The Credit-Risk Relevance of Loan Impairments Under IFRS 9 for CDS Pricing: Early Evidence

Abstract

Since 2018, banks have implemented the expected credit loss (ECL) model under International Financial Reporting Standard (IFRS) 9 to estimate loan losses, which replaces the incurred loss model under International Accounting Standard (IAS) 39. The key novelty of the ECL model is the incorporation of forward-looking information for recognizing accounting loan loss provisions (LLPs), which provides ample room for managerial discretion. Over the period 2014–2019, I first show that the shift to the ECL model improves the timeliness of loan loss recognition. However, under the IFRS 9 regime managers also use their accounting discretion more aggressively over LLP estimates to smooth earnings. I then investigate whether IFRS 9 improves the relevance of LLPs for credit default swap (CDS) pricing. I report that LLPs under IFRS 9 are incrementally more relevant than under IAS 39 for CDS pricing but mostly concentrated amongst banks with weaker pre-IFRS 9 information environments. I further show that under the IFRS 9 regime, LLPs are relevant for CDS pricing only when LLPs consistently reflect future expected losses while earnings smoothing via LLP generally impair the credit-risk relevance of LLPs. Finally, I find that strong governance is imperative for providing useful LLP estimates for CDS pricing.

Keywords:

• Credit Default Swaps
• Credit Risk
• IFRS 9
• Banks

JEL Codes:

• G21
• G28
• M41
• M48

Acknowledgements

I gratefully acknowledge the helpful comments received from Beatriz García Osma (editor) and two anonymous referees, whose insights have markedly improved the paper. Feedback from Zahn Bozanic (discussant) as well as the participants at the European Accounting Review 2020 Annual Conference was also very helpful in revising the paper. I also appreciate the suggestions I received from Elizabeth Demers, Minyue Dong, and Alain Schatt, as well as workshop participants at Laval University. I also would like to thank my family (especially Danièle, Juliette, Héloïse and François) as well as my friends for their support. Finally, I would like also to thank The Pixies, Christian Löffler, Cercle and Boom Festival (this list is not exhaustive) for remembering me that life is more than just ‘debit and credit’. This project was financially supported by a research grant from the University Laval.

Supplemental Data and Research Materials

Additional materials are available in an online Supplement at the journal’s Taylor and Francis website. All errors are my own.

• Appendix OA: Sample selection and distribution

• Appendix OB: Detailed descriptions of variables used in this study.

• Appendix OC: Figures

• Appendix OD: Additional robustness checks.

Notes

1 The literature often refers to loan impairments (i.e., loss recognized in the income statement) as LLPs, which is the term used in banking regulation but not in accounting standards. For comparability purposes I follow the terminology used in previous studies and refer to these impairments as LLPs.

2 The U.S. Financial Accounting Standards Board (FASB) and the IASB have worked jointly towards the development of an ECL model to measure loan loss allowances (LLAs) (i.e., the accumulated LLP at the end of the period). However, they have failed to develop a converged standard. In 2016, the FASB published the ASC topic 326, which includes the ‘current expected credit losses’ (CECL) model that takes effect in 2020 for listed companies and in 2021 for all other firms (see also Giner and Mora (2019) and Hashim et al. (2016)).

3 IFRS 9, paragraph B5.5.17, provides a list of forward-looking indicators that may be relevant in assessing changes in credit risk. This list includes several external market indicators of credit risk such as changes in the price of a borrower's debt and equity instruments, credit spread, and credit default swap prices for the borrower.

4 The CDS is an over-the-counter credit derivative contract between two parties, used to hedge the credit risk that a specific borrower represents for fixed-income investors. A CDS is quoted in basis points.

5 Using the economic value of a loan as a benchmark, Novotny-Farkas (2016) shows that at day-one, IFRS 9 overstates the LLA for Stage 1 financial assets. However, as credit risk increases, the LLA is understated. For Stage 2 and 3 financial assets, IFRS 9 again overstates the LLA. The overstatement (as opposed to the understatement) of LLA is likely to be preferred by creditors (Gebhardt, 2016). For a detailed discussion of how ECL information is reflected in LLA under IFRS 9 and how it differs from IAS 39, see Gebhardt (2016) and Novotny-Farkas (2016).

6 In an additional analysis, López-Espinosa et al. (2021) report that the ‘day-one impact’ of IFRS 9 is only reflected in the CDS market in countries experiencing a more negative change in expected credit conditions. This paper adds to this by focusing on LLPs reported under IFRS 9 beyond day-one as well as by investigating several channels that influence the usefulness of accounting information for CDS market participants.

7 Other metrics are affected by IFRS 9, such as the regulatory capital ratio. I do not discuss this latter metric further because (1) it is unclear whether the regulatory capital ratio is useful for CDS market participants (Kanagaretnam et al., 2016), (2) evidence of capital management through LLPs is mixed (e.g., Ahmed et al., 1999; Leventis et al., 2011), and (3) banks can neutralize the effect of the ECL model on the regulatory capital ratio through the Capital Transitional Arrangement (Basel Committee on Banking Supervision, 2017; Dong & Oberson, 2021).

8 IFRS 9 may also affect the comparability of earnings because of the judgmental nature of the IFRS 9 ECL model (Gebhardt, 2016). Moreover, IFRS 9 influences earnings conservatism (conditional and unconditional), which is also likely to influence the information content of earnings. For a detailed discussion see Giner and Mora (2019).

9 CDS contracts are written on standardized horizons, the most common being 5 years, followed by 1, 3, 7, and 10 years (e.g., Shivakumar et al., 2011).

10 Several studies examine the credit-risk relevance of accounting information investigating changes in explanatory power of an accounting-based credit-risk model following an accounting reform (e.g., Florou et al., 2017; Kraft et al., 2021). The objective of this paper is to examine changes in the credit-risk relevance of the largest accrual for most banks (i.e., LLPs) as a result of IFRS 9 adoption. However, I do not attempt to investigate general changes in the credit-risk relevance of accounting information following IFRS 9 adoption as IFRS 9 does not only reform the LLP measurement. Consequently, I investigate the impact of IFRS 9 on the credit-risk relevance of LLPs on CDS spreads using a level regression analysis in the spirit of Callen et al. (2009) who also focus on the relevance of one accounting metric (i.e., earnings) for CDS pricing.

11 If no CDS spread is available on the 45th day after the fiscal quarter-end, I use the first CDS spread available on the 44th, 46th, 43rd, and 47th days (in that order) after the quarter-end.

12 Most banks applied IFRS 9 for the first time on January 1, 2018, with the exception of Canadian banks, which first applied IFRS 9 on November 1, 2017, as well as banks with non-December fiscal year-end (Mediobanca on July 1, 2018 and Nationwide Building Society on April 5, 2018).

13 In a few cases where the S&P ratings are missing, I use Moody’s ratings as in Liu and Jiraporn (2010).

14 Doshi et al. (2017) report that local risk factors have more explanatory power for short maturity than global factors do. I therefore use short-term sovereign CDS spreads to control for a country-specific economic condition, while the fixed-effect structure will capture the effect of global factors on bank CDS spreads. The time fixed effects specifically control for the risk-free rate (Acharya et al., 2014), which is a determinant of CDS spreads implied by structural models.

15 The $WEAKIE$ measures are averaged over 2014–2016 to avoid capturing changes in the information environment due to earlier information release with respect to IFRS 9 adoption (see also footnote 24). I use yearly LLP forecasts because quarterly LLP forecasts are much less frequent.

16 Mayordomo et al. (2014) report that the CMA-quoted CDS spreads lead the credit-risk price discovery process with respect to the other databases.

17 CMA reports bid, ask, and mid prices, as well as tenors, seniorities, and restructuring types. The definition of debt restructuring differs slightly around the world; however, they do not vary over time. Consequently, the fixed-effects structure controls for such heterogeneity in CDS contract features. To further maintain contractual uniformity, I only kept spreads for CDS referencing senior debt.

18 Restrictions (1) and (3) are sufficient to constrain a bank to have at least one observation in the pre- and post-IFRS 9 adoption periods.

19 Sovereign CDS quotes for Singapore are only available from 2018. The exclusion of Singapore is also justified as the reporting framework is only identical to IFRS for annual periods beginning on or after 1 January 2018.

20 These are the National Australia Bank as well as the Bank of Bahrain and Kuwait that adopted IFRS 9 early and Hitachi Capital Corporation that adopted IFRS in 2015 after the publication of IFRS 9 and, thus, can be considered a voluntary adopter.

21 Panel A of Table OA1 in the Online Appendix summarizes the sample-selection process. Panel B of Table OA1 in the Online Appendix provides a breakdown of the sample composition by country.

22 Untabulated statistics indicate that the largest variance inflation factor (VIF) among the independent variables is less than 3.1. The mean VIF equals 2.1. Consequently, multicollinearity is likely not a concern.

23 I also investigate the capital management hypothesis by including the beginning-of-quarter regulatory capital ratio and its interaction with the variable $POST$ in Equation (4) in place of the leverage ratio; none of the coefficients loads significantly (unreported).

24 Ample evidence exists to suggest that market participants had early access to information relative to the implementation of IFRS 9. First, banks provided market participants with early information about the IFRS 9 effect. For instance, in its 2016 annual report, Deutsche Bank expects ‘an overall reduction in the Group’s total shareholders’ equity of approximately €1 billion before tax. This reduction is predominately driven by the impairment requirements of IFRS 9.’ Accountancy Europe (2018) reveals that across 62 large European banks, 19% of audit reports had a key audit matter related to ‘disclosure of IFRS 9 impact’, suggesting that for more than 80% of banks, auditors believed their financial statements had provided relevant information about the impact of IFRS 9.

25 Callen et al. (2009) provide evidence that earnings are not priced linearly into CDS spreads. Therefore, I use two alternative definitions of earnings before LLP ($ROA$): (1) a quarterly dummy variable for loss firms, and (2) the log transformation of the variable $ROA$; inferences remain qualitatively similar.

26 IFRS 9 paragraph B5.5.50 recognizes that information relevance decreases as the forecast horizon increases: ‘An entity is not required to incorporate forecasts of future conditions over the entire expected life of a financial instrument. […] As the forecast horizon increases, the availability of detailed information decreases and the degree of judgement required to estimate ECLs increases’.

27 I also include the interaction between the variables $ROA$ and $POST$. Inferences regarding changes in the credit-risk relevance of earnings are mixed (unreported). Inferences regarding the forward-looking properties of LLP remain qualitatively unchanged. The major difference is that the coefficient on $LLP\ast POST$ loads significantly for the pricing of 1-year CDS contracts.

28 While the total effect of the LLP accounting metric is never statistically significant under IAS 39 (unreported), Figure OC3 and Figure OC4 in the Online Appendix confirmed an increased relevance of LLPs for CDS pricing only for banks that were operating in a particularly weak information environment.

29 Inferences relative to the credit-risk relevance of LLPs are qualitatively similar if I exclude the interaction terms involving $ROA$, $POST$, and $WEAKIE$ as implied by Equation (3).

30 The endogeneity introduced by the lagged dependent variable (Nickell, 1981) is unlikely to affect my inferences (see Table OD7 in the Online Appendix).

31 That is because IFRS 9 adoption was not staggered (not taking into account the few banks that have non-December fiscal year-end).

32 Inferences are robust to using $EPS\mathrm{\_}ERROR\mathrm{\_}PRE$ or $EPS\mathrm{\_}DISP\mathrm{\_}PRE$ instead of $LLP\mathrm{\_}ERROR\mathrm{\_}PRE$ and $LLP\mathrm{\_}DISP\mathrm{\_}PRE$(unreported).

33 An unreported Chow test reveals that the difference in the magnitude of ${\text{β}}_1$ is statistically significant at the 1% level across Columns 2 and 3.

34 Figure OC5 and Figure OC6 in the Online Appendix present the change in CDS volatility following IFRS 9 adoption conditional on pre-IFRS 9 bank information environment quality based on the estimated coefficients displayed in Columns 4 and 5 of Panel B. The decrease in CDS volatility is larger for banks that had weaker pre-IFRS 9 information environments but only statistically negative for firms that were operating in a relatively weak pre-IFRS 9 information environment.

35 To save space in Section 4.3, I only present results excluding bank-quarter observations in the year prior to mandatory IFRS 9 adoption. Inferences are qualitatively similar if I include those observations.

36 Figure OC7 (Figure OC8) in the Online Appendix plots the total effect of $LLP$ on $LogCDS5Y$ conditional on the level of abnormal LLPs under the IFRS 9 (IAS 39) regime. As shown in Figure OC7, the association between the LLP accounting metric and CDS spreads decreases monotonically with the level of abnormal LLPs but remains statistically positive within plausible ranges of $ABN\mathrm{\_}LLP$ under IFRS 9.

37 An alternative explanation could be that, under IFRS 9, an increase in LLPs for banks with large abnormal LLPs results in a lower adjustment of CDS spreads because these banks are perceived to be relatively safer rather than because the LLP accounting metric being perceived as less credible. This could be possible if large abnormal LLPs consistently capture an overestimation of the LLP accounting metric by banks under IFRS 9. Indeed, creditors prefer LLP overstatements as they may diminish bank failure (Costello et al., 2019; Donovan et al., 2015; Gebhardt, 2016). Consequently, I replace $ABN\mathrm{\_}LLP$ with a dummy variable $\left(PosResid\right)$ that takes the value 1 for banks with strictly positive post-IFRS 9 median residuals (i.e., $ϵ$). Banks assigned a value of 1 for $PosResid$ should have a tendency to overestimate LLPs under IFRS 9 compared to other banks. The three-way interaction term ($LLP\ast POST\ast PosResid$) is positive and statistically significant at the 1% level while the total effect of LLPs on CDS spreads is positive but statistically insignificant (significant) for banks that underestimate (overestimate) LLPs under the IFRS 9 regime (unreported). Overall, these results are hard to reconcile with the fact that banks with large abnormal LLPs could be perceived as safer, which could explain the decreasing magnitude of LLPs displayed in Figure OC7 in the Online Appendix.

38 To be consistent, accounting flow variables ($LLP$ and $ROA$) are measured on a yearly basis.

39 This primary result is in line with Donoher et al. (2007), who show that board tenure is positively associated with financial reporting quality, and Arnaboldi et al. (2020), who report that board tenure has a positive impact on European bank performance, suggesting that tenure of board members affects their level of firm-specific knowledge and expertise, which ultimately allows for better monitoring of bank managers.

40 A tenure score between 50 and 59 represents an average board tenure of 6 years. A score between 70 and 79 represents an average board tenure of 10 years. The results are qualitatively similar if I use the ‘raw’ average board tenure value rather than the score. Huang and Hilary (2018) focus on the tenure of nonexecutive directors while I use the readily available average board tenure score provided by Thomson Reuters, which is based on all directors on the board. Nevertheless, boards are composed on average of 82% of nonexecutive directors (in-sample) which mitigates concerns that the results are unrelated to corporate governance mechanisms highlighted in Huang and Hilary (2018). Moreover, Arnaboldi et al. (2020) and Donoher et al. (2007) also focus on the tenure of all directors (see footnote 39).

41 The $POST$ variable, as well as the time and bank fixed effects, are excluded; see the Online Appendix OB.

42 Earnings smoothing can also mitigate pro-cyclicality of the financial system by allowing banks to provide more (less) for expected losses when earnings are high (low) (Bikker & Metzemakers, 2005; Laeven & Majnoni, 2003), which could explain a negative association between LLPs and CDS spreads.

43 The assessment of a significant increase in credit risk is based on forward-looking information (if obtaining it does not involve undue cost or effort) and reflects changes in risk of defaults that is key information for CDS market participants rather than changes in ECLs (IFRS 9, paragraph 5.5.9).

44 See IFRS 9, paragraph B5.5.22–B5.5.24.

Additional information

Funding

This work was supported by Université Laval: [Grant Number DC124452].