The Usefulness of Fair Value Accounting in Executive Compensation

Transcription

1 The Usefulness of Fair Value Accounting in Executive Compensation Mark DeFond University of Southern California Jinshuai Hu Xiamen University Mingyi Hung The Hong Kong University of Science and Technology Siqi Li Santa Clara University January 2018 Abstract We investigate the effect of fair value accounting on the usefulness of earnings in executive compensation contracts. Our analysis uses a shock-based difference-in-differences research design that exploits the 2005 worldwide mandatory adoption of IFRS, and employs a firm-level measure of the fair value treatment effect. We find that earnings pay-performance sensitivity (PPS) declines among the IFRS adopters that are most affected by IFRS s fair value provisions relative to the IFRS adopters that are least affected by its fair value provision. Our findings are consistent with the notion that IFRS, on average, improves the usefulness of earnings in executive compensation contracts, but that its fair value provisions offset this improvement among the firms most affected by those provisions. These results are primarily driven by firms in countries with strong enforcement, and are generally robust for industrial firms but hold only weakly for financial firms. We further find that increased earnings management, rather than increased earnings volatility, is the most likely channel through which fair value accounting impairs earnings PPS. Our findings contribute to the literature on the contracting usefulness of fair value accounting by presenting evidence that suggests fair value accounting impairs the usefulness of earnings in compensation contracts. Acknowledgments: We thank the following for their helpful comments: Kevin Chen, Yongtae Kim, Charles Shi, Haifeng You, and workshop participants at National University of Singapore.

2 The Usefulness of Fair Value Accounting in Executive Compensation I. INTRODUCTION A large body of literature addresses the use of historical cost versus fair value accounting as the measurement basis for financial statements. Most of this research focuses on equity market valuation and suggests that fair value-based accounting numbers are informative to investors (Landsman 2007). In contrast, research on the contractual usefulness of fair value accounting is limited and inconclusive. While several studies investigate this issue from a debt contracting perspective (e.g., Ball, Li, and Shivakumar 2015; Demerjian, Donovan, and Larson 2016), few studies examine compensation contracting issues, and those that do are confined to a limited subset of fair value items. 1 Given that a principal objective of GAAP is to provide information that is useful for performance evaluation (Kothari, Ramanna, and Skinner 2010), we attempt to provide large sample evidence on whether and how the increased use of fair value accounting affects the usefulness of earnings in executive compensation. Our approach differs from prior studies by using a difference-in-differences (DiD) research design that exploits a global financial reporting shock that incorporates a firm-level measure of the fair value treatment effect. Our study takes advantage of the 2005 worldwide mandatory adoption of IFRS, a setting that offers several advantages for testing the usefulness of fair value accounting in executive compensation contracts. First, for most of the affected firms, the IFRS mandate represents a sharp regime shift from a historical cost-oriented accounting system to a fair value-oriented accounting 1 As far as we are aware, prior published research that examines the sensitivity of executive compensation to fair value accounting is limited to just three studies. Livne, Markarian, and Milne (2011) find no evidence that fair value affects executive compensation, while Manchiraju, Hamlin, Kross and Suk (2016) and Chen and Tang (2017), which study firms in the oil and gas industry, and Hong Kong property companies, respectively, conclude that fair value gains, but not losses, affect executive compensation. 1

3 system (Ball et al. 2015). Second, first-time IFRS adopters are required to restate their prior year s financial statements from local GAAP to IFRS, allowing us to identify the firms most affected by IFRS s fair value provisions. These advantages enable us to employ a firm-level measure of the treatment effect in our DiD research design. This is an improvement upon most prior IFRS research, which tends to employ country-level measures of their treatment effects (e.g., Li 2010; Ozkan, Singer, and You 2012; Ball et al. 2015). Using firm-level effects in the context of a DiD model helps to mitigate the influence of contemporaneous confounding events and country-level correlated omitted variables. Further, because IFRS was adopted across a large number of countries, we are also able to study the role of country-level institutional factors, thereby allowing for a more in-depth understanding of the impact of fair value accounting on executive compensation when compared to single-country studies. Our approach to evaluating the usefulness of historical cost versus fair value accounting in executive compensation is to examine the sensitivity of executive compensation to reported earnings, which is referred to in the literature as earnings pay-performance sensitivity (hereafter, earnings PPS). In designing executive compensation contracts, contracting theory argues that firms will place relatively greater weight on performance measures that are more sensitive to the agent s effort (e.g., Banker and Datar 1989; Holmstrom 1979). This suggests that the optimal performance measures used in executive compensation contracts may differ from those used for equity valuation purposes (which focus on future cash flows), or debt contracts (which focus on repayment ability) (Natarajan 1996; Ball et al. 2015). If fair value accounting is better than historical cost accounting at capturing managerial effort, then the adoption of IFRS s fair value provisions should cause firms to put greater weight on accounting-based earnings when compensating executives. If this is indeed the case, we predict that IFRS adoption increases earnings PPS among the firms most 2

4 affected by IFRS s fair value provisions. Conversely, if fair value accounting reduces the informativeness of earnings with regard to managerial performance, by adding noise and/or by facilitating more opportunistic financial reporting, then the adoption of IFRS s fair value provisions should lead to firms placing less weight on accounting-based earnings. If this is the case, we predict that IFRS adoption decreases earnings PPS among the firms most affected by IFRS s fair value provisions. Our empirical analysis examines a sample of 13,438 executive-year observations from 1,608 unique industrial firms, and 3,057 executive-year observations from 410 unique financial firms, in 20 countries that mandate IFRS adoption in We compare the three years before the adoption (the pre-period), with the three years after the adoption (the post period). We separately examine industrial and financial firms because the fair value provisions in IFRS affect them differently (DeFond et al. 2015). In particular, the fair value provisions of IFRS affect a large number of accounts for industrial firms, while their effect on financial firms is concentrated among financial instruments. We begin by performing a baseline analysis that examines the overall effects of mandatory IFRS adoption on earnings PPS, before considering the firm-level treatment effect of IFRS s fair value provisions. We find that earnings PPS modestly increases after IFRS adoption for industrial firms, but does not change for financial firms. Our findings for industrial firms are consistent with Ozkan et al. (2012), who conclude that IFRS makes earnings more contractually useful in management compensation contracts. We test our predictions using a DiD design that compares the changes in earnings PPS of the firms most affected by IFRS s fair value provisions (our treatment firms), with the firms least affected by IFRS s fair value provisions (our benchmark firms). We use the reconciliations from local GAAP to IFRS reported during the transition year to identify the firms most affected by 3

5 IFRS s fair value provisions (e.g., PP&E, investments, and provisions). We classify our treatment firms as those with absolute reconciliation amounts greater than the sample median, and our benchmark firms as those below the median. Our tests yield two primary findings. First, for the benchmark firms, earnings PPS increases for both industrial and financial firms. This is consistent with the non-fair value effects of IFRS resulting in improved earnings informativeness, which leads to an increase in earnings PPS. Second, relative to the benchmark firms, the treatment firms experience a decrease in earnings PPS subsequent to the mandate. This relative decline is the result of a significant increase in earnings PPS among the benchmark firms, combined with an insignificant decrease in earnings PPS among the treatment firms. However, the relative decline among the treatment firms is not entirely explained by the increase in PPS among the benchmark firms. For industrial firms in countries with strong enforcement, the treatment firms experience a significant decrease in earnings PPS, while the benchmark firms continue to experience a significant increase in PPS. Taken together, our findings are consistent with IFRS adoption, on average, improving the usefulness of earnings in executive compensation contracts, but with its fair value provisions offsetting this improvement. A potential concern in our analysis is that our treatment firms may not be comparable to our benchmark firms. To address this, we repeat our primary analysis using a propensity-scorematched (PSM) sample, and find consistent results. We also assess the validity of the parallel trends assumption underlying our DiD estimation with two tests. Our first test plots the annual earnings PPS of the treatment and benchmark firms over the pre- and post-ifrs adoption years. For industrial firms, we find similar pre- and post-period trends in earnings PPS for the treatment and benchmark firms. We also find that after adopting IFRS, earnings PPS increases for the 4

6 benchmark firms and decreases for the treatment firms. The increase in earnings PPS among the benchmark firms is consistent with the non-fair value effects of IFRS improving earnings quality, while the decline in earnings PPS among the treatment firms is consistent with the fair value effects of IFRS reducing earnings quality. The results for financial firms are less conclusive. Our second test assesses the pre- and post-period trend by performing placebo tests using pseudo adoption years. This analysis finds no evidence of changes in earnings PPS subsequent to the pseudo adoption year for both industrial and financial firms. We also investigate the channels through which fair value accounting impairs the usefulness of earnings in compensation contracting. We find a relative increase in the likelihood of meeting or beating earnings targets for the industrial firms that are most impacted by IFRS s fair value provisions subsequent to the mandate as compared to those that are least affected. We also find a relative decrease in earnings volatility for both the industrial and financial firms that are most impacted by the fair value provisions. Together, these findings suggest that increased earnings management, rather than increased earnings volatility, is a likely channel through which fair value accounting affects the usefulness of earnings in compensation contracting. We evaluate the role of country-level institutions by performing an analysis conditional on country-level enforcement. An underlying assumption for our prediction is that boards are acting on behalf of shareholders interests to set optimal incentive-compatible contracts. Prior literature suggests that this assumption is more likely to hold in countries with stronger legal institutions (La Porta et al. 2000). Consistent with this expectation, we find that the relative decline in earnings PPS after IFRS adoption is driven primarily by treatment firms located in countries with strong enforcement. We also find evidence of increased earnings manipulation among the industrial firms 5

7 most impacted by the fair value provisions in these countries, consistent with the notion that there is greater capital market pressure for firms to meet or beat earnings targets in these countries. We note that our interest is in how compensation committees react to the increased use of fair value accounting, regardless of whether firms voluntarily use fair value measures, or whether they are mandated under IFRS. However, if firms that voluntarily adopt IFRS s fair value provisions differ systematically from those that do not, it raises the concern that this selection may confound our analysis. While the results of our PSM and parallel trend analysis help alleviate this potential concern, we also perform additional tests aimed at isolating the effects of IFRS s mandatory fair value provisions. While many fair value provisions are optional under IFRS, IAS 39 mandates fair value measurement of investment securities, and affects a large number of both industrial and financial firms in our sample. Thus, we repeat our primary analysis using two measures of the effect of fair value provisions: one that captures the effects of IAS 39, which are primarily mandatory, and the other captures the effects of the other fair value provisions, which are primarily voluntary. This analysis finds that our results hold for both types of fair value changes, suggesting that our results are not exclusively driven by the discretionary fair value provisions of IFRS. Finally, we find that our results for industrial firms are robust to a variety of sensitivity tests, including the use of alternative samples, standard error clustering schemes, and control variables. We caution, however, that the results for financial firms are weaker and more sensitive to our design choices, consistent with the effect of fair value accounting on financial firms being more selective. 6

8 Our study contributes to the literature in several ways. First, we add to the scarce and inconclusive evidence on the effect of fair value accounting on executive compensation. 2 Livne et al. (2011), using a sample of 152 U.S. banks, find that fair value income resulting from SFAS 115 has no incremental power beyond stock returns in explaining executive compensation. Manchiraju et al. (2016), find that fair value based derivative gains, but not losses, affect executive compensation among a sample of oil and gas firms. Chen and Tang (2017), looking at a sample of 70 Hong Kong property companies, find that revaluation gains, but not revaluation losses, become a significant determinant of executive compensation following IFRS adoption. Because these studies examine the incremental usefulness of fair value gains and losses to other earnings components, data constraints require them to focus on single-industry and single-country settings. In contrast, we examine the usefulness of overall earnings in response to the change from an historical cost-oriented to a fair value-oriented accounting system. As a result, we are able to focus on a large sample of firms across a variety of industries and institutional environments and shed light on the consequences of fair value accounting on the use of earnings in executive contracting. Second, our study contributes to the literature that examines the stewardship role of accounting information in executive compensation. This literature suggests that earnings PPS depends on the ability of earnings to reflect managerial effort and provide incentive alignment (Baber, Kang, and Kumar 1998; Bushman, Engel, and Smith 2006). Because earnings properties are determined by managerial actions, the endogeneity issue is inherently challenging to tackle using a cross-sectional 2 A related stream of literature examines how executive compensation contracts affect management s use of fair value accounting (e.g., Shalev, Zhang, and Zhang 2013). In addition, some studies use a cross-sectional design to examine the association between fair value-based earnings and compensation. For example, Dechow, Myers and Shakespeare (2010), using a sample of 96 U.S. companies, employ a cross-sectional research design, and find no evidence that the association between fair value-based asset securitization gains and executive compensation differs from its association with other (non-fair value-based) earnings components. 7

9 research design. We complement prior research by employing a shock-based research design that employs a DiD analysis and a firm-specific treatment measure, which collectively help alleviate endogeneity concerns (Atanasov and Black 2016). Third, we add to the literature that examines the impact of IFRS adoption on the decision usefulness of accounting information. Most prior studies focus on the value relevance of financial statement information under IFRS. Among the limited studies examining whether mandatory IFRS adoption affects the contracting usefulness of accounting information, the evidence is mixed. Ball et al. (2015) find a decline in the use of accounting debt covenants after IFRS adoption and conclude that IFRS reduces the contractibility of accounting information in debt contracting. Ozkan et al. (2012) find weak evidence of an increase in the usefulness of earnings in executive compensation after IFRS adoption, and conclude that IFRS facilitates executive compensation contracting. 3 Our study extends this literature by finding that IFRS adoption has a more nuanced effect on the use of accounting information in executive compensation. While mandatory IFRS adoption, on average, increases earnings PPS for industrial firms and has no effect for financial firms, firms that are most affected by IFRS s fair value provisions experience a decrease in earnings PPS for both industrial and financial firms, relative to those least affected by IFRS s fair value provisions. Our study highlights the importance of separately considering the effect of IFRS s fair value provisions. Further, our finding suggests that increased earnings management, rather than increased earnings volatility, is the most likely channel through which fair value accounting impairs earnings PPS. 4 3 Like our study, Ozkan et al. (2012) and Ball et al. (2015) are cross-country studies. Among single-country studies, Ke, Li, and Yuan (2016) find earnings PPS declines for central government-controlled firms after IFRS convergence in China. 4 In their review of IFRS adoption literature, De George, Li, and Shivakumar (2016, p. 954) state A major limitation of IFRS studies focusing on stewardship is that they do not pinpoint the mechanisms through which IFRS adoption 8

10 II. HYPOTHESIS DEVELOPMENT Mandatory IFRS Adoption and Fair Value Accounting The 2005 mandatory IFRS adoption represents a watershed event for the financial reporting practices of thousands of companies worldwide and establishes IFRS as the single most widely used accounting standard in the world. It is widely argued that a capital market benefit of IFRS adoption is increased reporting transparency (GAAP 2001). Consistent with this notion, prior studies find that IFRS, when credibly implemented, results in a reduction in the cost of capital and an improvement in a firm s information environment (e.g., Li 2010; Tan, Wang, and Welker 2011). A defining feature of IFRS is its emphasis on fair value accounting. While most local GAAP is historical cost oriented, IFRS is fair value oriented, resulting in a strong shift toward fair value accounting upon its adoption (Ball et al. 2015). 5 Specifically, IFRS incorporates fair value measurement in valuing various short-term and long-term assets and liabilities such as PP&E (IAS 16), employee benefits (IAS 19), provisions (IAS 37), and financial instruments (IAS 39). The costs and benefits associated with IFRS s fair value provisions are highly debated (DeFond et al. 2015). Proponents of the fair value provisions argue that fair value information is relevant affects stewardship or identify the specific accounting attributes that drive these changes. Another common problem associated with IFRS studies that focus on stewardship is their inability to find a substitution effect among accountingand non-accounting-based performance measures. Our study addresses these limitations by (1) separately considering the effect of IFRS s fair value provisions, (2) identifying earnings management as the channel through which fair value impairs the usefulness of IFRS for compensation contracting, and (3) providing evidence on the substitution effect of the use of earnings versus cash flows in executive compensation. 5 IFRS defines fair value as the price that would be received to sell an asset or paid to transfer a liability in an orderly transaction between market participants at the measurement date (IFRS 13: Fair value measurement). Its fair value hierarchy includes three inputs: Level one uses quoted prices in an active market (often referred to as mark-tomarket ), Level two allows reported values to be based on the quoted price of a comparable asset or liability, and Level three allows reported values based on unobservable inputs (often referred to as mark-to-model ). 9

11 because it better reflects a firm s underlying risk profiles and economic performance. This view is held, for example, by the IASB: In many accounting pronouncements, the Board has concluded that fair value information is relevant, and users of financial statements generally have agreed. (IASB 2006, p. 57) IAS 39 requires derivatives to be reported at their fair or market value, rather than at cost. This overcomes the problem that the cost of a derivative is often nil or immaterial and hence if, derivatives are measured at cost, they are often not included in the balance sheet at all and their success (or otherwise) in reducing risk is not visible. In contrast, measuring derivatives at fair value ensures that their leveraged nature and their success (or otherwise) in reducing risk are reported. (IASB 2003, p.1). Opponents of fair value provisions, however, argue that they introduce noise and obscure true performance. For example, an article from the Financial Times states: Many company directors are still disputing whether fair value accounting gives a more meaningful insight into a company's economic performance than other measures Many EU banks and financial services companies complained about the introduction of market values for some of their businesses, pointing out that it would lead to volatility in their income statements. This volatility would not necessarily capture the underlying economic performance of the company, but would reflect its exposure to oscillating derivatives market Many companies are also uncomfortable with the way IAS 39 restricts hedge accounting by making them show that the instruments they use are genuinely for hedging purposes. (Hargreaves 2005). Consistent with this view, Association of French Financial Analysts expresses the following view in the introduction of fair value accounting in France: The use of fair value can confuse interpretation of a company s operational results. Fair value accounting is less reliable, allows greater manipulation of results and introduces volatility. (Hawkins, Dessain, and Barron 2008). It is worth noting that the IASB s view assumes that equity valuation is the primary objective of GAAP, an assumption that is highly controversial. Kothari et al. (2010) argue that the objective of GAAP is to facilitate the efficient allocation of capital. Under this economic framework, GAAP s principal focus should be performance evaluation and stewardship, which may be more accurately measured using an accounting system that focuses primarily on historical costs. In the 10

12 next section, we discuss how fair value accounting may affect the role of earnings in performance evaluation and executive compensation. The Effect of Fair Value Accounting on Earnings PPS Contracting theory suggests that when multiple performance measures are present, optimal incentive-compatible contracts should place greater weight on measures that are more precise and more sensitive to the agent s effort (e.g., Banker and Datar 1989; Holmstrom 1979). Thus, the use of the performance measures in executive compensation contracting can differ from equity valuation, where the focus is on the information about future cash flows (Natarajan 1996). It can also differ from debt contracting, where the important consideration is about a firm s contractual obligation to pay (Ball et al. 2015). Fair value accounting may increase the usefulness of earnings in executive compensation contracting by reflecting management s contribution to firm performance in a timelier manner (De George et al. 2016). This is because historical cost accounting is inherently less timely than current fair values, thereby increasing the information asymmetry between managers and investors. Specifically, while fair value accounting permits both write-downs and write-ups that reflect current asset values, historical cost accounting requires assets to be valued at their initial cost, and only permits write-downs when their recoverability is in question. We also expect fair value to be more useful in compensation contracts if it improves transparency and comparability of earnings, thereby making earnings a more precise measure of managerial effort. On the other hand, there are various reasons why fair value accounting may decrease earnings PPS. One reason is that fair value accounting leaves more room for earnings management, which impairs the ability of earnings to reflect managers true performance and hence its usefulness in 11

13 executive compensation. This is consistent with Dechow et al. (2010), who suggest that managers use discretion obtained from fair value accounting rules for securitization to smooth or boost earnings. It is also consistent with Kothari et al. (2010), who observe that the lack of verifiability of many fair value measurements impairs the contracting usefulness of fair value accounting. As an example, they state: the current requirement under GAAP to impair goodwill, because of its inherent subjectivity, seems ill-suited to improving management s accountability. (p. 263) Another reason why fair value accounting may impair the usefulness of earnings in compensation, is that it introduces greater earnings volatility, and incorporates uncontrollable market-wide movement, which makes earnings a noisy measure of management performance (Sloan 1993; Baber et al. 1998). Related to this argument, De George et al. (2016) point out that the use of fair value accounting lowers the distinction between earnings and stock prices. Thus, earnings may lose its advantage relative to stock returns for use in compensation contracts. In summary, if fair value accounting improves the ability of earnings information to reflect managerial effort and management s contribution to firm profitability, earnings are expected to become more useful in executive compensation among firms for which the fair value provisions of IFRS have a relatively greater impact. Alternatively, if the use of fair value accounting introduces more opportunistic reporting and/or adds noise to accounting numbers, the contractual usefulness of earnings in executive compensation is expected to decrease among these firms. Thus, we predict the following non-directional hypothesis: Hypothesis: Earnings PPS may either increase or decrease subsequent to the increased use of fair value accounting under IFRS adoption. 12

14 We note that we are agnostic about whether corporate boards respond to IFRS adoption by changing the weight they place on reported earnings, or by making adjustments to the earnings numbers they use in compensation contracts. While the use of earnings-based performance measures is common in compensation contracts, it is well recognized that contracting parties often make adjustments to the reported earnings or use alternative performance measures to suit their purpose (Sloan 1993; Kothari et al. 2010). If boards believe that fair value gains or losses are beyond managers control, they may exclude these components or use non-accounting based measures, which in turn leads to a decrease in earnings PPS. Making such adjustments, however, is not costless, and their necessity is consistent with a reduction in contracting efficiency. We also expect that fair value accounting may be relatively more useful in executive compensation contracts for financial firms than for industrial firms. This is because risk trading and risk management are critical tasks for executives in financial firms. As a result, since fair value accounting is better at capturing the risk associated with financial instruments, boards of financial firms may be more likely to find that performance measures based on fair values are superior to measures based on historical cost. It is also notable that the boards of financial firms may find it easier to adjust reported earnings for the effects of the fair value provisions. This is because IFRS s fair value provisions affect financial firms primarily through their impact on financial instruments, which are disclosed separately, whereas IFRS s fair value provisions affect industrial firms through a variety of accounts, most of which are not separately disclosed. Data and Sample Selection III. DATA AND RESEARCH DESIGN 13

15 We collect financial data from Worldscope for firms in countries that mandated IFRS adoption in Our sample period consists of the last three fiscal years before the IFRS mandate (i.e., the pre-adoption period) and the first three fiscal years after the IFRS mandate (i.e., the post-adoption period). Following prior literature, we exclude the first year of IFRS adoption (e.g., 2005 for a December year-end firm) because firms may need time to adjust their compensation contracts after adopting IFRS (Ozkan et al. 2012). Our sample consists of mandatory IFRS adopters that report under local accounting standards in the three-year pre-adoption period and under IFRS in the threeyear post-adoption period. For December year-end firms, for example, the pre-adoption period is and the post-adoption period is We focus on mandatory IFRS adoption countries because we wish to exploit the firm-level variation in fair value treatment effects within each country in our DiD analyses. This approach has the advantage of using benchmark firms with the economic and regulatory commonality as the treatment firms, and therefore overcomes the limitation in prior studies that benchmark mandatory IFRS adopters with non-ifrs adopters in other countries. We obtain the executive-level data on key executives from S&P Capital IQ. 6 This database contains detailed compensation and personnel information on firms executives in more than 100 countries starting in To make our sample firms more comparable across countries, we require sample firms to have positive net income before interest and taxes, common equity and net sales greater than ten million U.S. dollars, and executives with annual cash compensation of no less than ten thousand U.S. dollars. After excluding outliers of firm-level continuous variables at the top and bottom one percentile, our final sample consists of 13,438 executive-year observations 6 Capital IQ reports compensation information for top key executives, including: Chairman or Co-Chairman of the Board, Chief or Co-Chief Executive Officer, President or Co-President, Chief or Co-Chief Financial Officer, Chief or Co-Chief Operating Officer, and Chairman & Vice Chairman of Management Board. 14

16 from 1,608 unique industrial firms (one-digit SIC code is not 6 ), and 3,057 executive-year observations from 410 unique financial firms (one-digit SIC code is 6 ), from 20 IFRS adoption countries. Our sample is substantially larger than the sample in Ozkan et al. (2012) due to the larger coverage in the database used in our study (Capital IQ) than the database used in their study (BoardEx). Table 1 reports the distribution of our sample by country and year. As shown in Panel A, the numbers of firms and executive-years vary widely across countries. For example, for industrial firms, the U.K. (5,047 executive-years and 448 firms) and Australia (4,364 executive-years and 465 firms) have the two largest numbers of executive-years and firms, while Austria (three executive-years and one firm) has the smallest. The sample of financial firms has fewer observations than industrial firms, but the distribution across countries is similar to that of the industrial firms. Panel A also reports our two country-level enforcement measures, Regulatory Quality and Supervisory Power, where higher values denote stronger enforcement. Hong Kong has the highest score (1.88) in Regulatory Quality and Philippines has the lowest (-0.13). Portugal and Switzerland have the highest score (14.0) in Supervisory Power while Italy has the lowest (7.0). Panel B shows that the number of distinct firms (executives) ranges from 66 (308) in 2002 to 559 (3,044) in 2007 for industrial firms, and from 12 (81) in 2002 to 124 (762) in 2007 for financial firms. This pattern may be due, in part, to the increased coverage of Capital IQ over the sample years. 7 7 We address the potential concern of increased coverage by Capital IQ in the sensitivity tests of Section

17 Measuring the Impact of Fair Value Accounting under IFRS To create our treatment variables we begin by constructing two firm-level continuous variables, FV_IND for industrial firms and FV_FIN for financial firms. Following Hung, Li, and Wang (2015), we measure these variables using the extent to which a firm restates key financial statement items in transitioning from local GAAP to IFRS (Deloitte 2009). The magnitude of these restatements (i.e., reconciliation amounts) have the advantage of representing the actual outcome-based measures of the financial reporting changes imposed under IFRS, and reflects permanent changes in financial reporting practices. While Hung et al. (2015) focus on the most frequently restated financial statement items to capture the overall changes in accounting standards, we focus on the financial statement items that reflect accounting changes under IFRS s fair value provisions. We identify these items using Ball et al. (2015), which provide a comprehensive list of IFRS standards that incorporate fair value measures. We create our measures of the effects of IFRS s fair value provisions using the reconciliation amounts reported by our sample firms for each of the following eight financial statement items, with the IFRS standards shown in parentheses: (1) PP&E (IAS 16: Property, Plant, and Equipment; IAS 40: Investment Property), (2) Short-term investment (IAS 39: Financial Instruments), (3) Long-term investment (IAS 39: Financial Instruments), (4) Intangibles (IAS 22: Business Combinations; IAS 38: Intangible Assets), (5) Provisions (IAS 37: Provisions, Contingent Liabilities and Contingent Assets), (6) Post-retirement benefits (IAS 19: Employee Benefits), (7) Stock options (IFRS 2: Share-based Payment), and (8) Discontinued operations (IFRS 5: Non-current Assets Held for Sale and Discontinued Operations). 8 We measure FV_IND 8 The first six financial statement items are balance sheet accounts and the last two are income statement accounts. We include specific accounts, rather than the aggregated equity account, to avoid double counting the impact of fair value provisions. Among the list of IFRS s fair value provisions in Ball et al. (2015, p. 924), we are unable to identify 16

18 as the sum of the absolute value of the difference between these accounts reported under local GAAP and IFRS, scaled by shareholders equity (Hung and Subramanyam 2007). We expect that IFRS s fair value provisions are likely to have a relatively greater impact on firms with larger values of FV_IND. For financial firms, we measure FV_FIN using the same procedure, except that we replace short-term and long-term investments with total investments (including trading securities, securities available for sale, real estate assets, and others). To mitigate measurement errors, we transform FV_IND and FV_FIN into binary indicator variables, High FV_IND (High FV_FIN), which takes a value of one if FV_IND ( FV_FIN) is greater than the sample median, and zero otherwise. Research Design Our baseline analysis assesses the impact of mandatory IFRS adoption on earnings PPS. The baseline model regresses the change in the natural logarithm of annual cash compensation ( COMP) on three performance measures (changes in earnings E, changes in operating cash flows C, and stock returns RET), an indicator variable that captures the post-adoption period (POST), their interactions, and control variables. We include operating cash flows and stock returns in the regression because theory suggests that the precision of earnings in capturing management performance affects the relative weight placed on earnings versus non-earningsbased performance measures in managerial compensation. Thus, including both accounting- and non-accounting-based performance measures allows us to observe whether there is a shift in the relative weights placed on the performance measures (De George et al. 2016). Our baseline analysis estimates the following model: financial statement items reflecting two industry-specific provisions (IAS 41: Agriculture and IFRS 4: Insurance contracts) and one provision that pertains to the implementation of IFRS (IFRS 13: Fair value measurement). 17

19 COMP i,j,t = β 0 + β 1 E j,t + β 2 E j,t POST + β 3 C j,t + β 4 C j,t POST + β 5 RET j,t + β 6 RET j,t POST+ β 7 POST + β 8 BM j,t + β 9 SIZE j,t + β 10 CEO i,j,t + β 11 LnAGE i,j,t + β 12 LnTENURE i,j,t + Country and Industrial Fixed Effects + µ (1) where the dependent variable COMP i,j,t is the change in the natural logarithm of annual cash compensation for executive i of firm j from year t-1 to year t; E j,t is the change in net income before taxes and interest expenses divided by total assets for firm j from year t-1 to t; C j,t is the change in cash flows from operating activities divided by total assets for firm j from year t-1 to t; RET j,t is the market-adjusted annual stock returns for firm j over fiscal year t; POST is an indicator variable that takes the value of one for the post-ifrs period and zero for the pre-ifrs period. We include the following control variables, as in Ozkan et al. (2012): (1) BM j,t, the ratio of the book value of equity to the market value of equity for firm j at the end of year t, (2) SIZE j,t, the natural logarithm of the market value of equity in million U.S. dollars for firm j at the end of year t, (3) CEO i,j,t, an indicator variable set to one if executive i is the chief executive officer of firm j in year t, and zero otherwise, (4) LnAGE i,j,t, the natural logarithm of executive i s age in year t, and (5) LnTENURE i,j,t, the natural logarithm of the number of years executive i serves in the position for firm j in year t. BM, SIZE and CEO control for firm- or executive-level characteristics that may affect executive compensation. LnAGE and LnTENURE are included because changes in compensation may be larger for young executives or executives in the earlier period of their tenure. In addition, we control for country and industry fixed effects. A positive (negative) coefficient on E POST indicates that IFRS adoption increases (decreases) earnings PPS. We test our hypothesis using a DiD design that compares the changes in earnings PPS among the treatment firms with changes in earnings PPS among the benchmark firms, subsequent to the mandatory IFRS adoption. This design helps mitigate the effects of potentially confounding events concurrent with IFRS adoption. Our DiD regression consists of equation (1), with the addition of 18

20 an indicator variable that captures our treatment firms (High FV j ), along with its interaction with our three performance measures ( E, C, and RET) and the post-ifrs indicator (POST), as follows: COMP i,j,t = β 0 + β 1 E j,t + β 2 E j,t POST + β 3 E j,t POST High FV j + β 4 E j,t High FV j + β 5 C j,t + β 6 C j,t POST + β 7 C j,t POST High FV j + β 8 C j,t High FV j + β 9 RET j,t + β 10 RET j,t POST + β 11 RET j,t POST High FV j + β 12 RET j,t High FV j + β 13 POST High FV j + β 14 POST + β 15 High FV j + β 16 BM j,t + β 17 SIZE j,t + β 18 CEO i,j,t + β 19 LnAGE i,j,t + β 20 LnTENURE i,j,t + Country and Industrial Fixed Effects + µ (2) where High FV j equals High FV_IND j for industrial firms and High FV_FIN j for financial firms. A positive (negative) coefficient on E POST High FV indicates that, relative to the benchmark firms, earnings PPS for the treatment firms increases (decreases) subsequent to IFRS adoption. We use robust standard errors clustered by firm to evaluate the significance of regression coefficients. A potential concern with this DiD design is that the benchmark firms may not be comparable to the treatment firms in their use of earnings in executive compensation. While we control for an array of factors that are associated with earnings PPS, it may not adequately address potential misspecification of the functional form of the underlying relation between the effects of fair value accounting and earnings PPS (Armstrong, Jagolinzer, and Larcker 2010). We address this concern by using propensity-score-matching (PSM) to select our benchmark group (DeFond, Erkens, and Zhang 2017; Shipman, Swanquist, and Whited 2017). Specifically, we estimate a logistic model to predict the probability of being classified as a treatment firm (i.e., High FV = 1), using data in year t-1 (the year before the IFRS adoption). By matching firms in year t-1, we also ensure that we have a balanced sample, that is, our sample firms in this test appear both in the pre- and postperiods. We include the following variables in our PSM prediction model: (1) the three 19

21 performance measures ( E, C, and RET); (2) LEV, because the fair value option under IFRS is associated with debt financing (Christensen and Nikolaev 2013); and (3) two additional firm-level control variables, BM and SIZE. We then match treatment firms (High FV = 1) with benchmark firms (High FV = 0), without replacement. 9 We then use these matched firms to re-estimate equation (2). This procedure results in our PSM sample of 3,082 executive-year observations for 238 industrial firms, and 768 executive-year observations from 62 financial firms. Appendix B provides details of the construction of the PSM sample. Panel A reports the results of the logistic regressions used to compute the propensity scores. For industrial (financial) firms, the logistic model has an explanatory power of 6.80% (12.19%) prior to the match, which is reduced to 1.08% (3.95%) after the match. Panel B presents the covariate balance metrics of the PSM samples in the year of matching, year t-1. The mean differences between the treatment firms and the PSM benchmark firms are all insignificant across the firm characteristics used in the logistic model. In addition, the L1 statistics, calculated as the difference between the histograms of the covariates (DeFond et al. 2017; Iacus, King, and Porro 2011), is closer to zero than to one for all the characteristics, consistent with the treatment firms and benchmark firms having similar univariate distributions. In sum, Appendix B indicates that the PSM procedure is successful in achieving balance in the covariates and hence a high quality match. 9 The sample size used in the PSM prediction model is larger than the number of sample firms in year t-1 reported in Table 1 because we do not require the sample for the PSM prediction model to have compensation data. We use a larger sample in the PSM prediction model in order to increase the precision of the estimated coefficients. Following prior studies (Rosenbaum and Rubin 1984; Austin 2011), we start with a caliper width equal to 30% of the standard deviation of the propensity score (yielding a caliper width of approximately 0.05). We then narrow the width until we find a high quality match between the treatment and benchmark firms, which we define as (1) insignificant differences between the mean and median covariates, and (2) comparable levels of pre-period earnings PPS, as indicated by an insignificant coefficient on E High FV. This procedure generates caliper widths of for the industrial sample and for the financial sample. 20

22 Descriptive Statistics Table 2 presents descriptive statistics for the variables used in our primary analyses. As shown in Panel A, the average annual cash-based executive compensation (COMP) is thousand U.S. dollars, with a median value of thousand U.S. dollars, which indicates that executive compensation is left-skewed. The mean and median of the change in the logarithm of cash compensation ( COMP) are and 0.099, respectively, indicating that the average (median) ratio of cash compensation in year t over that in year t-1 is 1.19 (1.10). 10 The performance variables are changes in earnings before taxes and interest expenses ( E), changes in cash flows from operations ( C) and market-adjusted stock returns (RET), which have a mean (median) of (0.014), (0.008), (0.000), respectively. Our sample has an average book-to-market ratio (BM) of and an average natural logarithm of market capitalization (SIZE) of CEOs (CEO) constitute 33 percent of our sample executive-years. The age for our average sample executive (AGE) is 54 years and the average tenure (TENURE) is five years. Overall, the descriptive statistics for these variables are similar to those reported in prior studies (e.g., Ozkan et al. 2012). Our firm-level fair value accounting measure for industrial firms ( FV_IND) has a mean (median) of (0.149), and for financial firms ( FV_FIN) has a mean (median) of (0.094). Panel B of Table 2 reports the correlation matrix for the variables used in the main regressions. The upper (lower) diagonal of the panel presents the Pearson (Spearman) correlation coefficients. Consistent with prior studies, the change in cash compensation ( COMP) is positively correlated with the three performance measures ( E, C, and RET), firm size (SIZE) and the CEO indicator 10 COMP = LnCOMP t - LnCOMP t-1 = Ln (COMP t /COMP t-1 ). Thus, the mean and median values of COMP t /COMP t-1 are 1.19 (Exp(0.176)) and 1.10 (Exp(0.099)), respectively. 21

23 (CEO), and negatively correlated with book-to-market (BM), and the executive s age (LnAGE) and tenure (LnTENURE). IV. EMPIRICAL RESULTS Baseline Results: the Effect of IFRS Adoption on Earnings PPS Table 3 presents the baseline regression results estimating equation (1). In Panel A, we test whether earnings PPS changes after mandatory IFRS adoption, without considering the potential effects of the fair value provisions. Column (1) shows the baseline regression without control variables. We find that the coefficients on the three performance measures, E, C, and RET, are all positive and significant at the 1% level, suggesting that executive compensation is linked to firm performance for our sample firms prior to IFRS adoption. More importantly, we find a positive and significant coefficient on the interaction between earnings performance and the post- IFRS indicator, E POST (at the 1% level). This finding is consistent with Ozkan et al. (2012) and indicates that, on average, earnings PPS increases after mandatory IFRS adoption. We also find that the coefficient on the interaction between cash flow performance and the post-ifrs indicator, C POST, is negative and significant at the 5% level. This indicates that cash flows become relatively less important in determining executive compensation after IFRS adoption. In addition, the coefficient on the interaction of stock returns and the post-ifrs indicator, RET POST, is insignificant at conventional levels, suggesting that IFRS adoption has no effect on the usefulness of market-based performance measures in executive compensation. Column (2) reports the baseline regression after adding control variables, and finds similar results on the three performance measures ( E, C, and RET) and their interactions with POST as in Column (1). In addition, we find a positive coefficient on SIZE and negative coefficients on LnAGE and 22

24 LnTENURE, consistent with executives in larger companies experiencing greater increases in cash compensation, while older executives and executives later in their careers tend to experience smaller increases. In Panel B of Table 3, we re-estimate equation (1) separately for industrial and financial firms. Similar to Panel A for the combined sample, the coefficients on E POST and C POST for industrial firms are positive and negative, respectively, although both become only weakly significant at the 10% level. For financial firms, on the other hand, the coefficients on E POST and C POST are insignificant. This suggests that the effects of IFRS adoption on earnings and cash flow PPS are not uniform across industrial and financial firms. The coefficient on RET POST is insignificant for both industrial and financial firms. Overall, the results in Table 3 are consistent with corporate boards, on average, relying more on earnings and less on cash flows after IFRS adoption when evaluating executive performance, but only for industrial firms. The Effect of Fair Value Accounting on Earnings PPS Summary of the Effects of IFRS s Fair Value Provisions In Panel A of Table 4, we provide descriptive statistics for the absolute values of the reconciled amounts (the difference between original and restated values, scaled by shareholders equity), for the financial statement items related to fair value accounting under IFRS ( FV_IND) for industrial firms. For each financial statement item, the panel reports the number of observations with nonzero values, the corresponding mean and median of the absolute values, and the related IFRS standards. We find that the mean of the absolute values of the reconciled amounts scaled by total equity is 0.085, 0.089, 0.064, 0.066, 0.123, 0.164, 0.004, and for PP&E, short-term 23