Price Momentum and Trading Volume

Charles M. C. Lee and Bhaskaran Swaminathan*

This Draft: August 29, 1999

*
Both Charles M.C. Lee and Bhaskaran Swaminathan are from the Johnson Graduate School of
Management, Cornell University. We thank Yakov Amihud, Hal Bierman, Larry Brown, Tom Dyckman,
David Easley, John Elliott, Eugene Fama, Wayne Ferson, Maureen O’Hara, Jay Ritter, Andrei Shleifer,
René Stulz (editor), Avanidhar Subrahmanyam, Yutaka Soejima, two anonymous referees, and workshop
participants at Barclays Global Investors, 1998 Berkeley Program in Finance, Carnegie Mellon
University, Chicago Quantitative Alliance’s Fall 1998 conference, Cornell University, University of
Florida, George Washington University, University of Illinois at Urbana-Champaign, the Mitsui Life
Finance Conference at the University of Michigan, the 1998 NBER Behavioral Finance Meeting, the
Ninth Financial, Economics and Accounting Conference, UNC-Chapel Hill, 1998 Prudential Securities
Conference, the 1999 Q-Group Spring Conference, the Summer of Accounting and Finance Conference at
Tel Aviv University, and the 1999 Western Finance Association Annual Meeting for helpful comments.
We also thank Bill Gebhardt for his expert research assistance. Data on analyst following and long-term
earnings growth forecasts are from I/B/E/S, Inc. Any errors are our own.
 Price Momentum and Trading Volume

Abstract

This study shows that past trading volume provides an important link between

"momentum" and "value" strategies. Specifically, we find that firms with high (low) past

turnover ratios exhibit many glamour (value) characteristics, earn lower (higher) future

returns, and have consistently more negative (positive) earnings surprises over the next

eight quarters. Past trading volume also predicts both the magnitude and persistence of

price momentum. Specifically, price momentum effects reverse over the next five years

and high (low) volume winners (losers) experience faster reversals. Collectively, our

findings show that past volume helps to reconcile intermediate-horizon "underreaction"

and long-horizon "overreaction" effects.

Financial academics and practitioners have long recognized that past trading volume may
provide valuable information about a security. However, there is little agreement on how
volume information should be handled and interpreted. Even less is known about how
past trading volume interacts with past returns in the prediction of future stock returns.
Stock returns and trading volume are jointly determined by the same market dynamics,
and are inextricably linked in theory (e.g., Blume, Easley, and O’Hara (1994)). Yet prior
empirical studies have generally accorded them separate treatment.

In this study, we investigate the usefulness of trading volume in predicting cross-sectional

returns for various price momentum portfolios. The study is organized into two parts. In
the first part, we document the interaction between past returns and past trading volume in
predicting future returns over intermediate and long horizons. 1 In the second part, we
evaluate alternative explanations for these empirical regularities. Our findings extend the
literature on both price momentum and trading volume. In addition, we establish an
important link between intermediate-horizon "momentum" and long-horizon "value"
strategies.

We contribute to the literature on price momentum in two ways. First, we show

that the price momentum effect documented by Jegadeesh and Titman (1993,
hereafter JT) reverses over long horizons. Like JT, we find no significant price
reversals through the third year following portfolio formation. However, over
years 3 through 5, we find that initial winner portfolios significantly underperform
initial loser portfolios. This finding is important because it refutes the common
presumption that price momentum is simply a market underreaction. Instead, the
evidence suggests that at least a portion of the initial momentum gain is better
characterized as an overreaction.2

Secondly, we show that past trading volume predicts both the magnitude and the
persistence of future price momentum. Specifically, high (low) volume winners
(losers) experience faster momentum reversals. Conditional on past volume, we

1
can create JT-type momentum portfolios (winners minus losers) that either exhibit
long-horizon return reversals or long-horizon return continuations. This evidence
shows that the information contained in past trading volume can be useful in
reconciling intermediate-horizon "underreaction" and long-horizon "overreaction"
effects.

Our findings also extend the trading volume literature. Prior research (e.g., Datar et. al.
(1998)) shows that low (high) volume firms earn higher (lower) future returns. We show
that this volume effect is long-lived (i.e., it is observable over the next three to five years),
and is most pronounced among the extreme winner and loser portfolios. More
importantly, our evidence contradicts the common interpretation of trading volume as
simply a liquidity proxy. These findings instead show that past trading volume is related
to various "value" strategies.

Contrary to the liquidity explanation, we find that high (low) volume stocks earn higher
(lower) average returns in each of the five years prior to portfolio formation. We show
that trading volume is only weakly correlated with traditional liquidity proxies, and that
the volume effect is robust to various risk adjustments. We find that the volume-based
momentum effect holds even in a sub-sample of the largest 50 percent of New York
(NYSE) and American Stock Exchange (AMEX) firms. Finally, we show that most of the
excess returns to volume-based strategies is attributable to changes in trading volume.
Firms whose recent volume is higher (lower) than volume four years ago experience
significantly lower (higher) future returns. The change in volume measures abnormal
trading activity, and is unlikely to be a liquidity proxy.

On the other hand, we find that low (high) volume stocks display many characteristics
commonly associated with value (glamour) investing. Specifically, lower (higher) trading
volume is associated with worse (better) current operating performance, larger (smaller)
declines in past operating performance, higher (lower) book-to-market ratios, lower
(higher) analyst followings, lower (higher) long-term earnings growth estimates, higher

2
(lower) factor loadings on the Fama-French HML factor, and lower (higher) stock returns
over the previous five years.

Further analyses show that the higher (lower) future returns experienced by low (high)
volume stocks are related to investor misperceptions about future earnings. Analysts
provide lower (higher) long-term earnings growth forecasts for low (high) volume stocks.
However, low (high) volume firms experience significantly better (worse) future operating
performance. Moreover, we find that short-window earnings announcement returns are
significantly more positive (negative) for low (high) volume firms over each of the next
eight quarters. The same pattern is observed for both past winners and past losers.
Evidently the market is "surprised" by the systematically higher (lower) future earnings of
low (high) volume firms.

The fact that a market statistic widely used in technical analysis can provide information
about relative under- or over-valuation is surprising, and is difficult to reconcile with
existing theoretical work. To help explain these results, we evaluate the predictions of
several behavioral models. We conclude that each model has specific features that help
explain some aspects of our findings, but that no single model accommodates all our
findings. We also discuss an interesting illustrative tool, dubbed the momentum life cycle
(MLC) hypothesis, which captures some of the most salient features of our empirical
results.

The remainder of the paper is organized as follows. In the next section, we discuss related
literature. In Section II, we describe our sample and methodology. In Section III we
present our empirical results. In Section IV, we further explore the information content of
trading volume, and relate these findings to several behavioral models. Finally, in Section
V, we conclude with a summary of the evidence and a discussion of the implications.

3
 I. Related Literature
In recent years, a number of researchers have presented evidence that cross-sectional
stock returns are predictable based on past returns. For example, DeBondt and Thaler
(1985, 1987) document long-term price reversals in which long-term past losers
outperform long-term past winners over the subsequent three to five years. Similarly,
Jegadeesh (1990) and Lehmann (1990) report price reversals at monthly and weekly
intervals.

But perhaps the most puzzling results are the intermediate-horizon return continuations
reported by Jegadeesh and Titman (1993). Forming portfolios based on past three to
twelve month returns they show that past winners on average continue to outperform past
losers over the next three to 12 months. While many competing explanations have been
suggested for the long-horizon price reversal patterns, 3 far fewer explanations have been
advanced to explain the intermediate-horizon price momentum effect.

For example, Fama and French (1996) show that a three-factor model of returns fails to
explain intermediate horizon price momentum. Chan, Jegadeesh, and Lakonishok (1996)
show that intermediate-horizon return continuation can be partially explained by
underreaction to earnings news, but price momentum is not subsumed by earnings
momentum. Rouwenhorst (1998) finds a similar pattern of intermediate horizon price
momentum in 12 other countries, suggesting that the effect is not likely due to a data
snooping bias.

More recently, Conrad and Kaul (1998) suggest that the momentum effect may be due to
cross-sectional variation in the mean returns of individual securities. Moskowitz and
Grinblatt (1999) claim that a significant component of firm-specific momentum can be
explained by industry momentum. However, the evidence in Grundy and Martin (1998)
suggests momentum effects are not explained by time-varying factor exposures, cross-
sectional differences in expected returns, or industry effects.4 None of these studies

4
examine the interaction between past trading volume and past price movements in
predicting cross-sectional returns.

At least two theoretical papers suggest that past trading volume may provide valuable
information about a security. Campbell, Grossman, and Wang (1993) present a model in
which trading volume proxies for the aggregate demand of liquidity traders. However,
their model focuses on short-run liquidity imbalances (or volume shocks) of a daily or
weekly duration, and makes no predictions about longer-term returns. Blume, Easley, and
O'Hara (1994) present a model in which traders can learn valuable information about a
security by observing both past price and past volume information. However, their model
does not specify the nature of the information that might be derived from past volume.
We provide empirical evidence on the nature of this information.

Our study is also tangentially related to Conrad, Hameed, and Niden (1994). Conrad et al.
show that, at weekly intervals, the price reversal pattern is observed only for heavily
traded stocks; less traded stocks exhibit return continuation.5 The Conrad et al. study
focuses on short-term price movements, because it is motivated by market microstructure
concerns raised in Campbell, Grossman, and Wang (1993). Our interest lies in the
prediction of cross-sectional returns over longer (three month and longer) horizons. In the
intermediate time horizon, the empirical puzzle is not return reversal, but return
continuation. Given the longer time horizons, these price continuations are unlikely to be
due to the short-term liquidity shocks. In fact, we deliberately form our portfolios with a
one-week (or a one-month) lag to minimize the effect of bid-ask bounce and short-horizon
return reversals.

In a related study, Datar, Naik, and Radcliffe (1998) show that low turnover stocks
generally earn higher returns than high turnover stocks. They interpret this result as
providing support for the liquidity hypothesis of Amihud and Mendelson (1986).6
According to the liquidity hypothesis, firms with relatively low trading volume are less
liquid, and therefore command a higher expected return. We build on the finding of Datar

5
et al. (1998) by examining the interaction between past price momentum and trading
volume in predicting cross-sectional returns. We confirm their findings but also present
additional evidence, which is difficult to reconcile with the liquidity hypothesis.

In sum, prior studies have documented a striking pattern of price momentum in the
intermediate horizon. Other studies have examined the relation between trading volume
and future returns. We integrate these two lines of research and report the joint
distribution of future returns conditional on both past trading volume and past returns.
More importantly, as we show later, our results provide a bridge between past studies on
market over- and underreaction, as well as a link to recent theoretical studies in behavioral
finance.

II. Sample and Methodology

Our sample consists of all firms listed on the New York (NYSE) and American (AMEX)
Stock Exchange during the period January 1965 through December 1995 with at least two
years of data prior to the portfolio formation date. We exclude Nasdaq firms from our
analysis for two reasons. First, Nasdaq firms tend to be smaller and more difficult to trade
in momentum-based strategies. Second, trading volume for Nasdaq stocks is inflated
relative to NYSE and AMEX stocks due to the double counting of dealer trades (Gould
and Kleidon (1994)). Because we rank our firms by average turnover, mixing Nasdaq and
NYSE firms will result in inconsistent treatment of firms across these different markets.7

We eliminate any firm that was a prime, a closed-end fund, a real estate investment trust
(REIT), an American Depository Receipt (ADR), or a foreign company. We also
eliminate firms that were delisted within five days of the portfolio formation date and firms
whose stock price as of the portfolio formation date was less than a dollar. Finally, to be
included in our sample a stock must also have available information on past returns,
trading volume, market capitalization, and stock price. Trading volume (Volume) is
defined as the average daily turnover in percent during the portfolio formation period,
where daily turnover is the ratio of the number of shares traded each day to the number of

6
shares outstanding at the end of the day.8 Descriptive statistics that require accounting
data (e.g., the B/M ratio and the return-on-equity) are based on the subset of firms in each
portfolio that also are in the COMPUSTAT database. Tests involving long-term earnings
forecasts or number of analysts are based on firms that are in the I/B/E/S database.

At the beginning of each month, from January 1965 to December 1995, we rank all
eligible stocks independently on the basis of past returns and past trading volume. The
stocks are then assigned to one of 10 portfolios based on returns over the previous J
months and one of three portfolios based on the trading volume over the same time
period.9 The intersections resulting from the two independent rankings give rise to 30
price momentum-volume portfolios. We focus our attention on the monthly returns of
extreme winner and loser deciles over the next K months (K=3,6,9, or 12) and over the
next five years.

Similar to Jegadeesh and Titman (1993), the monthly return for a K-month holding period
is based on an equal-weighted average of portfolio returns from strategies implemented in
the current month and the previous K-1 months. For example, the monthly return for a
three-month holding period is based on an equal-weighted average of portfolio returns
from this month’s strategy, last month’s strategy, and the strategy from two months ago.
This is equivalent to revising the weights of (approximately) 1/3rd of the portfolio each
month and carrying over the rest from the previous month. The technique allows us to
use simple t-statistics for monthly returns. To avoid potential microstructure biases, we
impose a one-week lag between the end of the portfolio formation period (J) and the
beginning of the performance measurement period (K).10

III. Results for Volume Based Price Momentum Strategies

In this section, we discuss the empirical results for volume based price momentum
strategies. In Subsection A, we confirm the price momentum strategy for our sample of
firms. We also ensure that our results are consistent with the stylized facts from prior
volume studies. In Subsection B, we introduce volume-based price momentum portfolios

7
and examine their predictive power for cross-sectional returns over intermediate horizons.
In Subsection C, we provide results of robustness checks for volume based price
momentum strategies. In Subsection D, we provide results from Fama-French three-
factor regressions. In Subsection E, we examine long-horizon (one to five year) returns to
various volume-based price momentum portfolios. Finally in Subsection F, we provide
evidence on the usefulness of trading volume in predicting the timing of price momentum
reversals.

A. Price Momentum
Table I summarizes results from several price momentum portfolio strategies. Each
January, stocks are ranked and grouped into decile portfolios on the basis of their returns
over the previous 3, 6, 9, and 12 months. We report results for the bottom decile
portfolio of extreme losers (R1), the top decile of extreme winners (R10), and one
intermediate portfolio (R5). The other intermediate portfolio results are consistent with
findings in prior papers (Jegadeesh and Titman (1993)) and are omitted for simplicity of
presentation.

For each portfolio, Table I reports the mean return, and volume during the portfolio
formation period, the time-series average of the median size decile of the portfolio based
on NYSE/AMEX cutoffs (SzRnk), and the time-series average of the median stock price
(Price) as of portfolio formation date. At the portfolio formation date, stocks in the
winner portfolio are typically larger (Column 5) and have higher price (Column 6) than
stocks in the loser portfolio. This is not surprising given the difference in recent returns.
For example, for the six month formation period (J=6), losers lost an average of 6.36
percent per month over the past six months while winners gained 8.30 percent per month
(Column 3).

The results in Columns 3 and 4 confirm stylized facts about price movements and trading
volume observed in prior studies. As expected, trading volume is positively correlated
with absolute returns, so that the extreme price momentum portfolios exhibit higher

8
trading volume. For example, the average daily turnover for the R1 and R10 portfolios in
the six-month portfolio formation period are 0.17 percent and 0.23 percent respectively,
compared to 0.12 percent for the intermediate (R5) portfolio. In addition, we find that the
positive relation between absolute returns and trading volume is asymmetric, in that
extreme winners have a higher trading volume than extreme losers (see Lakonishok and
Smidt (1986)).

Columns 7 through 10 report equal-weighted average monthly returns over the next K
months (K= 3, 6, 9, 12). In addition, for each portfolio formation period (J) and holding
period (K), we report the mean return from a dollar-neutral strategy of buying the extreme
winners and selling the extreme losers (R10-R1). These results confirm the presence of
price momentum in our sample. For example, with a six-month portfolio formation period
(J=6), past winners gain an average of 1.65 percent per month over the next nine-months
(K=9). Past losers gain an average of only 0.57 percent per month over the same time
period. The difference between R10 and R1 is 1.08 percent per month. The difference in
average monthly returns between R10 and R1 is significantly positive in all (J,K)
combinations.

The last five columns of Table I report the annual event-time returns for each portfolio for
five 12-month periods following the portfolio formation date, with t-statistics based on the
Hansen and Hodrick (1980) correction for autocorrelation up to lag 11. In Year 1, the
R10-R1 portfolio yields a statistically significant return of between 10.62 percent and
12.70 percent per year. Consistent with Jegadeesh and Titman (1993), we observe a
modest reversal to momentum profits in Years 2 and 3. Like their study, we find that the
negative returns in these two years are not statistically significant and are not sufficient to
explain the initial momentum gains in Year 1.

When we extend the event time to Years 4 and 5, a pattern of price reversal begins to
emerge. The last two columns show that R10-R1 returns are negative in Years 4 and 5
for all formation periods. The reversal pattern becomes stronger monotonically as the

9
formation period (J) increases. For the longest formation period (J=12), the sum of the
losses in Years 2 through 5 (10.95 percent) almost offsets the entire gain from Year 1
(11.56 percent). This reversal pattern is not documented in prior studies that limit return
prediction to Year 3.

In sum, Table I confirms prior findings on price momentum. It also extends prior results
by documenting significant long-term price reversals in Years 4 and 5. Our results show
intermediate-horizon price momentum effects do eventually reverse. Moreover, the longer
the estimation period for past returns, the more imminent the future price reversals. We
will expand on this theme later when we introduce autocorrelation evidence based on
regression tests (see Table VIII).

B. Volume-based Price Momentum

Table II reports returns to portfolios formed on the basis of a two-way sort between price
momentum and past trading volume. To create this table, we sort all sample firms at the
beginning of each month based on their returns over the past J months, and divide them
into 10 portfolios (R1 to R10). We then independently sort these same firms based on
their average daily turnover rate over the past J months, and divide them into three
volume portfolios (V1 to V3). V1 represents the lowest trading volume portfolio and V3
represents the highest trading volume portfolio. Table values represent the average
monthly return over the next K months (K=3, 6, 9, 12).

Several key results emerge from Table II. First, conditional on past returns, low volume
stocks generally do better than high volume stocks over the next 12 months. This is seen
in the consistently negative returns to the V3-V1 portfolio. For example, with a nine-
month portfolio formation period and six month holding period (J=9, K=6), low volume
losers outperform high volume losers by 1.02 percent per month while low volume
winners outperform high volume winners by 0.26 percent per month. We find similar
results in almost every (J, K) cell. Apparently firms that experience low trading volume in
the recent past tend to outperform firms that experience high trading volume.

10
The finding that low volume firms earn higher expected returns is consistent with Datar,
Naik, and Radcliffe (1998). In that paper, this finding is interpreted as evidence that low
volume firms command a greater illiquidity premium. However, Table II also contains
evidence that is difficult to explain by the liquidity explanation. The bottom row of each
cell in this table shows the return to a dollar-neutral, price momentum strategy (R10-R1).
Focusing on this row, it is clear that R10-R1 returns are higher for high volume (V3) firms
than for low volume (V1) firms. For example, for J=6 and K=6, the price momentum
spread is 1.46 percent for high volume firms and only 0.54 percent for low volume firms.
The difference of 0.91 percent per month is both economically and statistically significant.
The other cells illustrate qualitatively the same effect. The price momentum premium is
clearly higher in high volume (presumably more liquid) firms.

According to the liquidity hypothesis, the portfolio with lower liquidity should earn higher
expected returns. It is difficult to understand why a dollar-neutral portfolio of high
turnover stocks should be less liquid than a dollar-neutral portfolio of low turnover stocks.
Moreover, the magnitude of the difference is too large to be explained by illiquidity. For
example, for J=6, K=6, the difference in momentum premium between V3 and V1 is 0.91
percent per month or approximately 11 percent annualized. For the liquidity hypothesis to
hold, high volume winners would have to be much more illiquid than are high volume
losers.

A closer examination shows that this counter-intuitive result is driven primarily by the
return differential in the loser portfolio (R1). Low volume losers (R1V1) rebound
strongly in the next 12 months relative to high volume losers, averaging more than one
percent per month in virtually all (J, K) combinations. In contrast, high volume losers
(R1V3) earn an average return of between -0.21 percent and +0.41 percent per month.
The return differential between high and low volume winners is not nearly as large. In
most cells the difference in returns between low volume winners and high volume winners
is small and statistically insignificant.11 Nevertheless, high volume winners generally

11
underperform low volume winners, so buying high volume winners does not enhance the
performance of the price momentum strategy.

In sum, Table II shows that over the next 12 months, price momentum is more
pronounced among high volume stocks. In addition, we find that controlling for price
momentum, low volume stocks generally outperform high volume stocks. This effect is
most pronounced among losers in the intermediate-horizon.

C. Robustness Tests
Table III presents various robustness checks on these basic intermediate-horizon results.
Panel A confirms these patterns for three subperiods. The first subperiod spans 1965 to
1975, the second subperiod covers 1976 to 1985, and the last subperiod covers 1986 to
1995. We report results for the six-month formation period (J=6), but results are similar
for other formation periods. In all three subperiods, winners outperform losers, low
volume stocks outperform high volume stocks, and momentum is stronger among high
volume stocks. In fact, the result is strongest in the more recent subperiod.

Our earlier results are based on 10 price momentum portfolios and three trading volume
portfolios (10 × 3). Table III shows that our results are not specific to this partitioning.
Panel B reports results using three price momentum portfolios and 10 trading volume
portfolios (3 × 10), while Panel C reports results using five price momentum and five
volume portfolios (5 × 5). Generally, the volume-based results are as strong or stronger
than those reported in Table II. In fact, in these partitions, low volume winners generally
outperform high volume winners by a wider margin than was evident in Table II.

To ensure that these results are not driven by a few small stocks, Panel D of Table III
reports the volume based price momentum results using only the largest 50 percent of all
NYSE/AMEX stocks. Not surprisingly, both the momentum and volume effects are
weaker for this restricted sample. However, the volume-based results continue to obtain.
For example, for J=6, K=6, the momentum spread is 0.95 percent for high volume firms

12
and only 0.24 percent for low volume firms. This effect is again driven by low volume
losers that gain 1.09 percent per month, as compared to high volume losers that gain only
0.44 percent per month. 12

D. Risk Adjustments
Table IV reports descriptive characteristics for various price momentum and volume
portfolios. Looking down each column, we see that losers are generally smaller firms with
lower stock prices. This is not surprising given the losses they recently sustained.
Looking across each row, we see that high and low volume portfolios do not differ
significantly in terms of their median stock price or firm size. High volume firms tend to
be somewhat larger and more highly priced, but the difference is not large. For example,
Panel A shows that high volume losers (R1V3) have a median price of $10.64 while low
volume losers have a median price of $7.65. In later tests, we provide more formal
controls for firm size and industry differences.

Table V provides additional evidence on the source of abnormal returns for the various
volume based price momentum strategies. In this table, we report the results from time-
series regressions based on the Fama-French (1993) three-factor model.13 Specifically, we
run the following time-series regression using monthly portfolio returns:

(ri - rf ) = ai + bi (rm - rf) + si SMB + hi HML + ei , (1)

where ri is the monthly return for portfolio i, rf is the monthly return on one-month T-bill
obtained from the Ibbotson Associates’ Stocks, Bonds, Bills, Inflation (SBBI) series, rm is
the value-weighted return on the NYSE/AMEX/ Nasdaq market index, SMB is the Fama-
French small firm factor, HML is the Fama-French book-to-market (value) factor, bi, si, hi
are the corresponding factor loadings, and ai is the intercept or the alpha of the
portfolio.14 For parsimony, we report results for symmetrical combinations of portfolio
formation and holding periods (J and K= 6 and 12 months). The first cell on the left in
each panel reports the estimated intercept coefficient, the subsequent cells report

13
estimated coefficients for bi, si, and hi, respectively. The last cell of each panel reports the
adjusted R2.

The estimated intercept coefficients from these regressions (ai) are interpretable as the
risk-adjusted return of the portfolio relative to the three-factor model. Focusing on these
intercepts, it is clear that our earlier results cannot be explained by the Fama-French
factors. The intercepts corresponding to R10-R1 are positive across all three volume
categories. The return differential between winners and losers remains much higher for
high volume (V3) firms than for low volume (V1) firms. Finally, a strategy of buying low
volume winners and selling high volume losers yields average abnormal returns of between
one percent and two percent per month in both panels.

Even more revealing are the estimated factor loadings on the SMB and HML factors. First
focus on the estimated coefficients for the HML factor (hi) in the six-month horizon. Here
we see that low volume stocks (V1 portfolios) have a much more positive loading on the
HML factor. This applies to winners (R10), losers (R1), and even the intermediate
portfolio (R5). Apparently low volume stocks behave more like value stocks, i.e., stocks
with high book-to-market ratios. High volume stocks, on the other hand, behave more
like glamour stocks, i.e., stocks with low book-to-market ratios. In fact, high volume
winners (R10V3) tend to have negative loadings on the HML factor.

The magnitudes of the HML loadings correspond to those obtained for value and glamour
stocks (highest and lowest 40 percent by book-to-market ratio) in Fama and French
(1993). For (J=6, K=6), the difference in estimated HML loadings for our low and high
volume winner portfolios (R10V3-R10V1) is -0.51. This is comparable to the spread
Fama and French obtain when they separate firms on the basis of high versus low book-to-
market ratios.15 In short, low (high) volume stocks earn positive (negative) excess returns
when high B/M stocks do well. These results lay a foundation for our attempt to interpret
the findings of the paper and understand the nature of the information provided by trading
volume (See Section III).

14
The factor loadings on SMB also provide interesting information. Table V shows that our
high and low volume portfolios exhibit virtually no difference in their sensitivity to the
SMB factor. In the winner and loser portfolios (R10 and R1), differences in trading
volume have no explanatory power for a stock's sensitivity to firm size. In the
intermediate return portfolio (R5), there is in fact some evidence that high volume stocks
actually behave more like small stocks than low volume stocks. Since small stocks are
generally more illiquid, this evidence runs counter to the liquidity explanation for the
volume effect.

E. Long-horizon Results
Table VI presents long-term (event time) annual returns to various trading volume and
price momentum portfolios over the next five years. These results are based on the six
month portfolio formation period (J=6), 10 momentum portfolios and three volume
portfolios (10 × 3). Year 1, Year 2, Year 3, Year 4, and Year 5 represent the annual
returns of each portfolio in the five 12-month periods following the portfolio formation
date. To correct for spurious autocorrelation from overlapping observations, we compute
t-statistics using the Hansen and Hodrick (1980) correction for autocorrelation up to lag
11. Panel A presents raw returns, Panel B reports industry-adjusted returns, and Panel C
reports size-adjusted returns.

The industry adjustment is based on 25 equal-weighted industry portfolios formed by

grouping two-digit SIC codes (see Appendix).16 The size adjustment is based on equal-
weighted size decile portfolios. The benchmark portfolios are formed monthly using all
NYSE/AMEX firms available at that time. Each firm’s benchmark-adjusted return is
computed by subtracting the annual return of the appropriate benchmark portfolio (a
portfolio that corresponds to the industry grouping of the stock, or the size decile of the
stock, as of the portfolio formation date) from the individual stock’s annual return.
Annual benchmark-adjusted portfolio returns are computed as an equal-weighted average
of the adjusted returns of individual stocks.

15
The bottom row of each panel reports the annual returns to the price momentum strategy,
after controlling for trading volume (R10-R1). The results in Panels A, B, and C show
that the price momentum effect dissipates after 12 months in all three volume groups. As
we noted earlier, the spread between winners and losers (R10-R1) is higher for high
volume firms in the first year. However, this effect does not persist beyond Year 1.

The last five columns of Table VI report the difference between high and low volume
firms (V3-V1), controlling for price momentum. The results show that low volume losers
outperform high volume losers for each of the next five years. On the winner side, low
volume stocks take a little longer to outperform high volume stocks. As we saw earlier,
the difference in returns between high and low volume winners is not significant in the first
year. However, low volume winners begin to outperform high volume winners in Year 2
and this difference is seen through to Year 5.

Moskowitz and Grinblatt (1999) show that a portion of the returns from momentum
strategies is due to industry effects. Panel B shows that industry adjustment decreases
first year price momentum returns in our sample from 12.5 percent to an average of 10.1
percent (also see Table VII), a decline of about 20 percent. More importantly, industry
adjustments have virtually no effect on the volume results. The last five columns of Panel
B show that low volume firms continue to outperform high volume firms in the next five
years even after industry adjustment. This effect is clearly seen in both winner and loser
portfolios, and is also robust to firm size adjustments (Panel C). Thus, trading volume
does not appear to be a proxy for firm size or industry effects.

F. Price Momentum Reversals

Table VI results (see Panel B and Panel C) suggest that the magnitude and persistence of
price momentum is a function of past trading volume. Price reversals are more
pronounced among low volume losers (R1V1) and high volume winners (R10V3).
Conversely, price momentum is more pronounced among high volume losers (R1V3) and

16
low volume winners (R10V1). These observations suggest two volume-based price
momentum strategies. We refer to the first, which involves buying low volume winners
and selling high volume losers, as the early-stage strategy, to capture the idea that stocks
in these portfolios exhibit future price momentum over a longer horizon. We refer to the
second strategy, which involves buying high volume winners and selling low volume
losers, as the late-stage momentum strategy to capture the notion that the price
momentum in these stocks reverses faster.

Table VII compares the annual returns of the simple price momentum strategy (Simple) to
those of the early-stage (Early) and the late-stage (Late) strategies. Panel A shows that
the Simple strategy earns 12.5 percent in Year 1, but the momentum dissipates after 12
months. The Late strategy earns 6.8 percent in Year 1, but immediately begins losing in
subsequent years. In contrast, the Early strategy earns significant positive returns for
Years 1, 2, and 3 before the effect dissipates. Compared to the Simple strategy, Early
(Late) momentum strategies earn significantly higher (lower) returns in each of the next
four years. Panels B and C show that these effects are robust when returns are adjusted
for industry and size effects. Panel D shows the effect is weaker but still quite evident,
when both firm size and book-to-market are controlled for. Similarly, Panels E through G
show this effect holds for various alternative partitions of the data and even for firms in
the largest 50 percent of the NYSE/ AMEX population.

Figures 1 and 2 provide graphical representations of buy-and-hold returns to these three

strategies.17 Figure 1 reports buy-and-hold industry-adjusted returns while Figure 2
reports buy-and-hold size-adjusted returns. These graphs show that both long-horizon
underreaction and overreaction can occur in the data, and indeed can be reconciled
through judicious use of past trading volume. Looking at the returns for late-stage stocks
in isolation, we might be tempted to conclude that price momentum is an overreaction to
fundamental news. Yet the same graph shows that the early-stage momentum stocks
exhibit price continuation for three to five years. Looking at the early-stage momentum
stocks in isolation, we might conclude markets generally underreact to information. In

17
fact, both effects are part of a more general process by which information is incorporated
into prices. More generally, this evidence shows that the duration and magnitude of price
momentum can be predicted based on firm characteristics, such as trading volume.

Table VIII provides additional evidence on the timing of momentum reversals conditional
on trading volume and firm size. This table reports the time-series average of slope
coefficients estimated from monthly Fama-MacBeth cross-sectional regressions of the
following model:

rt + K ,i = a K + bK rt ,i + ut + K ,i

where subscript i refers to stock i, rt+K,i is the annual return K years ahead and rt,i is prior
year's (pre-portfolio formation) return, where K=1,2,3,4, or 5. The time-series average of
bK is an estimate of the average autocorrelation (across all stocks) between last year’s
return and future returns. The cross-sectional regression is run each month using all stocks
available at the beginning of the month, and the standard errors of the time-series means
are computed using the Hansen and Hodrick (1980) correction with 11 lags.

Reported table values represent the average slope coefficient estimated with various sub-
samples. The first two columns provide a more formal test of the momentum reversal
phenomenon reported in Table I. In column 1 involving all stocks, the slope coefficient is
positive and significant in Year 1, negative and insignificant in Years 2 and 3, and negative
and significant in Years 4 and 5. These return autocorrelation patterns confirm the
presence of price momentum in Year 1 and strong price reversals in Years 4 and 5.
Column 2 provides similar evidence using only winner and loser stocks.

Columns 3 and 4 report slope coefficient estimates when only early stage or late stage
stocks, based on past trading volume, are included in the regressions. These results show
that the Year 1 price momentum is much stronger for early stage firms. Moreover, early
stage firms show significant price momentum up to the third year, while late stage firms

18
show strong price reversal starting in Year 2. Evidently, the magnitude and persistence of
price momentum is a function of trading volume. We obtain similar results using size or
industry adjusted returns (not reported in the paper).

Columns 5 and 6 conduct the same test using firm size rather than trading volume as the
conditioning variable. The results for this subset of firms are weak and inconsistent,
indicating that firm size is not a good substitute for trading volume in the prediction of
return autocorrelation patterns. In other words, the information conveyed by trading
volume about the persistence of future price momentum is not driven by its correlation
with firm size.

It is useful to summarize the empirical facts at this point. Thus far we have seen that low
volume stocks generally earn higher returns than high volume stocks. This fact is
consistent with the liquidity effect. However, we have also seen that the price momentum
effect is stronger among high volume stocks, raising questions about the liquidity
explanation. We find that past trading volume predicts the magnitude and timing of price
momentum reversals. Finally, we have seen evidence that the information in trading
volume is not about firm size or industry effects. In the next section, we provide
additional evidence on the nature of the information provided by trading volume.

IV. Information Content of Trading Volume

Why does trading volume predict future returns? Is it a proxy for differences in liquidity,
or the rate of information diffusion, or something else? In this section, we conduct
additional tests to help answer these questions. Our goal is to better understand the
information content of trading volume, and to evaluate this evidence in the light of existing
models of investor behavior.

A. Volume as a Liquidity Proxy

First we explore the relationship between trading volume (turnover) and other proxies of
liquidity. While average daily dollar volume is an intuitive proxy for liquidity, it does not

19
necessarily follow that the average daily turnover is also a liquidity proxy. The turnover
measure we use is, in effect, average daily dollar volume scaled by a firm's total market
capitalization. The effect of dividing by firm size is to create a volume measure that may
not have strong correlations with traditional liquidity proxies.

To provide more direct evidence on the relation between liquidity and turnover, the
following table reports cross-sectional Spearman rank correlations of trading volume to
firm size, stock price, and relative spread.18 The sample period is 1964 to 1995, except for
the relative spread results, which are based on the 1979 to 1989 time-period (the data is
the same as that used in Eleswarapu and Reinganum (1993)).

Firm Size Stock Price Relative Spread

Turnover 0.20 0.11 -0.12

This chart shows that trading volume (as measured by average daily turnover) is not highly
correlated with common proxies for market liquidity. The low degree of correlation with
these variables suggests turnover may be providing information about something other
than market liquidity.

Guidance on where to look emerges from the results in Table V. Recall Table V shows
that the returns on low volume stocks are more positively correlated with HML than are
returns on high volume stocks. In other words, high volume stocks behave like glamour
stocks while low volume stocks behave like value stocks. Thus, trading volume seems to
provide information about relative under- or over-valuation of stocks. In the next section,
we investigate this possibility by conducting additional tests that focus on a possible link
between trading volume and measures of under- or over-valuation.

B. Firm Characteristics Related to Profitability and Under- or Over-valuation

Table IX reports the pattern of past and future profitability, and firm characteristics that
proxy for under- or over-valuation across high and low volume portfolios. Panel A

20
reports the results when firms are divided into ten price momentum and three volume
portfolios (the 10 × 3 partition), and Panel B reports the results when firms are divided
into five price momentum and five volume portfolios (the 5 × 5 partition).

Each panel reports the number of analysts following the firm (NANA), the forecasted
long-term earnings growth rate (Ltg), the cumulative buy-and-hold returns over the five
years prior to the portfolio formation date (LtRet), and the book-to-market ratio just
before the formation date (B/M). In addition, Table IX provides the return-on-equity
from the most recent fiscal year end (ROE(0)), as well as the change in ROE over the last
three years (DROE(-)) and the next three years (DROE(+)). All values represent time-
series averages of portfolio medians. For NANA and Ltg, we used the subset of firms
covered by I/B/E/S (sample period 1979 to 1995).

The most striking fact that emerges from Table IX is that high volume stocks are generally
glamour stocks and low volume stocks are generally value or neglected stocks. For
example, Panel B shows that high volume stocks have greater analyst coverage (NANA is
3.6 for low volume losers and 9.6 for high volume losers), have higher forecasted earnings
(long-term growth per year, Ltg, is 9.33 percent for low volume losers and 12.85 percent
for high volume losers), lower book-to-market ratios (B/M is 0.815 for high volume losers
and 1.125 for low volume losers), and higher return-on-equity (ROE(0) is 11.3 percent
for low volume losers and 7.3 percent for high volume losers). These differences are all
statistically significant. The results for winner portfolios are symmetrical and comparable
in magnitude.

Table IX shows that high volume firms and low volume firms also differ significantly in
terms of their past operating and price performance. In general, low volume losers have
experienced a greater decline in ROE over the past three years compared to high volume
losers. The pattern is symmetrical, but reversed in direction, among winners: high volume
winners have experienced an increase in ROE while low volume winners have experienced
a decline in ROE. In spite of this, high volume stocks – both winners and losers – have

21
significantly worse operating performance (significantly lower ROE increases) in the
future than low volume stocks. This result is striking, because, ex ante, analysts forecast
higher earnings growth (Ltg) for high volume firms. High volume firms have also
experienced a recent increase in ROE. In other words, analysts seem to consistently over-
estimate (under-estimate) the future profitability of high (low) volume firms, perhaps
because they naïvely extrapolate recent operating performance.

Finally, over the past five years, high volume losers (winners) have significantly
outperformed low volume losers (winners). For example, in Panel B, LtRet is 28.97
percent for low volume losers and 108.87 percent for high volume losers. Among winners,
LtRet is 133.53 percent for low volume stocks and 247.62 percent for high volume stocks.
Thus, high volume winners and low volume losers are long-term winners and long-term
losers respectively, whereas low volume winners and high volume losers are more recent
winners and losers.

Figure 3 provides further evidence on the past and future performance of volume based
portfolio strategies. This figure examines the average annual return for various volume
portfolios from Year -4 to Year +5 around the portfolio formation date. Figure 3a shows
that while high volume firms (V5) underperform low volume firms (V1) in the future they
have outperformed low volume firms in the past. Thus, high volume stocks appear to be
long-term winners relative to low volume stocks. This is consistent with the result in
Table IX that high volume stocks have lower B/M ratios than low volume stocks.

The pattern across high volume and low volume stocks is seen in winner (Figure 3b) and
loser (Figure 3c) portfolios as well. Once again, high volume stocks earned higher returns
than low volume firms in each of the past five years. 19 Figure 3b shows that among
winners the difference in performance between low volume and high volume stocks is
most pronounced in the immediate past (Year 0). Specifically, high volume winners have
been "long-term" winners, while low volume winners are only "recent" winners.

22
Figure 3c shows that among losers the performance gap is most pronounced in the more
distant past (on or before Year -1). Specifically, this figure provides striking evidence that
prior to Year 0, high volume losers have in fact been big winners (+37.34 percent in Year
–1 Vs –17.30 percent in Year 0). Therefore, it is very appropriate to refer to high volume
losers as "recent" losers. In contrast, low volume losers have been underperforming
consistently over the last five years, indicating that they are "long-term" losers.

These results fit our earlier characterization of volume based momentum strategies as
early and late stage strategies. Specifically, low volume winners only became winners in
the recent past, and exhibit positive momentum for a longer time in the future. Similarly,
high volume losers only became losers in the recent past and exhibit negative momentum
for a longer time in the future. In contrast, high volume winners and low volume losers
are "long-term" winners and losers, respectively. Our earlier results show they tend to
exhibit faster reversals in the future.

In sum, Table IX shows that past trading volume is related to a firm's past performance
measures, current valuation ratios, and analysts' future forecast errors. Along all these
dimensions, low (high) volume firms display value (glamour) characteristics. Controlling
for price momentum, low (high) volume firms have underperformed their peers in the past,
they possess lower (higher) valuation ratios today, and tend to over (under) perform
analyst expectations in the future.

C. Abnormal Returns Around Quarterly Earnings Announcements

Table IX shows that analysts are more optimistic (pessimistic) for high (low) volume
stocks, but future changes in profitability fail to meet expectations. Thus, trading volume
seems to provide information about investors' misperception of future earnings. In this
subsection, we further explore this issue by examining the stock price reactions around
future quarterly earnings announcements. If trading volume serves as a proxy for investor
misperceptions of future earnings, these misperceptions should correct themselves around
subsequent earnings announcement dates. Specifically, we might expect to see more

23
negative (positive) price reactions for high (low) volume stocks. Risk differences should
have little effect on short-window returns. Therefore, this technique provides a direct test
that distinguishes between the mispricing hypothesis and other risk-based explanations.

Table X reports the abnormal returns around quarterly earnings announcements for
various price momentum-trading volume portfolios. Table values represent four-day (day
-2 to +1) cumulative abnormal returns (CAR) in percent around quarterly earnings
announcements. Returns are reported for the eight quarters before and after the most
recent earnings announcement date just prior to portfolio formation. The NYSE/ AMEX/
Nasdaq value-weighted index is used as the benchmark in computing the CAR. The
numbers in parentheses are t-statistics computed using the Hansen and Hodrick (1980)
autocorrelation correction with six moving average lags. The six-month strategy (J=6,
K=6) results are reported. Results for other holding and formation periods are similar.

Row 1 of Panel A shows that loser (R1) portfolios have significant negative earnings
announcement abnormal returns in the past three quarters (-2 to 0). These losers continue
to exhibit losses in the next two quarters, before staging a modest recovery in quarters t+4
to t+8. The next two rows show that the recovery among losers is driven almost entirely
by the low volume (late-stage) stocks. The high volume losers keep losing for three
quarters and exhibit no significant CARs beyond quarter t+3. The difference in CARs
between high volume and low volume losers averages more than one percent per
announcement and is highly significant for each of the next eight quarters.

Panel B shows a similar pattern for the winner portfolios. Specifically, low volume (early-
stage) winners experience significant positive announcement period CARs for the next
eight quarters. High volume (late-stage) winners, on the other hand, experience negative
CARs starting from quarter t+4. The difference in CARs between low volume winners
and high volume winners averages between 0.50 percent to 1.22 percent per
announcement and is highly significant for each of the next eight quarters. Taken
together, the evidence in Panels A and B show clearly that the ability of low volume

24
stocks to outperform high volume stocks is related to the better earnings news received by
the low volume stocks in the future.

Panel C reports abnormal returns when firms are sorted only on the basis of past trading
volume. The bottom row shows that high volume (V3) and low volume (V1) firms do not
have significantly different returns around earnings announcements in the past (quarters -8
to -1). However, low (high) volume firms exhibit significantly more positive (negative)
earnings announcement returns in the future (quarters 0 to +8). On average, V1 firms
earn approximately 0.60 percent more than do V3 firms around each of the next eight
quarterly announcements. Although significant, this difference is much lower than what
was reported in Panels A and B. In sum, Panel C shows that an independent volume
effect exists, but that the volume effect is most pronounced among extreme winners and
losers.

Panel D augments these findings by comparing the announcement period CARs from three
different trading strategies. The results in this panel show that the early-stage momentum
strategy (R10V1-R1V3) has significantly more positive announcement period returns than
a simple price momentum strategy (R10-R1) in each of the next eight quarters.
Conversely, the late-stage strategy (R10V3-R1V1) results in sharply lower earnings
announcement period returns than the simple price momentum strategy.

In sum, we find that the short-window returns also support the view that trading volume
provides information about market misperceptions of future earnings. Specifically, short-
window earnings announcement returns are more positive for low volume stocks than for
high volume stocks. We observe this difference for the next eight quarters. The effect is
strong for both winners and losers.

D. Changes in Trading Volume

As a final test, we form portfolios based on price momentum and changes in trading
volume. Our goal is to compute a volume measure that purges each firm of its normal

25
level of trading activity (i.e., a measure of the abnormal trading volume). If the
information content of trading volume is due to intertemporal variations in a firm's normal
trading activity, this measure should also predict returns. Furthermore, by using changes
in trading volume, we address any lingering concern that our original results are driven by
liquidity effects.

In Table XI, we replicate our industry-adjusted return prediction tests, replacing trading
volume with the actual change in trading volume. To construct this table, firms are
independently sorted into five price momentum portfolios and five portfolios based on
changes in trading volume over the past four years (∆V). Specifically, if the 12 month
period just prior to the portfolio formation date is defined as year t, then we define change
in volume as the average daily turnover over the past six months minus the average daily
turnover in year t-4 (∆V=V(6,t)-V(t-4)).20 Using percentage change in trading volume
rather than actual change ((∆V=(V(6,t)-V(t-4))/V(t-4)) yields very similar results. We
find that the level of trading volume is positively correlated with the change in trading
volume. The Spearman rank correlation between these two variables is 0.48.

Table XI shows that portfolios ranked on price momentum and changes in trading volume
exhibit the same patterns in future returns as those ranked on price momentum and level of
trading volume. For example, the bottom row shows that returns to simple price
momentum strategies dissipate in 12 months. However, the last five columns show that
firms with the most increase in volume significantly underperform firms with the least
increase (or the most decline) in volume. The difference ranges from two percent to five
percent over the next five years and is equally strong in winner and loser portfolios.

Table XII compares the predictive power of 1) average trading volume from the past six
months, 2) changes in trading volume over the past four years, and 3) lagged trading
volume from four years ago (we ensure comparability by using only a sub-sample of
stocks for which all three volume measures are available). Panel A reports early and late
strategy returns based on last six-month trading volume. Panel B reports the results for

26
changes in trading volume measured relative to the trading volume in Year t-4. Finally,
Panel C reports the results using only trading volume from Year t-4.

The last two rows of each panel reports the incremental returns to the volume metric,
controlling for price momentum. These results show that most of the predictive power
comes from changes in trading volume, rather than lagged volume. The last two rows of
Panel C show that lagged volume from four years ago has some predictive power for
future returns, but the effect is not statistically significant in any of the next five years.
Conversely, Panel B shows that the change in trading volume has significant incremental
predictive power. With two exceptions (Years 2 and 3 in the early strategy), this
predictive power is statistically significant over each of the next five years for both the late
strategy and the early strategy.

It is important to recognize the imprecise nature of this test. Specifically, this test assumes
that we can parse past trading volume into a "normal" and an "abnormal" component using
a fixed time interval of four years for all firms. This is a strong assumption, and the
imprecision it introduces may explain why both the changes and lagged volume variables
have some predictive power. Despite this limitation, Table XII results show that most of
the predictive power of trading volume is attributable to recent changes in the level of
trading activity rather than lagged volume. This evidence further supports the notion that
past turnover is a measure of fluctuating investor sentiment, and not a liquidity proxy.

E. Relation to Existing Behavioral Models of Under- and Over-Reaction

Our results show trading volume is an important empirical link between intermediate-
horizon momentum and long-horizon return reversal. Recently, several behavioral models
have attempted to provide a framework for integrating these two empirical phenomena
(e.g., Daniel, Hirshleifer, and Subrahmanyam (1998) (DHS), Barberis, Shleifer, and
Vishny (1998) (BSV), and Hong and Stein (1999) (HS)). In this subsection, we briefly
summarize each model and discuss their relation to our findings.

27
Daniel, Hirshleifer, and Subrahmanyam (1998) focus on the overconfidence bias. They
argue that stocks that are more difficult to value tend to generate greater overconfidence
among investors. Therefore, according to their model, mispricing should be more severe
among securities that are hard to value (i.e., growth or glamour stocks) or where feedback
is slow or ambiguous (i.e., small, illiquid stocks). If high volume stocks tend to be growth
or glamour stocks, then DHS would predict that price momentum profits should be
stronger among the high volume stocks. This is consistent with our finding that high
volume stocks tend to behave like glamour stocks (see Table V and for more direct
evidence, Table VIII). It may also help explain our intermediate horizon finding that
momentum spreads (profits on R10-R1 strategies) are greater among high volume firms.

In Barberis, Shleifer, and Vishny (1998), the conservatism bias of the representative
investor causes him to update his priors insufficiently when he observes new public
information about a firm. This leads to an initial market underreaction. However, due to
the representativeness bias, when an investor receives a long sequence of good (or bad)
news he tends to become too optimistic (or pessimistic) about the future profitability of
the firm. As a result, firms experiencing prolonged periods of increasing earnings tend to
become over-valued, and those experiencing long periods of declining earnings tend to
become under-valued. The prices of these stocks ultimately undergo reversals as realized
earnings fail to meet expectations.

In Hong and Stein (1999) there are two types of investors: news watchers and momentum
traders. The news watchers trade only on private information about fundamentals, while
the momentum traders trade only on past price movements. Both are boundedly rational
in the sense they ignore all other information. Given these rationality constraints, HS
show that if firm specific information diffuses gradually across news watchers, there will
be an initial underreaction. This underreaction in turn allows momentum traders to make
money by trend chasing. As more and more momentum traders arrive in the market, the
initial underreaction inevitably turns into overreaction at longer horizons. In short, HS

28
provides a context for reconciling the dynamics of intermediate-horizon underreaction and
long-horizon overreaction.

The main appeal of these models is their synthesis of intermediate-horizon underreaction

and long-horizon overreaction. Each model presents a plausible explanation for these
empirical observations. In addition, each model has specific features that help explain
some aspect of our findings. However, the main limitation of these models, as they pertain
to our tests, is that none have an explicit role for trading volume. Therefore, the
directional predictions we discuss below are inferred from each model's underlying
assumptions.

The models fall into two camps in terms of their explanation of the intermediate-horizon
momentum effect. In DHS (as well as in DeLong, Shleifer, Summers, and Waldmann,
1990, (DSSW)), prices initially overreact to news about fundamentals, and continue to
move further away, before ultimately reverting to fundamentals. Therefore, in DHS and
DSSW, the positive autocorrelation in intermediate horizon returns is due to a market
overreaction. In contrast, both BSV and HS characterize the intermediate horizon
momentum effect as a market underreaction. In BSV, the underreaction arises because
the representative investor does not update sufficiently when he observes a firm-specific
public news event. In HS, insufficient diffusion of information across news watchers
results in a gradual incorporation of information into prices.

Our volume-based results do not fit neatly into either of these frameworks. For example,
the HS model predicts that momentum profits should be larger for stocks with slower
information diffusion. If we make the assumption that scarcity of trading leads to
insufficient diffusion of information, then the HS model would predict a greater
momentum effect among low volume stocks. Our results indicate this to be true among
winners, but not among losers. That is, low volume winners have greater momentum, but
low volume losers actually have less momentum. In addition, our results show that price
momentum strategies actually perform better among high volume stocks. Therefore, the

29
evidence does not seem to support the view that volume is an information diffusion proxy
at intermediate and long horizons.

Conversely, in DHS and DSSW, the implicit assumption is that high trading volume will
"fuel" momentum. For example, in DSSW, momentum arises from positive feedback
traders that seek to capitalize on an initial price move by buying (selling) on good (bad)
news. If we assume that trading volume is a proxy for positive feedback trading, or the
activity of overconfident traders, then these models predict greater momentum among
high volume stocks (in the case of DHS this is because high volume stocks are glamour
stocks that are more difficult to value). We find this is true among losers but not among
winners both in intermediate and long horizons. High volume losers do continue to lose
longer (and lose more) than low volume losers. However, among winners, the opposite is
true: high volume winners continue to win for a shorter period than low volume winners;
indeed, high volume winners do worse than low volume winners over the next two to five
years. Thus, the fact that at intermediate horizons, momentum profits (R10-R1) are
higher among high volume stocks is not because volume “fuels” price momentum.

F. Momentum Life Cycle (MLC)

An intriguing explanation for the above findings is depicted in Figure 4. This figure
presents a simple conceptual diagram that helps to integrate the evidence in this paper.
We refer to this diagram as the Momentum Life Cycle (MLC) hypothesis.21 The main
benefit of this graph is that it presents the interaction between price momentum, reversals,
and trading volume in a single framework. The main disadvantage is that it implies more
rigidity and regularity than is warranted by the evidence to date. We present it here as an
intriguing possibility that merits further research.

According to this hypothesis, stocks experience periods of investor favoritism and neglect.
A stock with positive price and/or earning momentum (past winner) would be on the left
half of the cycle while a stock with negative price and/or earning momentum (past loser)
would be on the right half of the cycle. Growth stocks that experience positive news

30
move up the cycle, but eventually these stocks disappoint the market and are "torpedoed."
Stocks that disappoint begin a downward slide and eventually experience general neglect.
If they fall far enough in price, they may become attractive to contrarian investors.

Given this framework, our evidence suggests trading volume may provide information
useful in locating a given stock in its momentum/expectation life cycle. Generally, when a
stock falls into disfavor, its trading volume declines. Conversely, when a stock is popular,
its trading volume increases. Viewed in this light, trading volume provides information on
the degree of investor favoritism (or neglect) in a stock, or more precisely, the extent to
which market sentiment favors the stock at a particular point in time.

The MLC would characterize high volume winners and low volume losers as late stage
momentum stocks, in the sense that their price momentum is more likely to reverse in the
near future. Conversely low volume winners and high volume losers are early stage
momentum stocks, in the sense that their momentum is more likely to persist in the near
future. The MLC also implies that trading volume should be correlated with
value/glamour characteristics. As a stock moves up the cycle, trading volume increases
and it becomes more "expensive" in terms of price-to-value measures. The higher (lower)
number of analysts following high (low) volume stocks is also consistent with this
explanation. In fact, many of the relations between volume and value characteristics are
difficult to accommodate in any other explanation that we are aware of.

We wish to stress the limitations of Figure 4. We have shown that, on average, firms in
each of the four quadrants of this cycle exhibit characteristics that are consistent with the
MLC hypothesis. However, these results describe general tendencies at the portfolio
level. For individual firms, things are far less deterministic than the figure implies.
Individual firms do not necessarily exhibit expectation cycles of the same frequency. Nor
does each firm need to pass through all phases of the cycle each time. The turning points
for individual firms may appear random and difficult to pinpoint, even though the
portfolios in each quadrant conform to the predictions of the MLC hypothesis.

31
The MLC diagram also does not explain the asymmetric volume effect in Year 1.
Specifically, the fact that the price momentum effect is more pronounced among high
volume stocks is not predicted by this explanation. Nor does the diagram explain why
volume might decline as a stock falls out of favor. There are no extant models that
formally address this question. One possibility is the disposition effect, or the tendency of
investors to hold on to losing investments too long (see Odean (1998)). According to the
disposition effect, as a stock falls out of favor, investors who own the stock become more
reluctant to realize their losses. This unwillingness to sell by its owners, coupled with
general neglect from potential investors, may be the reason for the decline in trading
volume.

Another possibility, suggested by asymmetric information models, is that trading volume

captures investors’ disagreement about a stock’s intrinsic value. In general, glamour
stocks tend to be high growth stocks that are difficult to value. This could result in
greater disagreement among investors about their intrinsic values, and therefore higher
trading volume. Interpreted in this context, stocks at the bottom of the MLC tend to have
less investor disagreement, while stocks at the top of their MLC (late winners and early
losers) tend to have more investor disagreement. The question that remains is why the
degree of investor disagreement would vary over the MLC. Clearly, a more complete
theoretical framework would be helpful.

V. Conclusion
Price and volume are simultaneously determined in equilibrium. Whatever process
generates price also gives rise to the accompanying trading volume. Trading volume is
also a widely available market statistic. Therefore, it is perhaps not surprising that both
financial academics (e.g., Blume, Easley, and O'Hara (1994)) and practitioners (e.g.,
various technical chartists) have recognized the potential usefulness of trading volume in
investment decisions. What is surprising is how little we really know about trading
volume.

32
In this study, we have begun the process of understanding the role of trading volume in
the prediction of cross-sectional stock returns. Our findings establish several important
regularities about the role of trading volume in predicting cross-sectional returns. First,
we show that trading volume, as measured by the turnover ratio, is unlikely to be a
liquidity proxy. Although high (low) volume firms earn lower (higher) future returns, the
opposite is true in the past. Trading volume is not highly correlated with firm size or
relative bid-ask spread, and the volume effect is independent of the firm size effect.

Rather, our evidence shows that the information content of trading volume is related to
market misperceptions of firms' future earnings prospects. Specifically, we provide strong
evidence that low (high) volume stocks tend to be under (over) valued by the market.
This evidence includes past operating and market performance, current valuation multiples
and operating performance, as well as future operating performance and earnings
surprises. One implication of our finding is that investor expectations affect not only a
stock’s returns, but also its trading activity.

Second, our results show that the effect of trading volume on price momentum is
more complex than prior research suggests. Neither of the two most common
views about volume's effect on price momentum (i.e., the "fueling" hypothesis and
the "diffusion" hypothesis) captures the stylized facts. In fact, we find volume
"fuels" momentum only for losers, and it helps information "diffusion" only for
winners. These facts should provide further guidance to researchers interested in
modeling the market dynamics that give rise to returns and volume.

Third, we show that the price momentum effect reported by Jegadeesh and Titman (1993)
eventually reverses, and that the timing of this reversal is predictable based on past trading
volume. Specifically, we show that it is possible to create JT-type momentum portfolios
(winners minus losers) that exhibit long-horizon return reversals of the type first
documented by DeBondt and Thaler (1985). This finding represents an important

33
conceptual shift in the literature. Previous studies have generally viewed intermediate-
horizon momentum and long-horizon price reversal as two separate phenomena. Our
results show that trading volume provides an important link between these two effects.

Finally, we show that existing theories of investor behavior do not fully account for all of
the evidence. Models presented in Daniel, Hirshleifer, and Subrahmanyam (1998),
Barberis, Shleifer, and Vishny (1998), and Hong and Stein (1999) capture the spirit of our
findings, in that they model prices as initially underreacting, and ultimately overreacting, to
fundamental news. However, none of these models incorporate trading volume explicitly
and, therefore, they cannot fully explain why trading volume is able to predict the
magnitude and persistence of future price momentum.

To summarize our results, we suggest a simple conceptual diagram, which we have

dubbed the momentum life cycle (MLC) hypothesis. According to the MLC hypothesis,
firms move through periods of relative glamour and neglect. We suggest that trading
volume may play a useful role in identifying where a stock is in this cycle. When stocks
decline in popularity, their trading volume drops and they become neglected. When stocks
increase in popularity, their trading volume increases.22

Our findings have important implications for the debate on market efficiency. The ability
of past trading volume to predict future returns (and earnings surprises) implies prices do
not generally equal fundamental values. Indeed, our results suggest that the market is
better characterized as being in a constant state of convergence toward intrinsic value.23
Viewed in this light, intermediate-horizon “underreaction” and long-horizon
“overreaction” are simply two elements of the same continuous process by which prices
impound new information. This characterization of the price adjustment process is
consistent with our findings and with the behavioral models we discuss in this paper.

Our results also raise at least three interesting questions for future research. First,
the asymmetry in the timing of momentum reversals between winners and losers

34
remains a puzzle. We show that low volume losers rebound quickly, and
outperform high volume losers within the next three to 12 months. However, it
takes low volume winners longer (more than 12 months) to significantly
outperform high volume winners. We know of no explanation for this timing
difference. Second, with the possible exception of the disposition effect from the
behavioral literature, we know of no explanation for why trading volume should
decline when firms fall out of favor. We believe more robust models of investor
behavior, which incorporate fluctuations in the level of trading activity, are needed
to explain this finding.

Finally, we find it remarkable that measures as readily available as past returns and trading
volume can have such strong predictive power for returns. The magnitude of these
returns is likely to be lower under practical implementation. However, given the
popularity of price momentum strategies, the improvement gained by also conditioning on
past volume appears economically significant. Why this information is not fully reflected
in current prices is another puzzle we leave for future research. In the meantime, we
remain agnostic as to the prediction that this phenomenon will yield positive abnormal
returns in future periods.

35
References

Amihud, Yakov and Haim Mendelson, 1986, Asset pricing and the bid-ask spread,
Journal of Financial Economics 17, 223-249.

Asness, Clifford A., 1997, The interaction of value and momentum strategies, Financial
Analysts Journal 53, 29-36.

Ball, Ray and S. P. Kothari, 1989, Nonstationary expected returns: Implications for tests
of market efficiency and serial correlation in returns, Journal of Financial
Economics 25, 51-74.

Ball, Ray, S. P. Kothari, and Jay Shanken, 1995, Problems in measuring portfolio
performance: An application to contrarian investment strategies, Journal of
Financial Economics 38, 79-107.

Barberis, Nicholas, Andrei Shleifer, and Robert Vishny, 1998, A model of investor
sentiment, Journal of Financial Economics 49, 307-343.

Bernstein, Richard, 1993, The earnings expectations life cycle, Financial Analysts
Journal, 49, 90-93.

Bernstein, Richard, 1995, Style Investing:Unique Insight Into Equity Management (John
Wiley & Sons, New York).

Blume, Lawrence, David Easley, and Maureen O’Hara, 1994, Market statistics and
technical analysis: The role of volume, Journal of Finance 49, 153-181.

36
Brennan, Michael J., Tarun Chordia, and Avanidhar Subrahmanyam, 1998, Alternative
factor specifications, security characteristics, and the cross-section of expected stock
returns, Journal of Financial Economics 49, 345-373.

Campbell, John Y., Sandord J. Grossman, and Jiang Wang, 1993, Trading volume and
serial correlation in stock returns, Quarterly Journal of Economics 107, 905-939.

Chan, Louis K., Narasimhan Jegadeesh, and Josef Lakonishok, 1996, Momentum
strategies, Journal of Finance 51, 1681-1713.

Chopra, Navin, Josef Lakonishok, and Jay Ritter, 1992, Measuring abnormal returns: Do
stocks overreact? Journal of Financial Economics 31, 235-268.

Chordia, Tarun, and Bhaskaran Swaminathan, 1999, Trading volume and cross-
autocorrelation in stock returns, Journal of Finance 54, forthcoming.

Conrad, Jennifer S., Allaudeen Hameed, and Cathy Niden, 1994, Volume and
autocovariances in short-horizon individual security returns, Journal of Finance 49,
1305-1330.

Conrad, Jennifer S., and Gautam Kaul, 1993, Long-term market overreaction or biases in
computed returns? Journal of Finance 48, 39-64.

Conrad, Jennifer S. and Gautam Kaul, 1998, An anatomy of trading strategies, Review of
Financial Studies 11, 489-519.

Daniel, Kent, David Hirshleifer, and Avanidhar Subrahmanyam, 1998, A theory of

overconfidence, self-attribution, and security market under- and overreactions,
Journal of Finance 53, 1839-1886.

37
Datar, Vinay, Narayan Naik, and Robert Radcliffe, 1998, Liquidity and asset returns: An
alternative test, Journal of Financial Markets 1, 203-220.

DeBondt, Werner F. M., and Richard Thaler, 1985, Does the stock market overreact?
Journal of Finance 40, 793-805.

DeBondt, Werner F. M., and Richard Thaler, 1987, Further evidence of investor
overreaction and stock market seasonality, Journal of Finance 42, 557-581.

De Long, J. Bradford, Andrei Shleifer, Lawrence H. Summers, and Robert J. Waldmann,

Positive feedback investment strategies and destabilizing rational speculation,
Journal of Finance 45, 379-395.

Eleswarapu, Venkat R., and Mark Reinganum, 1993, The seasonal behavior of the
liquidity premium in asset pricing, Journal of Financial Economics 34, 373-386.

Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on
stocks and bonds, Journal of Financial Economics 33, 3-56.

Fama, Eugene F., and Kenneth R. French, 1996, Multifactor explanations of asset pricing
anomalies, Journal of Finance 51, 55-84.

Fama, Eugene F., and Kenneth R. French, 1997, Industry costs of equity, Journal of
Financial Economics 43, 153-193.

Gould, John F., and Alan W. Kleidon, 1994, Market maker activity on Nasdaq:
Implications for trading volume, Stanford Journal of Law, Business & Finance 1, 1-
17.

38
Grundy, Bruce D., and J. Spencer Martin, 1998, Understanding the nature of the risks and
the source of the rewards to momentum investing, Working paper, the Wharton
School.

Hansen, Lars P., and Robert J. Hodrick, 1980, Forward exchange rates as optimal
predictors of future spot rates: An econometric analysis, Journal of Political
Economy, 88, 829-853.

Hong, Harrison, and Jeremy C. Stein, 1999, A unified theory of underreaction, momentum
trading and overreaction in asset markets, Journal of Finance, forthcoming.

Jegadeesh, Narasimhan, 1990, Evidence of predictable behavior of security returns,

Journal of Finance 45, 881-898.

Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying winners and selling
losers: Implications for stock market efficiency, Journal of Finance 48, 65-91.

Lakonishok, Josef, and Seymour Smidt, 1986, Volume for winners and losers: Taxation
and other motives for stock trading, Journal of Finance 41, 951-974.

Lee, Charles M. C., James Myers, and Bhaskaran Swaminathan, 1999, What is the
intrinsic value of the Dow? Journal of Finance 54, forthcoming.

Lehmann, Bruce N., 1990, Fads, martingales, and market efficiency, Quarterly Journal of
Economics 60, 1-28.

Moskowitz, Tobias J., and Mark Grinblatt, 1999, Do industries explain momentum?
Journal of Finance 54, 1249-1290.

39
Odean, Terrance, 1998, Are investors reluctant to realize their losses? Journal of Finance
53, 1775-1798.

Rouwenhorst, Geert K., 1998, International momentum strategies, Journal of Finance 53,
267-284.

40
 Appendix
Industry Benchmarks
To control for industry effects in our return calculations, we construct 25 equal-weighted
industry portfolios. The industry portfolios are formed monthly, from January 1965 to
December 1995, using 2-digit CRSP SIC codes. The following table lists the industry
groupings and their corresponding SIC codes. All NYSE/AMEX firms available at the
time of portfolio formation are included. The benchmark-adjusted returns are computed
by subtracting the annual returns of the appropriate benchmark portfolio (a portfolio that
corresponds to the industry grouping of the stock at the time of the portfolio formation)
from each individual stock’s annual returns. The annual portfolio returns are computed as
an equal-weighted average of the returns of individual stocks.

Industry SIC Code

1 Agriculture, forestry and fishing 01-09
2 Mining, minerals, oil and gas 10-14
3 Construction 15-17, 25, 32
4 Food and beverage 20
5 Tobacco products 21
6 Textile and apparels 22-23
7 Paper products 26
8 Printing and publishing 27
9 Chemicals 28
10 Petroleum 29
11 Rubber 30
12 Leather 31
13 Primary and fabricated metals 33-34
14 Machinery and electrical equipment 35-36
15 Transportation equipment 37
16 Manufacturing 38-39
17 Transportation 40-47
18 Communication and Utilities 48-49
19 Wholesale 50-51
20 Retail 52-59
21 Finance and real estate 60-67
22 Services 70-76, 81-89
23 Entertainment 78-79
24 Health care 80
25 Other >89
 Buy-and-Hold Industry Adjusted Abnormal
Returns: 1965-1995

40.00

30.00

20.00
Return in %

Early
10.00 Simple
Late
0.00

-10.00

-20.00
0 1 2 3 4 5 6

Year

Figure 1. Buy-and-hold industry adjusted long-term returns to various momentum

strategies. This graph depicts buy-and-hold industry-adjusted returns to three price momentum
strategies, formulated using past returns and trading volume from the previous six months (J=6).
Each month, stocks are independently sorted into 10 price momentum portfolios and 3 volume
portfolios. The simple price momentum strategy (Simple) buys top decile winners and sells
bottom decile losers (R10-R1). The early-stage momentum strategy (Early) buys low volume
winners and sells high volume losers (R10V1-R1V3). The late stage momentum strategy (Late)
buys high volume winners and sells low volume losers (R10V3-R1V1). The industry adjustment
is based on 25 equal-weighted industry portfolios formed by grouping 2-digit SIC codes (see
Appendix). Each firm’s benchmark-adjusted return is computed by subtracting the annual return
of the appropriate benchmark portfolio (a portfolio that corresponds to the industry grouping of
the stock as of the portfolio formation date) from the individual stock’s annual return.
 Buy-and-hold Size-Adjusted Abnormal Returns: 1965-
1995
40.00

35.00

30.00

25.00
Return in %

20.00
Early
15.00 Simple
10.00 Late

5.00

0.00

-5.00

-10.00
0 1 2 3 4 5 6

Year

Figure 2. Buy-and-hold size adjusted long-term returns to various momentum strategies.

This graph depicts the buy-and-hold size-adjusted returns to three price momentum strategies,
formulated using past returns and trading volume from the previous six months (J=6). Each
month, stocks are independently sorted into 10 price momentum portfolios and 3 volume
portfolios. The simple price momentum strategy (Simple) buys top decile winners and sells
bottom decile losers (R10-R1). The early-stage momentum strategy (Early) buys low volume
winners and sells high volume losers (R10V1-R1V3). The late stage momentum strategy (Late)
buys high volume winners and sells low volume losers (R10V3-R1V1). The size adjustment is
based on 10 equal-weighted size. Each firm’s benchmark-adjusted return is computed by
subtracting the annual return of the appropriate benchmark portfolio (a portfolio that corresponds
to the size decile of the stock as of the portfolio formation date) from the individual stock’s
annual return.
 Low Volume Stocks (V1) and High Volume Stocks (V5)

30.00

20.00

10.00

0.00
-4 -3 -2 -1 0 1 2 3 4 5

V1 15.68 14.05 12.88 10.84 10.98 18.04 17.87 17.66 16.16 16.07
V5 21.72 24.20 26.49 30.71 26.56 12.84 13.65 13.09 12.00 13.80
Annual Return in %, Figure 3a.

Low Volume Winners (R5V1) and High Volume

Winners (R5V5)
80.00

60.00

40.00

20.00

0.00
-4 -3 -2 -1 0 1 2 3 4 5

R5V1 13.94 13.07 13.03 10.48 43.91 21.37 18.88 19.07 14.56 13.78
R5V5 16.79 21.31 21.07 25.09 85.62 17.46 12.92 13.13 11.45 12.57
Annual Return in %, Figure 3b.

Low Volume Losers (R1V1) and High Volume Losers

(R1V5)
35.00

15.00

-5.00

-25.00
-4 -3 -2 -1 0 1 2 3 4 5

R1V1 17.34 14.41 11.14 9.64 -17.30 14.49 17.84 17.86 17.35 17.56
R1V5 25.77 28.33 33.52 37.34 -16.20 5.37 12.92 11.65 12.69 14.83
Annual Return in %, Figure 3c.

Figure 3. Annual returns from Year -4 to Year +5 for Various Volume Portfolios. These charts report
the annual returns for various volume and volume-based momentum portfolios from Year -4 to Year +5
around the portfolio formation date. Figure 3a reports the average annual return for high (V5) and low
(V1) volume quintile firms. Figures 3b and 3c report similar statistics for winner portfolios (R5) and loser
portfolios (R1), respectively. Price momentum and volume are based on past six-month data in Year 0.
 High-Volume Stocks

High-volume
High-Volume losers
winners

Winners Losers

Low-volume Low-volume
winners losers

Low-Volume Stocks

Figure 4. Momentum Investing Based on Past Price and

Volume Information. This figure illustrates some of the more
salient features of our empirical findings. We find that low-volume
stocks generally outperform high-volume stocks. Among winners, low-
volume stocks show greater persistence in price momentum. Among
losers, high-volume stocks show greater persistence in price momentum.
In addition, low-volume (high-volume) firms exhibit many
characteristics most commonly associated with value (glamour) stocks.
1
We use average daily turnover as a measure of trading volume. Turnover is defined as the ratio of the number of

shares traded to the number of shares outstanding. Any unqualified reference to trading volume henceforth refers

to this definition.
2
Studies that characterize price momentum as an underreaction include Jegadeesh and Titman (1993), Chan,

Jegadeesh, and Lakonishok (1996), Barberis, Shleifer, and Vishny (1998), and Hong and Stein (1999).

Conversely, studies that characterize price momentum to be the result of overreaction include DeLong, Shleifer,

Summers, and Waldmann (1990), and Daniel, Hirshleifer, and Subrahmanyam (1998).
3
For example DeBondt and Thaler (1985, 1987) and Chopra, Lakonishok, and Ritter (1992) attribute long-term

price reversals to investor overreaction. In contrast, Ball, Kothari, and Shanken (1995), Conrad and Kaul (1993)

and Ball and Kothari (1989) point to market microstructure biases or time-varying returns as the most likely

causes. Similarly, short-horizon price reversals have been attributed to return cross-autocorrelations (Lo and

MacKinlay (1990)), and transaction costs (Lehmann (1990), Conrad, Gultekin, and Kaul (1991)).
4
As we show later, industry adjustments accounts for approximately 20 percent of the price momentum effect in

our sample, but has no effect on the predictive power of volume.

5
Along the same lines, Chordia and Swaminathan (1999) find that at short-horizons low volume stocks exhibit

more underreaction than high volume stocks.

6
In a similar study, Brennan, Chordia, and Subrahmanyam (1998) use dollar-trading volume as a proxy of

liquidity and find similar results.

7
We have also replicated our results using a holdout sample consisting of just Nasdaq -NMS firms from 1983 to

1996. The predictive power of trading volume is even stronger among Nasdaq -NMS firms. However, we suspect

illiquidity problems are more pervasive with the Nasdaq -NMS sample.
8
Most previous studies have used turnover as a measure of the trading volume in a stock (see Campbell,

Grossman, and Wang (1993)). Note also that raw trading volume is unscaled and, therefore, is likely to be highly

correlated with firm size.

9
We have also formed intersections with five momentum and five volume portfolios and three momentum and ten

volume portfolios. These results are presented in Table III.

10
We have also replicated our tests with a one-month lag and found similar results.
11
As we show later, this result is specific to the first year after portfolio formation. Table VI reports that low

volume winners outperform high volume winners by two percent to six percent per year beyond Year 1.
12
We have also replicated the results in Table II using value-weighted portfolio returns. We obtain similar but

(not surprisingly) slightly weaker results. For instance, for the (J=6, K=6) strategy, the R10-R1 returns for low,

medium, and high volume portfolios are respectively 0.35 percent, 1.04 percent, and 1.15 percent per month

respectively. The difference in R10-R1 between high and low volume is 0.80 percent which is statistically

significant at the 1 percent level.

13
We have also performed characteristics based risk-adjustment and computed size-adjusted, industry-adjusted,

and size and book-to-market adjusted returns to check the robustness of results in Table II. These results confirm

the basic findings. We do not report these to conserve space.

14
SMB is small firm return minus large firm return and HML is high book-to-market portfolio return minus low

book-to-market portfolio return. For details on portfolio construction, see Fama and French (1993).
15
See Fama and French (1993), Table 6. Combining the estimated hi coefficient for the top two and bottom two

book-to-market quintiles, the Fama-French HML factor differential between low book-to-market and high book-to-

market firms is around -0.7.

16
Our industry partitions are slightly finer than the 20 industries used by Moskowitz and Grinblatt (1999), but

not as fine as the 48 industries used by Fama and French (1997).

17
The buy-and-hold abnormal returns (BHAR) are computed as follows:

T T

BHARi = ∏ (1 + rit ) −∏ (1 + rmt )

t =1 t =1

where rit is the annual return in stock i and rmt is the annul return on the benchmark. The benchmark annual

returns are computed by equally-weighting the annual returns of constituent securities. The time-series average of

the cross-sectional mean buy-and-hold abnormal return is depicted in Figures 2 and 3.

18
These results are based on annual pooled cross-sectional data measured as of June 30th each year; computing

year-by-year correlations and averaging the annual estimates yields similar results.
19
Table IX provides formal tests that these differences in past returns are statistically significant.
20
We chose the four-year horizon because it measures changes in trading volume over a fairly long period but not

so long that we have data availability problems. The use of a longer horizon is also driven by the empirical fact
that the level of trading volume (turnover) is a very slowly mean reverting process. Changes measured over a three-

year horizon also provide similar results.

21
This diagram closely parallels the intuition presented in Bernstein (1993, 1995). However, Bernstein does not

discuss the role of trading volume.

22
We stress that this framework applies only at the portfolio level. We show that the firms in each quadrant of the

cycle behave, on average, as predicted by the MLC hypothesis. However, these results reflect mean behavior at the

portfolio level. At the individual firm level, things are far less deterministic than the figure implies, and turning

points are far less predictable.

23
See Lee, Myers, and Swaminathan (1999) for a more formal econometric specification of this concept.