Impact of IFRS 9 on the Cost of Funding of Banks in Europe

On implementation, IFRS 9 increases credit loss (impairment) charges and reduces after-tax profits of banks. This makes retained earnings and hence capital resources lower than what they would be under IAS 39. To maintain their capital ratios under IFRS 9, banks could elect to hold higher levels of equity capital. This paper uses a modified version of CAPM, which accounts for the low-risk anomaly (as suggested by Baker and Wurgler (2015)), to estimate the impact of this potential increase in capital levels on the cost of funding of banks in six European countries, the UK, Germany, France, Italy, Spain and Switzerland. We confirm the existence of low-risk anomaly for banks’ equity in the six countries, except France. The magnitude of the anomaly varies across countries, but is generally low relative to the long-run cost of equity for banks. Our results show that, on day 1, the implementation of IFRS 9 has minor impact on the cost of funding of banks in the six countries.<br>


Introduction
In 2018, the International Accounting Standards Board's (IASB) Standard 9 (IFRS 9 4 ) came into effect, replacing IAS 39 5 standard. IFRS 9 has important implications especially for banks, as they mostly hold financial assets. The incurred loss model of IAS 39 allows recognition of credit losses only when there is 'objective evidence' of credit impairment, causing delayed identification of potential credit losses.
This, as many argue, could reinforce the pro-cyclicality effects of financial regulation (Novotny-Farkas (2016)). To mitigate the delayed recognition of credit losses, IFRS 9 introduces a forward-looking provisioning model, under which credit loss provisions (or impairment charges) are equal to the expected credit losses. The expected loss model of IFRS 9 is anticipated to reduce the pro-cyclicality of financial regulations, and hence improve financial stability.
IFRS 9 increases credit loss provisions charges for banks 6 . This rise in provision reduces after-tax profits and retained earnings, which represent a key component of Common Equity Tier 1 (CET1) resources for banks. Other things equal, this leads to higher levels of leverage. Adrian and Shin (2010) point out that a bank determines its leverage level depending on the implicit maximum leverage permitted by collateralized creditors. Hence, if the better asset quality transparency under IFRS 9 does not lower that implicit maximum level of leverage, banks would have to preserve their pre-IFRS 9 levels of leverage. To maintain their capital ratios under IFRS 9, banks may choose to hold higher levels of equity capital 7 . In a standard Modigliani-Miller environment, this increase in equity capital would not affect the overall cost of capital 8 . In efficient and integrated capital markets, the lower cost of equity and debt completely offsets the larger share of equity in the capital structure, leaving the weighted average cost of capital (WACC) unchanged. However, a number of inefficiencies or frictions have been 4 International Financial Reporting Standards 9: Financial Instruments. 5 International Accounting Standard 39: Financial Instruments: Recognition and Measurement. 6 This is based on a naïve logic. Because it cover more loans in its scope (Stage 1 and Stage 2 loans), IFRS 9 would make impairment charges, in a given year, higher than what they would be under IAS 39. However, in certain circumstances, IFRS 9 might lead to lower impairment charges and higher profits. This could be the case in an upturn if the stock of loan loss provisions at the start of a year exceeds that required at the end of the year (due to more optimism in expectations). 7 Banks could maintain their capital ratios by deleveraging and/or de-risking, rather than holding more equity capital. 8 For example, Brigham and Ehrhardt (2014): "As leverage increases, more weight is given to low-cost debt, but equity becomes riskier. Under Modigliani-Miller assumptions, the cost of equity capital increases by exactly enough to keep the cost of capital constant" (p. 597).
observed in actual capital markets, challenging the validity of this proposition. One of these frictions, the "low-risk" anomaly, refers to the empirical observation that historical returns and, hence, realised cost of equity are higher for shares with lower betas. In other words, lower levels of leverage are not necessarily associated with proportionally lower cost of equity. This means that increasing the share of equity in the capital structure would increase WACC. Several authors find a negative relationship between the returns on equity and the systemic (Fama and French (1992), Baker eta al. (2011b), and Baker et al (2014)) and idiosyncratic risks (Ang et al. (2006)) of the issuer, suggesting that the low-risk anomaly appears in equity markets. The low-risk anomaly has been observed in each of the G7 countries (Ang et al. (2009)), and across 23 developed economies (Baker and Wurgler (2016)). Baker et al (2011a) attribute this anomaly to a combination of irrational investor demand for highly volatile shares 9 and limited arbitrage 10 , resulting from "delegated investment management with fixed benchmarks and no leverage". Frazzini and Pedersen (2014) indicate that funding constraints (such as leverage constraints and margin requirements) affect risk-preferences of investors. They point out that while constrained investors tend to invest in risker shares, unconstrained investors hold portfolios with betas below one, on average. Karceski (2002) specifies an alternative explanation for the anomaly. He points out that the strategy of mutual fund managers of investing in high-beta stocks, which offer higher return in bull markets, reduces the risk premia for the high-beta shares, and can inverse the risk-return relationship.
We estimate the impact of the potentially higher levels of capital under IFRS 9 on the cost of funding of banks in six European countries, the UK, Germany, France, Italy, Spain and Switzerland. This would help us better understand the costs of IFRS 9 11 . These costs have to be compared to the benefits of earlier recognition of credit losses. 09 Authors attribute this irrational preference for shares with higher volatility to investor overconfidence (Cornell (2008)), and lottery preferences (Kumar (2009), Barberis andHuang (2008), andBali et al. (2011)). 10 Baker et al (2011a) explain the limited arbitrage preventing sophisticated institutional investors from exploiting any low-risk anomaly by the following. First, shorting highly volatile shares can be hard, especially for smaller companies with small number of shares available to borrow in the market. Moreover, institutional investors do not act on their own behalf in most cases. As their customers want to ensure they can compare different investors among asset classes, the investors must perform relative to a benchmark. This benchmarking restricts their ability to exploit the low-risk anomaly. 11 Our analysis covers the potential microeconomic costs only, and doesn't investigates the macroeconomic costs. For instance, if banks chose to pass the higher costs to their customers, this could lead to lower lending and possibly lower output.
We follow the method suggested by Baker and Wurgler (2015), which adopts a modified version of CAPM. This method allows us to check whether the low-risk anomaly exists for banks' equity, and to estimate the impact of changes in the capital structure of banks on their costs of funding, in the presence of such anomaly. This paper contributes to the literature studying the impact of capital structure on the cost of capital, especially those investigating the inadequacies of CAPM predictions.
Consistent with past literature (for instance, Baker and Wurgler (2016) and Arakelyan and Karapetyan (2014)), our results confirm that low-risk anomaly exists for bank equity in the UK, Germany, Italy, Spain and Switzerland. However, the results do not provide a robust evidence of the anomaly for French banks' equity. The annual magnitude of the anomaly varies across countries, but is generally low relative to the long-run cost of equity for banks. We show that the possible higher capital levels under IFRS 9 may slightly increase banks cost of funding in the six countries except France, where the cost of funding may fall.
It is important to note that our estimates of IFRS 9 impact on the cost of funding should be viewed as the "day 1" impact. Their validity as estimates for the longer-term impact of IFRS 9 relies on two main factors that are out of the scope of this paper. First, our analysis does not investigate whether the impact of IFRS 9 on the level of equity capital would be stable across different stages of the credit cycle. Moreover, our analysis does not account for the potential increase in asset quality transparency under IFRS 9. Yet, our analysis provides important insights about the impact of IFRS 9 on the cost of funding.
The reminder of the paper proceeds as follows. Section 2 outlines our model, Section 3 describes the data we use in the estimation, Section 4 presents the empirical analysis, and Section 5 concludes.

The Model
For the purposes of measuring the low-risk anomaly, we use equity beta as a measure of equity risk, defined as the covariance of returns on equity with the risk premium of the market, divided by the variance of the risk premium of the market. Under CAPM, asset beta is the weighted average of equity and debt betas: Where, βa: asset beta; βe: equity beta; βd: debt beta; and e: ratio of equity to total assets (the leverage ratio). By rearranging Equation (1), we can write βe as follows: Assuming approximately riskless debt, asset beta can be defined as the slope of the linear relationship between the inverse of leverage ratio and equity beta. As debt beta is normally significantly lower than asset beta, the estimation of βa will allow us to assess how reasonable the assumption of riskless debt for banks is. Additionally, assuming the CAPM holds for equity and debt, the returns on equity and debt are: Where: re: return on equity; rd: return on debt; rf: the risk-free rate; and rp: excess return on the market portfolio. However, following Baker and Wurgler (2015), we assume that there is an anomaly in the sense that lower beta shares outperform their CAPM benchmark, whereas high beta shares underperform it. That is: Where, = . ( − 1) is the low-risk anomaly term, and = < 0 is the magnitude of the lowrisk anomaly. Using the returns on equity and debt (Equations (4) and (5)) and equity beta (Equation (2)), WACC is given by the following 12 : Our aim is to calculate the impact of the potential increase in the level of equity capital, induced by the implementation of IFRS 9, on the cost of funding of banks. In other words, we are interested in calculating the change in WACC when the level of capital increases from e to e*. We can derive the change in WACC from Equation (6) as follows: 6 Our benchmark is the special case where bank debt is riskless (βd = 0) 13 . In Equation (7), ΔWACC becomes a function of the low-risk anomaly and the change in the level of capital. That is: We also explore another plausible scenario, under which debt is risky (βd > 0) but not very responsive to changes in leverage levels (βd isn't sensitive to small changes in leverage). Hence, ΔWACC would be: For the purposes of this paper, we assume that there is no low-risk anomaly in debt markets, and that there are no government guarantees for bank debt. The presence of the low-risk anomaly in debt markets reduces the impact of the low-risk anomaly in share markets on the change of WACC 14 . If an anomaly existed in debt market with the same magnitude of that in equity markets, then changes in capital structure would not affect WACC. Frazzini and Pedersen (2014) and Baker and Wurgler (2016) indicate that the low-risk anomaly in debt markets, if existed, is much smaller than that in share markets. Conversely, the presence of government guarantees for bank debt (such as deposit insurance) in practice increases the low-risk anomaly impact on the change of WACC. This is because such guarantees reduce the riskiness of banks debt and weaken the relationship between the riskiness of debt and the level of leverage (debt beta does not fall as much as what CAPM would predict) 15 .
Consequently, as Baker and Wurgler (2015) indicate, the estimated level of γ is a plausible approximation of the magnitude of the low-risk anomaly for banks.

The data
Our sample consists of 75 publicly traded banks from six European countries. It includes 10 UK banks, 8 German banks, 8 French banks, 9 Spanish banks, 17 Italian banks, and 23 Swiss banks. Other large banks could not be included as there is no publicly traded equity available for them. We collect daily data on banks' share price, market indices, and the yield of 10-year government bonds (used as the 13 If bank debt were risky, the change in the cost of funding would be reduced to the extent that the increased level of capital reduces βd, as Equation (7) shows. Although the assumption of riskless debt is a reasonable for banks, it does not hold true for very highly levered banks. Yet, we can drop such extreme cases, as banking regulations, which requires certain equity ratios, would prevent those cases. 14 Stronger low-risk anomaly in debt markets and/or higher riskiness of bank debt reduce the impact of the low-risk anomaly in share markets on WACC, as Baker and Wurgler (2015) indicate. 15 Baker and Wurgler (2015) indicate that the "calibration with riskless debt remains a reasonable estimate in the presence of such factors". 7 risk-free rate) from Refinitiv Eikon®. Our dataset spreads over the period between 1997 and 2017, with about 322,000 daily observations across the 75 banks. Table 1 presents country-level statistics of the market data. We also use quarterly balance sheet data between 1999 and 2017 from Eikon® to calculate three capital ratios used in the estimation of asset betas of the banks: the leverage ratio, common equity Tier-1 (CET1) ratio, and Tier-1 (T1) capital ratio. We define the leverage ratio as the ratio of total common equity to total assets, whereas CET1 and T1 capital ratios are calculated by dividing total common equity and T1 capital by total risk-weighted assets. The capital ratios dataset comprises around 5,000 observations across the 75 banks.  Source: Refinitiv Eikon®. 16 We estimate the average risk weight by dividing the leverage ratio by the CET1 ratio. 8 For the impact of IFRS 9 on the level of equity capital, we use two sources: the European Banking Authority's (EBA) report on results from its second impact assessment of IFRS 9 17 (July 2017), and Mazars's quantified impacts of IFRS 9 18 (March 2018). Both studies estimate the impact of the IFRS 9 on the CET1 ratios of a sample of European banks. The EBA's estimates use a sample of 54 banks, compared to 27 banks in Mazars's analysis. However, Mazars's estimates are more recent and presents bank-level estimates, allowing us to extract the IFRS 9 impact for 20 of the banks in our sample. A common feature of the two studies is that banks at the top quartile have very large negative estimated impacts compared to the average and median in each of the samples. Table 3 presents the data extracted from the two studies. Source: 1 Report on Results from the Second EBA Impact Assessment of IFRS 9. 2 Quantified Impacts of IFRS 9: Initial Findings. 3 The estimates for Santander UK parent (Banco Santander SA) are used.

Empirical Analysis and Results
We concentrate, in our estimation of the impact of the IFRS 9-induced potential increase in the level of equity capital on the cost of funding of banks, on the case where bank debt is riskless. We also explore another plausible scenario, under which debt is risky but not very responsive to changes in the level of leverage. As Equation 9 shows, the impact of IFRS 9 on the cost of funding would be lower, under this scenario, if debt beta is positive. Allowing the debt riskiness to vary with the level of leverage would decrease the impact of IFRS 9 on the cost of funding to the extent lower levels of leverage reduce the riskiness of bank debt. 9

Estimation of Asset Beta:
To estimate asset beta (Equation (1)), we regress equity betas on the inverse of leverage ratios for the banks in our sample, on quarterly basis. Equity betas are estimated by regressing the daily excess returns on equity for each bank on the excess returns of market indices over the two years preceding the observation period. We also estimate assets betas using other capital ratios, namely CET1 ratio and T1 capital ratio.  Asset beta is the weighted average of equity and debt betas. For most banks in our sample, asset betas (Appendix A2) are close to zero and very low relative to equity betas (Appendix A7). This means debt betas are very close to zero, indicating that riskless debt assumption is a reasonable approximation for the banks in our sample. However, it is important to note here that leverage levels at banks are partially influenced by bank regulations. This weakens the relationship between risk and leverage and flattens the cross-sectional relationship between risk and leverage. Hence, as Baker and Wurgler (2015) indicate, estimated βa would represent a plausible lower bound of asset beta.

Estimation of Magnitude of the Low-Risk Anomaly (γ)
To estimate γ, we start by estimating alphas and betas in Equation (4) for three portfolios 20 for each country: large banks (the largest 30% of the banks), small banks (the smallest 30%), and medium banks 10 (the remaining 40%). This is done by regressing excess returns on each of these portfolios on market excess returns. We then estimate the country-level values of γ by plotting the resulting alphas and betas for each country 21 . As the values in Table 5   To put these estimates in a context, we should relate them to the cost of equity of banks. King (2009) estimates the real cost of equity for banks in the UK, Germany and France at 6.6%, 9% and 7.3%, respectively. Based on these estimates, a percentage point increase in the share of equity in capital structure could raise the cost of funding of UK and German banks by about 1.61% (0.1061%/6.6%) and 1.63%, respectively. A similar increase in the share of equity capital could decrease the cost of funding of French banks by 0.53% of their long-term cost of equity. The annualised magnitude of the low-risk anomaly increases slightly, when debt is risky but not very responsive to changes in the level of leverage (Equation (9)), as Table 5 shows. This is because the weighted average of the debt betas of the country-level portfolios is negative in all countries 22 .

Estimation of the impact of IFRS 9 on the cost of funding
Having estimated the magnitude of the low-risk anomaly, we can assess the impact of the implementation of IFRS 9 on the cost of funding for banks. We do that by multiplying the annual magnitude of the low risk anomaly for each country (Table 5) by the average impacts of IFRS 9 on the levels of equity capital extracted from the EBA's report 9 and Mazars's study (Table 3). In addition to the two samples of banks reported in the two studies, we create country-level subsamples containing the banks included in Mazars's study from each of the six countries in our sample. Table 6 presents the estimated impact of IFRS 9 on banks' cost of funding in each country. As the Table shows, IFRS 9 may slightly affect the cost of funding of banks in the six countries 23 . Specifically, assuming that banks would fill any shortfalls by increasing equity, IFRS 9 may increase the cost of funding for all banks except French banks, whose cost of funding would fall. For example, based on the EBA's estimated impact of IFRS 9 on capital ratios, the cost of funding of UK banks may increase by about 5 bps. This impact decreases to about 3 bps when using Mazars's estimates. It is important here to note that we should interpret these estimates as the "day 1" impact of IFRS 9 on the cost of funding. Their validity as estimates for the longer-term impact of IFRS 9 relies on two main factors that are out of the scope of this paper. First, our analysis does not investigate whether the impact of IFRS 9 on the level of equity capital would be stable across different stages of the credit cycle. Moreover, our analysis does not account for the potential increase in asset quality transparency under IFRS 9. This increase in transparency might induce a reduction in the cost of equity and debt for banks at different levels of leverage 24 . Had this been the case, our estimates would have overstated 12 the impact of IFRS 9 on the cost of funding of banks. However, our analysis delivers important insights about the potential implications of IFRS 9. It also provides a good start for similar analyses in the future when IFRS 9 becomes more established.

Robustness checks
We re-estimate γ in five alternative ways. In the first experiment, we use excess returns on six portfolios including all the banks in each country. We estimate γ values by substituting the resulting alpha and beta into x = . ( − 1). In the second experiment, we use excess returns on two portfolios consisting of the largest 50% and smallest 50% of the banks, in each country. In the third experiment, we regress excess returns for each bank in the sample on market excess returns. As in the baseline case, we estimate γ in the second and third experiments by plotting the resulting alphas and betas. In the fourth experiment, we estimate alphas and betas using panels of excess returns for all banks in each country, and then estimate γ values as in the first experiment. As in the baseline, we use three portfolios in the fifth experiment. However, in this experiment, we classify banks in terms of their riskiness (equity betas), following Baker and Wurgler (2015). Thus, for each country, we create three portfolios: high-risk banks (the 30% most risky banks), low-risk banks (the 30% least risky banks), and medium-risk banks (the remaining 40% of the banks). the low-risk anomaly appears in the five experiments, but its magnitude is considerably higher than the baseline estimate in the first experiment and somehow lower in the fourth experiment. This could be due to a relatively higher anomaly for the largest two German banks in our sample (Deutsche Bank and Commerzbank), compared to other German banks. Compared to the baseline, the two banks have stronger influence on the magnitude of the anomaly in the first experiment and weaker influence in the fourth experiment 26 . Additionally, γ estimates for Italian banks are consistent with the baseline 25 Appendices A5 to A9 includes more information about γ estimation under each of the five robustness experiments. 26 In the first experiment, we have only one portfolio comprising all German banks, in which the two banks have combined weight of 96.5%. As a result, the two banks would have a stronger contribution to the values of the alpha and beta of the 13 estimates in the last three experiments but not in the first two experiments. This is especially true in the second experiment, when γ becomes positive. In the first experiment, γ is negative but has a very weak annual magnitude of 0.1bps. Lastly, γ estimates for French banks are strongly affected by the portfolio choice in each experiment. They are inconsistent and fluctuate in sign and magnitude across experiments. Overall, the results of the five experiments confirm our baseline assessment that the low-risk anomaly exists for banks equity in the six countries in our sample, except France.  portfolio, and hence the estimated magnitude of the low-risk anomaly. In the baseline, the two banks affect only one of three sets of alpha and beta which have equal impacts on the estimated anomaly, making their influence on it weaker. Their influence is even weaker in the fourth experiment where the banks have equal weights in the panel.

Conclusion
IFRS 9 replaces the incurred loss model of IAS 39 with a forward-looking expected loss model, under which credit loss provisions are equal to the expected credit losses. The expected loss model is likely to increase credit loss charges for banks, reducing after-tax profits, and, hence retained earnings; the main source of equity capital of banks. Thus, to maintain their capital ratios, banks may choose to hold higher levels of equity capital under IFRS 9. To estimate the impact of this IFRS 9-induced potential increase in equity capital on the cost of funding of banks, we followed Baker and Wurgler (2015) by adjusting CAPM to account for the low-risk anomaly.
Consistent with past literature, we confirm that the low-risk anomaly exists for bank equity in the UK, Germany, Italy, Spain and Switzerland. However, the results do not provide a robust evidence of the anomaly for French banks' equity. The annual magnitude of the anomaly varies across countries, but is generally low relative to the long-run cost of equity for banks. We show that the implementation of IFRS 9 may slightly increase the cost of funding for banks in the six countries except France, where the cost of funding for banks could fall.
Whether we can view this impact as an estimate of the longer-term impact of IFRS 9 depend on two elements, which are out of the scope of our analysis. First, we did not investigate whether the impact of IFRS 9 on the level of equity capital would be stable across different stages of the credit cycle.
Likewise, we did not account for the potential positive effects of the early recognition of losses under IFRS 9 on asset quality transparency. An increase in this transparency might reduce (or increase) the cost of equity and debt for banks at different levels of leverage. In this case, our estimates would have overstated the impact of IFRS 9. Nevertheless, our analysis provides important insights about the potential implications of IFRS 9 on the cost of funding of banks. It also represents a good start for similar analyses in the future when IFRS 9 becomes more established. • Start from the definition of WACC:
• Substitute in the values of re and rd from Equations (5) and (4), respectively: • Substitute in the value of βe from Equation (2): • For a given level of βa, Since βd is a function of e:

A2.1 Asset betas for UK banks
the leverage ratio CET1 ratio Tier-1 Capital ratio Bank quarterly data annual data quarterly data annual data quarterly data annual data

A2.4 Asset betas for Italian banks
the leverage ratio CET1 ratio Tier-1 Capital ratio Bank quarterly data annual data quarterly data annual data quarterly data annual data