BOOK AND TAXATION
Journal of International Accounting, Auditing and Taxation
15 (2006) 92–108
The relationship between returns and unexpected
earnings: A global analysis by accounting regimes
Department of Accounting, Miller College of Business, Ball State University, Muncie, IN 47306, United States
Numerous studies have documented a long-term association between earnings and returns. Surprisingly,
few attempts have been made to internationally examine market reactions to earnings releases over return
windows less than 12 months. This paper globally explores the market reaction to unexpected earnings defined
by both the change in earnings per share (EPS) and analyst forecast errors (AFE) using a 1-month return
window. First, the existence of the earnings–returns relationship is examined using a sample of firms from 32
countries grouped into accounting regimes. Accounting regimes represent groups of countries that exhibit
similarities in accounting standards, stock market characteristics, corporate governance mechanisms, and
economic conditions. Thus, similar reactions to earnings are expected within regimes. Next, the incremental
information content of analyst forecasts, a proxy for investors’ earnings expectations, is examined. Finally,
changes in the structure of the earnings–returns relationship over time are investigated. Results support
the existence of a relationship between earnings and returns in all accounting regimes. In addition, analyst
forecast errors appear to be incorporated into earnings expectations in most developed countries. Finally,
evidence suggests that the significance and explanatory power in the earnings–returns relationship has
increased in recent years.
© 2006 Elsevier Inc. All rights reserved.
Keywords:Capital markets; Earnings–returns relationship; International accounting
Assessing the usefulness of financial information has become a primary goal of accounting
research. In this study, I use the earnings–returns relationship to examine the usefulness of earnings
in an international setting. Specifically, this paper globally examines the monthly market reaction
to unexpected earnings defined by both change in earnings per share (CEPS) and analyst forecast
errors (AFE). It adds to prior literature in two ways. First, the incremental explanatory power of
AFE versus the change in EPS is internationally explored. Given the documented superiority of
analyst earnings forecasts over other earnings estimates, efficient investors will likely incorporate
analyst forecasts into earnings expectations.1 The degree that analyst forecasts are incorporated
into earnings expectations is measured by the strength of the relationship between analyst forecast
errors and returns around the earnings announcement. The use of analyst forecasts as proxies for
earnings expectations provides some insight into the process investors use in forming earnings
The next contribution of this paper is the international examination of the stability of the
earnings–returns relationship over time. Though research in the U.S. has examined changes
in the earnings–returns relationship across time to determine if there have been variations in
the value relevance of accounting information (e.g.,Francis & Schipper, 1999), little international
research addresses this issue. This is an interesting issue because there have arguably
been greater changes in factors that may impact the earnings–returns relationship outside the
U.S. (e.g., quality of accounting information, characteristics of capital markets, access to financial
and non-financial information) as compared to the changes that have occurred within
the U.S. I extend this research to an international setting to determine if there have been
changes in the value-relevance of earnings and the information content of analyst forecast
The results of analyses using data from the entire time period pooled by accounting
regimes reveal a significant market reaction to the announcement of earnings (either AFE or
CEPS) in all regimes. This indicates that investors view accounting information as value relevant
and react when earnings do not meet expectations. In addition, analyst forecast errors
appear to have incremental explanatory power over the change in EPS in most regimes. This
suggests that investors around the world act efficiently to incorporate analyst earnings forecasts
into earnings expectations. Results of the multi-period analysis reveal a trend toward
increased significance of the reaction to the earnings release. This implies that in recent years,
investors are more likely to react to the release of earnings. In addition, the results suggest
that investors are more likely to incorporate analyst forecasts into earnings estimates in recent
The results of this study are likely to be of interest to researchers, investors, and educators. The
analysis suggests that there has been an increase in the market reaction to the release of earnings.
New research aimed at identifying factors which have driven this change would contribute to the
literature. Investors can use the understanding of how market earnings expectations are formed to
improve investment decisions. The increase in significance of earnings and/or analyst forecasts
in recent time periods may be partially caused by improvements in accounting standards. If so,
such standards improvements may provide guidance to less developed countries attempting to
develop accounting standards. Educators could integrate the results of this paper in upper level or
graduate accounting classes.
1Analyst forecasts of earnings have been shown to be more accurate than time-series estimates of earnings (Brown,
Hagerman, Griffin, & Zmijewski, 1987a). In addition, analyst forecasts of earnings explain greater abnormal returns
around earnings announcements in the U.S. than other proxies for market expectations of earnings (Brown, Hagerman,
Griffin, & Zmijewski, 1987b). Analytical research shows that analyst forecasts serve as a reasonable proxy for investor
beliefs with error (Abarbanell, Lanen, & Verrecchia, 1995).
2Prior literature has not used analyst earnings forecasts because it has not been widely available until recent years (see
Meek & Thomas, 2004for a discussion of this issue).
2. Literature review and the role of accounting regimes
An extensive amount of research has examined the earnings–returns relationship. Early
research attempted to identify an association between earnings and returns in a single country.
Using data from the U.S.,Ball and Brown (1968) were the first to confirm the relationship between
returns and earnings. Similar analyses have been conducted in many other countries, with results
typically consistent with those in the U.S.3 More recent international earnings–returns research has
attempted to identify cross-country differences in the value relevance of accounting information.
Though value relevance has been examined using various methodologies, a consistent component
of these studies is an international comparison of the estimated coefficients and explanatory power
derived from earnings–returns regressions. Specifically, samples of firms from various countries
are often used to estimate a regression of 12 to 18 month returns on EPS and/or the change in EPS.
Numerous studies have followed this methodology. Using a sample of 17 developed countries,
Alford, Jones, Leftwich, and Zmijewski (1993)document a significant relationship between
change in earnings and returns in only six countries.Joos and Lang (1994) examine the value
relevance of earnings in Germany, France and the United Kingdom. Results suggest accounting
information produced using French GAAP is more relevant than that produced under U.K. or
German GAAP. Using both price and return models,Harris, Lang, and Moller (1994) find that
accounting information produced under U.S. GAAP is more relevant than that produced under
German GAAP.Hall, Hamao, and Harris (1994) show that the earnings–returns relationship is
weaker in Japan than the U.S.Ali and Hwang (2000) document a relationship between the value
relevance of accounting data and country-specific factors: bank versus market orientation, private
versus public standard settings, tax-financial reporting conformity, accounting regime and
spending on auditors.
Though this research makes a significant contribution to the understanding of the international
capital markets, it has limitations. Specifically, the majority of this analysis fails to use analyst
forecasts as proxies for unexpected earnings. In addition, returns are aggregated over periods
often in excess of 1 year. Finally, most previous international research does not examine changes
in the value relevance of accounting.
The analysis for this paper is conducted by pooling data by accounting regimes. Accounting
regimes represent groups of countries that exhibit similarities in accounting standards. Countries
within these regimes are also likely to exhibit similarities in market characteristics, corporate
governance mechanisms and capital market attributes. The grouping of countries into the regimes
used in this study is based on a combination of the classification systems proposed byDoupnik
and Salter (1993)and Gray (1988). Using these two studies as a basis for categorizing, eight
accounting regimes are identified: (1) North America (2) Other Anglo-Saxon (3) Nordic, (4)
Developed Latin, (5) Emerging Latin, (6) Germanic, (7) Japanese and (8) Emerging Asian/Near
Substantial variation across regimes exists in the characteristics of accounting systems, strength
of capital market attributes, corporate governance and characteristics of analyst forecasts.Fig. 1
3See, for example, Lev and Yahalomi (1972) for examination of the earnings-returns relationship in Israel; Kubota
(1980)for Japan; Cheung and Sami (2000) for Hong Kong among others.
4The European Union’s Directives impact on accounting standards was considered when formulating these regimes.
Previous research which was relied upon to generate these regimes considered similarities and differences in the post-
Directive period. Thus, any impact of the Directives were considered in the formation of the groupings. In addition,
research shows that directives were not successful in harmonizing standards (Bindon & Gernon, 1995).
documents some of these differences. The highest use of accruals and the greatest
between financial accounting systems and tax reporting occurs in the North America and Other-
Anglo Saxon regime countries (Francis, Khurana, & Pereira, 2003). The highest degree of
conformity between tax and financial reporting accounting occurs in Germanic, Developed Latin
and Japanese regimes. The importance of capital markets, as measured by stock market capitalization
to GDP, is highest in the North American, Other Anglo Saxon and Japanese regimes
(Francis et al., 2003). Corporate governance, as measured by the strength of investor protection
laws and law enforcement institutions, is strongest in the North American, Other Anglo
Saxon, and Nordic regimes and is weakest in the Latin and Emerging Asian and Near East
regimes (Defond & Hung, 2004).5 Finally, international differences exist in the number of analysts
providing forecasts and the accuracy of these forecasts (Basu, Hwang, & Jan, 1998; Hope,
Variations between accounting regimes may cause differences in the earnings–returns relationship.
Differences in the quality of accounting standards have been shown to affect the
relationship between earnings and returns (Ali & Hwang, 2000). The relative importance of
a country’s equity market to the economy may impact the relationship between earnings and
returns. In addition, the strength of corporate governance mechanisms has been shown to
affect the strength of the relationship between earnings and returns (Ball, Kothari, & Robin,
2000). The availability and/or accuracy of earnings forecast may impact the earnings–returns
3. Hypotheses development
Two sets of hypotheses are used in this study. The first set of hypotheses examines the market
reaction to the announcement of earnings using data from the entire research period. This serves
as a replication of previous research and provides base-line evidence that a market reaction exists.
5Despite the recent financial reporting failures, research indicates that two key measures of corporate governance,
investment protection law and enforcement of these laws, are strongest in countries the North American and Other Anglo-
Saxon regime (Defond & Hung, 2004). Further, Lang (2003) states that post-Enron corporate governance in the U.S.
remains at least as strong as any other country in the world and the Enron scandal is likely to further strengthen corporate
governance in the U.S.
Cross regime differences in the strength of the earnings–returns relationship may be caused by
differences in the quality of accounting information, strength of corporate governance, and the
nature of capital markets. Any of these conditions (or a combination of conditions) may cause
the reduction of a market response to earnings announcements. Thus, the first hypothesis can be
stated as (in null form):
H1.No market reaction to the release of earnings will exist in any accounting regime.
The second hypothesis provides a more detailed exploration of the information used by
investors when forming earnings expectations. Previous research in the U.S. has primarily used
two proxies for investors’ earnings expectations: prior year’s earnings and mean analyst earnings
forecasts. Using prior year’s earnings as a proxy for earnings expectations implies that
investors base their expectation of current year’s earnings on the prior year’s earnings alone.
The use of analyst forecasts as proxies for earnings expectations assumes that investors have
access to analyst forecasts of earnings and incorporate that information into earnings expectations
or, alternatively, have access to the same information as analysts and process it in a
similar manner. International variation could exist because investors may not possess equal access
to financial and non-financial information. In addition, differences in the accuracy of earnings
forecasts may impact the earnings returns relation. This leads to the second hypothesis (in null
H2.Analyst forecast errors do not provide any incremental information content over change in
EPS within each accounting regime.
The second set of hypotheses examines the changes in the market reaction to earnings over time.
The first two hypotheses in this set (H1mp and H2mp) are multi-period replications of the two
single-period hypotheses. The first multi-period hypothesis examines the existence of a market
reaction to earnings announcements during the month of release over each of the three time periods.
The market reaction to the release of earnings may change over time for many reasons. As time
passes, accounting standards change and often improve. With such improvements, earnings may
become more reflective of a firm’s current performance and future potential. In addition, there
has been some evidence that corporate governance procedures have converged in recent years
(Gillan & Stark, 2003). This may improve the quality of accounting information. There have
also been increases in the accessibility to both foreign and domestic stock markets. Specifically,
reports that the dollar volume of cross-border equity offerings tripled between 1995 and 1999
(PricewaterhouseCoopers, 2001). Hence, as time passes, it is more likely that the market will
react to the release of earnings information. This leads to the first multi-period hypothesis (in null
H1mp.No market reaction to the release of earnings will exist in any time period for any
The second multi-period hypothesis provides an examination of the information content of
analyst forecast errors relative to the change in EPS to explain returns around earnings announcements
is examined. As previously stated, analyst forecast errors are expected to have greater
explanatory power in regimes where investors have access to accounting and non-accounting
information, including analyst forecasts, and have the ability to effectively and efficiently
process such information. The increase in analysts following companies is likely to increase
investor’s exposure to forecasted earnings information. Thus, it is possible the relationship
between the explanatory power of the change in EPS versus analyst forecast errors that is iden
tified in H2 is not constant over time. This leads to the second multi-period hypothesis (in null
H2mp.Analyst forecast errors do not provide any incremental information content over the
change in EPS within any time period across each accounting regime.
The final hypothesis attempts to explicitly capture the relationship between time, estimated
coefficients and unexpected earnings of the earnings–returns relationship. The purpose
of this hypothesis is to examine whether time, which serves as a proxy for development
in accounting standards, corporate governance mechanisms and capital markets characteristics
affect the earnings–returns relation. If capital markets’ efficiency has improved
over time, the earnings–returns relationship should be stronger. Similarly, enhancement
of accounting standards and corporate governance mechanisms are likely to strengthen
the earnings–returns relation. Finally, expansion of information available to investors may
strengthen the earnings–returns relationship. This leads to the final hypothesis (in null
H3mp.The earnings–returns relationship does not change over the time periods examined.
The methodology used in this study is designed to examine both the stock market reactions to
the release of earnings information and investors’ expectations of earnings. The first hypothesis
requires estimation of the market response to earnings releases across accounting regimes. This is
accomplished by the use of two models. In the first model, 1-month excess returns are regressed
on the change in EPS as shown below:
ERjt = α0 + α1CEPSjτ + ejt (1)
where ERjt is 1-month excess returns for firm j and month t, CEPSjτ is change in EPS for firm j
and yearτ deflated by price, ejτ is regression error term for firm j and year τ.
Excess returns are calculated using the market-adjusted model for the month of the earnings
release. The market return is the return on an equally weighted portfolio of stocks in the
In the second model, excess returns are regressed on analyst forecast errors as shown in the
ERjt = β0 + β1AFEjτ + μjτ (2)
where AFEjτ is analyst forecast error for firm j and year τ deflated by price and μjτ is regression
error term for firmj and year τ.
AFEjτ is calculated using the mean consensus forecast of annual earnings. The most recent
consensus forecast prior to the earnings announcement release date is used.
H1 is tested by pooling data over the entire research period and estimating Eqs.(1) and (2) on
an accounting regime-by-accounting regime basis. If eitherα1 or β1 is significant in a regime,
then there is a market reaction to the release of accounting earnings, leading to the rejection
of H1 for that regime. H1mp is tested by estimating Eqs.(1) and (2) on accounting regime-byaccounting
regime basis with data pooled over each of the three sub-periods. If eitherα1 or β1
is found to be significant in a given sub-period, then there is a market reaction to the release
of accounting information, leading to the rejection of H1mp for that accounting regime in
H2 and H2mp explore the information content of change in EPS and AFE. Tests of incremental
information content are used to determine whether AFE is associated with excess returns after
controlling for the change in EPS. This is accomplished through a regression of excess returns
on the two measures of unexpected earnings: the change in annual EPS and AFE as shown in the
ERjt = ϕ0 + ϕ1AFEjτ + ϕ2CEPSjτ + ωjt (3)
whereωjt is the regression error term for firm j at year τ.
H2 is tested by estimating Eq.(3) using accounting data from the entire research period pooled
by accounting regime. Similarly, H2mp is tested by estimating Eq.(3) using data from each of the
three sub-periods pooled by accounting regime. AFE possess incremental information content
over the change in EPS in accounting regimes ifϕ1 is significant. More specifically, if ϕ1 is
significant andϕ2 is not, then analyst forecasts have explanatory power, suggesting that analyst
forecasts approximate investors’ expectations of current year’s earnings. Ifϕ1 and ϕ2 are both
significant, then the change in EPS has incremental explanatory power over AFE and AFE has
incremental explanatory power over the change in EPS. In either case H2 (H2mp) would be
rejected for the period (sub-period).
H3mp is tested by explicitly examining the relationship between time and various attributes
of the earnings–returns relationship. To perform this test, a time variable is defined to take a
value of one for the first sub-period, a value of two for the second sub-period, and a value of
three for the last sub-period (Aboody & Lev, 1998). Spearman rank correlations between time,
regression coefficients,t-statistics, and adjusted-R2s (from all models) are calculated. Significant
correlation between any of these variables and time is indicative of changes in the structure of the
earnings–returns relationship and leads to the rejection of H3mp.
H3mp is also tested by examining the significance of differences in adjusted-R2s. Using the
Vuong test, adjusted-R2s for all three models are compared across the time periods. Significant
increases in adjusted-R2s over time suggest a stronger association between earnings and returns,
leading to the rejection of the H3mp.
5. Data and descriptive statistics
The sample used in this study is selected based on data availability from Compustat Global
Vantage (GV) and Institutional Brokers Estimation System (I/B/E/S). The GV database contains
monthly stock prices and dividends for U.S. and foreign companies. The I/B/E/S database contains
analyst forecasts of earnings and actual earnings for U.S. and foreign companies. Data are collected
for the years 1987–1998. The final sample used in this analysis consists of a total of 66,123 firmyears
(10,096 companies) from 34 countries and 8 accounting regimes.7
6Landsman and Maydew (2002) suggest that the use of change in earnings per share is superior to analyst forecasts
when examining the earnings–returns relation across time. However, because this paper seeks to determine changes in
the information investors use to formulate earnings expectations, analyst forecasts are retained in the analysis.
7Previous research shows that the market response to earnings releases varies due to firm-specific characteristics (see
Kothari, 2001for a summary of factors that influence the earnings–returns relation). Over time, the composition of the
sample is this study has changed. To mitigate the impact of changes in the sample on the earning-returns relation, the
multi-period analysis conducted in this paper is re-estimated using a constant sample of firms. That is to say, only firms
Analysis (not reported) indicates regimes are similar in terms of industries represented. To
reduce the effect of extreme observations, each variable used in the analysis is winsorized
Descriptive statistics are presented inTable 1. Mean excess returns are positive for six of the
eight regimes and are significantly different from zero in two regimes. The median excess return
tends to vary slightly around zero. The mean (median) CEPS is positive in five (seven) of the eight
accounting regimes, indicating that EPS has tended to increase during the research period. Mean
analyst forecast errors are negative and significantly different from zero in all eight regimes. This
indicates that analysts, on average, are overly optimistic in forecasting earnings. Median analyst
forecast errors tend to be zero or slightly negative, again indicative of overly optimistic analyst
forecasts. Interestingly, analyst forecasts tend to be the most overly optimistic in the less developed
countries of the Emerging Latin and Emerging Asian/Near East regimes. Analyst forecast errors
possess a great deal of variability, as standard deviations range from 3.5 to 10.7 larger than the
mean analyst forecast error.
The results of tests of the first hypothesis are presented inTable 2.9 The CEPS regressions
reveal a positive and significant relationship between CEPS and excess returns in all eight regimes.
The results of the AFE regression reveal a weaker relationship with five of the eight coefficients
for the AFE variable positive and significant. These results provide support for the rejection of
H1 (under either the CEPS or AFE specification) in all regimes. In addition, the results suggest
that investors react to at least one of the two proxies for unexpected earnings in all regimes.
H2 examines the incremental explanatory power of AFE over CEPS as a proxy for earnings
expectations. Results of the regression of excess returns onAFEandCEPSare presented inTable 3.
The coefficients on CEPS are all positive and significant. More importantly, the coefficients
on AFE are positive in all regimes and significant in all regimes except the Emerging Latin
and Japanese regimes. This indicates that AFE has incremental explanatory power for these six
regimes, supporting the rejection of H2 in these regimes.
The results of H2 substantially conform with expectations. As expected, analyst forecast errors
have incremental explanatory power in six regimes, including all “developed” regimes with the
exception of the Japanese regime. In these six regimes both AFE and CEPS are significant,
indicating that the stock market reacts to both definitions of unexpected earnings. The results for
the Japanese and Emerging Latin regimes reveal a significant reaction to the release of unexpected
earnings information as defined by the CEPS but not as defined by AFE.10
included in the first sample period are used in this analysis for all regimes other than Emerging Latin. Because there was
insufficient data to provide results for the early time period, the constant sample is comprised of only firms with valid
observations in the second time period. The results of this analysis (not reported) are quantitatively similar to the results
reported in this paper.
8Winsoration at 5% is used to control for the large number of outliers in data of this type. Analysis of data windsorized
at the 1% level yields similar but slightly weaker results. However, the overall conclusions of analysis remain unchanged.
9Inclusion of year indicator variables in model results in marginal changes in adjusted R2 but does not impact the
findings of the study.
10Hall et al. (1994) find correlations between return on equity and both price to earnings and price to book ratios to
be lower in Japan than the U.S. In addition, the relationship between annual (and long-term returns) and earnings is
substantially weaker in Japan than the U.S. The authors conclude that investors in Japan pay less attention to earnings
than U.S.Ali and Hwang (2000) find a similarly weak relationship between earnings and returns for Japanese companies.
Pooled cross-sectional descriptive statistics for dependent and independent variables
Regime Mean Median S.D. First quartile Third quartile
North America (n = 22531)
Excess return 0.0032−0.0017 0.0088 −0.0563 0.0596
CEPS 0.0018 0.0063 0.0443−0.0096 0.0175
AFE−0.0037 0.0000 0.0167 −0.0038 0.0023
Other Anglo-Saxon (n = 13604)
Excess return 0.0102 0.0065 0.0680−0.0370 0.0554
CEPS 0.0009 0.0059 0.0368−0.0083 0.0162
AFE−0.0021 0.0002 0.0139 −0.0028 0.0031
Nordic (n = 1918)
Excess return−0.0003 −0.0025 0.0655 −0.0463 0.0424
CEPS 0.0065 0.0044 0.0807−0.0201 0.0329
AFE−0.0023 0.0000 0.0246 −0.0075 0.0057
Developed Latin (n = 4163)
Excess return 0.0018 0.0000 0.0656−0.0455 0.0464
CEPS 0.0009 0.0041 0.0554−0.0113 0.0166
AFE−0.0044 −0.0003 0.0197 −0.0068 0.0031
Emerging Latin (n = 420)
Excess return 0.0018 0.0039 0.0748−0.0522 0.0555
CEPS−0.0013 0.0048 0.1031 −0.0200 0.0245
AFE−0.0147 −0.0011 0.0517 −0.0108 0.0048
Germanic (n = 3722)
Excess return 0.0012−0.0027 0.0590 −0.0400 0.0391
CEPS 0.0031 0.0024 0.0451−0.0085 0.0149
AFE−0.0035 0.0000 0.0164 −0.0049 0.0026
Japanese (n = 16,024)
Excess return−0.0002 −0.0055 0.0628 −0.0472 0.0422
CEPS−0.0043 −0.0001 0.0246 −0.0094 0.0050
AFE−0.0016 −0.0002 0.0067 −0.0031 0.0016
Emerging Asian/Near East (n = 3741)
Excess return 0.0001−0.0026 0.0913 −0.0618 0.0580
CEPS−0.0107 0.0046 0.1109 −0.0148 0.0190
AFE−0.0079 −0.0002 0.0330 −0.0074 0.0031
Regimes defined inFig. 1. Excess returns are 1-month excess returns during the month of earnings release. CEPS are
change in annual earnings per share between timeτ and τ −1 deflated by price. AFE are actual I/B/E/S annual earnings
per share less I/B/E/S forecast annual EPS deflated by price.
Multi-period hypotheses examine the earnings–returns relationship using the data pooled over
three periods, the early period (1987–1990) the middle period (1991–1994), and the late period
(1995–1998).11 Results of tests of the first multi-period hypothesis, H1mp, are presented in
Table 4. As shown in Panel A, the regression of excess returns on either AFE or CEPS reveals
11The three time periods are used to examine changes in the earnings–returns relationship over time. They represent
the total sample period broken down into three equal 4 year periods. Ideally, yearly regressions would have been used to
examine this issue. However, due to limited data (especially early in the research periods), results are aggregated across
these time periods.
نوع مطلب :
برچسب ها :
مهندسی صنایع و مدیریت صنعتی
مهندسی صنایع و مدیزیت صنعتی یعنی طلوع صنعت نوین
مدیر وبلاگ : ALI MAHDAVI
کل بازدید :
بازدید امروز :
بازدید دیروز :
بازدید این ماه :
بازدید ماه قبل :
تعداد نویسندگان :
تعداد کل پست ها :
آخرین بازدید :
آخرین بروز رسانی :