G roup I nstitutional & R egulatory A ffairs

Group Institutional & Regulatory Affairs Banking Business Models Zeno Rotondi February 2015

Index • Introduction • Performance of alternative business models • Risk assessment of alternative business models • Contribution to the real economy during the financial crisis • Business modelsand government aid to banksduring the financialcrisis • Conclusions

Traditional business model of commercial banks Introduction FIGURE 1

Traditional Banking and Securitization in the Shadow Banking System: Originate-To-Hold model vs Originate-To-Distribute model Introduction FIGURE 2 Source: Bouwman (2015)

OTH model and relationship banking Introduction • Relationship bankingis defined as the provision of financial services by a financial intermediary that: • invests in obtaining customer-specific information, often proprietary in nature; and • evaluates the profitability of these investments through multiple interactions with the same customer over time and/or across products. • The first part highlights that banks obtain information while providing screening and/or monitoring services. The second part emphasizes the fact that information can be used in multiple interactions with the same customer, which allows the bank to reuse information. • Relationship banking involves the bank making the loan and holding it on its balance sheet. This is the so-called “originate-to-hold” model, in which banks fund relationship loans with core deposits. The loans are illiquid – banks keep them on their balance sheets until maturity. This reduces moral hazard on the side of the bank – the fact that the loans stay on the balance sheet gives the bank incentives to perform upfront screening and then monitor on an ongoing basis. Source: Bouwman (2015)

OTD model and shadow banking system Introduction • With securitization the bank originates loans as it does in the traditional OTH model, but then transfers the loans to a trust called a Special Purpose Vehicle (SPV), which issues various tranches of debt claims called asset-backed securities (ABS) against this pool of loans. These ABS are sold to institutional investors and the money received by the SPV is transferred in part to the bank. Thus, like loan sales, securitization, provides banks with extra funding that can be used to originate new loans. • The OTD model fuelled the development of the so-called shadow banking system, defined as financial entities other than regulated depository institutions (e.g. commercial banks) that serve as intermediaries to channel savings into investment. • The shadow banking system includes: • institutions such as investment banks, brokerage houses, and finance companies, • securitization structures such as asset-backed securities (ABS) and asset-backed commercial paper (ABCP), • key investors in securitized structures, such as money market mutual funds (MMMFs), which heavily rely on short-term funding like tri-party repurchase agreements (repos) and commercial paper (CP). Source: Bouwman (2015)

Maindifferences b/w OTH and OTD models Introduction • As in the case of the OTH model, banks create liquidity in the OTD model as well, since they continue to originate the loans that are subsequently sold or securitized. There are two key differences, however, from a liquidity creation perspective: • The first is that with the OTH model, the risk associated with liquidity creation is borne by the bank, whereas with the OTD model this risk is borne largely by the investors who purchase the loans or securities created by securitization. • The second difference is that funding for loans in the OTH model tends to come from (core) deposits, while funding for securitized structures in the OTD model typically comes eventually from repos and CP, even though the pre-securitization origination of the loan may have involved deposit funding. • Unlike core deposits, there is no deposit insurance backing repos and CP funding, so runs are possible. Such runs indeed occurred during the Subprime financial crisis in 2007-2008 Source: Bouwman (2015)

But securitization per se is not that evil after all Introduction • A growing number of studies on the US subprime market indicate that, due to asymmetric information, credit risk transfer activities have perverse effects on banks’ lending standards. • Nevertheless, securitization is means of aggregating loans to small firms, helping to facilitate their indirect access to the markets. For banks, this instrument is essential in disposing of part of their assets and freeing up funds for new loans. The success of such initiatives depends on a high level of transparency, enabling final investors to manage the associated credit risk more efficiently than in the case of direct investment in single issues. Simple structures, low leverage and limited maturity transformation are essential. • In the case of securitizations, there are advantages to be gained from Italy’s past experience, which has been extremely positive. Albertazziet al (2011) investigate a large part of the market for securitized assets (“prime mortgages”) in Italy, a country with a regulatory framework analogous to the one prevalent in Europe. Information on over a million mortgages consists of loan-level variables, characteristics of the originating bank and, most importantly, contractual features of the securitization deal, including the seniority structure of the ABSs issued by the Special Purpose Vehicle and the amount retained by the originator. They borrow a robust way to test for the effects of asymmetric information from the empirical contract theory literature. Overall, their evidence suggests that banks can effectively counter the negative effects of asymmetric information in the securitization market by selling less opaque loans, using signaling devices (i.e. retaining a share of the equity tranche of the ABSs issued by the SPV) and building up a reputation for not undermining their own lending standards. Source: Albertazzi et al (2011)

Differenttypes of banks: retailbanks, universalbanks and investment banks (1) Introduction • In Table 1 is reported a sample of banks divided in three sub-groups by degree of income diversification: • Mainly retail banks, • Mainly investment banks, • Truly universal banks. • The attribution to one of the three classes (retail, investment and universal banks) was calculated according to the following rule: retail banks have an average share of non-interest revenues below 40% of total revenues; investment banks have an average share of non-interest revenues beyond 60% of total revenues; while universal banks have an average share of non-interest revenues between 40% and 60% of total revenues. Average shares of non-interest revenues of total revenues are calculated over the whole time-sample in order to get rid of temporary effects on revenues • Retail banks still rely mainly on interest income, while for investment banks interest income plays a minor role, with the bulk coming from trading and commissions; finally, universal banks provide a high degree of diversification among the different sources of income. Source: Masciantonio and Tisenoa (2013)

Differenttypes of banks: retailbanks, universalbanks and investment banks (2) Introduction TABLE 1 Source: Masciantonio and Tisenoa (2013)

Differenttypes of banks: retailbanks, universalbanks and investment banks (3) Introduction • The deregulation of the banking sector that started in the 1980s has blurred the borders between traditional and investment banking, paving the way for the development of universal banks, which are involved in any kind of banking and financial activity. This process gained momentum from the end of the 1990s as a result of several factors: • liberalization of capital flows beginning in the previous decade, which allowed a rise in crossborder banking activities, • changes in accounting and regulatory standards, • technical innovation, • a higher degree of financial innovation. • The banking sector has thus undergone a significant and rapid transformation since the late 1990s, with several of its consequences still subject to investigation. The global financial crisis, which began in 2007, has proved to be a major breakpoint in universal banking activity and performance. Its impact and legacy are still unfolding and are worth analysing in detail. Moreover, the state of the banking sector is deeply intertwined with the state of the broader economy. Global banks enhance the international transmission of shocks though their activities, contributing to more integrated global business cycles. Source: Masciantonio and Tisenoa (2013)

Stylized facts concerning the sector’s transformation Introduction • Large banks have experienced more radical changes compared with small, savings or regional banks: • The former were big enough to compete in an increasingly globalized and integrated financial system, and had a more appropriate structure to react quickly to regulatory changes and financial innovation and compete against banks with different business models, • Smaller banks were less affected by this evolutionary path. Moreover, the largest banks typically feature among those defined systemic. • Given the massive degree of financial innovation during the last decade and its profound impact on banks’ activities, the label Large and Complex Financial Institutions (LCFI) looks particularly well-suited to these institutions. Source: Masciantonio and Tisenoa (2013)

Stylized facts in the run-up to crisis in the summer of 2007 Introduction • Sheer asset growth: banks’ balance sheets grew to enormous sizes compared with their home countries’ GDP. This was particularly true for banks in Europe, where the financial system is more bank-based. Balance-sheet growth was impressive in the runup to the global financial crisis, slipping back only slightly in the most recent years. Again, European banks provided the bulk of this growth during the last decade. • Leverage grew massively, fuelling asset growth and changing direction during the crisis years. The consolidation that occurred throughout the period under examination was another main factor behind asset growth. • The massive debt-fuelled asset growth underpinned the expansion of retail and universal banks’ activities in fields traditionally the preserve of investment banks, providing more revenues but also more risks. • Maturity mismatches and funding risk also grew in the run-up to the crisis. European banks, which are more reliant on wholesale funding than their North American counterparts, proved to be more exposed to these risks. More in general, the investment bank funding model, which relies heavily on wholesale funding, was intrinsically prone to these risks. • These developments led to an ever increasing profitability (as measured by ROE) during the boom years that followed the burst of the Dot-com Bubble. Source: Masciantonio and Tisenoa (2013)

Since 2007, when the crisis broke, things have changed swiftly: stylized facts for the crisis years (1) Introduction • Profitability dropped sharply and turned negative for several quarters, revealing all the fragilities embedded in the balance sheets of banks, which came out of the crisis severely weakened. The surge and subsequent fall of profitability was sharper for European banks. Profitability then recovered only to a fraction of the pre-crisis level in a completely changed financial landscape. • The losses experienced by banks emerged via three main channels. First, they arose from lending activity connected with the slump in the housing market, especially in the US, and the global recession in 2009. Second, trading activity – once the golden goose of universal banking – suffered enormous losses, affecting almost every bank. Third, an unprecedented level of extraordinary losses, connected with large writedowns and exceptional activities, was experienced. • The sovereign debt crisis severely affected European banks, which recovered at a slower pace than their North American counterparts. Source: Masciantonio and Tisenoa (2013)

Since 2007, when the crisis broke, things have changed swiftly: stylized facts for the crisis years (2) Introduction • Deleveraging is one of the main challenges banks have faced in the aftermath of the crisis. Undercapitalization – coupled with excessive asset growth – was identified as one of the banks’ main weaknesses. As a result, the bulk of regulatory efforts concentrated on strengthening capital bases. The increase in their capital bases by all the banks in the sample, together with a stand-still in asset growth, has been at the root of the massive deleveraging of recent years. • The dysfunctional interbank markets and the liquidity crunch encouraged banks to pile up increasing amounts of unallocated liquidity. However the sample is very heterogeneous: most of French, Italian and Spanish banks maintained a liquidity ratio broadly in line with its pre-crisis level, while some other banks (like English and Swiss banks) accounted for most of the increase in the average liquidity ratio. Another cause behind this surge in the liquidity ratio may be identified in regulatory changes. Source: Masciantonio and Tisenoa (2013)

The fall of universal banking? Introduction • Taken together, the above stylized facts have put the universal banking model under serious stress. Making the banking system safer might be inconsistent with the return to pre-crisis profitability levels. • The US and UK authorities have identified the excessive risk-taking embedded in trading activity as one of the main threats to banking stability and have proposed – in various forms – the separation of investment banking from retail banking. • This could even mark the end of the universal banking model as it evolved during the 2000s. Source: Masciantonio and Tisenoa (2013)

Differentbanks' business models and different performances Performance • Banks choose to be different from one another. They engage strategically in different intermediation activities and select their balance sheet structure to fit their business objectives. In a competitive pursuit of growth opportunities, banks choose a business model to leverage the strengths of their organisation. • We define and characterise banks' business models and examine their evolution during the financial crisis: • We identify a small set of key ratios that differentiate banks' business profiles and use a broader set of variables to provide a more complete characterisation of these profiles. • The second objective is to analyse the performance of these business models in terms of profitability and operating costs. • The final objective is to track how banks changed their business models before and after the recent crisis. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Identification of business models: threemaintypes Performance • We identify three business models: a retail-funded commercial bank, a wholesale-funded commercial bank and a capital markets-oriented bank. The first two models differ mainly in terms of banks' funding mix, while the third category stands out primarily because of banks' greater engagement in trading activities. On average, retail-focused commercial banks exhibit the least volatile earnings, while wholesale funded commercial banks are the most efficient. On the other hand, trading banks struggle to consistently outperform the other two business types. • Banks' profiles evolve over time in response to changes in the economic environment and to new rules and regulations. We find that transition patterns changed around the recent financial crisis. While several banks increased their reliance on wholesale funding prior to the crisis, in its wake more banks have adopted more traditional business profiles geared towards commercial banking. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Classifying banks: the methodology (1) Performance • The procedure we use to classify banks into distinct business models is based on Roengpitya, Tarashevand Tsatsaronis (2014) • The procedure is primarily driven by data but incorporates judgmental elements. It shares many technical aspects with the procedure employed by Ayadi and de Groen (2014), but differs in terms of the judgmental elements and the data used. In contrast to their analysis, which focuses exclusively on European banks, we use annual data for 222 individual banks from 34 countries, covering the period between 2005 and 2013. The unit of our analysis (i.e. a data point) is a bank in a given year (bank/year pair). Given that the available data do not cover the entire period for each bank, we work with 1,299 bank/year observations. By focusing on bank/year pairs our approach allows institutions to switch between business models at any point in the period of analysis. • The inputs to the classification are bank characteristics. These are balance sheet ratios, which we interpret as reflecting strategic management choices. We use eight ratios expressed in terms of balance sheet size and evenly split between assets and liabilities. They relate to the share of loans, traded securities, deposits and wholesale debt, as well as the interbank activity of the firm. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Classifying banks: the methodology (2) Performance • We distinguish this set of variables from other variables that we use to characterise the performance of different business models. We view these other variables, which capture profitability, income composition, leverage and cost efficiency, as reflecting the interaction between banks' strategic choices and the market environment. We thus treat them as variables that relate to outcomes as opposed to choices. • The core of the methodology is a statistical clustering algorithm. Based on a pre-specified set of input variables, the algorithm partitions the 1,299 bank/year observations into distinct groups. We select inputs from the set of choice variables. The idea is that banks with similar business model strategies have made similar choices regarding the composition of their assets and liabilities. We make no a priori decisions as to which choice variables are more important in defining business models or as to the general profile of these models. In that sense, the methodology is data-driven. We rely on the repeated use of the clustering algorithm and a goodness-of-fit metric to guide the selection of the most appropriate partitioning of the observations universe into a small number of distinct business model groups. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Three distinct business models: the characteristics that matter (1) Performance • The classification process identifies three distinct business models and selects three ratios as the key differentiating choice variables: the share of loans, the share of non-deposit debt and the share of interbank liabilities to total assets (net of derivatives exposures). This partition satisfies our criteria of robustness, parsimony and stability. The share of gross loans is the only variable relating to the composition of the banks' assets. The other two ratios differentiate banks in terms of their funding structure. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Three distinct business models: the characteristics that matter (2) Performance • Table 2 characterises the three business model profiles in terms of all eight choice variables (rows). The cells report the average ratio for all banks that were classified in the corresponding business model (columns). For comparison, the last column provides the average value of the corresponding ratio for the universe of observations. • The first business model group we label commercial "retail-funded", and it is characterised by a high share of loans on the balance sheet and high reliance on stable funding sources including deposits. In fact, customer deposits are about two thirds of the overall liabilities of the average bank in this group. This is the largest group in our universe with 737 bank/year observations over the entire period. • The second business model group we label commercial "wholesale-funded". The average bank in this group has an asset profile that is remarkably similar to the profile of the retail funded banks in the first group. The main differences between the two relate to the funding mix. Wholesale-funded banks have a higher share of interbank liabilities (13.8% versus 7.8%) and a much higher share of wholesale debt (36.7% versus 10.8%), with the balance being a lower reliance on customer deposits (35.6% versus 66.7%). There are half as many observations in the wholesale-funded group compared to the retail-funded group. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Three distinct business models: the characteristics that matter (3) Performance • The third group is more capital markets-oriented. Banks in this category hold half of their assets in the form of tradable securities and are predominately funded in wholesale markets. In fact, the average bank in this group is most active in the interbank market, with related assets and liabilities accounting for about one fifth of the balance sheet. We label this business model "trading bank". It is the smallest group in terms of observations (203 bank/years) in our sample. • By comparison, Ayadi and de Groen (2014) classify European banks into four business models, which they label as investment banks, wholesale banks, diversified retail and focused retail. Drawing rough parallels with the present classification, which involves a more global universe of banks, their investment bank model corresponds to our trading model, the two wholesale models correspond to each other, and the diversified and focused retail models together correspond to our retail-funded model. That said, an exact comparison would require comparing individual banks in the two universes. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Three distinct business models: the characteristics that matter (4) Performance • We find that the popularity of business models differs with banks' nationality (Table 2). Looking only at the last year of our data (2013), the North American banks in our universe had either a retail-funded or trading profile; none belonged to the wholesale-funded group. At the same time, one third of the European banks had a wholesale-funded model. In turn, banks domiciled in emerging market economies (EMEs) clearly preferred the retail-funded model (90%). • We also look at the distribution of global systemically important banks (G-SIBs) across business models (Table 3). Our data for 2013 cover 28 firms that were part of the banking organisations designated as G-SIBs by international policymakers (Financial Stability Board (2014)). The list - which includes institutions from both advanced and emerging market economies - was roughly equally split between the retail-funded and trading models. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Three distinct business models: the characteristics that matter (5) Performance TABLE 2 Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Three distinct business models: the characteristics that matter (6) Performance TABLE 3 Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Business models and bank performance (1) Performance • Are there systematic differences in the performance of banks with different business models? The question is pertinent for understanding the impact of banks' choices on shareholder value but also on financial stability, which depends on sustainable performance by financial intermediaries. In this section we examine the performance of banks in the different business model categories both in a cross section and over time. • In analysing the performance of different bank models, we use what we label "outcome" variables. In contrast to the choice variables that we used to define the business models, we interpret outcome variables as the result of the interaction between the strategic choices made by the bank in terms of business area focus and the market environment. Examples of such variables are indicators of profitability, (for example, banks' return-on-equity, RoE), the composition of bank earnings (for instance, the share of interest income in total income) and indicators of efficiency (for example, the cost-to-income ratio). • Profitability and efficiency have varied markedly across models as well as over time (Figure 3). The outbreak of the recent crisis marked a steep drop in advanced economy banks' RoE across all business models (Figure 3, left-hand panel). But while RoEstabilised for retail banks after 2009, it remained volatile for trading and wholesale-funded banks. In fact, trading banks as a group show the highest volatility of RoE across the three groups, swinging repeatedly between the top and bottom of the relative ranking. The story is qualitatively similar in terms of return-on-assets (RoA, not reported here), an alternative metric of profitability that is insensitive to leverage (see also Table 3). Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Business models and bank performance (2) Performance • All three business models show relatively stable costs in relation to income (Figure 3, centre panel). A spike in the cost-to-income ratio around 2008 is readily explained by the drop in earnings in the midst of the crisis. Compared to the other two business models, trading banks had a persistently high cost base throughout the period of analysis, despite their more mixed record in terms of profitability. Interestingly, high costs relative to income have persisted post-crisis despite the decline in these banks' profitability. A possible explanation can be found in staff remuneration rates, although this would be difficult to decipher from our data. • Post-crisis markets appear rather sceptical about the prospects of all three business models, judging from the price-to-book ratio of banks in advanced economies (Figure 3, right-hand panel). This ratio relates the banks' stock market capitalisation to the equity they report in their financial accounts. A value higher than unity suggests that the equity market has a more positive view on the franchise value of the bank than what is recorded on the basis of accounting rules. A value below unity suggests the opposite. The ratio declined dramatically around the crisis for banks in all three business models. In fact, it has been persistently below unity since 2009 for most advanced economy banks, reflecting market scepticism about their prospects. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Business models and bank performance (3) Performance • Banks domiciled in EMEs (dashed lines in Figure 3) remained largely unscathed by the 2007-09 crisis. These lenders are almost exclusively classified in the retail-funded model. But even compared to their advanced economy peers with a similar business model, they achieved a more stable performance. And while a more favourable macroeconomic environment has certainly contributed to their higher profitability in recent years, the overall stability of their performance is underpinned by greater cost efficiency, ie a lower cost-to-income ratio. In line with these results, market valuations are quite generous for EME banks with price-to-book ratios persistently higher than unity, although they are on a declining trend. • Table 4 compares the three business models in terms of a number of other outcome variables across the entire sample period. Besides RoA and RoE, which confirm the ranking from Figure 3, we also calculate risk-adjusted versions of these profitability statistics, which subtract from the earnings variable (the numerator of the ratio) the cost of capital that is necessary to cover for the risk inherent to the activity of the bank. The approach follows closely the rationale of standard industry approaches to calculate the risk-adjusted return on capital (or RAROC). More specifically, we subtract from the bank's gross earnings the associated operational expenses and losses (including credit losses and provisions) as well as the cost of capital set aside to cover possible future losses. This last component is the product of the quantity of capital held by the bank (proxied by the regulatory capital requirement linked to risk-weighted assets) multiplied by the cost of equity capital (estimated by a standard capital asset pricing model). Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Business models and bank performance (4) Performance • Regardless of the profitability metric, the retail-funded model is the top performer. This is true in almost every year in our sample (not reported here). The top performance of retail-funded banks is consistent with the findings in Altunbas et al (2011), who document that banks with a greater share of deposits in their funding mix fared significantly better in the crisis than their peers. In particular, they exploit the 2007-2009 financial crisis to analyze how risk relates to bank business models. They find that banks with higher risk exposure had less capital, larger size, greater reliance on short-term market funding, and aggressive credit growth. Business models related to significantly reduced bank risk were characterized by a strong deposit base and greater income diversification. • Trading banks come in second place, with the exception of the risk-adjusted RoE, which penalises the volatility of their earnings base. Trading banks differ very significantly from their commercial bank peers in terms of the source of revenue. They collect about 44% of their total profit through fees, a share that is almost double that of the average other bank. • Wholesale-funded banks have the thinnest capital buffers among the three business models, while they also have the lowest cost of equity. Somewhat surprisingly, trading banks do not seem to be too different from retail-funded banks in terms of these yardsticks. However, they do stand out in terms of total asset size. The average trading bank is more than twice as large as the average commercial bank, even those that are primarily funded in the wholesale markets. Source: Altunbas et al (2011); Roengpitya, Tarashev and Tsatsaronis (2014)

Business models and bank performance (5) Performance FIGURE 3 Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Business models and bank performance (6) Performance TABLE 4 Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Shifting popularity of bank business models (1) Performance • The crisis-driven reshaping of the banking sector has affected its concentration and business model mix. A number of institutions failed or were absorbed by others, thus increasing the concentration in the sector despite tighter regulatory constraints on banks with a large systemic footprint. And many of the surviving banks adjusted their strategies in line with the business models' relative performance. • Table 5 presents a summary of banks' shifts across different business models before and after the crisis. Each cell reports the number of banks that started the period in the model identified by the row heading and finished it in the model named in the column heading. The large numbers along the diagonal indicate that there is considerable persistence in the classification of banks, as the majority of institutions remain in the same business model group over time. • In recent years, most of the transitions have been between the retail- and wholesale-funded models of commercial banks. The group of trading-oriented banks is fairly constant throughout the period. The direction of change in bank business models, however, is very different post-crisis from that prevailing prior to 2007. During the boom period, market forces favored wholesale funding, as bankers tapped debt and interbank market sources of finance. About one in six retail banks in our 2005 universe increased their capital market funding share to the point that they could be reclassified as wholesale-funded by 2007 (first row of Table 5). Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Shifting popularity of bank business models (2) Performance • The opposite trend characterizes the post-crisis period. About two fifths of the banks that entered the crisis in 2007 as wholesale-funded or trading banks (ie 19 out of 50 institutions) ended up with a retail-funded business model in 2013. Meanwhile, only one bank switched from retail-funded to another business model post-crisis, confirming the relative appeal of stable income and funding sources. • While we observe transformations of banks in ways that result in their reclassification under a different business model, we cannot pinpoint the underlying economic drivers. We can, however, look at performance statistics to examine whether bank shifts correlate with a turnaround of the firm. We find that a change in bank business model actually hurts profitability, but improves efficiency relative to the firm's peer group. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Shifting popularity of bank business models (3) Performance • To do this, we select all the banks in our sample that switched models and for which we have data for at least two years before and two years after the switch. We focus on two performance ratios: RoE and cost-to-income. We benchmark the performance of the bank against a comparator group that comprises all banks that belonged to the same business model as the switching bank prior to the switch and remained in that model. We determine that the switching bank outperformed its old peers if the difference between its average post-switch and average pre-switch RoE is greater than the difference between the corresponding averages in the comparator group. On the basis of this criterion, we find that only a third of the banks that switched their business model outperformed their old peers in terms of profitability. The remaining two thirds underperformed. However, applying the same criterion to the cost-to-income ratio reveals that, among the banks that switched business model, two thirds registered post-switch efficiency gains relative to their peers. • In conclusion, we identified bank business models that have had different experiences over the past decade. Given the consistently stable performance of retail-funded banks engaging in traditional activities, it comes as little surprise that their model has recently gained in popularity. More surprising is the stability of the group of trading banks, which exhibited sub-par return-on-equity over most of the sample, both in absolute and risk-adjusted terms. While further analysis is needed to uncover the clear benefits to these banks' shareholders, high cost-to-income ratios suggest outsize benefits to their managers. Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Shifting popularity of bank business models (4) Performance TABLE 5 Source: Roengpitya, Tarashev and Tsatsaronis (2014)

Alternative identification of banks' business models (1) Risk assessment • In Ayadi and de Groen (2014), by using a sample of European banks spanning from 2006 to 2013, the clustering analysis identified 4 models as the most distinct form of clustering. • Model 1 groups together large investment-oriented banks and contains the largest banks, both in terms of total and average assets. The average size of a bank in this cluster was approximately € 583 billion in 2013, about quadruple for an average wholesale or focused retail bank and almost double the amount of a diversified retail bank. In what follows, Model 1 will be referred to as the cluster of ‘investment banks’. As is clear from the name, these banks have substantial trading activities. The cluster averages for trading assets and derivative exposures - representing 51.2% and 15.2% of total assets, respectively - stand between 1.3 and 1.5 standard deviations above the relevant sample means. In funding, the focus is on less stable and less traditional sources, such as debt liabilities and, more importantly, repurchase agreements, which came under severe stress during the financial crisis. The investment banks also tend to be highly leveraged, with an average tangible common equity ratio of 3.9%. Source: Ayadi and de Groen (2014)

Alternative identification of banks' business models (2) Risk assessment • Model 2 includes banks with a heavy reliance on interbank funding and lending. The liabilities of an average bank under this bank model to other banks, including both deposits and other interbank debt, represent, on average, 37.4% of the total balance sheet, towering above the interbank liabilities of other bank models. In turn, traditional customer deposits represent only 16.0% of the total balance sheet, the lowest among the four groups. Other funding sources come from debt liabilities, which exclude traditional deposits and interbank funding. The Model 2 banks, which will henceforth be referred to as ‘wholesale’, are also very active in non-traditional uses of these funds, including trading assets (i.e. all assets excluding cash, loans and intangible assets). On average, trading assets account for 28.1% of their balance sheets and interbank lending represents 38.4% of total assets. These banks are substantially less leveraged than their peers, with the highest tangible common equity ratio of 5.9% among the four clusters studied. The total size of the wholesale banking group, which is the smallest group, has declined over time, partly as a result of shrinking average sizes in the midst of the financial crisis in 2008 and partially due to a migration to other business models. Lastly, the expenditures on staff are the lowest in the wholesale banking group, with median personnel expenditures remaining at €3.0 per €1,000 of assets, less than half of the sample median. Source: Ayadi and de Groen (2014)

Alternative identification of banks' business models (3) Risk assessment • Model 3 is composed of retail-oriented banks, which use relatively non-traditional funding sources. Hence, customer loans and debt liabilities account for 48.0% and 67.5% of the total balance sheet on average, surpassing the sample averages. The greater diversification of funding sources is most likely an attempt to maintain a larger size. In line with this description, the Model 3 banks have, after a hic-up in 2009, continued to expand during the crisis, implying that the reliance on multiple sources of financing has reinforced the group’s growth prospects. Source: Ayadi and de Groen (2014)

Alternative identification of banks' business models (4) Risk assessment • Model 4 shares several similarities with Model 3. First, and foremost, the group is comprised of retail-oriented banks, with traditional customer loans representing on average 60% or more of the balance sheet totals in both groups. Moreover, the ratio of cash and cash-like liquid assets remains above the sample average. Models 3 and 4 also spend about twice as much as investment and wholesale banks on staff, with median personnel expenditures at €7.5 and €8.8 per €1,000 of assets, respectively. The higher staff costs may possibly reflect a larger geographical coverage through a larger number of branches and personnel. However, the two models do differ in funding sources. While the Model 3 banks have a greater reliance on debt markets, Model 4 banks rely primarily on customer deposits. The average size of predominantly focused retail banks under Model 4, as measured by average total assets, tends to be around half of the sample average, the smallest banks in the sample. The quest for a larger size of Model 3 banks is also expressed in higher leverage ratios; the average tangible common equity ratios are 4.7 and 5.5 for Model 3 and 4 banks, respectively. In order to distinguish between the two retail-oriented groups, models 3 and 4 will be referred to as the ‘diversified retail’ and ‘focused retail’ models, respectively. Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (1) Risk assessment • The retail-diversified banks rely more on stable forms of funding and limit risky investments, while wholesale and investment banks tend to be better at resisting default risks. The diversified retail banking model does well under most measures, with low default risks, a level of capitalisation close to the sample median, and moderate liquidity risks. The focused retail banks face the highest default risks, although these risks appear to be shielded by relatively strong capital levels and limited liquidity mismatch risks. Italian banks belong mainly to the diversified retail and focused retail models. • The first indicator (Figure 4), Z-score, provides an estimate of a bank’s distance to default. In essence, the risk measure uses historical earnings volatility and returns as well as current capital levels to construct the level of a (one-time) shock beyond the historical average that would lead to default. The greater the Z-score, the less probable is the likelihood of a default. The diversified retail banks appear safer, with a higher distance to default and a high level of net stable funding. The distribution of the Z-scores for diversified retail banks are significantly different from wholesale and retail-focused banks. In turn, focused retail banks have effectively lower Z-scores, implying the highest risks. Figure 4 shows that the differences in median Z-scores across business models have primarily been created in the most recent years. Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (2) Risk assessment • The second indicator (Figure 5), the ratio of risk-weighted assets (RWA) to total assets, or the average risk-weights, provides a regulatory measure of risk. Banks with higher RWA are expected to be more sensitive to risks and are thus required to hold more regulatory capital to account for their risk-weighted balance sheet26, without counting the risk pertaining to the off-balance sheet. According to the statistical analysis, both investment and wholesale banks appear to be less risky, with distinct median risk weights of 29% and 38% respectively, which is substantially lower than the risk weights of the retail-oriented banks (between 53% and 57%). The finding that wholesale banks have less exposure to risks in their assets is intriguing and clearly inconsistent with the Z-score findings, which indicates higher default risks than diversified retail banks. Moreover, Figure 5 shows that the median level of risk weighted assets across all business models has gradually been declining. The largest change was observed in diversified retail banks, which decreased the median risk weight from above the focused retail banks in 2007 to a level close to or below wholesale banks in 2011. Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (3) Risk assessment • The third indicator (Figure 6), risk-costs as a share of non-trading assets, is a proxy-measure for the credit losses. Since there is no harmonised definition of credit losses reported and the separation between credit losses and losses on other types of assets is often opaque, the measure also includes other non-trading assets (besides loans to banks and other customers). The results displayed in Figure 6 show that the pre-crisis risk-costs of investment and wholesale banks were substantially lower than those of retail banks. During the financial crisis of 2008 and 2009, all business models posted higher risk-costs. Afterwards, during the economic crisis, the credit losses of most business models dropped, with the exception of focused retail. The difference might be explained by a difference in the composition of the credit portfolio. The wholesale and, to a lesser extent, investment banks have relatively more credit outstanding to banks compared to other customers. Notwithstanding some high-profile cases like the collapse of Lehman Brothers, the losses on loans to banks have historically been lower than on loans to other customers. Even during the crisis, the banks were largely shielded from barring losses on loans to banks, primarily due to the various government- and central bank-interventions that prevented banks from going bankrupt and limited the burden sharing to equity holders and junior debt holders. Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (4) Risk assessment • The fourth indicator (Figure 7), the median CDS spreads for senior securities, displays a significant higher CDS spread for the small and least financially integrated focused retail banks than all other banking business models. The difference between the investment-, wholesale-, and diversified retail banks is not significant, implying that the underlying distributions may be identical. The market participants do not appear to distinguish among these three models in terms of their inherent risks. Provided that other indicators do find substantial differences for the underlying risks, it is likely that the market participants have already factored in the likelihood of government interventions, resulting in the comparability of the markets’ perception of default risks. Once again, these findings give support to the elevation of moral hazard risks due to the dilution of market discipline in the eventuality of bank bail-outs or state guarantees. • The fifth indicator measures the loss-absorption capacity of banks under the Basel capital rules (Figure 8). For any given level of risk, holding more capital could imply greater stability. The results show that Tier 1 ratios have been gradually increasing since 2007. However, the ratios are statistically almost indistinguishable among the four banking groups, implying a more or less identical absorption capacity. Only the Tier 1 ratio of diversified retail banks is significantly lower than that of investment- and retail-focused banks. The fact that the differences in risk and absorption capacity are not reflected in the Tier 1 ratios is intriguing and invites the possibility that the main regulatory instrument currently in use may not be adequate for capturing (or signalling) the loss-absorption capacity of a bank. Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (5) Risk assessment • The sixth indicator measures the loss-absorption capacity using a simple leverage ratio (Figure 9). The tangible-common equity ratios are statistically distinct for all business models. Retail banks hold substantially more tangible common equity, which made them able to absorb more losses (at least for the period of observation under investigation). Moreover, the results suggest that wholesale banks can absorb significantly more losses than investment banks. Figure 9 shows that banks across all business models, except for focused retail banks, have increased the tangible-common equity ratios. • The seventh indicator (Figure 10), the net stable funding ratio (NSFR), is an estimate of the proposed long-term liquidity risk measure proposed under the Basel III rules. Expressed simply, the measure gives an estimate of the available stable funding sources as a share of required stable funding, which is constructed with available data. Although the measure should be interpreted with caution, a greater value should point to lower liquidity risks. Figure 10 shows that the retail oriented banking models face relatively lower liquidity risks, while wholesale and investment banks may face higher risks. It is important to note that no single model satisfies the 100% funding requirement, as proposed under Basel III. Moreover, Figure 10 shows that liquidity conditions have gradually worsened for most models up to 2012, particularly for the investment and focused retail banks that took severe liquidity hits in 2008. Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (6) Risk assessment • An alternative assessment of default risks follows the ‘top-down’ approach to calibrating regulatory minimum capital requirements under stress conditions. More specifically, the quantiles of the return to risk-weighted assets (RoRWA) are used to construct expected losses that banks may face under a stress scenario. If the most loss-absorbing parts of equity (i.e. the tangible common capital ratio) remain below or close to such a measure, then the likelihood of a default would be equally higher under those stress conditions. • A comparison of the mean values for RoRWA, reported in Figure 11, shows that the distinctions between clusters are only clear for diversified retail banks when tested using Wilcoxon-Mann-Whitney non-parametric two-sample tests. The results for all years show that the diversified retail banks, on average, reported distinctly higher RoRWAs than banks belonging to one of the three other models. Although the same is true for all crises years (2008-12), the results for the financial and Eurozone crises on a stand-alone basis are only partially significant. On average, both types of retail banks reported significantly higher returns than wholesale banks during the 2008-09 financial crisis. During the 2010-12 Eurozone crisis, the returns of both diversified retail banks were distinct from the returns of investment and focused retail banks. • The findings show clear distinctions across business models in terms of riskiness which suggests that the average risk weights are not a good indicator of the underlying risks. In particular, wholesale banks and focused retail banks faced severe default risks during the financial and economic crises. Nevertheless, these differences appear in the underlying risks, not in the average risk weights. Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (7) Risk assessment • Table 6 provides the results of censored regressions to assess whether the average risk weights explain distance from default (Z-Score). To be a good regulatory risk measure, there should be a strong relation between the risk weighted assets and the underlying risk. Notwithstanding differences in capital levels, the relationship between Z-score and RWA to assets should be negative, which implies that banks with a higher RWA are closer to default. • The estimation results for the pooled sample as well as for retail banks show a persistent significant negative relation between the regulatory risk measure and distance to default. In turn, the results for wholesale banks show a significant positive relation, which implies that RWA are inversely related to underlying risks. The estimates for investment banks are also positive but insignificant at the 10% level. The relationship becomes stronger and more negative when capital is controlled, except for wholesale banks. This implies that banks with greater RWA are holding more capital, which can partly offset their lower risk profile. • Overall, RWA does appear to be able to capture the underlying risks for the two retail business models. In turn, it fails to do so for wholesale and investment banks. The relationship between the two measures of risk is ambiguous for investment banks and the reverse for wholesale banks, even after controlling for capital levels. The findings suggest that the risk weighted assets of wholesale and investment banks are not well calibrated. Hence, this implies that the risk weights of certain assets or activities conducted primarily by wholesale and investment banks are incorrect. The wholesale and investment banks, for example, engage more in trading activities. Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (8) Risk assessment • One explanation for the finding (see Figure 11 and Table 6) that regulatory measures appear to be misaligned with underlying risks is the possibility that greater risk-weights are associated with more capital. If banks with greater RWA also hold more capital, partly to fulfil the binding regulatory requirements, they may face lower default risks, possibly explaining the distorted relationship. • An alternative explanation is that banks may be engaging in ’risk optimisation’ to reduce their risk-weights (and the implied capital charges) without shedding any risks. Indeed, despite sound arguments for making capital requirements risk-sensitive, the complexity and flexibility of these rules have led to concerns over the potential for regulatory arbitrage. Since raising capital is not always possible during the crisis periods, some banks choose to respond to regulatory shortfalls by decreasing their risk-weighted assets. This can be done through deleveraging or changing the calibration of the risk-weights (i.e. changing from standard to internal models with lower average ratios or changing the internal models) or by changing the composition of the assets to assets with lower risk-weights. There is a concern among researchers, supervisors and policy makers about the usage of internal models, which implies that the risk-weights and thus capital requirements are reduced without reducing the underlying risks (i.e. regulatory arbitrage). • The ECB plans to inspect the internal models that banks use to calculate their risks to ensure that these systems behave consistently. Regulators have had internal models in their sights in the wake of the financial crisis due to concerns some banks may attempt to downplay the riskiness of their assets and hold less capital than they should. The ECB, which took over as the euro zone's leading financial regulator in November 2014, aims to compare these models using regional data and look for inconsistencies. The ECB's attention to these models is part of a wider push to impose order on the sector, which could lead to changes in the way banks do business. In fact, the ECB is also charged with regularly checking that banks have viable business models, something that has not been done consistently in all the euro zone states. The central bank would be intrusive when it checks banks' business models. Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (9) Risk assessment FIGURE 4 Source: Ayadi and de Groen (2014)

Risk attributes of banks' business models (10) Risk assessment FIGURE 5 Source: Ayadi and de Groen (2014)

G roup I nstitutional & R egulatory A ffairs