Capital market development and the (perceived) strength of financial auditing and reporting standards

Transcription

1 Capital market development and the (perceived) strength of financial auditing and reporting standards Henry L. Friedman Anderson School of Management University of California, Los Angeles 110 Westwood Plaza, D4.02 Los Angeles, CA May 2015 Early version Comments welcome I thank David Aboody, Jack Hughes, Sarah McVay, Bugra Ozel, Suhas Sridharan and participants at the 2014 UCLA Spring Accounting Mini Conference. I also thank UCLA s Anderson School of Management, Center for Global Management, and Fink Center for Finance and Investments for financial support. Amalia Merino, Tiffany Park, and Justin Wang provided valuable research assistance.

2 Capital market development and the (perceived) strength of financial auditing and reporting standards Abstract: This paper examines the effects of financial auditing and reporting standards on capital market development using a country-year measure based on executive perceptions. Unlike frequently-used country-level measures with limited coverage, the measure used here is available for over 1,400 country-years, allowing for identification based on both within- and across-country variation, and is distinct from other country-year measures based on IFRS adoption or firm-year earnings quality proxies. In a panel of over 140 countries from , I find that stronger auditing and reporting standards are positively associated with subsequent levels of equity market development. Additionally, public trust in politicians and managers complements stronger auditing and reporting standards in facilitating both equity and credit market development. JEL Codes: E22, G15, G18, G38, M41, M42, M48 Keywords: Financial auditing standards, financial reporting standards, capital market development, executive perceptions

3 1 Introduction Every year, the World Economic Forum administers their Executive Opinion Survey (EOS) to thousands of executives worldwide. Since 2002, their survey has asked executives to rate the strength of financial auditing and reporting standards in their home country. This study uses their answers and real-world data from numerous sources to address the question of whether better financial auditing and reporting standards, as measured by the perceptions of industry leaders, help promote capital market development reflected in larger and more active stock markets, private firms borrowing more, and banks increased lending to firms. 1 Using additional data from the World Economic Forum, I examine how trust in politicians and executives moderates the relation between financial auditing/reporting standards and capital market development. The primary motivation for the study is that prior empirical results on the relation between financial auditing/reporting standards and capital market development may suffer from identification problems. Prior studies used measures based on cross-sectional differences (e.g., the CIFAR index), specific institutional changes (e.g., the introduction of SOX or adoption of IFRS), or firm-level auditor choices. 2 Inferences based on cross-sectional differences and institutional changes are potentially confounded by cross-sectional variation in governance norms or contemporaneous institutional changes (Christensen, Hail, and Leuz 2013). The EOSderived perceived strength of auditing and reporting standards (PSARS) measure used here, in contrast, is available annually for a broad panel of countries from 2002 through 2013, allowing 1 A potential benefit of the measure used here is that it relies on industry leaders perceptions. Perceptions are a necessary mediator if information quality is to affect markets through agents economic actions, since their actions are guided by beliefs. For example, for risk related to information quality to be priced, the marginal investor must perceive variation in this quality across firms and use this belief about variation as an input in her financial decisions. This is not necessarily a behavioral (i.e., non-bayesian) explanation as beliefs can be based on rational Bayesian updating. Additionally, survey respondents perceptions of standards strength is likely to incorporate enforcement and the standards in effect rather than just the standards as written. 2 Prior studies have used the following cross-sectional proxies: the CIFAR index (e.g., La Porta et al. 1998); the PWC Opacity Index (Gelos and Wei 2005); the S&P Transparency scores (Khanna, Palepu, and Srinivasan 2004); and disclosure measures based on disclosures in prospectuses (La Porta, Lopez de Silanes, and Shleifer 2006). Studies exploiting the adoption of IFRS include Leuz and Verrecchia (2000) and Daske et al. (2008). Lang, Lins, and Maffett (2012) use auditor choice as one of several proxies for firm-level transparency. 1

4 for both within- and across-country identification not limited to windows around potentiallyconfounded institutional changes. Inferences based on auditor-level choices are subject to endogeneity concerns because firms select their auditors and the selection likely depends on other institutional features. The measure used in this study, based on executive perceptions of their surroundings rather than firms optimizing choices, mitigating concerns about selectionbased endogeneity, as managers do not select their beliefs in the same way that firms select their auditors. Preliminary analyses show that PSARS is significantly correlated in the cross-section with existing and oft-used purely cross-sectional measures of accounting and disclosure quality, including CIFAR index (e.g., La Porta et al. 1998), the PWC Opacity Index (Gelos and Wei 2005), S&P Transparency scores (Khanna et al. 2004), and disclosure measures based on disclosures in prospectuses (La Porta et al. 2006). The significant correlations between PSARS and these measures, and the use of PSARS as a disclosure quality proxy in prior studies helps establish construct validity (e.g., Fernandes and Ferreira 2009; Gelos and Wei 2005; Jin and Myers 2006). At the country-year level, PSARS is also significantly correlated with the probability that firms use high-quality auditors, the probability that firms use high-quality reporting standards (i.e., US GAAP or IFRS), and average analyst following, but not with accruals-based measures of earnings quality. The correlations pattern suggests that PSARS captures auditing and reporting standards quality, but is distinct from earnings-based measures of financial reporting quality, transparency, or opacity. The main tests focus on associations between PSARS and capital market properties related to aggregate development, including the size of equity and debt markets and trading activity relative to the size of the national economy, while controlling for identified institutional factors that are associated with variation in PSARS. In the main analyses, I use panel-based estimations to investigate the potential effects of PSARS on equity and debt market development. In regressions of indicators of stock market development on lagged PSARS, I find that PSARS has significant explanatory power over and above macroeconomic and capital market factors as well as country and time indicators. Specifically, lagged PSARS is positively associated with stock market capitalization and annual trading volume relative to GDP. In tests focused on credit market development, measured using the amount of credit provided to private entities or the amount of credit provided by banks, I find weak evidence for associations between PSARS and 2

5 future levels of credit market development, but these associations are not robust to controls for country effects or existing credit market development. The evidence suggests that stronger auditing and reporting standards facilitate the development of equity markets, but not the development of credit and lending markets. This is consistent with the idea that relational creditors like banks rely on private information (Rajan and Zingales 1998), which is less likely to be affected by financial auditing and reporting standards that improve public information. Additional tests allow the association between PSARS and subsequent capital market development to depend on existing levels of trust in politicians, perceptions of the ethical behavior of firms, and trust in general. I focus on these because trust in politicians, firms, or in general should facilitate the effects of PSARS, because market participants will be less likely to believe that the system is rigged (due to unscrupulous politicians and bureaucrats) or that financial reports are faulty (due to unethical managers). I find that in most specifications the positive associations between PSARS and subsequent capital market development are concentrated in regimes with higher trust in politicians and managers, and that the associations between PSARS and subsequent capital market development tend to be stronger when public trust in politicians or managers is high. Results focusing on variation in general trust, in contrast, are mixed, highlighting the importance of domain-specific trust. The empirical evidence in this paper is demonstrated in regressions that: control for variation across countries and across years; control for within-country variation associated with macroeconomic fundamentals and existing financial market development; and cluster standard errors by country to account for correlated errors across compatriot observations. As such, the findings related to PSARS are not subject to concerns about correlated omitted variables that arise in studies that make inferences based on cross-sectional measures of accounting and disclosure standards, like the CIFAR index. Additional analysis shows that the results are robust to alternative estimation methods (i.e., System GMM as in Blundell and Bond, 1998), and additional controls capturing variation in broader institutional protections potentially correlated with reporting quality (i.e., property rights), cultural-ethical norms that can improve reporting quality (i.e., corporate ethics), and growth expectations based on innovation that leads to more positive beliefs about firms generally (i.e., innovative capacity). These additional controls also help mitigate concerns about common response bias driving the main results, as the additional controls come from the same survey instrument as PSARS. 3

6 The paper contributes to the literature along several dimensions. First, it provides robust evidence on the determinants and implications of the strength of auditing and reporting standards in capital markets. Leuz and Wysocki (2008) highlight the paucity of research relating marketlevel outcomes to reporting and disclosure regulation. The analysis in this study helps address this gap in the literature by providing evidence on the associations between PSARS and macrolevel measures related to both economic and capital market development. A benefit of exploring variation in perceptions of accounting and disclosure quality in the international setting is that it is possible to capture variation driven by fundamentals that are observable (e.g., GDP) and not measured using firms accounting systems. Dechow, Ge, and Schrand (2010) identify this measurement issue as a significant problem in empirical work involving financial reporting quality. Second, the results contribute to our understanding of factors influencing capital market development. Several studies have shown that legal protections for investors are associated with capital market development (Djankov, McLiesh, and Shleifer 2007; Spamann 2010). Some studies in this stream of literature have suggested that accounting and disclosure quality is associated with capital market development, based on cross-sectional regression estimates (La Porta et al. 2006; La Porta et al. 1998). This study provides more robust evidence that financial auditing and reporting quality, and perceptions of such quality in particular, are associated with both the market capitalization of and trading activity on national stock markets. Additionally, this study provides evidence that the potential effects of PSARS on subsequent capital market development depend on public trust in politicians and firms, suggesting that the returns to investments in PSARS will vary with such slow-moving institutional-cultural factors. Third and closely related, the paper presents a novel country-year measure of the strength of financial auditing and reporting standards to the accounting literature. 3 Such a measure can be useful across a number of contexts because time-series variation allows for tests using PSARS to control for non-time-varying features that potentially confound prior research on accounting standards at the country level (e.g., unobservable country-level characteristics). Additionally, the 3 Questions from the EOS related to those that form the basis for PSARS have been used to measure corporate opacity in the cross section of countries in Gelos and Wei (2005) and Jin and Myers (2006). Bushee and Friedman (2014) use the PSARS questions as part of a proxy for disclosure standards in a study of the relation between disclosure standards and stock return noise. 4

7 breadth of coverage of the PSARS measure (up to 145 countries) allows for research on accounting and auditing standards to extend beyond the oft-used CIFAR sample which is restricted to around countries. 2 Related literature and hypothesis development Several studies examine the determinants and implications of accounting and disclosure standards (see Leuz and Wysocki (2008) for a review focused on regulation). These studies generally focus on cross-sectional measures available at a single point in time, including the CIFAR index (e.g., La Porta et al. 1998), the PWC Opacity Index (Gelos and Wei 2005), S&P Transparency scores (Khanna et al. 2004), and disclosure measures based on disclosures in prospectuses (La Porta et al. 2006). Djankov et al. (2008) construct an index of disclosures available at the country-year level, but their measure focuses on disclosure related to self-dealing transactions that can allow managers to expropriate value from shareholders and debtors. In contrast this study utilizes a proxy for accounting and disclosure standards that applies to reporting financial performance more generally and is available at the country-year level for over 1,400 country-years between 2002 and Bushman, Piotroski, and Smith (2004) examine the legal and political determinants of firm transparency using cross-sections of countries and data on transparency from CIFAR (Center for Financial Analysis and Research) based on financial reports from the early 1990s. They focus on country-level legal background and institutional quality as predictors of accounting and disclosure quality. In contrast, in this study I focus on relations between macroeconomic development, time-varying institutional quality, and capital market development as consequences of accounting and disclosure quality, while controlling for country-level factors with country fixed effects. Several studies in the international Law and Finance stream of literature have explored various facets of the institutional environment, including accounting and disclosure quality, and their relation to capital market development, often with a focus on legal protections for armslength equity and debt investors (e.g., Djankov et al. 2007; La Porta et al. 1998; Spamann 2010). The general finding from this literature is that stronger institutions and better investor protections facilitate capital market development. Studies have also examined more specifically how the availability of information to capital market participants is associated with market development. 5

8 However, the focus tends to be on information that prevents expropriation (e.g., Djankov et al. 2008) or information disclosed in prospectuses (La Porta et al. 2006) rather than information on financial performance more generally. Furthermore, the majority of this literature relies on proxies that are constant at the country-level. Because they cannot control for country fixed effects, inferences are subject to concerns about correlated omitted variables. The use of the time-varying PSARS measure allows for country fixed effects, mitigating concerns related to country-level confounds when exploring the relations between accounting and disclosure standards and both macroeconomic and capital market development. Related studies in the accounting literature focus on the determinants and implications of firm-level accounting quality in an international setting. These papers frequently use proxies for accounting quality related to firm-level earnings characteristics and relations between earnings and other fundamentals, including earnings smoothing, earnings predictability, and timely loss recognition (Bhattacharya, Daouk, and Welker 2003; Lang et al. 2012). Dechow et al. (2010) discuss issues with earnings quality proxies in international settings, including problems related to measurement noise and the effects of institutional heterogeneity. They note how institutional differences can imply that earnings- and accruals- based proxies could be interpreted differently across countries (i.e., in some cases reflecting opportunism, in other cases reflecting transparency). In contrast to these studies, I focus on a country-year measure of the strength of financial auditing and reporting standards based on a survey of executives. In Section 3.2.2, I show how the primary measure I use relates to several measures of earnings quality in addition to other transparency proxies related to analyst following and auditor choice. Accounting and disclosure standards have been linked to stock properties including liquidity, idiosyncratic volatility, and synchronicity (Leuz and Wysocki 2008). Several studies use the adoption of IFRS as a positive shock to the strength of standards at the firm or country level. Leuz and Verrecchia (2000) and Daske et al. (2008), for example, show that the adoption of IFRS is associated with increased liquidity. Internationally, Jin and Myers (2006) find a negative relation between disclosure transparency and synchronicity, which they argue reflects the degree of firm-specific information in prices (Morck, Yeung, and Yu 2000). Consistent with the present study, Jin and Myers (2006) use disclosure scores from the 1999 and 2000 Global Competitiveness Reports to capture cross-sectional variation in transparency. However, they use 6

9 only a cross-section of scores, while the present study exploits the panel nature of the PSARS data available from 2002 through Hypothesis development The specific hypotheses build on the extensive literature streams on capital market development, institutional quality, and potential effects of financial auditing and reporting quality discussed in the prior section. Better financial auditing and reporting standards can facilitate the development of capital markets in several ways. The first-order effect of improving financial auditing and reporting is likely a reduction in information asymmetry, which can help level the playing field between investors and managers, between directors and management, and across investor types. The reduction in information asymmetry can mitigate losses due to agency problems and loosen financial constraints caused by investors price-protecting. When investors expect better information quality, potentially due to stronger financial auditing and reporting standards, their needs to price protect should be lower. This should be associated with more equity market trading activity, because price protection acts as a volume-reducing transaction cost when impounded into bid-ask spreads. Lower price protection should also facilitate higher valuations, easier firm access to capital, and potentially a greater number of firms listing their equity on the market. These should all lead to larger domestic stock markets, as measured by market capitalization. 4 For creditors, lower price protection should translate into more loans being granted and expansions in credit markets as well. Based on these mechanisms, which are consistent with prior literature linking institutional features to capital market development (Djankov et al. 2008; La Porta et al. 2006; La Porta et al. 1998), I propose the first hypothesis. Hypothesis 1 PSARS is positively associated with capital market development. Financial auditing and reporting standards do not exist in a vacuum. There are additional features that can facilitate the potential effects of PSARS on capital market development, including trust that the primary players will dutifully carry out their responsibilities related to 4 There could also be a dampening effect due to compliance costs of more extensive auditing and reporting standards, which may be perceived to be better and therefore be associated with higher PSARS scores. 7

10 auditing and reporting. Two sets of important players are politicians, who generally exercise some control over the regulatory regime, and managers, who as preparers bear primary responsibility for financial reporting quality. When politicians and managers are more trustworthy, financial reports can be expected to be more reliable, amplifying the positive effect of stronger standards on capital market development. 5 Prior research finds that generalized trust (i.e., trust in others in general rather than in a specific set of economic agents) matters in capital markets, in that it is positively associated with stock market participation (Guiso, Sapienza, and Zingales 2008) and cross-border investment (Guiso, Sapienza, and Zingales 2009). Generalized trust can also enhance the creditability of financial reports or the credulity of capital market participants, in that people who have more generalized trust might be more likely to believe financial reports. These observations lead to the second, third, and fourth hypotheses. Hypothesis 2 The association between PSARS and capital market development is stronger when politicians are more trustworthy. Hypothesis 3 The association between PSARS and capital market development is stronger when managers are more trustworthy. Hypothesis 4 The association between PSARS and capital market development is stronger when people are more trusting in general. 3 Sample construction and descriptive statistics 3.1 Source and definition of PSARS The measure of the strength of auditing and reporting standards, PSARS, comes from the Executive Opinion Survey (EOS) conducted by the World Economic Forum (WEF), which is 5 Hypotheses 2-4 are based on a complementary relation between trustworthiness and the strength of financial auditing and reporting standards. If these are instead substitutes, then trustworthiness would be expected to be negatively related to the association between PSARS and capital market development. I expect a complementary relation as trustworthy regulators/politicians and managers/preparers are an important part of ensuring that strong financial reporting standards translate into reliable financial reports. If individuals, including investors, are more trusting in general, then the substitutive relation may play a larger role. 8

11 used as an input to and reported in the annual GCRs. The EOS is conducted annually and, since 2002, has included a question on the strength of financial auditing and reporting standards. These questions are presented in Table 1. The questions vary slightly in wording but generally ask respondents to rate the strength of financial auditing and reporting standards related to company financial performance in the respondents countries on a scale from 1 (extremely weak) to 7 (extremely strong the best in the world). Each annually-published GCR provides country-year averages of the responses to these questions. Gelos and Wei (2005), Jin and Myers (2006), and Fernandes and Ferreira (2009) use related scores from 1999 and 2000 to measure country-level opacity related to disclosure standards. 6 Bushee and Friedman (2014) use the scores as part of a proxy for the quality of disclosure standards. I use the country-year responses as a measure of the strength of financial auditing and reporting standards, broadly defined. This definition is consistent with the use of the survey responses in prior work and a natural interpretation of the survey questions. The EOS is administered in the first quarter of each year in a varying number of countries, with a generally increasing number of countries over time. In each year survey responses are collected from thousands of executives through a network of partner institutions of the WEF. Most respondents are CEOs or at a similar executive level. The 2013 EOS was administered to over 13,000 executives in 148 countries. Respondents come from a range of industries, sectors, and firm sizes. Greater detail on the EOS can be found in each annual GCR. The Global Competitiveness Report for 2013 was posted online on September 3, Values for PSARS from are downloadable from the data platform on the GCR website. 8 Values prior to 2006 were hand collected from the appendices of published GCRs. Values used in this study are based on values reported by the World Economic Forum rather than averages calculated by the author based on raw survey data because the WEF does not make the 6 They use responses to The level of financial disclosure required is extensive and detailed (1 = strongly disagree; 7 = strongly agree), which was asked in 1999 and 2000 only. This question about financial disclosure was not asked after 2000, but the responses to it are highly correlated with the PSARS questions from Bushee and Friedman (2014) use the disclosure questions from 1999 and 2000, the PSARS questions from , and the CIFAR index from the mid-1990s in their index of disclosure quality. 7 The 2013 GCR is titled The Global Competitiveness Report , but corresponds to data from 2013 and before, as the publication date is in late Accessed at on February 6,

12 raw survey data available. Reported country-year scores prior to 2006 were computed simply as equal-weighted averages. In 2007 the GCR began reporting country-year scores as two-year moving weighted averages of the country-year average responses using the current and prior year responses, with weights calculated to emphasize current-year responses while taking into account the country-specific number of responses in both years. In 2008, the GCR began computing country-year averages using a weighting process based on respondents firms sectors (Agriculture, Manufacturing industry, Non-manufacturing industry, and Services), where the weight applied to responses is based on the sector s contribution to GDP. Reported scores since 2008 are a result of sector-weighting at the country-year level and averaging the current and prior years responses to form a moving average score. This moving averaging process is expected to weaken the power of country-year regression analyses by smoothing variation in PSARS across country-years and reducing the independence of within-country observations, necessitating clustering of standard errors by country. The main sample is based on country-years reported in the World Economic Forum s Global Competitiveness Reports (GCR) with data available on the strength of financial auditing and reporting standards. The GCR reports data at the entity-year level, where entities are generally countries, but there are exceptions (e.g., Hong Kong). In the paper I use the term country for clarity even though the term entity might be technically more descriptive. Analyses are conducted with country-year level observations. Based on the timing of the surveys, PSARS taken from the GCR in year t is matched to independent variables measured in year t-1. So, PSARS taken from the 2010 Global Competitiveness Report is matched to GDP, capital market development, etc., captured and calculated for calendar In other words, the 2009 value for PSARS for each country is taken from the 2010 GCR, which itself is based on a weighted average of survey responses received in early 2009 and This matching procedure is consistent with the WEF s matching procedure in calculating their index of global competitiveness in that they match surveys from year t to data (e.g., real GDP) corresponding to year t-1. 10

13 PSARS evolves slowly, leading to significant autocorrelation. 9 The regression analyses presented below address this autocorrelation in two ways. First, standard errors are clustered by country to account for nonzero correlation across country-specific residuals. Second, all models on which inferences are based are estimated with country fixed effects. In some tables, I present regression estimates without country fixed effects, since juxtaposition of estimates with and without country fixed effects provides information about associations identified using acrossversus within-country variation. The across-country variation is more comparable to prior studies using static measures of financial auditing and reporting quality. Figure 1 shows time-series plots of PSARS for four countries from 2002 through The four countries are the United States, Italy, Ireland, and India. Their time-series of PSARS scores are shown to provide a better understanding of potential drivers of PSARS. The United States is mainly shown for reference, since there were numerous events that plausibly influenced PSARS in this time period. From 2002 through 2013, there were several accounting scandals (e.g., WorldCom, Enron, HealthSouth, Lehman Brothers) in addition to regulatory responses aimed at improving financial auditing and reporting quality (e.g., the Sarbanes-Oxley and Dodd- Frank Acts). Other countries provide starker examples. Italy s PSARS score dropped considerably from 2003 to 2004, around the time of the Parmalat accounting scandal. Ireland from 2008 through 2011 shows a decline of similar magnitude, coincident with the Anglo Irish Bank accounting scandal around This decline also coincides with the global financial crisis and macroeconomic slowdown, illustrating the need to control in subsequent tests for macroeconomic factors that can influence PSARS. India s time-series of PSARS scores shows a modest decline from 2009 to 2011 around the time of the Satyam accounting scandal. Two caveats must be pointed out before proceeding. First, the EOS focuses on opinions of executives, mainly CEOs. While their opinions are important, they are far from the only participants in capital markets or the only producers and users of financial reports. That being said, executives are likely to be wealthy individuals who are investors in capital markets, and are likely to use accounting information and disclosures of rivals when formulating strategy (e.g., 9 The AR(1) coefficient from the regression PSARS PSARS is with a k, t k, t 1 k, t standard error of The AR(1) coefficient from the regression PSARS PSARS in changes is with a standard error of k, t k, t 1 k, t 11

14 acquisitions, expansions, production, etc.). Therefore while executives might be thought of primarily as preparers, their role as users of financial information should not be disregarded. See also Dichev et al. (2013) and Nelson and Skinner (2013) for arguments in favor of and against inference based on surveys of executives. Because executives are preparers of financial reports, they may answer with positive bias in an attempt to influence survey-readers to view local accounting and auditing standards as stronger or more favorable. To the extent that such biases are country-specific and constant, their effects will be absorbed by country fixed effects in the empirical tests. Second, the EOS measures executive opinions with noise. This noise could be due to measurement noise (e.g., common response bias) or other factors influencing executives opinions (e.g., general optimism or a desire to influence survey readers). While the analysis attempts to control for factors that plausibly influence managerial opinions, this noise could still attenuate statistical relations between proxies and is expected to work against positive findings in the empirical tests. Robustness checks in Section 4.4 address concerns about biases in estimates due to noise in the survey-based proxy. 3.2 Relations between PSARS and extant proxies for auditing and reporting quality Given the relative novelty of the PSARS measure, this section examines correlations between PSARS and existing empirical proxies to provide a sense of what PSARS is and is not. In this section I focus on correlations between PSARS and previously-used measures of accounting, auditing, financial reporting, and disclosure standards and quality. Section 4 examines relations between PSARS and indicators for macroeconomic development, capital market development, and institutional quality before introducing the main analyses focusing on potential effects of PSARS on capital market development PSARS and country-level disclosure quality proxies First, I explore correlations between country-mean PSARS and several measures used in prior literature to capture financial reporting or disclosure quality available at the country level. These prior measures include the extensively-used CIFAR score (e.g., Bushman et al. 2004), the prospectus disclosure index (La Porta et al. 2006), the O-factor opacity index constructed by PriceWaterhouseCoopers (Gelos and Wei 2005), the S&P Transparency and Disclosure Survey mean country score (Khanna et al. 2004), and the anti-self-dealing business disclosure index 12

15 available from the World Bank (Djankov et al. 2008). Consistent with PSARS capturing variation in financial reporting quality, country-mean PSARS scores are positive and significantly correlated with CIFAR, S&P Transparency, Prospectus Disclosure, and Self-Dealing Disclosure, and negatively correlated with opacity as captured by PwC s opacity index. A benefit of the PSARS score over these measures is that the PSARS score is available at the country-year level, allowing for time-series variation, and for over 100 countries, representing much broader coverage PSARS and earnings quality, analyst following, auditor choice, and accounting standards Second, I examine simple correlations between PSARS and proxies for earnings quality, the fraction of listed firms using Big-N auditors to audit their financial statements, the fraction of listed firms choosing to report with high-quality accounting standards (i.e., US GAAP or IFRS), and analyst following. These proxies are based on annual firm-level data from Compustat and IBES. Firm-level data based on fiscal years ending in January through August (September through December) of year t are matched with country-level PSARS corresponding to year t (year t+1). Firm-level data is winsorized at the first and 99 th percentiles, consistent with prior research. The earnings or accruals quality proxies are calculated as country-year medians of: absolute accruals (Absolute Accruals), signed accruals (Signed Accruals), timely loss recognition (TLR), the Dechow and Dichev (2002) accruals quality measure (DD AQ), the Francis et al. (2005) measure of discretionary accruals quality (FLOS AQ), and two discretionary smoothing proxies that Lang et al. (2012) calculate (SMTH1 and SMTH2). Details regarding the construction of the earnings quality proxies can be found in the Appendix. Three additional proxies are the median number of analysts making annual earnings forecasts in the 90 days prior to a firm s earnings announcement from IBES (Analysts Following), the fraction of firms preparing their annual financial statements using US GAAP or 10 While the Self-Dealing Disclosure index is also available at the country-year level, it focuses on disclosure related to managerial diversion rather than financial auditing and reporting standards more broadly and is available only from 2005 rather than from As this study uses lagged variables in small-t (i.e., few years), large-n (i.e., many countries) regressions, the extra years of coverage are particularly useful in contributing power to statistical tests. 13

16 IFRS (High Acct Std Frac) from Compustat, and the fraction of firms employing Big-N auditing firms as the primary auditors of their annual reports (Big N Frac) from Compustat. Correlations between PSARS and proxies for earnings or accruals quality, analyst following, accounting standard adoption, and auditor choice are shown in Table 3. Pearson (Spearman) correlations are above (below) the diagonal, and correlations in bold are significantly different from zero at the five percent level. Overall, PSARS is significantly positively correlated with Signed Accruals, Analyst Following, High Acct Std Frac, and Big N Frac. PSARS has a significantly negative Pearson correlation with Absolute Accruals, with a negative but insignificant Spearman correlation. PSARS is not significantly correlated with country-year median values of the earnings quality proxies. The correlations presented in Table 3 suggest that PSARS is associated with more positive accruals, potentially smaller absolute accruals, greater analyst following, more firms reporting under IFRS or US GAAP, and more firms choosing big- N auditors. These correlations are consistent with PSARS capturing greater auditing and financial reporting quality, although potentially a dimension that is orthogonal to the features captured by common earnings quality proxies. The correlations are also consistent with Dechow et al. (2010) s observation that earnings quality proxies in international settings are noisy, due in part to potentially heterogeneous interpretations across countries and institutional settings (i.e., in some cases high accruals are more likely to reflect information, in other cases they are more likely to reflect obfuscation). In contrast, because it is based on survey responses, PSARS is likely to have a consistent interpretation across countries and institutional settings. 3.3 Definitions and sources of dependent and control variables Variables representing macroeconomic fundamentals include GDP, GDP Growth, Unemployment, Inflation, and FDI, which are taken from the World Bank s World Development Indicators data set. GDPk,t is GDP per capita in year t for country k measured using current US dollars. 11 GDP Growthk,t is the year-on-year change in GDP in percent. Unemployment is the unemployment rate in percent, and Inflationk,t is the annual inflation rate in percent based on the local consumer price index. FDIk,t is foreign direct investment (net inflows) as a percent of GDP. 11 GDP per capita measured in constant US dollars is available for a smaller set of country-years. In regression tests, year indicators will absorb variation in GDP driven by $US inflation over years. 14

17 Proxies for institutional quality are taken from the International Country Risk Guide (ICRG). ICRG reports these monthly, and monthly values are averaged to form annual values. The indicators taken from ICRG include Corruptionk,t, Law and Orderk,t, Bureaucracy Qualityk,t, and Investment Profilek,t. These and similar indicators have been used extensively in prior research on institutional factors (e.g., La Porta et al. 1998). At the time of data collection, the ICRG variables were available in a monthly time series from January 1984 through April Analyses with ICRG data therefore exclude years after Proxies for capital market development include Stock Market Cap, Stock Trading Volume, Bank Credit, and Private Credit. Stock Market Capk,t is the equity market s total capitalization as a percentage of GDP in year t. Stock Trading Volumek,t is the value of listed company stock traded during year t divided by year-t GDP. Bank Creditk,t is domestic credit provided by the banking sector as a percentage of GDP in year t. Private Creditk,t is domestic credit provided to the private sector (not just from banks) as a percentage of GDP in year t. Stock Market Cap, Stock Trading Volume, Bank Credit, and Private Credit come from the World Bank s World Development Indicators data set. Additional proxies are taken from the EOS, the World Values Surveys, and The European Values Surveys to facilitate tests of Hypotheses 2, 3, and 4, and robustness checks on the tests of Hypothesis 1. The additional proxies from the EOS include PTIP (public trust in politicians), EBOF (ethical behavior of firms), Property Rights, and Capacity for Innovation. Each of these is available at the country-year level and has coverage similar to that of PSARS, as discussed in Section 3.1. The numerical scores for these proxies are country-year average responses to prompts in the Executive Opinion Survey as reported in Global Competitiveness Reports. The prompts are scored on Likert scales ranging from one to seven. The prompt from the 2013 EOS for PTIP is, In your country, how would you rate the ethical standards of politicians? [1 = extremely low; 7 = extremely high]. The prompt for EBOF from the 2013 EOS is, In your country, how would you rate the corporate ethics of companies (ethical behavior in interactions with public officials, politicians and other firms)? [1 = extremely poor among the worst in the world; 7 = excellent among the best in the world]. Property Rights are taken as average responses to In your country, how strong is the protection of property rights, including financial assets? [1 = extremely weak; 7 = extremely strong]. The prompt from the 2013 EOS for Capacity for Innovation is, In your country, to what extent do companies have the capacity 15

18 to innovate? [1 = not at all; 7 = to a great extent]. As with PSARS, reported values from the GCR from year t are matched to macroeconomic data in year t 1. General Trust is taken from the World Values Surveys (WVS) and European Values Surveys (EVS) as the fraction of survey respondents in each country indicating that they believe most people can be trusted, rather than believing that one needs to be very careful when dealing with others. The WVS and EVS surveys are not conducted every year, so General Trust is used as a country-level variable. 3.4 Sample descriptive statistics Table 4 presents a list of the countries with PSARS available, the number of years available, and the average country value for PSARS. Finland and New Zealand have the highest average values of PSARS, at 6.21 and 6.18, respectively. Myanmar and Angola have the lowest, at 2.29 and 2.48, respectively. The number of observations for each country ranges from 1 for Myanmar and Bhutan to 12 for much of the developed and developing world. Some crosscountry comparisons of average PSARS values (e.g., the United States 5.66 compared to South Africa s 6.28) might defy expectations. These cross-country differences in means highlight the importance of controlling for country effects that can mitigate variation based on, for example, country-specific respondents biases. The upper limit of 12 observations per country is based on the number of years for which the EOS included the PSARS question as of the time of writing. Also included in Table 4 are the average GDP per capita in US dollars (GDP), and the average country values for Stock Market Cap, Stock Trading Volume, Private Credit, and Bank Credit. 12 Descriptive statistics are presented in Table 5. There are 1,443 country-year observations with PSARS available. These country-years form the basis of the sample. PSARS has a sample 12 Several countries (42 out of 147, or roughly 29%) have missing values for Stock Market Cap and Stock Trading Volume, in most cases because they lack stock markets. Because these countries tend to be less developed (average GDP per capita of $3,848 for these countries versus $12,501 for all sample countries), they are less likely to be covered by the EOS, and they represent only 285 out of 1443 country-year observations (less than 20%). Only Lao PDR, Rwanda, and Myanmar lack data on Private Credit and Bank Credit, representing 6 country-year observations. Negative values for Bank Credit are driven by government holdings of international reserves as deposits in banks other than the central bank, as deposits have a negative effect on Bank Credit. 16

19 mean of 4.72 (on a one to seven scale) and a standard deviation of PSARS does not appear to be skewed, as the median is 4.70, essentially equal to the mean. Average GDP per capita is $13,632.69, and average GDP growth is 4.00 percent. Average unemployment, inflation, and FDI as a percentage of GDP are, respectively, 8.66, 6.72, and 5.19 percent. The institutional quality variables from the ICRG are available only for 1034 country-years, as coverage stops in Mean Stock Market Cap is 57.52, implying that the average stock market capitalization is nearly 60 percent of annual GDP. Stock Trading Volume is, on average, percent of annual GDP. Bank Credit and Private Credit have means of and percent, respectively. The proxies for stock market development tend to have means larger than medians and standard deviations on the order of or larger than the mean. This suggests that these variables tend to be right-skewed. The additional proxies from the EOS, Public Trust in Politicians, Ethical Behavior of Firms, Property Rights, and Capacity for Innovation, have means and medians that are similar to each other and standard deviations around 1, like PSARS. General Trust, from the WVS and EVS, has a mean of 0.27 and a standard deviation of Correlations are shown in Table 6. The unit of analysis in Table 6 is the country-year. Pearson product-moment (Spearman rank) correlation coefficients are shown below (above) the diagonal. Correlations that are significantly different from zero at the one percent level are displayed in bold. Statistical tests of whether correlations are different from zero do not take into account potential correlations across observations (e.g., within-country across years), as the later analysis does. As such the statistical significance of the correlations is expected to be inflated. PSARS is positively correlated with proxies for macroeconomic and institutional development (e.g., GDP, Bureaucracy Quality, Corruption, Law and Order, and Investment Profile) 14. PSARS is also positively correlated with measures of capital market development (Stock Market Cap, Stock Trading Volume, Bank Credit, Private Credit), other proxies from the Executive Opinion Survey (Public Trust in Politicians, Ethical Behavior of Firms, Property Rights, and Capacity for Innovation), and General Trust. PSARS is not significantly correlated 13 The means reported in Table 5 are means of country-year observations. The means reported at the bottom of Table 4 are means of country-means, and therefore do not match exactly. 14 Recall that higher values of Corruption represent country-years less susceptible to corruption, consistent with ICRG s reporting methodology. 17

20 with FDI. Again, these correlations are susceptible to confounds from omitted correlated variables (i.e., GDP is correlated with everything) and inflated test statistics due to multiple observations of the same country, both of which are addressed in the subsequent regression analyses. 3.5 PSARS and contemporary macroeconomic, institutional, and capital market features This section explores macroeconomic, capital market, and institutional fundamentals as contemporary determinants of PSARS. Table 7 presents regression analyses that explore the relation between PSARS and fundamentals related to the macroeconomy, institutions, and market features that might drive demand for accounting and disclosure quality. The estimates in Table 3 are from regression estimates of the following equation PSARS ΓX (1) k, t t k, t k, t where Xk,t is a vector of proxies for macroeconomic fundamentals, institutional quality, and capital market development. Regressions presented in Table 7 are estimated on variables standardized to be mean-zero and unit-variance, so that the coefficients can be interpreted as expected standard deviation changes in expected PSARS with standard-deviation changes in underlying variables. All regressions shown in Table 7 include year fixed effects to account for global shocks, and the even-numbered regressions also include country-level fixed effects to account for unobservable factors at the country level. The odd-numbered models do not include country fixed effects in order to allow for comparisons of regression coefficients identified using all variation with coefficients identified from only within-country variation. Standard errors are clustered at the country level to adjust for potential correlation in the error terms across same-country observations. Model 1 of Table 7 includes the macroeconomic fundamentals and capital market development variables in Xk,t. The coefficients on log(gdp) and Stock Market Cap are positive and significantly different from zero at the one percent level, and the coefficient on Stock Trading Volume is negative and significant at the ten percent level. The coefficients on the remaining independent variables are not significantly different from zero, and the R 2 for the model is Model 2 adds country fixed effects. The coefficient on log(gdp) increases in 18

21 magnitude and remains significant at the one percent level, while the coefficient on Stock Market Cap declines, but remains positive and significant at the ten percent level. The coefficient on Stock Trading Volume in Model 2 is positive, but not significantly different from zero, and the coefficient on Unemployment becomes negative and significant at the five percent level. The R 2 for Model 2 is 0.927, implying that country-level fixed effects explain a substantial amount of the variation in PSARS over and above that explained by macroeconomic fundamentals and capital market characteristics. Models 3 and 4 of Table 7 are similar to Models 1 and 2, respectively, but with the addition of the institutional quality proxies. In Model 3, without country fixed effects, the coefficients on Bureaucracy Quality, Corruption, and Investment Profile are positive and significant at the one percent level. With the addition of country fixed effects in Model 4, however, the coefficients on these proxies move towards zero and lose statistical significance at conventional levels. As in Model 2, the coefficients on log(gdp) and Stock Market Cap are positive and significant, while the coefficient on Unemployment is negative and significant. Models 5 and 6 add the lagged value of PSARS to Models 3 and 4, respectively. The coefficient on the lagged value of PSARS is positive and significant at the one percent level in both regressions. In Model 6, which includes country fixed effects, the coefficients on log(gdp), GDP Growth, and Stock Market Cap are also positive and significant, all at the five percent level. The regressions in Table 7 overall suggest that much of the variation in PSARS can be explained by GDP, a state variable related to economic recession or growth (Unemployment, Inflation, or GDP Growth), the size of the stock market relative to the country s GDP, and country and year fixed effects. These factors are also plausibly associated with subsequent capital market development (Garcia and Liu 1999), meaning that it is crucial to control for them when investigating potential effects of PSARS on capital market development. The results in Table 7 further highlight the importance of using panel data to investigate the effects of PSARS on capital market development, as a cross-sectional study bears a significant risk of attributing variation in capital market development to variation in accounting, auditing, or disclosure factors when it is plausibly driven by other country-specific features (i.e., fixed effects). As the inclusion of country fixed effects makes associations between PSARS and each of FDI, Bureaucracy Quality, Corruption, Law and Order, and Investment Profile insignificant, I do not include these 19