The decrease of accrual anomaly : evidence from the European market

(1)

The Decrease of Accrual Anomaly

Evidence from the European market

by

Ihsan Akbarali

Student number: 10800271

Master’s thesis submitted in support of the degree of Master of Science in Accountancy and Control

Track: Accountancy

University Of Amsterdam Faculty of Economics and Business

Thesis supervisor: dr. S.W. Bissessur Word count: 13,287

(2)

Statement of Originality

This document is written by student Ihsan Akbarali who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Abstract

This empirical thesis investigates whether there has been a decrease of accrual hedge return on the European market. Data from eight European countries is gathered for the fiscal years 2002-2016. It is shown that the accrual component of earnings is less persistent than the cash flow component. With applying the Mishkin test, the behavior of investors is examined. It is shown that investors underweight the accrual and cash flow component, which is not in line with prior literature. Furthermore, it is shown that taking a long position in firms with relatively low accruals and a short position in firms with relatively high accruals leads to negative abnormal stock return. In addition, results show that the annual accrual hedge return has increased throughout the years. However, this result should be interpreted with caution, as the return is still negative on average. This study contributes to the literature as being an extension of prior international evidence gathered on accrual anomaly.

(3)

1 Introduction

Accrual anomaly, which finds its origin in Sloan (1996), had a considerable impact on the literature as it invalidates the efficient market hypothesis. It is shown that investors are earnings fixated, which means that they are unaware that the underlying components of earnings are not equally persistent. The underlying components of earnings are accruals and cash flows, and it is shown that accruals are less persistent than cash flows. Consequently, firms with relatively low accruals are underpriced and firms with relatively high accruals are overpriced. This mispricing of shares is called accrual anomaly in the literature (Battalio, Lerman & Mendenhall, 2012).

Furthermore, it is shown that investors can achieve returns on this mispricing through hedging. These so-called accrual hedge returns can be achieved by ranking the firms on the magnitude of the accrual component of earnings and assigning them in equal numbers to ten portfolios. In the first portfolio the firms with relative low accruals are included, while in the tenth portfolio the firms with relative high accruals are included. In this hedged portfolio, a long position will be taken in the lowest portfolio and a short position will be taken in the highest portfolio. The possibility of arbitrage invalidates the efficient market hypothesis. Therefore, to retain the efficient market hypothesis, various studies give other reasons for accrual anomaly. An example of this is Khan (2008), who tries to explain it through a risk-based model.

Further research, Bradshaw, Richardson & Sloan (2001) show supporting evidence for Sloan (1996). They show a persistence of accrual hedge return, which is also in line with the findings of Lev & Nissim (2006). However, recent research of Green, Hand & Soliman (2011) show that accrual hedge return has decreased. Shi & Zhang (2012) and Leippold & Lohre (2012) also find this. Green et al. (2011) show evidence that this decrease is due to the exploitation of hedge funds and the implementation of SOx. However, this decrease in accrual hedge return has only been shown on the financial markets in the United States. Some international research has been conducted (including the European market) about accrual anomaly, however a decrease in accrual hedge return has not yet been shown. The purpose of this thesis is to further investigate accrual hedge return on the European market. This leads to the following research question:

(5)

5 Providing an answer to this research question is of essence, as there is a gap in the literature. Most studies regarding accrual anomaly have been conducted using United States data, so not much awareness has been raised for accrual hedge returns on the European market. As a consequence, it could well be that accrual hedge return has not been exploited in the European market by hedge funds. Furthermore, Green et al. (2011) have shown that the decrease of accrual hedge return is also attributable to the implementation of SOx. However, this was only implemented for firms listed in the United States and it could be that it had less impact on the European market, as not all those firms are listed in the United States.

This thesis contributes to the current literature by extending prior international evidence about accrual anomaly. First, it is shown that there is statistical evidence to verify that the accrual component of earnings is less persistent than the cash flow component in the European market. Furthermore, I find that market efficiency has been rejected as an underweighting for accruals and cash flows was found by applying the Mishkin test. After that, the average accrual hedge return is examined and the results show a return of -2.82 percent. Lastly, I examine whether the average accrual hedge return has decreased by demarcating the sample into two subperiods. The results show that that the return has, contrary to the expectations, increased over the past years. However, caution must be paid when interpreting this result, as the average annual hedge returns are still negative in both periods. The increase in return was due to a lower negative accrual hedge return and not a positive accrual hedge return.

The remainder of this thesis is as follows. First, the relevant literature regarding the accrual anomaly is described in the following chapter. Based on this literature, the hypotheses of this thesis are formulated. Thereafter, in chapter three the research design is described. It is explained how the data is gathered and the measurement of the financial statement variables is described. Furthermore, the empirical models that are used to test the hypotheses are also described. After that, the findings of this study are described in chapter four. Lastly, in chapter five the conclusion and limitations of this study are stated.

(6)

6

2 Prior literature and hypothesis development

This chapter commences with describing the term accruals. After that, the origin of accrual anomaly is examined. Thereafter, alternative explanations and further research about accrual anomaly are examined. By combining the literature around accrual anomaly, the hypotheses are presented in the last paragraph of this chapter.

2.1 Defining accruals

According to Dechow & Dichev (2002), accruals account for revenues and expenses that are not yet received or paid. Typical accrual accounts are accounts receivable and accounts payable. Accruals are necessary to implement the matching principle, with this principle it is required to match revenues with the expenses incurred to earn those revenues, so they are reported simultaneously. Besides the matching principle, accruals provide a solution to the timing problem as it timely recognizes the added economic value (Dechow, 1994 and Dechow & Dichev, 2002). An illustration: a service is provided in 2017, but the payment is due in 2018. If there were no accrual accounting, which means that only cash flows would be relevant, then the service would have been recognized in 2018. This leads to a timing problem, as the service was provided in 2017. Accrual accounting solves this timing problem by recognizing the service in 2017, through creating a balance sheet item called accounts receivables.

However, besides the benefits of accruals, they also have some issues as they are subject to managerial expectations. Due to this, accruals are much used for earnings management, which is called accruals-based earnings management (McNichols & Wilson, 1988). With earnings management, the management manipulates the earnings to enable earnings smoothing. With this, the management shows stable profits to the shareholders, which have a positive effect on share price. Another reason for earnings management could be to achieve a target of compensation contracts for managers. By achieving these targets, the managers receive bonuses. So there is an incentive to manipulate the earnings if otherwise the target will not be achieved (Hadani, Goranova & Khan, 2011).

Accruals are split into two components in the literature, discretionary and non-discretionary accruals. Discretionary accruals, also known as abnormal accruals, are accruals where the management has discretion. Due to this discretion, there is a chance of earnings management. On the other hand, there are the non-discretionary accruals where the

(7)

7 management has no discretion over. Earnings management is not possible with non-discretionary accruals, these are also known as normal accruals (Subramanyam, 1996 and Healy, 1985).

2.2 Accrual anomaly

Accrual anomaly is first shown by the research of Sloan (1996), in which the underlying components of earnings are examined. These underlying components of earnings are accruals and cash flows. It is shown that accruals are less persistent in future period earnings than cash flows and investors do not react to this fact. The fact that investors do not react on the weaker underlying component of earnings, which are accruals, is called earnings fixation. This causes mispricing of shares, more precisely; it causes an underpricing of firms with low accruals and an overpricing of firms with high accruals. This mispricing is known as accrual anomaly.

However, Sloan (1996) was not the first research to state that accruals are less persistent than cash flows. This can be traced back to the research of Graham & Dodd (1934), where it was shown through illustrative cases that accruals are less persistent than cash flows. This idea is further examined by Graham, Dodd & Cottle (1962), where the term earnings power is defined. This is calculated by subtracting some accruals from the earnings, which implies that accruals are not consistent. Furthermore, Bernstein (1993) also argues that accruals are more subject to distortion than cash flows. Consequently, analysts prefer to relate cash flows from operations (CFO) to earnings, in order to determine the quality of the earnings. This also implies that accruals are less persistent than cash flows.

Ball & Brown (1968) was the first study to document a positive association between stock returns and earnings. This association exists, for earnings have an ability to summarize value-relevant information. Earnings are value relevant as they provide an indication of the management’s performance, which decreases the information asymmetry between the management and investors. However, further research presents evidence that the value-relevant information, which earnings provide, are not entirely used by investors in forecasting future earnings (Ou & Penman, 1989; Bernard & Thomas, 1990; Hand, 1990; Maines & Hand, 1996). Sloan (1996) indeed shows that the investors use not all value-relevant information, as they are unaware of the weaker persistence of accruals.

Sloan (1996) shows that, through a hedge portfolio, a return of 10.40 percent can be achieved on accrual anomaly. To build this portfolio, firms are ranked on the magnitude of

(8)

8 the accrual components of earnings and assigned in equal numbers to ten portfolios. In the first portfolio the firms with relative low accruals are included, while in the tenth portfolio the firms with relative high accruals are included. In this hedged portfolio, a long position will be taken in the lowest portfolio and a short position will be taken in the highest portfolio. The return of 10.40 percent is composed of 4.90 percent for the long position and 5.50 percent for the short position. The return of 10.40 percent drops to 4.80 percent in the second year and 2.90 percent in the third year. So accrual anomaly proves that the efficient market hypothesis is not true, as share prices do not fully reflect all public information. Still, it does not necessarily mean that there are unexploited profit opportunities, as there are acquisition costs and processing costs associated with building hedge portfolios. As a result of investing in different types of shares, the acquisition costs are not negligible. Furthermore, the processing costs are also not negligible considering that portfolios must be built for a substantial amount of firms, whereby the accruals must be measured for every firm.

2.3 Alternative explanations

The idea that one can earn returns by analyzing accruals invalidates the efficient market hypothesis. As a consequence, the research of Sloan (1996) lead to a spark in the academic literature. Many of the researchers searched for alternative explanations for accrual anomaly, as the efficient market hypothesis is an entrenched paradigm. One of the researchers is Khan (2008), who tries to explain accrual anomaly through risk-based models. Within these models, shares with higher (lower) returns must be more (less) risky. To explain accrual anomaly through risk-based models, investors ought to find shares with low accruals more (less) risky, which will explain the (overpricing) underpricing.

Sloan (1996) already ruled out two risk-based models, the Capital Asset Pricing Model (CAPM) and the Fama-French three-factor model. The CAPM is often used for valuing shares, whereby it lays a link between systematic risk (beta) and the expected return of a share. Systematic risk is the market risk, which is unpredictable and impossible to avoid entirely. Examples of these are interest rate changes and inflation. The Fama-French three-factor model is an extension of CAPM, where also the three-factors ‘size’ and ‘value’ explain the expected return of a share (Fama & French, 1993). Sloan (1996) has shown for both of these risk-based models that they do not explain accrual anomaly. Furthermore, also Bhojraj & Swaminathan (2009) show that the Fama-French three-factor model does not explain accrual anomaly.

(9)

9 Coming back to Khan (2008), he builds a four-factor model, which is inspired by the Fama-French three-factor model. In this model, there are four factors that explain the expected return of a share: news about future expected dividends on the market portfolio, news about future expected returns on the market portfolio, size and value. The research shows that accrual anomaly is explained through the four-factor model by showing that it rationally priced economic and financial distress characteristics that are correlated with accruals. However, according to Dechow, Khimich & Sloan (2011), these findings suffer from shortcomings. First of all, Khan (2008) assigns firms to five portfolios instead of ten portfolios, which mechanically has an effect on accrual anomaly. Secondly, the four-factor model does not directly measure accruals. It just claims to explain economic and financial characteristics that are linked to accruals and as a consequence to accrual anomaly. There are also some other alternative explanations, but the majority of them have been refuted like the risk-based model theory. This thesis will not further examine alternative explanations, as it is not the focus of this thesis.

2.4 Further research

2.4.1 Persistence accrual hedge returns

A lot of literature reinforces the findings of Sloan (1996). Bradshaw et al. (2001) provide evidence that there is earnings fixation, through analyzing analyst’s and auditor’s understanding of accruals. The results show that analysts are largely unaware of the lower persistence of accruals and therefore they incorrectly inform investors. Teoh & Wong (2002) and Barth & Hutton (2004) also show that analysts are earnings fixated. Bradshaw et al. (2001) also show that accountants are unaware regarding the lower persistence of accruals. Accrual anomaly states that firms with high accruals have overstated their earnings, so accountants who are aware of this should consider this in his audit report. However, no evidence of this is found, which leads to the conclusion that accountants are also earnings fixated. Therefore, if analysts and auditors are earnings fixated, it can be implicitly concluded that individual investors are also.

Lev & Nissim (2006) explicitly show that individual investors are earnings fixated and are unable to earn accrual hedge returns, as there are high acquisition and processing costs. To achieve a positive accrual hedge return, investment in many shares is necessary, which leads to high acquisition costs. In addition, the investors must measure accruals for different companies and sort them in portfolio, thus leading to high processing costs.

(10)

10 Mashruwala, Rajgopal and Shevlin (2006) reinforce the findings of Lev & Nissim (2006), as they show that taking a short position is costly and risky for an individual investor.

The previous paragraph has shown that individual investors are earnings fixated. However, it is shown in the literature that institutional investors are not earnings fixated. According to Easley & O’hara (1987), Walther (1997) and Bartov, Radhakrishnan & Krinsky (2000), institutional investors are more sophisticated and therefore, firms with more institutional ownership have more accurate share prices (so less mispricing). Furthermore, Balsam, Bartov & Marquardt (2002) show that institutional investors detect earnings management, which is measured by accruals, on an earlier stage than individual investors. In addition, Collins, Gong & Hribar (2003) examine a direct relation between institutional ownership and accrual anomaly. Their results show that firms with more institutional ownership have less mispriced accruals and as a consequence more accurate share prices. Therefore, this implies that institutional investors are less earnings fixated than individual investors are. Battalio et al. (2012) also find that small trade investors are more earnings fixated than large trade investors, which are according to them usually institutional investors. The previous paragraph shows that institutional investors are less earnings fixated than individual investors, as they are more sophisticated, which implies that they could earn accrual hedge return. However, according to Lev & Nissim (2006), most institutional investors do not trade on accrual anomaly as they avoid firms with extreme-accruals. Those firms have volatile shares, which are not desirable to institutional investors. This section has shown that the early research on accrual anomaly shows a persistence in accrual hedge return, as investors do not trade on it. However, further research shows a decrease in accrual hedge returns, as will be discussed in the next paragraph.

2.4.2 Decrease accrual hedge returns

Pincus, Rajgopal & Venkatachalam (2007) show evidence that accrual hedge returns have decreased. It is shown that an accrual hedge return of 8.40 percent can be earned for the United States, which is a lower return than the original 10.40 percent of Sloan (1996). In addition, Leippold & Lohre (2012) show a decrease in the accrual hedge returns. They show an accrual hedge return of 7.92 percent, which is also lower than the original 10.40 percent. Pincus et al. (2007) use as their sample the period 1994-2002, whilst Leippold & Lohre (2012) use 1994-2008. So, Leippold & Lohre (2012) use a six year longer period and find

(11)

11 accrual hedge returns even lower than Pincus et al. (2007) find. This could also be an indication that accrual hedge returns are continually decreasing.

Furthermore, the research of Shi & Zhang (2012) shows an accrual hedge return of 5.93 percent, which also implies a decrease. Also, Dechow et al. (2011) and Richardson, Tuna & Wysocki (2010) show a decrease in accrual hedge returns. Green et al. (2011) even show evidence of a decay of accrual hedge returns, as they are on average no longer reliably positive. According to them, the decrease in accrual hedge return finds its root in two factors: an increase in the capital invested by hedge funds in exploiting accrual anomaly and a decline in the size of mispricing.

Hedge funds existed at least from the 1940s but did not trade much until the 2000s (Stulz, 2007). This emergence of trade caused an exploitation of accrual hedge returns. Hedge funds are able to earn a return on accrual anomaly, unlike institutional and small investors, as they are unregulated, sophisticated, and can short sell securities at low cost. Furthermore, a decline in the size of mispricing, which was due to lower accruals-based earnings management, caused a decrease in accrual hedge returns. Bhojraj, Sengupta & Zhang (2009) and Cohen, Dey & Lys (2008) show that accruals-based earnings management has decreased due to the implementation of Sarbanes-Oxley (SOx), which means that accruals have become more persistent. The SOx law was implemented with the purpose to decrease accruals-based earnings management, by setting new or expanded requirements for United States public firms. Intuitively, more persistent accruals mean a decline in the size of mispricing as the effect of earnings fixation is less. Keskek et al. (2011) show that indeed, the implementation of SOx lead to a decline in the size of mispricing.

2.4.3 International evidence

Most of the research regarding accrual anomaly has been conducted in the United States, but some international evidence is also found. Pincus et al. (2007) is an example, they show positive accrual hedge return in 17 out of 20 countries, of which 11 are significant results. The most robust results were found for the United States, Australia, Canada and the United Kingdom. In addition, Leippold & Lohre (2012) find positive accrual hedge returns in 22 out of 26 countries, whereof 12 are significant results. LaFond (2005), who shows accrual anomaly in 15 out of 17 countries, also conducts international research. Furthermore, Clinch, Fuller, Govendir & Wells (2012) show accrual anomaly in Australia.

(12)

12

2.5 Hypothesis development

As there has been a lot of research conducted on accrual anomaly in the United States, the question arises whether the accrual hedge return also decreased in the European market. Pincus et al. (2007) and Leippold & Lohre (2007) have already shown that accrual anomaly exists on the European market. However, different countries and fiscal years are used in those researches. Therefore, the existence of accrual anomaly has to be confirmed again, before a decrease in accrual hedge return can be examined. Consequently, I hypothesize:

H1: The hedge strategy whereby a long position is taken in the stock of firms

reporting relatively low levels of accruals and a short position is taken in the stock of firms reporting relatively high levels of accruals leads to positive abnormal stock return on the European market.

After confirming an accrual hedge return, there will be examined whether there has been a decrease in the return. This is interesting to examine, as it is possible that there has not been a decrease in accrual hedge returns on the European market. Since most research is conducted in the United States, there has not been raised much awareness for accrual hedge returns on the European market. Therefore, it could well be that accrual hedge return has not been arbitraged in the European market by hedge funds. Furthermore, it is shown that the decrease in accrual hedge returns is also due to the implementation of SOx. This caused a decrease in the size of mispricing, as accruals became more persistent. However, SOx is only applicable for firms listed in the United States and it could have had a lesser impact on the European market, as not all of these firms are listed in the United States. Consequently, I hypothesize:

(13)

13

3 Research design

In this chapter, the research design will be discussed. First, it is described which databases are used to retrieve the required data. Thereafter, the process for attaining the final sample is explained. After this, the variables and empirical models are described, by which the two hypotheses will be tested.

3.1 Data and sample

Initially, the empirical analysis is conducted using all firms with available data over the fiscal years 2002-2016 on WRDS for eight countries: Austria, Belgium, Germany, Spain, Finland, France, Italy and the Netherlands. The initial idea was to investigate 12 European countries for the years 2002-2016, as these 12 countries are the first to introduce the euro in the year 2002. These countries are Austria, Belgium, Germany, Spain, Finland, France, Italy, the Netherlands, Greece, Ireland, Luxembourg and Portugal. Unfortunately, it was not possible to obtain data about the stock prices for Greece and Portugal from Compustat Global – Index Prices and therefore those are left out of the sample. Lastly, Ireland and Luxembourg are left out of the sample, as it was not possible to retrieve MSCI indexes from the database, so the final sample consists of eight countries.

In order to have consistent data, countries that joined the European Union at a later time, for example Slovenia in 2007, are left out. In addition, Denmark, United Kingdom and Sweden, who were in the European Union in 2002, are left out to have consistent data, as they did not implement the euro at that time. Computstat Global - Fundamentals annuals will be used to retrieve data about the accruals, cash flows and earnings. Compustat Global – Security Prices will be used to retrieve data about the stock prices inclusive of dividends. Following prior literature (Pincus et al., 2007) MSCI indexes will be employed to determine the market return, which will be retrieved from Compustat Global – Index prices. The way in which the variables are measured will be explained in the following paragraph.

The initial sample consists of 3,901 firms and 39,967 firm-years, as can be observed in Table 1. Financial firms are left out, SIC codes 6000-6999, as their nature of accruals are different (lack significant levels of inventory). This leads to an elimination of 623 firms and 8,069 firm-years. Furthermore, firms with missing annual statement data are removed. So, firms with missing data about total assets, current assets, cash, debt, current liabilities, income tax payable equity and operating income after depreciation are eliminated.

(14)

14 Consequently, it results in the elimination of the whole fiscal year 2002, as the financial statements of the preceding year are required in order to calculate the mutation of assets and average total assets. This results in the elimination of 349 firms and 4,583 firm-years.

The sample, so data about the financial statements, is then merged with Compustat Global – Daily Prices, which results in some missing data. These missing data are eliminated, among which the whole fiscal year 2016, as the stock price tests requires at least one year of future returns data. This results in the elimination of 471 firms and 9,150 firm-years, as can be observed in Table 1 and therefore the final sample consists of 2,458 firms and 18,165 firm-years.

Table 1 Sample selection

Description Firms Firm-years

Firms on Compustat Global - Annual 2002-2016 3,901 39,967

Less financial institutions 623 8,069

Less missing annual statement data 349 4,583

Less: no match with Compustat Global - Daily Prices 471 9,150

Final sample 2,458 18,165

3.2 Variables and empirical models

Following Sloan (1996), Pincus et al. (2007) and Leippold & Lohre (2002) there will be three financial variables used for this empirical analysis, which are earnings, accruals and cash flows. Earnings will be measured by the operating income after depreciation. This excludes non-recurring items, such as extraordinary items, discontinued operations, special items and non-operating income. These items are problematic, as Computstat does not provide enough information to split them into accruals and cash flows. Accruals will be measured with the definition of Sloan (1996):

(15)

15 where:

∆CA = change in current assets

∆Cash = change in cash/cash equivalents ∆CL = change in current liabilities

∆STD = change in debt included in current liabilities ∆TP = change in income taxes payable

Dep = depreciation and amortization expense.

Some clarifications about the measure for accruals: as debt included in current liabilities is related to financial transactions, it is subtracted from accruals. In addition, income taxes payable is excluded for consistency with the definition of earnings employed in the tests. The cash flow component is calculated by subtracting accruals from the earnings. It can also be defined as the net cash flow of operating activities, but to ensure these results are comparable with Sloan (1996), Pincus et al. (2007) and Leippold & Lohre (2012), this definition is not used. These three components are then standardized, for comparison reasons, by dividing them with the average total assets. The average total assets are measured as the average of the beginning and end-of-fiscal-year book value of total assets. Hence, the following measures are used:

Earnings = Income from continuing operations Average total assets

Accrual component = Accruals

Average total assets

Cash flow component = Income from continuing operations – accruals Average total assets

The variables for earnings, accruals and cash flows have been winsorized at the 1 st and 99th percentile values, as there were some outliers. In order to test the first hypothesis, there is a need to determine whether accruals are less persistent than cash flows in the European market. Prior literature has shown that accounting rates of return are mean reverting and earnings are here defined as an accounting rate of return. This is because it has been defined

(16)

16 as operating income scaled by total assets, so it is the return on assets. The relation between current earnings performance and future earnings performance can be stated as (Sloan 1996):

Earningst+1 = γ0 + γ1 Earningst+ υt+1 (1)

Prior research shows that useful information can be obtained by splitting the earnings into an accrual component and a cash flow component. To test the persistence of the underlying components of earnings, model (1) can be extended:

Earningst+1 = γ0 + γ1 Accrualst + γ2 Cash flowst + υt+1 (2)

If in model (2) γ1 is less than γ2, then the accruals are less persistent than cash flows. After

determining the difference in predictive power of the accrual and cash flow component, with regards to future earnings, there is a need to verify earnings fixation of investors. If there is earnings fixation, then this will be reflected in future stock returns. This will be shown, following prior literature using the Mishkin test1, which jointly estimates a linear forecast model together with a rational market pricing model. In other words, it compares the weights placed on the variable tested, which is in this case the earnings. If these placed weights differ, then the assumption that the market is efficient, will be rejected. The following model will be jointly estimated (Sloan, 1996, Mishkin, 1983):

Earningst+1 = γ0 + γ1 Earningst+ υt+1 (1)

Abnormal returnst+1 = β (Earningst+1 – γ0*– γ1*Earnings) + εt+1 (3)

The abnormal returns are determined by subtracting the market return, as obtained from MSCI, from the stock returns. No correction is made for the risk-free rate, as the effect has been shown to be minimum in prior literature (Pincus et al., 2007). Following prior research, stock returns are measured from four months after the fiscal year until one year later. By that time, it is assumed that all publicly available information should be incorporated in the share prices. The variable abnormal returns has been trimmed at the 5th and 95th percentile values,

1_{This test is conducted by using the ADO file of Judson A. Caskey, which is retrieved from:} https://sites.google.com/site/judsoncaskey/data

(17)

17 so unlike the other variables it is not winsorized at the 1st and 99th percentile values, as there were a lot of extreme abnormal return.

Model (1) is known as the forecasting equation and model (3) as the returns equation). The β in model (3) is the valuation multiplier, also known as earnings response coefficient, which represents the reaction on unexpected earnings. These unexpected earnings are calculated in the parentheses. Model (3) implies that abnormal returns should be zero in an efficient market, as all public information is already included in share prices. In this case, the investors correctly anticipate the average persistence of earnings performance.

If investors correctly anticipate the difference in the persistence of accruals and cash flows this should also already be embedded in share prices in case there is no earnings fixation (i.e. γ= γ*). In order to determine this model (3) will be extended and the model will look as follow:

Earningst+1 = γ0 + γ1 Accrualst + γ2 Cash flowst + υt+1 (2)

Abnormal returnst+1 = β (Earningst+1 – γ0*– γ1*Accrualst – γ2* Cash flowst) + εt+1 (4)

Following Sloan (1996), the combined models are estimated using iterative weighted non-linear least squares. Market efficiency is determined by using a likelihood ratio statistic, which is distributed asymptotically Chi-square (q):

2n log (SSRc_{+ SSR}u₎ ₍₄₎

where:

q = the number of constraints imposed by market efficiency

n = the number of observations

SSRc = the sum of squared residuals from the constrained weighted system SSRu = the sum of squared residuals from the unconstrained weighted system

Kraft, Leone & Wasley (2007) and Lewellen (2010) have criticized the use of Mishkin test. In their view, this model leads to an omitted variable problem, so they support the use of Ordinary Least Squares (OLS). However, according to Dechow et al. (2011), the use of OLS leads to a lack of direct estimates and associated standard errors for γ1* and γ2*. Furthermore, Pincus et al. (2007) and Leippold & Lohre (2012), support the use of Mishkin test. Therefore, this test is also used for this thesis.

(18)

18 After determining the earnings fixation of investors, the first hypothesis can now be directly tested. In order to determine the accrual hedge return, the firm-years will be assigned yearly in equal numbers to ten portfolios, where relative low accruals are placed in group one and relative high accruals are placed in group ten. Then a long position will be taken in the first portfolio and a short position in the tenth portfolio. Subsequently, the returns of the first and tenth portfolio will be determined. The second hypothesis will be tested in the same way as the first hypothesis. However, the full sample will now be demarcated into two subperiods, in order to examine a decrease in accrual hedge return. The first subperiod will be 2003-2009 and the second subperiod 2010-2015. After the year 2009, the peak of the financial crisis had passed and therefore it is decided to start the second subperiod in 2010. It could be the case that accruals have become more persistent after the financial crisis, as the crisis was partly caused by accruals. Therefore, it could be that auditors, analyst, investors and the management of the firms themselves pay more care to the accruals, which could lead to increased persistence. In paragraph 4.3, a robustness check is conducted to determine whether the persistence of accruals has indeed changed.

(19)

19

4 Results

In this chapter, the results will be discussed. First, the descriptive statistics and correlations are discussed. Thereafter, the regressions for the earnings persistence are tested. After discussing the regression results, a Mishkin test is carried out to determine whether investors are earnings fixated. In addition, the average accrual hedge returns are established. Lastly, two robustness tests are performed.

4.1 Descriptive statistics and correlations

Before providing the empirical results, some statistics on various firm-year characteristics are described. These characteristics are earnings, cash flows, accruals and abnormal returns. Table 2 shows the means for each of the preceding variable by country for the fiscal years 2002-2016. On average, the earnings for all countries are positive, 0.04, which is in this case the return on assets. Cash flows are on average 0.08 and accruals -0.04. The abnormal returns are on average also negative, as it states an average return of -0.02, which is -2 percent. These results are in line with the findings of Sloan (1996) and Pincus et al. (2007), as they also found similar values for these variables.

Table 2

Means of various firm-year characteristics across countries

Country N Earnings Cash flows Accruals Abnormal return

All countries 18,165 0.04 0.08 -0.04 -0.02 Austria 617 0.05 0.09 -0.04 0.03 Belgium 846 0.06 0.11 -0.05 -0.02 Germany 5,508 0.04 0.08 -0.04 -0.02 Spain 1,144 0.05 0.09 -0.04 -0.03 Finland 1,146 0.06 0.10 -0.04 0.01 France 5,462 0.04 0.07 -0.03 -0.02 Italy 2,235 0.03 0.08 -0.04 -0.02 The Netherlands 1,207 0.05 0.10 -0.05 -0.05

Notes: Earnings is income from continuing operations divided by average total assets. Accruals is the change in non-cash current assets, less the change in current liabilities (exclusive of short-term debt and taxes payable), less depreciation expense, all divided by average total assets. Cash flows is the difference between earnings and accruals, which are both divided by average total assets. Abnormal returns are computed by taking the stock return, inclusive of dividends and subtracting it by the market return, as calculated by MSCI. The stock returns are measured begins four months after the fiscal year-end in which the financial variables are measured. N are the number of observations. All numbers are rounded up to second decimal place.

(20)

20 When paying attention to the country-specific results in Table 2, it is observed that the earnings are the highest in Belgium and Finland and the lowest in Italy. Cash flows are the highest for Belgium, while the lowest are from France. The mean for accruals is negative across all countries, with the Netherlands having the most negative accruals and France having the least negative accruals. As can be seen from Table 2, six out of the eight countries provide on average negative abnormal returns. These countries are Belgium, Germany, Spain, Finland, France, Italy and the Netherlands. Two out of the eight countries provide positive abnormal returns, which are Austria and Finland.

The correlation statistics between earnings, accruals and cash flows are provided in Table 3. It is observed from Table 3 that accruals and cash flows have an apparent negative relation as the pooled sample of all countries provides a correlation of -0.55 with a one percent significant level. This also applies to every individual country, for example: Austria provides the highest negative correlation of -0.73 and the Netherlands the lowest negative correlation of -0.44 with both having a one percent significance level. The correlation between earnings and cash flows is 0.71 for the pooled sample of all countries. When the individual countries are examined, it is found that the highest correlation of 0.78 is from the Netherlands and the lowest correlation of 0.54 is from Austria.

Table 3

Correlation statistics between earnings, accruals and cash flows

Country Accruals/Cash flows Earnings/Cash flows Earnings/Accruals

All countries -0.55*** 0.71*** 0.16*** Austria -0.73*** 0.54*** 0.17*** Belgium -0.54*** 0.76*** 0.10*** Germany -0.55*** 0.69*** 0.19*** Spain -0.65*** 0.67*** 0.11*** Finland -0.52*** 0.77*** 0.12*** France -0.56*** 0.74*** 0.13*** Italy -0.60*** 0.61*** 0.24*** The Netherlands -0.44*** 0.78*** 0.18***

Notes: Earnings is income from continuing operations divided by average total assets. Accruals is the change in non-cash current assets, less the change in current liabilities (exclusive of short-term debt and taxes payable), less depreciation expense, all divided by average total assets. Cash flows is the difference between earnings and accruals. All numbers are rounded up to second decimal place.

Sample consists of 18,165 firm year observations for the fiscal years 2002-2016 *** indicate significance at the 1% level

** indicate significance at the 5% level * indicate significance at the 10% level

(21)

21 Lastly, the correlation between earnings and accruals is described in Table 3. It is found that this correlation is 0.16 for the pooled sample, wherein all countries are examined. When the country-specific correlations are examined for earnings and accruals, it is found that Italy provides the highest correlation of 0.24 and Belgium the lowest of 0.10. These findings are also confirm the findings of Sloan (1996) and Leippold & Lohre (2012), as they found similar relationships between earnings, accruals and cash flows. Now that it is established that the various firm-years characteristics still have approximately the same values/implications as prior research, the empirical results will be described.

4.2 Empirical results

Following the descriptive statistics and correlations, the results of the regressions will be explained. Table 4 shows the results from the regression of current earnings performance with respect to one-year-ahead earnings, also known as future earnings performance, which is model (1) as explained in paragraph 3.2. As observed in Table 4, the relationship between earnings and future earnings has a coefficient of 0.802. The p-value is significant, as it indicates significance at one percent level. When controlling for fixed effects of the individual countries, a coefficient of 0.800 is found at a 1 percent significance level. So, confirming to prior literature (Pincus et al., 2007) it is shown that earnings are mean reverting, with an average persistence of 0.802, through applying model (1).

Table 4

Results from regressions of future earnings performance on current earnings performance Earningst+1 = γ0 + γ1 Earningst+ υt+1 Earnings 0.802*** 0.800*** (0.012) (0.011) Constant 0.006*** 0.006*** (0.001) (0.001) Observations 18,165 18,165 R-squared 0.640 0.641

Fixed effect for country No Yes

Notes: Earnings are income from continuing operations divided by average total assets. Robust standard errors clustered by firm are displayed in parentheses. All numbers are rounded up to third decimal place.

Sample consists of 18,165 firm-year observations for the fiscal years 2002-2016 *** indicate significance at the 1% level

(22)

22 Table 5 provides coefficients for model (2), where the persistence of the accrual and cash flow component is tested. It is shown that the coefficient on the accrual component of earnings is 0.721, while the coefficient on the cash component of earnings is 0.774. In other words, an increase of 1 percent for the accruals will lead to an increase of 0.721 percent in future earnings, whilst an increase of 1 percent for the cash flows will lead to an increase of 0.774 percent in future earnings. Both the coefficients are extremely significant, as they indicate significance at the 1% level. An F-test confirms that both coefficients are indeed unequal (F=22.18). As can be observed in Table 5, controlling for fixed effects of the individual countries does not lead to other implications as the coefficient are similar. Hence, these results indeed confirm that the accrual component of earnings is still less persistence than the cash flow component of earnings, as has been shown in prior literature (Shi & Zhang, 2012).

Table 5

Results from regressions of future earnings performance on the components of current earnings performance

Earningst+1 = γ0 + γ1 Accrualst + γ2 Cash flowst + υt+1

Accruals 0.721*** 0.720*** (0.015) 0.015 Cash flows 0.774*** 0.773*** (0.012) (0.012) Constant 0.006*** 0.006*** (0.001) (0.002) Observations 18,165 18,165 R-squared 0.631 0.631

Fixed effect for country No Yes

Notes: Earnings are income from continuing operations divided by average total assets. Accruals are the change in non-cash current assets, less the change in current liabilities (exclusive of short-term debt and taxes payable), less depreciation expense, all divided by average total assets. Cash flows is the difference between earnings and accruals. Robust standard errors clustered by firm are displayed in parentheses. All numbers are rounded up to third decimal place.

Sample consists of 18,165 firm-year observations for the fiscal years 2002-2016 *** indicate significance at the 1% level

(23)

23 Green et al. (2011) and Keskek et al. (2011) state that a reason for the decrease in accrual hedge returns is the increased persistence of the accrual components, as a consequence of laws and regulations (i.e. SOx). However, in Table 5 it is shown that the accrual component is still less persistence than the cash flow component, so if investors are still unaware of this, which will be reflected in future stock returns, then achieving accrual hedge returns should still be possible. To test this, the Mishkin test will be applied. The results of the Mishkin test are stated in Table 6. As can be seen in Table 6, the coefficient for the forecasting equation, γ1, is the same as the result from Table 4. The coefficient is equal to 0.802, which should be the case. On the other side, some unexpected results for γ1* are found. The coefficient on earnings in the stock price equation is equal to -0.499 and is significant at the 1 percent level. The likelihood ratio statistic is 146.14 with a marginal significance level of 0.00, which results in a rejection of the market efficiency. This result is surprising, as it has not yet been found in prior literature.

Table 6

Results from Mishkin test for earnings

Earningst+1 = γ0 + γ1 Earningst+ υt+1

Abnormal returnst+1 = β (Earningst+1 – γ0*– γ1*Earnings) + εt+1

Parameter Estimate Asymptotic standard error

γ1 0.802^^^ 0.0282

γ1* -0.499^^^ 0.1075

Test of market efficiency: γ1=γ1* Likelihood ratio statistic: 146.14 Marginal significance level: 0.00

Notes: Earnings are the income from continuing operations divided by average total assets.Accruals are the change in non-cash current assets, less the change in current liabilities (exclusive of short-term debt and taxes payable), less depreciation expense, all divided by average total assets. Cash flows are the difference between earnings and accruals. Abnormal returns are computed by taking the stock return, inclusive of dividends subtracting it by the market return, as calculated by MSCI. The stock returns are measured begins four months after the fiscal year-end in which the financial variables are measured. All numbers are rounded up to third decimal place. Instead of asterisks, ^ is used to indicate the significance level in order to ensure that there is no confusion with the asterisks used in the Mishkin test.

Sample consists of 18,165 firm-year observations for the fiscal years 2002-2016 ^^^ indicate significance at the 1% level

^^ indicate significance at the 5% level ^ indicate significance at the 10% level

(24)

24 The Mishkin test is also applied to the individual components of earnings, which are accruals and cash flows. These results are described in Table 7. The market efficiency is tested by determining whether γ1=γ1* and γ2=γ2*, as in an efficient market the difference in persistence for accruals and cash flows should be reflected in stock prices. As can be observed, the coefficient for γ1 and γ2 are the same as in Table 5, which should be the case. As expected, the coefficients for γ1* and γ2* are negative, as the γ1* parameter in Table 6 was also negative. Both the parameters, γ1* and γ2*, have a coefficient of -0.334 and are significant at the one percent level. The likelihood ratio statistic is 170.66 with a marginal significance level of 0.00, which again results in a rejection of the market efficiency, so stock prices do not correctly anticipate of the persistence of accruals and cash flows. It can be concluded from the stated results, that investors underprice the accrual component, as well as the cash flow component.

As the results of the Mishkin test are contrary to prior literature, additional testing is performed to ensure no extreme observations drive this result. Following Sloan (1996) the variables are placed into ten deciles and subsequently the earnings persistence regression and Mishkin test are performed. However, this does not lead to other confirmations as provided above. It is shown that the coefficient, using model (2), of the accrual component (0.536) is still less persistence than the cash flow component (0.840) with a significance at the 1 percent level. The Mishkin test for model (2) and (4) shows negative coefficients for γ1*(-0.213) and γ2*(-0.390) at a likelihood ratio statistic of 388.76 and thus still rejecting the null hypothesis of market efficiency. Furthermore, following Kraft et al. (2007), an OLS regression is also used to test the market efficiency. In this test, the relation between accruals and cash flows with the future earnings is examined. Unreported findings show that the results provide the same implications as the Mishkin test. Therefore, market efficiency is rejected and an underweighting for accruals and cash flows is found. It is unclear what explains these uncommon results for the Mishkin test. As has been described in paragraph 4.1, the statistics for the various firm-year characteristics are according to prior literature, so that cannot be the reason for these results. Since the focus of this thesis is to examine the accrual hedge returns, and not the behavior of investors, no further investigation is conducted for this result. Now that the behavior of investors has been examined, the next paragraph will examine the accrual hedge return.

(25)

25

Table 7

Results from Mishkin test for the components of earnings

Earningst+1 = γ0 + γ1 Accrualst + γ2 Cash flowst + υt+1

Abnormal returnst+1 = β (Earningst+1 – γ0*– γ1*Accrualst – γ2* Cash flowst) + εt+1

Parameter Estimate Asymptotic standard error

γ1 0.721^^^ 0.007

γ1* -0.334^^^ 0.093

γ2 0.774^^^ 0.004

γ2* -0.334^^^ 0.085

Test of market efficiency: γ1=γ1* and γ2=γ2* Likelihood ratio statistic: 170.66

Marginal significance level: 0.00

Notes: Earnings are the income from continuing operations divided by average total assets .Accruals are the change in non-cash current assets, less the change in current liabilities (exclusive of short-term debt and taxes payable), less depreciation expense, all divided by average total assets. Cash flows are the difference between earnings and accruals. Abnormal returns are computed by taking the stock return, inclusive of dividends subtracting it by the market return, as calculated by MSCI. The stock returns are measured begins four months after the fiscal year-end in which the financial variables are measured. All numbers are rounded up to third decimal place. Instead of asterisks, ^ is used to indicate the significance level in order to ensure that there is no confusion with the asterisks used in the Mishkin test.

Sample consists of 18,165 firm-year observations for the fiscal years 2002-2016 ^^^ indicate significance at the 1% level

^^ indicate significance at the 5% level ^ indicate significance at the 10% level

Recall that the first hypothesis investigates whether taking a long position in the stock of firms reporting relatively low levels of accruals and taking a short position in the stock of firms reporting relatively high levels of accruals leads to a positive abnormal stock return on the European market. In order to examine this, the firms are ranked on the magnitude of accruals and placed yearly in ten portfolios. Then the abnormal returns are calculated for the first and the last portfolio, whereby a long position is taken in the first portfolio and a short position in the tenth portfolio. The results for these abnormal returns are provided in Table 8. As can be observed in Table 8, the average annual the hedge return over the full sample, so including all countries, has a return of -2.82 percent. It is explored whether this result is attributable to one extreme year, but this has not been found, as Figure 1 shows evidence of a consistent instability of accrual hedge return. It plots the yearly average annual hedge returns for the first fiscal year until the last, which is 2003 until 2015. As a result, the average of these 12 years is equal to the average accrual hedge return of -2.82%. As can be observed in

(26)

26 Figure 1, the average annual hedge return is negative in seven of the 13 years. The years with positive average annual hedge return are also a lot weaker than the return found by Sloan (1996). For example, the year 2006 gives us an average annual hedge return of 1.30 percent, whilst the return found by Sloan (1996) was 10.40 percent.

It must be noted that this accrual hedge strategy leads to negative returns, which implies that positive returns can be achieved by swapping the position of the trade. Thus, by going long on firms with relative high accruals and short on firms with relative low accruals a hedge return of 2.82 percent can be achieved. This finding has not yet been found in prior literature. An explanation for this result can be derived from the implications made from the Mishkin test. Recall that the Mishkin test showed an underweighting of both components of earnings by investors, so the cash flow component as well as the accrual component. This is contrary to findings in prior literature, as usually an overweighting of the accrual component is found. Hence, this could explain this result.

Figure 1 Calendar-time average annual hedge returns

Notes: This figure plots the yearly hedge portfolio returns, where a long position is taken each year in the lowest portfolio and a short position is taken in the highest portfolio. These deciles are made by sorting firms equally into ten portfolios based on their relative amount of accruals. The vertical line after 2009 demarcate two subperiods within the overall 2002-2016 period: 1) the fiscal years 2003-2009; 2) the fiscal years 2010-2015. The countries in this sample are Austria, Belgium, Germany, Spain, Finland, France, Italy and the Netherlands. All numbers are rounded up to second decimal place.

Sample consists of 18,165 firm-year observations for the fiscal years 2002-2016

2.15% -8.55% -6.69% 1.30% 6.53% -0.67% 0.75% -9.81% -1.17% -2.86% -6.50% -1.27% 3.76% -12.00% -10.00% -8.00% -6.00% -4.00% -2.00% 0.00% 2.00% 4.00% 6.00% 8.00% 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 H ed ge re tu rn Fiscal year

(27)

27 Further research of the data explores whether these results could be attributable to a specific country, which would then be seen as an outlier and therefore left out. However, as can be seen in Table 8, on average six out of the eight countries have a negative average annual hedge return. These countries are Belgium, Germany, Finland, France, Italy and the Netherlands. Within these countries, Belgium has the highest negative return of 10.30 percent and France the lowest of 2.25 percent. Two countries that have a positive average annual hedge return are Austria and Spain, but they have a small positive average annual hedge return of respectively 2.40 percent and 2.61 percent. Therefore, there is evident proof to reject the first hypothesis and state that on average taking a long position in the stock of firms reporting relatively low levels of accruals and taking a short position in the stock of firms reporting relatively high levels of accruals leads to a negative abnormal stock return on the European market.

These results are not in line with Pincus et al. (2007), as they show positive accrual hedge returns in 17 out of 20 countries. This difference in results could be due to the fiscal years used in the sample. Pincus et al. (2007), use firm year-observations from 1994-2002, whilst this thesis uses the fiscal years 2002-2016. Furthermore, the examined countries are also different, as only four out of the eight countries have an overlap with Pincus et al. (2007). In addition, little confirmation is found with the results of Leippold & Lohre (2012) as they find positive accrual hedge returns in 22 out of 26 countries. This difference in results could also be attributable to the difference in fiscal years, as they examined the fiscal years 1994-2008. Furthermore, like Pincus et al. (2007), only four out of the eight countries have an overlap with this research. Therefore, the provided results are in line with Green et al. (2011), as on average no positive accrual hedge return has been found. Nevertheless, it must be noted that the reason for this result is not attributable to the higher persistence of accruals, as Green et al. (2011) and Keskek et al. (2011) state. Recall that the earnings persistence test found that the accrual component was still less persistent than the cash flow component. So, this result must be attributable to the fact that investors underweight the cash flow and the accrual component of earnings, as was found in the Mishkin test.

It must also be noted that during the sample period, more specifically in the year 2005, IFRS became mandatory for countries within the European Union. As stated before, all countries used in this sample are part of the European Union. This could have had an effect on the accrual anomaly, as IFRS claims to enhance the quality of the financial statements. Consequently, it could be that accruals have become more persistent and as a result the

(28)

28 accrual hedge returns have decreased. However, there seems little reason to presume that the IFRS implementation had effect on accrual hedge return, when observing Figure 1. As can be observed, the returns stay volatile and thus do not show a consistent pattern. Hence, no further investigation is conducted concerning IFRS.

Table 8

Average annual hedge return

Country N Full sample Period 1: 2003-2009 Period 2: 2010-2015 Difference

All countries 18,165 -2.82% -2.65% -3.00% -0.35% Austria 617 2.40% -3.24% 7.85% 11.09% Belgium 846 -10.30% -11.19% -9.45% 1.74% Germany 5,508 -3.78% -2.14% -5.62% -3.48% Spain 1,144 2.61% 1.96% 3.30% 1.34% Finland 1,146 -5.64% -6.34% -4.90% 1.44% France 5,462 -2.25% -3.48% -0.97% 2.51% Italy 2,235 -3.41% -4.87% -1.79% 3.08% The Netherlands 1,207 -2.44% -3.36% -1.24% 2.13%

Notes: This figure plots the average annual hedge portfolio returns, where a long position is taken each year in the lowest portfolio and a short position is taken in the highest portfolio. These portfolios are made by sorting firms equally into ten portfolios on the basis of their accruals. The full sample consists of 18,165 firm-year observations for the fiscal years 2002-2016, which are afterward demarcated into two periods: 1) the fiscal years 2003-2009; 2) the fiscal years 2010-2015. The difference is the average annual hedge return in period 2 minus the average annual hedge return in period 1. The countries in this sample are Austria, Belgium, Germany, Spain, Finland, France, Italy and the Netherlands. All numbers are rounded up to second decimal place.

Recall that the second hypothesis investigates whether accrual hedge returns have decreased in the European market. To determine the preceding hypothesis, the sample is demarcated into two subperiods: period 1) fiscal years 2003-2009 and period 2) fiscal years 2010-2015. After demarcating the sample, the difference of average annual hedge return between those two subperiods is determined. These results are also stated in Table 8. As can be observed in Table 8, the average annual hedge return for all countries in the first period is -2.65 percent and in the second period -3.00 percent. Thus, the average accrual hedge returns have decreased by 0.35 percent. In order to verify whether this result is representable for all countries, country-specific returns are again examined. Table 8 shows that the average annual hedge return, contrary to the weak decrease of average annual hedge return in all countries, has increased in seven out of the eight countries. These seven countries are Austria, Belgium Spain, Finland, France, Italy and the Netherlands. The highest increase is for Austria, which is an increase of 11.09% and the lowest increase is for Belgium, which is an increase of 1.74%.

(29)

29 As Austria shows an extreme difference in the average annual hedge return between period 1 and 2, the preceding implications are also verified by leaving out Austria, in order to be sure that the results are not driven by one country. This had no impact on the implications, as the reported average annual hedge returns would then be -2.86 percent and for period 1 and 2 this would respectively be -2.75 and -2.99 percent. As stated earlier, it can be observed from Table 8 that Germany is the only country with a negative difference. Therefore, the results are also verified by leaving out Germany. It is then found that the average annual hedge return for all the years is 2.70 percent and for period 1 and 2 this is respectively 4.01 and -1.30 percent. Hence, this results in an increase of the annual average hedge return of 2.71 percent. As a result, the second hypothesis is rejected as there is shown to be an increase of 2.71 percent in the average annual hedge return by leaving out Germany. However, caution must be paid when interpreting this result, as the average annual hedge returns are still negative in both periods. As a matter of fact, the average annual hedge return is only less negative in period 1 than period 2. This leads to the implication that it is not possible to achieve on average positive abnormal returns with the accrual anomaly. This result has also been found in the first hypothesis and are confirm the results of Green et al. (2011).

4.3 Robustness analysis

The use of the balance sheet method, in order to define accruals, has been criticized. In order to ensure that this is a good measure, a sensitivity check is conducted using another definition for accruals. Following Shi & Zhang (2012), accruals will be measured by subtracting the cash flows from operating activities from the earnings. The results show that this change of definition has no impact for the earnings persistence test, as accruals appear still to be less persistent than the cash flow component. A coefficient of 0.832 is found for the cash flows and 0.705 for the accruals, both at a significance level of one percent. It must be noted that the difference between the coefficients for accruals and cash flows is more extreme, compared to the balance sheet method.

Furthermore, the implications for the Mishkin test also stay the same, when using this definition of accruals. Both components of earnings seem to be underweighted and have a negative coefficient, like observed in paragraph 4.2. The coefficient in the returns equation are -0.108 for accruals and -0.477 for cash flows, where accruals are significant at the five percent level and cash flows at the one percent level.

(30)

30 The results for the accrual hedge returns, whilst calculating accruals using the cash flow statement, can be observed in Table 9. As can be observed in Table 9, the average annual hedge return for the full sample, so including all countries, is -5.17 percent. When examining the country-specific data, a negative return is found for all countries. For example, Belgium has a return of -8.43 percent and Germany of -5.90 percent. As can be noticed, these returns are more negative than the results stated in Table 8. This confirms that it is indeed possible to achieve a return by switching the trading position of Sloan (1996), so going long on firms with relative high accruals and short on firms with relative low accruals. Therefore, the first hypothesis is rejected again.

Table 9

Average annual hedge return

Country N Full sample Period 1: 2003-2009 Period 2: 2010-2015 Difference

All countries 17,370 -5.17% -5.16% -5.17% -0.01% Austria 589 -5.07% -5.08% -5.06% 0.02% Belgium 779 -8.43% -22.73% 5.11% 27.85% Germany 5,247 -5.90% -2.65% -9.55% -6.91% Spain 894 -3.13% 3.56% -7.33% -10.89% Finland 1,172 -8.07% -5.38% -10.98% -5.60% France 5,299 -3.57% -5.30% -1.76% 3.54% Italy 2,164 -8.15% -8.91% -7.33% 1.58% The Netherlands 1,212 -6.60% -5.28% -8.27% -2.99%

Notes: This figure plots the average annual hedge portfolio returns, where a long position is taken each year in the lowest portfolio and a short position is taken in the highest portfolio. These portfolios are made by sorting firms equally into ten portfolios on the basis of their accruals. The full sample consists of 18,165 firm-year observations for the fiscal years 2002-2016, which are afterward demarcated into two periods: 1) the fiscal years 2003-2009; 2) the fiscal years 2010-2015. The difference is the average annual hedge return in period 2 minus the average annual hedge return in period 1. The countries in this sample are Austria, Belgium, Germany, Spain, Finland, France, Italy and the Netherlands. All numbers are rounded up to second decimal place.

The results obtained when demarcating the sample, can also be observed in Table 9. As can be observed, the average annual hedge return for all countries in the first period is -5.16 percent and in the second period -5.17 percent, which yields in a difference of -0.01 percent. When examining the country-specific results, it is found that four out of the eight countries have had an increase in their accrual hedge return. These results are somewhat different from the results obtained in paragraph 4.2, nevertheless this sensitivity check does still not show a decrease of the accrual hedge return.

When demarcating the sample, the first subperiod was defined as the fiscal years 2003-2009 and the second subperiod as the fiscal years 2010-2015. In order to support the

(31)

31 demarcation of the two periods, it will be examined whether the persistence of accruals indeed increased after the financial crisis. The results of this test can be observed in Table 10. As can be observed, the coefficient for accruals is 0.699 and for cash flows is 0.745 in period 1. Both coefficients are significant at the one percent level. When period 2 is examined, similar implications are found for both coefficients. The coefficient for accruals is 0.742, whilst the coefficient for cash flows is 0.799. So, accruals have indeed become more persistent after the financial crisis. However, it must be noted that accruals have become more persistent in period 2, but cash flows have had almost an evenly increase in its persistence.

Table 10

Results from regressions of future earnings performance on the components of current earnings performance

Earningst+1 = γ0 + γ1 Accrualst + γ2 Cash flowst + υt+1

Period 1: 2003-2009 Period 2: 2010-2015 Accruals 0.699 *** 0.742*** (-0.020) 0.022 Cash flows 0.745*** 0.799*** -0.016 (0.014) Constant 0.009*** 0.003** (0.001) (0.001)

Notes: Earnings are income from continuing operations divided by average total assets. Accruals are the change in non-cash current assets, less the change in current liabilities (exclusive of short-term debt and taxes payable), less depreciation expense, all divided by average total assets. Cash flows is the difference between earnings and accruals. Robust standard errors clustered by firm are displayed in parentheses. All numbers are rounded up to third decimal place.

Sample consists of 18,165 firm-year observations for the fiscal years 2002-2016, which are afterward demarcated into two periods: 1) the fiscal years 2003-2009; 2) the fiscal years 2010-2015

*** indicate significance at the 1% level ** indicate significance at the 5% level * indicate significance at the 10% level

The decrease of accrual anomaly : evidence from the European market