The real business effects of quarterly reporting

Transcription

1 The real business effects of quarterly reporting Benedikt Link* Oliver Vogler Jürgen Ernstberger Ruhr-University Bochum April Preliminary and incomplete version - * Corresponding author. Friedenheimer Str. 19, D München, Germany, Tel , [email protected]

2 The real business effects of quarterly reporting ABSTRACT: This study examines the real business effects of mandatory quarterly reporting. Specifically, we examine whether a higher mandated interim disclosure frequency has an effect on managers tendency to use real activities manipulation. We find that in years in which a firm just meets past years' earnings or the zero earnings benchmark (suspect years) mandatory quarterly reporters exhibit higher real activities manipulation compared to semiannual reporters. We also find significant cross-sectional differences of this effect depending on the industry and capital market characteristics in which the firm is operating. Institutional differences, however, only play a minor role in determining the size of the effect. We interpret these findings as evidence for managerial myopia as a result of increased pressure to meet or beat earnings benchmarks due to mandated interim disclosure. Our study contributes to the literature on the real effects of disclosure regulation. Keywords: Mandatory disclosure, interim reporting, quarterly reporting, real business effects, real activities manipulation, management myopia Data Availability: Data are available from public sources identified in the paper - 1 -

3 The real business effects of quarterly reporting I) INTRODUCTION This study investigates the real business effects of mandatory quarterly reporting. Specifically, we test whether interim reporting frequency influences management decisions by inducing myopic behavior. We argue that shorter reporting cycles lead managers to deviate from normal operational practices because in their attempt to meeting or beating earnings benchmarks they are tempted to use more earnings management and activities manipulation than they otherwise would. This argument is in line with empirical evidence from a recent survey among executives in the U.S., where 80% of CFOs admit they would decrease discretionary spending (e.g., R&D) given the pressure to meet quarterly earnings targets. Almost 40% would even provide incentives for customers to buy more products in the current period by, e.g., increasing discount levels (Graham et al., 2005). The authors argue that managers are willing to sacrifice long term value in order not to fail meeting short term expectations of analysts and investors. There is also ample anecdotal evidence, which suggests that managers, when required to issue quarterly financial reports, are forced to make short-sighted business decisions to meet earnings targets, often at the expense of long-term value. The most prominent example is probably German car manufacturer Porsche, who s CEO refused to issue quarterly reports in 2003, claiming that it triggers short-sighted management decisions. 1 As indicated by the evidence presented, we are particularly interested in the real effects of interim disclosure regulation. In order to test empirically whether reporting frequency leads to a deviation from normal operational practices, we need to compare quarterly reporters with companies that report semi-annually. The European Union (EU) constitutes a unique regulatory setup to test for these differences between reporting frequency regimes because quarterly reporting is mandatory only in 6 countries and 2 stock market segments 1 This resulted in a delisting of the company from the German stock market index for mid-caps (M-DAX). Other prominent examples of companies returning to semi-annual reporting include BAA (world's leading airport operator) and BHP Billiton (global leader in resource industry)

4 of the EU-15 countries. 2 Each member state and stock market operator in the EU can decide upon the required frequency of financial reporting individually. Hence, quarterly reporting is one of the few regulatory differences that remain, after the EU in general has harmonized accounting standards (IFRS), disclosure and enforcement regulations for publicly listed firms since This makes it easier to isolate the impact of disclosure frequency. Other settings, notably the U.S., do not offer such a variation in the frequency of reporting and/or have greater cross-country differences to control for. In order to measure real business effects we rely on the previously established measures for real activities manipulation (RAM) defined as deviations from normal operational practices for the purpose of avoiding earnings surprises and other adverse effects such as, e.g., loss in reputation. Prior research (Dechow et al., 1998; Roychowdhury, 2006) uses cash flow from operations (ACFO), abnormal production cost (APROD), and abnormal discretionary expenses (ADISC) as measures of RAM. We propose a new consolidated RAM measure, as previous studies have only partially consolidated these individual RAM measures. For our empirical study we hand-collected the regulatory requirements on interim disclosure in the EU-15 countries based on an extensive review of documents from the EU, national regulatory authorities, stock exchanges, and from interviews with financial analysts and stock market operators. Using a sample of 7,372 firm-year observations from EU-15 countries, we apply a propensity score based matching procedure to account for endogeneity and a potential lack of comparability between quarterly and semi-annual reporters. Moreover, we use the suspect year concept introduced in prior research to ensure that we primarily analyze years in which the probability of RAM is high. We also control for determinants of accounting earnings management (AEM) in our regressions to take into account the interaction between RAM and AEM. 2 EU-15 refers to the 15 countries that were part of the EU before its enlargement in We focus on these countries as these are mature economies that have integrated capital markets and harmonized disclosure regulation for several years already. The 12 new member states are mostly transitory economies and therefore less comparable

5 We find that mandatory quarterly reporters generally exhibit higher levels of RAM compared to semi-annual reporters, especially in suspect firm-years. We also document crosssectional differences depending on industry characteristics, such as short operating cycles and high competition, where the effect is most pronounced. While capital market characteristics are also important in determining the size of the effect, i.e., the effect is strongest when shareholders have a short investment horizon and the level of monitoring by analysts is low, institutional characteristics seem to play a minor role. Our findings suggest that mandated increased disclosure in form of interim reporting frequency is associated with indirect costs in terms of real activities manipulation. We conduct extensive additional analyses to test the robustness of our findings. In an additional analysis, we analyze the effect of the EU Transparency Directive (TD) introducing the requirement for semi-annual reporters to disclose additional quarterly financial information. Moreover, we rerun our main regressions using the individual RAM measures established by Roychowdhury (2006). Both tests corroborate our findings. Our study contributes to the literature in several ways. First, the findings contribute to the growing "real effects" literature by presenting evidence on real business implications of regulation. Roychowdhury (2006) shows that when companies are close to zero-earnings, they tend to deviate from normal operational practices by, e.g., price discounts and a reduction in discretionary expenses. Other studies have analyzed the real effects of accounting regulation recently in different contexts (e.g., Cohen et al., 2008; Gunny, 2010; Zang, 2010; Cohen and Zarowin, 2010). For example, Cohen et al. (2008) show that RAM has been used more extensively compared to accounting earnings management after the introduction of Sarbanes- Oxley. Cohen and Zarowin (2010) show that RAM is also used at the time of seasoned equity offerings. We contribute to this literature as the first study to present evidence on the "real" effects of interim reporting frequency. We also add to this literature by introducing a new aggregate RAM measure that consolidates all individual RAM measures and helps detecting RAM in general. Previous studies have only combined two of the three measures (e.g., Cohen - 4 -

6 and Zarowin, 2010). Second, we add to the literature by showing how these real effects depend on external characteristics, such as industry, capital market and institutional characteristics. We provide evidence of cross-sectional differences in the magnitude of disclosure frequency-induced real activities manipulation. Third, we add to the literature of mandatory reporting regulation by providing evidence on potential externalities of mandatory disclosure. The remainder of our paper is organized as follows. Section II discusses the related literature and states the hypotheses. Section III presents the methodology, sample and descriptive statistics. Section IV discusses the results of our empirical tests. Section V presents the results of additional and sensitivity analyses. Section 6 concludes. II) RELATED LITERATURE AND HYPOTHESIS DEVELOPMENT In 2003, the European Commission, the executive body of the European Union also in charge of proposing legislation changes to the European Parliament and the European Council, proposed to adopt a U.S.-like system of mandatory quarterly reporting for all listed companies in Europe in order to increase transparency towards investors and analysts. This triggered a discussion on the benefits and costs of increased disclosure frequency. Despite largely undisputed benefits of quarterly reporting (such as higher transparency and a reduction in asymmetric information), several countries (e.g., UK, Netherlands, Denmark and Austria) in Europe opposed the proposal. One of the main arguments was that mandated quarterly disclosure could induce myopic management behavior. The proposal was finally amended and today companies are not mandated to provide full quarterly reports by the EU. However, several national regulators and stock market operators do require quarterly financial reporting. This debate raises the question whether shorter reporting cycles influence real business decisions. We contribute to the discussion by investigating potential deviations from normal operational practices, i.e., a higher tendency towards real activities manipulation (RAM), as a reaction to an increase in mandatory reporting frequency

7 Economic effects of mandatory disclosure regulation The objective of disclosure regulation is to mitigate information asymmetries and the adverse selection problem between suppliers and users of capital and thus facilitate efficient capital allocation (e.g., Watts and Zimmerman, 1986). For instance, major benefits of disclosure include higher liquidity, lower cost of capital and higher firm valuation (e.g., Verrechia, 2001; Shleifer and Wolfenzon, 2002). Firm-specific costs include direct costs such as collecting, preparing and disseminating the respective reports as well as indirect costs such as the release of proprietary information to the market, which might benefit competitors (e.g., Butler et al., 2007). In the absence of disclosure regulation, firms voluntarily disclose information as long as the corresponding benefits exceed the costs. Besides these firmspecific costs, however, disclosure also has market-wide benefits and costs, which render a potential firm-specific disclosure equilibrium suboptimal. Previous literature finds that mandatory disclosure is particularly beneficial in the case of positive externalities, potential cost savings (e.g., through scale and standardization), stricter sanctions and sanction enforcement as well as costs from fraud and agency conflicts that could be mitigated by disclosure (Leuz and Wysocki, 2008) 3. However, mandatory disclosure can also have negative externalities such as unexpected interactions with other (disclosure) mechanisms or institutional features and it can ultimately also impact real decisions made, e.g., by managers (Beyer et al., 2010). While a variety of studies has investigated the firm-specific effects of voluntary disclosure (in particular the corresponding capital market effects), these studies cannot provide insights into the overall desirability, economic efficiency and potential negative externalities of regulating these disclosures (Leuz and Wysocki, 2008). Some studies have therefore compared voluntary and mandatory settings to test for potential differences caused by regulation. For instance, Butler et al. (2007) show that the timeliness of earnings in the 3 The authors provide an extensive review and framework of the existing literature on disclosure, its cost and benefits as well as corresponding economic consequences

8 U.S. only significantly increased for voluntary adopters of quarterly reporting and not for mandatory adopters. A large range of studies focuses on the firm-specific, capital market and economic effects of all kinds of disclosure regulation as well as changes to this regulation. 4 The recent literature on the effect of the Sarbanes-Oxeley Act or the mandatory IFRS introduction are only two examples (e.g., Zhang, 2007; Chhaochharia and Grinstein, 2007; Cohen at al., 2008; Daske et al., 2008). We contribute to this literature by providing further insights into potential externalities of one particular type of mandatory disclosure, i.e., mandatory quarterly reporting Management myopia and earnings benchmarks Economic theory suggests several reasons for a myopic behavior of managers, which cannot directly be explained by Jensen s (1986) agency-theoretic view of the firm. Stein (1988) proposes a managerial myopia theory, which predicts myopic management behavior by focusing on short-term actions to inflate earnings and to increase the stock price assuming the prevalence of market imperfections. 5 The empirical literature documents myopic behavior of managers, especially under certain conditions like pending stock issuances (Bhojraj and Libby, 2005), IPOs (Teoh et al., 1998), employment concerns (Fudenberg and Tirole, 1995) or compensation issues (Hall and Murphy, 2003), as well as in the presence of institutional investors (Bushee, 1998). Graham et al. (2005) document myopic behavior of managers to meet expectations of analysts and investors when disclosing financial reports. A significant part of previous literature has relied on the concept of meeting or beating earnings benchmarks to identify years in which myopic behavior is highly likely, so-called suspect years (e.g., Roychowdhury, 2006; Gunny, 2010; Zang, 2010). 4 Leuz and Wysocki (2008) provide a comprehensive review of the existing literature. 5 Further important research on the prevalence and conditions of management myopia includes, e.g., Laffont and Tirole, 1987; Stein, 1989; De Long et al., 1990; Sloan, 1996; Bar-Gill and Bebchuk, 2003; Ellis,

9 We contribute to this literature by testing if myopic behavior proxied by real activities manipulation to meet earnings benchmarks is more prevalent if reporting frequency is higher. Interim disclosure frequency, earnings management and real business effects Managerial myopia can lead to earnings management when managers use judgment and discretion in financial reporting to mislead stakeholders about the underlying economic performance (Healy and Whalen, 1999). Managers have at least two options to achieve this: They can either use accounting based earnings management by accruals manipulation and classification shifting or they can use "real" earnings management defined as adapting their operational practices to influence financial performance. On the one hand, a variety of studies have shown that managers influence financial reporting through accounting discretion (commonly known as "earnings management") and classification shifting (e.g., Jones, 1991; Dechow and Dichev, 2002; Kothari et al., 2005; McVay, 2006; Fan et al., 2010). Some of these studies focus on reporting frequency and quarterly reports. Degeorge et al. (1999) propose a quarterly earnings threshold hierarchy, i.e. managers seek to avoid quarterly losses or quarterly earnings decreases more than meeting or beating financial analysts quarterly earnings forecasts. Brown and Caylor (2005) examine and find temporal changes in this hierarchy. Since the mid-1990s firms appear to focus more on avoiding quarterly earnings surprises due to the increased importance of analysts forecasts. Das et al. (2009) document that patterns in quarterly earnings changes reflect accounting earnings management behavior. Companies that perform poorly in interim quarters try to increase earnings in the last quarter to reach earnings benchmarks and vice versa (earnings reversals). On the other hand, a growing stream of literature also investigates managers' willingness to depart from normal operational practices ostensibly to achieve certain reporting goals. These practices include, for example, price discounts to increase short term revenues, production increases to reduce average cost per unit, and the reduction of discretionary - 8 -

10 expenses to boost short term profit (Roychowdhury, 2006). Previous research on real activities manipulation (RAM) focuses on opportunistic R&D expenditure reduction to meet earnings forecasts or avoid share price dilution (e.g., Bens et al., 2002; Dechow and Sloan, 1991). Other findings include the acceleration of sales and the delay of R&D and other discretionary expenses (e.g., Healy and Wahlen, 1999; Fudenberg and Tirole, 1995; Dichev and Skinner, 2002). Roychowdhury (2006) finds evidence for abnormal levels of cash flow from operations, production cost and discretionary expenses triggered by managers trying to avoid reporting losses. Using similar measures, Cohen et al. (2008) find that in the post- Sabanes-Oxley period, many firms switched from accrual-based to real activities manipulation methods. This is consistent with the model of Ewert and Wagenhofer (2005) who suggest that tighter accounting standards might increase the expected total cost of earnings management due to a switch from accounting to more costly real activities manipulation. Our study focuses on real activities manipulation rather than accounting earnings management, since we investigate the real business effects of mandatory disclosure, particularly of higher reporting frequency. By doing so, we add to the existing literature on externalities of mandatory disclosure beyond capital market effects to provide evidence about real effects of disclosure regulation (Leuz and Wysocki, 2008). Reporting frequency-induced real business effects Large-scale empirical studies investigating the real effects of mandatory disclosure, specifically reporting frequency, are sparse. In an experimental study with experienced financial managers, Bhojraj and Libby (2005) investigate the investment preferences of managers and find that only if capital market pressure is high, an increase in reporting frequency results in myopic management behavior. In a survey study among executives in the U.S., Graham et al. (2005) find that given the pressure to meet quarterly earnings targets, 80% - 9 -

11 of the responding managers would decrease discretionary spending (e.g. R&D). Almost 40% would even provide incentives for customers to buy more products in the current quarter. Gigler et al. (2009) examine the impact of more frequent mandatory reporting in the presence of multiple market imperfections. They assume that managers cannot disclose credibly whether they are investing in short-term or long-term project. While the short-term project generates stochastically higher cash-flows in the early periods and lower cash-flows in the later periods, the long-term project overall has a higher net present value. In such a setting, more frequent financial reports lead to more efficient market prices (i.e., prices better reflect the firm's underlying cash flows), but also induces managers to engage in short-term projects which overall may not be socially desirable. They conclude that more frequent disclosure could be associated with a trade-off between higher price efficiency and lower economic efficiency, because the more frequent disclosure regimes can change the business activities they report about and thus can have strategic consequences. This evidence suggests that managerial myopia is a consequence of a higher disclosure frequency, in particular when capital market pressures are present. We contribute to the literature on the real effects of mandatory disclosure regulation in arguing that mandatory quarterly reporting leads managers to deviate from normal operation practices. In line with previous research, we measure real activities manipulation of managers in suspect years, i.e., when firms meet or beat earnings forecasts. We conjecture that a higher frequency of reporting is associated with higher RAM for several reasons: First, the main difference between semi-annual and quarterly reporting is the number of times a company releases full financial statements to the market. This increase in frequency leads to an increase in the number of potential earnings benchmarks that a company is expected to meet. 6 Following previous research (Roychowdhury, 2006; Gunny, 2010; Zang, 2010), managers use real activities manipulation to meet earnings benchmarks. Due to the 6 In line with Gunny (2010), we define earnings benchmarks as meeting or beating zero earnings as well as last period earnings. Quarterly reporters therefore have 8 (4x2) potential interim benchmarks and 2 yearly benchmarks, whereas semi-annual reporters only have 4 (2x2) interim benchmarks and 2 yearly benchmarks

12 increased number of earnings benchmarks, we expect firms to more heavily use RAM. Alternatively, one could also argue that since quarterly reporters only have to reach half of the semi-annual benchmark managers only have to use half of the level of RAM. However, RAM does not reverse automatically, at least not quickly, so the effect might be escalating the more often it is used in shorter successive time windows. For example, assume that a manager uses abnormal sales discounts in one period to meet a certain earnings benchmark. The market will probably adapt to the new price level quickly and it will be hard to restore the original price in the next period. In order to reach the target in the next period, the manager might have to reduce the price further or at least keep it at the lower level. The longer time horizon of semi-annual reporters should help in returning to normal operational practices and potentially also in restoring the price level in this example. 7 Second, a higher reporting frequency can create additional pressure due to more timely enforcement of contractual arrangements (e.g., performance-based compensation of managers, credit contracts). Penalties for missing performance targets or breaching debt covenants might therefore play an important role in triggering myopic decision making of managers of quarterly reporters. If we assume that performance is randomly distributed and missing a contractual arrangement only once will lead to sanction, then quarterly reporters incur a higher risk of missing at least one such benchmark because contracts are enforced four times instead of twice a year. This effect should be reinforced by the shorter time window to compensate for potential deline in performance during the period. Consequently, managers of quarterly reporters are likely to use disproportionally more RAM compared to semi-annual reporters. Third, quarterly reporters are supposedly subject to a higher level of capital market pressure because firms that report more frequently are potentially more attractive to investors 7 Another example of an escalating effect of RAM would be an increase in production with the intention to reduce average COGS: In the next period, one could either decrease production (which would be an economic reversal of RAM) at the cost of increasing average COGS. However, managers might be tempted/have to reduce the price of the units in stock to increase sales while keeping production constant which would further increase RAM

13 with shorter investment horizons due to the more timely information updates and the increased level of transparency associated with more frequent disclosure. If the group of short-term oriented shareholders is relatively large, investor myopia might induce myopic management behavior if management acts in the best interests of its average shareholders. 8 Taken together, these arguments suggest that a higher reporting frequency induces myopic management behavior. Following previous studies, we use RAM measures as proxies for testing the following hypothesis: H1: Mandatory quarterly reporting is associated with higher real activities manipulation (RAM) compared to semi-annual reporting. Cross-sectional differences of frequency-induced real activities manipulation Prior theoretical work (Stein, 1989; Bar-Gill and Bebchuk, 2003) suggests that the degree of myopic behavior will be influenced by the extent to which managers care about short-term price relative to long-term value. We identify several incentives and monitoring mechanisms arising from institutional differences, capital market differences and industry differences that can exacerbate or mitigate the impact of quarterly reporting on RAM. At the institutional level, we hypothesize that so-called insider economies (i.e., weaker securities regulation, weaker investor protection and lower securities regulation enforcement) exhibit a stronger effect than outsider economies due to an overall lower level of institutional quality resulting in higher RAM (Leuz, 2010). Similarly, we expect the effect to decrease in the level of overall regulatory enforcement (Kaufmann et al., 2010), because of the role of enforcement in monitoring the compliance of rules and in mitigating manipulations (Leuz et al., 2003). 8 The spiral of silence theory for financial markets proposed by Aspara et al. (2008) even predicts managerial myopia when the majority of financial analysts and (institutional) investors is not short-term oriented. The theory assumes that financial market participants could fear disadvantages from behaving according to a long-term orientation when they perceive short-term orientation as the dominant orientation and a larger proportion of market participants could shift to behave myopic. This effect is reinforced by a herding behavior, by short-term evaluations of investment managers and analysts and by potentially biased reports of the mass media. Quarterly reporting can act as a catalyst in this dynamic and self-reinforcing process, because it provokes mass media reports about short-term performance changes and a short-term evaluation of managerial performance by internal and external stakeholders. As a result short-term orientation could become the predominant orientation and could trigger managerial myopia even when the majority of financial market participants is not short-term oriented

14 Capital market differences might have an effect on the results. The shareholder horizon is likely to be a strong determinant of management decision horizon. A higher share of shortterm oriented investors correspondingly induces managers to behave myopically creating an additional incentive for higher RAM. In addition, the number of financial analysts following a firm should reduce the reporting frequency-induced RAM because of their external monitoring function (Yu, 2008). 9 Finally, on the industry level, we expect that shorter operating cycles lead to increased managerial myopia because of a smaller time window to compensate for a decline in performance and because investors of firms with short operating cycles generally tend to value short-term information more heavily (Zang, 2010). We also conjecture that reporting frequency-induced RAM increases in industry competition because firms have to use RAM more heavily due to higher peer pressure. We correspondingly predict: H2: Reporting frequency-induced RAM is higher in weak institutional environments, for firms with more short-term oriented investors and a lower number of analysts that monitor managers as well as in industries with shorter operating cylces and higher competition III) METHODOLOGY, DATA, AND DESCRIPTIVE STATISTICS Regulatory background We exploit the institutional setting of the European Union to test our hypotheses. Europe has a high level of harmonization with respect to financial disclosure and reporting regulation. 10 One of the few choices left to the individual member states is to whether or not the national regulators mandate quarterly financial reporting. In this setting, it is therefore 9 An alternative view would be that a higher number of analysts creates additional pressure on managers to meet targets. These competing hypotheses concerning the role of financial analysts is examined by Yu (2008). Following his results, we conjecture a monitoring function of analysts and predict that the number of analysts decreases the magnitude of reporting frequency-induced RAM. 10 All publicly-traded companies have to report according to IFRS. Capital markets are highly integrated and there is a common supervisory body next to the national regulators

15 easier to isolate the effect of different reporting frequency regimes than in comparable international studies. Table 1 gives an overview about the specific requirements in each EU-15 country, which we compiled from an extensive review of documents from the EU, national regulatory authorities, stock exchanges, and from interviews with financial analysts and stock market operators in the respective countries. In Finland, Greece, Italy, Portugal, and Spain, the regulatory authorities require listed firms to publish full quarterly financial reports. In Sweden, the stock market operator on both regulated stock markets requires full quarterly financial reports for all companies. In Austria and Germany, companies listed in specific stock market segments ("Prime Market" in Austria and "Prime Standard" in Germany) are also obliged to publish full quarterly reports. The remaining seven countries (Belgium, Denmark, France, Great Britain, Ireland, Luxembourg, The Netherlands) do not mandate full quarterly reports or quarterly earnings announcements. We classify companies into semi-annual reporters and quarterly reporters based on the corresponding Datastream item WC05200 ("earnings reporting frequency"). 11 [place Table 1 here] Measuring real business effects To test for real business effects of interim reporting frequency, we develop a consolidated real activities manipulation measure. We use the model from Dechow et al. (1998), further refined by Roychowdhury (2006) to calculate specific components of RAM, i.e., abnormal cash flow from operations (ACFO), abnormal production costs (APROD) and abnormal discretionary expenses (ADISC). We derive normal levels of the corresponding measures based on industry-year regression similar to Roychowdhury (2006). The corresponding residuals of these regressions represent the abnormal level of the measures. 11 We perform sensitivity checks on the validity of the Datastream item. Changing the definition of quarterly reporters to those companies for which quarterly net earnings are actually available in Datastream does not affect our results

16 The intuition behind real activities manipulation and its calculation has been well established in previous literature. We have compiled explanations and formulas for the generation of ACFO, APROD and ADISC in Appendix 1. A substantial amount of studies has validated, consolidated and refined the measures (e.g., Zang, 2010; Cohen et al., 2008; Gunny, 2010, Cohen and Zarowin, 2010). We establish a new consolidated RAM measure, because previous studies have only partially consolidated the individual RAM components. In order to derive our consolidated RAM measure (TOTALRAM), we rely on the previously identified RAM components, i.e., sales manipulation, overproduction and a reduction in discretionary expenses. TOTALRAM is defined as the sum of all individual RAM components: TOTALRAM = RAM ( Salesmanip) + APROD + ADISC (1) where all variables are defined in Appendix 2. While APROD and ADISC can be calculated directly and are not affected by other RAM activities, sales manipulation needs to be calculated via ACFO, which consolidates the effect of sales manipulation, overproduction and a reduction in discretionary expenses. In line with Roychowdhury (2006), ACFO is defined as follows: ACFO = RAM ( Salesmanip) + APROD ADISC where all variables are defined in Appendix 2. As established by Roychowdhury (2006), excessive discounts or more lenient credit terms and increases in COGS or inventory increase ACFO. As opposed to that, a reduction in discretionary expenses increases cash flows such that ACFO is biased downwards by ADISC. RAM(Salesmanip) can therefore be rewritten as follows: RAM ( Salesmanip) = ACFO APROD + ADISC (2) where all variables are defined in Appendix 2. In order to calculate TOTALRAM, we combine equation (1) and (2): TotalRAM = ACFO + 2*( ADISC) (3)

17 where all variables are defined in Appendix 2. While previous research has combined two of the three individual measures presented by Roychowdhury (2006),e.g., Cohen and Zarowin (2010), this is the first paper that consolidates all individual RAM activities into one measure. We use TOTALRAM throughout the paper for our main analyses. As additional sensitivity analysis, we also use the individual RAM measures to validate our results and determine the individual effect of each of the manipulation activities. We always calculate RAM on an annual level as it is commonly done in the literature. In addition, interim financial data other than sales and net income is not available on Datastream. We rely on the yearly RAM measure to identify differences between the two interim reporting frequency regimes. This should however not impede the robustness of our findings for two reasons: First, as opposed to accounting earnings management, there is no automatic reversal of RAM. For instance, a price discount that is granted in any given quarter is not necessarily off-set by a price increase in the following quarter.. We argue that once a manager has started to ride the tiger, it is comparatively hard to get back to normal operational practices, in particular if reporting cylces are short. Second, as is described below, by using suspect year definitions based on meeting or beating interim reporting targets, we are able to distinguish between interim and yearly RAM effects. Therefore, the effect should also be measurable on a yearly basis. Research design We perform extensive analyses to validate our findings. First, we use a propensity based score matching procedure to mitigate endogeneity concerns. Second, we employ the suspect year concept introduced in prior research to ensure that we primarily analyze years in which the probability of RAM is high. Third, we employ two fundamentally different regression analyses to reduce concerns on correlated omitted variable bias

18 Matching procedure As described above, our sample consists of both, countries that mandate quarterly reporting and countries that leave the choice of the interim reporting frequency to the individual companies. Quarterly reporters that voluntarily opt to do so by definition self-select into the group of quarterly reporters. These might be firms with a set of particular characteristics, which raises endogeneity concerns. We therefore exclude all voluntary quarterly reporters from the sample. However, even in mandatory reporting environments, there might be firms that would opt to report voluntarily anyways. Our study investigates the effects of mandatory quarterly reporting so if these companies had adopted the same reporting frequency without regulation, they are unaffected and including them in the sample might bias our results. We therefore intend to create a group of pure mandatory reporters, those that would presumably not report quarterly if they did not have to. We therefore use a probit model to determine the characteristics of the voluntary quarterly reporters to eliminate similar firms from our sample of mandatory reporters. We assume that firms with the following characteristics are more likely to adopt quarterly reporting voluntarily, which we estimated based on a probit model: Pr obit( voladopt) = α + α SIZE + α LEV + α DIV + α CLHELD + α FORSALES + α DELFORSALES + α OPCYC + α HERF + α INDSIZE + α TOPIND + α ANFOL + α INDEX α13lagroa + Fixedeffects + ε where all variables are defined in Appendix After plugging the characteristics of our mandatory firms into the probit model, we identify firm-years in the mandatory sample that have a hypothetical probability of voluntarily adopting quarterly reporting in excess of 50%. We exclude these from the sample. After this initial matching, our sample should consist of firms that report semi-annually and pure mandatory reporters, i.e., those that would (most probably) report semi-annually in a voluntary environment. One potential challenge of the setting we are using in the first set of analyses are structural differences in the characteristics of different companies across different countries

19 In order to account for this, we use a second propensity based score matching to ensure that industry and size of the firms is comparable. Based on this analysis, we further exclude firmyear observations that do not have a comparable counterpart in the other reporting frequency regime. To further address concerns about country differences, all of our regressions include country fixed effects as well as standard errors clustered by country (Christensen et al., 2010). Moreover, our measures are constructed such that industry and year fixed effects are also accounted for. Suspect years Previous research has extensively used the suspect year concept to identify years in which increased RAM is more likely than in others. This is important as it reduces potential measurement bias. While some studies investigate activities manipulation around seasoned equity offerings (e.g., Cohen and Zarowin, 2010), most studies have relied on years in which firms meet or beat earnings benchmarks (e.g., Roychowdhury, 2006; Zang, 2010; Gunny, 2010). We follow Gunny (2010) and define suspect years as firm-year observations that just meet of beat past years' earnings or the zero earnings benchmark. We define observations as suspect years if either of the two following conditions holds: Firm-years in which a company s net income before extraordinary items as percentage of lagged total assets is between 0 and 1% or firm-years in which delta net income as percentage of lagged total assets is between 0 and 1% as suspect years (SUSPECT_Y; Gunny, 2010). In the RAM literature, activities manipulation is usually calculated based on annual figures similar to suspect years. Given our research question, however, we are also interested in meeting or beating earnings benchmarks of interim reports. This gives us the opportunity to disentangle potential RAM drivers more appropriately than just comparing annual suspect years with annual RAM levels. Specifically, we are interested in firm-years where quarterly reporters and semi-annual reporters only meet or beat interim earnings benchmarks and

20 explicitly not the yearly benchmark. 12 We calculate quarterly and semi-annul earnings benchmarks by using only a forth and half of the above range, respectively. We therefore define two more types of suspect years: First, firm-years in which quarterly reporters meet or beat benchmarks in 2 successive quarters in the same half-year as well as semi-annual reporters meeting or beating the first or second half year target (SUSPECT_2Q, basic interim suspect year). This suspect year definition provides the basis for an important test: It allows us to compare quarterly reporters RAM level when they meet two successive quarterly benchmarks in the same half-year to semi-annual reporters meeting or beating one benchmark over the same period. Second, to make our analysis even more robust, we compare firm-years in which quarterly reporters meet or beat only one quarterly benchmark to semi-annual reporters that meet or beat one half-year benchmark (SUSPECT_1Q, strict interim suspect year). Both of the interim tests are particularly interesting because they allow us to compare quarterly and semi-annual reporters RAM activities directly: Finding that quarterly reporters exhibit higher RAM in SUSPECT_2Q or SUSPECT_1Q firm-years, would indicate that RAM for quarterly reporters is much higher for quarterly reporters even when comparable benchmarks (or even lower benchmarks for quarterly reporters in the case of SUSPECT_1Q) are met. All suspect year definitions are summarized in Appendix 1 and descriptive statistics are provided in Table 3. Regression setup In order to test our first hypothesis, we run extensive cross-sectional analyses to compare quarterly and semi-annual reporters RAM in different suspect years. We therefore run the following regression: Controls + Fixedeffects + TOTALRAM = α + α SUSPECTX + QR + QR * SUSPECTX + ε 0 1 where all variables are defined in Appendix 2. (4) 12 For instance, a firm can very well meet or beat several interim benchmarks, without meeting or beating the yearly benchmark and vice versa

21 We use a wide of control variables to account for potential differences in company characteristics. These include several measures of accounting earnings management, a potential substitute of RAM, which might bias our results if it is not controlled for (e.g., Zang, 2010; Gunny, 2010). As our main explanatory variable (QR) varies on country level, we use country fixed effects and standard errors clustered by country to account for country differences similar to Christensen et al. (2010). As TOTALRAM is measured by industryyear, fixed industry and year effects are also be accounted for. Despite including a variety of fixed effects, we cannot completely rule out potential correlated omitted variable bias. In section 5, we therefore make an additional test using a pre-treatment/post-treatment setup of a disclosure shock that primarily affected semi-annual reporters (the Transparency Directive) To validate our first hypothesis, we also combine the different suspect year definitions in one equation in order to separate the RAM effects of meeting or beating yearly and interim earnings benchmarks. We run the following regression: TOTALRAM = α + α SUSPECT _ Y + α SUSPECT _ XQ + α QR + α QR * SUSPECT _ Y + α QR * SUSPECT _ XQ Controls + Fixedeffects + ε 3 4 (5) where all variables are defined in Appendix 2 and SUSPECT_XQ represents both, SUSPECT_2Q as well as SUSPECT_1Q. If SUSPECT_2Q and SUSPECT_1Q are found to be significant next SUSPECT_Y in explaining TOTALRAM, this can be interpreted as additional evidence for the importance of interim earnings benchmarks in explaining overall RAM levels and thereby differences in RAM levels across different interim reporting frequency regimes. In order to test our second hypothesis, we examine whether in how our results depend on institutional, capital market and industry characteristics. This analysis is particularly important for at least two reasons. First, if the results differ in particular settings, this reduces concerns about omitted variables driving the results. Second, it is important to better understand the supporting factors and conditions of real effects of disclosure frequency,

22 e.g., to derive potential policy implications. We test hypothesis H2 using 6 different institutional, capital market and industry factors. Institutional factors include enforcement (low=1) as defined by Kaufmann (2010), general regulatory differences based on country clusters defined by Leuz (2010) 13. Capital market differences include shareholder horizon proxied by stock turnover (Polk and Sapienza, 2008; high=1) and abnormal number of analyst following (low=1) 14 Industry differences include operating cycle (short=1) and industry competition (high=1) measured using the Herfindahl index. Each of the cross-sectional variables is split into two groups depending on its position below or above the median. We use the following regression to test the hypothesis: TOTALRAM = α1 SUSPECT _ Y * QR * CROSS1 + α 2SUSPECT _ Y * QR * CROSS 0 + Controls + Fixedeffects + where all variables are defined in Appendix 2. ε (6) We compare QR in suspect years between the two groups. In addition to a large range of controls including several controls for the level of accounting earnings management, we include country fixed effects to again account for country differences within the two groups. Control variables We use a wide range of controls to mitigate omitted variable concerns and to account for other known RAM drivers. The controls can be classified broadly into four categories: Company characteristics, capital market effects, industry characteristics and the level of accounting earnings management. First, we include a range of company characteristics to further improve comparability, although the matching procedure as well as the elimination of cross-listed shares should already substantially reduce potential bias from structural differences between the companies. SIZE (natural logarithm of total assets) is included because bigger firms supposedly have 13 We split the 5 country clusters in table 4 (Leuz, 2010) into 2 groups, cluster ( outsider economies) as well as cluster ( insider economies ). No countries from cluster 5 are included in our sample. 14 Based on regressing of analyst following on size, book-to-market, market share and common shares outstanding by industry-year

23 more RAM, for instance because of their high visibility and the increased importance of meeting benchmarks. Firms with higher short term LEV (leverage, i.e., current liabilities divided by lagged total assets) are expected to have more RAM because meeting and beating earnings benchmarks is particularly important for these firms to get refinancing or not to breach debt covenants (Roychowdhury, 2006). For NI (net income divided by lagged total assets), we expect a negative effect because firms with higher net income are less likely to engage in RAM. CLHELD (closely held shares, i.e., shares held by shares held by officers, directors and their immediate families, shares held in trust, shares of the company held by any other corporation, shares held by pension/benefit plans and shares held by individuals who hold 5% or more of the outstanding shares) is included to account for differences in governance that might explain the level of RAM. We expect RAM to be negatively correlated to the amount of shares closely held. Second, capital market controls include SHARES (common shares outstanding) to account for the capital market orientation of firms. A greater number of shares can imply higher RAM as more activity is needed to achieve a given per share earnings target. Alternatively, it can discourage managers to use RAM because a given target is more difficult to reach. (Cohen and Zarowin, 2010). We also include ANFOL (natural logarithm of one plus analyst following) because financial analysts are amongst the most important recipients of interim and annual financial statements. Analysts can fill one of two roles: They can either have a monitoring function similar to effect of CLHELD, which reduces average RAM levels (Yu, 2008). Alternatively, they can exercise pressure on firms and thereby force managers to use more RAM. We follow Yu (2008) and expect the monitoring function to preponderate. Third, industry characteristics include MKTSHARE (market share which equals company sales divided by total industry sales) where firms with higher market share to have lower RAM because competition is lower so the pressure to meet or beat earnings benchmarks is lower. Zang (2010) argues that for firms with higher market share, RAM is comparatively less costly, so we are agnostic about the sign. In addition, we include an

24 indicator variable for LITIND (high litigation industries). The effect could be either positive or negative because firms could potentially be sued for missing earnings targets or for excessively using RAM, if detected. We also include OPCYC (a firms operating cycle measured as average receivables over the last two years divided by sales plus average inventory over the last two years to COGS; Biddle et al., 2009). We expect it to be negative because a longer operating cycle leaves more flexibility to use accounting earnings management due to smaller accrual accounts and a shorter period for accruals to reverse and therefore requires less RAM (Zang, 2010). Fourth, we account for a potential trade-off between accounting earnings management and RAM which is an important determinant of the RAM level (e.g., Ewert and Wagenfofer, 2005). Building upon previous research (Barton and Simko, 2002; Zang, 2010; Cohen and Zarowin, 2010), we include NOA (net operating assets, i.e., shareholders equity less cash and marketable securities plus total debt at the beginning of the year, divided by lagged total assets). This is a proxy of firm s balance sheet constraint to use accounting earnings management based upon its previous accounting choices. We expect a higher value of NOA to be positively correlated with RAM, as firms are willing to use more RAM if they are constrained in their opportunity to use accounting earnings management. We also include BIG4, i.e., an indicator variable that is one if the firm s auditor is a BIG4 auditor. (Zang, 2010; Cohen and Zarowin, 2010). We are agnostic about the sign because BIG4 can either force the company to reduce accounting earnings management and thereby have a positive effect on RAM. Alternatively, BIG4 could also have a negative effect on overall earnings management (accounting and RAM) due to higher level and quality of monitoring of earnings management activities. Sample selection We collect a sample of all shares covered by Datastream/Worldscope and I/B/E/S in EU-15 countries between January 2005 and December 2009, also including shares that were

25 delisted during the period in order to avoid survivorship bias. 15 To ensure a common financial reporting and disclosure setup, we do not use data prior to January 2005, as IFRS was not mandatory before that date and national reporting standards might deter our results significantly. We require that firms have data on total assets for at least two consecutive years from 2004 until 2009, distinct fiscal year end information and unique ISIN identifiers. This results in a starting sample of 16,517 firm-year observations from 2005 to In deriving our final sample, we first exclude financial firms (SIC ), firms from highly regulated industries (SIC ), public administrative firms (SIC > 9000) and firm-years in financial distress (common equity < 0). Next, we drop all firms that are cross-listed in the U.S. and are therefore subject to SEC quarterly reporting requirements. To further clean our sample, we also delete firm-years of voluntary quarterly reporters (i.e., quarterly reporters in voluntary reporting regimes) that are cross-listed on stock exchanges in mandatory countries as well as firm-years with non-quarterly observations in mandatory countries. This results in 9,412 firm-year observations. The application of our matching procedure reduces our sample further. We drop firms from the mandatory sample that would be likely voluntarily adopters in a voluntary regime ("non-pure" mandatory firms) to generate a sample of pure mandatory adopters, i.e., those firms that are really affected by disclosure regulation. In addition, we drop all voluntary reporters, since we focus on the effect of mandatory quarterly reporting only and to mitigate endogeneity concerns. Finally, when matching pure mandatory reporters and semi-annual reporters based on industry and size, we need to further restrict our sample to 7,372 firm-year observations for our empirical analyses. 15 Datastream/Worldscope is widely used in international studies. In their paper on the "effects of database choice on international accounting research", Lara et al. (2006) list 33 key research papers for international accounting, of which 15 use Datastream or Worldscope

26 In order to mitigate the influence of outliers all continuous variables are winsorized at the 1% and 99% level. 16 The sample selection process is summarized in Table 2. [place Table 2 here] Table 3 gives an overview of quarterly and semi-annual observations in our sample by country (Panel A) and by suspect year (Panel B). After the matching, our sample is equally split between quarterly and semi-annual reporters (3,686 each, Panel A). After having excluded the voluntary adopters, Austria and Germany are the only two remaining countries with both quarterly and semi-annual observations due to the specific listing requirements of the stock exchanges (Prime Market in Austria, Prime Standard in Germany). The distribution by suspect years (Panel B) reveals that there are 607 suspect firm-years of semi-annual reporters and 573 suspect firm-years of quarterly reporters, using the yearly suspect year definition (SUSPECT_Y), i.e., firm-years in which a company exclusively meets or beats the yearly earnings benchmark. Looking at interim suspect years, semi-annual reporters meet or beat earnings benchmarks in one (two) half-year(s) in 800 (138) firm-years. Quarterly reporters meet or beat one, two, three and four quarterly benchmarks per year in 902, 302, 79, 15 firm-years, respectively. We aggregate these interim suspect firm-years in the two ways described above to generate SUSPECT_2Q and SUSPECT_1Q. In both cases we exclude all observations of companies that meet or beat the yearly benchmark in addition to the interim benchmarks to test only the effect of interim benchmarks on RAM. [place Table 3 here] Descriptive statistics Table 4 provides the descriptive statistics as well as correlation coefficients for the continuous variables. Panel A summarizes the results. By definition, the overall average of TOTALRAM is (very close to) zero. Splitting up TOTALRAM in sub-groups, however, 16 Similar to Verdi (2006), we winsorize by year

27 already reveals some indication about potential cross-sectional differences. The mean for the quarterly observations is positive and negative for the semi-annual reporters. In suspect years the average is positive, while in non-suspect years it is negative. TOTALRAM is the most restrictive variable, with 3,317 firm-year observations. Panel C presents correlations for the measure and continuous control variables used in the regression models. The only variable that exhibits substantial correlation with other variables is SIZE. Since SIZE is probably amongst the economically most relevant controls, however, we keep it in the sample alongside the other controls. [place Table 4 here] IV) MAIN RESULTS Reporting frequency-induced real business effects In order test our hypothesis H1 which predicts that quarterly reporters exhibit higher RAM levels in suspect years than semi-annual reporters, we perform multiple regression analyses. Table 5 presents the results of estimating Equation 4, i.e., the effect of quarterly reporting on TOTALRAM under 3 different suspect year setups (SUSPECT_Y, SUSPECT_2Q, SUSPECT_1Q). For the yearly suspect year regression (SUSPECT_Y), we find that both of our experimental variables, quarterly reporting and suspect years, have a significantly positive effect on TOTALRAM (t-statistics of 3.19 and 3.18, respectively). Testing the joint effect using a one-sided F-test, we find that quarterly reporters exhibit significantly more real activities manipulation in suspect years (F-statistic of 8.02). We repeat the same regression also for our two interim suspect year definitions. For the basic interim test (SY_2Q), we also find significant coefficients for suspect years and quarterly reporting (t-statistics of 2.39 and 3.27, respectively). Testing again the joint effect of both, reveals a highly significant effect (F-statistic of 14.72). Also using the strict interim

28 suspect year definition (SY_1Q), we can confirm these findings. The coefficients for suspect years (t-statistic of 2.39) and quarterly reporting (t-statistic of 2.73) are both positive and significant, as well as the joint test (F-statistic of 9.44). In all three regressions, we control for all other factors discussed in section III. We find the expected signs for the coefficients of our controls. We also include country fixed effects and cluster the standard errors by country. The adjusted R 2 is between 8.3% and 8.5% and the sample size is 2,557 firm-year observations. Overall, all three regressions summarized in table 5 support hypothesis H1 that quarterly reporters exhibit more RAM than semi-annual reporters in suspect years. Further, looking at the interim definition of suspect years, we find that the effect also holds if managers only manage their interim results. This is particularly interesting, because it allows us to compare RAM levels in similar periods and for the same earnings benchmarks between semi-annual and quarterly reporters. Despite similar thresholds, quarterly reporters have a higher level of RAM compared to semi-annual reporters. Moreover, the strict interim regression even demonstrates that one suspect quarter alone induces more RAM than one half-year. [place Table 5 here] In order to mitigate concerns that omitting either yearly or interim suspect years in Equation 4 might drive the results, we estimate two regressions based on Equation 5 which include yearly and interim suspect years simultaneously. The results are summarized in summarized in Table 6. First, we regress TOTALRAM on quarterly reporting, yearly suspect years, and the basic interim suspect years measure (SY_2Q). We find that the individual effect of all three experimental variables is significantly different from zero (t-statistics of 3.30, 3.97, and 2.81, respectively). Testing the joint effect of quarterly reporting with yearly suspect years and quarterly reporting with basic interim suspect years, the one-sided test (F-statistics of 7.62 and 16.36, respectively) reveals that the effect of quarterly reporting is particularly strong in suspect years and that both suspect years are significant at the same time. This can be

29 interpreted as evidence for interim earnings benchmarks driving RAM even after controlling for yearly benchmarks, which mitigates concerns of omitted variable bias. With the second regression using the strict interim suspect year definition (SY_1Q), we can even use a more challenging test on the effect of quarterly reporting on RAM. Since the results also hold for this setup, in which meeting or beating one quarterly benchmark by quarterly reporters causes more RAM than meeting or beating one semi-annual benchmark by semi-annual reporters, we conclude that quarterly reporting is associated with very strong TOTALRAM premium compared to semi-annual reporters. The corresponding F-statistics are 7.20 and 9.02, respectively. Also in these two regressions we carefully control for other effects, which signs are in the expected direction, as discussed in section III. Again, we include country fixed effects and clustered standard errors by country. Taking together the results from both table 5 and table 6, we find strong support for hypothesis H1. [place Table 6 here] Cross-sectional differences of frequency-induced real activities manipulation The results of the analysis of cross-sectional differences in the investigated effect are presented in Table 7. The regressions are based on Equation 6 and separate the effect of quarterly reporting using yearly suspect years into two groups of the respective crosssectional variable at hand (e.g., "high" and "low" enforcement quality). We test two institutional, two capital market and two industry characteristics. Looking at all cross-sectional analyses at once, we are able to confirm the overall effect of quarterly reporting in suspect years, as found before. Except for "enforcement quality", for which the F-statistic is slightly below the significance level (1.43), all other cross-sectional difference variables exhibit highly significant QR coefficients. (F-statistics ranging from 9.62 for "shareholder horizon" to for "monitoring by analysts")

30 The more interesting test, however, is to look whether the split of the sample into two different groups reveals significant differences for these groups. We use a one-sided F-test to test these differences. We find the strongest cross-sectional differences as a result of industry characteristics. Ceteris paribus, quarterly reporters with short operating cycles (F-statistic of 12.70) and strong industry competition (F-statistic of 6.23) exhibit significantly higher RAM in suspect years compared to their cross-sectional counterparts with longer operating cycles and weaker industry competition. Concerning capital market differences, we also document significant cross-sectional differences. Quarterly reporters with an abnormally low number of analyst, i.e. low monitoring, (F-statistic of 3.37) and short shareholder horizons (F-statistic of 2.48) experience higher real activities manipulation in suspect years compared to the control group with the opposite characteristics. The institutional differences exhibit the least significant cross-sectional differences between treatment and control group. While in low quality enforcement countries, quarterly reporters use significantly more RAM (F-statistic of 2.02), the distinction between insider and outsider economies does not reveal significantly different RAM levels for quarterly reporters in suspect years (F-statistic of 1.70). In untabulated regressions, we repeat the two institutional differences regressions also excluding country fixed effects. The institutional differences become even less pronounced. The adjusted R 2 of these regressions lies between 7.1% ("monitoring by analysts") and 8.3% ("operating cycle"). We use the same regression setup and a similar set of control variables as in Table 5. Overall we conclude that there are indeed significant cross-sectional differences in the effect of quarterly reporting on RAM in suspect years, with industry differences playing the most important role. Capital market characteristics are important, while the institutional differences only marginally discriminate the overall effect. These findings are important for better understanding how quarterly reporting affects RAM. While the overall effect is

31 prevalent in almost all settings, it seems to be significantly depending on industry characteristics but remains rather robust across institutional setups. [place table 7 here] V) ADDITIONAL AND SENSITIVITY ANALYSES In order to test the validity of our results, we perform an additional analysis on isolating the effect of interim reporting frequency on total RAM that exploits a pre-post setup based on a recent regulatory change. Moreover, we include a sensitivity analysis regarding our total RAM measure by comparing it to the existing real activities manipulation measures. Finally, we also test our specification regarding the matching procedure and different time periods. Pre-/Post effect of the Transparency Directive on Total RAM We perform a pre-treatment vs. post-treatment test of the adoption of a recent EU Directive, the Transparency Directive (TD), in order to analyze differences in real effects of reporting frequency before and after the adoption of the Directive. The TD aims at enhancing the level of information to investors, setting minimum standards to the publication requirements for publicly-traded companies in the member states, and improves the dissemination of information on issuers. After opposition from several member states (e.g., UK, Netherlands and Denmark), the European Commission gave up on its original plans to introduce mandatory quarterly reporting for all companies to increase transparency towards investors and analysts. However, the Directive requires all companies in member states to publish "Interim Management Statements" (IMS) after the first and third quarter of the financial year. 17 While IMS do not require the preparation of a full set of financial statements, the TD requires all listed companies on regulated markets to provide an explanation of material events and transactions that have taken place during the relevant period and their impact on the financial position of the issuer and its controlled undertakings, and a general 17 In Appendix A, we provide an example of an IMS

32 description of the financial position and performance of the issuer and its controlled undertakings during the relevant period. We argue that the level of information and disclosure provided by semi-annual reporters in particular has increased after the adoption of the TD. Companies that already issue full quarterly financial statements are not affected by the introduction of IMS as their disclosure levels go well beyond what is required by IMS. This setup allows us to perform a pre- vs. post-adoption analysis, which controls for potential weaknesses in the cross-sectional analyses because firms and countries are used as their "own" controls. Although the TD had to be officially adopted by 2007, not all member states managed to transpose the Directive into national law in time. We rely on previous research and assume the implementation dates compiled by Christensen et al. (2010). According to our hypothesis, the difference in RAM between the two groups should be lower after the introduction of the TD because semi-annual reporters have to disclose some quarterly information, which should increase management s tendency to use RAM. While this setting better accounts for potential correlated omitted variable bias than the cross-sectional analysis, there are two potential weaknesses of the analysis: First, the TD does not exclusively deal with interim statement regulation, so other aspects of the Directive could also drive the results. Second, it might be hard to see the hypothesized effect using a yearly measure (RAM) to account for pre-/post-treatment differences. We estimate the following regression for suspect years only (SUSPECT_Y): Controls + Fixedeffects + TOTALRAM = α 0 + α1td + α 2HY + α 3QR * HY + ε (7) where all variables are defined in Appendix 2. We summarize the corresponding results in table 8. We chose 3 different models to test the effect of the TD and semi-annual reporting in suspect years that vary only in the set of controls (model 1 vs. model 2) and the inclusion of fixed effects and standard error clustering

33 While models 1 and 2 use country fixed effects and country standard error clustering, model 3 uses a two-dimensional firm and year clustering. In all three models we find that both the TD and half-yearly reporting is associated with lower RAM compared to before the TD and quarterly reporters. The two-sided t-statistics vary from to for TD and to for semi-annual reporting. Of particular interest, however, is whether the joint effect is also significant. Since after the introduction of the TD semi-annual reporters need to publish Interim Management Statement and thus have higher interim disclosure levels, we except that the marginal effect is associated with higher TOTALRAM which we document in all three models. The corresponding t-statistic varies from 3.00 to The adjusted R 2 ranges from 6.3% to 8.5%. The sample sizes are comparatively small including between 405 to 518 observations. Overall, this analysis also supports our inference that a higher frequency of interim reporting is associated with more real activities manipulation. [place table 8 here] Sensitivity for different real activities manipulation measures We also perform a sensitivity test for validity of our TOTALRAM measure. Since we build this measure based on the commonly used real activities manipulations measures, i.e. abnormal cash flow from operations (ACFO), abnormal production costs (APROD), and abnormal discretionary expenses (ADISC), we also test these measures individually. The calculation of these measures is summarized in Appendix 1. Table 9 summarizes the regression results. We use the same setup as in Table 5 for the yearly suspect year definition. Overall we find that also the individual measures point in the same direction as the TOTALRAM measure. The main test of the joint effect of suspect years and quarterly reporting exhibits significantly higher real activities manipulation, with the F-statistics ranging from 4.40 (for ADISC) to (for ACFO). Also the individual coefficients for

34 suspect years and quarterly reporting have the expected positive sign and are significantly different from zero (t-statistics ranging from 1.84 to 8.63) with the exception of suspect years for ACFO, where the t-statistics of 0.79 does not reveal a significance level of at least 10%. The corresponding adjusted R² for ACFO (35.3%) and APROD (12.7%) is higher than for the previous regressions in Table 5. Summarizing, we conclude that our inferences for hypothesis H1 are also confirmed by using the individual real activities manipulation measures. [place table 9 here] Specification tests In order to further test the overall robustness of our inferences we perform a set of untabulated specification tests, in which we analyze the sensitivity of our results towards the matching procedure as well as different sample periods. One of the research design innovations in our study is the matching procedure, in which we first correct the mandatory quarterly reporters by those firms that have a high likelihood of voluntarily adopting quarterly reporting and second match the semi-annual reporters to these "pure mandatory" adopters according to industry and size. Leaving out the first matching step has the expected effect on the results presented in tables 5 and 6. Without the "pure mandatory" correction, the F-test for the yearly suspect year measure (SUSPECT_Y) does not reveal a significant inference level anymore (F-statistic of below 1.0), while the inference on the interim suspect years (SUSPECT_2Q, SUSPECT_1Q) remains strong (all F-tests significant on 1% level). Our explanation is that by leaving out the first matching, our treatment group comprises of both, firms that were mandated to report quarterly as well as firms that would have adopted quarterly reporting also in the absence of regulation. In line with expectations, this reduces the power of the tests and weakens the overall effect

35 In order to test the robustness of our second matching, the propensity score-based matching on industry and size, we also add standardized net income (NI) to control for financial performance as a further firm characteristics in the matching procedure. This results in a similar number of firm-years (2,579) for our main regressions in tables 5 and 6. The results remain unchanged although some of the inferences are slightly less significant. In another test, we split our regressions into two sub-periods, i.e., the period before ( ) and after ( ) the adoption of the Transparency Directive (TD) by the EU 18. As a consequence, the number of observations in the regressions decreases to 1,396 firm-years ( ) and 1,663 firm-years ( ). In line with our results on the impact of the TD on the semi-annual adopters as part of the additional analyses, we document quarterly reporting is associated with more RAM in suspect years also in the two sub-periods, but that this effect is more strongly pronounced in the pre-td adoption period (all F-statistics significant). After the introduction of the TD, the difference of the yearly suspect year measure (SUSPECT_Y) decreases while that for interim suspect years (SUSPECT_2Q, SUSPECT_1Q) remains similar. We conclude that in the period following the TD, the RAM difference between quarterly and semi-annual reporters has decreased because semi-annual reporters have to issue interim management statements every quarter. VI) CONCLUSION We investigate the real business effects of increased mandated reporting frequency using a sample of 15 EU countries. We find that mandatory quarterly reporting is associated with higher real activities manipulation (RAM) compared to semi-annual reporting in years when managers try to meet or beat earnings benchmarks. While institutional differences have only limited impact on this effect, RAM is particularly pronounced when monitoring by analysts is low, the shareholder investment horizon and the operating cycle are short and 18 Christensen et al. (2010) argue that the transposition date of the TD into national law differs between the countries. In the pre-post analysis, we have therefore use the country-specific dates

36 industry competition is high. We associate these effects with the increased pressure on managers to meet or beat earnings benchmarks in ever shorter time intervals. We account for country, industry and year fixed effects as well as standard errors clustered by country. Our findings are also robust to a range of sensitivity analyses The underlying setting in our study is unique because all included countries exhibit a common minimum regulatory and institutional base that minimizes deterring effects usually present in cross-country studies in international accounting (i.e., all countries use IFRS). As the reporting regulation between countries only differs in few respects, the most important of which is reporting frequency, the sample setup allows to reasonably isolate the frequencyrelated effect on our dependent variable real activities manipulation. This paper contributes to the "real effects" literature as suggested by Beyer et al. (2010) and the non-u.s. focused international accounting literature as suggested by Leuz and Wysocki (2008). We also present a new aggregate RAM measure that combines all individual RAM measures and thereby helps identifying manipulation in general. The paper also contributes to the literature on mandatory reporting presenting evidence on potential externalities of mandatory disclosure. The results provide empirical evidence in line with interview-based evidence (Graham et al., 2005) as well as concerns of many practitioners who claim that mandatory quarterly reporting might lead managers to focus on short-term goals by adversely impacting real business decisions. We conclude that the regulatory debate on interim financial disclosure thus has to consider these impacts on economic efficiency instead of focusing on information efficiency arguments only. However, our analysis does not permit us to draw conclusions on the overall desirability of mandatory versus voluntary reporting regimes. Several undisputed benefits of increased disclosure such as higher transparency towards shareholders and analysts and a reduction in information asymmetry are not considered in our study. As we investigate the effects of reporting frequency regimes solely on the firm/manager level, there are

37 probably many unobserved effects that inhibit drawing conclusion on the macro-level. This could be areas of potential further research

38 REFERENCES Aspara, J., Pajunen, K., Tikkanen, H. and R. Tainio Explaining sacrifice of long-term investments for short-term earnings: Spiral of silence in financial markets. Working Paper. Bar-Gill, O. and L. A. Bebchuck Misreporting Corporate Performance. Working paper. Bens, D. A., Nagar, V., Skinner, D. J., and M. H. F. Wong Employee stock options, EPS dilution, and stock repurchases. Journal of Accounting and Economics 36 (December): Beyer, A., Cohen, A., Lys, T. and B. Walther The financial reporting environment: Review of the recent literature. Working paper. Bhojraj, S., and R. Libby Capital Market Pressure, Disclosure Frequency-Induced Earnings/Cash Flow Conflict, and Managerial Myopia. The Accounting Review 80 (January): Biddle, G. C., Hilary, G., and R. S. Verdi How does financial reporting quality relate to investment efficiency? Journal of Accounting and Economics 48 (December): Brown, L. D., and M. L. Caylor A Temporal Analysis of Quarterly Earnings Thresholds: Propensities and Valuation Consequences. The Accounting Review 80 (April): Bushee, B. J The influence of institutional investors on myopic R&D investment behavior. The Accounting Review 73 (July): Butler, M., Kraft, A., and I. S. Weiss The effect of reporting frequency on the timeliness of earnings: The cases of voluntary and mandatory interim reports. Journal of Accounting and Economics 43 (July): Chhaochharia, V., and Y. Grinstein, Corporate Governance and Firm Value: The Impact of the 2002 Governance Rules. Journal of Finance 62, Christensen, H. B., Hail, L., and Leuz, C. (2010). Capital-Market Effects of Securities Regulation: The Role of Implementation and Enforcement. Working Paper. Cohen, D., Dey, A., and T. Lys Real and accrual-based earnings management in the pre and post-sarbanes-oxley periods. The Accounting Review 83 (May): Committee of European Securities Regulators (CESR) Summary of responses to Questionnaire on Transposition of the Transparency Directive. Ref: CESR/08-514b, September Daske, H., Hail, L., Leuz, C. and R. Verdi Mandatory IFRS Reporting around the World: Early Evidence on the Economic Consequences. Journal of Accounting Research, 46: Das, S., Shroff, P. K., and H. Zhang Quarterly Earnings Patterns and Earnings Management. Contemporary Accounting Research 26 (Fall): Dechow, P. M., and R. Sloan Executive incentives and the horizon problem: an empirical investigation. Journal of Accounting and Economics 14 (March): , Kothari, S. P., and R. L. Watts The relation between earnings and cash flows. Journal of Accounting and Economics 25 (May): , and I. Dichev The Quality of Accruals and Earnings: The Role of Accrual Estimation Errors. The Accounting Review, 77 (Supplement): Degeorge, F., Patel, J., and R. Zeckhauser Earnings management to exceed thresholds. Journal of Business 72 (January): De Long, J. B., Shleifer, A., Summers, L. H., and R. J. Waldmann: Noise Trader Risk in Financial Markets. The Journal of Political Economy Vol. 98, No. 4 (Aug., 1990): Dichev, I., and D. J. Skinner Large-sample evidence on the debt covenant hypothesis. Journal of Accounting Research 40 (September): Ellis, D. Capital: The Story of Long-Term Investment Excellence. Hoboken, N.J.: John Wiley & Sons, Ernst & Young Interim management statements achieving comparability in Europe? July Ewert, R., and A. Wagenhofer Economic Effects of Tightening Accounting Standards to Restrict Earnings Management. The Accounting Review 43 (October): Fan, Y., Barua, A., Cready, W. M., and W. B. Thomas Managing Earnings Using Classification Shifting: Evidence from Quarterly Special Items. The Accounting Review 85 (July):

39 Fudenberg, D., and J. Tirole A theory of income and dividend smoothing based on incumbency rents. Journal of Political Economy 103 (February): Graham, J. R., Harvey, C. R., and S. Rajgopal The economic implications of corporate financial reporting. Journal of Accounting and Economics 40 (December): Hall, B. J. and K. J. Murphy The trouble with stock options. Working paper. Healy, P. M., and J. M. Wahlen A review of the earnings management literature and its implications for standard setting. Accounting Horizons 13 (December): Jensen Michael, The Takeover Controversy: Analysis and Evidence, Midland Corporate Finance Journal, IW (19m): Jones, J. J Earnings management during import relief investigations. Journal of Accounting Research, 29 (Autumn): Kothari, S. P., Leoneb, A. J. and C. E. Wasley Performance matched discretionary accrual measures, Journal of Accounting and Economics, 33 (February): Laffont, J. J., and J. Tirole Managerial Switching and Myopia, Working Paper Lara, J. M. G., Osma, G. B., and B. G. A. Noguer Effects of database choice on international accounting research. Abacus 42 (September/December): Leuz, C Different Approaches to Corporate Reporting Regulation: How Jurisdictions Differ and Why. Working Paper, Working Paper, University of Chicago., Nanda, D., and P. D. Wysocki Investor Protection and Earnings Management: An International Comparison. Journal of Financial Economics 69 (September): , and P. D. Wysocki Economic consequences of financial reporting and disclosure regulation: A review and suggestions for future research. Working Paper. Mazars Transparency Directive Assessment Report. June McVay, S Earnings management using classification shifting: An examination of core earnings and special items. The Accounting Review 81 (May): Roychowdhury, S Earnings management through real activities manipulation. Journal of Accounting and Economics 42 (December): Shleifer, A. and D. Wolfenzon Investor protection and equity markets. Journal of Financial Economics 66 (October): Sloan, R.G Do Stock Prices Fully Reflect Information in Cash Flows and Accruals About Future Earnings? The Accounting Review (July): Stein, J. C Takeover Threats and Managerial Myopia. Journal of Political Economy 96 (February): Stein, J. C Efficient capital markets, inefficient firms: A model of myopic corporate behavior. The Quarterly Journal of Economics 104 (November): Tassell, T Brussels gives quarter on data frequency trends. The Financial Times (November). Teoh, S. H., Welch, I., and T. J. Wong Earnings Management and the Long-Run Market Performance of Initial Public Offerings. Journal of Finance 53 (December): Verdi, R. S Financial Reporting Quality and Investment Efficiency. Working Paper, MIT. Yu, F Analyst coverage and earnings management. Journal of Financial Economics 88 (May): Verrechia, R. E Essays on disclosure. Journal of Accounting and Economics 32 (December): Watts, R.L. and J.L. Zimmerman Positive accounting theory, Prentice-Hall, Englewood Cliffs, NJ. Zang, A. Y Evidence on the trade-off between real activities manipulation and accrual-based earnings management. Working Paper. Zhang, X Economic Consequences of the Sarbanes-Oxley Act of Journal of Accounting and Economics 44:

40 Appendix 1 Roychowdhury (2006) presents three measures to proxy for real activities manipulation (RAM), ACFO, APROD and ADISC. While ACFO consolidates the effect of different earnings manipulation activities like sales manipulation, overproduction and reduction of discretionary expenses, APROD accounts for overproduction and ADISC for an abnormal reduction in discretionary expenses. Following Roychowdhury (2006), sales manipulation is defined as the acceleration of current period sales through price discounts and more lenient credit terms. Price discounts are often used to pull part of next-period sales into the current period to temporarily boost (absolute) earnings. As a percentage of sales, however, ACFO will decline due to declining margins as a consequence of price discounts. A second option to boost current period sales is offering more lenient credit terms. Many manufacturing companies (e.g., automotive manufacturers) offer lower interest rates towards the end of the fiscal year for financing their products. This is similar to price discounts and leads to reduced cash flow at given sales levels. As a result of sales manipulation, we expect current period ACFO to be negative. We define overproduction as managers' tendency to increase production above the necessary level to reduce cost of goods sold (COGS) per unit by spreading fixed cost over a larger amount of units (under the assumption that the increase in marginal cost does not offset this effect). However, the firm incurs additional production and holding costs that are not matched with corresponding sales. 19 This results in higher production costs and we expect abnormally high production cost for a given level of sales. This has a negative effect on cash flow from operations at given sales levels. We define discretionary expenses as the sum of R&D, advertising, and SG&A expenses. In order to increase earnings, managers can reduce discretionary expenses because a reduction does not immediately affect sales. If managers decrease discretionary expenses to meet earnings targets, they will experience abnormally high earnings in the respective period. We therefore expect abnormally low discretionary expenses for a given level of sales. This, in turn, increases cash flow from operations. To estimate the "normal levels" of the dependent variables, we run the following crosssectional regressions for every industry and year. In line with Roychowdhury (2006), we require at least 15 firm-year observations per 2-digit SIC industry classification group for each measure: 20 CFOit 1 SALES it SALES it = a + a + a + a + ε (8) it TAi, t 1 TAi, t 1 TAi, t 1 TAi, t 1 where all variables are defined in Appendix B. 21 ACFO is defined as the actual cash flow from operations minus the "normal" cash flow from operations calculated using the estimated coefficients above. The residuals of equation (1) are used as dependent variable in the regression models discussed in this section to test the influence of quarterly reporting. In line with the aforementioned reasoning, we expect ACFO to be more negative for quarterly reporters. Production costs are defined as the sum of cost of goods sold (COGS) and change in inventory in the respective year. Following Roychowdhury (2006) and Cohen et al. (2008), we calculate normal production cost using the following regression: Pr odit 1 Salesit Sales Sales it i, t 1 = a0 + a1 + a2 + a2 + a3 + εit (9) TA TA TA TA TA i, t 1 i, t 1 i, t 1 i, t 1 i, t 1 19 As stated by Roychowdhury (2006), managers only raise production if the benefits from decreasing average COGS are higher than the additional holding costs incurred in the respective period. 20 We also perform a sensitivity test using 30 firm-year observations per industry and a one-digit SIC code. Our results are robust for this specification. 21 In accordance with Roychowdhury (2006), we employ not only a scaled intercept when estimating nondiscretionary accruals (as is general convention in the literature), but also an unscaled intercept to ensure that the mean abnormal value for every industry year is zero

41 where all variables are defined in Appendix B. As outlined above, we expect production costs to be abnormally high in the current period with in turn reduces cash flow from operations as a function of sales. Discretionary expenses include SG&A, R&D and advertising expenses. As advertising expenses are not available for European companies on Datatastream/Worldscope, we calculate the measure using SG&A and R&D figures only. In line with Roychowdhury, we calculate discretionary expenses as long as SG&A is available. If R&D is not available, it is set to zero. We model the normal level of discretionary expenses as follows: DISC TA = a + a + a Sales it i, t i, t 1 TAi, t 1 TAi, t 1 where all variables are defined in Appendix B. We expect discretionary expenses to be abnormally low, which in turn reduces cash flow from operations as a function of sales. 1 + ε it (10)

42 Variable Definition (with Datastream/Worldscope data items) Dependent variable TOTALRAM Total real activities manipulation, calculated based on equation (XXX). ACFO Abnormal cash flow from operations, calculated based on equation (xxx). APROD Abnormal production cost, calculated based on equation (xxx). ADISC Abnormal discretionary expenses, calculated based on equation (xxx). Experimental variables QR Quarterly reporting (Y/N), derived from the earnings reporting frequency (WC05200). TD SY_Y SY_2Q SY_1Q Regulation (Low = 1) Enforcement Quality (Low = 1) Shareholder horizon (Short = 1) Monitoring by analysts (Low = 1) Industry competition (Strong = 1) Operating cycle (Short = 1) Appendix 2: Definition of Variables Transparency Directive (Y/N), with value of 1 for firm-years after the introduction by country, as collected by Christensen et al. (2010). Yearly suspect years, with value of 1 if yearly standardized net income before extraordinary items (wc01551/lagged wc02999) is greater than zero and smaller than 0.01 or yearly standardized delta net income greater than zero and smaller than Interim suspect years, with value of 1 if interim standardized net income before extraordinary items (wc01551/lagged wc02999) is greater than zero and smaller than (HY)/ (QR) or yearly standardized delta net income greater than zero and smaller than (HY)/ (QR). It is further restricted to only those cases where there is exactly one HY observations per year and exactly 2 QR observations per year. Interim suspect years, with value of 1 if interim standardized net income before extraordinary items (wc01551/lagged wc02999) is greater than zero and smaller than (HY)/ (QR) or yearly standardized delta net income greater than zero and smaller than (HY)/ (QR). It is further restricted to only those cases where there is exactly one HY observations per year and exactly 1 QR observations per year. Regulation country cluster (Low/High), with value of 1 for countries from low regulation clusters 3 and 4 from the Leuz (2010) "Institutional Clusters Around the World (k05). Enforcement quality (Low/High) is proxied by the Regulation Quality score from Kaufmann (2010), with value of 1 for countries with below EU-15 median scores. Shareholder horizon (Short/Long) proxied by stock turnover (VO / NOSH), with value of 1 for firms with above median performance. Abnormal analyst following (Low/High), with value of 1 for abnormal number of analysts following below median, and 0 else. Abnormal number of analysts following is the residuum from the regression of number of analysts following (IBES) on MVE, BTM, MKTSHARE, and SHARES. Industry competition (Strong/Weak) proxied by the Herfindahl index, with value of 1 for firms from industry-years with below median score. Operating cycle (Short/Long), with value of 1 for firm-years with below median OPCYC. Continued on next page

43 Appendix 2: Definition of Variables (ctd.) Control variables SIZE Size, defined as natural logarithm of total assets (WC02999) LEV Accounting leverage, calculated as total current liabilities (WC03101), divided by total assets (WC02999) NI Standardized net income, defined as net income available to common equity (WC01751), divided by lagged total assets (WC02999) CLHELD Ownership structure, defined as percentage of closely held shares (WC08021). SHARES Common shares outstanding (WC05301). ANFOL Natural logarithm of one plus analyst following. Analyst following is the total number of estimates in the mean associated with the FY1 forecast (N1NE, I/B/E/S). MKTSHARE Market share which equals company SALES divided by total industry SALES. LITIND High litigation industries, with SIC in [2833:2836, 8731:8734, 7371:7379, 3570:3577, 3600:3674]. OPCYC Operating cycle, calculated as receivables (WC02051) divided by sales (WC01001) plus inventory (WC02101) divided by COGS (WC01051), both divided by 360 NOA Balance sheet constraint, calculated as net operating assets divided by SALES at the beginning of the year. Net operating assets is calculated as common equity (wc03501) minus cash & cash equivalents (wc02001) plus the sum of short-term liabilities (wc03051) and long-term debt (wc03251). BIG4 Big 4 auditor, with value of 1 if auditor of the year is Deloitte, Ernst&Young, KPMG, PWC using data from Amadeus (Bureau van Dijk). Other variables BTM Book-to-market value of equity, calculated as common equity (WC03501), divided by market value of equity (MVE) CFO Standardized cash flow from operations (CFO), defined as CFO (WC04860), divided by lagged total assets (WC02999) CONST Constant, defined as 1 over lagged total assets (WC02999) DISC Discretionary expenses, calculated as the sum of R&D (WC01201) and SG&A (WC01101) expenses, divided by lagged total assets (WC02999). DIV Dummy for dividends, with value of 1 for firm-years in which cash dividends were paid (WC04551) in previous year FORSALES Foreign sales in percent of overall sales (WC08731). GEOSALES Spread of sales among geographies, calculated as squared sales by geographies (WC19601, WC19611,... WC19691) divided by squared total SALES. HERF Industry concentration (Herfindahl Index), calculated as sum of squared market shares (based on annual sales) in the firm's industry sector (defined by 2-digit SIC code) per year INDEX Dummy for index membership, with value of 1 for firms that are members of stock indexes (WC05661) INDSIZE Natural logarithm of sum of SALES per industry. MVE Market value of equity, calculated as number of shares outstanding (nosh) times share price at fiscal year end (WC05001) PROD Standardized production cost, calculated as cost of goods sold (WC01051) plus change in inventory (WC02101), divided by lagged total assets (WC02999) ROA Return on assets (WC08326) SALES Standardized sales, defined as total sales (WC01001), divided by lagged total assets (WC02999) SIZE Size, defined as natural logarithm of market value of equity TA Total assets (WC02999) TOPIND Dummy for top industry quartile, with value of 1 if firm-year in top quartile of operating income (WC01250) divided by sales (WC01001) VOL Volume traded (VO)

44 Table 1: Quarterly Reporting Environment in EU-15 Countries Country Quarterly Reporting Mandatory? Quarterly Reporting Rules Regulator Stock Market Austria (AT) Yes (A) Austrian stock market (Wiener Börse) requires firms Finanzmarktaufsicht Wiener Börse listed in "Prime Market" segment to publish full Quarterly Reports (in acc. with IAS 34) (FMA) Belgium (BE) No Listed firms are not required to publish quarterly earnings Commission Euronext Brussels Bancaire, Financière et des Assurances (CBFA) Denmark (DK) No Listed firms are not required to publish quarterly earnings (However, before Copenhagen Stock Exchange was taken over first by OMX in 2005 and then later by NASDAQ in 2008, firms were incentivized by the stock exchange to publish quarterly reports; this leads to a still high QR practice in Denmark today) Finanstilsynet (Danish FSA) NASDAQ OMX Copenhagen Finland (FI) Yes (B) The Finish Securities Market Act still requires listed Finanssivalvonta companies to present interim results for the first three, six (FIVA) and nine months of the financial period (chapter 2, section 5); there are only very limited exceptions France (FR) No Listed firms are not required to publish quarterly earnings (However, the French Monetary and Financial Code requires listed firms to publish quarterly net sales by subsidiaries) Autorité des Marchés Financiers (AMF) NASDAQ OMX Helsinki Euronext Paris Germany (DE) Yes (A) German stock market operator Deutsche Börse requires firms listed in "Prime Standard" segment to publish full Quarterly Reports (in acc. with IAS 34) Great Britain (GB) Bundesanstalt für Finanzdienstleistungsaufsicht (BaFin) No Listed firms are not required to publish quarterly earnings Financial Services Authority (FSA) Frankfurter Wertpapierbörse NYSE Euronext London Greece (GR) Yes (B) Greek regulation authority (HCMC) principally requires all Hellenic Capital listed firms to publish quarterly reports (with limited exceptions) Market Commission (HCMC) Ireland (IE) No Listed firms are not required to publish quarterly earnings Irish Financial Services Regulatory Authority (IFSRA) Italy (IT) Yes (B) Italian regulation authority (CONSOB) principally requires all listed firms to publish quarterly reports (with limited exceptions) Luxembourg (LU) The Netherlands (NL) Commissione Nazionale per le Società e la Borsa (CONSOB) No Listed firms are not required to publish quarterly earnings Commission de Surveillance du Secteur Financier (CSSF) No Listed firms are not required to publish quarterly earnings Autoriteit Financiële Markten (AFM) Portugal (PT) Yes (B) Portuguese regulation authority (CMVM) principally requires all listed firms to either publish full quarterly reports (in acc. with IAS 34) or specific quarterly financials (including quarterly earnings) Spain (ES) Yes (B) Spanish regulation authority (CNMV) principally requires all listed firms to publish specific quarterly financials (including quarterly earnings) Sweden (SE) Yes (C) Under the listing rules for the two regulated markets in Sweden, listed firms are required to publish full quarterly reports (in acc. with IAS 34) Athens Exchange Securities Market Irish Stock Exchange Borsa Italiana Bourse de Luxembourg Euronext Amsterdam Comissão do Euronext Lisbon Mercado de Valores Mobiliários (CMVM) Comisión Nacional del Mercado de Valores (CNMV) Finansinspektionen (Swedish FSA) Bolsa de Madrid NASDAQ OMX Stockholm This tables presents the different quarterly reporting regulation by country in the EU-15 countries. Yes (A) indicates required by stock exchange segment, Yes (B) indicated required by regulator, and Yes (C) indicates required by stock exchange

45 Table 2: Sample composition Step Starting sample: EU-15 regulated markets, Firm-years 16,517 Deleting financial firms (SIC )./. 3,798 Deleting firms from highly regulated industries (SIC )./. 1,612 Deleting pubic administrative firms (SIC > 9000)./. 24 Deleting firm-years in distress (Common Equity < 0)./ ,423 Deleting firms cross-listed in the U.S../. 830 Deleting VOL firms listed on MAND exchanges./. 63 Deleting non-quarterly obs. in MAND countries./ ,412 Deleting "non-pure" MAND firms./. 234 Deleting VOL firms./. 816 Deleting non-matched firms./ ,372 This tables presents the sample composition process. We use a sample of EU-15 firms listed on regulated markets from

46 Panel A: Distribution by country Table 3: Sample overview Country QR HY Total Austria (AT) Belgium (BE) Denmark (DK) Finland (FI) France (FR) 0 1,333 1,333 Germany (DE) ,133 Great Britain (GB) 0 1,506 1,506 Greece (GR) Ireland (IE) Italy (IT) Luxembourg (LU) The Netherlands (NL) Portugal (PT) Spain (ES) Sweden (SE) Total 3,686 3,686 7,372 Panel B: Distribution by suspect observations Suspect Non-Suspect Suspect definitions HY QR HY QR Total Yearly data [SY_Y] ,079 3,113 7,372 Interim data 1x per year ,702 2x per year x per year x per year Allocation for basic interim test: 2 quarters vs. 1 half-year [SY_2Q] ,057 3,621 7,372 Allocation for strict interim test: 1 quarter vs. 1 half-year [SY_1Q] ,057 2,875 7,372 This table presents a overview of the sample by country (Panel A) and by different suspect year definitions (Panel B). Variable definitions are in the Appendix

47 Panel A: Descriptive Statistics Variable Obs. Mean Std. Dev. 1% 25% Median 75% 99% TOTALRAM 3, If QR= 1 If QR= 0 If SUSPECT_Y = 1 If SUSPECT_2Q = 1 If SUSPECT_1Q = 1 If SUSPECT_Y = 0 If SUSPECT_2Q = 0 If SUSPECT_1Q = 0 Table 4: Descriptive statistics and correlations for continuous variables 1, , , , , SIZE 7, LEV 7, NI 6, CLHELD 5, SHARES 7,367 99, , ,875 28,536 97,117 1,142,371 ANFOL 7, MKTSHARE 7, OPCYC 6, NOA 6, Panel B: Pearson Correlations Matrix Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) TOTALRAM (1) 1 SIZE (2) LEV (3) NI (4) CLHELD (5) SHARES (6) ANFOL (7) MKTSHARE (8) OPCYC (9) NOA (10) This table presents descriptive statistics (Panel A) as well as Pearson correlation coefficients (Panel B) for continuous variables. Variable definitions are in the Appendix. All continuous variables are winsorized at 1st and 99th percentile

48 Basic Strict Yearly SY interim SY interim SY Experimental Variables Exp.Sgn. [SUSPECT_Y] [SUSPECT_2Q] [SUSPECT_1Q] Suspect Year [SUSPECT_*] *** ** ** (3.19) (2.39) (2.39) QR (1) *** *** ** (3.18) (3.27) (2.73) Suspect Year * QR (2) (0.52) (-0.01) (0.01) QR effect in Suspect Year: (1) + (2) *** *** *** [8.02] [14.72] [9.44] Control Variables SIZE *** *** *** (5.49) (5.50) (5.40) LEV (0.80) (0.82) (0.81) NI ** ** ** (-2.68) (-2.63) (-2.64) CLHELD ** ** ** (-2.81) (-2.89) (-2.89) SHARES +/ (-1.28) (-1.20) (-1.19) ANFOL *** *** *** (-4.14) (-4.33) (-4.28) MKTSHARE +/ (-0.76) (-0.76) (-0.72) LITIND +/ (-0.33) (-0.35) (-0.36) OPCYC (-0.74) (-0.79) (-0.79) NOA * (1.76) (1.77) (1.79) BIG4 +/ *** *** *** (-3.99) (-4.09) (-3.98) CONST +/ *** *** *** (-4.84) (-4.85) (-4.74) Country Fixed Effects Country SE clustering Table 5: Regression of QR and different Suspect Years on Total RAM Number of observations Adjusted R 2 2,557 2,557 2, % 8.4% 8.5% This table presents results from regression analyses examining the effect of QR and different Suspect Years on Total RAM. The estimation is a fixed-effects panel regression with clustered standard errors. It is of the following form: Variable definitions are in the Appendix. All variables are winsorized at 1st and 99th percentile. Amounts reported are regression coefficients with t-statistics in (parantheses) and F-statistics in [brackets]. *, **, *** indicates significance at the 0.10, 0.05 and 0.01 levels, respectively, using a two-tailed (t-statistics) and onetailed (F-statistics) test yes yes yes yes yes yes Controls + Fixedeffec ts + TOTALRAM = α + α SUSPECTX + QR + QR * SUSPECTX + ε 0 1

49 Table 6: Regression of QR and different Suspect Years combinations on Total RAM Basic Strict interim SY interim SY Experimental Variables Exp.Sgn. [SY_2Q] [SY_1Q] Suspect Year, Interim [SY_2Q] ** (2.81) Suspect Year, Interim [SY_1Q] ** (2.83) Suspect Year, Yearly [SY_Y] *** *** (3.97) (3.95) QR (1) *** ** (3.30) (2.47) Suspecy Year, Interim * QR (2) (-0.07) (0.11) Suspecy Year, Yearly * QR (3) (-0.13) (0.73) QR effect in Interim SY: (1) + (2) *** *** [16.36] [9.02] QR effect in Yearly SY: (1) + (3) *** *** [7.62] [7.20] Control Variables SIZE *** *** (5.45) (5.31) LEV (0.80) (0.78) NI ** ** (-2.66) (-2.68) CLHELD ** ** (-2.92) (-2.91) SHARES +/ (-1.26) (-1.26) ANFOL +/ *** *** (-4.22) (-4.16) MKTSHARE +/ (-0.76) (-0.70) LITIND +/ (-0.32) (-0.32) OPCYC (-0.75) (-0.74) NOA * * (1.80) (1.82) BIG4 +/ *** *** (-3.97) (-3.77) CONST +/ *** *** (-4.88) (-4.74) Country Fixed Effects Country SE clustering Number of observations Adjusted R 2 This table presents results from regression analyses examining the effect of QR and different Suspect Year combinations between Yearly SY and Inteirm SYs on Total RAM. The estimation is a fixed-effects panel regression with clustered standard errors. It is of the following form: α QR * SUSPECT _ XQ + Controls + Fixedeffects + Variable definitions are in the Appendix. All variables are winsorized at 1st and 99th percentile. Amounts reported are regression coefficients with t-statistics in (parantheses) and F-statistics in [brackets]. *, **, *** indicates significance at the 0.10, 0.05 and 0.01 levels, respectively, using a two-tailed (t-statistics) and one-tailed (F-statistics) test yes yes yes yes 2,557 2, % 8.8% TOTALRAM = α + α SUSPECT _ Y + α SUSPECT _ XQ + α QR + α QR* SUSPECT _ Y ε 3 4

50 Table 7: Regression of Cross-sectional differences of Quarterly Reporting in Suspect Firm-Years on Total RAM Experimental variables: Regulation Enforcement Shareholder Monitoring by Industry Operating Quality horizon analysts competition cycle Exp.Sgn. (Low = 1) (Low = 1) (Short = 1) (Low = 1) (Strong = 1) (Short = 1) SY * QR * Exp. Variable = 1 (1) *** *** ** *** *** *** (3.44) (4.78) (2.35) (3.88) (4.08) (4.95) SY * QR * Exp. Variable = 0 (2) (1.17) (0.03) (0.46) (0.32) (-0.37) (0.13) Overall SY * QR Effect: (1) + (2) *** *** *** *** *** [12.03] [1.43] [9.62] [38.10] [14.28] [14.93] Cross-sect. Difference: (1) - (2) * * ** ** *** [1.70] [2.02] [2.48] [3.37] [6.23] [12.70] Control Variables Country Fixed Effects Country SE clustering Number of observations Adjusted R 2 Institutional diff. Capital market diff. Industry diff. yes yes yes 1) yes 2) yes 3) yes 4) yes yes yes yes yes yes 2,557 2,557 2,557 2,557 2,557 2, % 8.2% 8.1% 7.1% 8.2% 8.3% This table presents results from regression analyses examining the joint effect of cross-sectional differences and quarterly reporting in suspect firm-years on Total RAM. The estimation is a fixed-effects panel regression with clustered standard errors. It is of the following form: TOTALRAM = α1 SUSPECT _ Y * QR * CROSS1 + α 2SUSPECT _ Y * QR * CROSS0 + Controls + Fixedeffects + ε We use the same set of standard control variables as in table 5, with the following adjustements: 1) w/o COMSHARES, 2) w/o ANFOL, 3) w/o MKTSHARE, 4) w/o OPCYC. Variable definitions are in the Appendix. All variables are winsorized at 1st and 99th percentile. Amounts reported are regression coefficients with t- statistics in (parantheses) and F-statistics in [brackets]. *, **, *** indicates significance at the 0.10, 0.05 and 0.01 levels, respectively, using a two-tailed (tstatistics) and one-tailed (F-statistics) test. yes yes yes yes yes yes

51 Table 8: Pre-/Post Regressions for Transparency Directive on Total RAM in Suspect Years Experimental Variables Exp.Sgn. [Model 1] [Model 2] [Model 3] Transparency Directive (Y/N) *** ** * (-3.29) (-2.95) (-1.71) HY (Y/N) ** ** * (-2.55) (-2.46) (-1.74) TD * HY *** *** *** (3.65) (4.31) (3.00) Control Variables SIZE ** ** (2.46) (1.31) (2.06) LEV ** (2.74) (1.16) NI ** (-1.73) (-2.54) (-0.92) CLHELD (-0.96) (-0.57) MKTSHARE +/ (-1.39) (-1.23) NOA (1.52) (-0.35) (0.76) BIG4 +/ * *** (-2.16) (-3.67) (-1.26) CONST +/ * (-1.78) (-0.25) (-0.91) Country Fixed Effects Country SE clustering Firm/year SE clustering Number of observations Adjusted R 2 yes yes no yes yes no no no yes % 6.3% 7.2% This table presents results from regression analyses examining the Pre-/Post effect of the Transparency Directive on Total RAM in Suspect Years. The estimation is a fixed-effects panel regression with clustered standard errors or a OLS panel regression with 2-dimensional standard error clusters, respectively. It is of the following form: Controls + Fixedeffects + TOTALRAM = α + α TD + α HY + α QR * HY + ε Compared to table 5, we account for the reduced sample size by excluding the least significant controls (SHARES, ANFOL, LITIND, OPCYC). Variable definitions are in the Appendix. All variables are winsorized at 1st and 99th percentile. Amounts reported are regression coefficients with t-statistics in (parantheses) and F-statistics in [brackets]. *, **, *** indicates significance at the 0.10, 0.05 and 0.01 levels, respectively, using a two-tailed (t-statistics) and one-tailed (Fstatistics) test

52 Table 9: Regression of QR and Suspect Years on ACFO, APROD and ADISC Experimental Variables Exp.Sgn. ACFO APROD ADISC Suspect Year [SY_Y] ** * (0.79) (2.67) (1.84) QR (1) *** *** * (8.63) (5.54) (1.93) Suspecy Year * QR (2) (-0.39) (0.75) (1.15) QR effect in SY: (1) + (2) *** *** ** [17.24] [15.44] [4.40] Control Variables SIZE *** (-1.75) (1.73) (5.49) LEV *** *** (6.22) (3.32) (0.07) NI *** *** (-9.09) (-8.44) (-0.50) CLHELD ** (-1.20) (-1.30) (-2.55) SHARES +/ ** (2.58) (1.13) (-1.39) ANFOL +/ *** *** *** (-3.95) (-5.32) (-3.66) MKTSHARE +/ ** *** (2.56) (-5.47) (-0.97) LITIND +/ (-0.53) (-1.05) (-0.32) OPCYC * (1.51) (-1.92) (-1.01) NOA * (-1.03) (-0.60) (2.04) BIG4 +/ *** ** ** (-4.82) (-2.55) (-3.03) CONST +/ * *** (2.02) (-1.62) (-4.82) Country Fixed Effects Country SE clustering Number of observations Adjusted R 2 yes yes yes yes yes yes 4,608 4,287 2, % 12.7% 7.0% This table presents results from regression analyses examining the effect of QR and Suspect Years on ACFO, APROD and ADISC. The estimation is a fixed-effects panel regression with clustered standard errors. It is of the following form: Variable definitions are in the Appendix. All variables are winsorized at 1st and 99th percentile. Amounts reported are regression coefficients with t-statistics in (parantheses) and F-statistics in [brackets]. *, **, *** indicates significance at the 0.10, 0.05 and 0.01 levels, respectively, using a two-tailed (t-statistics) and one-tailed (F-statistics) test Controls + Fixedeffects + Measure = α + α SUSPECTX + QR + QR * SUSPECTX + ε 0 1