Capital Allocation and Timely Accounting Recognition of Economic Losses: International Evidence

Transcription

1 Capital Allocation and Timely Accounting Recognition of Economic Losses: International Evidence Robert M. Bushman The University of North Carolina at Chapel Hill Kenan-Flagler Business School Chapel Hill, North Carolina Joseph D. Piotroski The University of Chicago Graduate School of Business Chicago, Illinois Abbie J. Smith The University of Chicago Graduate School of Business Chicago, Illinois Current draft: March 2005 Corresponding author. We appreciate the financial support of the Kenan-Flagler Business School, University of North Carolina at Chapel Hill, and the Graduate School of Business at the University of Chicago 1

2 Capital Allocation and Timely Accounting Recognition of Economic Losses: International Evidence Abstract We examine the role of financial accounting information in facilitating the avoidance of bad projects and promoting quick withdrawal of capital from failing projects. We investigate whether firms in countries whose accounting regimes are characterized by more timely recognition of economic losses tend to withdraw capital from losing projects relatively faster than firms in countries with less timely loss recognition. Our evidence suggests a relation between timely accounting recognition of economic losses (TLR) (or the underlying institutions that create a demand for or facilitate the supply of TLR accounting practices) and observed economic outcomes as reflected by the properties of returns distributions and investment elasticities. First, we find that the proportion of firms in an economy exhibiting economic losses in a given year (as measured by annual stock returns) is higher in economies without TLR accounting practices and lower in economies with TLR accounting practices, consistent with TLR (or the pressures that lead to TLR) leading to early intervention of bad projects. Further evidence suggests that the TLR-loss incidence relation is stronger in those regimes with diffuse ownership and monitoring structures, precisely where accounting numbers and the covenants they support are likely to be a primary means of monitoring the firm and its investment decisions. We also investigate how TLR practices could influence the economic income patterns that we document by examining the association between TLR and proxies for the efficient allocation of capital. We document that TLR measures are positively associated with a proxy for the speed with which firms withdraw capital from shrinking sectors and with a proxy for the severity of a country s control problems, but not with proxies for the speed with which capital flows to growing sectors. That is, firms in countries characterized by higher TLR tend to withdraw capital from losing projects relatively faster than firms in countries with less timely loss recognition. 2

3 1. Introduction Fundamental to wealth creation in an economy is the efficiency with which capital is allocated to investment opportunities. Efficiency is a function of the extent to which capital is invested in projects expected to have high returns, and the extent to which negative NPV projects are avoided or capital promptly withdrawn from projects discovered to be losers at some point after project initiation. Given information asymmetry and self-interested behavior by managers, agency theories argue that pressures from external investors, as well as formal contracting arrangements, are needed to encourage managers to pursue value-maximizing investment policies. While a wide range of economy-wide and firm-specific mechanisms arise to facilitate efficient capital allocation, and causal relations are difficult to isolate, credible, public information about firms lies at the heart of many economic theories of firm investment behavior. 1 Objective, verifiable accounting information facilitates monitoring and supplies a rich array of contractible variables useful for aligning executives and investors financial interests. In this paper, we examine the role of financial accounting information in facilitating the avoidance of bad projects and promoting timely withdrawal of capital from failing projects. We investigate whether firms in countries whose accounting regimes are characterized by more timely accounting recognition of economic losses tend to withdraw capital from losing projects relatively faster than firms in countries with less timely loss recognition. We also examine the relation between more timely recognition of economic losses and the distribution of countries stock returns. To the extent that the probability and/or severity of an economic loss within a firm is commensurate with the timeliness of project abandonment, firms that abandon poor projects early (and/or are denied capital early) should generate fewer and smaller losses than firms which delay abandoning poor projects and continue to sink capital into 1 Classic examples include Jensen and Meckling (1976), Holmstrom (1979) and Myers and Majluf (1984). 3

4 losers. Timely accounting recognition of losses is hypothesized to facilitate early termination of poor projects. We predict that firms in countries with more timely loss recognition should generate fewer and smaller economic losses (proxied by stock returns) than firms which delay abandoning poor projects and continue to sink capital into these projects. That managers may pursue losing projects has long been recognized in the economics, psychology and organizational behavior literatures. This phenomenon has been approached using a wide range of perspectives. These include, among others, perquisite consumption (Jensen and Meckling (1976)), free cash flow problems (Jensen (1986)), the Pain Avoidance Model (Jensen (1994)), principal-agent problems (Holmstrom (1979)), the sunk cost phenomenon and escalation of commitment (e.g., Staw (1981), Kanodia, Bushman and Dickhaut (1989), Heath (1995), Prendergast and Stole (1996), Camerer and Weber (1999)). For example, escalation of commitment" is said to occur when managers who have committed resources to a project are inclined to "throw good money after bad" and maintain or increase their commitment to the project, even when its marginal costs exceed marginal benefits (Camerer and Weber (1999)). Jensen (2000) discusses at length significant problems associated with designing control systems that lead managers to exit from activities when exiting is warranted by economic realities. We do not attempt to distinguish between extant theories of why managers pursue losing projects. Instead we take the phenomenon as an inherent problem faced by all economies, and explore the extent to which more timely recognition of economic losses in accounting reports is associated with more timely withdrawal of capital from losing projects. Our hypothesis that such a relation exists is based on the arguments articulated in Ball (2001). Ball (2001) develops a set of infrastructure requirements necessary to support an economically efficient system of public financial reporting and disclosure. His second criterion for economically efficient public 4

5 financial reporting is the timely incorporation of adverse information about future cash flows into published financial statements. Ball argues that such timely incorporation of economic losses increases the effectiveness of corporate governance, compensation systems and debt agreements to motivate and monitor managers. For example, since lenders and borrowers contract on financial reports through accounting-based covenants, timely accounting loss recognition provides lenders with more timely signals of deteriorating performance through a tightening of covenants or triggering of covenant violations. 2 Ball argues that the timely incorporation of economic losses (1) decreases the ex ante likelihood that managers undertake negative-npv projects whose earnings consequences are passed on to a subsequent generation and (2) increases the incentive of the current managers to incur the personal cost of abandoning investments and strategies that have ex post negative NPVs. We estimate timely accounting recognition of economic losses (measures of TLR) at the country level using two techniques. The first technique follows Basu (1997), Ball, Kothari and Robin (2000) and Ball, Robin and Wu (2003), and regresses net income on stock returns and stock returns interacted with an indicator variable that is set to one for negative returns and zero otherwise. This technique in essence proxies for economic income with stock returns, where the coefficient on stock returns interacted with the indicator variable captures the incremental speed of bad news recognition in earnings relative to the speed of good news recognition. The second technique follows Ball and Shivakumar (2005) and examines the asymmetric timeliness of earnings without reference to security prices. This technique regresses current period accruals on current period operating cash flows and current period operating cash flows interacted with an 2 Several papers empirically examine the efficiency gains from accounting conservatism in the debt contracting process. See for example, Ahmed et al. (2002) and Zhang (2004). 5

6 indicator variable that is set to one for negative cash flows and zero otherwise. The coefficient on the interaction term (i.e. negative cash flows) is interpreted as the incremental speed of bad news recognition in earnings relative to the speed of good news recognition. Our first set of analyses compares the distribution of stock returns in countries with faster speed of bad news accounting recognition relative to countries with lower speed of bad news accounting recognition. We classify countries as being either in a timely loss recognition (high TLR) or an untimely loss recognition (low TLR) accounting reporting regime. We hypothesize that the probability of a firm in an economy experiencing an economic loss is a function of whether has high or low TLR. We find that that the probability of a firm experiencing a negative return is significantly lower in countries characterized as having high TLR, both before and after controlling for firm-specific and country-level differences in risk, differences in sample composition and differences in underlying economic attributes. We also find that the proportion of firms in a country experiencing economic losses (i.e., negative returns) in a given year is significantly lower in countries with high TLR. Finally, we document that the relation between TLR measures and negative returns is stronger in countries with low concentration of ownership relative to countries with highly concentrated ownership, suggesting that timely loss recognition, as a governance mechanism, offers its greatest benefits where diffuse ownership structure allow for classical agency conflicts. In our second set of analyses, we investigate whether countries characterized by more timely recognition of economic losses tend to withdraw capital from losing projects relatively faster than firms in countries with less timely loss recognition practices. Wurgler (2000) computes an investment elasticity measure that gauges the extent to which a country increases investment in growing industries and decreases investment in declining ones. He shows that 6

7 countries with a higher level of financial development increase investment more in growing industries and decrease investment more in declining industries than financially underdeveloped countries. Beck and Levine (2002) use Wurgler s elasticity measures to examine the relative importance of banks versus markets in promoting efficient capital allocation. 3 Building from these papers, we measure efficiency of capital allocation decisions using Wurgler s estimates of the elasticity of gross capital investment to value added. This elasticity measure, η k, is interpreted as the extent to which firms in country k reduce investment in declining industries and increase investment in growing industries. Further, Wurgler disaggregates η k by separately estimating the elasticity in country k for industry-year observations reflecting increasing value added (η + k ) and those reflecting shrinking value added (η - k ). The difference (η - k - η + k ) proxies for the severity of the control problem. We examine relations between our TLR proxies and η k, η + k, η - k, and (η - k - η + k ). We find evidence that TLR is unrelated to the elasticity of investing in growing sectors (η + k ), but positively related to the elasticity of withdrawing capital from shrinking sectors (η - k ) and to the severity of the control problem (η - k - η + k ). While the associations documented in our analyses are suggestive of TLR playing a role in disciplining managers exit decisions, we do not presume to have established casual relations. As shown in Bushman and Piotroski (2005), Ball, Kothari and Robin (2000), and others, TLR is influenced by key aspects of a country s institutional structure. However, whether our results are caused by TLR or the underlying institutions that create a demand for or facilitate the supply of TLR reporting behavior, it is indeed interesting that TLR appears as part of equilibrium outcomes associated with timely exit from losing projects and lower severity of control 3 They find that while financial development in general appears to facilitate efficient capital allocation, having a bank-based or market-based system per se does not seem to matter. 7

8 problems. Moreover, ex ante our empirical design has the potential to cast meaningful doubt on the empirical validity of the theory that timely recognition of economic losses decreases the ex ante likelihood that managers undertake negative-npv projects and increases the incentive of current managers to abandon investments and strategies that have ex post negative NPVs. A lack of association between TLR and timely withdrawal of capital from losers would cast doubt on this theory. The rest of the paper is organized as follows. Section 2 discusses our conceptual framework and the relation of our study to the prior literature. Section 3 describes the data, our sample and the research design. Section 4 presents our empirical evidence on associations between timely loss recognition practices and the prevalence of economic losses, while section 5 presents evidence on relations between investment flow efficiency and timely loss recognition. Section 6 summarizes the paper and concludes. 2. Conceptual framework and prior literature In this section, we describe the conceptual framework underlying our hypotheses and relate our study to the extant literature. We empirically investigate two hypotheses: H1: The probability of a firm in an economy experiencing an economic loss is a decreasing function of timely loss recognition; and H2: Countries characterized by more timely recognition of economic losses tend to withdraw capital from losing projects relatively faster than firms in countries with less timely loss recognition. The importance of efficient exit for the productivity and economic vitality of an economy has long been recognized, dating back to at least Schumpeter s (1942) ideas about creative 8

9 destruction. Jensen (2000) emphasizes the importance of devising control systems to effectively deal with the requirement for exit and downsizing in the face of declining costs, increasing average (but decreasing marginal) productivity of labor, reduced growth rates of labor income, and excess capacity. But why are managers reluctant to exit losing projects and why at times do they pursue negative NPV projects? While no definitive answer exists, a number of potential explanations have been offered in the economics, psychology and organizational behavior literatures. Explanations include: 1) Agency cost explanations where expected marginal benefits of a project are less than expected marginal costs, but private benefits accruing to the manager favor escalation (see e.g., Kanodia, Bushman, and Dickhaut (1989) and Prendergast and Stole (1996)). This explanation also includes notions of private benefits of control (Jensen and Meckling (1976) and empire building (Jensen (1986)), among others ; 2) Gambling in the domain of losses, where it is argued (e.g., Bazerman (1984)) that escalating commitment is utility-maximizing if people have a convex disutility function for losses, so that they prefer to gamble in the domain of losses to break even; and 3) Overconfidence, where agents simply overestimate marginal benefits and underestimate marginal costs. 4 We do not attempt to distinguish between these theories and simply take the phenomenon as an important problem faced by all economies. We focus on the role of the financial accounting system in alleviating managerial reluctance to exit losing projects and tendencies to pursue negative NPV projects, recognizing that accounting is only one of many mechanisms used to align the interests of managers with investors and other key stakeholders (see Shleifer and Vishny (1997) and Bushman and Smith (2001) for extensive reviews of the governance 4 See Camerer and Weber (1999) for an interesting discussion of this literature and difficulties in empirically isolating escalation of commitment. 9

10 literature). Thus, while there are compelling reasons to believe that timely accounting recognition of economic losses disciplines managers capital allocation decisions (discussed in detail next), we cannot claim to empirically establish such a causal relation. However, ex-ante, our empirical design has the potential to cast meaningful doubt on the empirical validity of the theory that timely recognition of economic losses decreases the ex ante likelihood that managers undertake negative-npv projects and increases the incentive of current managers to incur the personal cost of abandoning investments and strategies that have ex post negative NPVs. The persistent influence of conservatism on accounting practice suggests that it confers benefits to economic agents who use, prepare or regulate financial reports. 5 Watts (2003a) argues that conservative accounting is part of efficient contracting design useful for constraining opportunistic behavior by managers, and that timely incorporation of losses in accounting income leads managers to address losses more quickly, and allows debt covenants and dividend restrictions bind more quickly. Ball (2001) argues that the use of financial statement information in debt agreements and in corporate governance creates a demand for recognizing and reporting economic losses in a timely fashion, even though this recognition is based on expectational information. Relative to debt, because lenders and borrowers contract on the financial reports through accounting-based covenants, timely recognition of losses in accounting reports enable lenders to receive more timely signals of deteriorating performance through the tightening of covenants or triggering of covenant violations. As such, timely loss recognition accelerates the transfer of decision rights to debt holders, and thus promotes early intervention behavior. Ball (2001) argues that corporate governance gives rise to a similar demand for accounting asymmetry. As discussed above, take as given that managers are reluctant to exit 5 For example, Basu (1997) notes that conservatism has influenced accounting practice for at least five hundred years, and Sterling (1970, p. 256) stresses the highly influential impact of conservatism on the principles of valuation in accounting. 10

11 losing projects and at times pursue negative NPV projects. This behavior can be mitigated by accounting rules that quickly identify economic losses and charge them against reported income. If economic losses are charged against income, there is no incremental income penalty to actual abandonment, reducing the incentives of managers to prolong losing investments and strategies. Managers agree to an accounting system that incorporates economic losses in a timely fashion as it allows them to bond themselves ex ante to act more in the interests of the owners of the firm, making their employment contract more valuable. Ball (2004) illustrates this phenomenon in a case study of Daimler-Benz, arguing that by listing its stock on the New York Stock Exchange, Daimler bonded itself to publicly report its economic losses in a timely fashion. Following this listing, Ball notes that Daimler drastically reduced its workforce, closed several plants, and ruthlessly discarded several loss-making businesses. Finally, Ball (2001) notes that early reporting of economic losses also brings quicker pressure on managers from directors, large shareholders, security analysts and the press. Furthermore, ex ante knowledge that unrealized losses will be charged against reported income reduces the incentive of managers to undertake negative-npv projects in the first place. We turn now to our empirical analysis. 3. Data and research design The primary objective of this paper is to identify whether or not a relation exists between timely loss recognition accounting practices and early intervention behavior across economies. To address this question, we measure four proxies for timely loss recognition at the country level, and focus on two economic outcomes likely to be correlated with early intervention 11

12 behavior: the likelihood of negative economic income and the elasticity of investment to value added in that country. 3.1 Data and variable definitions Our four measures of timely loss recognition, as well as our two measures of the incidences of economic losses, are estimated using observable returns and accounting realizations. The accounting and return data employed in this study are drawn from Global Vantage. Accounting income and dividends data are gathered from the Global Vantage Industrial / Commercial (IC) files. Stock price data are gathered from the Global Vantage Issues file. Consistent with prior international research (e.g., Ball, Robin and Kothari (2000); Ball Robin and Wu (2003); Bushman and Piotroski (2005)), accounting earnings (NI i,t ) is defined as net income before extraordinary items (IC data 32), and dividends (DIV i,t ) is defined as dividends paid (IC data 36). Operating cash flow (CFO i,t ) is measured as net income plus depreciation (IC data 11), increases in current operating liabilities (IC data 104 minus IC data 94) and decreases in current operating assets (IC data 75 minus IC data 60). Operating accruals (ACCRUALS i,t ) are measured as the difference between NI i,t and CFO i,t. All accounting variables (in local currency) are scaled by the beginning market value of equity (in local currency), defined as price times number of shares outstanding, adjusted for stock splits and dividends using the Global Advantage adjustment factor. Stock return (RET i,t ) is the holding period return, including dividends, over the firm s fiscal year. We assign firm-year observations to countries based on the ISO country-code of incorporation. Several additional control variables are estimated using Global Vantage data. First, each firm s book-to-market ratio (BM i,t ) is measured as the firm s shareholder s equity at the 12

13 beginning of the fiscal year (IC data 135) scaled by beginning market value of equity. Second, firm size (MVE) is measured as the firm s market value of equity at the beginning of the fiscal year, translated into U.S. dollars for all firms using average foreign currency exchange rates for the calendar year ending closest in time to the measurement of the market value of equity. 6 Third, at the country level, we annually estimate average firm size in U.S. dollars (AVEMVE), average market-to-book ratios (AVEBM) and average stock returns (AVERET) for use in our country-level estimations of the likelihood of loss incidences. Fourth, we measure the percentage of each country s domestic listed firms covered by Global Vantage (MKTPCT) in a given year to control for any coverage / selection biases in our data. The number of domestic listed firms in a country at the end of each calendar year is gathered through the World Development Indicators database. Finally, we measure various country-level attributes to proxy for country s general level of economic development (EMERGING), institutional structure of the economy (CIVILLAW), financial development (FD), fraction of the economy s output attributable to state owned enterprise (SOE), protection of investor rights (RIGHTS) and information flow proxied by stock return synchronicity (FRACTION). See appendix 1 for a complete definition of each variable and its measurement. 3.2 Measurement of timely loss recognition in accounting earnings To measure and classify each country s accounting regime along the dimension of timely loss recognition, notated generically as TLR k, we construct variables derived from two separate methodologies: the non-linear earnings-return model of Ball, Kothari and Robin (2000) and Basu (1997), and the non-linear accruals-cash flow model of Ball and Shivakumar (2005). Our four 6 Information on foreign currency exchange rates are gathered through the World Development Indicators database. 13

14 country-level timely loss recognition attributes are estimated for 37 countries with sufficient firm-level accounting and returns data on Global Vantage over the period 1992 to 2001 to estimate the respective model. 7 The methodologies for estimating each country s accounting properties are outlined below; the resultant country-level estimates of timely loss recognition (TLR k ) are presented in Appendix Country-level measures of timely loss recognition: Ball, Kothari and Robin (2000) earnings-return estimations Prior research (e.g., Ball, Kothari and Robin (2000); Pope and Walker (1999)) suggests the timeliness of loss recognition in a country (i.e., bad news sensitivity of earnings), as well as the incremental timeliness of loss recognition relative to timeliness of gain recognition in a country, can be estimated using coefficients from the following cross-sectional Basu (1997) model: NI i,t = α+ β 1 NEG i,t + β 2 RET i,t + β 3 NEG i,t *RET i,t + ε i,t (1) where NI i,t is annual earnings, RET i,t is the annual holding period stock return over the firm s fiscal year, and NEG i,t is an indicator variable equal to one if RET i,t is less than zero, zero otherwise. In this non-linear earnings-return framework, the timeliness of loss recognition in earnings is equal to sum of estimated coefficients β 2 + β 3, while the incremental timeliness of loss recognition (i.e., frequently conceptualized as one dimension of conservative accounting practices) is measured by the estimated coefficient β 3. We estimate equation (1), by country, using pooled, cross-sectional data over the time period 1992 to 2001 for all countries with at least 100 observations over this time period. Our 7 Specifically, our timely-loss recognition variables directly correspond to the country-level estimates of bad news sensitivity and incremental bad news sensitivity of earnings and accruals used in Bushman and Piotroski (2005). 14

15 first measure of timely loss recognition, BKR_TIME k, is defined as the sum of estimated coefficients β 2 + β 3 from a country-level estimation of equation (1). Our second measure of timely loss recognition is the incremental timeliness of loss recognition, BKR_INCR k, defined as the estimated coefficient β 3 from a respective country-level estimation of equation (1). To implement the non-linear earnings-return model, we followed the sample selection guidelines established in Ball, Robin and Wu (2003), as applied in Bushman and Piotroski (2005). First, we exclude the extreme percentiles of each model variable (RET i,t, NI i,t ) when estimating our timely-loss recognition measures. Second, we exclude all countries with less than 100 firm-year observations over the ten years. Third, we exclude China and Poland because these two countries have a socialist legal origin, and we exclude Bermuda, Cayman Islands and Turkey because we lack various institutional or accounting variables for these three countries. This results in estimates of BSK_TIME k and BKR_INCR k for 37 countries Country-level measures of timely loss recognition: Ball and Shivakumar (2005) accruals-cash flow estimations The measures of timely loss recognition derived from equation (1) rely on the implicit assumption that price changes reflect economic gains and losses, and that the price formation process is equally efficient across all markets. Recent evidence in Morck, Yeung and Yu (2000) suggests that securities in different economies reflect different levels of firm-specific information. Moreover, some countries place asymmetric restrictions on the price formation process, such as short-selling constraints, that impede the impounding of bad news into prices in a timely manner. To the extent that the information content of prices varies across countries, our estimates of timely loss recognition could be biased. Finally, Dietrich, Muller and Riedl (2004) 15

16 argue that results from the traditional earnings-returns asymmetric timeless estimations are induced by the research design itself. To mitigate these concerns, we implement an alternative method outlined in Ball and Shivakumar (2005) to measure the timeliness and incremental timeliness of loss recognition properties in earnings without referencing security prices. Specifically, Ball and Shivakumar estimate the following model: ACCRUALS i,t = α+ β 1 NEGCFO i,t + β 2 CFO i,t + β 3 NEGCFO i,t *CFO i,t + ε i,t (2) where ACCRUALS i,t is current period accruals, CFO i,t is current period operating cash flows, and NEGCFO i,t is an indicator variable equal to one if CFO i,t is less than zero. Dechow, Kothari and Watts (1998) document a negative correlation between contemporaneous accruals and cash flows. Ball and Shivakumar argue that if the cash flow effects of current news are persistent, then timely accounting recognition will be a source of positive correlation between accruals and current-period cash flows. Given the asymmetric treatment of gains and losses due to accounting conservatism, this asymmetric timeliness would suggest that a positive correlation effect would be stronger for losses than gains, or in other words, the traditional negative association between accruals and cash flows will be weaker in the presence of losses than for gains. By conditioning the relation between accruals and operating cash flows on the sign of current cash flows, Ball and Shivakumar find that the negative relation between accruals and operating cash flows is attenuated when cash flows are negative (i.e., β 3 > 0). 8 We estimate equation (2) by country using pooled, cross-sectional data over the time period 1992 to 2001 using all firm-years with sufficient cash flow and accrual information, and after eliminating the top and bottom percentile of CFO i,t and ACCRUAL i,t realizations annually. 8 Bushman and Piotroski (2005), using cross-country data, find similar relations. 16

17 To measure the timely incorporation of bad news property for a given country, we define BS_TIME k as the sum of estimated coefficients β 2 + β 3. Similarly, to measure the incremental timeliness of bad news recognition for given country, we define BS_INCR k as the estimated coefficient on NEGCFO i,t *CFO i,t (i.e., β 3 ). As noted earlier, our estimates of BS_TIME k and BS_INCR k do not depend upon the assumption of equally informed stock prices; however, this non-price-based technique of measuring timely loss recognition also has limitations. First, this methodology relies on an assumption that the cash flow implications from a current news event are present in the current year and are persistent (i.e., not a transitory, current period expenditure). Second, these innovations are assumed to be equally persistent across countries and regimes, which may not hold because different economies face different risks and different investment opportunities. Third, the methodology assumes that operating cash flows are unbiased. Existing research suggests that managers also have an incentive to manipulate real activities; cross-country variation in the level of real activities manipulation could potentially induce spurious relations. Lastly, unlike price changes, operating cash flows co-mingles the effects of both past and current news events. Despite the respective limitations of both the non-linear earnings-returns model and the non-linear accruals-cash model, consistent results using timely loss recognition measures (i.e., TLR k s) derived from both of these techniques will mitigate concerns about the potential limitations of each methodology. 3.3 Measurement of economic income and the incidence of economic losses We measure each firm-year s economic income realization as the firm s holding period stock return, including dividends, over the fiscal year (i.e., RET i,t ). To the extent that economic 17

18 income is negative, we set the indicator variable NEG i,t equal to one, zero otherwise. Our first set of tests examine the likelihood of a specific firm-year observation being associated with negative (versus non-negative) economic income given certain country-level and firm-specific attributes, including the presence of timely loss recognition practices in the economy. We also examine the relative incidence of economic losses annually at the country level. We define the variable PROPNEG k,t as the country-level average of NEG in a given fiscal year; as such, PROPNEG k,t reflects the percentage of country k s firms included in our sample that incurred negative returns in fiscal year t. Our second set of tests examine whether variation in the annual proportion of firms generating economic losses in a country is inversely associated with timely loss recognition practices (i.e., TLR k ) in that country. 3.4 Measuring Investment Flow Efficiency We use the approach developed by Wurgler (2000) to measure investment flow efficiency, or the extent to which capital in each country is allocated to value creating opportunities and withdrawn from value destroying ones. Wurgler (2000) estimates the elasticity of gross investment to value added as a measure of the efficiency of resource allocation in each country from equation (3): ln( I jkt / I jkt-1 ) = α k + η k ln( V jkt / V jkt-1 ) + ε, (3) where I jkt is gross fixed capital formation in industry j, country k, year t, V jkt is value added in industry j, country k, year t. 9 He interprets the elasticity, η k, for each country k, as a measure of 9 Value added is defined as the value of shipments of goods produced (output) minus the cost of intermediate goods and required services (but not including labor), with appropriate adjustments made for inventories of finished goods, work-in-progress, and raw materials. In other words, this value added measure reflects value added by labor as well as capital. The sum of value added across all firms in the economy is GDP. Value added growth is used as a proxy for investment opportunities. Gross fixed capital formation is defined as the cost of new and used fixed assets minus the value of sales of used fixed assets, where fixed assets include land, buildings, and machinery and equipment. 18

19 the extent to which country k reduces investment in declining industries and increases investment in growing industries. In addition, Wurgler disaggregates η k into two distinct components, estimating the elasticity in each country k separately for industry-year observations reflecting increasing value added (η + k ) and decreasing value added (η - + k ). Interpretatively, η k captures the overall tendency for managers in country k to allocate capital to growing sectors, and η - k the overall tendency for managers in country k to withdraw capital from shrinking sectors. The difference (η - k - η + k ) in each country can be viewed as a measure of the severity of the control problems in a country, as self-serving managers are less likely to downsize investments in declining projects than they are to increase investments in growth opportunities. (e.g., Jensen (1986)). We examine the relations between our proxies for TLR and η k, η + k, η - k, and (η - k - η + k ). Combining this elasticity data with our country-level TLR measures yields a maximum sample of 31 country-level observations for the elasticity-based tests. Descriptive statistics for these 31 country-level observations are presented in Table Empirical evidence: Economic losses and timely loss recognition practices 4.1 Descriptive evidence Table 1 presents descriptive evidence on the sample of firms used in our analysis of the incidences of economic losses. This sample consists of 78,333 firm-year observations drawn from 37 countries, and comprising 358 country-year observations. 10 Panels A and B present 10 The returns sample consists of all available firm-year observations on Global Vantage for the 37 countries with sufficient data to estimate our TLR measures, after excluding the top and bottom one-half of one percent of RET, MVE and BM observations. These observations were eliminated from the sample to mitigate the effect of outliers and data coding errors on our analysis. 19

20 descriptive statistics for the firm-year and country-year observations, respectively, while panels C and D present correlation matrices. In terms of sample composition, these observations are weighted toward firms drawn from larger, more developed countries (89.9 percent) and countries with a common law legal origin (53.4 percent). Moreover, 57.9 percent of all observations are drawn from the United States (n=22,983) and Japan (n=22,417). Given the potential for these observations to exert an undue influence on the research design, all results from the loss incidence tests are presented both including and excluding U.S. and Japanese observations. In terms of underlying economic performance, 50.9 percent of the firm-year observations in the sample are associated with negative economic income, with a mean one-year return of 7.81 percent. Consistent with prior cross-country research (e.g., Ball, Kothari and Robin, 2000; Bushman and Piotroski, 2005), our estimates of timely loss recognition practices display significant cross-country variation. For example, BKR_TIME k ranges from (Austria) to (Mexico), with a mean timeliness coefficient of (see Appendix 2 for TLR k estimates by country). Lastly, our evidence suggests that, on average, only one quarter of all domestic listed firms each year are included in our sample (average country-level MKT_PCT k,t realization is 23.6 percent). Given the significant variation in this percentage across country-year observations, these statistics suggest that Global Vantage coverage decisions, especially in smaller economies, could be tilted towards firms with strong historical performance, and therefore potentially introduce a bias into our returns-based analyses. Focusing on the firm level correlations (panel C), several interesting relations emerge. First, despite the different methodologies employed, the two unique sets of timely loss recognition measures are strongly correlated. For example, BKR_INCR k and BS_INCR k have a 20

21 spearman correlation of 0.708, while BKR_TIME k and BS_TIME k have a correlation of Thus, despite the different empirical assumptions underlying the two TLR estimation techniques, the relative ranking of countries into timely loss recognition regimes across these metrics appears to be quite similar. Second, firm-level correlations confirm that all TLR k s are lower in code law regimes and in emerging markets, consistent with prior research. Third, evidence from the country-level correlation matrix (panel D) confirms the preceding relations, as well as documents a fairly robust negative relation between our timely loss recognition measures and countries average book-to-market ratios. One explanation for these negative relations is a potential mechanical relation between conditional and unconditional conservatism practices in an economy. Ceteris paribus, differences in the amount of unconditional conservatism practiced in across countries will systematically alter average book values across countries, resulting in crosscountry variations in book-to-market ratios (e.g., Beaver and Ryan (2005)). Because accounting regimes with strong unconditional conservatism practices offer very fewer opportunities to exercise conditional conservatism (i.e., timely loss recognition practices), this mechanical relation between conditional and unconditional conservatism practices could induce the observed relation between book-to-market ratios and TLR k. Finally, these correlation matrices document an inverse relation between the presence of negative returns (NEG i,t and PROPNEG k,t ) and our four measures of timely loss recognition (TLR k ). These panels also illustrate, however, that stock returns and the presence of economic losses are both correlated with firm and country-level attributes, including the apparent bias in Global Vantage s coverage. As such, the univariate relations should be interpreted with caution. 4.2 Distribution of realized stock returns across time loss recognition regimes 21

22 Our first hypothesis suggests that the returns distribution in timely loss recognition regimes will systematically differ from the returns distribution in regimes without time loss recognition practices. Specifically, managers (and firms) operating in a strong timely loss recognition regime, and facing early intervention mechanisms driven by timely loss recognition, are less likely to invest in negative NPV projects and are more willing to abandon projects earlier (and will have their capital funding bad projects terminated quicker) than managers (and firms) operating in countries without timely loss recognition practices. If both better investment selection decisions and the early abandonment of bad projects lead to better economic outcomes for a firm, we expect to observe that return distributions are shifted to the right (i.e., the density function should have less mass to the left of zero) in timely loss recognition regimes. For comparison purposes, countries are classified as having high timely loss recognition practices if the country s respective bad news timeliness metric (i.e., TLR k ) is greater than or equal to the median country s realization; otherwise, the country is classified as having low timely loss recognition practices. Next, we estimate the approximate shape of the probability density function of realized returns in each accounting regime. Specifically, realized returns are classified into one of 21 bins, with each bin spanning a realized return interval of ten percent. This technique produces a measure of the percentage of firm year observations within a given TLR k regime falling within a given annual return interval. Under our hypotheses, low TLR k regimes should have a greater proportion of their observations falling into negative return bins relative to the proportions observed in the same intervals for high TLR regimes. Table 2 (along with Figure 1) presents evidence on the shape of the realized returns distribution across different TLR k regimes. 22

23 Panel A presents distribution evidence for the complete sample of 37 countries, with observations pooled into their TLR k regime, while panel B presents evidence excluding the presence of US and Japanese observations. The primary conclusion from these panels is that the proportion of negative returns is systematically lower in strong TLR k regimes, regardless of the TLR k measure employed or whether or not the analyses included U.S. and Japanese observations. 11 Moreover, this table illustrates that the mass of the distribution of these return realizations is systematically shifted to the right in the high TLR k regimes, as evidenced by the higher mean and median returns realized in these regimes, as well as (generally) the presence of a fewer proportion of observations in the negative return bins of the distribution. Although suggestive of systematic differences in economic outcomes across high and low TLR k regimes, this univariate evidence is subject to a host of omitted variable concerns. Only after controlling for firm-specific and country-level differences in risk, differences in sample composition and differences in underlying economic attributes can one reasonably infer that TLR k behavior is associated with better economic performance and more efficient capital allocation behavior. However, measuring differences in cross-country returns distributions for purposes of multivariate analyses is challenging. One empirical construct that seems to be both theoretically and empirically correlated with these differences in economic outcomes is the likelihood of observing a negative return (i.e., economic loss being realized). As illustrated in table 2, the proportion of firm year observations associated with negative returns is systematically higher in low TLR k regimes. As such, this statistic appears to be capturing meaningful differences in the distributional property and economic outcomes we desire to test across accounting regimes. The next section examines this issue further. 11 Differences in proportions across TLR regimes (reported in panel B) are all significantly different at the one percent level using a binomial test of proportions. 23

24 4.3 Multivariate evidence on the incidences of economic losses To test for differences in the incidences of loss realizations across TLR regimes, we perform two sets of tests. The first analysis examines the likelihood of a given firm-year observation being associated with a negative return given that country s timely loss recognition practices. The second analysis aggregates firm-level data to estimate the percentage of firms in a given economy generating economic losses in a given fiscal year Multivariate evidence: Firm-year estimations Table 3 presents coefficients from various estimations of the following cross-sectional logistic model: Prob(NEG i,t =1) = Logit (α γt t t= 1993 Y + β1emerge k + β 2 CIVILLAW k + β 3 MKTPCT k,t + β 4 log(mve + β 5 log(1+bm i,t ) + β 6 RNK_TLR k + ε) (4) In this model, Y t is an annual indicator variable to control for annual fixed (market) effects influencing return outcomes, EMERGE k is an indicator variable equal to one if the firm operates in an emerging market economy, CIVILLAW k is an indicator variable equal to one if the firm operates in a code law economy, MKTPCT k,t is the percentage of domestic listed firms in country k included in the sample in year t, log(mve i,t ) and log(1+bm i,t ) are logarithmic transformations of the firm s market capitalization of equity and market-to-book ratio, respectively, at the beginning of year t, and RNK_TLR k is the fractional rank (i.e., between zero and one) of the respective country-level timely loss recognition (TLR k ) measure. 12 i,t) 12 We rank our TLR k measures to eliminate any non-normality and cross-tlr metric scale effects in the underlying estimates of TLR, while preserving information about relative differences in TLR attributes across countries. Our results and inferences are qualitatively similar if the raw TLR measures (i.e., the estimated coefficients from the non-linear earnings-returns and accruals-cash flow models) are used in lieu of these rank transformations. 24

25 In these estimations, the variables EMERGE k and CIVILLAW k are designed to control for cross-country differences in economic development and other legal and/or financial architecture institutions that could influence realized levels of economic income (see LaPorta et al. (1998); (2002)). The variables log(mve i,t ) and log(1+bm i,t ) are designed to control for firmspecific differences in risk, as well as potential differences in investment opportunities and potential misvaluation attributes that could influence realized returns. Finally, the variable MKT_PCT k,t is designed to control for any potential positive historical performance / return bias associated with Global Vantage s selection criteria. Panel A of Table 3 presents coefficients and two-tailed p-values from a pooled, crosssectional estimation of equation (4) for each TLR k variable. To deal with cross-sectional dependence in the return data, panel B presents average coefficients from ten annual estimations of equation (4) for each TLR k variable. Reported p-values reflect t-statistics and standard errors based on the empirically-derived distribution of these ten annual coefficients. Results are presented both including and excluding US and Japanese observations. First, both size and book-to-market ratios have explanatory power, with smaller firms and low book-to-market firms having a greater likelihood of incurring negative returns. In the case of firm size, this could be an artifact of investment diversification; for low BM firms, this could be artifact of possible overvaluation associated with glamour firms (i.e., an international value - glamour effect; see Fama and French (1998)). Second, there appears to be a systematic bias between firm returns and Global Vantage coverage. Specifically, greater depth of coverage (i.e., higher MKTPCT) increases the likelihood that a given firm-year observation is associated with a negative return realization. This positive relation suggests that Global Vantages selection criteria among thinly covered countries is biased towards covering firms with strong actual 25

26 and/or expected performance. As a result, selective Global Vantage coverage results, ex post, in fewer negative return realizations for that country. Finally, after controlling for these structure relations in the return data, both sets of estimations, regardless of sample composition and metric of TLR k, continue to document an inverse relation between timely loss recognition practices and negative returns. Moreover, these relations are remarkably stable between the use of pooled and annual estimates. Although the p- values generated using average annual coefficients are larger than those generated in the pooled model, the coefficients themselves are quite consistent across specifications. In the most demanding methodological setting (panel B), where statistical significance is being assessed on the basis of only ten annual coefficient estimates, four (six) of these estimations generate average TLR k coefficients that are significantly different than zero at the five (ten) percent level of significance when assessed using a one-tailed t-test Multivariate evidence: Country-level estimations The preceding evidence supports the hypothesis that economic losses are less prevalent in economies where accounting practices recognize losses in a timely manner. However, in these estimations, the results could be disproportionately tilted to reflect relations driven by countries with more developed equity markets (i.e., countries with larger number of observations in our sample). An alternative research design is to measure attributes at the country level, effectively constructing a panel of country-year observations, and examining whether country level economic income attributes vary across TLR k regimes. Table 4 presents coefficients from various estimations of the following country-level, cross-sectional model: 26