Earnings Surprises, Growth Expectations, and Stock Returns or Don't Let An Earnings Torpedo Sink Your Portfolio

Review of Accounting Studies, 7, 289–312, 2002

C 2002 Kluwer Academic Publishers. Manufactured in The Netherlands.
Earnings Surprises, Growth Expectations, and Stock

Returns or Don’t Let an Earnings Torpedo Sink
Your Portfolio
DOUGLAS J. SKINNER∗ [email protected]
University of Michigan Business School
RICHARD G. SLOAN
University of Michigan Business School
Abstract. We provide new evidence that the inferior returns to growth stocks relative to value stocks are the
result of expectational errors about future earnings performance. Our evidence demonstrates that growth stocks
exhibit an asymmetric response to earnings surprises. We show that while growth stocks are at least as likely to
announce negative earnings surprises as positive earnings surprises, they exhibit an asymmetrically large negative
price response to negative earnings surprises. After controlling for this asymmetric price response, we find no
remaining evidence of a return differential between growth and value stocks. We conclude that the inferior return to
growth stocks is attributable to overoptimistic expectational errors that are corrected through subsequent negative
earnings surprises.
Keywords: earnings surprises, earnings torpedoes, value, growth, glamour
JEL Classification: G14, M41
1. Introduction
It is well-established that ‘growth’ or ‘glamour’ stocks have historically underperformed

other stocks in terms of realized stock returns over the five years after portfolio formation.
We demonstrate that this phenomenon can be explained by the fact that growth stocks
exhibit an asymmetric response to negative earnings surprises. We first show that growth
stocks issue at least as many negative earnings surprises as positive earnings surprises.
Rational expectations therefore implies that the negative expected return to negative earnings
surprises be no larger in magnitude than the positive expected return to positive earnings
surprises. In contrast, we find that the average realized negative return to negative earn-
ings surprises is significantly larger in magnitude than the average realized positive return
to positive earnings surprises. After controlling for the asymmetric response of growth
stocks to negative earnings surprises, we show that there is no remaining evidence of a
stock return differential between growth stocks and other stocks.
∗ Address correspondence to: University of Michigan Business School, 701 Tappan Street, Ann Arbor, MI 48109-
1234.
290 SKINNER AND SLOAN
Our results provide compelling evidence that the inferior returns to growth stocks are the
result of expectational errors about future earnings performance. Existing research focuses
on distinguishing among three explanations for the inferior returns to growth stocks. First,
some argue that investors have overly optimistic expectations about the prospects of growth
stocks, resulting in lower subsequent stock returns when these expectations are not met
(Lakonishok, Shleifer and Vishny (LSV), 1994). Second, some argue that growth stocks are
less risky (Fama and French (FF), 1992). Third, some argue that methodological problems
with the measurement of long-term abnormal returns create the illusion of inferior returns
to growth stocks (Fama, 1998; Kothari, Sabino and Zach, 1999). Our evidence is consistent
with the first explanation and inconsistent with the second and third explanations.
Our paper also resolves the inconclusive evidence reported by La Porta, Lakonishok,
Shleifer and Vishny (1997) and Bernard, Thomas and Wahlen (1997). These papers ex-
amine whether the differential stock returns between growth stocks and other stocks are
clustered around earnings announcements, but report weak and inconclusive results. We
provide more powerful tests by conditioning on the sign of the earnings surprise and by
incorporating the price response to preannouncements of earnings news. These features
of our research design are important, because negative earnings news is frequently prean-
nounced for growth stocks (Skinner, 1994, 1997; Soffer, Thiagarajan and Walther, 2000).
Consistent with the importance of controlling for preannouncements, we show that evi-
dence of an asymmetric reaction to negative earnings surprises in growth stocks weakens
considerably when focusing exclusively on earnings announcement date returns.
Finally, we show that the intertemporal performance of growth stocks relative to other
stocks is directly related to intertemporal patterns in the relative proportion of growth stocks
reporting negative earnings surprises. Thus, while growth stocks underperform on average,
they systematically outperform other stocks in ‘boom’ periods during which a relatively low
frequency of negative earnings surprises are reported. In short, our paper provides the most
compelling evidence to date that the inferior returns to growth stocks are directly linked to
expectational errors about future earnings performance.
The next section of the paper formulates our research hypothesis and empirical predictions.
Section 3 describes our sample and research design, Section 4 presents the empirical results
and Section 5 concludes.
2. Hypothesis and Empirical Predictions
Our primary objective is to test the hypothesis that the inferior stock returns experienced by
growth stocks are attributable to expectational errors made by investors. Following previous
research, we use the market-to-book ratio as our primary measure of growth (LSV, 1994). A
specific behavioral explanation for the expectational error hypothesis is developed in more
depth by LSV. The basic idea is that investors tend to set their expectations for long-run future
sales and earnings growth too high for stocks with high past growth and/or high expected
short-run future growth. These overoptimistic expectations lead to high pricing multiples,
such as high market-to-book ratios. Over time, subsequent earnings announcements reveal
that the optimistic expectations reflected in these pricing multiples are not sustainable,
resulting in the inferior stock price performance.
EARNINGS SURPRISES, GROWTH EXPECTATIONS, AND STOCK RETURNS 291
The key to testing the expectational errors hypothesis is to first identify the subsequent
events informing investors that their expectations are overoptimistic, and to then show that
the inferior stock price performance is concentrated around these events. Existing papers
employ two research designs in this respect. First, La Porta et al. (1997) and Bernard
et al. (1997) examine the stock returns around future earnings announcement dates. If
investors in growth stocks become aware of their expectational errors through subsequent
earnings announcements, then the lower stock returns associated with growth stocks should
be concentrated around these subsequent earnings announcements. These studies report
mixed results. For example, La Porta et al. find that only about 20% of the inferior returns
of growth stocks are concentrated around subsequent earnings announcements, and that
this figure drops to about 15% for large firms. Further, the studies provide no statistical
evidence that these percentages are significantly different from what would be expected by
chance if the inferior returns were distributed evenly over time. An obvious shortcoming
of this research design is that many firms preannounce earnings, and preannouncements
are particularly prevalent in the case of negative earnings surprises in large growth firms
(Skinner, 1994, 1997; Kasznik and Lev, 1995; Soffer, Thiagarajan and Walther, 2000).
The second type of research design focuses on the association between analysts’ long-
run forecasts of EPS growth and subsequent realizations of EPS growth and stock returns.
Research by La Porta (1996) and Dechow and Sloan (1997) shows that analysts’ long-run
EPS forecasts are systematically overoptimistic for growth firms, and that the magnitude
of the overoptimism in these forecasts is systematically related to the inferior stock price
performance of growth firms. While the results of these studies are consistent with the
expectational error hypothesis, they have been criticized on the grounds that analysts’
expectations may not be representative of investors’ expectations, and that the long-run
abnormal stock returns used in these studies may be misspecified (Fama, 1998; Kothari,
Sabino and Zach, 1999).
In this paper, we introduce a new research design that avoids the problems described
above. Our research design is based on the premise that investors will revise downward
their overoptimistic expectations for growth stocks in response to subsequent negative
earnings surprises. This premise leads to two predictions. First, we predict an asymmetrically
large negative stock return for growth stocks reporting negative earnings surprises. This
asymmetrically large stock return arises as the negative earnings surprises causes investors
to revise downwards their previously overoptimistic expectations. Second, we predict that
the entire return differential between growth stocks and other stocks will be realized during
periods in which negative earnings surprises are reported.
Our basic premise, that investors realize their expectations are overoptimistic upon ob-
serving negative earnings surprises, seems natural. There is little reason to believe that
investors will revise their overoptimistic expectations downward for firms that are meeting
or beating earnings expectations. Moreover, the financial press is replete with examples of
growth firms experiencing large stock price declines in response to quite pedestrian neg-
ative earnings surprises.1 Finally, we see no reason to believe that the competing risk and
misspecification explanations for the inferior returns to growth stocks would predict that
the inferior returns are concentrated around negative earnings surprises.
Our primary prediction is that growth stocks exhibit an asymmetrically large negative
price response to negative earnings surprises. The relation between growth stocks and
(a) Unrelated:
The return differential between value and growth stocks is the same regardless of the subsequent earnings surprise
Earnings Surprise
Stock Type
Negative Positive All
Value −4% 6% 1%
(25%) (25%) (50%)
Growth −6% 4% −1%
(25%) (25%) (50%)
All −5% 5% 0%
(50%) (50%) (100%)
(b) Asymmetric response to negative surprises:

The return differential between value and growth stocks is all concentrated in subsequent negative earnings
surprise quarters
Earnings Surprise
Stock Type
Negative Positive All
Value −3% 5% 1%
(25%) (25%) (50%)
Growth −7% 5% −1%
(25%) (25%) (50%)
All −5% 5% 0%
(50%) (50%) (100%)
Figure 1. Illustration of alternative hypothetical abnormal return combinations for portfolios of value and growth
stocks over subsequent quarters, stratified by the nature of the subsequent quarterly earnings surprises. The numbers
in parentheses represent the hypothetical relative frequencies with which stocks enter a cell.
earnings surprises has been previously studied by Basu (1977) and Dreman and Berry
(1995). However, both of these studies predict that stock returns will be more pronounced
for high (low) growth stocks reporting negative (positive) earnings surprises. In contrast,
our predictions pertain only to high growth stocks reporting negative earnings surprises.
This difference is crucial, because the stock return behavior predicted in these two prior
studies would be expected even if the reaction to an earnings surprise was unrelated to the
growth characteristics of the stock. We illustrate this point in Figure 1(a).
Figure 1(a) illustrates hypothetical average abnormal returns to growth and value stocks
under the assumption that the return differential to growth and value stocks is realized
regardless of the sign of the subsequent earnings surprise. The rows of the table report the
average abnormal returns for value and growth stocks over a one-quarter holding period.
For simplicity, we assume that value stocks have a 1% average abnormal return, while
growth stocks have a −1% average abnormal return, and that stocks are distributed in equal
numbers between the two categories. The columns report the abnormal returns stratified by
the nature of the earnings surprise reported during the quarter. For simplicity, we assume
that stocks reporting a positive earnings surprise have an average abnormal return of 5%
and stocks reporting a negative earnings surprise have an average abnormal return of −5%.
We also assume that stocks are distributed 50% in each of the surprise categories (i.e., firms
are equally likely to miss or beat expectations, but no firms exactly meet expectations).
The above assumptions provide the row and column totals of Figure 1(a). The distribution
of the returns among the other cells depends on the relation between the growth character-
istics and the stock price response to earnings surprises. Figure 1(a) is prepared under the
assumption that the 2% return differential between growth and value stocks occurs regard-
less of the earnings surprise. For example, the average abnormal return for firms reporting a
positive earnings surprise is 5%. Hence, growth firms reporting positive earnings surprises
have an average abnormal return of 5% plus −1% to give 4%, while value firms reporting
positive earnings surprises have an average abnormal return of 5% plus 1% to give 6%.
The key feature of the returns in Figure 1(a) is that the average return differential between
growth and value stocks is the same regardless of the sign of the earnings surprise. Thus, this
table presents exactly the relation that would be expected if the two effects were completely
unrelated, as predicted by Basu (1977) and Dreman and Berry (1995). Average abnormal
returns are more pronounced for growth (value) stocks reporting negative (positive) earnings
surprises.
Figure 1(b) illustrates the average abnormal returns to growth and value stocks under
the assumption that the return differential to growth and value stocks is concentrated in
subsequent negative earnings surprise quarters. The row and column totals are prepared
using the same assumptions used in Figure 1(a). However, the distribution of the returns
between the other cells is different from Figure 1(a). The average abnormal returns for
firms reporting positive earnings surprises are the same regardless of the value/growth
classification. The entire differential between value and growth stocks is concentrated in
firms reporting negative earnings surprises. Because only 50% of securities are assumed to
report negative surprises, the average return differential between value and growth stocks
is magnified to 4% for these securities, thus maintaining the average differential across all
stocks of 2%. The key feature of Figure 1(b) is that the differential returns for growth and
value stocks are only realized during quarters when negative earnings surprises are reported.
Figure 1(b) illustrates the two key predictions that we test in this study. First, we see a large
asymmetric negative reaction to negative earnings surprises in growth stocks. Second, there
is no evidence of a value/growth return differential in stocks reporting positive earnings
surprises, indicating that the value/growth return differential is entirely concentrated in
firms reporting negative earnings surprises.
Our third and final prediction concerns the timing of the return differential. Our version of
the expectational hypothesis predicts that the differential returns to growth and value stocks
will be concentrated around the release date of negative earnings news. Evidence in support
of this prediction would corroborate the link between the differential return behavior and
earnings surprises. Such evidence is not presented in either Basu (1977) or Dreman and
Berry (1995). Past research by La Porta et al. (1997) and Bernard et al. (1997) focuses on
the returns to growth and value stocks during short (2–3 day) windows centered on quarterly
earnings announcement dates. However, as described earlier, there has been a growing trend
for management to preannounce earnings (Skinner, 1994, 1997; Kasznik and Lev, 1995;
Soffer, Thiagarajan and Walther, 2000). The evidence indicates that preannouncements
predominantly convey adverse earnings news,2 and are more likely in litigation intensive
industries, which tend to be industries with high growth firms (e.g., computer hardware and
software, drugs, electrical equipment, and retail). By announcing adverse earnings news
early, these firms accelerate the associated stock price decline, thus avoiding large stock
price declines on the earnings announcement date and reducing the expected costs of any
potential stockholder litigation.3 This evidence suggests that negative earnings surprises
in growth firms are particularly prone to preannouncement. Since these observations are
also those that we hypothesize will exhibit an asymmetrically large stock price response to
earnings news, our research design uses a return measurement interval designed to capture
preannouncements.
3. Sample and Research Design
We obtain quarterly earnings forecasts from the I/B/E/S historical database, which contains
139,027 observations with non-missing data on the consensus forecast of quarterly earnings,
realized quarterly earnings, and stock prices between 1984 and 1996. We use the consensus
forecast provided by I/B/E/S in the final month of the fiscal quarter for which earnings is
being forecast. I/B/E/S collects the forecast data through the first half of the month and
releases the forecast data around the middle of the month. Thus, we can be sure that the
forecasts do not contain any information from earnings preannouncements made after the
middle of the final month of the quarter.4 We also require that sample firms have the required
data to compute the growth/value measures (described below) on COMPUSTAT and daily
stock return data for at least one quarter on CRSP. These requirements reduce the final
sample size to 103,274 firm-quarter observations.
Our research design consists of classifying firm-quarters on the basis of growth/value
characteristics and tracking their subsequent stock return and earnings surprise characteris-
tics. Prior research shows that the differential returns to growth and value stocks persist for
five years after the date the growth/value characteristics are measured (LSV). We therefore
track stock return and earnings surprise characteristics for the 20 quarters that follow the
measurement of the growth/value characteristics. For example, growth/value characteristics
measured using data from the fourth quarter of 1989 are related to stock returns and earnings
surprises for each of the subsequent 20 quarters (i.e., the first quarter of 1990 through the
fourth quarter of 1994).
We measure growth/value characteristics in a similar manner to previous research. We
focus on the market-to-book ratio, since this variable has received the most attention in
previous research. We measure market-to-book (MB) as the market value of outstanding
shares at the end of the quarter divided by book value of common equity at the end of the
quarter. We also report results using the price-to-trailing earnings ratio (PE) and the I/B/E/S
median analyst forecast of long-term earnings growth.
We measure the earnings surprise for a quarter by subtracting the median forecast of
quarterly EPS from realized quarterly EPS. We then create three indicator variables, which
we label SURPRISE, GOOD, and BAD. SURPRISE takes on the value of −1 if the earnings
surprise is negative, 0 if the earnings surprise is 0, and 1 if the earnings surprise is positive.
GOOD takes on the value of 1 if the earnings surprise is positive and zero otherwise. BAD
takes on the value of 1 if the earnings surprise is negative and zero otherwise. Finally, we
create a continuous variable that captures both the sign and magnitude of the forecast error,
which we label FE. FE is the earnings surprise divided by the stock price at the end of the
final month of the fiscal quarter for which earnings is being forecast. We winsorize the 1%
tails of this variable to mitigate outlier problems.
Throughout the paper we compute stock returns as buy-hold with-dividend stock returns
and compute abnormal returns by subtracting the return over the corresponding period on
a size-matched portfolio.5 The size-matched portfolio is constructed by allocating all firm-
quarter observations in our sample to decile portfolios on the basis of market capitalization
at the beginning of the quarter. An equal-weighted portfolio return is computed for each size
portfolio in each quarter. Raw buy-hold returns for individual securities are then adjusted
by subtracting the return on the portfolio to which the security belongs based on its market
capitalization at the beginning of the quarter. Our objective is to examine stock return
behavior over the 20 quarters following the measurement of the growth/value characteristics
and to relate the returns to the earnings surprises reported in each of these 20 quarters. To
this end, we cumulate abnormal returns over four different intervals for each quarter. These
intervals are illustrated in Figure 2.
The first abnormal return measurement interval begins two days after the announcement
of earnings for the previous quarter and ends the day after the announcement of earnings for
the current quarter. We obtain quarterly earnings announcement dates from COMPUSTAT.
We refer to the quarterly return measured over this interval as ‘fullret.’ This interval averages
63 trading days in length. We next divide this interval into two sub-intervals, the later of
which is designed to capture earnings-related announcements. The first interval begins two
days after the announcement of earnings for the prior quarter and ends thirteen trading days
before the end of the current fiscal quarter. The second interval begins twelve trading
days before the end of the current fiscal quarter and ends the day after the announcement
Figure 2. Illustration of the alternative intervals over which the abnormal stock return relating to the announcement
of earnings for quarter t is measured.
of earnings for the current quarter. Evidence in Skinner (1997) and Soffer et al. (2000)
indicates that over 75% of all earnings preannouncements occur within two weeks on either
side of the fiscal quarter end, so we expect the large majority of earning surprises to be
announced during this second interval. The two intervals each average 31 trading days in
length, so return comparisons across the two intervals are simplified. We refer to the stock
returns cumulated over the first interval as ‘preret’ and over the second interval as ‘postret.’
Finally, we measure stock returns around the quarterly earnings announcement date, which
we define as the three-day period beginning one day prior to the earnings announcement
date and ending on the day after the announcement date. We use this return measurement
interval for comparisons with prior research that also uses this interval (La Porta et al.,
1997; Bernard et al., 1997). We expect this interval to miss much of the response to negative
earnings surprises since most adverse earnings news tends to be preannounced. We refer to
the return measured over this interval as ‘aret.’
4. Empirical Results
We begin by reporting descriptive evidence on each of our predictions after which we

provide formal statistical tests of our predictions using regression analysis. We then conduct
robustness tests using alternative measures to classify firms as ‘growth’ or ‘glamour’ stocks.
Finally, we report on the intertemporal relation between earnings surprises and the return
differential between growth and value stocks.
4.1. Descriptive Evidence
Table 1 provides descriptive evidence on the relation between the MB effect and earnings
surprises. This table stratifies our sample of firm-quarter observations into quintiles based
on the MB ratio and then divides each quintile into three categories based on the sign of
the earnings surprise. Each of the resulting 15 cells in Table 1 reports the mean quarterly
abnormal stock returns (fullret). Each cell also reports the number of observations falling
into that cell and the proportion of each row’s observations falling into that cell. The column
at the far right and the row at the bottom of the table report the grand averages across the
earnings surprise portfolios and the growth portfolios respectively.
Focusing first on the rightmost column, we see clear evidence of the previously docu-
mented MB effect in returns. The average abnormal return declines monotonically from
0.66% for the low growth quintile to −0.58% for the high growth quintile. This represents a
1.24% quarterly differential, which translates into a 5.05% compound annual return differ-
ential. This return differential is somewhat smaller than the 8–10% differential reported in
previous research, such as Lakonishok et al. (1994). However, their research design is based
on decile portfolios, and is not restricted to firms for which analysts’ forecasts are available.
Moving to the bottom row, we see the well-documented return differential between firms
reporting negative versus positive earnings surprises. The average quarterly abnormal return
for firms reporting negative surprises is −5.04% while the corresponding return for positive
surprises is 5.50%. Firms reporting a zero surprise report a positive return of 1.63%. This lat-
ter result reflects the fact that firms are more likely to report a negative surprise (47.8%) than
Table 1. Mean quarterly abnormal stock returns over the subsequent twenty quarters for portfolios of stocks
formed on growth and the sign (positive, negative, or zero) of the subsequent quarterly earnings surprise.
Earnings Surprise Portfolio
Negative Zero Positive All
Growth portfolio
1 (Low growth) −3.57% 1.13% 5.44% 0.66%
138,752 17,143 121,439 277,334
(50.0%) (6.2%) (43.8%) (100%)
2 −3.91% 2.01% 4.93% 0.35%
136,405 23,803 117,842 278,050
(49.0%) (8.6%) (42.4%) (100%)
3 −4.89% 1.71% 5.29% −0.03%
134,089 31,214 112,127 277,430
(48.3%) (11.3%) (40.4%) (100%)
4 −5.82% 1.54% 5.65% −0.40%
130,977 42,049 104,034 277,060
(47.3%) (15.2%) (37.5%) (100%)
5 (High growth) −7.32% 1.65% 6.32% −0.58%
122,099 52,789 102,051 276,939
(44.1%) (19.1%) (36.8%) (100%)
All growth portfolios −5.04% 1.63% 5.50% 0.00%
662,322 166,998 557,493 1,386,813
(47.8%) (12.0%) (40.2%) (100%)
Growth is measured using the MB ratio (low MB = low growth, high MB = high growth). Stock returns are
cumulated over the period beginning two days following the announcement of earnings for the previous quarter
and ending on the day following the announcement of earnings for the current quarter (Fullret). Each cell reports
the mean abnormal portfolio stock return, the number of observations in the portfolio, and the percentage of that
row’s observations falling into that cell. Abnormal returns are computed using a decile-based size adjustment.
a positive surprise (40.2%) so that a zero surprise for the remaining firms (12.0%) is actually
a better than expected outcome. The fact that there are more negative surprises than positive
surprises overall reflects the previously documented average over-optimism in sell-side ana-
lysts’ earnings forecasts for our sample period (Abarbanell and Lehavy, 2000; Brown, 2001).
Table 1 provides descriptive evidence on our first two predictions. Recall from Figure 1(b)
that these predictions require that all of the MB return differential is concentrated in the
negative earnings surprise portfolios. The evidence in Table 1 shows this to be the case.
The mean abnormal returns for the zero and positive surprise portfolios show no systematic
trend as a function of growth. If anything, the high growth portfolio returns actually seem
to be slightly higher than the low growth portfolio returns, opposite to what is necessary to
explain the overall value vs. growth effect. However, the negative surprise portfolios tell a
different story. The mean abnormal returns decline monotonically across growth portfolios
from a high of −3.57% for portfolio 1 to a low of −7.32% for portfolio 5. The pattern
of returns clearly coincides with the asymmetric response to negative surprises depicted
in Figure 1(b), rather than with the unrelated effects depicted in Figure 1(a). This pattern
indicates that the predictable lower returns for high MB firms are realized when these firms
subsequently report negative earnings surprises.
20
15
Cumulative Return
10
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
–5
Years Since Portfolio Formation
Figure 3. Cumulative average abnormal return for a MB (market-to-book) hedge portfolio over the five years
following portfolio formation. The hedge portfolio consists of a long position in the lowest quintile of MB stocks
and a short position in the highest quintile of MB stocks for each of the firm-quarters in our sample. Returns for
the first half of each year are cumulated over the four quarterly ‘Preret’ periods during which very little earnings
information is typically released. Returns for the second half of each year are cumulated over the four quarterly
‘Postret’ periods, during which most earnings information is typically released.
Our third prediction is that the asymmetric returns to growth and value are concentrated
around the release of earnings news. Figure 3 provides descriptive evidence on this predic-
tion. Figure 3 plots the returns to a hedge portfolio that takes a long position in a portfolio of
low MB stocks and an offsetting short position in a portfolio of high MB stocks. We assign
stocks to quintiles based on MB at the end of each quarter and then track the mean stock
returns for each quintile over the subsequent five years. The hedge portfolio returns are
computed by subtracting the highest quintile mean returns from the lowest quintile mean
returns. To distinguish between the stock price movements attributable to earnings news
versus other factors, such as risk, we divide each of the annual returns into two compo-
nents. The first component represents the cumulative abnormal return over the four quarterly
‘preret’ periods. Recall that the ‘preret’ return period begins two days after the announce-
ment of last quarter’s earnings and ends 13 days before the end of the fiscal quarter and so
excludes the release of earnings news, including earnings preannouncements. The second
component represents the cumulative abnormal return over the four quarterly ‘postret’ re-
turn periods which begins 12 days before the end of the quarter and ends on the day after
the earnings announcement, and so captures the release of earnings news, including any
preannouncements. Our third prediction is that the returns to the MB hedge portfolio will
be concentrated in the ‘postret’ period.
Consistent with previous research (Fama and French, 1992; Lakonishok et al., 1994),
Figure 3 demonstrates that our MB hedge portfolio yields systematic positive returns.
The cumulative five-year return is just below 20%. This return differential is somewhat
smaller than that documented in previous research for three reasons. First, we use quintiles
rather than deciles, so differences between the extreme portfolios are smaller. Second, by
restricting the sample to the larger, more closely followed stocks in I/B/E/S, we restrict
attention to stocks for which these types of strategies are typically less profitable.6 Third,
our sample period is concentrated in the 1980s and 1990s, where the MB effect is somewhat
weaker than in the 1960s and 1970s. The important feature of Figure 3 is that the return
differential is clearly concentrated in the second half of each ‘year’ (postret) as we see
from the steeper slope in that interval, which provides clear evidence that the MB return
differential is concentrated around the release of earnings news. Returns during the first
half of the ‘year’ (preret) account for less than 20% of the total predictable returns to the
MB hedge portfolio.
To illustrate the asymmetric response of returns to earnings news for growth stocks,
Figure 4 plots quarterly abnormal returns (fullret) against earnings surprises (FE) separately
for growth and value stocks.7 Figure 4 clearly shows how the relation between stock returns
and earnings surprises differs between growth and value stocks. For value stocks the relation
is fairly symmetric—for both positive and negative surprises the return/earnings relation
looks similar, with returns increasing in the magnitude of the earnings surprise to a maximum
of a little over 5% in absolute value for both good and bad news.8 For growth stocks we see
a very different response to positive and negative surprises. When earnings news is positive,
returns climb steeply over a small range of forecast error, to a maximum a little over 10%
(thus even when the news is good, the reaction is stronger for growth stocks). However,
when firms miss their forecasts the effect is dramatic. For even small forecast errors (of less
than 0.5% of stock price) the stock price reaction declines rapidly into the −10% to −15%
range, and continues to decline beyond this, into the −15% to −20% range. The sharp
Figure 4. Earnings surprise response functions for value and growth stocks. This graph plots the quarterly abnormal
returns for value and growth stocks respectively as a function of the magnitude of the quarterly earnings forecast
error. Each plot is formed by dividing the stocks into 20 portfolios based on the magnitude of the forecast error,
and then plotting the mean portfolio abnormal returns and forecast errors. The resulting points are joined using
smoothed lines.
drop is the earnings “torpedo”—the fact that missing analysts’ forecasts, even by small
amounts, causes disproportionately large stock price declines. It is also clear from Figure 4
that when earnings news is positive, growth stocks outperform value stocks, but that when
growth stocks disappoint, they underperform value stocks by substantially more than they
outperform when the news is good (i.e., the area between the two plots is much greater in
the negative forecast region than in the positive forecast region). As our regressions show, it
is this large differential reaction to bad news that accounts for the overall underperformance
of growth stocks. We turn to these regressions next.
4.2. Regression Analysis
In this section, we provide statistical tests of our predictions using regression analysis. We
begin in Table 2 by regressing stock returns on growth portfolio membership and both the
sign of the earnings surprise and the magnitude of the earnings surprise. We further allow
for a growth variable interaction with each of these explanatory variables. The purpose of
these regressions is to demonstrate that the negative relation between growth and future
stock returns (the MB effect) is robust to the inclusion of various earnings surprise metrics
including those that control for the magnitude of the earnings surprise. These regressions do
not allow for an asymmetric response to negative earnings surprises. In Table 3, we allow for
an asymmetric response to negative earnings surprises using the following regression model:
Ritτ = α + β1 · Growthit + β2 · Gooditτ + β3 · Baditτ + β4 · (Gooditτ ∗ Growthit )

+ β5 · (Baditτ ∗ Growthit ) + εitτ (1)
where
i indexes firms, t indexes calendar quarters in which growth portfolio assignments are made,
and τ indexes the 20 subsequent quarters over which we track returns and earnings
surprises for each growth (firm-quarter) observation;
Growthit = growth quintile to which firm i is assigned in quarter t (0 = low growth quintile,
. . . 4 = high growth quintile);
Ritτ = the announcement-to-announcement (fullret) abnormal stock return for
firm i in quarter t + τ ;
Gooditτ = indicator variable taking the value of 1 if the firm-quarter observation reports a
positive earnings surprise in quarter t + τ and 0 otherwise;
Baditτ = indicator variable taking the value of 1 if the firm-quarter observation reports a
negative earnings surprise in quarter t + τ and 0 otherwise.
As an alternative specification, we also estimate regressions of the following form, and

report these results in Table 4:
Ritτ = α + β1 · Growthit + β2 · Gooditτ + β3 · Baditτ + β4 · FEitτ

+ β5 · (Gooditτ ∗ Growthit ∗ FEitτ ) + β5 · (Baditτ ∗ Growthit ∗ FEitτ ) + εitτ (2)
Table 2. Estimated coefficients (adjusted t-statistics) from regressions of quarterly stock returns (‘fullret’) on
‘growth’ portfolio membership, the sign of the earnings surprise for that quarter (defined as −1, 0, or +1), and
the analysts’ forecast error for the quarter.
Quarters from: Forecast Forecast Adjusted

(Number of Obs.) Intercept Growth Surprise Surprise ∗ Growth Error Error ∗ Growth R-Squared
Year 1 0.0014 −0.0014 0.02%

(n = 349,678) (1.25) (−3.06)
0.0078 −0.0034 0.0476 0.0058 8.02%
(7.07) (−7.41) (41.38) (12.00)
0.0144 −0.0022 0.0110 0.0094 4.0602 0.5498 10.59%
(13.07) (−4.98) (6.78) (14.71) (29.76) (7.99)
Year 2 0.0059 −0.0029 0.05%

(n = 305,416) (5.01) (−6.13)
0.0085 −0.0020 0.0437 0.0054 7.50%
(7.54) (−4.42) (37.11) (10.96)
0.0135 −0.0020 0.0113 0.0077 3.5877 0.5867 9.96%
(11.99) (−4.45) (6.72) (11.69) (25.81) (8.87)
Year 3 0.0046 −0.0023 0.03%
(n = 269,864) (3.77) (−4.64)
0.0066 −0.0012 0.0405 0.0056 7.35%
(5.66) (−2.46) (33.47) (11.10)
0.0110 −0.0011 0.0093 0.0075 3.2185 0.6377 9.72%
(9.46) (−2.21) (5.31) (10.82) (22.14) (9.06)
Year 4 0.0045 −0.00239 0.03%

(n = 241,668) (3.67) (−4.50)
0.0056 −0.0009 0.0371 0.0062 7.12%
(4.65) (−1.74) (29.82) (11.97)
0.0094 −0.0006 0.0081 0.0069 2.8238 0.7371 9.62%
(7.91) (−1.22) (4.45) (9.62) (19.09) (10.47)
Year 5 0.0051 −0.0026 0.08%

(n = 220,185) (4.14) (−5.06)
0.0061 −0.0014 0.0358 0.0058 7.02%
(5.10) (−2.88) (28.45) (11.97)
0.0097 −0.0012 0.0080 0.0062 2.5912 0.6880 9.45%
(8.06) (−2.36) (4.25) (8.38) (17.53) (9.92)
All 20 quarters 0.0065 −0.0032 0.06%
(n = 1,386,813) (5.36) (−6.56)
0.0092 −0.0026 0.0414 0.0059 7.49%
(7.89) (−5.36) (34.12) (11.67)
0.0137 −0.0025 0.0123 0.0070 3.1803 0.5842 9.80%
(11.74) (−5.28) (7.15) (10.19) (21.53) (8.39)
Growth portfolios are MB quintiles (low MB = quintile 0, low MB = quintile 4). Growth portfolios are formed at the beginning of
Year 1, and regressions employ returns and earnings data over the subsequent twenty quarters, reported in annual blocks of four
quarters. We estimate regressions of the following form:
Ritτ = α + β1 · Growthit + β2 · (Surpriseitτ ∗ Growthit ) + β3 · FEitτ + β4 · (FEitτ ∗ Growthit ) + εitτ ,
where
Ritτ = the size-adjusted stock return (where the size adjustment is the return on the corresponding CRSP size-decile
portfolio) for firm i in quarter tτ , where t indexes calendar quarters and τ indexes the 20 subsequent quarters
over which we estimate these regressions;
Growthit = the growth quintile into which firm i was assigned in quarter t (where 0 denotes the low growth quintile and 4
denotes the high growth quintile) and growth is measured as the firm’s market-to-book (MB) ratio at the end of
quarter t;
FEitτ = Realized EPS for firm i in quarter tτ minus the corresponding consensus analyst forecast of EPS, deflated by
the firm’s stock price at the end of fiscal quarter tτ ; and
Surpriseitτ = −1 if FEitτ is negative, +1 if FEitτ is positive, and 0 otherwise.
The t-statistics are adjusted for cross-correlation in the residuals resulting from multiple appearances of the Rit observations.
‘growth’ portfolio membership, good (bad) news indicator variables coded one if the earnings surprise is positive
(negative) and zero otherwise, and interaction terms.
Adjusted
Quarters from: Good ∗ Bad ∗ F-Statistic Adjusted
(No. of Obs.) Intercept Growth Good Bad Growth Growth ( p-Value) R-Squared
Year 1 0.0163 0.0012 0.0388 −0.0561 0.0016 −0.0099 8.80 8.14%
(n = 349,678) (4.27) (0.97) (9.28) (−13.66) (1.04) (−6.72) (0.0030)
Year 2 0.0112 0.0019 0.0411 −0.0462 0.0003 −0.0103 11.24 7.58%
(n = 305,416) (2.81) (1.37) (9.44) (−10.71) (0.19) (−6.62) (0.0008)
Year 3 0.0132 0.0011 0.0338 −0.0473 0.0025 −0.0086 3.92 9.72%
(n = 269,864) (3.15) (0.74) (7.46) (−10.53) (1.52) (−5.34) (0.0477)
Year 4 0.0092 0.0023 0.0336 −0.0407 0.0021 −0.0102 6.48 7.63%
(n = 241,669) (2.14) (1.56) (7.20) (−8.78) (1.23) (−6.14) (0.0109)
Year 5 0.0090 0.0018 0.0330 −0.0386 0.0017 −0.0097 6.12 7.08%
(n = 220,186) (2.09) (1.16) (7.01) (−8.27) (1.00) (−5.76) (0.0137)
All 20 quarters 0.0158 0.0020 0.0347 −0.0481 0.0022 −0.0095 5.80 7.57%
(n = 1,386,813) (3.81) (1.48) (7.70) (−10.78) (1.33) (−5.95) (0.0160)
Growth portfolios are MB quintiles (low MB = quintile 0, high MB = quintile 4). Growth portfolios are formed at
the beginning of Year 1, and regressions employ returns and earnings data measured over the subsequent twenty
quarters, reported in annual blocks of four quarters. We estimate regressions of the following form:
Ritτ = α + β1 · Growthit + β2 · Gooditτ + β3 · Baditτ + β4 · (Gooditτ ∗ Growthit ) + β5 · (Baditτ ∗ Growthit ) + εitτ ,
where
Ritτ = the market-adjusted stock return (where the market return is the CRSP value-weighted market
index) for firm i in quarter tτ , where t indexes calendar quarters and τ indexes the 20 subsequent
quarters over which we estimate these regressions;
Growthit = the growth quintile into which firm i falls in quarter t (where 0 denotes the low growth quintile
and 4 denotes the high growth quintile) and growth is measured as the firm’s market-to-book
(MB) ratio at the end of quarter t;
Gooditτ = 1 if FEitτ is positive, and 0 otherwise; and
Baditτ = 1 if FEitτ is negative, and 0 otherwise.
The F-statistic is from an F-test is of the restriction that β4 = −β5 .
The t-statistics and F-statistics are adjusted for cross-correlation in the residuals resulting from multiple appear-
ances of the Rit observations.
where all variables are as defined above and FEitτ is the forecast error defined as realized
EPS for firm i in quarter t + τ minus the corresponding consensus analyst forecast of EPS,
deflated by the firm’s stock price at the end of fiscal quarter t + τ . This specification allows
for a differential (good vs. bad) stock price response per unit of earnings surprise across
growth quintiles, and so allows for an asymmetric response that is a function of the magni-
tude of the forecast error. Specification (2) is thus the appropriate specification if investors
react asymmetrically to both the sign and the magnitude of negative earnings surprises in
growth stocks, while (1) is more appropriate if the asymmetric reaction is a function of
the sign, but not the magnitude, of the surprise. We find that specification (1) explains the
value vs. growth phenomenon better than (2), consistent with the idea that missing analysts’
forecasts by even small amounts results in large stock price declines for growth stocks.
(negative) and zero otherwise, forecast error, and growth interaction terms conditioned on the sign of the earnings
surprise.
Good ∗ Bad ∗
Quarters from: Growth ∗ Growth ∗ Adjusted
(No. of Obs.) Intercept Growth Good Bad FE FE FE R-Squared
Year 1 0.0220 −0.0010 0.0232 −0.0400 2.7873 0.6220 1.4270 10.47%
(n = 349,678) (10.36) (−2.03) (10.74) (−19.07) (25.08) (6.89) (22.78)
Year 2 0.0183 −0.0009 0.0231 −0.0335 2.5341 0.5677 1.3300 9.87%
(n = 305,416) (8.20) (−1.73) (10.15) (−14.96) (21.95) (6.16) (21.31)
Year 3 0.0164 −0.0002 0.0200 −0.0316 2.2227 0.7265 1.3001 9.63%
(n = 269,864) (7.06) (−0.32) (8.40) (−13.48) (19.89) (7.98) (20.84)
Year 4 0.0146 0.0002 0.0177 −0.0291 1.9094 0.8574 1.3297 9.54%
(n = 241,669) (6.09) (0.38) (7.19) (−12.95) (16.54) (9.74) (21.48)
Year 5 0.0137 −0.0001 0.0176 −0.0262 1.7797 0.7000 1.2608 9.40%
(n = 220,186) (5.60) (−0.11) (7.00) (−10.51) (15.61) (8.34) (20.84)
All 20 quarters 0.0193 −0.0012 0.0217 −0.0337 2.2370 0.5207 1.2889 9.78%
(n = 1,386,813) (8.35) (−2.29) (9.18) (−14.43) (18.97) (6.05) (20.88)
quarters, reported in annual blocks of four quarters. We estimate regressions of the following form:
Ritτ = α + β1 · Growthit + β2 · Gooditτ + β3 · Baditτ + β4 · FEitτ + β5 · (Gooditτ ∗ Growthit ∗ FEitτ )
+ β5 · (Baditτ ∗ Growthit ∗ FEitτ ) + εitτ ,
where
FEitτ = Realized EPS for firm i in quarter tτ minus the corresponding consensus analyst forecast of EPS,
deflated by the firm’s stock price at the end of fiscal quarter tτ ; and
The t-statistics are adjusted for cross-correlation in the residuals resulting from multiple appearances of the Rit
observations.
We also estimate specification (1) using alternative return measurement intervals in the
dependent variable and report these results in Table 5. These regressions illustrate that the
MB effect and its relation to earnings surprises are concentrated in the ‘postret’ return
measurement interval, when most earnings news is released.
Our basic sample consists of approximately 103,000 firm-quarters, giving us potentially
2.06 million regression observations as we track each firm-quarter over the subsequent
20 quarters. The actual ‘full’ sample is on the order of 1.4 million observations, primarily
because we lose firm-quarters outside the end of our sample period as we move forward
through the 20 quarters, and because of missing earnings announcement dates. We conduct
Table 5. Estimated coefficients (adjusted t-statistics) from regressions of stock returns measured over various
intervals on ‘growth’ portfolio membership, good (bad) news indicator variables coded one if the earnings surprise
is positive (negative) and zero otherwise, and interaction terms.
Return Adjusted
Measurement Good ∗ Bad ∗ F-Statistic Adjusted
Interval Intercept Growth Good Bad Growth Growth ( p-Value) R-Squared
Fullret 0.0065 −0.0032 0.06%
(5.36) (−6.56)
Fullret 0.0158 0.0002 0.0347 −0.0481 0.0022 −0.0095 5.80 7.57%
(3.81) (0.15) (7.70) (−10.78) (1.33) (−5.95) (0.0160)
Preret 0.0009 −0.0005 0.00%
(1.20) (−1.47)
Preret 0.0048 0.0010 0.0097 −0.0195 0.0001 −0.0034 2.48 1.68%
(1.72) (0.97) (3.18) (−5.27) (0.12) (−3.14) (0.1153)
Postret 0.0053 −0.0026 0.08%
(5.94) (−7.27)
Postret 0.0106 −0.0003 0.0238 −0.0310 0.0020 −0.0063 4.11 6.34%
(3.46) (−0.27) (7.13) (−9.38) (1.65) (−5.32) (0.0426)
Aret 0.0010 −0.0005 0.01%
(2.48) (−3.04)
Aret 0.0023 −0.0006 0.0112 −0.0122 0.0015 −0.0010 0.16 4.26%
(1.53) (−1.09) (6.95) (−7.64) (2.54) (−1.80) (0.6892)
quarters, providing a sample of 1,386,813 observations. We estimate regressions of the following form:
where
our regression results both at the annual level, where we include each of the four firm-
quarters from each of the five subsequent years, and the five-year level, where we pool
observations across all 20 firm-quarters.
This regression approach leads to a dependence problem, because each quarterly return
can be included as the dependent variable up to four times in the annual regressions and up
to 20 times in the five-year regressions. To correct this problem, we adjust the t-statistics
by dividing by the square root of the maximum number of times √ each observation can
enter the regression. In the annual-level regressions, we divide by 4. In the five-year level
√
regressions, we divide by 20. If all observations entered the maximum number of times,
then this procedure would be asymptotically equivalent to using generalized least squares
with a residual variance-covariance matrix that sets each of the off-diagonal elements re-
lating to the same dependent variable observation equal to the residual variance. However,
because not all observations enter the maximum number of times, our procedure leads to a
slight downward bias in our ‘adjusted’ t-statistics. The F-statistics are adjusted in a similar
manner, dividing by 4 at the annual level and 20 at the five-year level.
Turning first to the regression results in Table 2, we first estimate a simple regression of
return on growth. As expected based on previous research, growth loads with a significantly
negative coefficient in each of the five years. The coefficients have a simple interpretation
in this regression. The intercept provides an estimate of the expected quarterly abnormal
return for the low growth quintile, and the coefficient on growth provides an estimate of the
expected quarterly abnormal return differential between adjacent growth quintiles. Focusing
on the ‘All 20 Quarters’ regression, the intercept is 0.0065(t = 5.36) and the coefficient on
growth is −0.0032 (t = −6.56). These coefficients indicate an annualized abnormal return
to the lowest growth quintile of 2.6% (4 × 0.65%) and an annualized abnormal return to the
highest growth quintile of −2.5% {4 × [0.65% − 4 × 0.32%]} for an annualized average
differential of 5.1%.
The next regression includes growth, surprise (defined earlier as a +1/0/−1 indicator
variable reflecting the sign of the earnings surprise), and a surprise ∗ growth interaction.
This regression allows the sensitivity of abnormal returns to earnings surprises to vary as
a function of the growth quintile to which the stock belongs. The coefficient on growth
remains negative in all regressions and is statistically significant in all regressions except
for year 4. As expected, the coefficient on surprise is consistently positive and highly
statistically significant, indicating that stock returns are correlated with the sign of earnings
surprises. In addition, the coefficient on the surprise ∗ growth interaction is consistently
positive and statistically significant, indicating that the stock returns of high growth firms are
more responsive to earnings surprises than those of low growth firms. The final regression in
Table 2 also includes the earnings forecast error (defined earlier) and a forecast error ∗ growth
interaction. Surprise, forecast error, and their respective growth interactions all load with
positive coefficients, indicating that stock returns respond to both the sign and the magnitude
of earnings forecast errors, and that these responses are increasing in growth. Nevertheless,
even after controlling for all of these effects (which substantially increase the explanatory
power of the regressions), the coefficient on the growth main-effect variable remains reliably
negative. Thus, none of the regressions in Table 2 explain the value vs. growth phenomenon.
However, none of these regressions allow for an asymmetric response to good and bad news
earnings surprises.
We now move on to Table 3, which estimates the regression specification in (1) that
allows for a differential response to good and bad earnings news. In this specification, the
intercept measures the expected abnormal quarterly return on a low growth, zero earnings
surprise observation. The coefficient on growth measures the return differential on zero
earnings surprise observations in adjacent growth quintiles. The coefficients on the good
(bad) indicator variables measure the incremental return for a low growth observation
reporting a positive (negative) earnings surprise. Finally, the coefficient on the good ∗ growth
(bad ∗ growth) interaction measures the return differential on positive (negative) earnings
surprises in adjacent growth quintiles. If the MB effect is independent of earnings surprises

(as depicted in Figure 1(a)), then we should simply continue to observe a significantly
negative coefficient on growth—i.e., the effect should manifest itself regardless of the
sign of the earnings surprise. However, if the MB effect is concentrated in firms reporting
negative earnings surprises (as depicted in Figure 1(b)), then we should see a significantly
negative coefficient on bad ∗ growth, and the coefficient on growth should no longer be
negative.
Consistent with our predictions, the results in Table 3 demonstrate that the MB effect is
concentrated in negative earnings surprise observations. None of the coefficients on growth
or good ∗ growth are significantly negative, and many are significantly positive. To the
extent these coefficients are zero or positive, they indicate that there is either no differential
performance between value and growth stocks or that growth stocks outperform value stocks
in those states of the world where earnings news is neutral or positive. Thus, the fact that
value outperforms growth in these data cannot be explained by the ‘no news’ or ‘good
news’ observations. In contrast, the coefficients on bad ∗ growth are consistently negative
and highly statistically significant, indicating that (consistent with Figure 1(b)) the stock
price response to bad news is much more pronounced for growth stocks. Thus, the return
differential must be embedded in this set of observations. We also test statistically whether
the absolute value of the coefficients on bad ∗ growth are larger than those on good ∗ growth.
If the response to earnings surprises were symmetric within growth quintiles, these two
coefficients would sum to zero. Table 3 reports an F-statistic to test the restriction that
they sum to zero, which is uniformly rejected at conventional significance levels. Thus, the
results in Table 3 provide clear evidence of an asymmetrically large response to negative
earnings surprises in high growth firms.
Table 4 next presents the results of specification (2), which modifies the Table 3 spec-
ification to include the magnitude of the forecast error (FE) in the asymmetric growth
interaction. Specification (2) does not perform as well as specification (1) in explaining the
asymmetric response of the value vs. growth phenomenon. In particular, the coefficient on
growth remains reliably negative in several of the Table 4 regressions, including the overall
results. In contrast, this coefficient is never negative in Table 3. In addition, the asymmetric
reaction to bad news for growth stocks is much more clearly evident in Table 3 than in Table
4. In Table 3, the coefficient on bad ∗ growth is consistently four to five times larger than
that on good ∗ growth, while in Table 4 the analogous coefficient for bad news is only one
to two times as large as that on good news. These results indicate that it is the simple fact
of an earnings disappointment that matters for investors in growth stocks, rather than the
magnitude of the disappointment.
To test our third prediction, Table 5 reports a subset of the regressions in Tables 2 and 3
using alternative return measurement intervals for the dependent variable. In the interest of
brevity, we only report results for the ‘All 20 Quarters’ sample.9 The table reports both the
simple regression of returns on growth and the full regression specification from Table 3 that
allows for an asymmetric response to earnings surprises. Each regression is first reported
using the same ‘fullret’ quarter returns as shown in Tables 2 and 3 as a benchmark. We then
report each of the regressions using the ‘preret,’ ‘postret,’ and ‘aret’ return measurement
intervals. Recall that ‘preret’ spans the first half of the period between formal earnings
announcements, when little earnings news is released, while ‘postret’ captures the second
half of this period, when most earnings news is released. Finally, ‘aret’ captures the three-day
announcement window itself, but excludes any preemptive earnings disclosures.
Focusing first on the simple regressions of returns on growth, we find a negative and
statistically significant coefficient for all of the return measurement periods except ‘preret.’
The relative coefficient magnitudes vary considerably. The coefficient in the ‘fullret’ re-
gression is −0.0032, versus −0.0005 in the ‘preret’ regression and −0.0026 in the ‘postret’
regression, so over 80% of the overall MB effect is concentrated in the ‘postret’ period,
despite the fact that ‘preret’ and ‘postret’ each average 31 days. The coefficient on ‘aret’ is
only −0.0005 implying that three-day earnings announcement window captures less than
20% of the total MB effect. La Porta et al. (1997) and Bernard et al. (1997) also find that only
a small portion of the total MB effect is concentrated in the formal earnings announcement
period. Overall, these results confirm that the MB effect is concentrated in the 31 days lead-
ing up to earnings announcements, but that only a small part of the effect is concentrated in
the three-day announcement window. This is consistent with much of the MB effect being
driven by preemptive earnings disclosures, and in particular with the tendency for managers
of growth firms to preannounce adverse earnings news.
The second set of regressions in Table 5 investigate how the asymmetric response of
growth stocks to negative earnings surprises varies across the different return measurement
intervals. The first regression uses the ‘fullret’ return measurement interval, and confirms our
previous (Table 3) finding that the MB effect is concentrated in growth firms that report neg-
ative earnings surprises. The regressions using ‘preret’ and ‘postret’ generally confirm this
finding, although the results are much stronger in the ‘postret’ returns. In both regressions,
the coefficients on growth and good ∗ growth are non-negative while those on bad ∗ growth
are again negative, relatively large in magnitude and strongly significant. The R-squared of
the ‘postret’ regression is almost four times as large as in the ‘preret’ regression, consistent
with most of the earnings-related variation in returns occurring during the ‘postret’ period
(the R-squareds are 6.3% and 1.7%, respectively). The results for the regression using
‘aret’ are somewhat different. In this regression, there is no evidence of an asymmetrically
large reaction to negative earnings surprises for growth firms, and the coefficient on the
bad ∗ growth interaction is not statistically significant. When combined with the strongly
significant results for the ‘postret’ period (which includes ‘aret’), these results imply that
most adverse earnings news for growth stocks is preannounced, so that the accompanying
stock price reactions generally occur before the earnings announcement period.
4.3. Alternative Measures of Growth
All tests conducted thus far have used the MB ratio as a measure of ‘growth’ or ‘glamour.’
Prior research identifies alternative measures of ‘growth’ or ‘glamour’ that also have pre-
dictive ability with respect to future stock returns. Two of the most frequently encountered
growth proxies are price-to-earnings ratios (Lakonishok et al., 1994) and analysts’ forecast
of long-term earnings growth (Dechow and Sloan, 1997; La Porta, 1996). In Table 6, we
provide our basic regression analysis using these alternative measures of growth to demon-
strate that our results are not sensitive to the particular measure of growth that is employed.
We measure the price-to-earnings ratio as the ratio of stock price to most recent annual EPS
(negative) and zero otherwise, and interaction terms.
Adjusted
Growth Good ∗ Bad ∗ F-Statistic Adjusted
Measure Intercept Growth Good Bad Growth Growth ( p-Value) R-Squared
MB 0.0065 −0.0032 0.06%
(5.36) (−6.57)
MB 0.0158 0.0002 0.0347 −0.0481 0.0022 −0.0095 5.79 7.57%
(3.81) (0.15) (7.70) (−10.78) (1.33) (−5.95) (0.0160)
PE 0.0051 −0.0026 0.04%
(4.45) (−5.46)
PE 0.0137 0.0008 0.0338 −0.0466 0.0025 −0.0093 5.61 7.83%
(3.77) (0.61) (8.42) (−11.76) (1.63) (−6.19) (0.0179)
LTG 0.0071 −0.0033 0.07%
(5.64) (−6.38)
LTG 0.0082 0.0031 0.0246 −0.0262 0.0086 −0.0190 10.59 8.61%
(2.00) (2.04) (5.48) (−5.89) (5.10) (−11.40) (0.0011)
Growth portfolios are measured using MB quintiles (low MB = quintile 0, high MB = quintile 4), PE quintiles (low
PE = quintile 0, high PE = quintile 4), and LTG quintiles (low LTG = quintile 0, high LTG = quintile 4). Growth
portfolios are formed at the beginning of Year 1, and regressions employ returns and earnings data measured over
the subsequent twenty quarters, providing a sample of 1,386,813 observations. We estimate regressions of the
following form:
where
at the end of each fiscal quarter. We measure long-term growth using the median forecast
of long-term growth provided by I/B/E/S in the last month of the fiscal quarter. We then
examine (as before) the relation between stock returns, growth portfolio membership, and
earnings surprises over the subsequent 20 quarters.
Table 6 reports remarkably similar results across all measures of growth. The first two
rows of Table 6 present our original results for the MB ratio for benchmarking purposes, and
then presents results for the price-to-earnings ratio (PE) and analyst forecasts of long-term-
growth (LTG). In all cases, a simple regression of quarterly abnormal returns on growth
yields significantly negative coefficients of similar magnitude, ranging from −0.0026 for PE
to −0.0033 for LTG. These coefficients translate to annual return differentials between the
lowest and highest growth quintiles of 4.16% and 5.12% respectively. When we allow for
an asymmetric response to negative earnings surprises, the negative coefficient on growth

disappears, and the coefficients on bad ∗ growth are all reliably negative and significantly
larger in absolute value than those on good ∗ growth. These results provide clear evidence
of a large asymmetric response to negative earnings surprises for growth firms and confirm
that all of the MB, PE, and LTG return differentials are realized in firm quarters when
negative earnings surprises are released.
4.4. Intertemporal Variation in the Relative Performance of Growth Stocks
The basic result in the paper—that value generally outperforms growth and that this dif-
ference is largely explained by a differential response to adverse earnings surprises—may
seem hard to reconcile with the stock market experience of the late 1990s, during which time
growth stocks substantially outperformed value stocks overall. Yet there is nothing about this
stock market era that is necessarily inconsistent with our arguments or evidence—it could
simply be that this period was characterized by unusually strong earnings performance. To
investigate whether there is a significant intertemporal relation between the value vs. growth
return differential and the nature of earnings surprises, we estimate a regression of hedge
portfolio returns on aggregate differences in earnings surprises. For each calendar quarter in
our sample period, we construct a hedge portfolio return we label MRET(HML)t (the aver-
age ‘fullret’ return for high growth firms, minus the average ‘fullret’ return for low growth
firms), and a net earnings surprise indicator we label MSURP(HML)t (the average value
of SURP for high growth firms minus the average value of SURP for low growth firms).
Results from a quarterly time-series regression of MRET(HML)t on MSURP(HML)t are
as follows:
MRET(HML)t = −0.010 + 0.090 ∗ MSURP(HML)t ; Adj. R 2 = 10.3%; Obs. = 150.

(t-statistic) (−1.57) (4.26)
Consistent with our arguments, the regression indicates that there is a reliably positive in-
tertemporal relation between the differential return on growth stocks versus value stocks
and the extent to which growth stocks report relatively good earnings news.10 Moreover,
the distribution of MSURP(HML) (not reported) indicates that growth strategies will out-
perform value strategies in about 25% of calendar quarters.11 Thus, intertemporal variation
in the relative frequency of good versus bad earnings surprises helps explain variation in
the relative performance of value and growth stocks. This confirms that in periods when
growth stocks experience unusually good earnings performance (such as the late 1990s),
growth stocks can outperform value stocks.
5. Conclusion
We demonstrate that growth stocks exhibit an asymmetrically large negative price response
to negative earnings surprises and show that this asymmetric response to negative earnings
surprises explains the return differential between ‘growth’ and ‘value’ stocks. Another way
of stating this result is that the lower returns of growth stocks relative to value stocks are
entirely attributable to quarters when negative earnings surprises are announced. Growth
stocks perform at least as well as value stocks in quarters when zero or positive earnings
surprises are announced. We also show that the inferior performance of growth stocks is
concentrated in the 31 days leading up to quarterly earnings announcements, when most
earnings-related news is released. Finally, we find that relatively little of the return differen-
tial is observed at the formal earnings announcement date, presumably because managers
of growth firms tend to preannounce negative earnings surprises (Skinner, 1997; Soffer
et al., 2000).
Our results provide strong support for the expectational errors hypothesis in explaining
the inferior returns to growth stocks. LSV (1994) argue that the return differential arises
because investors initially have overly optimistic expectations about the future earnings’
prospects of growth stocks, leading to subsequent price declines when these expectations
are not met. Our evidence is consistent with LSV’s argument—we show that these price
declines are sudden and occur during relatively short periods of time when adverse earnings
news is released, confirming that this is an earnings-related phenomenon.12 Others, such as
Fama and French (1992), argue that the lower returns to growth stocks reflect the fact that
these stocks are less risky on some dimension that has not been identified by academics, but
that is priced by investors. Our findings make this argument less plausible, since it implies
that the risk premium to value investors is only realized in those states of the world where
negative earnings surprises are announced.
Our evidence also has implications for other areas of capital markets research. First, our
research extends previous evidence on non-linearities in the relation between returns and
earnings (e.g., Hayn, 1995; Freeman and Tse, 1992). Second, our findings have implications
for managers’ financial reporting and disclosure strategies. If managers of growth firms are
aware that their firms’ stock prices suffer large downward adjustments when they report
earnings disappointments, they may have incentives to manage reported earnings and/or
manage analysts’ expectations of reported earnings to avoid negative earnings surprises
(e.g., see Bartov, Givoly and Hayn, 2002; Matsumoto, 2002). The evidence provided in this
paper offers a framework for understanding why managers engage in such behavior.
Acknowledgments
We are grateful for the comments of participants at the 2001 Review of Accounting
Studies Conference, particularly John Hand (discussant). We also appreciate the com-
ments of workshop participants at Cornell University, Harvard University, the University of
North Carolina, the University of Oregon, the University of Pennsylvania, the University of
Rochester, and the University of Washington, the 5th Annual Chicago Quantitative Alliance
Conference, and the 13th Annual Prudential Quantitative Conference. We thank I/B/E/S for
providing EPS forecast data. We appreciate financial support from KPMG (Skinner) and
PricewaterhouseCoopers (Sloan). All errors are our own.
Notes
1. Three recent anecdotes illustrate this phenomenon, often referred to by investment professionals as the ‘earnings
torpedo’ effect. First, on December 9, 1997, Oracle (market-to-book ratio = 12) experienced a 29% drop in
its stock price in response to the announcement of earnings of $0.19 versus consensus expectations of $0.23.
Second, on September 21, 2000, Intel (market-to-book ratio = 12) experienced a 22% drop in its stock price
in response to a preannouncement of earnings of about $0.38 versus consensus expectations of $0.41. Third,
on October 12, 2000 Home Depot (market-to-book = 8) experienced a 29% drop in its stock price in response
to a preannouncement of earnings of about $0.28 versus consensus expectations of $0.31.
2. For example, Soffer et al. (2000) report that 67% of the preannouncements in their sample convey adverse
earnings news.
3. Skinner (1997) provides evidence that earlier disclosure of adverse earnings news reduces expected litigation
costs. However, there are other potential reasons why managers might preannounce adverse earnings news
more often than other earnings news. For example, they view preannouncements as a means to preserving their
reputation and credibility with security analysts who follow their firm’s stock.
4. Note that one limitation of this procedure is that measures of earnings surprise derived using this forecast
date are unlikely to pick up surprise information that is revealed in the earlier part of the quarter. We also
replicated our results using the earliest consensus forecast available after the release of the prior quarter’s
earnings announcement. The results were almost identical, suggesting that very little earnings news is released
in the earlier part of the fiscal quarter.
5. Our results are robust to alternative methods of computing abnormal returns, including a simple market
adjustment and a market model adjustment. We explicitly avoid making an adjustment for the MB effect,
because our objective is to explain the MB effect.
6. The fact that the use of size-adjusted returns yields almost identical inferences to market-adjusted or market
model adjusted returns in our tests supports this explanation.
7. To create the plot for ‘value’ and ‘growth’ we take the bottom and top growth quintiles (as before) respectively
and within each of these form 20 portfolios by ranking the observations based on FE. We then plot the mean
returns against the mean forecast error for each of these 20 portfolios and join these points using the Excel
smooth line charting feature.
8. The non-linear, S-shaped relation between earnings and returns is noted by Freeman and Tse (1992), as well
as others since then.
9. The results display a consistent pattern during each of the five component years.
10. The regression also indicates that value outperforms growth on average for our sample. The mean value
of MSURP(HML) is −0.052, so the mean difference between growth and value is −0.010 + (0.090 ∗
−0.052) = −0.015 or about −1.5% per quarter.
11. The 75th percentile value of MSURP(HML) is 0.11, indicating that the expected value of MRET at this level
of MSURP(HML) is approximately 0[−0.010 + (0.090 ∗ 0.11) = 0]. Thus, the expected value of MRET
(HML) is greater than zero for the upper quartile of the MSURP(HML) distribution.
12. Our evidence may also shed light on recent psychological explanations for anomalous stock market behavior,
such as those offered by Daniel, Hirshleifer and Subrahmanyam (1998) or Barberis, Shleifer and Vishny
(1998). Barberis et al. (1998) argue that stock prices overreact to consistent patterns of good or bad news. This
is consistent with the notion that growth stocks gradually become overpriced as investors observe a series of
consistently good earnings reports, but then “fall to earth” when those stocks report earnings disappointments
and investors realize their expectations were overly optimistic.
References
Abarbanell, J. and R, Lehavy. (2000). “Biased Forecasts or Biased Earnings? The Role of Earnings Management in
Explaining Apparent Optimism and Inefficiency in Analysts’ Earnings Forecasts.” Unpublished paper, University
of North Carolina, Chapel Hill, NC, and University of California, Berkeley, Berkeley, CA.
Barberis, N., A. Shleifer and R. Vishny. (1998). “A Model of Investor Sentiment.” Journal of Financial Economics
49, 307–343.
Bartov, E., D. Givoly and C. Hayn. (2002). “The Rewards to Meeting or Beating Earnings Expectations.” Forth-
coming, Journal of Accounting and Economics.
Basu, S. (1977). “Investment Performance of Common Stocks in Relation to Their Price Earnings Ratios: A Test
of the Efficient Market Hypothesis.” Journal of Finance 32, 663–682.
Bernard, V. L., J. Thomas and J. Wahlen. (1997). “Accounting-Based Stock Price Anomalies: Separating Market
Inefficiencies from Risk.” Contemporary Accounting Research 14, 89–136.
Brown, L. D. (2001). “A Temporal Analysis of Earnings Surprises: Profits Versus Losses.” Journal of Accounting
Research 39, 221–241.
Daniel, K., D. Hirshleifer and A. Subrahmanyam. (1998). “Investor Psychology and Security Market Under- and
Overreactions.” Journal of Finance 53, 1839–1885.
Dechow, P. M. and R. G. Sloan. (1997). “Returns to Contrarian Investment Strategies: Tests of Naı̈ve Expectations
Hypotheses.” Journal of Financial Economics 43, 3–27.
Dreman, D. and M. Berry. (1995). “Overreaction, Underreaction, and the Low-P/E Effect.” Financial Analysts’
Journal 51, 21–30.
Fama, E. F. (1998). “Market Efficiency, Long-Term Returns, and Behavioral Finance.” Journal of Financial
Economics 49, 283–306.
Fama, E. F. and K. R. French. (1992). “The Cross-Section of Expected Stock Returns.” Journal of Finance 47,
427–465.
Freeman, R. N. and S. Y. Tse. (1992). “A Nonlinear Model of Security Price Responses to Unexpected Earnings.”
Journal of Accounting Research 30, 185–209.
Hayn, C., “The Information Content of Losses.” Journal of Accounting and Economics 20, 125–153.
Kasznik, R. and B. Lev. (1995). “To Warn or Not to Warn: Management Disclosures in the Face of an Earnings
Surprise.” The Accounting Review 70, 113–134.
Kothari, S. P., J. S. Sabino and T. Zach. (1999). “Implications of Data Restrictions on Performance Measurement
and Tests of Rational Pricing.” Unpublished paper, Massachusetts Institute of Technology, Cambridge, MA.
Lakonishok, J., A. Shleifer and R. Vishny. (1994). “Contrarian Investment, Extrapolation, and Risk.” Journal of
Finance 49, 1541–1578.
La Porta, R. (1996). “Expectations and the Cross-Section of Stock Returns.” Journal of Finance 51, 1715–1742.
La Porta, R., J. Lakonishok, A. Shleifer and R. Vishny. (1997). “Good News for Value Stocks: Further Evidence
of Market Efficiency.” Journal of Finance 52, 859–874.
Matsumoto, D. A. (2002). “Management’s Incentives to Avoid Negative Earnings Surprises.” Forthcoming, The
Accounting Review.
Skinner, D. J. (1994). “Why Firms Voluntarily Disclose Bad News.” Journal of Accounting Research 32, 38–60.
Skinner, D. J. (1997). “Earnings Disclosures and Stockholder Lawsuits.” Journal of Accounting & Economics 23,
249–282.
Soffer, L. C., S. R. Thiagarajan and B. R. Walther. (2000). “Earnings Preannouncement Strategies.” Review of
Accounting Studies 5, 5–26.

Earnings Surprises, Growth Expectations, and Stock Returns or Don't Let An Earnings Torpedo Sink Your Portfolio

Uploaded by

Copyright:

Available Formats

Earnings Surprises, Growth Expectations, and Stock Returns or Don't Let An Earnings Torpedo Sink Your Portfolio

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Earnings Surprises, Growth Expectations, and Stock Returns or Don't Let An Earnings Torpedo Sink Your Portfolio

Uploaded by

Copyright:

Available Formats

Review of Accounting Studies, 7, 289–312, 2002

Earnings Surprises, Growth Expectations, and Stock

Keywords: earnings surprises, earnings torpedoes, value, growth, glamour

JEL Classification: G14, M41

It is well-established that ‘growth’ or ‘glamour’ stocks have historically underperformed

2. Hypothesis and Empirical Predictions

(b) Asymmetric response to negative surprises:

3. Sample and Research Design

We begin by reporting descriptive evidence on each of our predictions after which we

4.1. Descriptive Evidence

Earnings Surprise Portfolio

Negative Zero Positive All

4.2. Regression Analysis

Ritτ = α + β1 · Growthit + β2 · Gooditτ + β3 · Baditτ + β4 · (Gooditτ ∗ Growthit )

As an alternative specification, we also estimate regressions of the following form, and

Ritτ = α + β1 · Growthit + β2 · Gooditτ + β3 · Baditτ + β4 · FEitτ

Quarters from: Forecast Forecast Adjusted

Year 1 0.0014 −0.0014 0.02%

Year 2 0.0059 −0.0029 0.05%

Year 4 0.0045 −0.00239 0.03%

Year 5 0.0051 −0.0026 0.08%

surprises in adjacent growth quintiles. If the MB effect is independent of earnings surprises

4.3. Alternative Measures of Growth

an asymmetric response to negative earnings surprises, the negative coefficient on growth

4.4. Intertemporal Variation in the Relative Performance of Growth Stocks

MRET(HML)t = −0.010 + 0.090 ∗ MSURP(HML)t ; Adj. R 2 = 10.3%; Obs. = 150.

You might also like