not find statistical evidence that there is a
difference due to the initial lockdowns. We cannot reject the null hypothesis that the
initial outbreak of Covid-19 has an effect on retail order flow and the predictability of
returns. Possibly due the fact that the initial market decline in March was regained over
4. RESULTS AND ANALYSIS 27
April. However, despite not being able to say with certainty that the initial impact of
Covid-19 is a root of our differing results, our results in general point towards Covid-19
having a notable impact.
Table 4.10: Predicting future return short period 2020-2021 - full model
Returnw+1 Returnw+1 Returnw+1 Returnw+1 Returnw+2 Returnw+4
Retail OIBw −0.024 −0.019 −0.002 0.024
∗∗∗ −0.006 −0.020
(0.015) (0.014) (0.004) (0.007) (0.021) (0.022)
Inst OIBw 0.010
∗ 0.010∗ 0.010∗ 0.133 0.015 −0.050
(0.006) (0.005) (0.005) (0.103) (0.079) (0.065)
Returnw −0.035 −0.033 −0.033 −0.034 0.062
(0.028) (0.044) (0.044) (0.028) (0.059)
OIBDqq −0.000 −0.000
∗∗ −0.000∗ −0.000∗ −0.000∗
(0.000) (0.000) (0.000) (0.000) (0.000)
Sales/Turnover (Net) 0.000 0.000 0.000∗∗∗ 0.000 −0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
RealizedVol 0.553 −0.015 −0.061 0.135 0.561∗∗
(0.319) (0.497) (0.508) (0.278) (0.194)
report 5=1 −0.006 −0.005 −0.034 −0.043
(0.051) (0.051) (0.044) (0.049)
meme=1 0.262∗∗∗ 0.273∗∗∗ 0.215∗∗∗ 0.137
(0.058) (0.062) (0.039) (0.119)
Covid=1 −0.034 −0.042 −0.037 −0.017
(0.031) (0.032) (0.049) (0.032)
report 5=1 × Retail OIBw −0.002 −0.003 0.014 0.027
(0.021) (0.021) (0.019) (0.022)
meme=1 × Retail OIBw −0.164
∗∗∗ −0.181∗∗∗ −0.131∗∗∗ −0.075
(0.015) (0.020) (0.021) (0.080)
Covid=1 × Retail OIBw 0.008 0.012 0.019 0.016
(0.014) (0.015) (0.020) (0.014)
Retail OIBw logSize −0.003
∗∗ −0.000 0.001
(0.001) (0.002) (0.002)
Inst OIBw Size −0.012 −0.002 0.005
(0.010) (0.007) (0.006)
Constant 0.066∗ 0.047 0.014 0.014 0.026 0.015
(0.037) (0.038) (0.012) (0.012) (0.017) (0.016)
Observations 4212 4147 4147 4147 4082 3952
Standard errors in parentheses
Dependent variables: Future return, in week 1,2 and,4.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
5Discussion
5.1 Results compared to previous research
Using a method developed by BJZZ, we identify trades executed by retail traders on a
sample of stocks and use this data to calculate a retail order imbalance measure which tells
us about the nature and direction of retail trades. We further the research of BJZZ, albeit
on a smaller sample size, by running regressions on specific time periods in connection
with quarterly reports and by including a sample of meme stocks to investigate whether
this leads to different findings.
Similarly to the conclusions of BJZZ and other previous literature, our results indicate
that retail order imbalances can be used to predict future returns. Using our full model
on S&P500 stocks between 2020-2021, we find statistically significant results showing
that retail order imbalance correlates positively with one-week returns. However, con-
tradictory to BJZZ’s findings, when we look at longer horizons such as two-week returns
or four-week returns, we find a negative correlation although not statistically significant.
Using our simplified model on 2016-2021, we find similar results as for the shorter period
but with much lower coefficients. Retail order imbalance has a statistically significant
positive correlation with one-week returns and the correlation turns negative during week
28
5. DISCUSSION 29
two and forward.
There are several reasons as to why we have different findings compared to previous
literature. Firstly, it could be due to the sample size; BJZZ look at 3000 stocks for a 6
year time period whereas we look at 12 stocks for a five year time period. This could also
in part explain why many of the results aren’t statistically significant.
Another explanation connected to our sample could be that it is biased towards larger
firms as, disregarding the meme stock sample, we only look at stocks within the S&P500
while BJZZ’s sample of 3000 stocks naturally includes stocks outside of the S&P500 as
well. However, according to BJZZ’s results, larger firms tend to show higher positive
coefficients, which is the opposite of what we observe, thus we can disregard this as an
explanation.
Another reason could be connected to bid-ask bounce, as our method of calculating
returns isn’t the most optimal for eliminating these errors. However, as noted in our
method we believe this should only be a marginal error due to the size of the companies
in our sample and the minimal effect witnessed due to bid-ask bounce in BJZZ’s results.
In contradiction to previous research on which components drive retail order imbalance
and return, we find no statistical evidence that our proxy for informed trading contributes
to return. We do however find that retail order imbalance can be explained by our
proxy for persistence as well as contrarian trading. However, we can only find suggestive
evidence that the persistence is explanatory to future return, and no statistical evidence
that contrarian trading has effect on future return.
For our meme stock sample, our findings on what direction retail order imbalances predict
returns stand in direct contradiction to BJZZ; we find that retail order imbalances have
a negative correlation with returns for all examined horizons with varying degrees of
statistical significance. This goes for both the shorter period where we apply our full
model and for the longer period with the simplified model. Interestingly, the coefficients
are considerably higher for the shorter time period and we specifically find a statistically
5. DISCUSSION 30
significant, strong negative correlation between retail order imbalances and one week
returns. We believe that the high negative correlation could be a direct effect of the
high risk, gambling-like nature, that meme stocks tend to posses. Similarly to what
is witnessed in the S&P500 sample, we see that one-week returns become much more
volatile in 2020 and in 2021 for the meme stock sample. (Figure 4.2) We also note a large
amount of outliers in one-week returns, especially in 2021.
We are unable to find any statistical evidence of retail order imbalances being more or
less explanatory of future returns in connection with the release of quarterly reports.
Therefore, we can not confirm our hypothesis that predictability of returns differs in
connection with quarterly reports. We do however note that it seems unlikely that there
is no real contribution from increased information since previous research suggests that
information is explanatory to retail trading behaviour (Boehmer et al. 2020; Farrell et al.
2020). Thus, we conclude that our sample size is a more likely reason for the insignificant
coefficients.
Finally, we hypothesized that the initial effect of Covid-19 would cause differing results.
We were unable to find statistical evidence of this. However, we find close to zero coeffi-
cients before 2020 and 2021 with low statistical significance and confidence intervals going
above zero. When we instead run regressions solely on 2020-2021 we find larger coeffi-
cients with similar statistical significance. Looking at a comparison of the distribution of
one week returns (Figure 4.1) between different years may in part provide an explanation
for this difference. Between 2016 and 2019 we mostly see one week returns distributed
evenly around the 0%-mark while the returns become much more volatile with a positive
tilt during both 2020 and 2021. With a larger amount of extreme negative or positive
returns within the investigated time horizons, we get an increased effect on our observed
coefficient. While this difference could be interpreted as an effect of Covid-19 we find no
statistical evidence which can attribute this to Covid-19.
5. DISCUSSION 31
5.2 The trend of increased volume of retail trades
A need for better financial literacy?
Table A.1 in appendix shows that realized volatility can explain an increased marketable
retail order imbalance. This could imply that the increased volatility in the market
during Covid-19 is the main explanation for the trend of increase in retail trades as seen
in Figure 4.1. However, when looking at the correlation between realized volatility and
marketable retail order imbalance for meme stocks during 2020 to mid 2021, we find
suggestive evidence, that an increase in volatility has a negative effect on marketable
retail order imbalance which contradicts the previous finding. We interpret this as retail
traders not choosing meme stocks purely on the grounds of them having a high volatility
but rather due to other factors catering to their gambling propensity. Without further
evidence we can only speculate and it is hard to draw any more conclusions, but one
reason for this could be the rise of forums such as Wall-Street Bets and an increased
coverage of meme stocks capturing the eye of more retail traders leading to the increased
activity.
When looking at our summary statistics, it becomes evident that the share of volume
traded by retail investors is increasing significantly. If you also couple this with the neg-
ative correlation between increased marketable retail order imbalance and future returns
within meme stocks between 2020 - to mid 2020, it highlights a further need and interest
in studying this area of research more extensively and with bigger scopes. Previous re-
search on liquidity provision and price pressures find that retail investors generally supply
liquidity, i.e pay prices at premium in transitory effects.(Barrot, Kaniel and, Sraere 2015)
While we do not find this in our current smaller data set it is fair to assume this is the case
in a general setting. We do however find that in especially meme stocks, retail investors
exhibit a contrarian-like trading behaviour where they buy losers and sell winners, which
in part indicates that they are losing money.
Even if several of our results are low in significance individually, the aggregate of the
5. DISCUSSION 32
findings suggest an increased role of retail investors in the financial markets. Given the
increase in retail trades, and the gambling-like features of meme stocks it is becoming
increasingly important to supply information to retail investors. For many people, their
first encounters with the financial markets might come through these meme stocks and
without prior knowledge, misguided decision-making caused by hausse in social media
forums could lead to the loss of large sums of money. However, the increased level of
retail investor activity should also, if coupled with an increased level of financial literacy,
be seen as something positive through an equality perspective as more people start taking
advantage of the opportunities that the financial markets can provide .
6Conclusion
We contribute to the research area surrounding the nature of retail trading activity by
investigating how future returns correlate with retail order imbalances and how this differs
for periods in connection with quarterly reports, for periods following the outbreak of
Covid-19 and for meme stocks. For the period of 2016 to 2021 with a sample of randomly
selected stocks from the S&P500 we find evidence supporting the findings of BJZZ that
retail investors order flows correlate positively with future returns, we do however only
find this to be statistically significant for returns one week in the future. Regarding the
periods of reporting we are unable to find statistical significance likely as a result of the
limited size of our sample. However, without rejecting the null hypothesis, we find that
the data suggests similar results as the longer period due to the sign of the coefficient,
albeit very close to zero. Interestingly, we find that for our group of meme stocks there is a
significant and negative correlation between one-week returns and retail order imbalances.
Although the sample is small, we find suggestive evidence that retail investors lose money
when investing in stocks with gambling characteristics. We investigate the possibility that
there is a correlation between volatility and retail order imbalances within meme stocks
and hence that the investors choose these stocks on the basis of volatility but find that
this isn’t specifically the case. Finally, while our results suggest an effect of Covid-19
in the shorter time frame, we are unable to support this with statistical evidence when
33
6. CONCLUSION 34
looking at the initial lockdown effect.
We suggest further research in stocks with gambling characteristics with larger samples
to find to further evidence of uninformed retail investors losing money in them. An
interesting route connected to meme stocks could be to perform sentiment analysis’ on
forums such as Wall-street Bets and performing a similar analysis as the one done by
Farrell et. Al (2020) on SeekingAlpha research. Although our method for identifying
retail trades relies on a circumstance specific for the US market, we believe an interesting
and important future subject would be to extend the research to other international
markets. In doing so, we would find whether these results and findings are exclusive for
the US market or if the nature of retail trading follows the same pattern irrespective of
geography and market.
In general, we believe that the nature of retail investing is an important subject to inves-
tigate further as it can provide an answer to whether retail traders need to be protected,
either through legislation or through increasing the general level of financial literacy
through education. A capital market where people have the tools to make informed de-
cisions and where they don’t lose money to others due to unawareness is, in our opinion,
absolutely vital in creating a financially sustainable society.
Bibliography
[1] Nicholas Barberis, Andrei Shleifer, and Robert Vishny. “A model of investor sen-
timent”. In: Journal of financial economics (1999). url: https://doi.org/10.
1016/S0304-405X(98)00027-0 (visited on 05/26/2021).
[2] Marshall Blume and Robert Stambaugh. “Biases in computed returns: An applica-
tion to the size effect”. In: Journal of financial economics 12 (1983). url: https:
//doi.org/10.1016/0304-405X(83)90056-9 (visited on 05/22/2021).
[3] Ekkehart Boehmer et al. “Tracking Retail Investor Activity”. In: Journal of Finance
(2020). url: https://doi.org/10.1111/jofi.130 (visited on 05/21/2021).
[4] Bidisha Chakrabarty, Roberto Pascual, and Andriy Shkilko. “Evaluating Trade
Classification Algorithms: Bulk Volume Classification Versus the Tick Rule and
the Lee-Ready Algorithm”. In: Journal of Financial Markets 25 (2015).
[5] Tarun Chordia, Richard Roll, and Avanidhar Subrahmanyam. “Order imbalance,
liquidity and market returns”. In: Journal of financial economics 65 (2000).
[6] Jennifer Conrad, Mustafa N. Gultekin, and Gautam Kaul. “Profitability of Short-
Term Contrarian Strategies: Implications for Market Efficiency”. In: Journal of
Business and Economic Statistics) 15 (1997).
[7] Kent Daniel, David Hirshleifer, and Avindhar Subrahmanyam. “Investor Psychol-
ogy and Security Market Under- and Overreactions”. In: The Journal of Finance
6 (2002).
35
BIBLIOGRAPHY 36
[8] David Easley, Marcos Lopez de Prado, and Maureen O’Hara. “Flow Toxicity and
Liquidity in a High Frequency World”. In: Review of Financial Studies (2012). url:
https://dx.doi.org/10.2139/ssrn.1695596 (visited on 05/14/2021).
[9] Michael Farrell et al. “The Democratization of Investment Research and the In-
formativeness of Retail Investor Trading”. In: Revise and Resubmit (Journal of
Financial Economics) (2020). url: https://papers.ssrn.com/sol3/papers.
cfm?abstract_id=3222841 (visited on 05/21/2021).
[10] Emilios C. Galariotis. “Contrarian and momentumtrading: a review of the litera-
ture”. In: Review of Behavioral Finance (2013). url: https://doi.org/10.1108/
RBF-12-2013-0043 (visited on 05/26/2021).
[11] Erin Gobler. “What is a Meme Stock”. In: The Balance (2021). url: https://
www.thebalance.com/what-is-a-meme-stock-5118074 (visited on 05/19/2021).
[12] Ron Kaniel, Gideon Saar, and Sheridan Titman. “Individual Investor Trading and
Stock Returns”. In: The Journal of the American Finance Association (1999). url:
https://doi.org/10.1111/j.1540-6261.2008.01316.x (visited on 05/25/2021).
[13] Eric K. Kelley and Paul C. Tetlock. “How Wise Are Crowds? Insights from Retail
Orders and Stock Returns”. In: Journal of Finance 68 (2013).
[14] Andrew Keshner. Tendies? Diamond hands? Your guide to the lingo on WallStreet-
Bets, the Reddit forum fueling Gamestops wild rise. 2021. url: https://www.
marketwatch . com / story / tendies - diamond - hands - your - guide - to - the -
lingo- on- wallstreetbets- the- reddit- forum- fueling- gamestops- rise-
11611780829 (visited on 05/22/2021).
[15] Craig Lazzara. “Efficient Markets and Irrational Exuberance”. In: S&P Global Re-
search Insights 49 (2021).
[16] Charles M. C. Lee and Mark J. Ready. “Inferring Trade Direction from Intraday
Data”. In: The Journal of the American Finance Association (1991). url: https:
//doi.org/10.1111/j.1540-6261.1991.tb02683.x (visited on 05/14/2021).
BIBLIOGRAPHY 37
[17] Bruce N. Lehmann. “Fads, Martingales, and Market Efficiency”. In: The Quarterly
Journal of Economics (1990). url: https://doi.org/10.2307/2937816 (visited
on 05/26/2021).
[18] Phil Mackintosh. “Who Counts as a Retail Investor?” In: Nasdaq (2020). url:
https://www.nasdaq.com/articles/who-counts-as-a-retail-investor-
2020-12-17 (visited on 05/22/2021).
[19] Katie Martin and Robin Wigglesworth. “Rise of the retail army: the amateur traders
transforming markets”. In: Financial Times (2021). url: https://www.ft.com/
content/7a91e3ea-b9ec-4611-9a03-a8dd3b8bddb5 (visited on 05/28/2021).
[20] Barclay Palmer. What determines a Stocks’s Bid-Ask Spread? 2021. url: https:
//www.investopedia.com/ask/answers/06/bidaskspread.asp (visited on
05/14/2021).
[21] Alasdair Sandford. Coronavirus: Half of humanity now on lockdown as 90 countries
call for confinement. 2020. url: https://www.euronews.com/2020/04/02/
coronavirus-in-europe-spain-s-death-toll-hits-10-000-after-record-
950-new-deaths-in-24-hou (visited on 05/28/2021).
[22] Ali Shalchi. “The rise of armchair retail trading: Risks and regulation”. In: UK
House of Commons Library (2021).
Appendix A
38
APPENDIX A. 39
Table A.1: Realized volatility as an explanatory variable for retail orderimbalance in
memestocks during 2020-2021.
Retail OIBw Retail OIBw
RealizedVol −1.296∗ −1.374∗
(0.596) (0.685)
Returnw−1 0.029 0.037
(0.034) (0.038)
Retail OIBw−1 0.340∗∗∗ 0.275∗∗∗
(0.068) (0.075)
OIBDqq −0.001∗∗ −0.001∗∗
(0.000) (0.000)
Sales/Turnoverq 0.000∗∗∗ 0.000∗∗∗
(0.000) (0.000)
Month=1 0.000
(.)
Month=2 0.295∗∗∗
(0.016)
Month=3 0.105∗∗∗
(0.020)
Month=4 −0.182∗∗∗
(0.039)
Month=5 −0.189∗∗
(0.073)
Month=6 0.012
(0.028)
Month=7 0.032
(0.032)
Month=8 0.166∗∗∗
(0.026)
Month=9 0.184∗∗∗
(0.033)
Month=10 0.218∗∗∗
(0.030)
Month=11 0.182∗∗∗
(0.032)
Month=12 0.093∗∗∗
(0.028)
Constant 0.354∗∗∗ 0.283∗∗∗
(0.076) (0.084)
Observations 960 960
Standard errors in parentheses
Dependent variables: Retail order imbalance and return.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
APPENDIX A. 40
Table A.2: Predicting future return long period 2016-2021 firm specific
Returnw+1 Returnw+1 Returnw+1 Returnw+1 Returnw+2 Returnw+4
Retail OIBw −0.005 −0.004 0.000 0.005 −0.002 −0.002
(0.004) (0.003) (0.001) (0.006) (0.006) (0.014)
AFL 0.000
(.)
AMC 0.007∗∗∗ 0.000 0.000 0.000 0.000 0.000
(0.000) (.) (.) (.) (.) (.)
BBY 0.003∗∗ 0.008 −0.024 −0.025 −0.029 −0.021
(0.001) (0.004) (0.032) (0.031) (0.033) (0.027)
CI 0.003∗∗ 0.011 −0.024 −0.024 −0.027 −0.019
(0.001) (0.007) (0.030) (0.029) (0.031) (0.026)
CTSH 0.000 0.008∗ −0.026 −0.026 −0.029 −0.020
(0.001) (0.004) (0.031) (0.031) (0.032) (0.026)
CZR 0.014∗∗∗ 0.013∗∗∗ −0.021 −0.024 −0.025 −0.020
(0.002) (0.002) (0.032) (0.030) (0.034) (0.028)
ECL 0.000 0.009∗ −0.024 −0.024 −0.028 −0.019
(0.000) (0.005) (0.031) (0.031) (0.032) (0.027)
ED −0.006∗ 0.005 −0.026 −0.026 −0.031 −0.021
(0.003) (0.005) (0.030) (0.030) (0.031) (0.026)
GME 0.026∗∗∗ 0.019∗∗∗ 0.012∗∗ 0.011∗∗ 0.014∗∗∗ 0.016∗∗
(0.001) (0.002) (0.005) (0.005) (0.004) (0.006)
HES 0.001 0.003 −0.031 −0.031 −0.035 −0.028
(0.001) (0.003) (0.032) (0.031) (0.034) (0.027)
OMC −0.002∗∗∗ 0.005 −0.029 −0.029 −0.032 −0.023
(0.000) (0.004) (0.032) (0.031) (0.033) (0.027)
PTC 0.008∗∗ 0.011∗∗∗ −0.025 −0.026 −0.027 −0.019
(0.003) (0.004) (0.033) (0.031) (0.034) (0.028)
T −0.005∗∗∗ 0.039 0.000 0.000 0.000 0.000
(0.001) (0.037) (.) (.) (.) (.)
Returnw −0.034 −0.034 −0.034 −0.037 0.048
(0.019) (0.020) (0.020) (0.029) (0.058)
OIBDqq −0.000 −0.000 −0.000 −0.000 −0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Sales/Turnoverq 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
RealizedVol 0.657∗∗∗ 0.565∗∗∗ 0.565∗∗∗ 0.686∗ 0.978∗∗∗
(0.199) (0.177) (0.178) (0.347) (0.159)
report 5=1 −0.005 −0.004 −0.013 −0.017
(0.017) (0.017) (0.013) (0.016)
meme=1 0.009 0.008 −0.009 −0.021
(0.038) (0.037) (0.038) (0.038)
report 5=1 × Retail OIBw −0.002 −0.002 0.004 0.008
(0.007) (0.007) (0.006) (0.007)
meme=1 × Retail OIBw −0.021
∗ −0.022∗ −0.015 −0.007
(0.010) (0.011) (0.009) (0.009)
Retail OIBw X logSize −0.000 0.000 0.000
(0.001) (0.001) (0.001)
Constant 0.012 −0.002 0.025 0.025 0.032 0.019
(0.007) (0.009) (0.033) (0.032) (0.034) (0.029)
Observations 15989 14699 14699 14699 14638 14518
Standard errors in parentheses
Dependent variables: Future return, in week 1,2 and,4.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
APPENDIX A. 41
Table A.3: Predicting future return long period 2016-2021 month specific
Returnw+1 Returnw+1 Returnw+1 Returnw+1 Returnw+2 Returnw+4
Retail OIBw −0.004 −0.003 0.001 0.009
∗∗∗ 0.005 −0.000
(0.004) (0.003) (0.001) (0.002) (0.006) (0.009)
Month=1 0.000 0.000 0.000 0.000 0.000 0.000
(.) (.) (.) (.) (.) (.)
Month=2 −0.026 −0.030∗ −0.028 −0.028 −0.022∗∗ −0.006
(0.015) (0.016) (0.016) (0.016) (0.009) (0.011)
Month=3 −0.037∗ −0.045∗∗ −0.044∗∗ −0.044∗∗ −0.026 0.017∗∗∗
(0.017) (0.019) (0.018) (0.018) (0.020) (0.005)
Month=4 −0.014 −0.019 −0.018 −0.018 −0.022 0.000
(0.017) (0.018) (0.018) (0.018) (0.016) (0.004)
Month=5 −0.025 −0.029 −0.027 −0.027 −0.018 0.005
(0.022) (0.023) (0.022) (0.022) (0.019) (0.008)
Month=6 −0.032 −0.035 −0.034 −0.034 −0.027 0.006
(0.019) (0.020) (0.019) (0.019) (0.017) (0.007)
Month=7 −0.032 −0.034 −0.033 −0.033 −0.026 0.008
(0.019) (0.019) (0.019) (0.019) (0.016) (0.008)
Month=8 −0.025 −0.026 −0.025 −0.025 −0.019 0.010
(0.016) (0.016) (0.015) (0.015) (0.013) (0.011)
Month=9 −0.029 −0.031 −0.031 −0.031 −0.028 −0.001
(0.018) (0.018) (0.018) (0.018) (0.016) (0.008)
Month=10 −0.036∗ −0.039∗ −0.038∗ −0.038∗ −0.024 0.017∗∗
(0.020) (0.020) (0.020) (0.020) (0.018) (0.007)
Month=11 −0.015 −0.017 −0.017 −0.017 −0.015 0.002
(0.018) (0.018) (0.018) (0.018) (0.015) (0.007)
Month=12 −0.032 −0.034 −0.035 −0.035 −0.022 0.045
(0.020) (0.021) (0.021) (0.021) (0.017) (0.028)
Returnw −0.045
∗∗ −0.044∗ −0.043∗ −0.042 0.055
(0.019) (0.021) (0.021) (0.031) (0.056)
OIBDqq −0.000
∗∗ −0.000∗ −0.000∗∗ −0.000∗∗ −0.000∗
(0.000) (0.000) (0.000) (0.000) (0.000)
Sales/Turnoverq 0.000 0.000 0.000
∗ 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
RealizedVol 0.730∗∗∗ 0.617∗∗∗ 0.603∗∗∗ 0.700∗ 0.956∗∗∗
(0.220) (0.194) (0.190) (0.357) (0.179)
report 5=1 −0.004 −0.004 −0.012 −0.013
(0.016) (0.016) (0.012) (0.017)
meme=1 0.042∗ 0.041∗ 0.030∗ 0.012
(0.020) (0.020) (0.015) (0.022)
report 5=1 × Retail OIBw −0.004 −0.004 0.004 0.008
(0.007) (0.007) (0.005) (0.007)
meme=1 × Retail OIBw −0.022
∗ −0.023∗∗ −0.017∗ −0.009
(0.010) (0.010) (0.009) (0.010)
Retail OIBw X logSize −0.001
∗∗∗ −0.001 −0.000
(0.000) (0.001) (0.001)
Constant 0.039 0.032 0.024 0.025 0.021 −0.013
(0.025) (0.023) (0.018) (0.018) (0.014) (0.009)
Observations 15989 14699 14699 14699 14638 14518
Standard errors in parentheses
Dependent variables: Future return, in week 1,2 and,4.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
APPENDIX A. 42
Table A.4: Predicting future return short period 2020-2021 firm specific
Returnw+1 Returnw+1 Returnw+1 Returnw+1 Returnw+2 Returnw+4
Retail OIBw −0.013 −0.014 −0.001 −0.006 −0.087 −0.133
∗∗
(0.009) (0.010) (0.002) (0.023) (0.052) (0.052)
Inst OIBw 0.009 0.009 0.011 0.135 0.011 −0.055
(0.006) (0.006) (0.007) (0.110) (0.088) (0.060)
AFL 0.000 0.000 0.000 0.000 0.000 0.000
(.) (.) (.) (.) (.) (.)
AMC 0.034∗∗∗ 0.006 −0.080∗∗∗ −0.083∗∗∗ −0.088∗∗∗ −0.089∗∗∗
(0.009) (0.020) (0.004) (0.004) (0.005) (0.010)
BBY 0.003∗∗∗ 0.039 0.034 0.032 0.042 0.043
(0.001) (0.027) (0.025) (0.024) (0.030) (0.033)
CI 0.009 0.177 0.151 0.127 0.143 0.147
(0.006) (0.138) (0.129) (0.111) (0.150) (0.162)
CTSH −0.001 0.006 0.004 0.008 0.013∗∗ 0.004
(0.002) (0.006) (0.006) (0.006) (0.006) (0.006)
CZR 0.019∗∗∗ 0.003 0.006 0.009 0.041∗ 0.044∗
(0.003) (0.016) (0.016) (0.012) (0.021) (0.021)
ECL 0.002 0.003 −0.010 0.001 0.000 −0.023
(0.005) (0.009) (0.013) (0.011) (0.013) (0.015)
ED −0.014∗ −0.005 0.000 0.006 0.007 0.001
(0.007) (0.011) (0.009) (0.010) (0.011) (0.011)
GME 0.115∗∗∗ 0.098∗∗∗ 0.000 0.000 0.000 0.000
(0.009) (0.018) (.) (.) (.) (.)
HES 0.004 −0.010 −0.012 −0.005 0.006 0.003
(0.003) (0.012) (0.014) (0.011) (0.019) (0.017)
OMC 0.006 0.007 −0.000 0.007 0.022 0.021
(0.005) (0.008) (0.006) (0.010) (0.016) (0.012)
PTC 0.012∗∗ −0.006 −0.011 −0.007 0.021 0.025
(0.004) (0.016) (0.019) (0.013) (0.021) (0.020)
T −0.007∗∗∗ 0.356 0.304 0.299 0.335 0.270
(0.001) (0.219) (0.197) (0.188) (0.224) (0.245)
Returnw −0.023 −0.033 −0.033 −0.040 0.053
(0.036) (0.040) (0.041) (0.027) (0.064)
OIBDqq −0.000
∗ −0.000∗ −0.000∗ −0.000∗∗ −0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Sales/Turnoverq −0.000 −0.000 −0.000 −0.000 −0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
RealizedVol 0.154 −0.093 −0.116 0.145 0.626∗∗
(0.351) (0.426) (0.456) (0.233) (0.225)
report 5=1 −0.005 −0.004 −0.035 −0.045
(0.052) (0.051) (0.045) (0.049)
meme=1 0.287∗∗∗ 0.301∗∗∗ 0.246∗∗∗ 0.165∗
(0.029) (0.035) (0.021) (0.086)
report 5=1 × Retail OIBw −0.002 −0.003 0.014 0.027
(0.021) (0.021) (0.020) (0.022)
meme=1 × Retail OIBw −0.161
∗∗∗ −0.165∗∗∗ −0.092∗∗∗ −0.024
(0.015) (0.015) (0.029) (0.072)
Retail OIBw X logSize 0.001 0.008 0.013
∗∗
(0.002) (0.005) (0.005)
Inst OIBw X logSize −0.012 −0.001 0.005
(0.010) (0.008) (0.006)
Constant 0.031 0.050 0.029 0.024 0.030 0.022
(0.019) (0.039) (0.026) (0.022) (0.027) (0.030)
Observations 4212 4147 4147 4147 4082 3952
Standard errors in parentheses
Dependent variables: Future return, in week 1,2 and,4.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
APPENDIX A. 43
Table A.5: Predicting future return short period 2020-2021 month specific
Returnw+1 Returnw+1 Returnw+1 Returnw+1 Returnw+2 Returnw+4
Retail OIBw −0.024 −0.019 −0.001 0.028
∗∗∗ −0.003 −0.011
(0.015) (0.014) (0.002) (0.009) (0.022) (0.016)
Inst OIBw 0.009 0.008 0.009 0.135 0.011 −0.056
(0.006) (0.005) (0.005) (0.102) (0.076) (0.065)
Month=1 0.000 0.000 0.000 0.000 0.000 0.000
(.) (.) (.) (.) (.) (.)
Month=2 −0.078 −0.094 −0.089 −0.088 −0.083∗∗ 0.001
(0.045) (0.056) (0.053) (0.052) (0.033) (0.031)
Month=3 −0.093∗ −0.118∗ −0.111∗ −0.111∗ −0.078 0.041∗∗
(0.050) (0.056) (0.055) (0.057) (0.058) (0.015)
Month=4 −0.043 −0.060 −0.063 −0.061 −0.070 0.020∗
(0.051) (0.059) (0.062) (0.063) (0.053) (0.011)
Month=5 −0.051 −0.067 −0.070 −0.070 −0.050 0.013
(0.063) (0.070) (0.073) (0.072) (0.061) (0.012)
Month=6 −0.090 −0.106 −0.109 −0.110 −0.104 0.014
(0.057) (0.065) (0.068) (0.070) (0.061) (0.013)
Month=7 −0.075 −0.089 −0.092 −0.092 −0.068 0.033
(0.055) (0.063) (0.066) (0.066) (0.056) (0.019)
Month=8 −0.054 −0.066 −0.068 −0.070 −0.072 0.021
(0.042) (0.051) (0.053) (0.055) (0.046) (0.028)
Month=9 −0.084 −0.099 −0.099 −0.101 −0.075 0.019
(0.054) (0.062) (0.063) (0.064) (0.052) (0.022)
Month=10 −0.082 −0.097 −0.094 −0.093 −0.069 0.058∗∗∗
(0.056) (0.065) (0.065) (0.064) (0.057) (0.016)
Month=11 −0.027 −0.041 −0.043 −0.040 −0.054 0.007
(0.046) (0.054) (0.057) (0.055) (0.050) (0.020)
Month=12 −0.082 −0.095 −0.096 −0.095 −0.067 0.174
(0.057) (0.067) (0.067) (0.067) (0.048) (0.115)
Returnw −0.062
∗∗ −0.054 −0.054 −0.051 0.065
(0.028) (0.042) (0.042) (0.033) (0.057)
OIBDqq −0.000 −0.000
∗∗ −0.000∗ −0.000∗∗ −0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Sales/Turnoverq 0.000 0.000 0.000
∗∗ 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
RealizedVol 0.687∗ 0.065 0.015 0.226 0.652∗∗∗
(0.353) (0.493) (0.509) (0.278) (0.150)
report 5=1 −0.001 0.001 −0.028 −0.035
(0.046) (0.046) (0.040) (0.057)
meme=1 0.257∗∗∗ 0.269∗∗∗ 0.211∗∗∗ 0.136
(0.058) (0.064) (0.042) (0.103)
report 5=1 × Retail OIBw −0.007 −0.007 0.010 0.029
(0.019) (0.019) (0.016) (0.026)
meme=1 × Retail OIBw −0.162
∗∗∗ −0.181∗∗∗ −0.131∗∗∗ −0.081
(0.016) (0.023) (0.025) (0.069)
Retail OIBw X logSize −0.003
∗∗ −0.001 0.000
(0.001) (0.002) (0.001)
Inst OIBw X logSize −0.012 −0.001 0.006
(0.010) (0.007) (0.006)
Constant 0.129 0.121 0.088 0.086 0.086 −0.022
(0.080) (0.091) (0.062) (0.062) (0.049) (0.020)
Observations 4212 4147 4147 4147 4082 3952
Standard errors in parentheses
Dependent variables: Future return, in week 1,2 and,4.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01