Moving toward the expected credit loss model under IFRS 9: capital transitional arrangement and bank systematic risk

Abstract

This paper examines banks’ option to adopt the capital transitional arrangement (CTA) set out by the Basel Committee on Banking Supervision, in response to the introduction of the International Financial Reporting Standard 9 (IFRS 9), which requires the use of an expected credit loss model instead of an incurred loss model to estimate the impairment of financial assets. Using a sample of publicly listed European banks from 2016 to 2019, we find that bank CTA adoption choice is associated with neutral factors captured by bank-specific fundamental characteristics, and potential opportunistic factors related to regulatory constraints implied by the application of IFRS 9. We further find that banks that adopted the CTA (CTA adopters) decrease their exposure to systematic risk during the transitional period. However, this relationship is only significant in countries with powerful banking authorities. In those with less powerful banking authorities, CTA adopters tend to exercise more aggressively their accounting discretion. Our study is the first to address banks’ voluntary choice to adopt the CTA policy under the mandatory application of IFRS 9.

Keywords:

• systematic risk
• regulatory capital
• IFRS 9
• expected credit loss
• banks

JEL Classifications:

• G21
• G28
• M41
• M48

Acknowledgements

We thank the IASB/ABR Research Forum 2020 participants, particularly Tadeu Cendon and Araceli Mora (the discussants) as well as Mark Clatworthy, Eric Jondeau and an anonymous referee for helpful comments and suggestions. We also appreciate the research assistance of Jiawei Lu and Cédrine Fuchs. This project was financially supported by a research grant from the University Laval. All errors are our own.

Disclosure statement

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

Supplemental data

Supplemental data for this article can be accessed at https://doi.org/10.1080/00014788.2021.1952060.

Notes

1 As defined in the International Accounting Standards Board's Conceptual Framework for Financial Reporting, the general purpose of financial reporting is to ‘provide financial information about the reporting entity that is useful to existing and potential investors, lenders and other creditors in making decisions relating to providing resources to the entity’ (IASB 2018, par. 1.2). In contrast, the primary mission of banking regulators is to protect the financial system as a whole by avoiding bank failure and limiting the frequency and cost of systemic crises (e.g. Barth and Landsman 2010, Acharya et al. 2017).

2 In a speech on June 4, 2012, IASB Chairman Hans Hoogervorst stated ‘Stability should be a consequence of greater transparency, but stability cannot be a primary goal of accounting standard-setters. It is not our remit and we simply lack the tools for fostering stability. […] What accounting standard setters can also not do is to develop standards that make items appear to be stable when they are not. And, quite frankly, we are sometimes suspicious that we are being asked to put a veneer of stability on instruments that are inherently volatile in value. Our standards should not create volatility that is not already there economically. But, if volatility exists, our standards should certainly not mask it.’ (Hoogervorst 2012) available at https://www.ifrs.org/.

3 The IASB and the Financial Accounting Standards Board (FASB) worked jointly on a converged standard on credit loss impairment, but ultimately failed to develop a single ECL model. In 2016, the FASB published the Accounting Standard Update (ASC) topic 326, which describes the new impairment model: the current expected credit loss (CECL) model. The FASB’s CECL model standard takes effect in 2020 for listed companies and in 2021 for all other firms. For detailed discussions of the development of ECL models and differences between the IFRS 9 ECL model and the FASB CECL model, see Giner and Mora (2019) and Hashim et al. (2016).

4 Loipersberger (2018) provides evidence consistent with market participants viewing the ECB as holding significant power over banking activities through the SSM.

5 Gebhardt and Novotny-Farkas (2011) report that the adoption of IAS 39 by European banks resulted in delayed recognition of loan losses. O’Hanlon (2013) provides evidence suggesting that stricter requirements for the recognition of loan losses for UK banks implied by the adoption of IAS 39 did not result in less timely loan-loss provisioning. These contrasting results imply that multiple stakeholders (e.g. managers, enforcers and regulators) are key to determine the timeliness of loan loss recognition beyond the standards per se (Mora and Walker 2015). To judge whether the IFRS 9 ECL model will effectively solve the issues related to the IL model is unclear. As the IFRS 9 ECL model requires the immediate establishment of loss allowances at day 1 (unconditional conservatism) that may lead to earnings management, Hashim et al. (2019) note that this approach is ‘consistent with the way in which bank regulators require expected losses on exposures to be reflected for the purpose of determining banks’ capital requirements, [but] this approach is not easily justified for the purpose of measuring credit-loss expense and loss allowances in financial statements’ (p. 713). Building on these arguments, they conclude that ‘it is unlikely that any ex-ante acceptable method of accounting for credit-loss impairment would have substantially mitigated the consequences of a shock of the magnitude that occurred in the crisis.’ (p. 715).

6 IFRS 9 also specifies a ‘simplified approach’ for trade receivables, contract assets recognised under IFRS 15 and lease receivables under IAS 17 (or IFRS 16). When adopting the simplified approach, the entity does not need to calculate a 12-month ECL nor to identify a significant increase in credit risk, but instead should recognize a loss allowance based on lifetime ECLs at each reporting date from origination.

7 Overall, the IFRS 9 impairment approach differs substantially from that under IAS 39. Only credit-impaired loans (Stage 3) are not modified, since this category of exposures also requires the estimation of lifetime expected losses under IAS 39. Consequently, the estimation of ECLs for Stage 1 and Stage 2 financial assets should result in greater accounting loan loss provisions (BCBS 2017).

8 In 1996, the BCBS extended the RWA requirement from credit risk to market risk (BCBS 1996), but the minimum capital adequacy ratio (i.e. regulatory capital over RWAs) remained unchanged at 8%. In 2004, aiming to improve the risk sensitivity of capital requirements, the Basel II accords extended the risk-weight categories from credit risk and market risk to operational risks, and changed the rules for assigning risk weights to assets.

9 Under Basel I, internal models were already available to banks for estimating market risk.

10 IRB yields two methods: the foundation internal ratings-based (F-IRB) and the advanced internal ratings-based (A-IRB) methods. Under the F-IRB method, banks are allowed to define only one parameter – probability of default – while under the A-IRB method, banks can use their own methodologies to estimate all three main parameters.

11 See the guidelines on the implementation, validation and assessment of Advanced Measurement (AMA) and Internal Ratings Based (IRB) Approaches (GL10) − https://eba.europa.eu/.

12 In its 2018 annual report on supervisory activities, available at www.bankingsupervision.europa.eu, the ECB notes, ‘TRIM is the largest project that ECB Banking Supervision has launched so far’ (p. 38).

13 The effect of accounting loan loss provisions on regulatory capital is conditional on whether the loan loss allowance is lower or higher than 1.25% of the risk-weighted asset since the introduction of the Basel framework. The inclusion of general provisions in Tier 2 capital is limited to 1.25% of credit RWAs. The regulatory capital treatment of provisions under the SA and IRB approaches differs slightly since the adoption of Basel II. For more information please refer to BCBS (2017).

14 It corresponds to the difference between accounting provisions under IAS 39, and prudential expected losses for portfolios under the IRB approach. If prudential expected losses under the IRB approach are higher than accounting provisions under IAS 39, the shortfall will absorb (totally or partially) the impact on CET1 of the increase in accounting provisions when IFRS 9 is first applied (which would not be the case for portfolios under the SA approach).

15 See footnote 14. Please also refer to Novotny-Farkas (2016) for a detailed discussion over the regulatory capital treatment of IFRS 9 impairments of IRB banks.

16 See details on https://www.bankingsupervision.europa.eu/, ‘Status update on TRIM: overview of outcome of general topics review and interim update on preliminary results of credit risk on-site investigations’.

17 The literature often refers to loan impairment as loan loss provision, which is the term used in the prudential bank regulation but not in accounting standards. In this study, we refer to loan impairment as the loss recognised in the income statement in the reporting period, and LLA is the accumulated impairment that appears in the balance sheet at the end of the reporting period.

18 More precisely, we focus on operating banks as of 31.12.2018 from developed European countries that are fully covered by S&P Global Market Intelligence.

19 Our inferences remain qualitatively unchanged if we exclude banks operating in Norway or Italy (unreported).

20 Our sample is characterised as 38% CTA adopters. The EBA reports that 43% of banks out of a sample of 54 (mostly) large banks opted for the CTA across 20 member states in 2018. Using all banks operating in the European Union, the percentage of CTA adopters increases to 57% (European Banking Authority [EBA] 2018).

21 The variable $\mathrm{IDIOSYNCRATIC}\mathrm{RISK}$ is multiplied by 100 for expositional convenience.

22 Our sample comprises 42.6% of banks in the SSM. Amongst the 38 CTA adopters, 19 are operating in the SSM. Of those 19 banks, 13 are operating in countries characterised by a powerful banking authority as measured with the 2019 ‘official supervisor power’ index ($\mathrm{SP}$> = 11). The correlation between the indicator variable $\mathrm{SSM}$ that captures banks in the SSM and SP is 0.34. It indicates that SSM banks are operating in countries with more powerful national banking authorities. Loipersberger (2018) shows that the stock market reacted more positively to announcements that regard the implementation of the SSM in countries with a less powerful banking authority, suggesting that the SSM, in providing the ECB with supervisory powers over individual banking institutions, can influence financial system stability. Overall, the country-level SP index might capture the effect of the SSM as suggested by the positive correlation between $\mathrm{SP}$ and $\mathrm{SSM}$ (unreported). However, our goal is not to evaluate the efficacy of national versus supra-national supervisors, but rather to investigate the impact of the power of banking authority in general.

23 To maximise statistical power, we do not split the sample between strong versus weak banking authority, as we restrict the sample size to the IFRS 9 period. However, our inferences are qualitatively similar if we apply this split (unreported). We do not include the pre-IFRS 9 period when measuring $\mathrm{ALI}$ and $\mathrm{ASYM}\mathrm{ALI}$ to ensure that the residuals ($\mathrm{ALI}$ and $\mathrm{ASYM}\mathrm{ALI}$) do not capture changes in ECL measurements.

24 To avoid selection bias, in an unreported analysis we exclude 6 IRB banks that adopted the IRB approach concurrently to or after the publication of IFRS 9 in 2014. Overall, our conjecture is not affected by the exclusion of those banks.

25 We acknowledge that we already control for bank capitalisation with the variable $\mathrm{CAPITAL}\mathrm{RATIO}$. In an unreported robustness test, we exclude the variable $\mathrm{CAPITAL}\mathrm{RATIO}$ and we find that $\mathrm{DIFF}$ remains negative and statistically significant at the 1% level. In addition, we replace $\mathrm{DIFF}$ with the risk-weight density measured as in Vallascas and Hagendorff (2013) (i.e. using the ratio of the risk-weighted assets over total assets). Again, consistent with the level of bank credit risk influencing the choice to adopt the CTA, we find that the coefficient on the risk-weight density is positive and significant, as long as we exclude the variable $\mathrm{CAPITAL}\mathrm{RATIO}$.

26 Using Equation (3), we decompose total risk (e.g. Holod et al. 2020) as: ${\beta }_i^2{\sigma }_{R\mathrm{\_MSCI}}^2$+${\sigma }_{\epsilon }^2$. We find that systematic risk (${\beta }_i^2{\sigma }_{R\mathrm{\_MSCI}}^2$) represents 12.4% of total risk (${\beta }_i^2{\sigma }_{R\mathrm{\_MSCI}}^2$+${\sigma }_{\epsilon }^2$). Our inferences are qualitatively similar if we use this decomposition for total, systematic, and idiosyncratic risks in our analysis.

27 This effect is likely to induce a ‘cliff effect’ in loan impairment (Novotny-Farkas 2016). Moreover, Gaffney and Mccann (2019) report that when the economy improves, it is likely that a large amount of loans that had been transferred to Stage 2 will be re-classified into Stage 1.

28 For instance, this effect is reflected in the UBS financial report of the second quarter 2020: ‘Total net credit loss expenses were USD 272 million during the second quarter of 2020, compared with USD 12 million in the prior-year quarter, reflecting net expenses of USD 202 million related to Stage 1 and 2 positions, and net expenses of USD 70 million related to credit-impaired (Stage 3) positions. Stage 1 and 2 net credit loss expenses of USD 202 million were primarily driven by a net expense of USD 127 million from an update to the forward-looking scenarios, factoring in updated macroeconomic assumptions to reflect the effects of the COVID-19 pandemic, in particular updated GDP and unemployment assumptions. This also led to exposure movements from stage 1 to stage 2 as probabilities of default increased’ (p. 12).

29 Systematic risk exposure can be key to bank contributions to systemic risk (e.g. Iannotta et al. 2019).

30 An unreported Chow test reveals that the difference in the magnitude of ${\beta }_3$ is not statistically significant (two-tailed p-value = 0.17 clustered at the bank level) across Columns 1 and 2. Nevertheless, the magnitude becomes statistically significant at conventional levels if we exclude banks with the highest score in the weak group (i.e., when SP = 10).

31 We also provide level of significance based on a one-tailed test, as we have a clear prediction and signs are consistent.

32 LRMES is the fraction of the bank’s loss when the MSCI World index declines 40% over a six-month window. Intuitively, if one multiplies the LRMES by the market value of equity, it results in the absolute market value loss due to a systemic financial crisis in millions of euros. The DCB used in the computation of the LRMES is estimated using generalised autoregressive conditional heteroscedasticity and dynamic conditional correlation. LRMES and DCB data are collected from V-Lab, maintained by the NYU Stern School of Business. The theoretical motivation of the measure is given in Acharya et al. (2012).

33 The use of the MSCI World index in our main tests should not bias our results, because European countries did not experience a ‘local’ crisis during the period 2016–2019 (Engle et al. 2015).

34 The implementation of this matching procedure is based on the ebalance command in Stata, further described in Hainmueller and Xu (2013).

35 In September 2016, the IASB issued an amendment to IFRS 4, introducing a temporary exemption from the adoption of IFRS 9 until 2021 for insurance companies that have not yet applied IFRS 9.

36 We follow the original sample-selection procedure described in Panel A of Table 1, selecting 30 insurance companies as an alternative control group. We restrict our sample to insurance companies that are located within the 19 European countries analyzed in this paper. As additional criteria, we excluded insurance companies involved with banking through subsidiaries, companies that do not qualify for temporary exemption under IFRS 4, and companies that adopted IFRS 9 early with respect to the temporary exemption.

37 We tried to match the first and second moment for all firm-specific covariates. However, the entropy balance maximum likelihood routine does not converge. The lack of convergence is primarily driven by the earnings variables, which might highlight structural differences in reported earnings across banks and insurance companies. Consequently, we excluded $\mathrm{ROA}\text{\%}$ and $\mathrm{ROA}\mathrm{SD}$ of the matching procedure.

38 This Interim Final Rule permits U.S. banks that were required to implement the CECL model before the end of 2020 to have a five-year CTA period: https://www.occ.treas.gov/.

39 https://www.bankingsupervision.europa.eu/.