Quality and Asymmetry of Information in the Capital Markets of Emerging Countries

Transcription

1 1 Quality and Asymmetry of Information in the Capital Markets of Emerging Countries Orleans Silva Martins 1 Federal Uniiversity of Paraíba Address: PPGCC/CCSA, UFPB, University City, João Pessoa/PB, Brazil, ZIP code: orleansmartins@yahoo.com.br Lucas Ayres Barreira de Campos Barros University of São Paulo Address: FEA/USP, Avenue Professor Luciano Gualberto, 908, Butantã, São Paulo/SP, Brazil, ZIP code lucasayres2002@gmail.com EFM classification code: 180 Abstract We investigate the relationship between the quality of accounting information and information asymmetry at the firm level and how this relationship is influenced by the quality of the information environment at the country level. Our sample comprises over 15,000 publicly traded firms from 22 countries that make up the MSCI Emerging Markets Index between the years 2000 and We construct an Information Asymmetry Index (IAI) using Principal Component Analysis (PCA) based on a set of firm characteristics usually associated with the degree of firm-level information asymmetry. We relate the IAI to five proxies for earnings quality in both the pre- and post-ifrs adoption periods. We also include in the analysis two proxies for the quality of the information environment, based on country governance and sovereign risk ratings. We show that a better information environment at the country level is associated with a lower degree of information asymmetry and with higher accounting information quality at the firm level. Lower information asymmetry is associated with higher information quality. However, this relationship is usually stronger in emerging countries with weaker information environment, suggesting that in these countries firm-level characteristics may be particularly relevant to investors. Keywords: Earnings Quality; Information Asymmetry; Information Environment. 1 Presenter and corresponding author.

2 1. Introduction As noted by Leland and Pyle (1977), financial markets around the globe are plagued by information asymmetry between buyers and sellers. In this way, Bushman and Smith (2001) point out that the use of financial accounting information can reduce information asymmetry among investors, because this disclosure guarantees greater investment protection. But for this, Palepu and Healy (2012) say that accounting effectively contributes to reducing this asymmetry only when it provides good information quality. Additionally, although the difficulty to distinguish good quality assets from bad quality assets is widespread, it might be even more pervasive in emerging markets, where the degree of information asymmetry is usually larger than in developed markets, and this is negatively related to the legal protection of investors (La Porta, Lopez de Silanes, Shleifer, & Vishny, 1998; Duarte & Young, 2009). Previous studies investigated the information environment of emerging countries and the quality and quantity of information provided by companies in these markets, usually within the framework of agency theory (Jensen and Meckling, 1976). For example, La Porta et al. (1998) highlight the importance of the information environment in which companies stocks are traded, as well as the legal protection that each country offers to investors, which varies significantly around the world. However, the association between the information quality and the degree of information asymmetry has rarely been studied (Richardson, 2000; Bhattacharya, Desai, & Venkataraman, 2013; Cormier, Houle, & Ledoux, 2013), notably in emerging countries, where the legal protection and the information environment available to investors tend to be weaker (La Porta et al., 1997, 1998). So, this paper is an effort to fill this gap. And to the best of our knowledge, this study is the first to investigate this issue in emerging stock markets. The motivation for this study is twofold. First, a number of previous studies examine the effects of information quality and information asymmetry on stock markets (e.g., on firm performance or stock returns), but the association between quality and asymmetry has rarely been explored (Richardson, 2000; Bhattacharya et al., 2013). Second, despite the substantial growth of emerging stock markets in the past decades, little attention has been paid to examining their information environment (Ghysels, Plazzi, & Valkanov, 2016; Yang, Ahn, Kim, & Ryu, 2017). In this way, we investigate the relationship between five earnings quality proxies and the information asymmetry index in different information environments in emerging countries. In this sense, we analyse a sample of firms from the main emerging stock markets to evaluate two basic and competing hypotheses. Based on the assumption that information quality and information asymmetry are negatively associated (Richardson, 2000; Brown, Hillegeist, & Lo, 2009; Bhattacharya et al., 2013), and considering that the information environment of the country can affect the quality and the asymmetry of the firm (Houqe, Zijl, Dunstan, & Karim, 2012; Bhattacharya et al., 2013; Cormier et al., 2013; Schiehll & Martins, 2016), the hypotheses of complementarity and substitution are analysed. Under the complementarity hypothesis, the stronger association between better information quality and lower information asymmetry is present in countries with stronger information environments (Cormier et al., 2013). On the other hand, under the substitution hypothesis, we expect a stronger association between better information quality and lower information asymmetry in countries with weaker information environments, as a compensate effect of the weaker legal protection (Bhattacharya et al., 2013; Schiehll & Martins, 2016). This study innovates when analysing in emerging markets the relationship between the accounting information quality and an Information Asymmetry Index (IAI), which is based on characteristics of companies with a lower level of information asymmetry. We identify in the extant literature the characteristics commonly used as proxies for information asymmetry and which are easily observed in emerging markets, and we construct the IAI using firm s age, 2

3 size, analysts coverage and depositary receipts traded in developed markets. Subsequently, we estimate the relationship between this index and the proxies for accounting information quality: earnings persistence, conservatism, value relevance, income smoothing and earnings management. For this, we use a panel data of 15,111 companies in 22 countries during the period from 2000 to In this period, the mandatory IFRS adoption is determinant for the accounting information quality, as Houqe et al. (2012) can observe. Finally, we analyse the relationship between the IAI and the variables related to the information environment at the country level. In this regard, our study complements the works of Richardson (2000), Brown et al. (2009) and Bhattacharya et al. (2013). Our main findings are the following. We can verify that age, size, analysts coverage and trading depositary receipts (DR) in developed markets are positively correlated characteristics that can be easily found in the countries of the analysed sample. We can observe that more mature firms tend to be larger, have more analysts accompanying them, and tend to trade their DR in developed markets. From these findings we construct the IAI through Principal Component Analysis (PCA). In general, we identify a positive association between information quality and IAI, which indicates that better information quality is positively associated with lower information asymmetry, as predicted in the extant literature. Additionally, we find differences in the periods pre- and post-ifrs, what is consistent with Houqe et al. (2012) and Bhattacharya et al. (2013), and this association remains positive. Our results suggest that information quality is higher in emerging countries with stronger legal protection to investors. But, when we narrow the analysis to firms with the lowest level of asymmetry, we find that in countries with weaker information environment the association between better information quality and lower information asymmetry is stronger, which is consistent with our predicted substitution hypothesis, where firms in emerging countries with weaker information environment try to compensate this weaker legal protection to investors with better information quality. The remainder of the paper is organized as follows. Section 2 presents the literature review and develops the hypotheses. Section 3 describes the study s data and method. Section 4 presents the results and their discussion. Finally, Section 5 provides the conclusions and potential implication of the results. 2. Literature Review In markets where information asymmetry is high, stakeholders do not have sufficient resources, incentives, or access to relevant information to monitor manager s actions, which may give rise, for example, to the practice of earnings management (Richardson, 2000). On the other hand, Brown et al. (2009) argue that higher quality accounting information associates with lower information asymmetry and thus reduces the uncertainty surrounding the operations of firms, leading its users to make better decisions and increasing economic efficiency. In addition, Bhattacharya et al. (2013) say that this relation interest to investors, managers, regulators, and standard-setters, and Cormier et al. (2013) add to this discussion the influence of the information environment, considering the reflection of the uncertainty of this environment. Bushman and Smith (2001) point out that the disclosure of accounting information may affect companies in three ways: (i) identification of promising investments, since the disclosure of this type of information supports the decision of the investor to allocate resources to opportunities that can create greater value; (ii) improvement of asset productivity, since, due to the greater quality and quantity of information, managers feel compelled to allocate resources to good projects, reducing the risk of expropriation; and (iii) reduction of liquidity and adverse selection risks, that is, reducing the information asymmetry among 3

4 stakeholders. Information asymmetry among managers and investors has profound implications for the liquidity and pricing of assets (Bushman & Smith, 2001). Brown et al. (2009) point out that earnings announcement can reduce information asymmetry in the capital market, reducing the informational advantage of the insider and increasing the pricing capacity of stakeholders. In order to facilitate the aggregation of more information to its users, the accounting profession has incorporated a series of concepts over time, such as: reliability, conservatism, comparability, accrual basis, relevance, among others, with the objective to describe the firm in the most reliable form. Information quality relates to the information environment in which the firm is inserted, including economic, political and social factors. Although Dechow and Schrand (2004) argue that information quality is contextual, since it means different things for the different users of the financial statements, the extant literature usually relates it to the excessive amount of non-recurring items published or the lack of transparency regarding the accounting choices made, in light of the current rules. In this context, Dechow and Schrand (2004) and Dechow, Ge and Schrand (2010) note that the most common accounting information quality indicators are: (i) earnings persistence; (ii) conservatism; (iii) value relevance; (iv) income smoothing; and (v) earnings management. Some of these indicators can be computed for individual firms, while others must be inferred from the estimation of models using a sample of firms. Earnings persistence, conservatism and value relevance are in the latter category, while income smoothing and earnings management are in the former. In our empirical analysis, we first compute the Information Asymmetry Index (IAI) for each firm and then relate it to each of these proxies for information quality, either directly or via interactions. In addition, in all cases we investigate how the quality of the information environment moderates the relationship of interest. Earnings persistence relates to profit sustainability. Dechow and Schrand (2004) argue that accounting information is of higher quality when it accurately demonstrates the intrinsic value of the firm, period after period, and is thus useful for evaluating its stocks and other securities in the capital market. However, Louis and Robinson (2005) warn that persistence may decrease in the presence of information asymmetry, while Bhattacharya et al. (2013) document that poor earnings quality is significantly associated with higher information asymmetry. So, it is natural to expect more persistence in firms with lower information asymmetry, indicating better information quality. Conservatism can be understood as the biased recognition of bad news, i.e., bad news tends to be recognized more quickly than good news (Basu, 1997; Ball & Shivakumar, 2005). It involves providing conservative (more reliable, in principle) information through nonoverly optimistic financial statements. As many estimates and judgments made by management cannot be verified, conservatism may become an efficient tool in establishing efficient contracts, since it restricts the opportunistic behaviour of managers, as asserted by Agency Theory (Watts, 2003). Therefore, in the face of lower information asymmetry, conservatism tends to increase, because the asymmetric recognition of events is a fundamental property of accounting rule and practice, and managers have an incentive to recognise firm s earnings in an asymmetric way (Ball & Shivakumar, 2005). Value relevance captures the association between accounting numbers and the firm s stock price, assessing the importance of accounting information for investment decision making (Collins, Maydew, & Weiss, 1997). The degree of information asymmetry between insiders and outsiders in the capital market may affect the relevance of accounting information because insiders may have privileged access to information, reducing the relevance of public disclosures. Bushman and Smith (2001) note that better disclosure reduces 4

5 information asymmetry and is key to determining the market value of the firm, indicating to investors the best opportunities for capital allocation. For this reason, we expect an association between lower information asymmetry and higher value relevance. Income smoothing is practiced to reduce the variability of earnings, thus reducing the uncertainty related to the results of firms. Eckel (1981) argues that income smoothing is a measure of quality and might be assessed based on the volatility of firms earnings divided by the volatility of their sales. Similarly, Dechow et al. (2010) state that it captures the reduction of the time differences between the recognition of revenues and expenses and the inflows and outflows of resources (minor accruals). Thus, income smoothing is a measure of accounting information quality, and firm s earnings become more predictable as it increases. This is consistent with Richardson (2000) and Bhattacharya et al. (2013), who note that in environments with greater uncertainty and volatility the information asymmetry also tends to be larger, which reduces the earnings quality. Thus, we expect that lower information asymmetry is associated with greater income smoothing. Dechow et al. (2010) claim that earnings management occurs when managers exercise judgment about the firm s financial information and operating activities in order to change its accounting information to suit their own interests. Higher earnings management indicates poorer information quality, because higher earnings volatility reduces the predictability of the firm s future cash flows (as opposed to income smoothing, which improves predictability). Additionally, there is a systematic relationship between the magnitude of information asymmetry and the level of earnings management (Richardson, 2000; Bhattacharya et al., 2013). Thus, we should expect that a decrease in information asymmetry is associated with an increment in income smoothing, and a consequent reduction in earnings management. The quality of the information environment relates to features such as the level of legal protection that the country offers to investors (La Porta et al. 1997, 1998). In this sense, Houqe et al. (2012) verify the importance of investor protection for accounting information quality. In Canada, a developed country, Cormier et al. (2013) point out that the relation between firm-level information quality and information asymmetry tends to be weaker in an environment with high uncertainty. However, especially in poor information environments, Bhattacharya et al. (2013) note that the information advantage of informed traders is greater, and the earnings quality has a negative association with information asymmetry. Additionally, although there is consistent evidence that investor protection affects financial market development (Schiehll & Martins, 2016), its influence on the relationship between information quality and information asymmetry is less clear (Yang et al., 2017). In this way, based on Bhattacharya et al. (2013), Cormier et al. (2013) and Schiehll and Martins (2016), and considering the influence of the information environment on the relationship between quality and asymmetry, especially in emerging stock markets, that receives little attention although their significantly growth in the last years (Yang et al., 2017), we evaluate two basic and competing hypotheses: (i) the complementarity hypothesis, which states that the predicted association between better information quality and lower information asymmetry is stronger in countries with stronger information environment, because in these countries firm-level characteristics work as a complement to country-level investor protection in the signalling of credibility to external investors; and (i) the substitution hypothesis, according to which the predicted association between better information quality and lower information asymmetry is stronger in countries with weaker information environment, because in these countries firm-level characteristics become particularly important to investors, serving to compensate the weaker legal protection. 5

6 3. Data and Method The sample analysed in this study is composed by firms that traded their stocks in the main stock exchanges of selected emerging countries. Our sample of countries is defined based on the MSCI Emerging Markets Index of Morgan Stanley Capital International (MSCI), which in 2017 has 24 countries. Thus, we collect the information available to all non-financial firms in these countries from the following sources: MSCI, Thomson Reuters DataStream (TRD), Worldwide Governance Indicators (WGI) and Standard & Poor s Database. Banks and financial institutions are excluded from the empirical analysis. To be included in the sample, the country must have at least one company with all the financial information available at TRD. And each firm must have financial information for at least one year during the period analysed. Under these conditions, we analysed the firms from the following countries: Brazil, Chile, China, Colombia, Czech Republic, Egypt, Greece, Hungary, India, Indonesia, Malaysia, Mexico, Peru, Poland, Qatar, Russia, Saudi Arabia, South Africa, South Korea, Taiwan, Turkey and the United Arab Emirates. Pakistan and Philippines were excluded from the main sample because their firms did not have information on at least one variable analysed in this study. The final sample consists of 256,581 firm-year observations, across 22 countries and 15,111 non-financial firms for the fiscal years 2000 to In Table 1 we summarize and describe the main variables. Table 1 Descriptions of the main variables. Proxy Description Reference Data Source Panel A: Information Asymmetry Firm Size Logarithm of Total Assets at year end. Easley, Hvidkjaer and O Hara (2002) Firm Age Analyst Coverage Depositary Receipt Panel B: Information Quality Persistence Number of Years from the IPO. Number of Analyst Forecasts issued in year t. Dummy that assumes value 1 if the firm has DR trading in developed markets and 0 otherwise. Earnings persistence from year t 1 to year t. Conservatism More timely recognition of Losses than Gains from year t 1 to year t. Value Sensitivity of Stock Prices to changes in Relevance Earnings Per Share and Book Value per Share. Dummy that assumes value 1 for firms whose Income coefficient of variation of net sales divided by Smoothing the coefficient of variation of net income is lower than 0.9, and 0 otherwise. Earnings Standard deviation of the error terms Management estimated by the DHKS model. Panel C: Information Environment Principal component of WGI s six dimensions (voice and accountability, political stability Worldwide and absence of violence, government Governance effectiveness, regulatory quality, rule of law Indicators and control of corruption), using a scale from 0 to 100. Sovereign Risk Indicators The SRI of each country, coded as follows: AAA = 17, AA+ = 16, AA = 15, AA- = 14,..., BB = 6, BB- = 5, B+ = 4, B = 3, B- = 2, CCC+ = 1, CCC and under = 0. Roudaki, Babajani and Tahriri (2016) Lin, Pantzalis and Park (2009) La Porta et al. (1998), Chan, Hong, and Subrahmanyam (2008) Dechow and Schrand (2004) Ball and Shivakumar (2005) Collins et al. (1997) Eckel (1981) Dechow, Hutton, Kim and Sloan (2012) González and García- Meca (2014) Oshiro and Saruwatari (2005) Thomson Reuters DataStream Thomson Reuters DataStream WGI Database 6 Standard & Poor s Database

7 We carry out the analysis in three stages. First, we construct the Information Asymmetry Index (IAI) based on a set of firm characteristics that are easily observable in emerging countries. Subsequently, we estimate the relationship between this index and the proxies for information quality and information environment. Finally, we analyse the relationship between the IAI and the proxies for information environment at the country level Information Asymmetry Index Information asymmetry in capital markets is a phenomenon that is affected by a variety of factors and cannot be measured directly. Several proxies have been used in the extant literature, either alone or in combination. Huang, Ma and Lan (2014) argue that it is still quite difficult to quantify information asymmetry, so that there are controversies about how to measure the degree of asymmetry. While many studies adopt a single proxy in their analyses, Huang et al. (2014) argue that different proxies likely reflect distinct dimensions of the phenomenon, although their information may overlap substantially. Lin, Pantzalis and Park (2009) point out that constructing an index is a useful way to use the information available from different proxies for information asymmetry and, at the same time, minimize redundancy and the influence of outliers. Brave and Butters (2011) argue that indices of this type have the advantage of capturing the interconnection of different variables, a desirable characteristic to allow the assessment of the systemic importance of each indicator. Huang et al. (2014) use a Principal Component Analysis (PCA) to construct an information asymmetry index based on 10 accounting variables, exclusively for Chinese firms. However, some of these variables are not easily obtained in most emerging markets, such as the market value of debt, and others are prone to measurement errors and timeliness problems (Pan, Lin, Lee, & Ho, 2015). Other indices are subject to similar problems, from those that use exclusively or excessively proxies related to analyst forecasts, to those based on market microstructure data. These indices may present bias or difficulty access to their proxies in less developed markets. According to Pan et al. (2015), the proxies for information asymmetry in the extant finance literature can be categorized into three groups: firm characteristics, dynamic, and sophisticated. The sophisticated proxies are arguably the most theoretically robust. They are, however, the hardest to access in emerging markets. On the other hand, firm characteristics are noisier, but more persistent over time and easier to access and process (Bharath, Pasquariello, & Wu, 2009), especially in emerging markets. Thus, in this study we develop an Information Asymmetry Index (IAI) based on firm characteristics. These proxies tend to be more accessible, less affected by measurement errors and market mood, and are easily observed in emerging markets, which increases the practical usefulness and replicability of the index. Specifically, we construct the IAI using firm s age, size, analyst s coverage and depositary receipts traded in developed markets. We show that these proxies are positively correlated with one another, and they are assumed to correlate negatively with the degree of information asymmetry. Therefore, a larger IAI indicates lower information asymmetry. Roudaki et al. (2016) point out that researchers have used firm s age as a measure of information asymmetry because older firms tend to be more mature and to use time-tested disclosure policies and practices. In addition, being more established firms, they are likely to receive more attention from market participants and regulators, thus becoming less opaque. Similarly, the larger the firm, the smaller the information asymmetry tends to be because they tend to have more stocks in circulation, a greater number of investors transacting with their securities, as well as more analyst coverage (Easley et al., 2002). In addition, as noted by Lin et al. (2009), numerous studies use analyst coverage directly as a proxy for information asymmetry. In this study, we use the number of analyst forecasts, under the assumption that 7

8 information diffusion increases with the number of analysts who cover the firm. The market development of a country, including legal protection of investors, may also substantially impact the degree of information asymmetry (La Porta et al., 1998). When a firm trades its securities in a foreign market, for example because of cross-listing through depositary receipts, its degree of opaqueness should be influenced by the institutional factors related to that foreign market, such as restrictions on ownership, short sale constraints, and accounting standards (Chan et al., 2008). Therefore, firms that trade in developed markets by issuing American Depositary Receipts (ADR) or Global Depositary Receipts (GDR) are expected to have lower information asymmetry. The IAI is constructed using Principal Component Analysis (PCA) applied to the several proxies for information asymmetry. The index is represented by the factor scores associated with the first principal component, presented in a standardized way in the interval [0, 1]. Firms with higher IAI should have lower information asymmetry. We hypothesize that a higher IAI is associated with better information quality Estimating Information Quality Using Models The relationship between information quality and information asymmetry is addressed first using three proxies for information quality (earnings persistence, conservatism and value relevance) that must be estimated using regression models. In these analyses, we first estimate the original models. Subsequently, we estimate models including the interactions of the explanatory variables with the IAI. Finally, we estimate alternative models including countrylevel variables related to the information environment, allowing us to investigate if the information environment is a relevant moderator of the relationship between information quality and information asymmetry. We use two variables to represent the information environment of each country. The Worldwide Governance Indicator (WGI jt ) computed by the World Bank reports aggregate governance indicators for each country j in year t, based on six dimensions: voice and accountability, political stability and absence of violence, government effectiveness, regulatory quality, rule of law and control of corruption. Each dimension ranges from 0 to 100 in each year, where the country with the best reputation in that dimension receives score 100 and the worst receives score 0. Therefore, similarly to Langbein and Knack (2010), we use Principal Component Analysis (PCA) to construct the WGI index from these six dimensions. Then, we present the WGI in a standardized form in the interval [0, 100], where greater WGI indicates better information environment. The second proxy for information environment is the Sovereign Risk Indicator (SRI jt ) issued by the major rating agencies, such as Standard & Poor s (S&P), that reflects country j s sovereign risk in year t (Oshiro & Saruwatari, 2005). Similarly to Oshiro and Saruwatari (2005), we recode these ratings as numbers, as follows: AAA = 17, AA+ = 16, AA = 15, AA- = 14,..., BB = 6, BB- = 5, B+ = 4, B = 3, B- = 2, CCC+ = 1, CCC and under = 0. Thus, we expect countries with better grades to have better information environment. It is important to note that 17 of the 22 countries in our sample adopted the International Financial Reporting Standards (IFRS) during the period analysed. Houqe et al. (2012) investigated the effects of both mandatory IFRS adoption and legal protection for the investor on information quality in 46 countries and found an increase in information quality after IFRS, but only in countries with stronger legal protection. This is consistent with Li and Yang (2016), who argue that the IFRS adoption can improve the information environment and benefit capital markets, possibly with different effects depending on the degree of legal protection of the country. For this reason, we run the analyses in two distinct periods, pre- IFRS and post-ifrs. 8

9 Earnings Persistence and Information Asymmetry Earnings persistence is assessed using the model of Dechow and Schrand (2004). We estimate the original (DS) model (eq. 1) and its modified versions by including the IAI (eq. 2) and a proxy for information environment (eq. 3), using fixed-effects estimation. For every firm i in country j, X ijt is the operating income in year t weighted by the total assets in year t 1; X ijt 1 is the operating income in year t 1 weighted by the total assets in year t 2; IAI ijt 1 is the Information Asymmetry Index in year t 1; IE jt 1 is an information environment proxy for country j in year t 1, alternatively WGI jt or SRI jt ; β 0i represents firm fixed-effects and γ t represents time fixed-effects, estimated using a set of year dummies; and ε ijt is the error term of the model. 9 X ijt = β 0i + β 1 X ijt 1 + γ t + ε ijt (1) X ijt = β 0i + β 1 X ijt 1 + β 2 IAI ijt 1 + β 3 (X ijt 1 IAI ijt 1 ) + γ t + ε ijt (2) X ijt = β 0i + β 1 X ijt 1 + β 2 IAI ijt 1 + β 3 (X ijt 1 IAI ijt 1 ) + β 4 IE jt 1 +β 5 (X ijt 1 IE jt 1 ) + β 6 (X ijt 1 IAI ijt 1 IE jt 1 ) + γ t + ε ijt (3) If earnings are persistent, β 1 should approach 1. This model is consistent with Dechow and Schrand (2004) and Dechow et al. (2010) because firms with more persistent earnings have greater predictive power of their earnings in an asset valuation model. In addition, if the earnings of firms with lower information asymmetry are more persistent, β 3 should be greater than zero (Richardson, 2000; Bhattacharya et al., 2013). Similarly, β 5 should be greater than zero if a better information environment leads to increased persistence (Cormier et al., 2013). Finally, the sign and magnitude of β 6 indicates how the association between earnings persistence and information asymmetry depends on the information environment of the country. Specifically, depending on the magnitudes of the estimates, β 6 should be greater than zero if there is a complementation effect, i.e., a better information environment predicting a more positive association between the IAI and persistence (Cormier et al., 2013). Analogously, a negative β 6 could imply, for example, a substitution effect, i.e., a better information environment predicting a less positive association between the IAI and persistence (Bhattacharya et al., 2013; Schiehll & Martins, 2016) Conservatism and Information Asymmetry To estimate the level of accounting conservatism, we employ the model proposed by Ball and Shivakumar (2005) (eq. 4), a modified version including the IAI (eq. 5) and two other specifications including alternative proxies for information environment (based on eq. 6). All models are estimated by fixed-effects estimation using panel data, where, every firm i in country j, ΔNI ijt is the net income change from year t 1 to year t, weighted by the total assets in year t 1; ΔNI ijt 1 is the net income change from year t 2 to year t 1, weighted by the total assets in year t 2; DΔNI ijt 1 is a dummy indicating if there is a negative change in net income from year t 1 to year t, assuming the value 1 if ΔNI ijt 1 < 0, and 0 otherwise. To save space we include IAI and its interactions with the other explanatory variables in the (column) vector W ijt 1 ; similarly, we include IAI, IE, and all their interactions with the other explanatory variables (and with each other) in the vector Z ijt 1 ; δ and γ are vectors of coefficients. NI ijt = β 0i + β 1 D NI ijt 1 + β 2 NI ijt 1 + β 3 (D NI ijt 1 NI ijt 1 ) + γ t + ε ijt (4)

10 NI ijt = β 0i + β 1 D NI ijt 1 + β 2 NI ijt 1 + β 3 (D NI ijt 1 NI ijt 1 ) + δ T W ijt 1 + γ t + ε ijt (5) NI ijt = β 0i + β 1 D NI ijt 1 + β 2 NI ijt 1 + β 3 (D NI ijt 1 NI ijt 1 ) + γ T Z ijt 1 + γ t + ε ijt (6) 10 In these models, the use of net income change as an explanatory variable has the advantage of providing an adequate specification to identify the transitory components of the earnings (Ball & Shivakumar, 2005). According to Ball and Shivakumar (2005), in more conservative firms there is greater reversal of losses in subsequent periods because these firms recognize losses in a more timely manner than they recognize gains. In the above models, the coefficient β 2 should either be equal to zero or lower than zero. β 2 = 0 if firms defer the recognition of earnings until the moment that cash flows are realized, meaning that positive results become a persistent component of net income, which tends not to be reversed (Ball & Shivakumar, 2005). On the other hand, β 2 < 0 if there is timely recognition, in which case positive results are a transient component of net income and tend to be reversed in subsequent periods. However, the timelier recognition of losses in comparison with gains implies that β 3 is lower than zero (β 3 < 0). Therefore, the timely recognition of losses implies transient decreases in the result and, consequently, tend to be reversed in the following periods, implying, a priori, that the sum of the coefficients β 2 and β 3 should be lower than zero (β 2 + β 3 < 0). It should be noted that there are no predictions for the parameters β 0 and β 1 in these models. To check if the firms with lower information asymmetry are more conservative, the key parameter is the coefficient of the interaction of D NI ijt 1, NI ijt 1, and IAI ijt 1. Moreover, to investigate the influence of the information environment on these relations, we add a set of interactions of these variables with the information environment proxies (WGI jt and SRI jt ). In this way, we expect that the main interaction coefficients involving IAI are negative, indicating higher conservatism (and quality) among firms with lower information asymmetry (Richardson, 2000; Brown et al., 2009; Bhattacharya et al., 2013). In addition, if the interaction coefficients involving IAI and IE are negative, the complementarity hypothesis is confirmed (Cormier et al., 2013); but, if they are positive, the substitution hypothesis is confirmed (Bhattacharya et al., 2013; Schiehll & Martins, 2016) Value Relevance and Information Asymmetry Collins, Maydew and Weiss (1997) (CMW, for short) argue that the value of a firm s equity can be expressed as a function of its earnings and book value. The value relevance of the accounting information may be captured by the explanatory power of regressions based on the CMV model (eq. 7), on the modified model including IAI (eq. 8), and on other two modified models including the proxies for the information environment (eq. 9), using panel data and fixed-effects estimation. In these equations, for every firm i in country j, P ijt is the price of the firm s stock three months after fiscal year-end t; E ijt is the earnings per share in year t; BV ijt is the equity book value per share at the end of year t; analogously to equations 5 and 6 above, W ijt 1 is a vector containing IAI and its interactions with the other explanatory variables; and Z ijt 1 is a vector containing IAI, IE, and all their interactions with the other explanatory variables (and with each other). P ijt = β 0i + β 1 E ijt + β 2 BV ijt + γ t + ε ijt (7) P ijt = β 0i + β 1 E ijt + β 2 BV ijt + δ T W ijt 1 + γ t + ε ijt (8) P ijt = β 0i + β 1 E ijt + β 2 BV ijt + γ T Z ijt 1 + γ t + ε ijt (9)

11 Analogously to CMW, the adjusted coefficients of determination from equations (7)- 2 2 (9) are denoted R CMW, R IAI and R 2 IE, respectively. We expect that models modified with the inclusion of information asymmetry and with the information environment present greater explanatory power. We use adjusted coefficients of determination to check this hypothesis because it penalizes the inclusion of irrelevant explanatory variables in the regression. Subsequently, we analyse the coefficients and we expect that the main interaction coefficients involving IAI are positive, because lower information asymmetry is associated with better information quality (Richardson, 2000; Brown et al., 2009; Bhattacharya et al., 2013). In addition, if the interaction coefficients involving IAI and IE are positive, the complementarity hypothesis is confirmed (Cormier et al., 2013); but, if they are negative, the substitution hypothesis is confirmed (Bhattacharya et al., 2013; Schiehll & Martins, 2016) Firm-level Proxies for Information Quality We investigate two firm-level proxies for information quality: income smoothing and earnings management Income Smoothing and Information Asymmetry The income smoothing proxy (IS) is computed for each firm following Eckel (1981), who assumes that earnings are a linear function of sales, while variable costs are constant over time. Eckel (1981) also assumes that fixed costs do not decrease, and gross sales cannot be smoothed artificially, so the coefficient of variation of sales (CVΔS%) should be lower than the coefficient of variation of earnings (CVΔE%). Algebraically, Eckel s model is given by IS CVΔE%/CVΔS%. According to Iñiguez and Poveda (2004), if IS < 0.9 the firm is assumed to smooth its earnings; if 0.9 IS 1.1 the firm is in the grey area and cannot be classified; and if IS > 1.1 the firm is assumed not to smooth its earnings. Thus, in this study IS ijt is a dummy that assumes value 1 if IS ijt < 0.9, and value 0 otherwise. To investigate how the binary proxy for income smoothing (IS ijt ) relates to our information asymmetry index and to the information environment proxies, we estimate the Logit models in equations 10 and 11 below, where p ijt represents the (conditional) probability that IS ijt = p ijt p ijt ln ( ) = β 1 p 0 + β 1 IAI ijt + γ t (10) ijt ln ( ) = β 1 p 0 + β 1 IAI ijt + β 2 IE jt + β 3 (IAI ijt IE jt ) + γ t (11) ijt As income smoothing is a desirable attribute of accounting information because it improves earnings predictability (Eckel, 1981), we expect a positive association between IS and IAI (β 1 ), indicating that better information quality is related with lower information asymmetry, as well as a positive association between IS and IE (β 2 ), indicating that in better information environment the information quality is greater (Richardson, 2000; Brown et al., 2009; Bhattacharya et al., 2013). In addition, if the interaction coefficient involving IAI and IE (β 3 ) is positive, the complementarity hypothesis is confirmed (Cormier et al., 2013); but, if it s negative, the substitution hypothesis is confirmed (Bhattacharya et al., 2013; Schiehll & Martins, 2016) Earnings Management and Information Asymmetry We compute a proxy for earnings management (EM) using the model by Dechow et al. (2012), which adjusts the Modified Jones model by including lagged accruals in order to

12 capture their natural reversal in subsequent periods. In equation 12, for every firm i in country j, the working capital accrual is calculated by WC_ACC ijt = ( CA ijt CL ijt Cash ijt + STD ijt )/A ijt 1, where CA ijt is the change in current assets from year t 1 to year t, CL ijt is the change in current liabilities from year t 1 to year t, Cash ijt is the change in cash from year t 1 to year t, STD ijt is the change in short-term debt from year t 1 to year t, and A ijt 1 is the total asset in year t 1. REV ijt is the change in revenues from year t 1 to year t, weighted by A ijt 1 ; REC ijt is the change in receivables from year t 1 to year t, weighted by A ijt 1 ; PPE ijt is the gross property, plant, and equipment, weighted by A ijt 1 ; and ε ijt is defined as the discretionary accruals. 12 WC_ACC ijt = β 0 + β 1 (1/A ijt 1 ) + β 2 ( REV ijt REC ijt ) + β 3 (PPE ijt ) + β 4 (WC_ACC ijt 1 ) + ε ijt (12) Specifically, we define EM ijt as the standard deviation of the squared residuals from the estimated model (eq. 12). We assume that firms with high standard deviation are more likely to manage earnings. To investigate how our earnings management proxy relates to our proxies for information asymmetry and information environment we estimate models 13 and 14 below using OLS estimation, which presents a better fit to these sub-samples. EM ijt = β 0 + β 1 IAI ijt + γ t + ε ijt (13) EM ijt = β 0 + β 1 IAI ijt + β 2 IE jt + β 3 (IAI ijt IE jt ) + γ t + ε ijt (14) According to Richardson (2000) and Bhattacharya et al. (2013), the existence of information asymmetry is a necessary condition for the practice of earnings management. For this reason, we expect a negative relationship between EM and IAI, as well as between EM and IE, indicating that in better information environment the information quality is greater. In addition, if the interaction coefficient involving IAI and IE (β 3 ) is negative, the complementarity hypothesis is confirmed (Cormier et al., 2013); but, if it s positive, the substitution hypothesis is confirmed (Bhattacharya et al., 2013; Schiehll & Martins, 2016) Information Asymmetry and Information Environment Finally, we explore the relationship between the firm-level information asymmetry index (IAI) and the country-level proxies for the quality of the information environment. In equation 15 below, estimated using the fixed-effects estimator, WGI jt is the Worldwide Governance Indicator for country j in year t and SRI jt is the Sovereign Risk Indicator for country j in year t. A set of year dummies is included to control for the effects of macroeconomic shocks that affect all firms equally in each year. IAI ijt = β 0i + β 1 WGI jt + β 2 SRI jt + β 3 (WGI jt SRI jt ) + γ t + ε ijt (15) According to Houqe et al. (2012) and Cormier et al. (2013), we expect a positive association between the information asymmetry index (IAI ijt ) and the quality of the information environment, whether considering the country s best governance environment (WGI jt ) or the country s lowest sovereign risk (SRI jt ), because the asymmetry tends to decline in stronger information environments.

13 4. Results and Discussion Table 2 presents the number of firm-year observations per country as well as descriptive statistics for the sample firms and countries. There is significant variation in the number of observations across countries due to differences in capital market development, country size, and availability of complete financial accounting data. China and India are the countries with the largest number of firms with available information, while the Czech Republic and Malaysia are the countries with the lowest number of firms in the sample. All financial information is displayed in millions of US dollars. To avoid the influence of outliers we present the medians of firm characteristics and proxies for the quality of the information environment. Table 2 also shows a substantial cross-country variation in firm size, age and analyst coverage. Moreover, there is substantial variation in the percentage of firms that trade depositary receipts, in the timing of the mandatory adoption of IFRS, and in the medians of sovereign risk and country governance indicators. Table 2 Descriptive statistics of sample firms and countries, from 2000 to Analysts Size Age Panel A: By Country # Obs. # Firms Coverage (median) (median) (median) DR (%) IAI (median) IFRS Adoption SRI (median) 13 WGI (median) Brazil 4, Chile 2, China 61,727 3, None Colombia 1, Czech Republic Egypt 3, None Greece 3, Hungary India 69,105 4, None Indonesia 7, None Malaysia Mexico 1, Peru 2, Poland 11, Qatar Russia 10, Saudi Arabia 2, None South Africa 3, South Korea 33,983 1, Taiwan 30,090 1, Turkey 4, United Arab Emirates Panel B: Full Sample Size Age Analysts DR IAI SRI WGI Coverage (%) Mean (median) th Percentile th Percentile th Percentile Note: The full sample consists of 256,581 firm-year observations for the fiscal years 2000 to 2016 across 22 countries and 15,111 non-financial firms. Panel A reports the information by country, including the total number of observations and firms, the median of selected firm characteristics, the percentage of firms in each country that trade DR, the timing of mandatory IFRS adoption, median sovereign risk and country governance quality indicators. Panel B reports the statistics for the full sample. # Obs. is the total number of observations. # Firms is the number of publicly traded firms in the sample. Size is the total assets at year-end (in US$ millions). Age is the number of years from the IPO. Analyst Coverage is the number of analyst forecasts per firm. DR is the percentage of firms that trade American Depositary Receipts (ADR) or Global Depositary Receipts (GDR) in a developed market. IAI is the Information Asymmetry Index, constructed using Principal Component Analysis and scaled between 0 and 1, where higher IAI indicates lower information asymmetry. IFRS Adoption indicates the year in which the full IFRS became mandatory. SRI is the Sovereign Risk Indicator for each country, from Standard & Poor s, coded as follows: AAA = 17, AA+ = 16, AA = 15, AA- = 14,..., BB = 6, BB- = 5, B+ = 4, B = 3, B- = 2, CCC+ = 1, CCC and under = 0 (Oshiro & Saruwatari, 2005). WGI is the Worldwide Governance Indicators index for each country, constructed using Principal Component Analysis and scaled between 0 and 100 (Langbein & Knack, 2010).

14 Considering the whole sample, the average firm has total assets of 938 million US dollars, has been a public company for about 11 years and has an average of 2 analyst forecasts per year. The country with the largest median firm size is Mexico (950 million US dollars), while Chile has the highest median firm age (20 years since the date of their IPO). Czech Republic (7) and Brazil (6) stand out in the median number of analyst forecasts. However, only 11 firms one should note that the median of the Czech Republic refers to, while in Brazil this figure is observed among 276 companies, indicating a better picture of the Brazilian market. Chile, China, Greece, Indonesia, Malaysia, South Korea and Taiwan show a median of analysts forecast equal to zero. Except for Malaysia, the other countries have several analysed firms that are equivalent to the size of their market. Therefore, the median zero can be explained by the fact that the sample includes many smaller firms that are not accompanied by analysts. South Africa (25%) followed by Mexico (24%) are the countries with the highest percentage of firms that trade ADR and GDR in developed markets, which can positively influence the information environment of these countries, due to the crosslisting effect (Chan et al., 2008). Apart from Malaysia, with only one firm, Qatar and Saudi Arabia do not have in this sample firms that trade depositary receipts. We construct the Information Asymmetry Index (IAI) via Principal Component Analysis (PCA) using the following firm-level characteristics: size, age, analyst coverage and DR trading in developed markets. These variables are available for most publicly traded firms in emerging markets and they are positively and significantly correlated with each other. Specifically, the IAI corresponds to the factor scores from the first principal component. Both the Kaiser-Meyer-Olkin measure (KMO = 0.651) and the Bartlett s test of sphericity (χ 2 = 49, , p < 0.01) suggest that our PCA procedure is adequate. The country with the highest median IAI in our sample is South Africa, which is also the country with the highest percentage of firms with DR traded in developed markets and ranks third in median firm age. As expected, IAI correlates positively with DR trading (0.707, p < 0.01), suggesting that cross-listing through depositary receipts reduces information asymmetry, in accordance with Chan et al. (2008). Similarly, there is a positive and significant correlation (0.313, p < 0.01) between firm age and IAI, in agreement with Roudaki et al. (2016), who argue that older firms tend to use time-tested disclosure policies and practices, are better established and closely followed by the market. Mandatory IFRS adoption in our sample countries starts in 2005 with the European Union member countries (Greece, Hungary and Poland), whereas the latest adoptions occur in 2015 (Colombia and the United Arab Emirates). The country with the highest median Sovereign Risk Indicator (SRI) is the United Arab Emirates (15, equivalent to AA), whereas the lowest median SRIs are from Indonesia and Turkey (5, equivalent to BB-). Malaysia had the highest governance indicator (WGI), equal to (from 0 to 100). It is important to note that this indicator is independent of the number of firms in the country. The country with the lowest median WGI is Russia (25.527). SRI and WGI are positively correlated (0.534, p < 0), suggesting that both measures correlate with a better information environment (Oshiro & Saruwatari, 2005; Langbein & Knack, 2010) Earnings Persistence, Conservatism, and Value Relevance In the empirical analysis reported below, we estimate all panel regression models separately in two periods: pre- and post-ifrs. The rationale is that mandatory full IFRS adoption may substantially affect the quality of accounting information (Houqe et al., 2012; Li & Yang, 2016) and the degree of information asymmetry of firms. We obtain similar results (unreported) when we estimate the models using the full sample period and including controls related to the IFRS adoption. All continuous variables included in the reported regressions were winsorized at 1%. 14