Information Uncertainty and the Post-Earnings Announcement Drift in Europe
Xavier Gerard
xavier.gerard@rbs.com
The Royal Bank of Scotland
250 Bishopsgate, London UK, EC2M 4AA
Abstract
This paper investigates the relation between earnings announcement abnormal return,
abnormal trading volume, and subsequent returns for a large sample of European firms. In
addition to bringing new insights for the dynamics of the abnormal return and abnormal
trading volume effects, our analysis provides out-of-sample confirmations of several prior US
findings. We show that each measure of market surprise is positively related to future
abnormal returns, and especially so when information uncertainty is high. These two effects
appear to be complementary as each retain some incremental predicting power for future
returns. Finally, our empirical evidence is not limited to small, illiquid, stocks; and it is robust
to controlling for potential market microstructure biases.
JEL classification: M4, G12, G14
Keywords: Market efficiency; Trading Volume; High-Volume Premium; Post-Earnings
Announcement Drift; Earnings Surprises; Information Uncertainty
2
1. Introduction
Since Ball and Brown (1968), over forty years ago, an extensive body of mainly US
academic research has documented a positive relation between earnings forecast errors,
computed using either a time series approach or analyst estimates, and abnormal post-
announcement returns. In contrast, due to several data limitations, the evidence for the post-
earnings announcement drift in Europe is relatively scarce. The historical coverage for analyst
forecasts of interim numbers, as well as for the actual interim numbers released by the firms,
is typically low or unreliable in Europe. Moreover, pan-European accounting studies are
plagued by differences of accounting practices across member states.
This paper extends the US evidence to a large sample of European firms with annual as
well as interim earnings announcements from 1997 to 2010. To avoid using analyst forecasts
of interim numbers or the actual interim numbers released by the firm, we resort to some
recent academic findings and quantify the degree of market surprise with some market related
information, namely, the abnormal return and the abnormal trading volume at the time of the
announcement. The use of market data implies that our metrics of market surprise can be
easily computed for a large number of European firms. Moreover, by using market data we
largely alleviate concerns related to differences of accounting practices across European
countries.
The current predominant belief for the post-earnings announcement drift is that it is
caused by some form of under-reaction to the information contained in earnings
announcements. Although previous empirical work has focused on the earnings surprise, it
has recently been noted that some important non-earnings information is also released at the
time of the announcement. For example, firms provide information about components of
earnings such as sales, and operating margins. Still further, earnings announcements tend to
be accompanied by conference calls and press releases where valuable information is
disseminated. Therefore, to the extent that the abnormal return captures a wide range of
earnings and non-earnings related news, it could be argued that it is a broader measure of
market surprise than traditional proxies.
In contrast with the post-earnings announcement drift of Ball and Brown (1968), the
abnormal volume anomaly has been discovered recently. It refers to the outperformance of
high volume stocks relative to low volume stocks following an earnings announcement.
Explanations for this anomaly includes models of capital market equilibrium (Merton, 1987),
behavioural biases arising from the limitations of individual investors to process large
3
amounts of information (Barber and Odean, 2004), and risk-based arguments where the
abnormal volume is seen as a proxy for opinion divergence (Garfinkel and Sokobin, 2006).
Given that the empirical evidence for this effect is still limited, extending the analysis to the
European market should help bring about some valuable insights.
Last but not least, we investigate the role played by information uncertainty in explaining
the pay-offs associated with the abnormal volume and the abnormal return effects. Our proxy
for information uncertainty is the idiosyncratic volatility of the securities in our sample.
Hirshleifer (2001) posits that greater uncertainty combined with the lack of accurate feedback
about fundamentals leave more room for behavioural biases. Although behavioural biases
could well be accentuated in settings of higher information uncertainty, it is also possible that
idiosyncratic risk constitutes a limit to arbitrage preventing investors from eliminating the
anomalies. In any case, we predict that high idiosyncratic risk is associated with larger drifts,
and this analysis should help further our understanding of the mechanisms at play behind
these two empirical patterns.
Our results are four-fold. First, we find, in line with the US evidence, that a measure of
market surprise, computed as the abnormal return around an earnings announcement, is
positively related with future returns in Europe. The effect is not short lived as we show that
firms with a positive surprise in one quarter continue to surprise the market, in the same
direction, up to one year after the announcement.
Second, we are able to demonstrate, to our knowledge for the first time in Europe, the
existence of a strong abnormal volume anomaly following earnings announcements. Firms
with high abnormal volume around their earnings announcement date outperform low
abnormal volume stocks for up to ninety days after the event. This outperformance is
followed by a short period of underperformance, which strongly reverts one year after the
event. Therefore, as with the abnormal return effect, we find that firms which surprise the
market in one quarter continue to do so one year later. This evidence is echoed by patterns of
abnormal trading volume following the earnings announcement date. Those firms with the
highest abnormal trading volume in one quarter also experience a surge in their trading
volume one year later. All in all, these results strongly suggest that investors in Europe fail to
understand the implications of current announcements for future ones.
Third, as predicted, we show that the anomalies tend to generate stronger premiums when
information uncertainty is larger. For instance, although significant abnormal returns are
observed within stocks that have high and low idiosyncratic risk, we find that the abnormal
4
return and abnormal volume effects are typically stronger within those stocks that experience
the highest level of idiosyncratic volatility. Similarly, we show that the abnormal volume
effect arises primarily during periods of high aggregate idiosyncratic risk.
Finally, the two effects appear to capture independent dimensions of the market surprise
around an earnings announcement. Controlling for the degree of abnormal volume or
abnormal return around an earnings announcement, we continue to find that the other effect is
associated with positive future returns. Given their independent information contents we
investigate their combined predictive power. We find that a trading strategy based on these
two indicators of market surprise earns a monthly quintile spread return of approximately
0.70%. The trading strategy is implemented with a monthly rebalancing frequency so that a
signal is used for the first time on average two weeks after the earnings announcement, hence
alleviating concerns that our findings are contaminated by potential market microstructure
biases. Controlling for risk does not explain away the premium earned by the combined
strategies. Moreover, these findings are not restricted to small firms, and they are robust to
taking into account stock illiquidity and the volatility of stock returns over the announcement
days.
The remainder of this paper is organised as follows. The next section discusses the
relevant literature and motivates our analysis. The third section describes the sample data and
introduce the main variables of this study. The fourth section presents and discusses our
empirical findings. We test the robustness of our results in section five. Finally, the last
section concludes the study.
2. Literature Review and Motivations
This analysis is related to prior US findings that show a positive association between
measures of earnings surprise and future abnormal returns. The earnings surprise indicator,
calculated at the time of an announcement, is typically defined as the scaled difference
between the actual earnings figure and a proxy for market expectations computed using either
analyst forecasts or the time series of prior earnings. Despite having been extensively
researched since its discovery more than forty years ago by Ball and Brown (1968), the
reason for the existence of the post-earnings announcement drift continues to be heavily
debated. The prior empirical evidence offers little support for a risk-based explanation or
potential flaws in research design (Bernard and Thomas, 1989). Instead, the current
predominant belief for the post-earnings announcement drift is that it is caused by some form
of under-reaction to the information contained in earnings announcements. The exact nature
5
of the under-reaction remains vague however. Bernard and Thomas (1990) point out that the
drift occurs at subsequent earnings announcement dates. This finding lead them and others
(see for instance Ball and Bartov, 1996) to argue that investors fail to understand the
implications of current earnings for future ones. However, Jacob et al (1999) offer an
alternative interpretation of the causes of the post-earnings announcement drift when looking
at the autocorrelation structure of forecast errors, and Livnat and Mendenhall (2006) put
forward an explanation based on a more general hypothesis of under-reaction to earnings
information.
Although the bulk of the prior evidence has focused on the earnings surprise, little
attention has been paid to non-earnings information. However, considering non-earnings
related news make sense for several reasons. For instance, Liu and Thomas (2000) show that
a significant portion of the market reaction around earnings announcement is due to non-
earnings related information. It is also obvious that some important non-earnings related news
are released at the time of an announcement. For example, firms provide information about
components of earnings such as sales, and operating margins (see Jegadeesh and Livnat,
2006). Still further, earnings announcements tend to be accompanied by conference calls and
press releases where some additional valuable information is disseminated. Finally, a number
of recent empirical findings suggest that market participants incorrectly value non-financial
information. For instance, Ragjopal et al (2003) find that investors overestimate the valuation
implications of order backlogs, while Gu (2005) and Deng et al (1999) show that investors
systematically underweight patent counts, as well as the level change in patent citations.
In this study we follow Brandt et al (2006) and use as our first measure of market surprise
the abnormal stock return around an earnings announcement date. To the extent that the
abnormal return captures a wide range of earnings and non-earnings related news, both
tangible and intangible, it could be argued that it is a broader measure of market surprise than
previous indicators. In line with earlier findings, Brandt et al (2006) report a strong post-
earnings announcement drift when ranking stocks on the basis of their abnormal return at the
time of the earnings announcement. However, it has to be noted that the magnitude of the
price change around the earnings announcement could also be increasing in the information
content of the news and decreasing in the degree of under-reaction.
Using market data as a proxy for the market surprise also achieves another objective, as it
helps alleviate several data issues that have plagued European studies. First, the historical
coverage for analyst forecasts of interim numbers, as well as for the actual interim numbers
released by the firms, is typically low or unreliable in Europe. As a result the empirical
