COMPARISON OF THE WATERFALL MODEL SPREADS

11. Comparison of the Waterfall Model

Naturally, the next question to ask is whether we can use the findings of our wa-terfall model to make statements about the spreads we observe in the market on a more general level – without going into too much detail and analyzing each indi-vidual bank. The answer is that we can. The results from our waterfall model show that, as in any other bond-pricing framework, credit spreads are seen as a compen-sation for default risk. This risk is driven by two factors: factors that influence the PD, and factors that influence the LGD. While we have found that both types of factors influence the absolute credit spreads, only LGD factors have an effect on the relative spread ratio. Our results further show that the main drivers behind absolute spreads are volatility of assets and the CET1 ratio, whereas relative spreads are driven mostly by the balance sheet composition, i.e. the amount of AT1, T2, and T3, the location of the PONV, and the costs of the resolution process.

Drivers of Absolute Spreads

There might be an intuitive reason why we see the aforementioned differences across banks in our waterfall model findings. Starting with the cases in which bonds trade at same relative levels but differ in absolute spreads, consider two hypothet-ical banks in the same jurisdiction. These banks are supervised by the same au-thority, are subject to (almost) the same MREL requirements, and optimize their balance sheets such that they hold approximately the same amount of AT1 and T2 capital. The banks may differ in their CET1 level, possibly due to the fact that one bank has a higher capital requirement because it is deemed to be systemically im-portant, or because it chooses to hold a greater CET1 management buffer. At the same time, the banks may also differ in their business operations, or have different asset compositions, such that they differ in terms of asset volatility.

Since the two banks are supervised by the same authority, they will most likely have approximately the same PONV. Assuming that the costs of resolution are random and expected to be the same, our waterfall model predicts that the two banks will trade at approximately the same relative spread ratio. However, due to their different asset volatility and their different CET1 ratios, the banks could potentially trade at highly different absolute levels. This may not be surprising, given the literature on bond pricing. However, that our model is indeed able to capture these “expected” dynamics of spreads does give us a greater confidence in its robustness.

Drivers of Relative Spreads

As in this thesis we are motivated to investigate relative spreads and the pricing of T3, it might be more interesting to consider the drivers behind relative spreads. To this end, our results suggest that we should examine the balance sheet composition, the location of the PONV, and the expected costs of resolution for an explanation.

We will therefore discuss these three identified factors in relation to the market observations. These factors are subject to a significant number of uncertainties due to regulatory discretion, random factors, and doubt regarding the final implemen-tation of the MREL. Therefore, our waterfall model provides a tool to quantify these uncertainties, which are unique to banks. In contrast, conventional models can be difficult to use in this matter.

11.3.1. Balance Sheet Composition

Our waterfall model shows that the T2/T3 spread ratio increases when the tranche size of both AT1 and T2 increases. At the same time, we see a substantial expansion in the T2/T3 spread ratio when the T3 tranche increases. Since there is still uncer-tainty about the final implementation of the MREL in many EU countries, includ-ing Denmark, our model can be used to make qualified predictions about how the relative spreads should vary with different levels of the MREL. Thus, instead of only considering the T2 spreads when pricing T3 as seems to be the current market standard (Morgan Stanley Survey, 2017), our model provides a theoretical expla-nation for how bond spreads should vary.

It is important to keep in mind that the T3 is currently a completely new liability class. This means that stack sizes between banks may vary substantially, which alone could drive some of the relative spread differences we observed across banks in chapter 4. Note at the same time, however, that the novelty of the liability class also could mean that the T2/T3 ratio we observe in the market at the moment might be biased downwards. This is because the stacks of MREL-eligible debt are still thin across European banks, as the latter are in the phase of building up their new capital structure. Thus, due to the thin debt tranche, the LGD might be “too”

high at the moment for T3, compared to a fully phased-in scenario. This effect is further amplified by the credit rating of these new instruments, which will most likely be low initially due to small stack sizes. According to Barclays, T3 bonds are likely to be rated in the BBB category when issued, which will push down prices and raise spreads such that the T2/T3 ratio will be lower (Barclays, 2017). In time,

the recovery prospects for the T3 bonds due to a lower LGD, as well as the ratings on the instruments.

Our model suggests that balance sheet composition is one of the main drivers of spread differences between banks. While it might be clear that spreads should de-crease in stack size, it is certainly not evident how much spreads should vary. This is due to the complexity of a bank’s capital structure, and the amount of capital regulation. However, this can be investigated using our model. We do see large differences in the balance sheet composition of banks in Europe. This could, for example, be due to different capital requirements across countries, or simply to different target capital ratios. Furthermore, the fact that the MREL will be set at country level and adjusted on a case-by-case basis means that different stack sizes of debt will continue to drive relative spread differences across Europe, also in the longer run.

11.3.2. Cost of Resolution

The results of our model show that credit spreads are highly sensitive to changes in the costs of the resolution process. A change of only 1 percentage point from 5%

to 6% increased spreads on T3 from 49bps to 93bpd, corresponding to an increase of 92%. With regard to the T2/T3 spread ratio, we found a large effect of changing the cost parameter. For low levels of costs, the ratio is high, while it is low for higher costs. These observations might help explain some of the differences in spread ratios we observed across European banks in chapter 4.

However, in practice, the exact costs of the resolution process are difficult to esti-mate in advance, as they depend on the circumstances, the resolution tools applied by authorities, and the market conditions at the time of resolution. The market may have perceptions about the costs being higher for some banks than for others, which may be due to their business activities or the liquidity of their assets. How-ever, a lesson learned from the financial crisis is that liquidity can suddenly dry up, such that many instruments for which there has previously been a liquid market can only be sold at fire-sale prices (Hull, 2015). Thus, such effects or perceptions can be difficult to price into bonds in advance.

In the context of the costs of resolution, it might be appropriate to touch briefly upon a “real life” situation that has unfolded in recent months in Italy, because it could present a different interpretation of the costs of the resolution parameter – at least temporarily.

11.3.3. To Bail-In or not to Bail-In?

Over the last months, as we have written this thesis, three Italian banks have requested a “precautionary recapitalization,” which is a feature included in Article 32(4) of the BRRD. The first bank to make this request was the larger bank Monte dei Paschi, in late December 2016. In April 2017, the two smaller Italian banks Veneto Banca and Banca Popolare di Vicenza made the same request. By request-ing a precautionary recapitalization, the banks hope to bypass enterrequest-ing resolution, and thereby bail-in.

Precautionary recapitalization describes the injection of own funds into a solvent bank, and it is meant to be approved only in exceptional cases when the financial stability of an EU Member State is at risk (ECB, 2017). The benefit of using pre-cautionary recapitalization is that resolution will not be triggered for the bank. In this context, it is at the same time important to note that a precautionary recapi-talization will be carried out in accordance with the state aid rules laid down by the European Commission. Under these rules, the loss sharing burden is reduced to only shareholders and subordinated bondholders; thus, in such an event, T3 will not be used for bail-in, and bondholders will not incur any loss. This has caused a significant amount of uncertainty regarding the credibility of the bail-in tool and the risk of the T3 bonds.

In April 2017, the ECB declared both Veneto Banca and Popolare di Vicenza sol-vent, because their last reported capital ratios complied with the Pillar 1 capital requirements. This declaration is important because it shows that the ECB is will-ing to view a bank as solvent with as low as a 4.5% CET1 ratio, thereby qualifywill-ing for a precautionary recapitalization.

This highlights the fact that the cost parameter might at the moment be interpreted differently than only the costs of the actual resolution process. Given the situation in Italy, the cost parameter could arguably also be seen as a likelihood that losses will in fact be imposed on bondholders all the way up through the balance sheet, i.e. a probability that the bail-in tool will be applied. This interpretation might help to explain some of our empirical observations.

In our waterfall model, we can roughly interpret a precautionary recapitalization as the case when the cost parameter is set between 0% and 4%. In such a case, T3 will not incur any losses and T2, AT1, and CET1 will absorb everything. This

in a high T2/T3 ratio. This suggests that if the precautionary recapitalization will be applied in practice¹⁷ (or just expected to be by the market), the relative spread between T2 and T3 should be large. This could in turn lead to relative spread differences, due to regulatory discretion across countries in Europe and the fact that regulators might have different views on the use of precautionary recapitaliza-tion versus full bail-in.

With this discussion in mind, let us consider the T2 and T3 spread differences between European and US banks.

Table 11: US TLAC and Tier 2 Bonds

Source: Bloomberg, Barclays Research

In Figure 11 above, we show spreads on T2 and T3 bonds for major European and US banks. Although the MREL does not apply in the US, these banks are subject to the TLAC, which is MREL’s G-SIB equivalent. We see than the relative spread ratios for US banks are much lower than for EU banks, all in the area of 1.2x.

Considering the history of bank failures in the US during the financial crisis, and the fact no feature such as precautionary recapitalization exists there, the reason why the US T3 and T2 bonds trade closer to each other is clear. It is difficult to conclude anything from these findings, since the shown effects may be due to many different factors. However, they might highlight some uncertainty regarding the application of the bail-in tool in the EU.

11.3.4. PONV

The failure of a bank is unlike the failure of a normal corporation. For banks, regulators set the default boundary. Therefore, it is important that one be able to measure the effects of this regulatory discretion. This can be done with our model.

17 We discuss this further in the end of this thesis. See section 12.4

Bond Description Z-spread (bps)

TLAC senior Tier 2 TLAC senior Tier 2 Ratio

C 3.887 '28 C 4.125 '28 146 187 1.3 x

BAC 3.824 '28 BAC 4.183 '26-27 144 163 1.1 x

GS 3.85 '27 GS 5.95 '27 150 170 1.1 x

JPM 3.782 '28 JPM 3.625 '26-27 133 157 1.2 x

MS 3.625 '27 MS 3.95 '27 138 172 1.2 x

WFC 3 '26 WFC 4.3 '27 121 146 1.2 x

We saw in the previous chapter, that the location of the PONV had significant impact on spreads. This could in turn lead to both absolute and relative spread differences across countries in the EU if the PONV decision is not taken uniformly.

Whether a bank has reached the PONV or not is officially decided by the relevant resolution authority, but it will most likely be the ECB’s SRB that makes the final decision on a bank’s fate (Barclays, 2017). The extent to which local authorities are able to influence the PONV decision might therefore cause differences across the EU. Furthermore, the ECB only has the power to make the PONV decision in the Eurozone. Therefore, we might see that the PONV decision in this region is different than in the Nordic countries, the UK, or Switzerland.

In this context, it is interesting to return to the current situation in Italy. As men-tioned above, the ECB chose to declare the two Italian banks solvent because their CET1 ratio was greater than the Pillar 1 minimum requirement. This decision effectively indicates that the ECB would consider an institution solvent with as low as a 4.5% CET1 ratio. Given the amount of industry literature we have reviewed and the experts with whom we have been in contact, including the EBA, in writing this thesis, it is our impression that the market has been assuming a PONV some-what higher than this level. This has also been our own assumption (note that we have used 6.0% CET1 in our benchmark scenario). If it is indeed the case that a 4.5% CET1 ratio will become the “market standard,” and we are correct in stating that the market may have assumed a higher PONV, then our model suggests that we, ceteris paribus, should see decreases in both T2 and T3 spreads and a lower relative ratio going forward.

Summary

In this chapter, we discussed how we can use our model to explain what influences and drives current spreads observed in the market. Different parameters have a different impact on spreads: some factors influence the absolute spreads, while oth-ers impact the relative spreads between T3 and T2. We also identified which of the factors in our model setting potentially impact spreads the most. However, the results show that many of the parameters affect spreads, thereby highlighting the importance of analyzing observed spreads using multiple factors. Finally, through our example of the recent event of potential precautionary recapitalization in Italy, we showed that our parameters can have different interpretations, and provided evidence that there may still be other factors that we do not consider in our current model setting.