Post-earnings announcement drift in crisis
conditions:
Evidence in Europe and the US in the period 2007 – 2012
Gijs C. Verheijke*1
ABSTRACT
Market conditions can potentially alter the relationship between earnings announcements and share prices. In this study we contribute to capital markets research in international markets by studying post-earnings announcement drift in Europe. We investigate the crisis period 2007 – 2012, over which this subject has not yet been studied. Besides comparing the evidence in Europe with the US, we subdivide our European sample to investigate the existence of international differences. Our evidence shows that the post-announcement drift anomaly seems to have disappeared in recent years. We also find preliminary evidence that market conditions affect the way investors treat the information in earnings announcements.
Keywords: earnings surprises, post-earnings announcement drift, international capital markets, financial crises
JEL classification: G01, G14, G15, M41
*
University of Groningen, Faculty of Economics and Business, Department of Finance, written as Master’s thesis for the MSc Business Administration program with specialisation Finance. Student number: 1480642
1
2
Introduction
How does information about earnings affect firm value? This question lies at the core of capital markets research in accounting. In this study we investigate whether the traditionally observed relation between earnings surprises and the share price has changed in the volatile period from 2007 onward. An earnings surprise (ES) entails the extent to which the earnings reported by a firm are surprising, or unexpected. An impressive array of studies have shown that the share price increases (decreases) in response to a positive (negative) earnings surprise. Two papers documenting this relation, one founding study and one fairly recent one, are Ball and Brown (1968) and Chudek et al. (2011)2. Besides the predictable immediate response to earnings, researchers have identified an interesting and puzzling phenomenon first spotted by Ball and Brown (1968). The share price predictably drifts in the same direction as the initial response for a period of about 60 days after earnings announcements. This phenomenon is known as post-earnings announcement drift (PEAD) and is the main focus of this paper. We investigate PEAD in (i) a time period in which it has not yet been studied and (ii), a geography (Europe) where it has not yet thoroughly been studied.
Our results are somewhat surprising and indicate that the crisis may have a significant impact on the way information is transferred from earnings announcements to share prices. PEAD seems to have disappeared in more recent years. This observation is robust across international capital markets and when controlling for firm size and industry.
Surely, it is not surprising that earnings announcements contain relevant information for investors. The value of the firm is equal to the discounted value of all expected future cash flows. Since new earnings provide an update of realised value the fact that the actual earnings numbers come available should affect firm value on its own. Moreover, information about earnings today also holds information about future earnings. The expected future cash flows therefore change as does the current realised value.
2
3 The presence of PEAD implies that the share price movement over a future period can be predicted from the value of a variable (ES) today. This cannot be consolidated with the general view that markets behave in an informationally efficient way, hence the puzzle.
The most recent period in which this subject has been studied is from 1994 to 2009 (Chudek et al., 2011) A mere two years of this period lie within the financial crisis period of the last 5 years. Various factors affecting PEAD have been proposed. Empirical evidence exists that PEAD is larger in smaller firms, due to the less transparent informational environment of smaller firms (Brown and Han, 2000). Another empirically supported driver for PEAD is investor sophistication. Bartov et al. (2000) showed that the drift is smaller for firms where a large percentage of the shares are held by institutional investors.
In this study we investigate whether the relation between ES and the capital market response has changed after the collapse of the housing bubble and the subsequent sovereign debt crisis, and whether this relation differs between surprise magnitudes, geographies and firm sizes. The hypotheses with regard to this research question are that the relation is the same as documented, that is: Some of the price response to earnings surprises is delayed, evidenced by the existence of PEAD, and PEAD is larger in subsets of observations with higher ES magnitudes, in less sophisticated capital markets and in smaller firms.
4 check if the effect is different in the UK and non-UK subsamples of the European data. Since the UK lies between the US and continental Europe in terms of capital market sophistication, PEAD could be different in the UK for the same reasons we expect it to be different in the US. This leads us to expect smaller PEAD in the UK subsample. Within continental Europe, different markets have been differently affected by the crisis. We compare a subsample of more developed and less affected capital markets consisting of The Netherlands, Belgium, Germany, Finland and Sweden, with a subsample of less developed and more affected markets consisting of Portugal, Ireland, Greece, Spain and Italy. We expect to find smaller PEAD in the subsample of Northern European countries3.
In addition to the above we perform various sensitivity analyses. Since the period under investigation contains several bouts of very high market volatility and a succession of rallies and crashes, we sensitise the results to effects from the reporting period. Lastly, since the diversification across different sectors is unequal in our European and US sample, we repeat the analysis of PEAD with samples where this inequality is excluded.
The remainder of this paper is structured in the following way: Section 1 contains a more in depth review of the literature and explanation of the relevant theory. In section 2 we explain the variables and methods used to calculate the results as well as provide a description of the data. Section 3 contains the empirical results of the study and the results of the additional analyses. Lastly, we present our conclusions in section 4.
I. Literature review and theory
Research on the information content of earnings announcements is often grouped under the term capital markets research. The economic theory is straightforward: Since earnings are an important determinant of firm value, an innovation in earnings information would be expected to change the value of the company. In addition, because different earnings today
3
5 also affect potential future earnings, the discounted value of the change in expected future earnings also impacts firm value. It can thus be inferred that there should be a positive relationship between earnings surprises and market value. We start this section with a short overview of the founding literature on this subject. In subsection B we discuss the definition of earnings surprise. More specifically, what part of earnings is surprising and how do we determine this? The third subsection is concerned with the question whether PEAD constitutes a departure from market rationality or if rational explanations are possible. Finally, in subsection D we proceed to discuss the characteristics of individual firms that determine PEAD.
A. Founding studies in capital markets research
The first empirical studies regarding the information content of earnings releases were efforts by Ball and Brown (1968) and Beaver (1968). At that time, the efficient market hypothesis was still quite new (Fama, 1965) and had spurred a flurry of new research. The key assumption in Fama’s efficient market hypothesis is that markets are informationally efficient, meaning that the share price reflects all publically available information. This enabled researchers to determine whether earnings announcements have informational content by observing the change in the share price around the announcement date.
6 information and that markets work in an efficient way in the sense that the price reflects the consensus opinion from all investors.
In the decades following these early studies the relation between earnings and price was analysed further. The association between earnings and share price was confirmed by among others: Brown and Kennelly (1972), Beaver, Clarke, and Wright (1979), Patell and Wolfson (1979, 1984), Beaver, Lambert, and Morse (1980), and Beaver, Lambert, and Ryan (1987). Naturally, it is necessary to find out more about the nature of this relation. A relevant question in this aspect is which firm characteristics determine the strength of the response to earnings announcements, referred to as the earnings response coefficient (ERC).
The earnings response coefficient measures the strength of the relationship between price and earnings. From Kormendi and Lipe (1987), Easton and Zmijewski (1989) and Collins and Kothari (1989) four economic determinants of the earnings response coefficient can be distilled: persistence4, risk, growth and the risk-free rate.
In addition to their proof that earnings contain information, Ball and Brown (1968) made a second interesting discovery; they found preliminary evidence of a post-earnings announcement drift. In other words, it seemed that the abnormal return continued for some time after the initial response to the earnings announcement.
Under efficient market conditions the assimilation of new information by the market is expected to be quick and complete. The consistent presence of PEAD could therefore be a departure from market efficiency.
PEAD is of the same sign as the earnings change, which indicates that the market may underreact to earnings announcements. The drift is most visible in the subsets of firms with the most extreme earnings announcements.
4
7
B. Which part of earnings is surprising?
An important aspect of research on earnings surprises is determining which percentage of an earnings announcement constitutes the surprise. Obviously, an earnings announcement can only change market value to the extent in which it provides new information to the market. The component of earnings that was expected should already be priced in, assuming that at least the weak form of market efficiency holds. A variety of models has been used to develop a good proxy for the market’s expected earnings. A popular method, employed in most of the literature cited in subsection A, was to predict future earnings from past earnings by using the observed time-series property of earnings. More recently, the use of earnings forecasts by analysts has gained popularity.
With the rise of equity research reports, a viable consensus expectation for earnings has come available. The earnings surprise is then easily calculated as ES = Ei - Fi or the
difference between the reported earnings and the consensus earnings forecast.
The usefulness of earnings forecasts as a proxy for expected earnings is debated by academics. Ramnath (2008) provides an overview of the literature concerned with analyst forecasts. Brown and Rozeff (1978) first reported that analyst forecasts significantly outperform time-series earnings forecasts in terms of accuracy.
8
C. Is PEAD really a departure from market efficiency?
Notable papers concerned with PEAD are Bernard and Thomas (1989, 1990), Bartov (1992) and Bartov et al. (2000). The fact that PEAD is predictable gives rise to a puzzling challenge to market efficiency. More specifically: in the months after an earnings announcement the share price predictably moves in the same direction as it did in direct response to the announcement. The fact that information at t=0 predicts the share price movement over the next period opens the door to a riskless profit opportunity. The apparent inefficiency has not yet been explained in a way consistent with rational capital markets, nor has a logically consistent behavioral explanation been found. Schwert (2003) has shown that this anomaly has survived decades of research whilst other anomalies such as the weekend effect have weakened or disappeared over the years.
This is not to say that no effort has been spent in attempting to find a rational explanation for PEAD. Some researchers tried to find explanations in methodological issues. Jacob et al. (1999), show that in some cases the drift may be an artifact of methodological problems. Whilst their results are convincing, they do not account fully for observed PEAD. Their study is concerned with problems with the specification of time-series in earnings. This means that when analyst forecasts are used to determine unexpected earnings, a method not relying on the time-series model, the issues described by Jacob et al. should disappear. However, with analyst forecasts, PEAD is still found by various authors (Chudek et al. (2011), Truong, (2011)).
9
D. Firm characteristics impacting PEAD
More recently, studies have tried to understand why this underreaction to earnings announcements persists. Firm size has been used as a proxy for information transparency. Brown and Han (2000) argue that PEAD is smaller for larger firms due to the more transparent information environment of larger firms. Another empirically supported explanation for PEAD is investor sophistication. Bartov et al. (2000) showed that the drift is smaller for firms where a large percentage of the shares are held by institutional investors.
Two of the most recent efforts analysing PEAD are Chudek et al. (2011) and Truong (2011). Interestingly, Chudek et al. (2011) show, that the relation between ES and PEAD can be exploited by going long in firms with positive earnings surprises whilst simultaneously going short in firms with negative earnings surprises. Following this strategy, Chudek et al. (2011) showed it has been possible to attain a 6% riskless return before transaction costs. It would thus seem that it is possible to profitably trade on earnings surprises.
Ke and Ramalingegowda (2005) indeed showed that institutional investors in the US do attempt to profit from PEAD and in doing so, accelerate the share price adjustment to the correct level. If this continues to happen, and more investors catch on to this arbitrage opportunity, it is likely that the PEAD market inefficiency will, over time, be arbitraged away. Because US capital markets are generally slightly more developed than their European counterparts, it can be argued that PEAD is likely to be smaller in the US due to the activity of arbitrageurs being more pronounced in US markets.
10 to protect their portfolios from firm specific risk. Therefore, high arbitrage risk can decrease the ability of investors to exploit PEAD trading opportunities.
As stated earlier, the PEAD anomaly is not exclusive to US markets. Chudek et al. (2011) provide an overview of studies that found PEAD in international capital markets. They report studies that found PEAD in UK firms (Hew et al., 1996 and Liu et al., 2003), Finnish firms (Booth et al., 1996) and emerging markets (Griffin et al., 2008) whilst Chudek et al. themselves are concerned with PEAD in Canada. Truong (who also collaborated on Chudek’s paper) found PEAD in New Zealand. In addition, we have also found a study confirming the presence of drift in India (Sen, 2009). Besides finding PEAD, Sen also investigates if Indian investors exploit the anomaly. He finds little evidence that they do. Assuming that the Indian capital market is less advanced than the US market, this can be interpreted as support for Bartov’s conjecture that investor sophistication is inversely related to PEAD. Another reasonable explanation might be that market frictions and high idiosyncratic risk in India prevent exploitation of PEAD.
II. Data and methodology
This section details the methodology employed to calculate the results and the data used in this study.
A. Definition of variables
A.1. Unexpected earnings
In accordance with earlier studies (Truong, 2010) we define unexpected earnings as actual earnings less forecast earnings, scaled by the market value 10 days prior to the earnings announcement5. This is in accordance with a methodology study by Christie (1987) where it is shown econometrically that using market value as the deflator/scalar in return
5
11 studies prevents errors in variables problems. In keeping with earlier studies we refer to this metric as standardised unexpected earnings based on analyst forecast (SUEAF). The following equation defines SUEAF for each firm quarter:
,= , ,
, (1)
Ei,y are the actual reported quarterly earnings per share for firm i in announcement y, Fi,y
is the mean consensus forecast of all broker reports released in the period between -30 and -2 days relative to the announcement date. Pi,y is the share price for firm i 10 trading days
prior to the announcement.
It has been noted by Livnat and Mendenhall (2006) that different (higher magnitude) results for PEAD are observed when a time-series model of earnings surprises is used. In addition, such a metric allows us to check the results with a method that does not rely on analyst forecasts, effectively controlling for any bias in the analyst forecasts. The time-series metric, called standardised unexpected earnings (SUE), is defined as:
, =
, ,
, (2)
or announced actual earnings less actual earnings from the same period one year earlier scaled by the share price 10 trading days prior to the announcement. We thus test both SUEAF and SUE to sensitise the results for the possibility that analyst forecasts are biased and/or do not correctly reflect market expectations.
In order to test whether the response to ES depends on the magnitude of the ES, the sample is divided in smaller subsamples, sorted by surprise. In keeping with Bernard and Thomas (1990) and more recent studies based on their work, we sort the SUEAFs and SUEs in deciles based on their rank within each period. To clarify, the lower 10% of all observations for period i receive a decile 1 score. By sorting the observations with this method we control for any seasonal effects in earnings surprises.
12 scores (DS), rather than the absolute values, as the explanatory variable in the regressions to reduce the effect of outliers and to eliminate possible nonlinearities.
A.2. Abnormal return
We are interested in the differential return in response to an ES; in other words, the return in excess of (or below) the expected return. Although this is not an event study, we draw on event study methodology in defining the abnormal return.
Abnormal return (AR) is defined as the excess return on the firm’s stock over the normal (expected) return, or AR = Ri – E[ri]. There are several ways of calculating abnormal return,
all of which are concerned with different definitions of the ‘normal’ return. Since the sample period includes several periods of high market volatility, a market corrected measure for abnormal return is the preferred method for this study.
Following the market-adjusted method outlined in MacKinlay (1997) we assume a zero intercept for the capital market line and a beta of 1 for all firms. This provides for ease of calculation and prevents the introduction of unknown and hard to spot bias in the returns resulting from faulty estimations of beta and the risk free rate (rf). Crisis conditions yield extra importance to the independence of beta and rf since both beta and rf probably changed over the course of the financial crisis. The expected return for security i under this model is thus given by the standard CAPM equation, which due to the assumptions that the risk free rate equals 0 and beta equals 1 for all firms, simplifies to:
[] = (3)
where rm is the return on the index (SP500 or Stoxx 600). The abnormal return on a given day is thus computed as:
, = ,− (4)
where ri,t is the logreturn for firm i on day t and rmt is the index return. To reduce the effect
13 within each decile the 2 largest values are replaced with the 3rd largest value and the 2 smallest values with the 3rd smallest value.
Notable about this definition of AR is the fact that there is inherent autocorrelation in this design. By taking the difference between the return on stock i and the return on the broad index containing i, a high return on stock i causes the return on the index to be higher as well. However, since these indices are so large, the impact of a single stock’s movement is negligible.
A similar concern is the clustering of earnings announcements. For example, quarter 1 earnings are reported in March or April for almost all firms. When firm A and firm B announce large surprises on the same date, the stock response for firm A is reflected in the market return used to determine firm B’s AR. With regard to this issue, we again assume that the size of the indices protects the results from this problem.
A.3. Post earnings announcement drift
Following Truong (2010), PEAD is defined as the 60 day cumulative (buy-and-hold) abnormal return following the earnings announcement. This is calculated as the compounded return on the stock minus the compounded return on the relevant index over the same period. Because we use logarithmic returns, the abnormal return can simply be summed over the PEAD window to obtain the cumulative abnormal return. PEAD is thus calculated as:
14 opportunity may present itself in going long the large positive surprise stocks and shorting the large negative surprise stocks.
To obtain the testing variable, we average the PEADs in each ES decile to calculate the Average Post Earnings Announcement Drift (APEADi) for each decile I. It is worthwhile to point out that the sum of all APEADi is equal to zero because together the APEADs constitute the market return. What we thus do, simply put, is split the APEAD in groups based on SUEAF score which we expect to perform differently from each other in such a way that arbitrage is possible.
B. Statistical tests and hypotheses
(i) Does abnormal return in the 60-day period after the announcement deviate from zero? To determine if PEAD is present, we test this hypothesis for each of the deciles. We perform a parametric t-test as well as a non-parametric alternative test for equal central tendency (Kruskal-Wallis test). We test this hypothesis separately for each decile, both for Europe and the US.
H0: APEADi = 0 H1: APEADi ≠ 0
(ii) Is there a positive relation between the SUEAF magnitude and PEAD? We regress the PEADs on the SUEAF decile scores and expect to find a positive relation between the SUEAF magnitude and PEAD. We estimate the following regression equation:
=∝ +% + & (6)
To determine whether a positive relation is present, we test if the coefficient on the SUEAF DS is larger than zero.
H0: β = 0 H1: β > 0
15 markets. To determine whether there is a difference we test the hypothesis that the sample means are equal.
H0: APEADi EU = APEADi US H1: APEADi EU > APEADi US
(iv) We further analyse PEAD in Europe by estimating the relation between return and SUEAF in Europe with and without the firms that are based in the UK. Since the UK capital market is closer to the US market in terms of sophistication, this could yield results that lie in between those seen in the US and those seen in continental Europe
(v) Because different parts of continental Europe have differing degrees of capital market sophistication and have been differently affected by the crisis, we determine if PEAD differs between a subsample of Northern countries and a subsample of Southern countries. We expect PEAD to be larger in the Southern countries since the capital markets in those countries are typically less sophisticated and also because these countries have been more severely affected by the crisis.
(vi) If firm size is a good proxy for investor sophistication, PEAD should be larger in smaller firms. We estimate a regression of PEAD on the SUEAF decile scores with size as a control variable. We subdivide the sample in quartiles based on size and create dummy variables for sizes 1, 2, 3 and 4. To avoid perfect multicollinearity (the dummy variable trap) the constant is dropped from the regression. This gives the following regression equation:
= %" + %'()1 + %+()2 + %-()3 + %/()4 + & (7)
16
C. Data selection and sample
It is important for this study to have datasets for the US and Europe that are reasonably similar in size and diversification across industries and firm sizes. In figure 1 we show the distribution of firm size in both indices.
Figure 1. The distribution of firm size is almost identical in the S&P500 and Euro Stoxx 600 indices.
For the US we use all constituents of the S&P500 index and for Europe we use the Euro Stoxx 600 index. All data is drawn from Factset6. For all constituents of both indices we obtained daily closing prices and quarterly earnings from 1-January 2007 up to 1-September 2012. This yields 1,635,700 daily closing prices in total7. The broker consensus earnings forecasts are based on brokers released between 30 days prior to the announcement up to 2 days before the announcement to ensure that no stale forecasts are used and that no forecasts are used that are contaminated by information leakage prior to the earnings announcement. Moreover, a minimum of three brokers must be available to yield a consensus. We employ earnings per share as our preferred metric because this is reported in nearly all analyst reports in contrast with operational profit metrics (EBITDA, EBIT). EPS is
6
Share prices and market values from Factset Global, Earnings from Factset Fundamentals and Announcement dates and broker consensus from Factset Estimates.
7 1487 days 0 100 200 300 400 500 1,000 101,000 201,000 301,000 401,000 501,000 601,000 N o . o f fi rm s
Market value (millions of local currency)
Distribution of firm size
17 also more relevant for investors in the shares of a firm since it represents the earnings directly attributable to the shareholders (i.e. including the effects of leverage).
The observations (firm quarters) are selected further based on data availability. To obtain a useful firm quarter for the SUEAF based analyses, reported earnings, analyst forecast and closing prices from -2 to +60 days relative to the announcement must all be available. In the SUE based analyses the criterion that the analyst forecast must be available is replaced by the criterion that previous year’s earnings in the same quarter are available.
A large amount of firm quarters are dropped from the sample because one of these criteria is not met, particularly in Europe. In Europe, the 600 firms in the Stoxx 600 index and the 22 quarters in the sample period translate to 13200 potential firm quarters. More than half of those are lost because no announcement date was available in Factset, or because both SUEAF and SUE are non-calculable due to missing data. The resulting sample size for Europe is 5795 firm quarters for SUEAF. Data availability is much better in the US. Of the 11000 potential firm quarters, we preserve a sizeable sample of 10845 firm quarters for SUEAF.
18
Figure 2. The distribution of firm sizes in the European sample is unaffected by the firm omissions arising from limited data
availability.
Figure 3. A relatively large amount of the omitted firms consists of industrial and financial sector firms. No problematic
sector bias has been introduced in the European sample due to the elimination of missing values.
With respect to sector bias, industrials and financials are among the sectors where most firms were eliminated. There are no sectors where all firms are lost however, so sector diversification is relatively intact. Figure 3 shows the omitted firms in the European sample split out per sector. The amount of industrial and financial sector firms in the sample is reduced the most, although these sectors still remain the largest. There is no reason to expect any problems arising from sector bias within Europe.
0 100 200 300 400 500 1,000 101,000 201,000 301,000 401,000 501,000 601,000 N o . o f fi rm s
Market value (millions €)
Distribution of firm size in European sample after omission of missing data
0 50 100 150 N o . o f fi rm s
Sector distribution in European sample before and after removal of missing values
19 Since we are comparing Europe and the US in this study, the sector distributions should ideally be the same in both samples. Figure 4 provides a comparison of the sectors included in the final samples in Europe and the US. Consumer services, Health Care and Technology are strongly overrepresented in the US sample. Therefore, we conduct a sensitivity analysis where we exclude these three sectors to check if our PEAD results are robust to this difference in sector distribution.
Figure 4. A side-by-side comparison of the sector distribution in the European and US samples. The figure shows that
there are significantly more Consumer Services, Health Care and Technology firms in the US sample. In all other sectors the samples are highly similar. This result leads us to decide on a sensitivity analysis where we control for the sector imbalance between Europe and the US.
The last bias that could result from the diminished European sample is country bias. The useable sample is quite different from the full Stoxx 600 in terms of country distribution, in a potentially material way. Whilst significantly different from the total index’ distribution, there is no reason to doubt the quality of international diversification in Europe. There is also still a sufficient amount of data in Southern European as well as Northern European countries to make a viable comparison between these two groups of firms. The international distribution within the European sample before and after missing value removal is shown in figure 5. UK firms account for the bulk of the excluded data. This is unfortunate considering the
0 50 100 N o . o f fi rm s
Sector distribution of included firms in Europe and the US
20 importance of the UK as the most developed European capital market. In our analysis of sector bias it was shown that mostly industrials and financials are omitted from the sample. Since these are large industries in the UK, it is likely that this result is driven by the fact that so many UK based firms are omitted.
Figure 5. The largest number of omitted firms is based in the UK. The second largest number of omissions is in French
firms. Despite the large amount of omissions in these countries, the resulting sample is still diversified across a broad range of countries. The large number of UK firms that dropped from the sample is also what likely drives the earlier finding that primarily industrial and financial sector firms are among the excluded firms.
The power of the comparison of PEAD in the UK versus continental Europe suffers as a result. However, aside from the significantly reduced amount of UK firms, the sample is still strongly internationally diversified. In conclusion, the poor data availability in Europe does not carry adverse consequences for this study, except for the fact that less weight can be attached to our conclusions concerning the UK subsample than would have been the case had we not needed to exclude so many UK firms.
Because earnings are more often close to expectation, a relatively high amount of observations lies close to zero. This gives rise to high kurtosis that requires me to reject the assumption that the data is normally distributed. This gives the parametric testing methods employed far less power and necessitates the use of a non-parametric check for robustness. Descriptive statistics for both samples can be found in Appendix A.
0 50 100 150 200 N o . o f fi rm s
Country distribution in European sample before and after removal of missing values
21
III. Empirical results
In this section we present the results of the main tests. Figures 6 and 7 show the share price response to earnings announcements for SUEAF deciles 1,2,6,9 and 10 in Europe and the US respectively8.
These figures clearly show the positive relation between earnings surprise and abnormal return. The lower deciles seem to underperform whilst the higher deciles outperform the index. Moreover, the response to earnings surprises seems to be more pronounced in the US compared with Europe. This cannot be accounted for by the magnitude of the SUEAF. The spread in SUEAF is fairly equal between the US and European samples.
Figure 6. The share price response to SUEAF announcements in Europe. SUEAF 1 to 10 refer to SUEAF deciles. It can be
seen that the market reacts to SUEAF announcements as predicted. Positive abnormal return follows announcements in the higher deciles whilst negative abnormal return is seen in the lower deciles.
8
We show only the highest and lowest 2 deciles plus 1 middle decile for the sake of readability of the graph. (3.0%) (2.0%) (1.0%) – 1.0% 2.0% 3.0% -2 8 18 28 38 48 58
Response to earnings - Europe
22
Figure 7. The share price response to SUEAF announcements in the US. As in figure 6 the numbers refer to the decile
numbers with SUEAF 1 representing the decile with the smallest (most negative) SUEAF. Looking at the figure, it seems that the response to earnings surprises is more pronounced in the US compared with Europe.
A. The presence of PEAD
In Table I, we present the results of the t-tests of APEAD based on SUEAF deciles.
Table I. t-statistics calculated based on SUEAF deciles. APEAD is only found in the lowest decile in Europe. In the US, APEAD
is significant and of the expected sign in both the lowest and highest decile
Europe*** US***
SUEAF Decile APEAD*** APEAD***
1 -1.971*** -2.032*** 2 -0.688*** -1.783*** 3 -0.470*** 0.094*** 4 0.334*** -1.294*** 5 -0.642*** -1.353*** 6 0.058*** 1.404*** 7 -0.898*** 1.162*** 8 -0.585*** 2.643*** 9 -0.598*** 0.361*** 10 0.792*** 1.957***
* = significant at 0.1 level, ** = significant at 0.5 level, *** = significant at 0.01 level
In Europe, we only find evidence for PEAD in the lower most decile. In the US, PEAD is present in more deciles, with larger t-values in the higher and lower deciles, which indicates the documented PEAD anomaly is present in the US data. However, the evidence in the US is not particularly strong either. No PEAD is found in decile 9 whilst significant PEAD is present in the three middle deciles. Nonetheless, the fact that PEAD is found in the US but not in Europe is contrary to our expectation that PEAD would be smaller in the US.
(3.0%) (2.0%) (1.0%) – 1.0% 2.0% 3.0% -2 8 18 28 38 48 58 Response to earnings US
23 Since the returns data is not normally distributed, we also perform non-parametric Kruskal-Wallis tests. With this method the results are the same as with the parametric method. We only find evidence for PEAD in the US. See appendix B for the results of the non-parametric tests.
In table II we present the results with the deciles based on SUE. This measure of earnings surprise does not rely on analyst forecasts.
Table II. t-statistics calculated based on SUE deciles. The evidence for PEAD in the US that we found with SUEAF deciles is
not present now that the deciles are based on SUE.
Europe*** US***
SUE Decile APEAD*** APEAD***
1 -1.381*** -1.743*** 2 -0.918*** -1.106*** 3 0.201*** 0.684*** 4 1.300*** -0.807*** 5 -0.066*** 0.206*** 6 -2.945*** 2.260*** 7 -1.481*** 1.407*** 8 -0.641*** 0.440*** 9 -0.734*** 1.216*** 10 0.697*** -0.872***
* = significant at 0.1 level, ** = significant at 0.5 level, *** = significant at 0.01 level
The PEAD we found in the US with the SUEAF method is no longer present with the SUE method. We do however observe highly significant PEAD in decile 6 in both Europe and the US, but in opposite directions. This could result from the fact that SUE contains no reflection of overall market conditions, whereas analysts do include this in their forecasts. In volatile markets this may result in errors in the classification of the surprises.
The analyst forecasts do not seem to be severely biased and yield results consistent with theory and previous research. The SUE deciles on the other hand, yield results that are hard to reconcile with theory. Therefore we are comfortable with using only SUEAF in the regression analyses.
B. Relation between SUEAF and APEAD
24
Table III. Regression of APEAD on SUEAF deciles. The highly significant positive coefficient in the US signifies a positive
relationship between SUEAF and APEAD. In other words, A larger earnings surprise gives rise to larger PEAD. No significant relation is found in Europe. The R-squared values are rather small, showing that the model is not a very close fit to the data. There is no evidence for auto-correlation since the Durbin-Watson statistics are close to 2.
Europe US***
Coefficient 0.011 0.021***
Adj. R squared 0.000 0.002***
Durbin-Watson 1.793 2.031***
Schwarz criterion -0.979 -1.079***
* = significant at 0.1 level, ** = significant at 0.5 level, *** = significant at 0.01 level
In table III the coefficients on DS and adjusted R-squared values are shown. The coefficients are positive in both Europe and the US but significant only in the US. Furthermore, the low R-squared values indicate that the model is not a terribly good fit to the data. PEAD is thus positively related to the magnitude of the earnings surprise in the US, although this relation is not very strong.
C. Difference between Europe and the US
In table IV the results of the two sample t-test of APEAD in Europe and the US are presented. A positive value means the dependent variable is smaller in the US compared with Europe.
Table IV. US and Europe comparison of APEAD. A positive value means APEAD is smaller in the US. APEAD does not seem
to be materially different in Europe compared with the US. In as far as there is a difference, the values are larger in the US.
SUEAF Decile APEAD*
1 -0.423* 2 -0.281* 3 0.173* 4 0.573* 5 -0.130* 6 -0.390* 7 -1.859* 8 -1.659* 9 -1.785* 10 -0.248*
* = significant at 0.1 level, ** = significant at 0.5 level, *** = significant at 0.01 level
25 in the first place, except in decile 7 in the US. In as far as there is a difference, PEAD is larger in the US which is contrary to our expectation.
D. PEAD differences within Europe
In this subsection we present the results of the additional analysis of PEAD in UK firms and in different European countries.
Table V. APEAD with and without UK firms. There is no significant difference in APEAD between UK firms and the full sample
or non-UK firms and the full sample.
Coefficient APEAD
Total 0.011
Non-UK firms -0.009
UK firms 0.012
Adj. R-squared 0.002
* = significant at 0.1 level, ** = significant at 0.5 level, *** = significant at 0.01 level
There is no significant difference in PEAD between UK firms and non-UK firms. This is surprising because we expected PEAD to be significantly smaller in the UK. This may however be explained by the large amount of UK firms that had to be excluded due to missing data. In a separate regression with only the UK firms we also find no evidence for the presence of PEAD.
The result of our investigation of PEAD in different countries within continental Europe is presented in tables VI and VII. We first use a country dummy variable to split the European data into ‘Northern countries’, ‘Southern countries’ and ‘Other’. The Northern countries subsample consists of The Netherlands, Belgium, Germany, Finland and Sweden whereas the Southern countries subsample is made up of Portugal, Ireland, Greece, Spain and Italy. Whilst we find no evidence for PEAD in the full sample, the coefficient on PEAD is far smaller in the Southern countries.
Table VI. PEAD in Southern- versus Northern Europe; full sample SUEAF deciles with country dummy. PEAD appears to be
significantly smaller in the Southern countries, although no evidence for PEAD is found in the full sample.
Coefficient APEAD***
Total 0.012***
Northern countries -0.002***
Southern countries -0.029***
Adj. R-squared 0.005***
26 To increase the fidelity of our analysis we separate the two country groups and redefine the SUEAF decile scores based on the new groups. The results are inconclusive. Whilst we find significant PEAD in the lowest and highest deciles in Southern Europe, the drift is negative in both deciles whereas we expect negative drift in the lowest decile and positive drift in the highest.
Table VII. PEAD in Southern- versus Northern Europe; new deciles per group. Significant drift is found in Southern Europe, but
the share price drifts downwards in all instances where PEAD is present. This cannot be interpreted as evidence for PEAD in Southern Europe since the pattern of PEAD does not correspond with the underreaction hypothesis.
Coefficient / Decile Northern countries** Southern countries***
Regression coefficient 0.002** 0.013*** Adj. R-squared 0.000** 0.000*** Decile 1 -0.276** -2.944*** Decile 2 -0.148** -1.331*** Decile 3 2.244** -1.675*** Decile 4 -0.491** -0.712*** Decile 5 0.795** -1.474*** Decile 6 1.872** -0.965*** Decile 7 -0.706** -0.939*** Decile 8 0.453** -1.821*** Decile 9 0.382** -1.242*** Decile 10 0.453** -2.345***
* = significant at 0.1 level, ** = significant at 0.5 level, *** = significant at 0.01 level
Interestingly, the normally observed share price jump in the low and high deciles on the day of the announcement prevails in the Northern European subsample whilst this relation appears to have broken down in the subsample consisting of Southern European firms9. It can also be seen that no PEAD is found with the regressions and that the R-squared values are virtually equal to 0. A possible explanation for these results is that the high volatility during crisis conditions hides the underlying relationship, or put simply; there may be competing information releases with more profound implications for firm value than a surprise in earnings.
E. Additional tests
In this section we present the results of three additional analyses. Because three sectors are highly overrepresented in the US, we analyse how our results change when we exclude
9
27 these sectors. In addition, we control for firm size and check if PEAD is different across the reporting periods.
Table VIII. PEAD in Europe and US with and without the sectors where the US is overrepresented. The results show that
PEAD in the US is significantly smaller when the sectors Consumer services, Healthcare and Technology are excluded. The fact that PEAD is found in the US but not Europe may thus be attributed to sector bias
Coefficient Europe** US***
Total 0.010** 0.021***
Excluded sectors only -0.001** -0.005***
Excluding overrepresented sectors -0.009** -0.012***
Adj. R-squared 0.001** 0.002***
* = significant at 0.1 level, ** = significant at 0.5 level, *** = significant at 0.01 level
In table VII we present the results of our analysis with and without the three sectors that are overrepresented in the US; consumer services, health care and technology. We see that the coefficient on PEAD is far smaller when the sectors are excluded, especially in the US, which means that the difference in PEAD between Europe and the US may be caused by sector bias.
Firm size has been used in literature as a theorised proxy for information transparancy. In this subsection we present the results of the regressions of PEAD with firm size as a control variable.
Table IX. Regression of PEAD on SUEAF controlled for firm size. PEAD is found more strongly in the US, but in the cross
section containing only the smallest firms, we also find strong evidence for PEAD in Europe.
Coefficient Europe*** US*** Total 0.010*** 0.019*** SIZE 1 -0.026*** -0.029*** SIZE 2 -0.001*** -0.014*** SIZE 3 -0.001*** -0.008*** SIZE 4 -0.002*** -0.002*** Adj. R-squared 0.009*** 0.014***
* = significant at 0.1 level, ** = significant at 0.5 level, *** = significant at 0.01 level
28 evidence for PEAD with the PEAD values in decile 4 highly significant and no significant PEAD in decile 10.
When estimating the regressions with a dummy variable to separate the reporting periods, we make an interesting discovery. PEAD is significantly different for several reporting quarters, and many of these quarters are the same in Europe and the US. The best way to visualise this result is with a graph.
Figure 8. The coefficients on the Period dummy variables in a regression of PEAD on SUEAF decile scores. The
coefficients in the US and Europe track each other and exhibit particular variability at the height of the financial crisis. This provides evidence that the market’s response to earnings announcements is dependent on overall market conditions.
The coefficients on the period dummy variable approximately track each other in Europe and the US. This indicates the presence of a period effect which can be connected with market conditions. PEAD is most different from the total in quarters 2, 3 and 4 of 2008, which was of course the period around the fall of Lehman Brothers when the market fell sharply. It is therefore possible that PEAD is overshadowed by the effects of macro events on stock prices and/or investor behaviour.
-0.15 -0.1 -0.05 0 0.05 0.1
PEAD Period coefficients in Europe and the US
29
IV. Conclusion
In this study we investigate post earnings announcement drift in crisis conditions in Europe and the US. Our conclusion is that the PEAD found in many earlier studies is not present in the current crisis period. This is true for the US market as well as the European market and several subsamples of the EU market. The evidence for differences in PEAD between Europe and the US is inconclusive. We cannot conclude that a significant difference is present over the investigated period, and to the extent that the we observe a non-significant difference, this seems to indicate that PEAD is larger in the US, rather than smaller. In addition, we find preliminary evidence that PEAD changes over time, dependent on market conditions.
These results are robust to different testing methods and sensitivity analyses controlling for firm size and sector. As we have used both analyst forecasts as well as a method independent of these, the results cannot be attributed to any bias in the forecasts. With regard to our methodology, we have replicated earlier studies in the definition and calculation of the variables, thereby making it unlikely that the results arise from econometric problems in the calculation of the results. The most likely conclusion is therefore that the relation between earnings surprises and the share price response has been different in the last 5.5 years compared with the years before. In any case, what we have definitely proved here is that an arbitrage strategy based on earnings surprises would not have yielded any positive alpha if employed over the last 5 years in the markets investigated.
A. Discussion and avenues for further research
I propose two theories that could explain our results.
1. Enough investors have finally wised up to the riskless profit opportunity that PEAD promises and have moved to arbitrage the market inefficiency away.
30 investors (are trying to) exploit it, one could convincingly argue that this was bound to happen at some point. The avenues for further research in this direction are simple: even more markets where PEAD has earlier been found can be retested over the 2007 – 2012 period. Another interesting idea is to look at the change in PEAD over time in a particular market; for example in the US for rolling 3 years periods over the last 21 years.
With regard to the second theory: it is possible that the underlying results are hidden from sight by market volatility. This could for example occur if many financial institutions post large negative earnings surprises in the same quarter, causing them all to end up in one of the lower deciles. Next month, the FED or ECB makes an announcement that has a strong positive effect on financial institution share prices. The resulting average PEAD for this decile would thus be far smaller than it would have been under more stable market conditions. One way to check whether this is driving the results is to test for PEAD within separate industries. An observation in support of this theory is that we found indications that (some) PEAD is present in the US market but none at all in the European markets. Since Europe has been the focus of continuing financial turmoil in the last two years, lower PEAD in Europe would make sense if depth and length of financial crisis drives the disappearance of PEAD.
31
V. References
Ball R., Brown P., 1968, An Empirical Evaluation of Accounting Income Numbers. Journal of
Accounting Research 6, 159–178.
Bamber, L.S., Barron, O.E., Stevens, D.E., 2011. Trading Volume Around Earnings Announcements and Other Financial Reports: Theory, Research Design, Empirical Evidence, and Directions for Future Research. Contemporary Accounting Research 28, Issue 2, 431–471.
Bartov, E., 1992. Patterns in unexpected earnings as an explanation for post-announcement drift. The Accounting Review 67, 610–622.
Bartov, E., Radhakrishnan, S., Krinsky, I., 2000. Investor sophistication and patterns in stock returns after earnings announcements. The Accounting Review 75, 43–63.
Beaver, W., 1968. The information content of annual earnings announcements. Journal of
Accounting Research Supplement 6, 67–92.
Beaver, W., Lambert, R., Ryan, S., 1987. The information content of security prices: a second look. Journal of Accounting and Economics 9, 139–157.
Beaver, W.; Clarke, R.; and Wright, W. 1979. The association between unsystematic security returns and the magnitude of earnings forecast errors. Journal of Accounting
Research 17, no. 2:316-40.
Beaver, W.; Lambert, R.; and Morse, D. 1980. The information content of security prices.
Journal of Accounting and Economics 2, no. 1:3-28.
Bernard, V., Thomas, J., 1989. Post-earnings-announcement drift: delayed price response or risk premium? Journal of Accounting Research 27, 1–48.
Bernard, V., Thomas, J., 1990. Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting and Economics 13, 305–3 Booth, G.G., Kallunki, J., Martikainen, T., 1996. Post-announcement drift and income smoothing: Finnish evidence. Journal of Business Finance and Accounting 23, 1197–1211. Brown, L., Rozeff, M., 1978. The superiority of analyst forecasts as measures of
expectations: evidence from earnings. Journal of Finance 33, 1–16.
Brown, L.D., Han, J.C.Y., 2000. Do stock prices fully reflect the implications of current earnings for future earnings for AR1 firms. Journal of Accounting Research 38, 149–164. Brown, P., and Kennelly, J.W. 1972. The informational content of quarterly earnings: An extension and some further evidence. Journal of Business 45, no. 3:403-15.
Christie, A., 1987. On cross-sectional analysis in accounting research. Journal of Accounting
and Economics 9, 231–258.
32 Fama, E., 1965. The behavior of stock market prices. Journal of Business 38, 34–105. Griffin, M.J., Nicholas, H.H., Patrick, J.K., 2008. Why does the reaction to news
announcements vary across countries? Working paper, University of Texas at Austin. Hew, D.L., Skerratt, L., Strong, N., Walker, M., 1996. Post earnings-announcement drift: some preliminary evidence for the UK. Accounting and Business Research 26, 283–293. Ke, B., Ramalingegowda, S., 2005. Do institutional investors exploit the post-earnings announcement drift? Journal of Accounting and Economics 39, 25–53.
Kormendi, R., Lipe, R., 1987. Earnings innovations, earnings persistence and stock returns.
Journal of Business 60, 323–345.
Liu, W., Strong, N., Xu, X., 2003. Post-earnings-announcement drift in the UK. European
Financial Management 9, 89–116.
Livnat, J., Mendenhall, R.R., 2006. Comparing the post-earnings announcement drift for surprises calculated from analyst and time series forecasts. Journal of Accounting Research 44, 177–205.
MacKinlay, A.C. 1997. Event Studies in Economics and Finance. Journal of Economic
Literature 35, 1.
Miller, M., Rock, K., 1985. Dividend policy under asymmetric information. Journal of Finance 40,1031–1051.
Ramnath, S., Rock S., and Shane, P.B. 2008. Financial Analysts' Forecasts and Stock Recommendations: A Review of the Research, Foundations and Trends in Finance 2, 311-421.
Schwert, W., 2003. Anomalies and market efficiency. Handbook of the Economics of
Finance 1, 939–974.
Truong, C., 2010. Post earnings announcement drift and the roles of drift-enhanced factors in New Zealand. Pacific-Basin Finance Journal 18, 139–157.
33
Appendix A
AI. Decile averages Europe
Europe SUEAF SUE PEAD
Decile SUEAF SUE
1 (11.4%) (8.6%) (1.6%) (1.1%) 2 (2.1%) (2.1%) (0.5%) (0.7%) 3 (1.1%) (1.1%) (0.3%) 0.1% 4 (0.6%) (0.5%) 0.2% 0.8% 5 (0.2%) (0.2%) (0.3%) (0.0%) 6 0.1% (0.0%) 0.0% (1.7%) 7 0.3% 0.5% (0.5%) (0.7%) 8 0.6% 1.2% (0.3%) (0.4%) 9 1.2% 2.3% (0.3%) (0.5%) 10 4.5% 11.2% 0.5% 0.5%
AII. Descriptive statistics based on SUEAF deciles
Europe PEAD
Decile N Avg Sigma Max Min
1 573 (1.6%) 19.2% 69.8% (82.5%) 2 560 (0.5%) 16.3% 78.5% (76.2%) 3 551 (0.3%) 13.6% 48.8% (82.5%) 4 586 0.2% 14.3% 60.1% (83.3%) 5 675 (0.3%) 12.1% 55.1% (45.2%) 6 602 0.0% 14.2% 57.6% (71.0%) 7 571 (0.5%) 13.8% 58.3% (70.5%) 8 559 (0.3%) 13.9% 70.6% (60.4%) 9 559 (0.3%) 13.7% 53.1% (76.9%) 10 561 0.5% 16.2% 45.4% (82.4%) total 5797 (0.3%) 14.8% 78.5% (83.3%)
AIII. Descriptive statistics based on SUE deciles
Europe PEAD
Decile N Avg Sigma Max Min
34
AIV. Decile averages US
US SUEAF SUE PEAD
Decile SUEAF SUE
1 (12.5%) (5.5%) (1.2%) (1.1%) 2 (1.3%) (0.9%) (0.8%) (0.6%) 3 (0.6%) (0.4%) 0.0% 0.3% 4 (0.3%) (0.1%) (0.5%) (0.3%) 5 (0.1%) 0.1% (0.5%) 0.1% 6 0.0% 0.2% 0.5% 0.9% 7 0.2% 0.4% 0.4% 0.6% 8 0.4% 0.7% 1.0% 0.2% 9 0.7% 1.4% 0.2% 0.5% 10 2.7% 12.8% 0.9% (0.5%)
AV. Descriptive statistics based on SUEAF deciles
US PEAD
Decile N Avg Sigma Max Min 1 1080 (1.2%) 19.9% 82.6% (108.5%) 2 1068 (0.8%) 14.3% 79.0% (78.1%) 3 1067 0.0% 13.4% 81.9% (91.4%) 4 1069 (0.5%) 13.1% 79.7% (75.8%) 5 1071 (0.5%) 12.0% 54.7% (82.5%) 6 1062 0.5% 12.0% 50.0% (70.7%) 7 1068 0.4% 12.4% 61.4% (63.3%) 8 1066 1.0% 12.3% 63.8% (53.5%) 9 1068 0.2% 13.8% 65.2% (60.2%) 10 1226 0.9% 15.6% 55.8% (98.4%) Total 10845 0.0% 14.1% 82.6% (108.5%)
AVI. Descriptive statistics based on SUE deciles
US PEAD
Decile N Avg Sigma Max Min
35
Appendix B
BI. Non-parametric test results on CAAR and APEAD
Europe US*** PEAD PEAD*** Kruskal-Wallis 11.220 30.913*** SUEAF Decile 1 -0.074 -0.071*** 2 -0.011 -0.057*** 3 0.009 -0.005*** 4 0.029 -0.048*** 5 -0.019 -0.050*** 6 0.033 0.035*** 7 -0.024 0.028*** 8 -0.011 0.068*** 9 -0.014 0.013*** 10 0.084 0.077***
