A49 I The Dark Side of Trading

Article
Journal of Accounting,
Auditing & Finance
The Dark Side of Trading 2014, Vol. 29(4) 492–518
ÓThe Author(s) 2014
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0148558X14537826
jaf.sagepub.com
Ilia D. Dichev1, Kelly Huang2, and

Dexin Zhou1
Abstract
This study investigates the effect of high trading volume on observed stock volatility con-
trolling for fundamental information. We investigate a number of settings, including two nat-
ural experiments (exchange-traded funds [ETFs] and dual-class shares), the aggregate time-
series of U.S. stocks since 1926, and the cross-section of U.S. stocks during the last 20
years. Our main finding is that there is an economically substantial positive relation
between trading volume and stock volatility, especially when trading volume is high. The
conclusion is that stock trading can inject volatility above and beyond that based on
fundamentals.
Keywords
stock trading, volatility, stock returns, volume, fundamentals
Introduction
We investigate the effect of high trading volume on stock volatility. Given existing work
that links volume and volatility as simultaneously driven by the flow of fundamental infor-
mation (e.g., Karpoff, 1987), we are specifically interested in the effect of high volumes of
trading holding fundamental information constant.1 The motivation is that stock trading
volume has increased tremendously during the last 50 years, from an annualized value-
weighted NYSE/AMEX turnover of less than 10% in 1960 to more than 300% in 2008-
2009 (see evidence in Figure 1). A change of this magnitude can be fairly characterized as
transforming the marketplace, and it is important to document and assess the parameters of
this transformation. The stock market setting is just one manifestation of a powerful trend
of great increases in trading volume across a number of investment assets, including bonds,
commodities, currencies, and derivatives. Thus, the findings of this study can possibly have
broad utility for the investment world at large.
A well-established existing literature maps out a positive relation between volume and
volatility. Generally speaking, this literature investigates the endogenous co-movement of
1
Emory University, Atlanta, GA, USA
2
Florida International University, Miami, FL, USA
Corresponding Author:
Ilia D. Dichev, Emory University, 201 Dowman Drive, Atlanta, GA 30322, USA.
Email: [email protected]
Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Dichev et al. 493
4.0
3.5
3.0
Annualized Trading Volume
2.5
2.0
1.5
1.0
0.5
0.0
1926
1928
1930
1932
1934
1936
1938
1940
1942
1944
1946
1948
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
NYSE/AMEX NASDAQ
Figure 1. Value-weighted stock trading volume from 1926 to 2012.

Note. This figure shows the annualized value-weighted trading volume for NYSE/AMEX (solid line) from 1926 to
2012 and Nasdaq (patterned line) from 1983 to 2012. Annualized value-weighted trading volume is the average
daily value-weighted market trading volume for calendar year t multiplied by the number of trading days in year t
(ranging from 226 to 302 days with 252 days for most years). Daily value-weighted trading volume is measured as
dollar-value traded (volume 3 price) on a trading day aggregated over all stocks on the corresponding exchanges
divided by aggregate market value (price 3 shrout) outstanding as of that day. Volume for stocks traded on
Nasdaq is volume on CRSP - the Center for Research in Security Prices scaled by two.
volume and returns, where the basic message is that new information sparks trades and trig-
gers corresponding price discovery (Schwert, 1989). However, other arguments and evi-
dence suggest that non-informational noise trading can lash prices away from
fundamentals, into temporary swings and reversals (Campbell, Grossman, & Wang, 1993).
For example, Shiller (1981) suggests that stock prices are ‘‘too volatile’’ given the variabil-
ity of underlying fundamentals. The interplay of these two opposing forces is not under-
stood well, and we have a poor idea of which effect dominates, especially in view of the
dramatic increase in trading during the last half century.
The most significant problem in this investigation is that information flow endogenously
drives both volatility and volume. We address this problem in two ways. First, we identify
two natural experiments, where the settings provide tight controls for information flow and
firm and business characteristics, while there is a significant exogenous variation in volume
of trading. Specifically, we look at matched pairs of exchange-traded funds (ETFs) that
mirror the same underlying index but have different trading volume, for example, the SPY
fund and the IVV fund both track the Standard & Poor’s (S&P) 500 index but the older
and better known SPY is much more heavily traded. We also examine dual-class U.S.
stocks where typically the two classes have identical cash flow rights but different control

494 Journal of Accounting, Auditing & Finance
rights and different trading volume. Our main finding is that in these matched pairs of
ETFs and dual-class stocks, higher volume of trading leads to higher return volatility.
Second, we explore the relation between volume of trading and stock volatility in the
aggregate time-series of U.S. stocks since 1926 and in the cross-section during the last 20
years, while controlling for information flow and other determinants of volatility. The
advantage of this broad setting over the previous two natural experiments is better calibra-
tion of the examined effects to the natural properties of U.S. stocks as we are interested in
the implications of the drastic increase of trading volume in recent years; the disadvantage
is losing some of the sharpness of the controls in the natural experiments. We find that the
correlation between annual aggregate measures of volume and volatility is on the magni-
tude of 50% in levels and 30% in changes in the aggregate time-series, which are highly
statistically significant and economically substantial. In addition, we perform an Actual/
Implied volatility ratio analysis and find that greater aggregate volume corresponds to
smaller aggregate Actual/Implied ratios, which indicates that high volume leads to greater
return reversals and less efficient securities prices. Using an extensive set of control vari-
ables, we also find a positive and convex relation between volume and volatility in the
cross-section of stocks, where the relation is much clearer and stronger for high volumes of
trading. In efforts to more precisely quantify and calibrate the effect of trading on volatility,
we estimate that in recent years trading-induced volatility accounts for about a quarter of
total observed stock volatility.
Summarizing, these results suggest that trading can create its own volatility above and
beyond the volatility due to fundamentals. Prior research finds that higher volumes are
highly prized and rewarded by investors; they are correlated with lower transaction costs,
easier creation and adjustment of investment positions, and lead to higher prices (e.g.,
Branch & Freed, 1977; Brennan, Chordia, & Subrahmanyam, 1998; Jones, 2002). One
implication of our study is that the benefits of increased trading are not a one-way street.
Given that existing evidence on benefits is mostly for relatively low levels of trading, the
results in this study suggest that there could be a point (or range) of optimal levels of trad-
ing, and that there are costs of going beyond that.
Our findings have implications for various market participants including managers,
investors, and regulators. For managers, excess stock volatility is mostly a nuisance, obfus-
cating the effect of managerial efforts on firm value, raising the cost of capital, and compli-
cating investor relations (Luo, 2005). But excess volatility can also create more windows
for issuing and repurchasing stock at favorable prices. On the investor side, excess volati-
lity is also generally undesirable, as volatility is closely related to risk, and to decreased
propensity to invest. But certain segments of sophisticated investors probably like and even
thrive on volatility, as price fluctuations allow them more opportunities for trading profits.
Finally, market-makers and regulators need to be cognizant and proactive about the fact
that high trading volume leads to high volatility, and is therefore destabilizing. To a certain
extent, such reactions already exist, for example, circuit-breakers dampen extreme price
moves by halting trading. There are also other possibilities, such as imposing a small trans-
action tax on trading to dampen speculation and reduce the diversion of resources into the
financial sector (Ripley, 1911; Summers & Summers, 1989). Overall, although excess vola-
tility probably expands the possibility for private gain for some stakeholders, it seems
mostly a net negative for society at large. Future research can help to further map out the
characteristics and consequences of trading-induced volatility.

Dichev et al. 495
Theory and Background

Our goal is to investigate the effect of high volumes of trading on stock volatility, with a
particular emphasis on the effect of intense trading controlling for the flow of the underly-
ing fundamental information. The motivation is that volume of stock trading has increased
tremendously during the last 30 to 50 years (e.g., Baker & Stein, 2004; Chordia, Roll, &
Subrahmanyam, 2011). Figure 1 provides an illustration of this phenomenon for the full
history of volume data on the major U.S. exchanges, 1926-2012 for NYSE/AMEX and
1983-2012 for Nasdaq; specifically, Figure 1 plots annualized value-weighted turnover
(volume/shares outstanding) over time.2 An examination of Figure 1 reveals a dizzying
growth in trading with NYSE/AMEX turnover of less than 10% a year during 1940-1970, a
somewhat uneven but forceful rise during 1970-2000, hitting highs of 200% to 300% in a
pronounced spike of trading in the late 2000s, and a retreat to 160% in the most recent
year.3 The Nasdaq time-series, although much shorter, reveals a similar pattern of sixfold
increase but with a less pronounced spike in the most recent years. A market in which secu-
rities change hands once in 10 years is likely to be qualitatively different from a market in
which securities change hands 2 or 3 times a year, and this difference likely leads to quali-
tatively different outcomes in fundamental issues like security valuation, equity risk, and
market efficiency. Our study assesses the implications of these material changes by exam-
ining the effect of high volumes of trading on stock volatility.
There is a large existing literature that maps out a positive relation between volume and
volatility. Generally speaking, this literature investigates the endogenous co-movement of
volume and returns, where the basic message is that ‘‘volume moves prices’’ (see Karpoff,
1987, for an early review). Although this literature is rather broad, its unifying intuition is
that new information sparks trades and triggers corresponding price revisions over rela-
tively short horizons. There have been significant accomplishments in this line of research,
which studies issues like the effects of private versus public information, information asym-
metry, and information with different precision on volume and security prices (Bamber,
Barron, & Stober, 1999; Easley, Kiefer, & O’Hara, 1996, 1997; Kandel & Pearson, 1995;
Morse, 1980).
Other arguments and evidence suggest that high volume can induce excessive non-fun-
damental driven volatility. Predictions along these lines have surfaced in various forms in
the literature but essentially the idea is that trading produces trading noise, and this noise
can lash prices away from fundamentals. For example, Shiller (1981) suggests that stock
prices are ‘‘too volatile’’ given the variability of underlying fundamentals. DeLong,
Shleifer, Summers, and Waldmann (1990) argue that positive feedback investment strate-
gies can result in excess volatility even in the presence of rational speculators. The fascinat-
ing finding that stock returns are on the magnitude of 10 times more volatile during trading
hours than during non-trading hours (French & Roll, 1986) is also consistent with the view
that trading produces its own volatility. A similar conclusion is reached by Black (1986),
who argues that noise traders increase trading and simultaneously introduce noise in prices,
and thus more trading and higher volatility go hand-in-hand.
The large increases in trading presented in Figure 1 motivate us to investigate the effect
of high volumes of trading holding fundamental information constant. A number of factors
have likely contributed to the leap in trading volume, including decreases in trading costs
(Chordia et al., 2011), easier access to stock trading because of the popularization of
Internet brokerages (Barber & Odean, 2002), and increased financial literacy among aver-
age people (van Rooij, Lusardi, & Alessie, 2011). There is also evidence that

fundamentally the economy today is more volatile than in the 1960s (Irvine & Pontiff,
2009; Wei & Zhang, 2006). Therefore, it is possible that newer and faster information
sources, easier access to stock markets, and more volatile firm fundamentals lead to more
news and more trading.
On some basic level, though, it seems implausible that the more than 20-fold increase in
trading since the 1960s is purely driven by more news about fundamentals. Even more tell-
ing in this regard is actually the comparison of the 1940-1970 period with the 1926-1940
period in Figure 1. Note that the 1926-1940 period also represents a prolonged episode of
heavy trading, and while its intensity is not as pronounced as in the most recent years, it is
remarkable that annualized turnover both before and after the 1929 crash was more than
100% a year, 10 times as much as during the 1940-1970 period (which includes watershed
information events like World War II and the Korean war). Differences of such magnitude
are difficult to square with just differences in the amount of fundamental information, and
it is highly unlikely that information sources were better in the 1920s than two decades
later.
In any case, in addition to the indirect and only suggestive evidence in Figure 1, there is
more specific evidence that a great amount of trading is not driven by fundamental infor-
mation, and that the amount of such trading has increased over time. One example of non-
information trading is ‘‘liquidity trading,’’ that is, trading driven by personal consumption
needs or windfalls as opposed to stock fundamentals. Consistent with the hypothesized
effects of non-fundamentals trading, Coval and Stafford (2007) document that fire sales or
purchases by mutual fund investors lead prices of stocks held by the funds to shift away
from fundamental value and result in significant return reversals. Other trading can be
thought of as triggered by a number of different reasons, which span a continuum between
trading purely driven by fundamental information to trading purely driven by non-informa-
tion motivations like sentiment. In fact, much trading falls in the gray area between pure-
information and non-information trading (Chordia, Huh, & Subrahmanyam, 2007). A vivid
illustration of this gray area is various types of algorithmic trading, which apparently
account for more than 70% of all trading today (Hendershott, Jones, & Menkveld, 2011).4
A trading algorithm based on momentum, for example, is based on information from the
past pattern of security prices, essentially from past trading itself. As momentum trading
also shapes prices, there is a lot of room for feedback loops and other interactions, which
affect prices but are not based on fundamental information. More generally, a lot of trading
is based on watching and reacting to the actions of other traders, and has little to do with
true underlying fundamentals. We attempt to capture the impact of this kind of non-infor-
mation trading by examining the relation between high trading volume and stock volatility
while controlling for firm fundamentals.5
A summary impression from the extant literature is that higher volume of trading is an
almost universally good thing. A number of studies have documented that increased
volume is reliably related to decreased transaction costs (bid-ask spreads, brokerage fees,
execution costs) where these two variables reinforce each other, and innovations in either
one can lead to changes in the other (Branch & Freed, 1977; Copeland & Galai, 1983).
Investors like higher volume because it allows them to build and adjust investment posi-
tions easier, faster, and cheaper. Market-makers also like volume because it generally
makes their job easier and less risky. Another reliable finding is that, everything else equal,
higher volume leads to lower cost of capital and higher prices (Amihud & Mendelson,
1986; Brennan et al., 1998; Liu, 2006).6

Dichev et al. 497
What has yet to receive much attention in the literature is that high volumes of trading
can be destabilizing, injecting a sizable layer of trading-induced volatility over and above
the unavoidable fundamentals-based volatility. Some observations from practice highlight
this potential link between trading and volatility. Stock exchanges often employ circuit-
breakers, a policy of shutting down trading for a pre-specified amount of time after large
price drops, either at the aggregate or at the individual security level. Such policies are
questionable and even counter-productive if one takes the view that large price drops indi-
cate dramatic revisions of information, and that it is in precisely such times that trading
and the associated pricing process are most needed and should be allowed to freely flow to
their new equilibrium levels. The counterpoint is that such policies are likely not accidental
and are really the evolutionary outcome of much historical trial-and-error, where the accu-
mulated wisdom indicates that sometimes trading can go haywire for no particular reason
related to fundamentals, and then a mandatory break allows everyone to cool off. Thus,
such policies are consistent with the view that trading can produce its own volatility, and
sometimes this volatility can get so out of hand that the simplest and most effective way to
tame it is to completely shut down trading.
Natural Experiments
We start our empirical investigation of the effect of trading volume on stock volatility with
two natural experiments. The advantage of this approach is that when an appropriate setting
is available, there is a natural and efficient control for potentially confounding variables.
Here, as discussed earlier, the most important variables to control for are those related to
information flow but an appropriate setting would also control for other influential vari-
ables like firm size, profitability, nature of business, corporate governance, investor clien-
tele variables, and so on. On a philosophical level, we choose to pursue relatively ‘‘tight’’
research settings, where the controls for fundamentals are as good as possible but at the
possible cost of limited generalizability. Specifically, our two settings rely on comparisons
of essentially the same underlying security across different trading environments, which
provide enough exogenous variation in trading intensity while holding fundamentals nearly
perfectly constant.7 Our first setting is ETFs, where we identify matched pairs of funds
which track the same underlying benchmark but have materially different trading volume,
for example, both the SPY fund and the IVV fund track the S&P 500 index but the older
and better known SPY fund is much more heavily traded. Our second setting is dual-class
stocks, where typically the two classes have the same cash flow rights but difference in
control rights leads to differential volumes of trading. The unifying feature and the main
advantage of these two settings is near-perfect control for the underlying fundamentals,
which provides a sharp test for the conjectured volume/volatility relation. The disadvantage
is that because the two instruments are near-identical substitutes, arbitrage is likely to keep
return and volatility differentials to a minimum. Thus, we view the natural experiments as
sharp tests of existence for the hypothesized phenomenon but at the cost of limited evi-
dence about the economic magnitude of the effects in the wider world of investments.
ETFs
Using CRSP data, we identify 16 fund-pairs traded on the major U.S. stock exchanges
(NYSE, AMEX, Nasdaq, and NYSE Arca), which track the same index for at least one full
year.8 Table 1, Panel A includes the names of the underlying indexes, the ticker symbols of

498
Table 1. Evidence From Exchange-Traded Funds.
Panel A: Sample Composition.

Index name Type Ticker (a) MV (a) Ticker (b) MV (b) Sample period
Barclays Capital U.S. Aggregate Index X AGG 11,840 LAG 246 June 2007-December 2010
Barclays Capital U.S. MBS Index X MBB 2,204 MBG 35 February 2009-December 2010
Dow Jones U.S. Large-Cap Growth Total Stock Market X ELG 192 SCHG 172 January 2010-December 2010
Dow Jones U.S. Large-Cap Total Stock Market X SCHX 386 ELR 41 November 2009-December 2010
Dow Jones U.S. Large-Cap Value Total Stock Market X ELV 172 SCHV 121 January 2010-December 2010
MSCI EAFE Index X EFA 36,518 VEA 5,134 August 2007-December 2010
MSCI Emerging Markets Index X EEM 48,153 VWO 42,853 April 2005-December 2010
Russell 2000 Index 22X TWM 468 RRZ 12 November 2007-April 2010
Russell 2000 Index 2X UWM 243 RRY 31 November 2007-April 2010
S&P 500 Index X SPY 84,702 IVV 24,229 June 2000-December 2010
S&P 500 Index 22X SDS 2,806 RSW 73 November 2007-December 2010
S&P 500 Index 2X SSO 1,592 RSU 88 November 2007-December 2010
S&P MidCap 400 Index X MDY 10,798 IJH 8,984 June 2000-December 2010
S&P MidCap 400 Index 2X MVV 149 RMM 22 November 2007-April 2010
S&P MidCap 400 Index 22X MZZ 40 RMS 3 November 2007-April 2010
S&P/Citigroup International Treasury Bond Index X IGOV 161 ISHG 113 February 2009-December 2010
Panel B: Descriptive Statistics.

n M SD 10% 25% 50% 75% 90%
DIF_VOLUME 683 642.5% 662.8% 42.6% 160.5% 412.2% 947.1% 1,606.7%
DIF_STDRET 683 3.8%*** 20.5% 26.1% 21.9% 0.6%*** 4.6% 13.1%
RETCORR 16 0.979 0.036 0.905 0.984 0.993 0.999 0.999
(continued)
Table 1. (continued)
Panel C: Difference in Trading Volume and Stock Volatility for ETFs.

M (%) Median (%)
Quintiles by DIF_VOLUME (sorted at the index level) DIF_VOLUME DIF_STDRET DIF_VOLUME DIF_STDRET
Q1 148.1 0.6 86.1 0.2
Q2 363.9 3.6** 240.8 1.0**
Q3 561.0 3.3* 382.6 0.6
Q4 777.3 6.3*** 583.1 0.6*
Q5 1,368.5 5.0*** 1,077.0 1.5**
Q5 2 Q1 diff. 1,220.4 4.4** 991.0 1.3
Note. This table reports results of analyses on exchange-traded funds. Panel A reports sample composition of ETF pairs that trace the same index. Type indicates how fund
returns correspond to the underlying index returns. X (2X, 22X) indicates that the fund before fees and expenses seeks investment results that correspond to 100% (200%,
2200%) of daily performance of the underlying index. MV is the market value of equity ($million) for each fund at the end of sample period. Panel B reports the descriptive sta-
tistics. For each index-month, shares in a pair are split into high and low groups based on their corresponding trading volume. DIF_VOLUME is the volume difference between
high- and low-volume funds, scaled by the volume of the low-volume fund. Trading volume is the sum of daily trading volume in a month, calculated as volume divided by shares
outstanding. DIF_STDRET is the stock volatility difference between the high- and the low-volume funds for each fund-month, scaled by the stock volatility of the low-volume

fund. Stock volatility is measured as the standard deviation of daily returns in a month. RETCORR is the within-pair return (Pearson) correlation between the two funds that
track the same index. Panel C reports mean and median DIF_VOLUME and DIF_STDRET across DIF_VOLUME quintiles sorted at the index level. The p value for the difference
between the top and bottom quintiles (Q5 2 Q1 diff.) is based on z statistics for medians and t statistics for means.
***, **, and * denote significance (two-tailed) at the 1%, 5%, and 10% levels, respectively.
499
the two ETFs that track each particular index, fund type, the ending market values of the
funds, and the available time span over which both funds are in operation. The data in
Panel A reveal that these are mostly large and well-known U.S. indexes and funds, with
data limited to the last 10 years. Panel B provides descriptive statistics for the main test
variables in our investigation, DIF_VOLUME and DIF_STDRET, which are defined at the
index-month level. Specifically, for each available index-month, we calculate the average
daily turnover for each of the two funds, tag them as ‘‘high’’ and ‘‘low’’ within each pair,
and create the variable DIF_VOLUME as the difference in volume between the high- and
low-volume funds, scaled by the volume of the low-volume fund. This variable is inter-
preted as the percentage by which the turnover of the high fund exceeds that of the low
fund, for example, a value of 100% implies that the high fund has turnover double that of
the low fund. Then, in a similar manner and with a similar interpretation, we create the
variable DIF_STDRET defined as the volatility difference between high- and low-volume
funds, scaled by the volatility of the low-volume fund.
An inspection of Panel B reveals a wide variation in the DIF_VOLUME variable, with a
median of 4.12 and mean of 6.43, which implies that the more heavily traded funds have
on the magnitude of 5 to 7 times higher turnover than their corresponding lightly traded
counterparts. The descriptive statistics also reveal that the two test variables are non-
normal, with positive skewness and heavy tails. Because of these non-normalities, many of
our subsequent specifications include robust measures of central tendency (e.g., medians)
and non-parametric tests. Panel B also provides some evidence on the limitations of our
investigation. Note that due to the short time-series and the low number of available fund-
pairs, the number of index-month observations is limited to only 683. In addition, the
returns within fund-pairs are highly correlated, with an average correlation of .98. On one
hand, this high correlation is re-assuring because it is consistent with our premise that these
pairs track the very same underlying fundamentals. On the other hand, it also reflects our
priors that arbitrage forces likely keep differential effects across the paired securities to a
minimum, so the observed differences in volatility are likely to be small. Finally, the
descriptive statistics provide preliminary evidence about our main research question. As
both mean and median DIF_STDRET are positive and significant, the implication is that,
holding fundamentals constant, higher trading volume is associated with higher return
volatility.
For the main tests in Panel C, we aim to more fully use the natural variation in the
sample by ranking the index-month observations into within-index quintiles on their
DIF_VOLUME variable, and reporting the resulting mean and median values of
DIF_STDRET across quintiles. If higher volume is indeed associated with higher volatility,
we expect to see positive and increasing values for DIF_STDRET across quintiles. An
inspection of Panel C is largely in line with this conjecture, although the patterns of
increase are non-monotonic and the economic magnitudes are small. Specifically, all
means and medians of DIF_STDRET across quintiles are positive, and most are significant
at least at the .10 level. We also see some evidence that DIF_STDRET is increasing across
quintiles, where the spread of mean DIF_STDRET between Quintiles 5 and 1 is 4.4% and
significant at the .05 level, and the spread in medians is 1.3% but insignificant.9
Overall, in trying to distinguish whether higher volume leads to lower or higher volati-
lity, the ETF evidence shows no evidence of the former, and moderate support for the
latter relation. In interpreting this evidence, it is good to keep in mind that the ETF setting
provides the desired tight controls but at the cost of limited number of observations and the

Dichev et al. 501
availability of easy arbitrage, which constrain the statistical and economic significance of
the results.
Dual-Class U.S. Stocks

Our second natural experiment relies on a comparison of volatility across dual-class U.S.
stocks, where the two classes usually have identical cash flow rights but different control
rights (e.g., A-class shares have 10 times the voting power of B-class shares) and often sub-
stantially different volumes of trading. Consistent with this intuition, several previous stud-
ies use the same or similar settings to control for underlying cash flows. For example,
Zingales (1994) uses dual-class firms to study the pricing of voting rights, whereas
Gompers, Ishii, and Metrick (2010) study the difference in insiders’ cash flow rights and
voting rights. There are, however, two limitations to the dual-class setting. The first limita-
tion is that—similar to ETFs—the two classes of shares are close substitutes, and thus
arbitrage forces keep their returns and volatility of returns within fairly tight bounds. The
second limitation is that there is usually a price difference between the two classes, which
reflects the value of the control premium. As the value of the control premium likely varies
over time, it creates a separate source of return differences over time, possibly confounding
our investigation. We have some priors, though, that the second limitation is unlikely to be
critical. Lease, McConnell, and Mikkelson (1983, 1984) document that superior voting
shares generally have a small (5%) premium over inferior voting shares.
Our sample of dual-class stocks is obtained by searching CRSP data from 1965 to 2009
for entries with the same PERMCO and company name but distinct PERMNOs. We also
require that both issues are common stocks, are listed on the same major U.S. exchange
(NYSE, AMEX, Nasdaq, or NYSE Arca), and have an overlap of at least 4 years of trad-
ing. Table 2, Panel A shows that our sample consists of 59 firms, 118 issues, and 7,322
firm-months. The number of firms used in our research is comparable with previous
research. Brief descriptive statistics in Panel B of Table 2 reveal that these are sizable
firms with mean (median) market cap of US$1,789 million (US$148 million). The average
correlations in monthly returns between the two share classes are about .79, which confirms
that the two classes are largely moved by the same underlying fundamental information.
Note that the larger sample size and the lower within-pair return correlations in the dual-
class setting suggest that the dual-class setting probably allows for greater differential stock
volatility effects than the ETF setting.
Panel B also reports descriptive statistics for the test variables. Similar to the ETF set-
ting, for each available pair-month, we calculate the volume for each of the two issues, tag
them as ‘‘high’’ and ‘‘low’’ within each pair, and create the variable DIF_VOLUME
defined as the trading volume difference between high- and low-volume issues, scaled by
the volume of low-volume issue. Then, we create the variable DIF_STDRET defined as the
stock return volatility difference between the high- and low-volume issues, scaled by the
return volatility of the low-volume issue. Panel B reveals that there are large differences in
liquidity between the two share classes, for example, the median (mean) DIF_VOLUME is
about 1.45 (45.47). Note that the median (mean) DIF_STDRET is 2.6% (20.9), both signifi-
cant at the .01 level, which provides preliminary evidence that shares with higher turnover
have higher return volatility as well.
In Panel C, we explore the natural variation in the sample by ranking the firm-month
pairs into within-firm quintiles on their DIF_VOLUME variable, and reporting the spread
in DIF_STDRET across quintiles, where the formal test is on the difference in

Table 2. Evidence From Dual-Class Firms.
Panel A: Sample Composition.

Number of firms 59
Number of share classes 118
Firm-month pairs 7,322
Panel B: Descriptive Statistics.

M 10% 25% 50% 75% 90%
MV 1,789 11 53 148 419 2,212
DIF_VOLUME 4,546.5% 16.4% 48.3% 144.6% 522.5% 2,361.9%
DIF_STDRET 20.9%*** 237.3% 216.5% 2.6%*** 28.5% 69.5%
RETCORR 0.790 0.500 0.666 0.852 0.932 0.978
Panel C: Difference in Trading Volume and Stock Volatility.

M (%) Median (%)
Quintiles by DIF_VOLUME
(sorted at the firm level) DIF_VOLUME DIF_STDRET DIF_VOLUME DIF_STDRET
1 127.0 10.9*** 17.2 0.0
2 381.3 10.0*** 59.2 2.0**
3 735.5 15.5** 121.2 2.4***
4 1,584.0 16.0*** 254.6 3.7***
5 19982.8 26.4*** 743.6 4.2***
Q5 2 Q1 Diff. 11855.8 15.5*** 726.1 4.2***
Note. This table reports results of analyses on dual-class firms. Panel A reports sample composition of dual-class
stocks listed on the same major U.S. exchanges (i.e., NYSE, Nasdaq, AMEX, or NYSE Arca) during 1965-2009.
Panel B reports descriptive statistics. MV is the market value of equity ($million). For each firm-month, shares in a
pair are split into high and low groups based on their corresponding trading volume. DIF_VOLUME is the trading
volume difference between the high- and the low-volume issues, scaled by the volume of the low-volume issue.
Trading volume is the sum of daily trading volume in a month, calculated as volume divided by shares outstanding.
DIF_STDRET is the stock volatility difference between the high- and the low-volume issues, scaled by the stock
volatility of the low-volume issue. Stock volatility is measured as the standard deviation of daily returns in a month.
RETCORR is the within-pair return (Pearson) correlation between the two classes of shares of the same firm. Panel
C reports mean and median DIF_VOLUME and DIF_STDRET across DIF_VOLUME quintiles sorted at the firm level.
The p value for the difference between the top and bottom quintiles (Q5 2 Q1 Diff.) is based on z statistics for
medians and t statistics for means.
DIF_STDRET means and medians between the two extreme quintiles. The results reveal a
strong and generally monotonic positive relation between DIF_VOLUME and
DIF_STDRET, where the difference in medians between the extreme quintiles is 4.2% and
difference in means is 15.5%, both highly statistically significant. In addition, the mean
DIF_STDRET in each volume quintile is positive and statistically significant. Except in the
lowest DIF_VOLUME quintile, all the medians of DIF_STDRET are significantly positive.
Summarizing, the results for dual-class shares are largely consistent with the results for
ETFs, and suggest that controlling for fundamentals higher volume is associated with
higher volatility. Compared with the ETF setting, the dual-class setting reveals a much
stronger positive relation between volume and volatility. This is consistent with the much

Dichev et al. 503
lower pairwise return correlations observed for dual-class pairs than for ETF index pairs
and is likely a reflection of less room for arbitrage.10
Large-Sample Evidence for U.S. Stocks

As previously discussed, the thorniest issue in investigating the hypothesized relation
between trading volume and stock volatility is how to control for information flow. This is
problematic because information flow follows a multitude of public and private channels
and is thus difficult to observe and measure. The preceding section relies on the two natural
experiments that aim to control for information flow and to establish the existence and the
direction of the trading/volatility relation. The disadvantage of these settings, however, is
that by definition they are fairly specialized and limited, and thus there is a question about
the generalizability and portability of these findings (especially their magnitudes) to the
wider world of stock trading. In this section, we address this question by extending the
investigation to the full universe of U.S. stocks. Generally speaking, this extended investi-
gation is more realistic and is well-calibrated to the naturally occurring properties of the
U.S. stock market; this benefit, however, comes at the cost of losing some of the sharp con-
trols in the earlier specifications.
Evidence From the Full Time-Series of U.S. Stocks

The first type of evidence for the broad stock market looks at the long-run record for a
sample of the 500 largest U.S. stocks over 1926-2012. We use 500 stocks because data
availability is limited to about this number in the early years of the sample, and we want to
preserve some measure of comparability over time. The evidence for this specification is
presented in Figure 2 and Table 3, based on annual observations of value-weighted turn-
over and stock return volatility. An inspection of Figure 2 reveals that the evolution of
volatility has a perceptible synchronicity with the broad ebbs and flows of trading volume.
When trading is lowest in the quiet years between 1940 and 1970, volatility is also lowest,
never exceeding 2% (daily measure) over this extended period that includes World War II,
the Korean War, and the various upheavals of the Cold War. Volatility is the highest
during the two periods with the most intense trading, peaking at over 4% during 1926-1940
and with the second and third highest peak occurring after the mid-1990s. To be sure, the
relation is far from lock-step and one can identify several instances where it is inadequate
to describe the empirical behavior of volatility, for example, volatility spikes during the
recession of 1973-1974 with no discernible change in volume of trading. The summary
impression from Figure 2, however, is that even at this broad-brush graphical level volume
of trading and volatility are substantially positively related. This impression is confirmed
by the statistical test in Table 3, with a Spearman correlation of .52 between these two vari-
ables, which is statistically significant and economically substantial. We also explore a
change specification as levels tests are often subject to confounding influences of other
variables, and a simple way to control for these confounding effects is to conduct the same
test in the innovations of the variables of interest. The Spearman correlation between the
time-series changes in volume and volatility is .35 in Table 3, again with high statistical
and economic significance, confirming the results in levels.
By necessity, the evidence from the long-run sample is limited because we lack data to
control for fundamental information, for example, earnings data are only available since
the 1960s. However, we provide one additional analysis that helps to sharpen the long-run

Table 3. Trading Volume and Stock Volatility for Largest 500 U.S. Stocks From 1926 to 2012.
Year VOLUME STDRET Year VOLUME STDRET Year VOLUME STDRET

1926 114.6% 0.014 1955 11.9% 0.014 1984 58.9% 0.016
1927 113.1% 0.013 1956 9.6% 0.012 1985 66.4% 0.014
1928 147.4% 0.016 1957 8.8% 0.013 1986 75.4% 0.017
1929 130.4% 0.029 1958 9.8% 0.012 1987 85.9% 0.026
1930 77.4% 0.023 1959 9.6% 0.013 1988 67.5% 0.016
1931 60.6% 0.032 1960 8.5% 0.014 1989 65.7% 0.014
1932 52.4% 0.043 1961 9.6% 0.013 1990 55.8% 0.017
1933 61.4% 0.037 1962 10.1% 0.017 1991 55.8% 0.016
1934 24.2% 0.021 1963 10.0% 0.011 1992 54.7% 0.015
1935 24.7% 0.018 1964 9.3% 0.010 1993 62.1% 0.015
1936 26.2% 0.016 1965 10.1% 0.011 1994 63.8% 0.015
1937 22.9% 0.024 1966 14.8% 0.015 1995 69.9% 0.014
1938 18.1% 0.024 1967 16.0% 0.014 1996 72.5% 0.016
1939 15.7% 0.020 1968 17.0% 0.015 1997 79.4% 0.019
1940 11.6% 0.018 1969 15.6% 0.015 1998 82.9% 0.023
1941 9.2% 0.014 1970 15.3% 0.018 1999 93.3% 0.024
1942 7.1% 0.014 1971 17.2% 0.014 2000 111.9% 0.031
1943 11.9% 0.012 1972 16.2% 0.013 2001 110.9% 0.024
1944 10.7% 0.010 1973 16.0% 0.019 2002 129.3% 0.026
1945 14.4% 0.012 1974 14.6% 0.023 2003 124.7% 0.016
1946 14.7% 0.018 1975 18.6% 0.018 2004 124.2% 0.013
1947 10.0% 0.013 1976 22.8% 0.014 2005 138.0% 0.013
1948 11.5% 0.013 1977 20.5% 0.012 2006 159.8% 0.013
1949 9.3% 0.011 1978 24.9% 0.014 2007 224.2% 0.016
1950 16.4% 0.013 1979 28.3% 0.014 2008 360.5% 0.037
1951 12.5% 0.011 1980 38.4% 0.020 2009 277.3% 0.028
1952 9.2% 0.010 1981 36.3% 0.018 2010 220.6% 0.017
1953 9.0% 0.010 1982 49.8% 0.020 2011 201.3% 0.020
1954 12.0% 0.011 1983 57.0% 0.017 2012 159.2% 0.013
Note. This table reports trading volume and stock volatility for the largest 500 stocks on NYSE/AMEX from 1926
to 2012. VOLUME and STDRET are defined in Figure 2. CH_VOLUME (CH_STDRET) is the difference of VOLUME
(STDRET) in year t and t21. Spearman Correlation (VOLUME, STDRET) = .520***; Spearman Correlation
(CH_VOLUME, CH_STDRET) = .348***.
evidence, and is perhaps the most direct evidence that high volumes of trading induce
noise in stock returns. This analysis is based on the idea that noise in stock returns eventu-
ally has to revert, and thus in the presence of noise long-window stock returns will be less
volatile than short-window stock returns. The major difficulty in implementing this idea is
deciding on the horizon of noise reversals, and here we use the technology and results in
French and Roll (1986) as a guide. Specifically, we construct a ratio of Actual/Implied
volatility for our sample at weekly and monthly horizons. The Actual volatility in the
numerator is the standard deviation of weekly and monthly returns measured over each cal-
endar year. The Implied volatility in the denominator is the hypothetical weekly and
monthly volatility implied by daily volatility assuming serial independence of returns, that
is, the standard deviation of daily returns over a year multiplied by the square root of the
number of trading days in a week or a month. The resulting Actual/Implied ratio has some

Dichev et al. 505
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
1926
1928
1930
1932
1934
1936
1938
1940
1942
1944
1946
1948
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
VOLUME STDRET
Figure 2. Trading volume and stock volatility for largest 500 U.S. stocks from 1926 to 2012.
Note. This figure shows trading volume (solid line) and stock volatility (dotted line) for the largest 500 stocks on
NYSE/AMEX from 1926 to 2012. VOLUME is annualized value-weighted trading volume as defined in Figure 1 with
the exception that the calculation is based on the largest 500 U.S. stocks. STDRET is the value-weighted stock
return volatility (multiplied by 50 for scaling), measured as the sum of return volatility for each of the 500 stocks
multiplied by its corresponding weight. Return volatility for stock i in year t is the standard deviation of its daily
returns in year t and weight for stock i in year t is the average of its beginning and ending market values in year t
divided by the total market values of the 500 stocks in year t.
nice properties and intuitive appeal. Under the null of no noise, which means no negative
autocorrelation of returns, this ratio should be close to one, and the magnitude of deviation
from this null indicates the magnitude of trading noise.
The results for the Actual/Implied specification are presented in Figure 3 (means across
our 500 firms) in two lines corresponding to the weekly and monthly horizons of noise
reversals. In addition to the jagged lines linking the actual observations, we also present the
same results after a second-order polynomial smoothing. An examination of Figure 3
reveals that the Actual/Implied ratio is mostly less than one, and thus indicates the presence
of negative autocorrelations in returns and therefore trading noise. In addition, the graph
reveals a distinct inverted-U shape over time, that is, the ratio approaches its peak and the
theoretical ideal of one in the low-volume middle years of the sample, and drops away
from that level in the high-volume early and late years in the sample. Note that this
inverted-U shape in Figure 3 is precisely the opposite of the U-shape observed for volume
in Figure 2. The implication is that high volumes of trading induce trading noise that
makes short-horizon returns considerably more volatile than long-horizon returns. The mag-
nitudes of the Actual/Implied ratio also allow an estimate of the amount of trading noise in
short-horizon returns. Using the estimates from the smoothed monthly line, trading noise
accounts for as much as 15% to 25% of the volatility of daily returns in the early years of
the sample and 10% to 15% in the late years.11

1.15
1.05
Actual/Implied Ratio
0.95
0.85
0.75
0.65
1926
1928
1930
1932
1934
1936
1938
1940
1942
1944
1946
1948
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
Weekly Monthly
Figure 3. Actual/Implied ratios for largest 500 U.S. stocks from 1926 to 2012.
Note. This figure shows the mean weekly Actual/Implied ratios (solid line) and monthly Actual/Implied ratios
(dotted line) for the largest 500 stocks on NYSE/AMEX from 1926 to 2012. Weekly (Monthly) Actual/Implied
ratio for stock i in year t is the actual weekly (monthly) return volatility divided by the implied weekly (monthly)
return volatility. Actual weekly (monthly) return volatility for stock i in year t is the standard deviation of weekly
(monthly) returns of stock i in year t. Implied weekly (monthly) stock volatility for stock i in year t is the standard
deviation of daily returns of stock i in year t multiplied by the square root of the number of trading days in a week
(month). The smooth trend lines are obtained from the second-order polynomial function.
Evidence From the Cross-Section of U.S. Stocks Over the Last 20 Years
The second type of evidence for the broad stock market is based on the cross-sectional var-
iation in trading intensity during recent years; specifically, we use a sample of all NYSE-
AMEX stocks over 1988-2007. Nasdaq stocks are excluded because of the well-known
problems in accurately measuring trading volume on Nasdaq (see Anderson & Dyl, 2005,
for a full account).12 We start with a simple specification that examines the univariate rela-
tion between volume and volatility. Stocks are sorted annually into deciles based on their
annualized daily trading volume, and we report median turnover and volatility by decile in
Panel A of Table 4. An inspection of Panel A reveals that there is a substantial cross-sec-
tional variation in turnover, with a low of about 10% for the bottom decile and a high of
235% for the top decile. There is also a substantial spread in volatility between the extreme
deciles, from about 2% (daily volatility) in the bottom decile to about 3% in the top decile,
which is both statistically significant and economically substantial. A closer look at the
results also reveals that this increase is not monotonic, and indeed there is little reliable
variation in volatility from the first decile until about the seventh decile, followed by a
quick rise and hitting a high in the top decile. The combined impression from these obser-
vations is that while the relation between volume and volatility is generally positive, it is
also decidedly non-linear, with volatility only clearly rising in the extremes of high trading.

Table 4. Stock Volatility Across Trading Volume Portfolios.
Panel A: Stock Volatility Formed on the Basis of Trading Volume.

D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 D10 2 D1
VOLUME deciles (low to high)
0.097 0.224 0.350 0.473 0.592 0.707 0.842 1.049 1.364 2.448 2.351***
STOCK VOLATILITY
0.020 0.021 0.020 0.019 0.019 0.019 0.020 0.022 0.025 0.031 0.011***
Panel B: Stock Volatility Formed on the Basis of Both Earnings Volatility and Trading Volume.
VOLUME deciles (low to high)
D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 D102D1
0.097 0.222 0.347 0.465 0.585 0.710 0.860 1.076 1.393 2.350 2.250***
STDEARN deciles (low to high) D1 0.001 0.014 0.015 0.015 0.015 0.015 0.016 0.016 0.018 0.018 0.021 0.007***
D2 0.003 0.015 0.016 0.014 0.015 0.015 0.015 0.016 0.018 0.019 0.023 0.008***
D3 0.004 0.017 0.017 0.017 0.016 0.016 0.017 0.018 0.019 0.021 0.024 0.007***
D4 0.006 0.018 0.019 0.018 0.017 0.017 0.017 0.018 0.020 0.021 0.026 0.008***
D5 0.008 0.019 0.019 0.018 0.018 0.018 0.019 0.020 0.021 0.024 0.027 0.008***
D6 0.011 0.020 0.021 0.020 0.019 0.018 0.019 0.021 0.022 0.025 0.028 0.008***
D7 0.014 0.022 0.022 0.021 0.022 0.020 0.021 0.023 0.024 0.026 0.030 0.008***
D8 0.021 0.025 0.026 0.025 0.024 0.024 0.024 0.023 0.026 0.028 0.033 0.008***
D9 0.036 0.028 0.033 0.031 0.029 0.029 0.028 0.028 0.029 0.031 0.036 0.008***

D10 0.096 0.037 0.043 0.042 0.047 0.040 0.041 0.043 0.040 0.039 0.046 0.009***
D12D10 0.095*** 0.023*** 0.027*** 0.027*** 0.032*** 0.025*** 0.025*** 0.026*** 0.023*** 0.021*** 0.024***
Note. This table reports median stock volatility for portfolios formed on the basis of trading volume (both trading volume and earnings volatility) in Panel A (Panel B). The
sample consists of 40,577 firm-year observations from all NYSE/AMEX stocks over 1988-2007 with available CRSP and Compustat data to calculate trading volume, stock
volatility, and earnings volatility at the firm-year level. Trading volume (VOLUME) is the annualized trading volume, calculated as average daily trading volume (volume/shares
outstanding) multiplied by 250 for firm i in year t. Stock volatility (STDRET) is the standard deviation of daily stock returns for firm i in year t. Earnings volatility (STDEARN) is
the standard deviation of quarterly earnings for firm i over years t, t + 1, and t + 2. Quarterly earnings are earnings before extraordinary items, scaled by the average of
beginning and ending total assets (Compustat data8/data44). In Panel A, all sample firms are sorted into deciles each year based on trading volume. In Panel B, all sample firms
are sorted into deciles each year based on earnings volatility and within each earnings volatility decile firms are further sorted into deciles based on trading volume. The p value
of the difference of medians between the top and bottom deciles (D10 2 D1 Diff.) is based on Wilcoxon z statistics.
507
Of course, the simple analysis in Panel A is inadequate because it does not control for
variation in volatility related to fundamentals. Broadly speaking, stock volatility due to fun-
damentals can come from two sources, changes in expectations about future cash flows and
changes in the discount rate. We make no formal attempt to control for discount rate
changes because our volatility observations are at the firm-year level, while the empirical
variation in discount rates within a year is likely small; in addition, discount rates are
notoriously difficult to measure. We control for changes in expectations about future cash
flows by using realized earnings variability over current and future periods as a proxy; spe-
cifically, for any firm i and year t, we use the standard deviation of realized quarterly earn-
ings over the current and two future years (i.e., years t, t + 1, and t + 2). Earnings are
defined as earnings before extraordinary items, scaled by the average of beginning and
ending total assets, where earnings and asset data are from Compustat. To explore for pos-
sible non-linearities in the examined relations, we rely on a portfolio specification to map
out the relation between trading volume and stock volatility, controlling for fundamentals-
based volatility. Specifically, we first sort the sample annually on fundamentals variability
into deciles, and then within these portfolios sort on volume into deciles. The result is a 10
3 10 matrix in Panel B of Table 4, with each cell reporting median stock volatility for that
portfolio; variation down the columns captures the effect of fundamental variability on
stock volatility, and variation across the columns captures the effect of trading volume on
stock volatility, controlling for fundamental variability.
An examination of the results in Panel B reveals that fundamental variability is the pri-
mary driver of observed stock volatility. The bottom line in Panel B captures differences in
the extreme deciles down the columns; while these differences vary, they average 2.5%
(daily volatility). This magnitude clearly dominates the corresponding numbers for the
effect of trading volume, captured in the extreme-right column, which average about 0.8%.
Of course, the dominance of fundamental variability is not surprising; in fact, in an effi-
cient market fundamental variability should be the only variable that affects stock volatility.
What is more remarkable, actually, is that the effect of trading intensity remains economi-
cally large after controlling for fundamental variability. If one thinks of total stock volati-
lity as the sum of volatility due to fundamental variability and volatility due to trading
intensity, using the averages above suggests that differences in trading intensity account for
about a quarter of total stock volatility, 0.8%/(0.8% + 2.5%), a rather significant amount.
A closer look at the results in Panel B also reveals the same non-linear pattern in the trad-
ing/volatility relation first observed for the univariate specification in Panel A. Moving
across columns, there is little reliable variation in volatility from column 1 until about
column 7, and then a clear and pronounced increase over the remaining columns, always
hitting a high in column 10.
We extend the analysis of the cross-sectional relation between volatility and volume
using a multivariate regression. The advantage of the regression specification is that it
allows for simultaneous control for a number of variables that have been shown to be deter-
minants of stock volatility. The disadvantage is the normality and linearity assumptions,
which are likely violated in this setting. We make appropriate adjustments to variable and
regression specification to overcome these limitations. Specifically, for the period 1988-
2007, we estimate coefficients in the following regression:
STDRETi, t = b0 + b1 HIGHi, t + b2 VOLUMEi, t + b3 VOLUME3HIGHi, t + b4 STDRETi, t 1

+ b5 RETi, t + b6 STDEARNi, t + 2 + b7 SIZEi, t 1 + b8 AGEi, t 1
+ b9 LEVERAGEi, t 1 + b10 BTMi, t 1
+ b11 ROEi, t 1 + b12 DIVIDENDi, t 1 + ei, t,

Dichev et al. 509
Table 5. Cross-Sectional Relation Between Stock Volatility and Trading Volume, Controlling for
Other Factors.
STDRETi, t = b0 + b1 HIGHi, t + b2 VOLUMEi, t + b3 VOLUME3HIGHi, t + b4 STDRETi, t 1
+ b5 RETi, t + b6 STDEARNi, t + 2 + b7 SIZEi, t 1 + b8 AGEi, t 1 + b9 LEVERAGEi, t 1
+ b10 BTMi, t 1 + b11 ROEi, t 1 + b12 DIVIDENDi, t 1 + ei, t :
Pred. sign (1) (2) (3) (4) (5)

Intercept + 38.366 44.526 69.993 25.771 25.133
(45.47) (54.65) (37.64) (18.51) (6.33)
HIGH 2 269.784 229.114 219.165 212.880
(221.70) (26.86) (27.52) (21.89)
VOLUME + 0.225 0.012 0.227 0.105 0.128
(13.20) (0.66) (21.56) (10.28) (6.20)
VOLUME 3 HIGH + 1.004 0.396 0.257 0.157
(25.46) (7.42) (8.41) (1.86)
LAGGED STDRET + 0.626 0.631
(61.66) (16.98)
RET 2 20.090 20.068 20.115
(23.60) (23.55) (24.84)
STDEARN + 0.189 0.086 0.097
(19.09) (15.16) (8.95)
SIZE 2 20.319 20.135 20.120
(215.28) (27.46) (24.32)
AGE 2 20.068 20.021 20.023
(26.01) (22.06) (22.18)
LEVERAGE + 20.021 0.001 0.011
(22.29) (0.25) (1.67)
BTM 2 20.079 20.031 20.025
(26.94) (24.42) (21.60)
ROE 2 20.065 20.045 20.034
(26.57) (26.41) (22.49)
DIVIDEND 2 216.922 24.370 24.866
(222.37) (29.86) (24.24)
(Average) number of observations 1,913 1,913 1,913 1,913 38,262
(Average) adjusted R2 .055 .096 .566 .758 .712
Note. This table reports cross-sectional regressions of stock volatility on trading volume along with control vari-
ables at the firm level for all available NYSE/AMEX firms over the period 1988-2007. STDRETt is the standard devia-
tion of returns for stock i in year t. VOLUME is the annualized trading volume. HIGH is an indicator variable coded
as 1 if volume is in the top quartile of the sample. STDRETt21 is the standard deviation of returns for stock i in
year t21. RET is the compounded daily return for stock i in year t. STDEARN is the standard deviation of quarterly
earnings scaled by average total assets (Compustat data8/data44) over years t, t + 1, and t + 2 with a minimum
requirement of eight quarters. SIZE is proxied by the market value of common equity, computed by multiplying the
common stock price at fiscal year-end (data199) by common shares outstanding (data25). AGE is the number of
years since the firm first appears in the CRSP database. LEVERAGE is the ratio of debt to assets (data9 + data34)
/ (data6). BTM is the book equity to market equity ratio. Book equity is constructed as stockholders’ equity plus
balance sheet deferred taxes and investment tax credit (data35) minus the book value of preferred stock.
Depending on availability, stockholder’s equity is computed as data216 or data60 + data130 or data6 2 data181,
in that order, and preferred stock is computed as data56 or data10 or data130, in that order. ROE is the earnings
to book equity ratio. Earnings are calculated as income before extraordinary items, available to common stock-
holders (data237), plus deferred taxes from the income statement (data50), plus investment tax credit (data51).
DIVIDEND is an indicator variable equal to one if the firm pays dividend in that year, zero otherwise. Regressions
(1) through (4) report the estimates from Fama–MacBeth cross-sectional regressions. The t values in parentheses
are based on Fama–MacBeth standard errors and the number of observations and R2 are the averages across the
20 annual regressions. Regression (5) reports estimates from pooled-cross-sectional regression to gauge the effect
of increasing volume over time on stock volatility. The t values reported in parentheses are based on standard
errors clustered by firm and year as suggested by Petersen (2009). To control for non-normalities in their distribu-
tions and to allow for direct comparison of their strength across variables, all variables in regressions (1) through
(4) are ranked into percentiles by year and all variables in regression (5) are ranked into percentiles without sort-
ing by year.

where STDRET is the standard deviation of daily stock returns for firm i, HIGH is an indi-
cator variable set to 1 if volume is in the top quartile in year t, VOLUME is the annualized
trading volume, RET is the compounded daily return in year t. STDEARN is the standard
deviation of quarterly earnings scaled by the average total assets over years t, t + 1, and t
+ 2 with a minimum requirement of eight quarters. SIZE is proxied by the market value
of common equity, AGE is the number of years since the firm first appeared in the CRSP
database, LEVERAGE is the ratio of debt to assets, BTM is the book to market ratio, ROE
is earnings to equity ratio, and DIVIDEND indicates whether the firm pays a dividend.
We introduce the HIGH variable to account for the convex relation between volatility
and volume shown in Table 4; thus, we expect a positive sign on HIGH 3 VOLUME.
Control variables are from Pastor and Veronesi (2003) and Wei and Zhang (2006). Briefly,
lagged value of STDRET is included because volatility is known to be positively autocorre-
lated, and essentially as a catch-all variable that captures omitted variables and other mis-
specifications. Contemporaneous return is included because expected return and risk are
positively correlated (Sharpe, 1964), and so are their realizations. As above, STDEARN
controls for volatility related to fundamentals and information flow, we expect a positive
sign. The rest of the variables are commonly found in asset pricing tests, and the predicted
signs are clear. We replace the original values of all continuous variables with their percen-
tile ranks to control for non-normalities in their distributions and to allow for direct com-
parison of their strength across variables. Thus, these regression coefficients can be
interpreted as the percentage change in volatility for 1% change in the corresponding vari-
able (controlling for all other variables).
The regression results are presented in Table 5, where regressions (1) through (4) use a
Fama–MacBeth specification to control for cross-sectional dependencies in the residuals.
We start with baseline specifications (1) and (2), which include only VOLUME and then
VOLUME interacted with HIGH. Consistent with the results in Table 4, regression (1) con-
firms that there is a positive relation between volatility and volume, while the positive and
significant coefficient on VOLUME 3 HIGH in regression (2) clarifies that this relation is
convex, that is, it is much stronger for high levels of volume. We expand these results to
regression (3), which includes all control variables except the catch-all lagged STDRET. An
inspection of regression (3) reveals that the relation between volatility and volume remains
statistically significant and economically substantial after the controls, with sizable coeffi-
cients on both VOLUME and VOLUME 3 HIGH. In fact, the coefficient on VOLUME 3
HIGH is larger than that of any other variable (except for the non-comparable DIVIDEND
dummy), dominating even the coefficients on SIZE and STDEARN.
The addition of Lagged STDRET in regression (4) leads to increased R2 and decreased
coefficients and significance on all other variables, consistent with the catch-all nature of
this variable. The coefficient on VOLUME 3 HIGH, however, remains larger than any
other continuous variable except Lagged STDRET, confirming the strength of the volume-
volatility effect at high volumes of trading. A disadvantage of the Fama–MacBeth specifi-
cation in regressions (1) to (4) is that it essentially assumes time-series stationarity in the
volume/volatility relation, and ignores much of the meaningful increase in volume over
time. We address this limitation in regression (5), which uses a panel specification with
standard errors clustered by firm and year as suggested by Petersen (2009). Compared with
the coefficient in the main specification (4), the coefficient on VOLUME 3 HIGH in
regression (5) is much smaller but remains significant at the 10% level, confirming that
these findings are reliable.

Dichev et al. 511
Table 6. Granger Causality Test of Trading Volume and Stock Volatility: Time-Series From 1926 to
2012.
Panel A: Trading Volume Causes Stock Volatility (N = 1,043).

Full model Restricted model
STDRETt STDRETt
Estimate t statistic Estimate t statistic F statistic p value
Intercept 0.003 10.08 0.003 10.05 8.87 .003

STDRETt21 0.770 37.81 0.796 41.87
VOLUMEt21 0.008 2.98
Adjusted R2 .631 .628
Panel B: Stock Volatility Causes Trading Volume (N = 1,043).

Full model Restricted model
VOLUMEt VOLUMEt
Estimate t statistic Estimate t statistic F statistic p value
Intercept 0.008 6.36 0.002 3.12 30.50 \.001

VOLUMEt21 0.978 94.51 0.955 99.71
STDRETt21 20.417 25.52
Adjusted R2 .911 .910
Note. This table reports Granger causality tests of stock volatility and trading volume using aggregate value-
weighted trading volume and stock volatility based on the largest 500 stocks on NYSE/AMEX in each month over
1926-2012. Panel A (B) presents results of a test whether VOLUME (STDRET) causes STDRET (VOLUME), using one
lag of each variable. VOLUME is the sum of daily value-weighted trading volume for month t, measured as dollar-
value traded (volume 3 price) on a trading day divided by aggregate market value (price 3 shrout) outstanding as
of that day for the largest 500 stocks. STDRET is the value-weighted average of stock volatility for month t,
measured as the sum of stock volatility for each of the 500 stocks multiplied by its corresponding weight. Stock
volatility for stock i is the standard deviation of daily stock returns in month t and weight for stock i is the
average of its beginning and ending market values in month t divided by the total market values of the 500 stocks
in month t.
Extensions and Robustness Checks

We perform a number of extensions and robustness checks for the large-sample results; for
parsimony, these results are discussed only briefly. First, there is a concern that as both
volume and volatility are endogenously determined, the hypothesized causality in the
volume/volatility relation may run in the opposite direction; specifically, the concern is that
environments of high stock volatility are also those with the highest potential for specula-
tive profits, and thus attract more traders and trading (Han & Kumar, 2013). Although this
story is appealing, it omits the consideration of opposing forces, namely as uncertainty and
volatility go up, market-makers widen the spreads to protect themselves against informed
trading, which kills volume. In any case, to further sort out these alternative stories, we per-
form Granger causality tests in our time-series sample and report the results in Table 6.
Panel A reveals that after controlling for lagged stock volatility, current stock volatility is
significantly positively related to lagged trading volume (coefficient of 0.008 with t statistic
of 2.98) but the incremental R2 is very small. In contrast, Panel B shows that after

controlling for lagged volume, current volume is significantly negatively related to lagged
stock volatility (coefficient of 20.417, t statistic of 25.52). Overall, the Granger causality
evidence suggests that volume drives up volatility (F statistic of 8.87) while volatility
drives down volume (F statistic of 30.50), which is inconsistent with the reverse-causality
conjecture.13
Another direction in which we extend the results is implementing the Actual/Implied
volatility specification used in the time-series analysis (in ‘‘Evidence From the Full Time-
Series of U.S. Stocks’’ section) for the cross-section of stocks (in ‘‘Evidence From the
Cross-Section of U.S. Stocks Over the Last 20 Years’’ section), where the expectation is to
find lower Actual/Implied ratios for the stocks with the highest trading volume. The cross-
sectional results yield a more complicated and nuanced picture; in fact, we find that the
Actual/Implied ratio increases with volumes of trading for smaller stocks and for most
years. However, the Actual/Implied ratio decreases with volumes of trading for larger
stocks and when the most recent years are included. These findings are consistent with the
earlier conjecture about the non-monotonic benefits of trading. Taken as a whole, the
results imply that increased volumes of trading reduce trading noise and are beneficial for
smaller stocks and for comparatively low levels of trading. But after a certain point, this
relation reverses and higher volumes of trading lead to more trading noise for large stocks
and for the most recent years. One caveat in interpreting these results is that the behavior
of trading noise and its reversals are more complicated and cover longer horizons than the
ones considered here. For example, momentum is a continuation of existing return trends at
intermediate horizons (up to a year), and thus any reversals of momentum ‘‘noise’’ have to
happen at rather long horizons, much longer than the weekly and monthly horizons consid-
ered here. We leave a more comprehensive investigation of the parameters and horizons of
noise reversals for future research.
As another implication of our results, we also explore for a ‘‘gone fishing’’ reduction in
summer volatility. Hong and Yu (2009) document a ‘‘gone fishing’’ effect in trading activ-
ity, where stock turnover is significantly lower during summer vacation months (July-
September) as compared with the rest of the year. We use this finding as providing a natu-
ral setting for exogenous variation in trading volume, and investigate whether the lower
trading volumes in summer leads to lower stock volatility as well. We first confirm that
volume of trading is lower during summer months; specifically, this pattern exists in 64 out
of 87 years for our long-term sample. Then, we find that stock volatility is also lower
during summer months as opposed to the rest of the year, in 54 out of 87 years. Finally, we
take the differences in summer/non-summer volume and volatility, and document a
Spearman correlation of .501 between them, which is highly statistically significant and
economically large. Summarizing, seasonal changes in volume of trading are reliably asso-
ciated with seasonal changes in stock volatility, consistent with the main results presented
earlier.
Discussion of Results
Although the results in this article span a number of specifications and offer many nuances,
they converge on some key themes. We find economically substantial evidence that, con-
trolling for fundamentals, more intensive stock trading is accompanied by increased return
volatility. This relation is weak to non-existent at low to moderate levels of trading but
becomes increasingly strong as volume of trading increases. Furthermore, our results on the
Actual/Implied volatility ratio (Figure 3) provide specific evidence that high levels of

Dichev et al. 513
trading lead to temporary swings of stock prices away from fundamentals, that is, high trad-
ing volumes correspond to Actual/Implied ratios trending down from the benchmark value
of one, indicating return reversals. The combined impression from these results is that high
volumes of trading inject a sizable layer of volatility over and above the unavoidable fun-
damentals-based volatility.
Two recent studies offer evidence that is largely in line with our findings. Foucalt,
Sraer, and Thesmar (2011) explores the effects of a reform in the French stock market that
triggers a drop in retail trading activity, and find that the daily return volatility of stocks
falls by 20 basis points. This evidence suggests that (noise) traders indeed affect the volati-
lity of stock returns. Zhang (2010) investigates the effect of high-frequency trading on
price discovery, and finds that it has some harmful effects, including inducing higher vola-
tility in stocks. Although these findings have more specialized motivations and methodolo-
gies, the agreement in the results provides further confidence in our more general findings,
which quantify the impact of the radically increasing trading volume in the U.S. stock
market on stock return volatility.
In considering the larger meaning of these results, it is useful to remember that existing
research documents a number of benefits from trading (Brennan et al., 1998; Chordia &
Swaminathan, 2000; Fang, Noe, & Tice, 2009). There is reliable evidence that traded assets
command higher valuations, lower transaction costs, and wider investor recognition, and
that these benefits increase with higher levels of trading. To be able to reconcile the dispa-
rate messages of this study and existing research, note that much of the previously docu-
mented benefits of liquidity come from environments with low trading intensity, for
example, newly listed stocks experience a substantial increase in price and decrease in bid-
ask spread (Kadlec & McConnell, 1994). In contrast, the evidence in this study comes from
the largest, most-traded environments and stocks of all time; generally, we examine promi-
nent companies on the major U.S. stock exchanges, mostly during the unprecedented surge
in trading activity over the last 20 years.
The results in this study provide an example of the new problems that start appearing
with the intensification of trading. Higher levels of trading can generate their own volati-
lity, with all ensuing consequences, including possible shifts in investor risk preferences
and risk management behavior, and possible destabilization of the market. At this point,
these possibilities are just conjectures, and it will be useful to explore them further in
future research. For example, it will be interesting to examine more closely the origin and
dynamics of trading-induced volatility and compare them with what we know about funda-
mentals-induced volatility.
A related question is whether market-makers and regulators need to be more cognizant
and proactive about the fact that high trading leads to high volatility. To a certain extent,
such reactions already exist, for example, circuit-breakers dampen extreme price moves
by halting trading, which is essentially a forced and extreme reversal of the forces docu-
mented in this study. There are also other ideas about possible reactions, and some of
them have a long history. For example, Ripley (1911) reviews a massive wave of specula-
tion in major U.S. railroad stocks at the turn of the last century, where annual turnover
for several stocks reached magnitudes of 10 to 20. Ripley suggests that one way to
dampen such speculative excesses is to impose taxes on trading, with the side benefit of
raising government funds. Similar ideas are developed in Summers and Summers (1989),
who argue that imposing a small security transaction tax will curb speculation and reduce
the diversion of resources into the financial sector of the economy. Most recently, various
forms of security transaction taxes have been seriously discussed in the European Union

(Steinhauser & Olsen, 2012). Currently, transaction tax ideas remain controversial, and
there is justified need to be cautious because the cure may prove to be worse than the
ailment.
A possible direction for research is to investigate the volume/volatility relation in invest-
ment assets beyond stocks. Corporate and government bonds, closed-end funds, commodi-
ties, currencies—all these instruments provide potential testing ground for the effects
documented here. Currency trading, for example, has grown 10-fold during the last 20
years, and today at US$4 trillion/day is arguably much higher than needs tied to the real
economy (e.g., total annual global trade is US$25 trillion and global money stock is only
US$12 trillion as of 2009).14 Finally, it is useful to consider the implications of the trading/
volatility results for other investor environments and stock exchanges.
Conclusion
This study investigates the effect of high trading volume on observed stock volatility con-
trolling for fundamental information. The motivation is that volumes of U.S. trading have
increased more than 20-fold over the last 50 years, truly transforming the marketplace, and
it is important to map out the effects of such a momentous change. First, we use matched
ETFs and dual-class stocks as settings with substantial variation in trading but good natural
controls for the underlying fundamentals. Our main finding is that in both settings
increased volume of trading triggers a small but reliable increase in stock volatility.
Second, we examine the aggregate time-series of U.S. stocks since 1926 and the cross-sec-
tion of stocks during the last 20 years to better calibrate the economic parameters of the
identified relation. Using annual measures, volume and volatility are correlated on the mag-
nitude of 30% to 50% in the aggregate time-series, suggesting that much of the historical
variation in volatility is driven by the prevailing volumes of trading. Tests in the cross-sec-
tion not only confirm the positive volume/volatility relation but also reveal a pronounced
convexity, where the relation is weak to non-existent for low levels of trading and becomes
much clearer and stronger for high levels of trading. Efforts quantifying the volume effect
reveal that trading-induced volatility accounts for about a quarter of total observed stock
volatility today. The combined impression from these results is that stock trading injects an
economically substantial layer of volatility above and beyond that based on fundamentals,
especially at high levels of trading.
Acknowledgment
We appreciate the helpful comments of workshop participants at Yale University, University of
Chicago, Duke University, University of Southern California, Florida State University, Southern
Methodist University, Wharton, Washington University, Norwegian School of Economics and
Business Administration (NHH), and especially those of Theo Sougiannis (the editor), an anonymous
referee, Linda Bamber, Tarun Chordia, Larry Harris, Jonathan Karpoff, Feng Li, Lasse Pedersen, and
Catherine Schrand.
Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/
or publication of this article.

Dichev et al. 515
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or
publication of this article: The authors gratefully acknowledge financial support from the Goizueta
Business School at Emory University and the College of Business at Florida International University.
Notes
1. Our interest is in total trading volume, defined as turnover (volume/shares outstanding), rather
than the volume of individual trades.
2. AMEX volume data are available since 1963, here combined with the NYSE data series for par-
simony. The value weighting is accomplished by calculating for each trading day the total
dollar-value traded that day (aggregated over all stocks) and dividing it by aggregate market
value outstanding as of that day. This measure is then annualized by multiplying the mean daily
turnover for that year by the actual number of trading days in that year.
3. At this point, it is not clear whether the retreat in trading volume after the financial crisis in
2008 is temporary or signals a structural break in the secular upward trajectory of trading. Note,
however, that even after this retreat volume of trading is very high by historical standards. Note
also that in Figure 1 trading volume subsided after the crash in 1987 for NYSE/AMEX, and after
the dotcom crash in the early 2000s for NASDAQ—but soon resumed its upward climb.
4. This trend toward algorithmic and technical-type trading has been turbocharged by the great reduc-
tion in transactions costs and improvements in technology during the last 20 to 30 years. Bid-ask
spreads and commissions are an order of magnitude lower than they were just a generation ago.
Computing and communication technology has enabled traders to execute thousands of orders a
minute, often completely automated. In most likelihood, sentiment has also played considerable
role in the increase of trading volume, where just a generation or two ago stock trading was a
fairly arcane and specialized activity but has since become much more accepted and even
embraced in society. Sentiment is also likely the chief driver of the early spike in volume of trad-
ing during the 1920s, when there is little room for the transaction cost and technology explanation.
5. Empirically, we do not attempt specific decomposition of trading volume into the component
driven by firm fundamentals and the component driven by other factors. Instead, we examine the
overall trading volume and stock volatility and use various methods to control for firm
fundamentals.
6. An exception to this generally positive view of the volume is a recent literature on stock bubbles
documenting that market valuations, which is ‘‘too high’’ compared with fundamentals are typi-
cally accompanied by ‘‘overtrading’’ (Hong & Stein, 2007), where euphoric investors bid up
prices solely in anticipation of even further price appreciation.
7. In untabulated work, we explored several other natural experiments settings which rely on significant
changes in trading triggered by an appropriate event, including exchange switches, S&P 500 addi-
tions and deletions, Russell 3000 additions and deletions, and the introduction of automated stock
quotes. Exchange switches and S&P 500 changes provide the desired material changes in trading;
however, they are problematic in controlling for fundamentals because of their own information con-
tent and the associated changes in investor clientele and information environment (e.g., analyst cov-
erage). Russell 3000 changes are attractive because they are entirely driven by market value changes,
so they are entirely predictable and with little intrinsic information content; unfortunately, they dis-
play little reliable change in long-term trading except for a brief burst of apparent portfolio reshuf-
fling on the effective date. Similar lack of material long-term changes in trading volume obtains with
the introduction of automated stock quotes. On balance, these deficiencies in ‘‘before and after the
event’’ settings leads us to emphasize the ‘‘simultaneous effects’’ settings presented in the article.
8. Because volume count on Nasdaq is determined differently from that on the other three stock
exchanges, we exclude fund-pairs where one fund is traded on Nasdaq and the other is traded on
any of the other three stock exchanges.

9. We deleted two observations with extreme values of DIF_STDRET. Including these observations
will lead to a spread of 12% in mean DIF_STDRET (vs. 4% tabulated in Panel C of Table 1)
across the top and bottom quintiles sorted by DIF_VOLUME.
10. Using similar variables and test design, we find similar results for a comparable setting, trading,
and volatility for American Depositary Receipts (ADRs) versus the underlying stocks.
11. The Actual/Implied ratio equals one when there is no autocorrelation of returns (French & Roll,
1986), and the magnitude of deviation from one indicates the magnitude of trading noise. For
example, the ratio ranging from 0.75 to 0.85 in the early years indicates a negative autocorrela-
tion in returns (returns reversals), and corresponds to 25% to 15% of the volatility of daily
returns explained by noise trading.
12. The most well-known problem with Nasdaq volume arises because Nasdaq is a dealer market,
and thus end-customer to end-customer transactions pass through a dealer, and are thus double
counted in volume. But not all trades are double counted, and there are substantial other practical
complications, see Anderson and Dyl (2005) for details.
13. In Table 6, we report results based on 1,043 monthly observations over 1926-2012. One potential
problem of this specification is that the coefficient on lagged volume is close to one in one of
the regressions, which suggests a possible unit-root problem. This concern, however, does not
seem critical because a Phillips–Perron test for unit root rejects at the 0.05 level. In any case, we
obtain similar results when we use an alternative specification based on yearly observations,
where the coefficients on lagged volume are reliably different from one.
14. Data from the Central Intelligence Agency (CIA) World Factbook.
References
Amihud, Y., & Mendelson, H. (1986). Asset pricing and the bid-ask spread. Journal of Financial
Economics, 17, 223-249.
Anderson, A., & Dyl, E. A. (2005). Market structure and trading volume. Journal of Financial
Research, 28, 115-131.
Baker, M., & Stein, J. C. (2004). Market liquidity as a sentiment indicator. Journal of Financial
Markets, 7, 271-299.
Bamber, L. S., Barron, O. E., & Stober, T. L. (1999). Differential interpretations and trading volume.
Journal of Financial and Quantitative Analysis, 34, 369-386.
Barber, B. M., & Odean, T. (2002). Online investors: Do the slow die first? Review of Financial
Studies, 15, 455-488.
Black, F. (1986). Noise. Journal of Finance, 41, 529-543.
Branch, B., & Freed, W. (1977). Bid-asked spreads on the AMEX and the big board. Journal of
Finance, 32, 159-163.
Brennan, M. J., Chordia, T., & Subrahmanyam, A. (1998). Alternative factor specifications, security
characteristics, and the cross-section of expected stock returns. Journal of Financial Economics,
49, 345-373.
Campbell, J. Y., Grossman, S. J., & Wang, J. (1993). Trading volume and serial correlation in stock
returns. Quarterly Journal of Economics, 108, 905-939.
Chordia, T., Huh, S., & Subrahmanyam, A. (2007). The cross-section of expected trading activity.
Review of Financial Studies, 20, 709-740.
Chordia, T., Roll, R., & Subrahmanyam, A. (2011). Recent trends in trading activity and market qual-
ity. Journal of Financial Economics, 101, 243-263.
Chordia, T., & Swaminathan, B. (2000). Trading volume and cross-autocorrelations in stock returns.
Journal of Finance, 55, 913-935.
Copeland, T. E., & Galai, D. (1983). Information effects on the bid-ask spread. Journal of Finance,
38, 1457-1469.
Coval, J., & Stafford, E. (2007). Asset fire sales (and purchases) in equity markets. Journal of
Financial Economics, 86, 479-512.

Dichev et al. 517
DeLong, J. B., Shleifer, A., Summers, L. H., & Waldmann, R. J. (1990). Positive feedback investment
strategies and destabilizing rational speculation. Journal of Finance, 45, 379-395.
Easley, D., Kiefer, N. M., & O’Hara, M. (1996). Cream-skimming or profit-sharing? The curious role
of purchased order flow. Journal of Finance, 51, 811-833.
Easley, D., Kiefer, N. M., & O’Hara, M. (1997). The information content of the trading process.
Journal of Empirical Finance, 4, 159-186.
Fang, V. W., Noe, T. H., & Tice, S. (2009). Stock market liquidity and firm value. Journal of
Financial Economics, 94, 150-169.
Foucalt, T., Sraer, D., & Thesmar, D. (2011). Individual investors and volatility. Journal of Finance,
66, 1369-1406.
French, K. R., & Roll, R. (1986). Stock return variances: The arrival of information and the reaction
of traders. Journal of Financial Economics, 17, 5-26.
Gompers, P. A., Ishii, J., & Metrick, A. (2010). Extreme governance: An analysis of dual-class firms
in the United States. Review of Financial Studies, 23, 1051-1088.
Han, B., & Kumar, A. (2013). Speculative retail trading and asset prices. Journal of Financial and
Quantitative Analysis, 48, 377-404.
Hendershott, T., Jones, C. M., & Menkveld, A. J. (2011). Does algorithmic trading improve liquidity?
Hong, H., & Stein, J. C. (2007). Disagreement and the stock market. Journal of Economic
Perspectives, 21, 109-128.
Hong, H., & Yu, J. (2009). Gone fishin’: Seasonality in trading activity and asset prices. Journal of
Financial Markets, 12, 672-702.
Irvine, P. J., & Pontiff, J. (2009). Idiosyncratic return volatility, cash flows, and product market com-
petition. Review of Financial Studies, 22, 1149-1177.
Jones, C. M. (2002). A century of stock market Liquidity and trading costs (Working paper). New
York, NY: Columbia University.
Kadlec, G. B., & McConnell, J. C. (1994). The effect of market segmentation and illiquidity on asset
prices: Evidence from exchange listings. Journal of Finance, 49, 611-636.
Kandel, E., & Pearson, N. D. (1995). Differential interpretation of public signals and trade in specula-
tive Markets. Journal of Political Economy, 103, 831-872.
Karpoff, J. M. (1987). The relation between price changes and trading volume: A survey. Journal of
Financial and Quantitative Analysis, 22, 109-126.
Lease, R. C., McConnell, J. J., & Mikkelson, W. H. (1983). The market value of control in publicly-
traded corporations. Journal of Financial Economics, 11, 439-471.
Lease, R. C., McConnell, J. J., & Mikkelson, W. H. (1984). The market value of differential voting
rights in closely held corporations. Journal of Business, 57, 443-467.
Liu, W. (2006). A liquidity-augmented capital asset pricing model. Journal of Financial Economics,
82, 631-671.
Luo, Y. (2005). Do insiders learn from outsiders? Evidence from Mergers and Acquisitions. Journal
of Finance, 60, 1951-1982.
Morse, D. (1980). Asymmetrical information in securities markets and trading volume. Journal of
Financial and Quantitative Analysis, 15, 1129-1148.
Pastor, L., & Veronesi, P. (2003). Stock valuation and learning about profitability. Journal of
Finance, 58, 1749-1790.
Petersen, M. A. (2009). Estimating standard errors in finance panel data sets: Comparing approaches.
Ripley, W. Z. (1911). Railway speculation. The Quarterly Journal of Economics, 25, 185-215.
Schwert, W. G. (1989). Why does stock market volatility change over time? Journal of Finance, 44,
1115-1153.
Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under conditions of risk.

Shiller, R. J. (1981). Do stock prices move too much to be justified by the subsequent changes in div-
idends? American Economic Review, 71, 421-436.
Steinhauser, G., & Olsen, J. (2012, March 31). EU eyes small tax on share purchases. Businessweek
Retrieved from http://www.businessweek.com/ap/2012-03/D9TREPG80.htm
Summers, L. H., & Summers, V. P. (1989). When financial markets work too well: A cautious case
for a securities transactions tax. Journal of Financial Services Research, 3, 261-286.
van Rooij, M., Lusardi, A., & Alessie, R. (2011). Financial literacy and stock market participation.
Journal of Financial Economics, 101, 449-472.
Wei, S. X., & Zhang, C. (2006). Why did individual stocks become more volatile? Journal of
Business, 79, 259-292.
Zhang, F. (2010). High-frequency trading, stock volatility, and price discovery (Working paper). New
Haven, CT: Yale University.
Zingales, L. (1994). The value of the voting right: A study of the Milan Stock Exchange experience.

A49 I The Dark Side of Trading

Uploaded by

Copyright:

Available Formats

A49 I The Dark Side of Trading

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A49 I The Dark Side of Trading

Uploaded by

Copyright:

Available Formats

Article

Ilia D. Dichev1, Kelly Huang2, and

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Figure 1. Value-weighted stock trading volume from 1926 to 2012.

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Theory and Background

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Panel A: Sample Composition.

Panel B: Descriptive Statistics.

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Panel C: Difference in Trading Volume and Stock Volatility for ETFs.

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Dual-Class U.S. Stocks

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Table 2. Evidence From Dual-Class Firms.

Panel A: Sample Composition.

Panel B: Descriptive Statistics.

Panel C: Difference in Trading Volume and Stock Volatility.

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Large-Sample Evidence for U.S. Stocks

Evidence From the Full Time-Series of U.S. Stocks

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Year VOLUME STDRET Year VOLUME STDRET Year VOLUME STDRET

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Panel A: Stock Volatility Formed on the Basis of Trading Volume.

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

STDRETi, t = b0 + b1 HIGHi, t + b2 VOLUMEi, t + b3 VOLUME3HIGHi, t + b4 STDRETi, t 1

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Pred. sign (1) (2) (3) (4) (5)

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Panel A: Trading Volume Causes Stock Volatility (N = 1,043).

Intercept 0.003 10.08 0.003 10.05 8.87 .003

Panel B: Stock Volatility Causes Trading Volume (N = 1,043).

Intercept 0.008 6.36 0.002 3.12 30.50 \.001

Extensions and Robustness Checks

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Declaration of Conflicting Interests

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

Downloaded from jaf.sagepub.com at RUTGERS UNIV on August 21, 2015

You might also like