Effect of Mandatory IFRS Adoption on Accounting-Based Prediction Models for CDS Spreads

Abstract

In this study, we examine the effects of mandatory IFRS adoption on accounting-based prediction models of CDS spreads for a sample of 292 firms in 16 countries. In our examination, we estimate the models for both financial and nonfinancial firms before and after mandatory IFRS adoption. We find that mean and median absolute percentage prediction errors are larger for both financial and non-financial firms after mandatory IFRS adoption. We also estimate accounting-based prediction models of CDS spreads separately for financial and non-financial US firms as a benchmark. Although US firms also show an increase in the mean and median absolute percentages of prediction errors over the same period, our findings from regressions that use a difference-in-difference design indicate that the increase is significantly greater for firms in countries that adopted IFRS mandatorily. We also find that in the post-adoption period, prediction errors are larger for firms in countries with weaker institutions such as low levels of property rights and more restrictive access to credit.

Keywords:

• International financial reporting standards
• Credit markets

Acknowledgments

We thank Minkwan Ahn, Gary Biddle, Mary Billings, Francois Brochet, Vincent Chen, Qiang Cheng, Xiumin Martin, Michel Magnan, Tiago Pinheiro, Darren Roulstone, Stephen Ryan, Dan Taylor, Holly Yang, Paul Zarowin, and two anonymous reviewers, as well as seminar participants at NYU, MIT, Ohio State, University of North Carolina, Chinese University of Hong Kong, Hong Kong University, Hong Kong University of Science and Technology, European Accounting Association Annual Congress, Norwegian School of Economics, National University of Singapore, and Singapore Management University for helpful comments. We are grateful for research assistance by Donny Zhao, Seil Kim, Maria Nikiforovich, and Travis Dyer. We appreciate funding from the Kenan Institute of Private Enterprise, Kenan-Flagler Business School, Labex Ecodec (ANR-11-LABX-0047), and HEC Paris Foundation.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 In the US, Rauh and Sufi (2010) provides evidence that, on average, publicly traded debt accounts for approximately 10% of all corporate debt.

2 Consistent with this, Easton et al. (2009) provide evidence that accounting information on earnings is important to investors in publicly traded bonds.

3 As another illustration, while treasury shares are often presented as marketable securities under many countries' domestic standards, IFRS require that they be deducted from shareholders' equity.

4 For example, until 2008 IFRS permitted the choice between capitalizing the interest that was incurred during construction and expensing it. This choice affected the reported interest expense that was used in interest coverage ratios. Furthermore, IFRS require that received and paid interest, received dividends, and paid income taxes be reported under different headings in the cash flow statement. However, users of financial statements can easily adjust their calculations of the ratios that involve cash flows to reflect these differences in firms' choices.

5 Not all studies conclude that mandatory adoption of IFRS reduced the usefulness of accounting information to debt contracts. For example, Florou and Kosi (2015) find that firms in countries that mandate IFRS are more likely to issue public rather than private debt, and that public debt issues have lower yields than before IFRS adoption. Taken at face value, these findings indicate that IFRS adoption and improvements in accounting quality made accessing the public debt markets less costly.

6 By design, our empirical measure of credit risk predictability is based on four financial statement-based ratios. In principle, we could extend the models to include additional ratios, although doing so generally results in a significant loss in sample observations with a resulting loss in power of our tests. Our accounting-based measure of credit risk predictability also ignores the relevant information on credit risk that is disclosed in the footnotes of financial statements. This omission potentially affects the inferences we draw because IFRS mandate more comprehensive disclosures of debt-like obligations than are mandated by most countries' domestic standards. To the extent that such information is relevant to debt holders in their assessments of credit worthiness and orthogonal to improvements in recognized measures, our measure of credit risk predictability will fail to reflect such information.

7 In robustness tests, we use different combinations of accounting variables, such as return on assets instead of return on equity, total assets instead of equity book value, and including earnings volatility and the ratio of cash flow from operations to debt, as additional explanatory variables. Although including additional variables causes significant reductions in sample size, untabulated findings from tests based on this version of Equation (1(1) $\begin{array}{r}Ln(CDS{)}_{i,t}={\alpha }_0+{\alpha }_{1}Ln(BV{)}_{i,t}+{\alpha }_{2}LE{V}_{i,t}+{\alpha }_{3}RO{E}_{i,t}+{\alpha }_{4}Interestcoverag{e}_{i,t}\end{array}$(1) ) result in no change in inferences from those based on the tabulated findings. In addition, we disaggregate earnings into accruals and cash flows and find no change in inferences.

8 We also calculated prediction errors based on a jack-knifing procedure in which, for each firm, we estimate a version of Equation (1(1) $\begin{array}{r}Ln(CDS{)}_{i,t}={\alpha }_0+{\alpha }_{1}Ln(BV{)}_{i,t}+{\alpha }_{2}LE{V}_{i,t}+{\alpha }_{3}RO{E}_{i,t}+{\alpha }_{4}Interestcoverag{e}_{i,t}\end{array}$(1) ) that excludes that firm. Not surprisingly, the untabulated prediction errors are virtually identical to those tabulated in the text, and inferences based on them are identical to those based on the tabulated findings.

9 A maintained assumption necessary to draw such inferences is that there is no change in the underlying economic relation between CDS spreads and the economic constructs captured by the accounting measures before and after 2005. For example, we assume that the relation between a firm's economic leverage and its credit risk remains constant during the sample period.

10 In untabulated tests, we find that the 5-year CDS spreads of all four document clauses are very highly correlated. In robustness tests, we use the CDS spreads of MM and MR contracts as dependent variable and find very similar results.

11 Standard errors are robust and clustered by firm.

12 To avoid the influence of extreme prediction errors, we also conduct analyses winsorizing GAP at the top 1%. Untabulated findings indicate that the inferences based on these statistics are the same as those based on the tabulated findings.

13 We also estimated Equation (1(1) $\begin{array}{r}Ln(CDS{)}_{i,t}={\alpha }_0+{\alpha }_{1}Ln(BV{)}_{i,t}+{\alpha }_{2}LE{V}_{i,t}+{\alpha }_{3}RO{E}_{i,t}+{\alpha }_{4}Interestcoverag{e}_{i,t}\end{array}$(1) ) for the post-adoption period by defining the financial crisis as 2008 and 2009, and 2007 through 2009. The inferences based on GAP statistics using these alternative definitions are the same as those using the 2008 definition.

14 Standard errors are robust and clustered by firm.

15 Untabulated findings based on estimations of Equation (2(2) $\begin{array}{r}GA{P}_{i,t}={\text{β}}_0+{\text{β}}_{1}IFR{S}_{i,t}+{\delta }_{0}y{2005}_{i,t}+{\delta }_{1}IFR{S}_{i,t}\ast y{2005}_{i,t}\end{array}$(2) ) using winsorized GAP measures result in the same inferences as those based on the tabulated findings.

16 We also estimate Equation (2(2) $\begin{array}{r}GA{P}_{i,t}={\text{β}}_0+{\text{β}}_{1}IFR{S}_{i,t}+{\delta }_{0}y{2005}_{i,t}+{\delta }_{1}IFR{S}_{i,t}\ast y{2005}_{i,t}\end{array}$(2) ) using only Post-I and Post-II sample years. Untabulated findings show that the mean absolute prediction error for IFRS firms increases by a significantly greater amount than that for the US firms.

17 We also estimate versions of Equation (3(3) $\begin{array}{rl}GA{P}_{i,t}& ={\gamma }_0+{\gamma }_{1}y{2005}_{i,t}+{\gamma }_{2}PC\mathrm{\_}GD{P}_{i,t}+{\gamma }_{3}y{2005}_{i,t}PC\mathrm{\_}GDPi,t\\ & +{\gamma }_{4}INSTITUTION{S}_{i,t}+{\gamma }_{5}y{2005}_{i,t}INSTITUTION{S}_{i,t}\end{array}$(3) ) using other alternative measures of country-level institutional variables including strength of legal rights, the ratio of private international bonds to GDP, and investor perceptions of the rule of law. Untabulated results show none of these variables is significantly related to changes in prediction errors for IFRS firms.

18 Untabulated findings that relate to estimations that exclude observations relating to the financial crisis years yield the same inferences as those based on the full sample.

19 Untabulated robustness tests give similar inferences when we use the four accounting variables in our main specification, and when we include earnings volatility as an additional explanatory variable.

20 Because the number of IFRS and US firms differ, random assignments are based on the relative proportion of IFRS and US firms in the pre- and post-IFRS adoption periods.

21 We also estimate Equation (5(5) $\begin{array}{r}RESIDUA{L}_{i,t}={\alpha }_0+{\alpha }_{1}Ln(SIZE{)}_{i,t}+{\alpha }_{2}LE{V}_{i,t}+{\alpha }_{3}RO{A}_{i,t}\end{array}$(5) ) excluding the financial crisis. Untabulated findings show the same inferences as those based on tabulated findings.

22 In particular, the coefficients on the absolute value of abnormal discretionary costs and the composite measure combining signed abnormal discretionary operating cash flow, abnormal production cost, and abnormal discretionary costs load in three-way interactions. The coefficients on the other eight proxies' three-way interactions are not significant. The results are the same if we exclude the financial crisis. We do not examine earnings management for financial firms because the traditional measures do not apply to them.

23 We also test for the possibility that firms that experience greater changes in accrual-based earnings management, real earnings-based management, and earnings volatility have greater prediction errors following IFRS adoption. Untabulated findings show no evidence of an association between changes in earnings management and volatility and prediction errors.