The world price of earnings opacity.
ABSTRACT: We analyze financial statements from 34 countries for the period 19841998 to construct a panel data set measuring three dimensions of reported accounting earnings for each country: earnings aggressiveness, loss avoidance, and earnings smoothing. We hypothesize that these three dimensions
Keywords: earnings opacity; earnings aggressiveness; loss avoidance; earnings smoothing; cost of equity; trading volume.
Data Availability: The data used in this study are publicly available from the sources indicated in the text.
JEL Classification: G15, G30, M41.
I. INTRODUCTION
The recent decline in equity values in the United States has been attributed to investor concerns over corporate governance and accounting. (1) This decline is consistent with investors perceiving a decreased informativeness in U.S. accounting numbers, and demanding higher rates of return to compensate for this additional component of informational risk. (2) The purpose of our paper is to investigate whether informational risk associated with accounting earnings impacts equity markets around the world. Of necessity, we do this using joint tests. We first measure distributional properties of accounting earnings that suggest poor correspondence between observable accounting earnings and unobservable economic earnings--which we dub "earnings opacity"--in a country every year, and we then determine whether our measures of earnings opacity in a country are associated with the return shareholders demand for holding equity in that country and are associated with shareholder trading of equity in that country. (3)
We define the earnings opacity of a country as the extent to which the distribution of reported earnings of firms in that country fails to provide information about the distribution of the true, but unobservable, economic earnings of firms in that country. As reported earnings of a particular firm in a country equals unobservable economic earnings plus a noise term, earnings opacity of a country is simply the average lack of informativeness of reported earnings in that country. (4)
Reported earnings in a country could be opaque because of a complex interaction among, at least, three factors: managerial motivation, accounting standards, and the enforcement of accounting standards (e.g., audit quality). It could be that earnings are opaque because managers are motivated to manipulate earnings, and they can do this either because accounting standards allow substantial flexibility, accounting standards do not exist to specify accounting principles related to some areas of business activity, or accounting standards, though rigorous, are weakly enforced. It could also be that earnings are opaque, not because managers manipulate earnings, but simply because accounting standards do not call for accounting treatments that transparently reflect underlying business activity and management is not willing or able to overcome these deficiencies by voluntarily providing more informative earnings reports.
Earnings opacity is inherently difficult to measure, particularly across countries, because it is not possible to pinpoint management's motives, and it is difficult to compare accounting standards and the enforcement of these accounting standards. In addition, it is not possible to capture all factors that might influence earnings opacity, or to model how they interact to produce more or less opaque earnings. So, instead of studying the inputs that determine earnings opacity, we analyze the outcome: the distributional properties of reported accounting numbers across countries and across time that suggest earnings opacity. Specifically, we use measures that are intended to capture three attributes of earnings numbers that could lead to earnings opacity: earnings aggressiveness, loss avoidance, and earnings smoothing. We focus on these three dimensions because the literature has suggested that these three dimensions may weaken the link between accounting performance and the true economic performance of a firm. We limit our analysis to industrial firms, so that differences in the underlying earnings process across different industry groups, and differences in the proportion of firms in various industry groups across countries and across time, do not affect the dimensions of reported earnings we examine. Finally, given the above mentioned difficulties in measuring earnings opacity, all our tests are inherently joint tests of two hypotheses: one, our three measures, or a composite of all three, are associated with uninformative or opaque earnings; and two, earnings opacity creates an informational risk that affects the cost of equity and trading in the stock market.
We construct a panel data set for each of these three measures of earnings opacity--earnings aggressiveness, loss avoidance, and earnings smoothing--and then combine them to obtain an overall earnings opacity time-series measure per country. We find that our estimates of average earnings opacity per country are significantly associated with variables that might impact the overall quality of a financial reporting regime of a country, namely the Center for International Financial Analysis and Research (CIFAR) disclosure index and the number of auditors per 100,000 population.
The second part of our paper goes on to investigate whether earnings opacity affects equity markets. We first examine the effect of our measures of earnings opacity on the return shareholders demand for holding equity (cost of equity). We measure the effect on the cost of equity employing two distinct approaches used in a companion paper (Bhattacharya and Daouk 2002). We discuss the details of these approaches, and their merits and demerits, in the next section of the paper.
The first approach is to extract the cost of equity from the dividend discount model. After controlling for other influences, we find in our panel data tests that the earnings aggressiveness dimension and overall earnings opacity have significant adverse effects on the cost of equity. An increase in our measure of overall earnings opacity from the 25th percentile rank to the 75th percentile rank is associated with a 2.8 percent increase in the cost of equity measured using dividend yields.
The second approach uses an international asset pricing factor model. This empirical specification allows for partial integration of a country to the world equity markets. After controlling for other influences, we find in our panel data tests that the loss avoidance dimension and overall earnings opacity have a significant effect on the cost of equity. An increase in our measure of overall earnings opacity from the 25th percentile rank to the 75th percentile rank is associated with a 3.2 percent increase in the cost of equity measured using this model.
Our last set of panel data tests examines the effect of earnings opacity on the level of trading. The details of the data set used to measure trade are discussed in the next section. After controlling for other influences, we find that earnings aggressiveness, earnings smoothing, and overall earnings opacity have significant adverse effects on trade. An increase in our measure of overall earnings opacity from the 25th percentile rank to the 75th percentile rank is associated with an 8.8 percent decrease in annual trade.
To summarize, we find that, after controlling for other influences, an increase in our measure of overall earnings opacity in a country is linked to an increase in the cost of equity and a decrease in trading in the stock market of that country. Both these effects are economically as well as statistically significant. However, our tests also reveal some inconsistencies in the relationship between our individual measures of earnings opacity and our cost of equity and trade measures. As all our tests are joint tests, this suggests that caution must be exercised in interpreting the links between our individual measures based on reported earnings and earnings opacity, as well as the links between these individual measures and the cost of equity and trade.
The rest of the paper is organized as follows. Section II provides the conceptual development underlying our empirical analyses. Section III discusses the methodological issues in the measurement of the earnings opacity variables as well as the stock market variables--cost of equity and trade. In Section IV we discuss the data and give some summary statistics. Section V contains the results of the main empirical tests. We conclude by discussing the implications and limitations of our analysis in Section VI.
II. CONCEPTUAL DEVELOPMENT
Bushman and Smith (2001), who call for more research using cross-country designs to explore the links between financial accounting information and corporate governance, identify three channels by which earnings opacity may affect financial markets. First, better accounting information helps investors distinguish between good and bad investments, which decreases estimation risk, which decreases the firm's cost of equity. Second, better accounting information helps investors distinguish between good and bad managers, which decreases agency costs, which decreases the firm's cost of equity. Third, earnings opacity, by weakening the link between reported accounting earnings and unobservable economic earnings, increases asymmetric information. An increase in asymmetric information leads to an increase in the adverse selection problem a liquidity provider faces when trading with insiders. The liquidity providers in such a market would protect themselves by increasing their sell price and decreasing their buy price. (5) This increases the transaction cost, which induces a shareholder to require an even higher return on equity for compensation,6 and also leads shareholders to trade less often or not trade at all. (7)
Our research investigates whether the three properties of reported accounting earnings we examine result in more opaque earnings, leading to increased information risk that is associated with increased cost of equity and decreased trading. Four assumptions underlie our above hypothesis. First, our measures actually measure what we claim they measure; that is, our measures of the three properties of reported accounting earnings are positively associated with how opaque, or uninformative, investors perceive reported accounting earnings to be. Second, markets are efficient in the sense that investors can detect the level of earnings opacity, but they cannot "see through" it. Third, the informational asymmetry created by earnings opacity is not completely resolved through some other communication mechanism, like alternate disclosures directed at large, affiliated stakeholders. Fourth, the informational risk caused by earnings opacity is an important factor relative to the other factors that affect equity markets, and so it is priced. None of these assumptions may hold. The above assumptions are tested in the second part of this paper.
A cross-country comparison of earnings opacity has many advantages over a cross-firm comparison of earnings opacity. First, because of considerable differences in accounting standards and audit quality across the globe, we can obtain an enviable dispersion in earnings opacity around the world. Second, as Bushman and Smith (2001) state, the cross-country differences in earnings opacity can be linked meaningfully to the cross-country differences in economic efficiency and institutional factors. A cross-country design also has some disadvantages. Two potential disadvantages--a more severe omitted variables problem and a more severe endogeneity problem--are discussed later in the paper.
Ours is not the first paper to exploit the advantages of cross-country comparisons. Our paper is a part of the growing international accounting literature that examines the value-relevance of accounting measures (Alford et al. 1993; Ali and Hwang 2000; Harris et al. 1994; Joos and Lang 1994; Land and Lang 2002). This value-relevance literature examines associations between economic income, measured using equity returns, and accounting data such as earnings. This literature reveals which countries experience greater associations between accounting data and equity returns, but it does not examine the potential consequences of variation in earnings informativeness on equity markets. Recent literature has examined international variation in analyst forecasts (Ashbaugh and Pincus 2001; Chang et al. 2000), but this literature also does not examine equity market consequences of this variation. Finally, examinations of earnings timeliness and conservatism (Ball, Kothari, and Robin 2000), or the effect of institutional factors on earnings management (Leuz et al. 2002), explore the potential causes of variation in earnings informativeness around the world, but again do not address the effects of this variation on equity markets. Our contribution to the above literature is that ours is the first paper, to our knowledge, that measures earnings opacity at a country level every year to form a panel data set and then uses panel data tests to check whether earnings opacity adversely affects the equity markets of that country. Our paper should be viewed as complementary to Leuz et al. (2002), who measure earnings management at a cross-sectional level across 31 countries and then explore whether institutional factors are linked to the cross-sectional differences in earnings management. Our paper should also be viewed as complementary to a recent survey conducted by PricewaterhouseCoopers (2001) that constructs a broad measure of opacity in a particular country, and links it to capital inflows and the country risk premium in sovereign bonds of that country.
III. VARIABLE SELECTION AND MEASUREMENT ISSUES
Earnings Opacity Measures
Earnings Aggressiveness Measure
Our first measure of earnings opacity is earnings aggressiveness. Ball, Kothari, and Robin (2000) argue that the opposite of aggressiveness, accounting conservatism, which is the more timely incorporation of economic losses versus economic gains into accounting earnings, arises to reduce information asymmetry. Specifically, they argue that three factors are expected to lead to accounting conservatism. First, accountants are aware that managers would like to report economic gains and suppress information about economic losses. Hence, accountants find negative information more credible, and are more likely to incorporate it into accounting income. Second, lenders are important users of financial statements, and lenders are more impacted by economic losses than by economic gains. Third, the timely incorporation of economic losses provides an important corporate governance role, providing quick feedback about bad investment decisions and strategies that managers may not wish to disclose. The first and third of these factors suggest that accounting conservatism is related to informativeness, since conservative accounting is expected to provide information that management might have incentives to withhold otherwise.
It is possible that earnings aggressiveness does not necessarily lead to earnings opacity. It could be argued that conservative accounting prevents good news from being transmitted quickly, thus adding noise. However, given that one might reasonably expect managerial incentives to overstate rather than understate earnings on average, our belief is that aggressive earnings are more opaque earnings, because such accounting reports are more likely to reflect biased and optimistic reporting on the part of management, adding noise to reported earnings and, hence, increasing earnings opacity. To understand these managerial motives, see, for example, Rangan (1998), Teoh et al. (1998), Shivakumar (2000), Healy (1985), Barth et al. (1999). Ultimately, whether or not earnings aggressiveness leads to earnings opacity is an empirical issue.
We use accruals to measure earnings aggressiveness. As earnings aggressiveness is the tendency to delay the recognition of losses and speed the recognition of gains, it implies that if cash flow realizations are held equal, then we would expect accruals to increase as earnings aggressiveness increases. For example, lower of cost or market rules, accounting procedures with a conservative bias, result in negative accruals. Aggressive accounting would be characterized by fewer such negative accruals that capture economic losses, and more positive accruals that capture economic gains, increasing the overall level of accruals. (8) Though it is true that unrealized gains and unrealized losses would eventually be recognized in accounting earnings in any clean surplus accounting system, the more conservative accounting system is expected to result in more negative accruals at any given point in time because a greater proportion of economic losses relative to economic gains will be reflected in accounting earnings at any point in time. This motivates us to measure earnings aggressiveness of a country at a point in time as the median for country i, year t, of accruals divided by lagged total assets. We use the median observation of scaled accruals to minimize the influence of extreme observations. The higher the median observation of scaled accruals of country i in year t, the higher the earnings aggressiveness. The effect of earnings aggressiveness on the distribution of accounting earnings vis-a-vis economic earnings is depicted in the Earnings Aggressiveness graph of Figure 1. (9)
[FIGURE 1 OMITTED]
Consistent with much of the past literature (e.g., Dechow et al. 1995; Healy 1985; Jones 1991; Leuz et al. 2002), we compute scaled accruals from balance sheet and income statement information, and then compute scaled cash flows as scaled operating income minus scaled accruals. We do not use information from the cash flow statement because of differences in the presentation of cash flow information across countries and time. In fact, many of our sample countries do not require the preparation or presentation of a statement of cash flows. (10) We define scaled accruals as:
(1) AC[C.sub.kt] = ([DELTA]C[A.sub.kt] - [DELTA]C[L.sub.kt] - [DELTA]CAS[H.sub.kt] + [DELTA]ST[D.sub.kt] - DE[P.sub.kt] + [DELTA]T[P.sub.kt],)/T[A.sub.kt-1]
where:
AC[C.sub.kt] = scaled accruals for firm k, year t;
[DELTA]C[A.sub.kt] = change in total current assets for firm k, year t;
[DELTA]C[L.sub.kt] = change in total current liabilities for firm k, year t;
[DELTA]CAS[H.sub.kt] = change in cash for firm k, year t;
[DELTA]ST[D.sub.kt] = change in current portion of long-term debt included in total current liabilities for firm k, year t;
DE[P.sub.kt] = depreciation and amortization expense for firm k, year t;
[DELTA]T[P.sub.kt] = change in income taxes payable for firm k, year t; and
T[A.sub.kt-1] = total assets for firm k, year t-1.
Loss Avoidance Measure
Our second measure of earnings opacity is loss avoidance behavior. Burgstahler and Dichev (1997) present persuasive evidence that U.S. firms engage in earnings management to avoid reporting negative earnings. Degeorge et al. (1999) provide evidence that suggests that the following hierarchy exists among three earnings thresholds: (1) avoiding negative earnings, (2) reporting increases in quarterly earnings, and (3) meeting analysts' earnings forecasts. As Burgstahler and Dichev (1997) and Degeorge et al. (1999) discuss, these results indicate that incentives to report positive earnings (i.e., beat a benchmark of zero earnings) exist for some sample firms. Such loss avoidance behavior obscures the relationship between earnings and economic performance, thus increasing earnings opacity.
We define firms with small positive earnings (small negative earnings) as firms with net income scaled by lagged total assets between 0 and 1 percent (between 0 and -1 percent). We find the ratio of the number of firms with small positive earnings minus the number of firms with small negative earnings divided by their sum. The higher is this ratio in country i, year t, the higher is the loss avoidance. The effect of loss avoidance on the distribution of accounting earnings vis-a-vis economic earnings is depicted in the Loss Avoidance graph of Figure 1.
Earnings Smoothing Measure
Our third measure of earnings opacity is earnings smoothing. Some accounting standards (for example, cases of high book/tax conformity) or some managerial motives (see, for example, Trueman and Titman 1988; Fudenberg and Tirole 1995) may lead to smooth earnings over time. If accounting earnings are artificially smooth, then they fail to depict the tree swings in underlying firm performance, thus decreasing the informativeness of reported earnings and, hence, increasing earnings opacity. This is consistent with the view of earnings smoothing taken in Leuz et al. (2002). An alternative view, as expressed in Zarowin (2002), is that earnings smoothing can be used by management as a means to convey information, potentially decreasing earnings opacity. While we believe that earnings smoothing at the country level is indicative of accounting that obscures information about economic volatility, whether or not earnings smoothing leads to earnings opacity and adverse capital market consequences is again an empirical issue.
Following Leuz et al. (2002), we find the cross-sectional correlation between the change in accruals and the change in cash flows, both scaled by lagged total assets, in country i, year t. Cash flows are obtained by subtracting accruals (which were obtained in Equation (1)) from operating earnings. Because some degree of earnings smoothing is a natural outcome of any accrual accounting process, this measure is expected to be negative on average. However, the more negative this correlation, the more likely it is that earnings smoothing is obscuring the variability in underlying economic performance, and the greater is the earnings opacity. The effect of earnings smoothing on the distribution of accounting earnings vis-a-vis economic earnings is depicted in the Earnings Smoothing graph of Figure 1.
Overall Earnings Opacity Measure
We rank all the raw time-series earnings aggressiveness median observations, across countries and years, into deciles, with higher ranks associated with greater earnings aggressiveness; we rank all the raw time-series loss avoidance ratios, across countries and years, into deciles, with higher ranks associated with greater loss avoidance; we rank all the raw time-series earnings smoothing correlations, across countries and years, into deciles, with higher ranks associated with greater earnings smoothing. Hence, each dimension of earnings opacity in a country each year is assigned a rank between 1 and 10, depending on which decile of the earnings opacity dimension distribution across all country-years that particular country for that year appears. For example, a rank of 3 for Australia in 1986 in the earnings smoothing dimension means that Australia in 1986 was in the third decile of all earnings smoothing measures across all countries and across all years. We then average the earnings aggressiveness rank, the loss avoidance rank, and the earnings smoothing rank in each country-year to obtain a time-series of overall earnings opacity for each country.
While there is no strong conceptual basis for aggregation into an overall earnings opacity measure, we perform the aggregation for two reasons. First, to the extent that each of our three earnings opacity dimensions measures the same underlying phenomenon with (uncorrelated) measurement error, our aggregate measure will suffer less from measurement error than each of the individual measures. Second, to the extent that each of our three earnings opacity dimensions measures a unique dimension of earnings opacity, then, intuitively, each separate dimension adds to the overall difficulty investors have in gleaning information from reported accounting earnings. We report all our results for each separate dimension of earnings opacity as well as for our aggregate measure.
To construct raw or rank cross-sectional measures of each individual dimension of earnings opacity per country, we average across all available years for each country the raw or the rank measure, respectively, of each dimension of earnings opacity. To construct rank cross-sectional measures of overall earnings opacity per country, we average across all available years the overall earnings opacity rank for each country.
Limitations of Our Earnings Opacity Measures
There are limitations of our measures of earnings opacity. First, we want to know how expected earnings opacity affects investor behavior, but expected earnings opacity is unobservable. We use lagged earnings opacity measures from year t-1 in all our tests to proxy for investor expectations about earnings opacity in year t, which implicitly assumes that investors observe earnings opacity after the fact, and their expectation of earnings opacity this year are based on their observation of last year's earnings opacity.
Second, it is possible that our measures of the distributional characteristics of reported earnings are impacted by factors that do not affect earnings opacity. For example, industry membership and growth may systematically impact the distributional characteristics of accruals, cash flows, and earnings, and these effects may be clear and predictable to investors, resulting in no effect on their perception of earnings opacity. As previously mentioned, we control for industry by limiting the sample to industrial firms. We control for growth by including the real GDP growth rates in all our empirical tests. These, however, may be over-controls if the variation in the distributional characteristics of reported earnings caused by industry membership and growth do indeed affect earnings opacity.
Third, while we have argued that earnings aggressiveness, loss avoidance, and earnings smoothing increase earnings opacity, counter arguments, mentioned above, also exist. Clear evidence linking our measures to informativeness in accounting numbers does not exist. Hence, our tests are tests of two joint hypotheses: our measures are correlated positively with earnings opacity, and earnings opacity affects cost of equity and trading in stock markets. Insignificant results could be due to a lack of a link between earnings opacity and cost of equity and trading, or it could be that our measures do not capture earnings opacity.
Stock Market Measures
Cost of Equity Measures
The cost of equity in a country is defined as the return shareholders require for holding shares in that country. This is an expectations variable, which we measure using ex post data. We employ two approaches used in a companion paper (Bhattacharya and Daouk 2002).
The first approach is to compute the cost of equity by backing it out from the classical constant growth dividend discount model. Bekaert and Harvey (2000, Appendix A) explore in great detail the relationship between dividend yields and the cost of equity for many models. Assuming that the best forecast for future growth rates in dividends is the most current dividend growth rate, which implies that we assume that dividend growth rates follow a random walk, it follows that the estimated cost of equity = current dividend yield x (1 + current growth rate of dividends) + current growth rate of dividends.
The advantages of using dividend yields to measure cost of equity are many. Dividend yields are observable, stable, and stationary. A sharp change in cost of equity should lead to a sharp change in dividend yields. The disadvantage of using dividend yields is that changes in dividend yields may come about because of repurchases of stock, and may come about because of changes in growth opportunities. The first factor is not much of a problem in emerging markets because repurchases are minor. We try to control for the second factor--growth opportunities--by including GDP growth rates as control variables.
If the earnings opacity variables have no incremental effect on the cost of equity, then those variables will be orthogonal to the above estimate of the cost of equity. We control for other influences.
The second approach to estimating the cost of equity accounts explicitly for risk. The international version of the capital asset pricing model does not hold up well in the data (see Harvey [1991] or Ferson and Harvey [1993]). The consensus seems to be that a country's beta with respect to the world market portfolio has some merit to explain expected returns for developed countries; the variance of return of the country's stock market does better in explaining expected returns for emerging markets (see Harvey 1995).
We adopt a simplified version of Bekaert and Harvey (1995) as our international asset pricing model. Their empirical specification allows for partial integration of a country to the world equity markets. Their model is very appealing because it permits a country to evolve from a developing segmented market (where risk is measured by the country's variance) to a developed country that is integrated to world equity markets (where risk is measured by the sensitivity of a country's equity returns to movements in the world market portfolio). The special case of complete integration, where the world factor is the only factor, is nested in their model. This international asset pricing model is expressed as follows:
(2) ([r.sub.i,t]) - ([r.sub.f,t]) = [[alpha].sub.0] + [[phi].sub.i,t][[lambda].sub.cov][h.sub.i,w,t] + (1 - [[phi].sub.i,t]))[[lambda].sub.var][h.sub.i,t] + [e.sub.i,t]
where:
[r.sub.i,t] = the dollar monthly return of the stock market index of country i at time t;
[r.sub.f,t] = the monthly return of the one-month U.S. T-Bill at time t;
[[alpha].sub.0] = a constant that would be estimated;
[[phi].sub.i,t] = a measure of the level of integration of country i at time t, 0 < [[phi].sub.i,t] < 1;
[[lambda].sub.cov] = the price of the covariance risk that would be estimated;
[h.sub.i,w,t] = the conditional covariance of the monthly return of the stock market index of country i with the monthly return of the world index at time t;
[[lambda].sub.var] = the price of own country variance risk that would be estimated (which we are restricting to be the same across all countries);
[h.sub.i,t] = the conditional variance of the monthly return of the stock market index of country i at time t; and
[e.sub.i,t] = the residual error term.
The independent variables in Equation (2)--conditional covariance [h.sub.i,w,t] and conditional variance [h.sub.i,t]--are separately estimated pair-wise for each country i and world pair from the multivariate ARCH model specified below:
(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where:
[r.sub.w,t] = the dollar monthly return of the stock market index of the world at time t;
[[epsilon].sub.i,t-j] = the innovation in monthly return of the stock market index of country i at time t-j, j [member of] {0,1,2,3};
[[epsilon].sub.w,t-j] = the innovation in monthly return of the stock market index of the world at time t-j, j [member of] {0,1,2,3 }; and
[h.sub.w,t] = the conditional variance of the monthly return of the stock market index of the world at time t.
Equation (3) was first introduced by Bollerslev et al. (1988). As in Engle et al. (1987), the weights of the lagged residual vectors are taken to be 1/2, 1/3, and 1/6, respectively. The constants [a.sub.2], [b.sub.2], and [c.sub.2] are constrained to be identical for all country-word pairs. Maximum likelihood is used to estimate Equation (3).
The other independent variable in Equation (2)--[[phi].sub.i,t]--measures the level of integration of country i at time t. We define it as follows:
(4) [[phi].sub.i,t] = exp ([a.sub.1] ([exports.sub.i,t] + [imports.sub.i,t]))/1 + exp ([a.sub.1] ([exports.sub.i,t] + [imports.sub.i,t]/gd[p.sub.i,t])).
The definition of [[phi].sub.i,t] in Equation (4) implies that it is a function of the ratio of the sum of exports and imports to gross domestic product. It is designed to take on values between 0 and 1. When its value is 0, the country is not integrated with world equity markets, and its equity is exposed only to local risk (own variance). When its value is 1, the country is fully integrated with world equity markets, and its equity is exposed only to global risk (covariance with world factor). Bekaert and Harvey (1997) find that increases in this ratio are empirically associated with increased importance of the word factor relative to local risk factors.
We use a two-step procedure (first remove the effect of risk, and then test the effect on residuals) instead of using a one-step procedure (include all independent variables in Equation (2) directly). We do so because of technical convergence problems in the one-step nonlinear estimation procedure. If the earnings opacity variables have no incremental effect on the cost of equity, then those variables will be orthogonal to the residuals from the model in Equation (2). We control for other influences on this residual. The advantage of using a well-specified asset pricing factor model like Equation (2) to measure cost of equity is that we explicitly account for risk. This comes at a price. Recall that all the independent variables in Equation (2) are estimates from other models. This introduces estimation error, which may introduce bias, and it reduces power.
Trade Measure
A good metric to capture the amount of trade in a market is turnover, which is defined as the ratio of volume of dollar trade per month to dollar market capitalization at the end of the month. To mitigate the effect of outliers, which occur because the denominator is small in some countries, we take the natural logarithm of this ratio.
IV. DATA AND DESCRIPTIVE STATISTICS
Earnings Opacity Measures
The data used in constructing the earnings opacity variables come from the Worldscope database for the years 1984 through 1998. We restrict the sample to industrial firms (SIC codes 2000-3999 and SIC codes 5000-5999) to increase the homogeneity of our sample across countries and across time. Since the underlying earnings process being represented by accounting earnings is similar for industrial firms, this restriction reduces the probability that the cross-country differences and time-differences we observe in our earnings opacity measures are caused by the difference in or changes in industrial composition in our sample. This sample restriction is also consistent with much of the accounting literature (e.g., Alford et al. 1993; Ali and Hwang 2000). (11) Because all our tests are panel data tests, we include countries that have data for more than three years, and have more than 20 firms per year. This yields 58,653 firm-year observations from 34 countries, spanning the years 1986 through 1998. (We lose 1984 and 1985 because the calculation of changes in accruals and cash flows requires data from years t-1 and t-2.)
The names of the countries for which we have data are given in Column (1) in the Appendix, the sample period for each country is given in Column (2), and the number of firm-years per country is given in Column (3). For each firm-year, we use the following variables from Worldscope: cash, total current assets, total current liabilities, income taxes payable, current portion of long-term debt included in total current liabilities, depreciation and amortization expense, operating income, net income, and total assets. Some firms do not have information on income taxes payable or on the current portion of long-term debt included in total current liabilities. Similar to Leuz et al. (2002), if these variables are missing, then we assume them to be zero. We include observations with fiscal years ending between July 1 of year t and June 30 of year t + 1 in the calculation of our earnings opacity variables for year t. So, for example, observations with fiscal years ending between July l, 1995 and June 30,, 1996 are considered year 1995 observations.
Descriptive information on each of the raw earnings opacity variables for each sample country is provided in Columns (2) through (4) of Table 1. Each column gives the average across the available years for each country for each measure. Column (2) provides the average accruals divided by lagged total assets for our sample countries. As expected, average accruals are negative, averaging about 2 percent of lagged total assets. Interestingly, three of the 34 countries in our sample--Greece, India, and Turkey--have positive accruals. The loss avoidance measure is presented in Column (3). Avoidance of small negative bottom-line earnings is observed in 32 of our 34 countries, implying that this is a global phenomenon. Finally, the earnings smoothing measure--the average cross-sectional correlation between the change in cash flows and the change in accruals--is presented in Column (4). As expected, the correlation is strongly negative in every country in our sample.
Other Financial Reporting Measures
We first analyze the relation between our cross-sectional measures of earnings opacity and alternative cross-country measures related to financial reporting quality that have been documented in the past literature. As discussed previously, we expect earnings opacity to be a complex function of at least three factors: accounting standards, enforcement of accounting standards, and managerial motivation. We identify four measures from the prior literature that might be related to earnings opacity through these three factors.
The first measure is the number of auditors per 100,000 population. The number of auditors per 100,000 population comes from Saudagaran and Diga (1997, Table 6). The original source is communication with the International Federation of Accountants (IFAC) Secretariat on August 13, 1996. This variable is intended to capture the enforcement of accounting standards, with enforcement rising and earnings opacity declining as the proportion of auditors in the population rises. Column (5) in Table 1 gives this variable. As our raw measures for earnings aggressiveness and loss avoidance increase and our raw measure for earnings smoothing decreases as earnings opacity increases, auditors per 100,000 population is expected to have a negative relationship with our measures of earnings aggressiveness and loss avoidance, and a positive relationship with our measure of earnings smoothing.
We also use two measures that capture aspects of the accounting standards themselves. The first such measure is a disclosure level variable that comes from Saudagaran and Diga (1997, Table 2). The original source is CIFAR (1995). It represents a disclosure score based on the inclusion of 90 items as required disclosures in annual reports for each country. Higher scores correspond to greater required disclosure. We expect that greater disclosure requirements in accounting standards will enhance the informativeness of earnings by reducing the manager's ability to manipulate earnings, thus decreasing earnings opacity. Column (6) in Table 1 gives this variable. As disclosure and earnings opacity are expected to be negatively correlated, we expect this variable to have a negative relationship with our measures of earnings aggressiveness and loss avoidance, and a positive relationship with our measure of earnings smoothing.
The second measure of accounting standards is the extent of compliance with International Accounting Standards (IAS). These data come from Choi et al. (1999, Exhibit 8.6). The original source is the International Accounting Standards Committee (IASC 1997). We assign a score of 0 to all countries that independently produce accounting standards and do not use international accounting standards as the basis for those standards (this corresponds to categories F and G in Choi et al. [1999]). We assign a score of 1 to all countries that use international accounting standards as the basis for their separately developed accounting standards, but promulgate some standards that offer more or less choice than international accounting standards (this corresponds to category E in Choi et al. [1999]). Finally, we assign a score of 2 to all countries that adopt international accounting standards with few, if any, modifications beyond additional explanatory material (this corresponds to categories A through D in Choi et al. [1999]). Column (7) in Table 1 gives this variable. The relationship of this variable to our earnings opacity measures depends on whether international accounting standards produce more or less informative earnings than local standards.
The final measure is the legal origin of the country. Ball, Kothari, and Robin (2000) argue that common-law countries have a demand for more transparent earnings. Further, Leuz et al. (2002) argue that legal protection of outside investors, which is greater in common-law countries, decreases incentives for earnings management. Column (8) in Table 1 gives this variable. These data come from the CIA World Factbook (2001). The above arguments and the above coding suggest a negative relationship between legal origin and our measures of earnings aggressiveness and loss avoidance, and a positive relationship with our measure of earnings smoothing.
Table 2 presents a correlation matrix between our earnings opacity variables and each of the above four variables. Loss avoidance decreases as the number of auditors per 100,000 population increases, while earnings aggressiveness and earnings smoothing have insignificant, but correctly signed, relationships with the number of auditors. Earnings aggressiveness and loss avoidance decrease as disclosure level increases; however, there seems to be no link between earnings smoothing and the disclosure variable. While this last result may seem surprising since lower disclosure may be necessary for earnings smoothing to be possible, it is likely that the demand for smooth accounting numbers and disclosure are related, thus obscuring the supply-side effect. Interestingly, there seems to be little link between legal origin and our earnings opacity variables, and little link between the extent of use of international accounting standards and our earnings opacity variables. This suggests that the use of international accounting standards does not help in making earnings numbers more transparent. This conclusion is similar in spirit to the conclusion of Ball, Robin, and Wu (2000). However, since the extent of the use of international accounting standards is a difficult variable to measure accurately, our finding of no link between international accounting standards and earnings opacity could simply be due to measurement error.
Table 2 also presents the correlation between each of our three earnings opacity variables. They range in absolute value from 0.15 to 0.45, indicating that though there is some relationship between our three earnings opacity variables, there is a distinct component to each measure. These correlations have signs consistent with our measures of earnings aggressiveness and loss avoidance being positively related to earnings opacity and our measure of earnings smoothing being negatively related to earnings opacity, with one exception. There is a significant positive relation between earnings aggressiveness and earnings smoothing. This suggests that more conservative accruals (i.e., more negative accruals) results in smoother earnings (i.e., more negative correlation between changes in accruals and changes in cash flows), and is consistent with growth producing both more negative accruals and a stronger negative correlation between changes in cash flows and changes in accruals. To the extent that there is a mechanical relationship between our measures of earnings aggressiveness and earnings smoothing that does not relate to earnings opacity, both of these measures capture earnings opacity with error, biasing our coefficient estimates toward zero.
Table 3 provides the rankings of earnings opacity across the countries in our sample for each of the three dimensions of earnings opacity we identify, and for overall earnings opacity. The United States has the least amount of earnings opacity, followed by Norway. Greece, South Korea, and Indonesia show the most severe earnings opacity in our sample.
Stock Market Measures
Data on monthly equity indices of 20 developed countries were obtained from Morgan Stanley Capital International (MSCI). Data on monthly equity indices of 14 emerging markets were obtained from International Financial Corporation (IFC). The fourth column in the Appendix gives the sample period that was available for these 34 monthly stock market indices in the 1986-1998 period. These indices are value-weighted, and are calculated with dividend reinvestment. As noted by Harvey (1991), the returns computed on the basis of these indices are highly correlated with popular country indices. The MSCI value-weighted World Index was used as a proxy for the world market portfolio.
We computed monthly returns of each country's stock market and the world market portfolio from these indices. These returns are used in our international asset pricing factor model. The ninth column in Table 1 gives the mean return scaled by the standard deviation of returns per country in the 1986-1998 sample period (some countries do not have data for the full period).
We obtained monthly data on the dividend yield for 32 of the 34 countries from the vendor Datastream. The dividend yield was on the Datastream-constructed indices. The seventh column in the Appendix gives the sample period that was available for these 32 monthly dividend yield time-series.
The measure of trading that we adopted was turnover, which is defined as the ratio of the volume of trade in the stock market to the market capitalization of the stock market. We took the natural logarithm of this ratio. We could obtain monthly data on the volume of trade and market capitalization for 30 of the 34 countries from Datastream. The fifth and sixth column in the Appendix gives the sample period that was available for these 30 monthly market capitalization and volume time-series. The tenth column in Table. 1 gives the mean of this variable per country in the 1986-1998 sample period (some countries do not have data for the full period).
Bekaert and Harvey (1997) divide the sum of exports and imports by the country's gross domestic product to obtain a variable that captures the level of integration of a country with the rest of the world. This is because the level of globalization does affect the cost of equity (see Stulz 1999a). We follow the same method. Monthly data on exports and imports for the 34 countries were obtained from the International Financial Statistics provided by the International Monetary Fund. For some countries the frequency of GDP was quarterly, and for some it was yearly. To obtain monthly GDP, we divided by 3 in the former case, and by 12 in the latter case. The eighth, ninth, and tenth columns in the Appendix give the sample period that was available for these 34 GDP, exports, and imports time-series, respectively.
As purchasing power parity is not observed in the data, standard international asset pricing models like Ferson and Harvey (1993) and Dumas and Solnik (1995) have a foreign exchange factor (FX factor). We include this control in our international asset pricing factor model as well. Monthly data on foreign exchange rates are obtained from the International Financial Statistics. The eleventh column in the Appendix gives the sample period that was available for these 34 monthly foreign exchange rate time-series.
As discussed before, our measures of earnings opacity may be biased against countries that exhibit fast economic growth. To control for this, we use real GDP growth as another independent variable in our panel data tests. GDP growth data comes from the World Bank. The average GDP growth exhibited during 1985-1998 in each of our 34 countries is documented in the eleventh column in Table 1.
Bhattacharya and Daouk (2002) document that the enforcement of insider trading laws reduces the cost of equity of a country. We obtain the insider trading enforcement date from Bhattacharya and Daouk (2002, Table 1). These are given in the twelfth column in Table 1. We control for the confounding effects of insider trading enforcement in all our tests.
When a country opens up its capital markets to foreigners, the cost of equity is reduced through two routes (Stulz 1999b). It reduces required return because risk sharing improves, and it reduces required return because corporate governance improves. Bekaert and Harvey (2000) and Henry (2000) confirm empirically that such liberalization reduces the cost of equity. We obtain official liberalization dates from Bekaert and Harvey (2000, Table I). These are given in the thirteenth column in Table 1. We control for the confounding effects of liberalization in all our tests.
V. DOES EARNINGS OPACITY AFFECT STOCK MARKETS?
We explore the effect of earnings opacity on two dimensions of an equity market in a country--the return the shareholders demand and how much shareholders trade. As can be seen from the descriptive statistics in Table 1, there is significant variation in these two variables among equity markets across the world. An attempt to answer whether earnings opacity across countries causes some of this variation has to address two challenges: missing explanatory variables and endogeneity.
The missing explanatory variables problem is serious. The differences in equity markets across the world come about because of a number of differences in country characteristics, not just because of earnings opacity. It could be further argued that some of these country characteristics, like its economic, political, and legal infrastructure, have a bigger influence on the stock market of the country than how opaque earnings are in that country. (12) It could be even further argued that it is impossible to control all these factors in cross-sectional tests.
The endogeneity problem is even more serious. It is possible that changes in institutional factors within a country intended to facilitate capital formation simultaneously impact the properties of reported accounting numbers and equity market measures such as the cost of equity capital and trading volume. As an example, a substantial commitment of government resources to securities regulation could result in less opaque accounting earnings as firms try to avoid increased regulatory scrutiny, as well as improved equity market performance due to improved corporate governance practices. In such circumstances, our tests could reveal a spurious association between earnings opacity and equity market measures, as all are impacted by a third variable. This limits our ability to draw clear causal inferences from our empirical analyses.
The above are valid criticisms. However, we believe that two crucial features of our research design substantially mitigate the concern that our empirical analyses are impacted severely by these two problems. First, all our tests are panel data tests corrected for country-fixed effects, country-specific heteroscedasticity, and country-specific autocorrelation. (13) Therefore, though it may be true that many country-specific factors impact the stock market and earnings opacity, as long as these country-specific factors remain stable during our period of study, their inclusion or non-inclusion has no effect on the coefficient estimates in panel data tests with the above corrections.
These panel data tests with fixed country effects also minimize the endogeneity problem. Himmelberg et al. (1999) argue that a similar endogeneity problem arises in studies that attempt to relate managerial ownership and firm performance. Specifically, they note that since firm performance and the level of managerial ownership both depend on observable and unobservable firm characteristics, ordinary least squares cannot be used to produce an unbiased estimate of the relation between ownership and performance. They also note that an instrumental variables solution to the problem is difficult to implement because it is difficult to find identifying instruments. If, however, the unobserved source of the endogeneity is constant over time, Himmelberg et al. (1999) note that panel data with fixed effects (in their case, firm-level fixed effects; in our case, country-level fixed effects) effectively eliminates the potential bias caused by endogeneity. They also note that if the primary source of measurement error is across rather than within (in our case, countries), then the bias caused by measurement error is reduced by our estimator. Hence, if the features that could create a endogeneity problem with earnings opacity and equity market measures are relatively constant through time (an explicit assumption over the decade of analysis in Leuz et al. [2002]), then our panel data tests with fixed country effects control for this endogeneity bias. (14)
Alternatively, it is possible that the missing country-specific variables or the institutional features creating simultaneity between the observed properties of accounting numbers and equity market measures such as cost of equity and trading volume do change over the period of our analyses. If so, then a second feature of our research design reduces the potential endogeneity bias caused by changing institutional features. As previously discussed, our empirical specification uses our earnings opacity measures lagged by one year. This specification, in addition to providing a measure of expected earnings opacity, has a side advantage: it mitigates endogeneity. In order for our test statistics to suffer from endogeneity bias, we have to assume that the institutional factors that could impact both earnings opacity and equity market measures change over time and that reported earnings numbers reflect these changes at time t-1, while the equity market impact is observed at time t. Since accounting changes are slower to occur than are equity market changes, this does not seem a very plausible assumption. It seems far more likely that the accounting measure response to institutional changes would lag the market response. (15) We conduct two additional sensitivity analyses, which we discuss later, to further allay concerns that our results are attributable to omitted variables or endogeneity.
Cost of Equity
Using Dividend Yields
As discussed before, we can back out the cost of equity from the dividend discount model. If we further assume that dividend growth rates follow a random walk, then the estimated cost of equity = current dividend yield x (1 + current growth rate of dividends) + current growth rate of dividends.
Using this estimate of the cost of equity as the dependent variable, we run four panel time-series regressions with country-fixed effects. Model (1) uses the "earnings aggressiveness" rank measure as the independent variable; Model (2) uses the "loss avoidance" rank measure as the independent variable; Model (3) uses the "earnings smoothing" rank measure as the independent variable; whereas Model (4) uses the "overall earnings opacity" rank measure as the independent variable. We correct for country-specific heteroscedasticity and country-specific autocorrelation in each case. As liberalization and insider trading enforcement have been empirically shown to affect the cost of equity, and as these institutional variables did change during our period of study (see Columns (12) and (13) in Table 1), we use an indicator for liberalization and an indicator for insider trading enforcement as control variables in each case. As discussed before, we also control for GDP growth rates. Note that institutional variables that did not change need not be included as controls, because in a panel time-series regression with fixed-effects, they will have no effect. The panel regressions use data for the 32 countries for which we have dividend yield data from January 1986 to December 1998 (some countries do not have data for the full time period).
Table 4 presents the results from this panel time-series regression. The coefficient of the overall earnings opacity measure (Model (4)) is positive and statistically significant at the 5 percent level. A detailed look at Models (1), (2), and (3) reveals that this significance is coming from the earnings aggressiveness variable, although the coefficients on the other earnings opacity variables have the right sign. This is consistent with our joint hypotheses that our earnings aggressiveness measure and our overall measure are correlated with earnings opacity, and earnings opacity adversely affects the cost of equity. The association is also economically significant. An increase in overall earnings opacity from the 25th percentile rank to the 75th percentile rank is associated with a 2.8 percent increase in the cost of equity. (16) The coefficient on the insider trading enforcement variable has the right sign and is statistically significant, implying that insider trading enforcement causes the cost of equity to drop as seen in Bhattacharya and Daouk (2002). The liberalization indicator has the correct negative sign, but is not significantly different from zero at conventional levels. GDP growth rates are positively related to this measure of the cost of equity.
Using an International Asset Pricing Model
We estimate Equation (2) using nonlinear least squares. The regressions use data for our 34 countries from December 1986 to December 1998 (some countries do not have data for the full time period). The results are given in Panel A of Table 5.
Panel A of Table 5 reveals that though covariance risk seems to have a positive price ([[lambda].sub.cov] is positive), the estimates are statistically significant only at the 11 percent level. It also reveals that though own country variance risk has a positive price ([[lambda].sub.var] is positive), the estimates are statistically significant only at the 12 percent level. These results contrast with the results of Bhattacharya and Daouk (2002), who use the same estimation technique and obtain statistical significance, but that is because their estimation was carried out for a longer 1969-1998 sample period.
Using the residuals from Equation (2) as the dependent variable, we run four panel time-series regressions with country-fixed effects. Models (1), (2), (3), and (4) are as previously defined. We correct for country-specific heteroscedasticity and country-specific autocorrelation in each case. We control for liberalization, insider trading enforcement, and GDP growth as before. We control for two other sources of risk that have been documented in the literature--foreign exchange risk (Dumas and Solnik 1995; Ferson and Harvey 1993) (17) as well as liquidity risk (Brennan and Subrahmanyam 1996) (18)--and that change in our sample period.
Panel B of Table 5 presents the results from this panel time-series regression. The coefficient of the overall earnings opacity measure (Model (4)) is positive and statistically significant at the 8 percent level. A detailed look at Models (1), (2), and (3) reveals that this significance is coming from the loss avoidance variable, in contrast with the results from the dividend yield model where the significance was primarily driven by the earnings aggressiveness variable. This is consistent with our joint hypotheses that our loss avoidance measure and our overall measure are correlated with earnings opacity, and earnings opacity adversely affects the cost of equity. Insider trading enforcement and GDP growth are insignificantly related to the cost of equity in this specification, though both variables have the same sign as in the dividend yield specification. The liberalization indicator again has a negative sign, and is statistically significant in this specification. An increase in overall earnings opacity from the 25th percentile rank to the 75th percentile rank is associated with a 3.2 percent increase in the cost of equity. (19) It should be noted here that the point estimates obtained by two completely different methods of estimating the cost of equity--the dividend yield method (implicitly controls for risk, but has less estimation risk) and an international asset pricing model (explicitly controls for risk, but has more estimation risk)--are very similar: 2.8 percent and 3.2 percent, respectively.
Trading
The measure of trade is turnover, which is defined as the ratio of volume of trade to market capitalization. Using the natural logarithm of this ratio as the dependent variable, we run four panel time-series regressions with country-fixed effects. Models (1), (2), (3), and (4) are as previously defined. We correct for country-specific heteroscedasticity and country-specific autocorrelation in each case. We control for liberalization, insider trading enforcement, and GDP growth as before. The panel regressions use data for the 30 countries for which we have trading data from January 1986 to December 1998 (some countries do not have data for the full time period).
Table 6 presents the results from this panel time-series regression. Except for Model (2) whose coefficient is insignificant, the coefficients of the individual earnings opacity measures (Models (1) and (3)) and the coefficient of the overall earnings opacity measure (Model (4)) are negative and statistically significant at the 5 percent level. This result is consistent with our joint hypotheses that our earnings aggressiveness measure, our earnings smoothing measure and our overall measure are correlated with earnings opacity, and earnings opacity adversely affects trading volume. An increase in overall earnings opacity from the 25th percentile rank to the 75th percentile rank is associated with an 8.8 percent decrease in annual trade. The coefficients on liberalization and insider trading enforcement are significant, and have the right sign.
Further Robustness Checks
Despite two features of our research design that we believe help mitigate concerns over omitted explanatory variables and endogeneity--panel data tests and use of lagged opacity variables--we run two tests to further allay concerns. (20) First, we reran all our tests including lagged equity market development, measured as stock market capitalization divided by GDP, as an additional control variable. This variable is measured on an annual basis and comes from Datastream. While there is no direct conceptual basis for including this variable in equations explaining either cost of equity or trade, we include this measure of equity market development as a proxy for unobserved institutional factors that might promote the development of equity markets within a country and thus impact both earnings opacity and our stock market variables. We find that equity market development is unrelated to either of our measures of the cost of equity capital, but it is positively related to trade. Our previous inferences on the effect of earnings opacity variables on cost of equity or trade are not affected by the addition of this control variable.
Second, we estimated a Vector Auto Regression (VAR) model proposed by Sims (1980) in order to explicitly model earnings opacity, the cost of equity, and trade as endogenously determined dependent variables. The endogenous variables are modeled as linear functions of lagged endogenous variables and all exogenous variables in the system. The system of equations in the VAR is estimated jointly. This means that the effect of the independent variables on each endogenous variable takes into account the endogenous nature of the other endogenous variables. (21)
Formally, the system of equations to estimate the effect on the cost of equity is:
(5) Cost of [Equity.sub.i,t] = [[beta].sub.10] + [[beta].sub.11] Dimension of Earnings [Opacity.sub.i,t-1] + [[beta].sub.2] [Liberalization.sub.i,t] + [[beta].sub.13] Insider Trading [Enforcement.sub.i,t] + [[beta].sub.14] GDP [Growth.sub.i,t] + [[beta].sub.15] Market Capitalization/GD[P.sub.i,t-1] + [u.sub.1i,t]
and
(6) Dimension of Earnings [Opacity.sub.i,t] = [[beta].sub.20] + [[beta].sub.21] Cost of [Equity.sub.i,t-1] + [[beta].sub.22] [Liberalization.sub.i,t] + [beta].sub.13] Insider Trading [Enforcement.sub.i,t] + [[beta].sub.14] GDP [Growth.sub.i,t] + [[beta].sub.15] Market Capitalization/GD[P.sub.i,t-1] + [u.sub.2i,t].
and the system of equations to estimate the effect on trade is:
[Trade.sub.i,t] = [[beta].sub.10] + [[beta].sub.11] Dimension of Earnings [Opacity.sub.i,t-1] + [[beta].sub.12] [Liberalization.sub.i,t] + [[beta].sub.13] Insider Trading [Enforcement.sub.i,t] + [[beta].sub.14] GDP [Growth.sub.i,t] + [[beta].sub.15] Market Capitalization/GD[P.sub.i,t-1] + [u.sub.1i,t]
and
(6) Dimension of Earnings [Opacity.sub.i,t] = [[beta].sub.20] + [[beta].sub.21] [Trade.sub.i,t-1] + [[beta].sub.22] [Liberalization.sub.i,t] + [[beta].sub.23] Insider Trading [Enforcement.sub.i,t] + [[beta].sub.24] GDP [Growth.sub.i,t] + [[beta].sub.25] Market Capitalization/GD[P.sub.i,t-1] + [u.sub.2i,t].
The system of equations is estimated jointly using Seemingly Unrelated Regressions (SUR). SUR computes estimates using the technique of joint GLS (Generalized Least Squares). The two error terms [u.sub.1i,t] and [u.sub.2i,t] are allowed to be correlated (see Enders [1996] for further details). The estimation allows for country-fixed effects, for country-specific heteroscedasticity, and for country-specific autocorrelation.
We find that endogeneity does exist. Overall earnings opacity as well as all dimensions of earnings opacity, except earnings smoothing, are positively affected by the cost of equity using the dividend yield method. Earnings aggressiveness and earnings smoothing are affected by trade, but loss avoidance as well as overall earnings opacity are not affected by trade. However, though endogeneity exists and we explicitly account for it, our previous inferences about the effect of earnings opacity variables on cost of equity (using the dividend yield method) or trade are not affected. (22)
VI. CONCLUSION
This paper explores the link between earning opacity and cost of equity and trading volume in a broad cross-section of countries. We attempt to measure earnings opacity from the financial statements of firms. We use distributional properties of reported earnings to estimate for each country, for each year, three dimensions of earnings opacity: earnings aggressiveness, loss avoidance, and earnings smoothing. We combine these three dimensions to obtain an overall earnings opacity time-series measure per country. While not all results are consistent for our individual earnings opacity measures, we document in our panel data tests that, after controlling for other influences, an increase in overall earnings opacity in a country is linked to an increase in the cost of equity and a decrease in trading in the stock market of that country.
Our analyses have important limitations that should be kept in mind when interpreting our results. First, it is possible that earnings opacity, the cost of equity capital, and trading volume are all impacted by some unknown third variable, resulting in a spurious association between earnings opacity and our equity market measures. While we have attempted to control for the factors suggested by the past literature, theoretical and empirical limitations prevent us from knowing whether all of the important influences have been controlled. While our research design mitigates concerns over the endogeneity of earnings opacity, we are unable to ensure that endogeneity does not impact our analyses. Second, we undoubtedly measure earnings opacity with error, and this measurement error could impact our analyses. Finally, our cross-sectional tests relating earnings opacity to various factors that might influence financial reporting quality are cross-sectional tests rather than panel data tests. As such, these tests lack power due to the lack of data, and are particularly susceptible to bias caused by correlated omitted variables.
Keeping these limitations in mind, there are important implications of our analyses for investors, securities regulators, and academics. Our cross-sectional analysis documents associations between the proportion of auditors in the population and disclosure level and earnings opacity, suggesting that increased enforcement of accounting standards through auditing, and increased disclosure may improve earnings transparency. We document economically and statistically important relations between our measures of the distributional properties of reported earnings and the cost of equity capital and trade. These findings are consistent with the joint hypotheses that our measures are associated with opaque earnings, and that investors perceive risk associated with opaque earnings. So our result is consistent with the widely held belief that recent sharp declines in U.S. equity prices are in response to widely publicized accounting "scandals" in the United States, scandals that have heightened investors' concerns over earnings opacity, prompting investors to demand greater premiums.
Our analyses clearly suggest that further research into the impact of informational risk in general, and earnings opacity specifically, on equity markets is warranted. One avenue of future research could develop techniques to assess earnings opacity at the individual firm level, and then test for links between earnings opacity and equity market variables at the firm level rather than at a country level. A second avenue for research would be to determine which institutional factors impact which dimension of earnings opacity. A third avenue for research, and in our view the most useful avenue, is to develop more refined earnings opacity measures from the distribution of reported earnings.
APPENDIX
Description of Data Used
(5)
(2) (4) Market Capi-
Financial Indices of talization
Statement Stock of Main
Data (3) Markets Exchange
(Annual) Number (Monthly) (Monthly)
(1) (Sample of (Sample (Sample
Country Period) Firm-Years Period) Period)
Australia 86Y-98Y 888 01/86-12/98 01/86-12/98
Austria 87Y-98Y 472 01/86-12/98 01/86-12/98
Belgium 86Y-98Y 567 01/86-12/98 01/86-12/98
Brazil 91Y-98Y 550 01/86-12/98 07/94-12/98
Canada 86Y-98Y 1,997 01/86-12/98 01/86-12/98
Chile 94Y-98Y 147 01/86-12/98 07/89-12/98
Denmark 88Y-98Y 953 01/86-12/98 01/86-12/98
Finland 86Y-98Y 704 12/87-12/98 03/88-12/98
France 86Y-98Y 3,834 01/86-12/98 01/86-12/98
Germany 86Y-98Y 3,847 01/86-12/98 01/86-12/98
Greece 90Y-98Y 491 01/86-12/98 01/88-12/98
Hong Kong 87Y-98Y 925 01/86-12/98 01/86-12/98
India 92Y-98Y 1,342 01/86-12/98 01/90-12/98
Indonesia 92Y-98Y 493 12/89-12/98 04/90-12/98
Ireland 86Y-98Y 445 12/87-12/98 01/86-12/98
Italy 86Y-98Y 1,146 01/86-12/98 01/86-12/98
Japan 86Y-98Y 8,762 01/86-12/98 01/86-12/98
Malaysia 86Y-98Y 1,233 01/86-12/98 01/86-12/98
Mexico 90Y-98Y 361 01/86-12/98 01/88-12/98
The Netherlands 86Y-98Y 1,367 01/86-12/98 01/86-12/98
Norway 88Y-98Y 502 01/86-12/98 01/86-12/98
Pakistan 92Y-98Y 361 01/86-12/98 NA
Portugal 93Y-98Y 165 01/86-12/98 01/90-12/98
Singapore 86Y-98Y 566 01/86-12/98 01/86-12/98
South Africa 86Y-98Y 889 12/92-12/98 01/86-12/98
South Korea 89Y-98Y 867 01/86-12/98 09/87-12/98
Spain 88Y-98Y 483 01/86-12/98 03/87-12/98
Sweden 86Y-98Y 1,004 01/86-12/98 01/86-12/98
Switzerland 86Y-98Y 1,261 01/86-12/98 01/86-12/98
Taiwan 93Y-98Y 577 01/86-12/98 09/87-12/98
Thailand 92Y-98Y 765 01/86-12/98 01/87-12/98
Turkey 93Y-98Y 188 12/86-12/98 01/88-12/98
United Kingdom 86Y-98Y 8,974 01/86-12/98 01/86-12/98
United States 86Y-98Y 11,527 01/86-12/98 01/86-12/98
All Countries 58,653
(6)
Dollar (8)
Volume (7) GDP of
in Main Dividend Country
Exchange Yield (Quarterly
(Monthly) (Monthly) or Annual)
(1) (Sample (Sample (Sample
Country Period) Period) Period)
Australia 01/86-12/98 01/86-12/98 85Q4-98Q4
Austria 08/86-12/98 01/86-12/98 85Q4-98Q4
Belgium 01/86-12/98 01/86-12/98 86Y-98Y
Brazil NA NA 86Y-98Y
Canada 01/86-12/98 01/86-12/98 85Q4-98Q4
Chile 07/89-12/98 01/86-12/98 86Y-98Y
Denmark 04/88-12/98 01/86-12/98 86Y-98Y
Finland NA 03/88-12/98 86Y-98Y
France 06/88-12/98 01/86-12/98 85Q4-98Q4
Germany 06/88-12/98 01/86-12/98 85Q4-98Q4
Greece 01/88-12/98 01/90-12/98 86Y-98Y
Hong Kong 06/88-12/98 01/86-12/98 86Y-98Y
India 01/95-12/98 01/90-12/98 86Y-98Y
Indonesia 04/90-12/97 04/90-12/98 86Y-98Y
Ireland NA 01/86-12/98 86Y-98Y
Italy 07/86-12/98 01/86-12/98 86Y-98Y
Japan 01/90-12/98 01/86-12/98 85Q4-98Q4
Malaysia 01/86-12/98 01/86-12/98 86Y-98Y
Mexico 01/88-12/98 05/89-12/98 86Y-98Y
The Netherlands 02/86-12/98 01/86-12/98 86Y-98Y
Norway 01/86-12/98 01/86-12/98 85Q4-98Q4
Pakistan NA NA 86Y-98Y
Portugal 01/90-12/98 01/90-12/98 86Y-98Y
Singapore 01/86-12/98 01/86-12/98 86Y-98Y
South Africa 01/90-12/98 01/86-12/98 85Q4-98Q4
South Korea 09/87-12/98 09/87-12/98 85Q4-98Q4
Spain 02/90-12/98 03/87-12/98 86Y-98Y
Sweden 01/86-12/98 01/86-12/98 86Y-98Y
Switzerland 01/89-12/98 01/86-12/98 86Y-98Y
Taiwan 04/91-12/98 05/88-12/98 85Q4-98Y
Thailand 01/87-12/98 01/87-12/98 86Y-98Y
Turkey 01/88-12/98 06/89-12/98 87Q1-98Q4
United Kingdom 10/86-12/98 01/86-12/98 85Q4-98Q4
United States 01/86-12/98 01/86-12/98 85Q4-98Q4
All Countries
(9) (10) (11)
Exports of Imports of Exchange
Country Country Rate
(Monthly) (Monthly) (Monthly)
(1) (Sample (Sample (Sample
Country Period) Period) Period)
Australia 01/86-12/98 01/86-12/98 01/86-12/98
Austria 01/86-12/98 01/86-12/98 01/86-12/98
Belgium 01/93-12/98 01/93-12/98 01/86-12/98
Brazil 01/86-12/98 01/86-12/98 01/86-12/98
Canada 01/86-12/98 01/86-12/98 01/86-12/98
Chile 01/86-12/98 01/86-12/98 01/86-12/98
Denmark 01/86-12/98 01/86-12/98 01/86-12/98
Finland 01/86-12/98 01/86-12/98 01/86-12/98
France 01/86-12/98 01/86-12/98 01/86-12/98
Germany 01/86-12/98 01/86-12/98 01/86-12/98
Greece 01/86-12/98 01/86-12/98 01/86-12/98
Hong Kong 01/86-12/98 01/86-12/98 01/86-12/98
India 01/86-12/98 01/86-12/98 01/86-12/98
Indonesia 01/86-12/98 01/86-12/98 01/86-12/98
Ireland 01/86-12/98 01/86-12/98 01/86-12/98
Italy 01/86-12/98 01/86-12/98 01/86-12/98
Japan 01/86-12/98 01/86-12/98 01/86-12/98
Malaysia 01/86-12/98 01/86-12/98 01/86-12/98
Mexico 01/86-12/98 01/86-12/98 01/86-12/98
The Netherlands 01/86-12/98 01/86-12/98 01/86-12/98
Norway 01/86-12/98 01/86-12/98 01/86-12/98
Pakistan 01/86-12/98 01/86-12/98 01/86-12/98
Portugal 01/86-12/98 01/86-12/98 01/86-12/98
Singapore 01/86-12/98 01/86-12/98 01/86-12/98
South Africa 01/86-12/98 01/86-12/98 01/86-12/98
South Korea 01/86-12/98 01/86-12/98 01/86-12/98
Spain 01/86-12/98 01/86-12/98 01/86-12/98
Sweden 01/86-12/98 01/86-12/98 01/86-12/98
Switzerland 01/86-12/98 01/86-12/98 01/86-12/98
Taiwan 01/88-12/98 01/88-12/98 12/93-12/98
Thailand 01/86-12/98 01/86-12/98 01/86-12/98
Turkey 01/86-12/98 01/86-12/98 01/86-12/98
United Kingdom 01/86-12/98 01/86-12/98 01/86-12/98
United States 01/86-12/98 01/86-12/98 01/86-12/98
All Countries
(1) Annual financial statement data for firms in 20 developed markets
and 14 emerging markets were obtained from Worldscope. These countries
are listed in Column (1). The sample period per country is given in
Column (2). The number of firm-years is given in Column (3).
(2) Data on monthly stock market indices for the 20 developed markets
were obtained from Morgan Stanley Capital Market International (MSCI).
Data on monthly stock market indices for the 14 emerging markets were
obtained from the International Financial Corporation (IFC). The
sample periods are given in Column (4).
(3) Data on monthly market capitalization, dollar volume, and monthly
dividend yields were obtained from Datastream. The sample periods are
given in Columns (5), (6), and (7).
(4) Data on quarterly/annual GDP, monthly exports, monthly imports,
and monthly foreign exchange rates were from the International
Financial Statistics of the International Monetary Fund. The
statistics for Taiwan come from Datastream. The sample periods are
given in Columns (8), (9), (10), and (11).
TABLE 1
Summary Statistics
Earnings Opacity
(2) (3) (4)
(1) Earnings Loss Earnings
Countries Aggressiveness Avoidance Smoothing
Australia -0.0213 -0.04615 -0.82374
Austria -0.03727 0.500397 -0.87909
Belgium -0.05467 0.317765 -0.87866
Brazil -0.0068 0.035416 -0.77614
Canada -0.03433 0.450318 -0.81781
Chile -0.01706 0.6 -0.91368
Denmark -0.03937 0.267444 -0.91274
Finland -0.03267 0.621092 -0.88223
France -0.03827 0.376352 -0.86549
Germany -0.04138 0.586525 -0.8978
Greece 0.01344 0.652206 -0.91468
Hong Kong -0.01194 0.17013 -0.85786
India 0.001681 0.735644 -0.86787
Indonesia -0.00098 0.733766 -0.85613
Ireland -0.024 0.153846 -0.86847
Italy -0.02733 0.505334 -0.92531
Japan -0.01247 0.642863 -0.92135
Korea (South) -0.0115 0.595265 -0.93793
Malaysia -0.01226 0.469553 -0.87234
Mexico -0.02058 -0.03333 -0.74486
The Netherlands -0.04506 0.378023 -0.9172
Norway -0.03786 0.178788 -0.72913
Pakistan -0.02584 0.616327 -0.91133
Portugal -0.06614 0.211112 -0.87479
Singapore -0.02534 0.484873 -0.88578
South Africa -0.02021 0.307692 -0.88157
Spain -0.0379 0.514142 -0.85582
Sweden -0.02256 0.340096 -0.84528
Switzerland -0.03963 0.589985 -0.87921
Taiwan -0.02405 0.691198 -0.85276
Thailand -0.03953 0.730403 -0.85693
Turkey 0.127142 0.5 -0.67144
United Kingdom -0.02924 0.372985 -0.8683
United States -0.03833 0.350638 -0.77688
All Countries -0.02141 0.38765 -0.86541
Other Variables
(5)
Auditors per (6) (7)
(1) 100,000 Disclosure IAS
Countries Population Level Use
Australia 539 80 0
Austria NA 62 0
Belgium 38 68 0
Brazil 1 NA 1
Canada 350 75 0
Chile 87 78 NA
Denmark 106 75 0
Finland 10 83 0
France 45 78 1
Germany 26 67 0
Greece 12 61 NA
Hong Kong 110 73 0
India 9 61 1
Indonesia 2 NA NA
Ireland 262 81 1
Italy 110 66 0
Japan 10 71 0
Korea (South) 7 68 0
Malaysia 48 79 2
Mexico 15 71 1
The Netherlands 52 74 1
Norway NA 75 1
Pakistan 2 73 2
Portugal NA NA 1
Singapore 273 79 1
South Africa 35 79 1
Spain 18 72 0
Sweden 41 83 0
Switzerland 53 80 1
Taiwan 17 58 NA
Thailand 5 66 2
Turkey NA 58 1
United Kingdom 352 85 0
United States 168 76 0
All Countries 93 73
Other Variables
(9) Mean
Monthly
Return/
Standard
(8) Deviation
(1) Common of Monthly (10)
Countries Law Return Trade
Australia Yes 0.1292 -3.3862
Austria No 0.0980 -3.5173
Belgium No 0.3282 -4.6080
Brazil No 0.0285 NA
Canada Yes 0.1446 -3.6756
Chile No 0.2628 -4.9177
Denmark No 0.2198 -5.0497
Finland No 0.1572 NA
France No 0.2164 -4.0970
Germany No 0.1671 -1.9795
Greece No 0.1708 -4.0892
Hong Kong Yes 0.1455 -3.401
India Yes 0.0301 -3.7879
Indonesia No -0.0876 -4.5513
Ireland Yes 0.2213 NA
Italy No 0.1339 -4.2832
Japan No 0.0698 -4.2534
Korea (South) No 0.0433 -2.9382
Malaysia Yes 0.0271 -4.3673
Mexico No 0.1243 -3.3118
The Netherlands No 0.3426 -2.8995
Norway No 0.0839 -3.4892
Pakistan Yes 0.0369 NA
Portugal No 0.0822 -5.1253
Singapore Yes 0.0958 -4.0609
South Africa No 0.0609 -4.2894
Spain No 0.2321 -3.3038
Sweden No 0.2242 -3.6422
Switzerland No 0.2602 -3.3903
Taiwan No 0.0950 -2.3109
Thailand Yes 0.0542 -3.3376
Turkey No 0.0698 -3.6251
United Kingdom Yes 0.2369 -3.1222
United States Yes 0.3097 -2.7766
All Countries 0.1416 -3.7196
Other Variables
(12)
(11) Insider
Real % GDP Trading (13)
(1) Growth Enforcement Liberalization
Countries (1985-1998) Date Date
Australia 3.14 1996 Before 01/86
Austria 2.49 No Before 01/86
Belgium 2.21 1994 Before 01/86
Brazil 2.70 1978 05/91
Canada 2.54 1976 Before 01/86
Chile 6.75 1996 01/92
Denmark 2.37 1996 Before 01/86
Finland 2.12 1993 Before 01/86
France 2.13 1975 Before 01/86
Germany 2.96 1995 Before 01/86
Greece 2.15 1996 12/87
Hong Kong 5.30 1994 Before 01/86
India 5.73 1998 11/92
Indonesia 4.55 1996 09/89
Ireland 5.95 No Before 01/86
Italy 1.93 1996 Before 01/86
Japan 2.74 1990 Before 01/86
Korea (South) 7.48 1988 12/88
Malaysia 6.50 1996 05/89
Mexico 3.27 No Before 01/86
The Netherlands 2.74 1994 Before 01/86
Norway 2.94 1990 02/91
Pakistan 4.26 No 07/86
Portugal 4.38 No Before 01/86
Singapore 6.97 1978 01/92
South Africa 1.30 No Before 01/86
Spain 3.18 1998 Before 01/86
Sweden 1.49 1990 Before 01/86
Switzerland 1.51 1995 Before 01/86
Taiwan 7.26 1989 01/91
Thailand 5.97 1993 09/87
Turkey 4.82 1996 08/89
United Kingdom 2.57 1981 Before 01/86
United States 2.73 1961 Before 01/86
All Countries 3.74
(1) Annual financial statement data for firms in 20 developed
markets and 14 emerging markets were obtained from Worldscope.
These countries are listed in Column (1).
(2) We scale accruals by lagged total assets for each firm,
determine its median in the cross-section of firms per country
per year, and then average across time to obtain the "earnings
aggressiveness" variable per country. This is listed in Column (2).
(3) We define firms with small positive earnings (small negative
earnings) as firms with net income scaled by lagged total assets
between 0 and 1% (between 0 and -1%). We subtract the number of firms
with small negative earnings from the number of firms with small
positive earnings per country per year, divide this difference by the
sum of the two, and then average this ratio across time to obtain the
"loss avoidance" variable per country. This is listed in Column (3).
(4) We find the correlation between the change in accruals and the
change in operating cash flows (both scaled by lagged total assets)
in the cross-section of firms per country per year, and then average
across time to obtain the "earnings smoothing" variable per country.
This is listed in Column (4).
(5) The number of auditors per 100,000 population in Column (5)
comes from Saudagaran and Diga (1997, Table 6). The original source
is the International Federation of Accountants (IFAC) Secretariat,
8/13/1996.
(6) Disclosure level data in Column (6) come from Saudagaran and
Diga (1997, Table 2). The original source is the Center for
International Financial Analysis and Research (CIFAR 1995). The
higher the number, more is the disclosure.
(7) International Accounting Standards (IAS) use data in Column (7)
come from Choi et al. (1999, exhibit 8.6). They took it from
International Accounting Standards Committee (IASC Insight dated
October, 1997.) 0 = completely independent standard setting, no
use of IAS except possibly a comparison with IAS; 1 = separate
accounting standards that are based on and similar to IAS in most
cases, however, some standards provide more or less choice; 2 = IAS
are used as national standards with some modification for local
conditions, standards not covered by IAS added.
(8) The common law data in Column (8) come from the CIA World
Factbook (2001), http://www.odci.gov/cia/publications/factbook.
(9) Data on monthly stock market indices for the 20 developed markets
were obtained from Morgan Stanley Capital Market International
(MSCI). Data on monthly stock market indices for the 14 emerging
markets were obtained from the International Financial Corporation
(IFC). The mean return scaled by the standard deviation of returns
is given in Column (9). The sample periods used to calculate these
statistics are given in Column (4) of the Appendix.
(10) Trade is defined as the natural logarithm of the ratio of volume
of dollar trade per month to dollar market capitalization at the end
of the month. Monthly data on these two variables were obtained from
Datastream. Their sample periods are given in Columns (5) and (6) in
the Appendix. The mean trade per country is given in Column (10).
(11) The real growth in Gross Domestic Product per country per year
is obtained from the World Bank, http://www.worldbank.org/research/
growth/GDNdata.htm. The average per country is given in Column (11).
(12) The insider trading enforcement date in Column (12) comes from
Bhattacharya and Daouk (2002, Table 1).
(13) The official liberalization date, which was obtained from
Bekaert and Harvey (2000), is given in Column (13).
TABLE 2
Relation between Earnings Opacity Measures and Other Financial
Reporting Measures
Correlation Matrix
Auditors per
Loss Earnings 100,000
Avoidance (b) Smoothing (c) Population (d)
Earnings
Aggressiveness (a) .152 .453 * -.135
Loss Avoidance (b) 1 -.431 * -.448 *
Earnings
Smoothing (c) 1 .210
Auditors per
100000
Population (d) 1
Disclosure Level (e)
IAS Use (f)
Common Law (g)
Correlation Matrix
Disclosure IAS Common
Level (e) Use (f) Law (g)
Earnings
Aggressiveness (a) -.404 * .135 -.034
Loss Avoidance (b) .429 * .096 -.069
Earnings
Smoothing (c) -.094 .138 .050
Auditors per
100000
Population (d) .458 * -.284 .581 *
Disclosure Level (e) 1 -.031 .251
IAS Use (f) 1 .250
Common Law (g) 1
* Significant at p < .05, two-tailed test.
(a) We scale accruals by lagged total assets for each firm,
determine its median in the cross-section of firms per
country per year, and then average across time to obtain
the "earnings aggressiveness" variable per country.
(b) We define firms with small positive earnings (small
negative earnings) as firms with net income scaled by lagged
total assets between 0 and 1 percent (between 0 and -1
percent). We subtract the number of firms with small
negative earnings from the number of firms with small
positive earnings per country per year, divide this
difference by the sum of the two, and then average
this ratio across time to obtain the "loss avoidance"
variable per country.
(c) We find the correlation between the change in accruals
and the change in operating cash flows (both scaled by
lagged total assets) in the cross-section of firms per
country per year, and then average across time to obtain
the "earnings smoothing" variable per country.
(d) The number of auditors per 100,000 population comes
from Saudagaran and Diga (1997, Table 6). The original
source is communication with the International Federation
of Accountants (IFAC) Secretariat, August 13, 1996.
(e) Disclosure level data come from Saudagaran and Diga
(1997, Table 2). The original source is the Center for
International Financial Analysis and Research (CIFAR
1995). The higher the number, more is the disclosure.
(f) International Accounting Standards (IAS) use data
come from Choi et al. (1999, exhibit 8.6). They took it
from International Accounting Standards Committee (IASC
Insight dated October, 1997.) 0 = completely independent
standard setting, no use of IAS except possibly a
comparison with IAS; 1 = separate accounting standards
that are based on and similar to IAS in most cases,
however, some standards provide more or less choice;
2 = IAS are used as national standards with some
modification for local conditions, standards not covered
by IAS added.
(g) The common law data come from the CIA World Factbook
(2001), http://www.odci.gov/cia/publications/
factbook.
TABLE 3
Earnings Opacity Ranking of Countries (a)
Earnings
Aggressiveness (b) Loss Avoidance (c)
Least, 1 Portugal Brazil
Belgium Mexico
The Netherlands Australia
Germany United States
Switzerland Norway
United States Ireland
Denmark Denmark
2 France France
Spain United Kingdom
Finland Belgium
Austria Sweden
Canada Portugal
Thailand Canada
Norway Hong Kong
3 Italy The Netherlands
United Kingdom South Africa
Pakistan Austria
Ireland Singapore
Australia South Korea
Sweden Malaysia
Singapore Germany
4 Taiwan Italy
Chile Spain
Japan Switzerland
South Africa Japan
Brazil Finland
Mexico Pakistan
Hong Kong Chile
Most, 5 Malaysia Greece
South Korea Turkey
Indonesia Taiwan
India Thailand
Greece India
Turkey Indonesia
Overall Earnings
Earnings Smoothing (d) Opacity (e)
Least, 1 Turkey United States
United States Norway
Brazil Portugal
Norway Brazil
Mexico Belgium
Canada Mexico
Australia Canada
2 Taiwan France
Spain Australia
France Spain
Thailand United Kingdom
Sweden Denmark
United Kingdom Switzerland
India Sweden
3 Hong Kong Germany
Portugal The Netherlands
Indonesia Finland
Malaysia Austria
Switzerland Thailand
Finland Ireland
Singapore Hong Kong
4 Belgium Singapore
South Africa Taiwan
Austria Turkey
Germany South Africa
Ireland Malaysia
Pakistan Italy
Denmark Pakistan
Most, 5 Chile Japan
Greece Chile
Japan India
The Netherlands Indonesia
Italy South Korea
South Korea Greece
(a) The data used to construct the earnings opacity variables
come from Worldscope.
(b) We scale accruals by lagged total assets for each firm,
determine its median in the cross-section of firms per
country per year, and then rank these medians across years
and across countries. This rank is the "earnings
aggressiveness" time-series variable per country.
(c) We define firms with small positive earnings (small
negative earnings) as firms with net income scaled by lagged
total assets between 0 and 1 percent (between 0 and -1
percent). We subtract the number of firms with small
negative earnings from the number of firms with small
positive earnings per country per year, divide this
difference by the sum of the two, and then rank this
ratio across years and across countries. This rank is the
"loss avoidance" time-series variable per country.
(d) We find the correlation between the change in accruals
and the change in operating cash flows (both scaled by
lagged total assets) in the cross-section of firms per
country per year, and then rank these correlations across
years and across countries. This rank is the "earnings
smoothing" time-series variable per country.
(e) The "overall earnings opacity" time-series variable
per country is the average of the "earnings aggressiveness"
time-series variable per country, the "loss avoidance"
time-series variable per country and the "earnings
smoothing" time-series variable per country.
TABLE 4
Effect of Earnings Opacity on the Cost of Equity
(Using Dividend Yields) (a)
MODEL: Cost of [Equity.sub.i,t] = [[beta].sub.0] + [[beta].sub.1]
Dimension of Earnings [Opacity.sub.i,t-1] + [[beta].sub.2]
[Liberalization.sub.i,t] + [[beta].sub.3] Insider Trading
[Enforcement.sub.i,t] + [[beta].sub.4] GDP [Growth.sub.i,t]
+ [u.sub.i,t]
Dependent Variable (b) Cost of Equity
Independent Variables (c) (1) (2)
Earnings aggressiveness (d) 0.0012
(0.0000)
Loss avoidance (c) 0.0000
(0.9479)
Earnings smoothing (f)
Overall earnings opacity (g)
Liberalization (h) -0.0045 -0.0040
(0.0862) (0.1301)
Insider trading enforcement (i) -0.0038 -0.0039
(0.0345) (0.0300)
GDP growth (j) 0.0011 0.0013
(0.0000) (0.0000)
Dependent Variable (b) Cost of Equity
Independent Variables (c) (3) (4)
Earnings aggressiveness (d)
Loss avoidance (c)
Earnings smoothing (f) 0.0001
(0.7443)
Overall earnings opacity (g) 0.0012
(0.0011)
Liberalization (h) -0.0040 -0.0043
(0.1293) (0.1033)
Insider trading enforcement (i) -0.0039 -0.0040
(0.0320) (0.0260)
GDP growth (j) 0.0013 0.0012
(0.0000) (0.0000)
(a) The numbers below are coefficient estimates from panel
regressions, and are corrected for country-fixed effects,
country-specific heteroscedasticity and country-specific
autocorrelation, p-values are in parentheses.
(b) The dependent variable "Cost of equity" is co