Regulation of G-SIBs. Does one size fit all? Working paper December 2013

Transcription

1 Małgorzata Iwanicz-Drozdowska Institute of Finance, Iwona Schab Institute of Statistics and Demography, Warsaw School of Economics Regulation of G-SIBs. Does one size fit all? Working paper December 2013 Abstract The ongoing global financial crisis has enhanced the interest in effective regulation as well as the supervision of internationally active and large financial institutions, named as global systematically important financial institutions or banks (G-SIFIs or G-SIBs). In this paper, we address two questions. Firstly, is additional capital enough to preserve the safety and soundness of G-SIBs? Secondly, is the quite uniform methodology adequate enough to regulate a diverse group of institutions? Following these questions and data analysis, we try to point out the weaknesses of BCBS/FSB approach and suggest the possible corrections. We use a panel data set covering G-SIBs and the time range of that demonstrates remarkable differences among the analyzed institutions. We find e.g. that the relationships between risk and profitability differ strongly between globally and not so globally acting G-SIBs, meaning that the one size fits all solution is likely to be improper. Key words: G-SIB, regulation, bank capital, risk, JEL classification: G21, G15 1. Introduction Although the issue of too big to fail banks (TBTF) is not new in the literature, the problem of G-SIBs regulation and supervision still generates more questions and doubts than answers. They are also called: too-big, too-complex and too-interconnected-to-fail, which properly reflects the problem faced by regulators and supervisors. For the first time regulators following the conclusions of the G-20 summit in Seoul in October 2010 decided to introduce additional requirements for large international financial

2 institutions as additional loss absorbency capacity [BCBS 2011] or in other words a higher level of common equity tier 1 solvency ratio, introduced by the Basel 3 approach. One should keep in mind that the ongoing global financial crisis forced many banks and other financial institutions to increase capital in order to cover the losses and/or to improve their quality (e.g. common equity capital instead of hybrid instruments) which was reflected also in a higher level of the solvency ratio. The higher the solvency ratio, the lower the return on equity which is strictly analyzed by investors. It is important in case of new capital injection into the banking sector. In November 2011 the Financial Stability Board published for the first time the list of 29 G- SIBs 1, which were selected with the use of specific criteria set up by the Basel Committee on Banking Supervision [BCBS 2011]. The results of the assessment as well as the refreshed list of G-SIBs were published in November 2012 and In November 2012 the number of G- SIBs was reduced to 28, 3 banks fell from the list (Dexia restructuring process, Lloyds and Commerzbank) and 2 new G-SIBs were added (Standard Chartered UK and BBVA - Spain). In November 2013 one G-SIB was added to the list (Industrial and Commercial Bank of China). The literature on G-SIBs is scarce. Dewenter and Hess [2013] compared the characteristics of bank designated as G-SIBs against other banks from the same countries as well as investors reactions and further financial performance. They focused on stocks exchange and financial statements data analysis. Conclusions indicated that investors reactions very negative, G-SIBs financial performance deteriorated and stock returns became for correlated. Barth et al. [2013] reviewed policies for G-SIBs and D-SIBs (domestic SIBs) around the globe for 135 countries. Other authors did not differentiate between G-SIBs and D-SIBs. Castro and Ferrari [2014] applied Adrian and Brunnermeier [2011] CoVaR approach to test systematical importance of 26 large European banks with the conclusion that very few banks may be recognized as systematically important with the use of that approach. Ueda and di Mauro [2013] analyzed worldwide data sample to evaluate subsidy values deriving from the government support (bail-out), reflected in rating and thus costs of financing. Tabak et al. 1 Barclays (UK), Commerzbank (Germany), Bank of America (USA), Bank of China (China), Bank of New York Mellon (USA), ING Bank (Netherlands), BNP Paribas (France), Citicorp (USA), Credit Suisse (Switzerland), Deutsche Bank (Germany), Dexia (France/Belgium), Goldman Sachs (USA), Credit Agricole (France), HSBC (UK), JP Morgan Chase (USA), UBS (Switzerland), Lloyds (UK), Nordea (Sweden), Mizuho (Japan), Mitsubishi (Japan), Morgan Stanley (USA), RBS (UK), Societe Generale (France), Santander (Spain), State Street (USA), Sumitomo Mitsui (Japan), Unicredit (Italy), Wells Fargo (USA), BCPE (France).

3 [2013] analyzed 17 SIBs operating in Latin America concluding that they did not take more risk, however they outperformed other banks in terms of profitability. Our research relates to aspects that has been not yet taken under consideration with reference to G-SIBs. In this paper, we address two questions. Firstly, is additional capital enough to preserve the safety and soundness of G-SIBs? Secondly, is the quite uniform methodology adequate enough to regulate a diverse group of institutions? We analyze the approach undertaken for the G-SIBs regulation in the context of theoretical and empirical literature as well as the analysis of G-SIBs financial data collected from annual financial statements for In section 2 we review the relationship between banks capital and their safety based on the literature and some data analysis. Section 3 describes BCBS methodology for additional loss absorbency. In the next section we explain the data and methodology used to identify similarities and differences among G-SIBs. Section 5 formulates proposals for changes and concludes. 2. Capital and banks safety The Basel Committee on Banking Supervision plays an important role in setting standards for bank regulation and supervision. In 2004 it presented the final version of Basel 2 capital framework which entered into force in most industrialized (except the US) and emerging economies in The outbreak of the global financial crisis questioned the Basel 2 solutions, especially the quality of tier I capital, the way in which VaR models has been used (model risk), lack of liquidity and leverage regulations and securitization. In the wake of criticism, the BCBS introduced necessary changes called Basel 2.5 (2009) and presented the Basel 3 framework (2010, will be implemented progressively from 2013 to 2019) in order to reduce acute imperfections in banking sector regulations. Quantitative impact studies (QISs) conducted by the BCBS and the European Commission proved that big international banks are not ready to comply quickly with the Basel 3 regulations, especially in the case of high quality capital (common equity tier 1 and capital buffers) and liquidity (BCBS 2010, CEBS 2010). Therefore G-SIBs like other large international banks are required to improve their capital and liquidity, but additionally, after the high level political pressure, they shall comply with other new requirements (like additional loss absorbency, living wills and resolution regime), which are supposed to curtail the excessive risk taking and moral hazard. All those regulations are supposed to improve safety and soundness of the global banking system.

4 Would this work like this with no doubt? The answer to this question in the literature is rather ambiguous. In case of relation between bank risk and capital we shall distinguish two streams of research: (1) changes in risk exposure and (2) changes in banks default risk or assets risk (table 1). From the practical point of view financial data on risk exposure are presented in the financial statements, and are easily available, but it is more difficult to measure inherent assets risk and thus default risk.

5 Table 1. Review of the literature on the relation between capital and risk Author(-s) R. Merton (1977) W. Sharp (1978) M. Keeley, F. Furlong (1990) Type of analysis and population Theoretical Risk Indirect answer: The higher the level of debt to assets, the exposure higher the cost of deposit guarantee (insurance); in order and default to maximize value of deposit insurance option, banks are risk increasing leverage and assets risk Theoretical Default risk Indirect answer: liability of deposit insurer may be reduced by new capital injection; not related to risk taking behavior Theoretical Default risk Direct answer: no, for insured, value-maximizing banks what was opposite to utility maximization models which assumed that higher capital increases probability of bank failure Stream Does more capital increase risk? Comments Option pricing theory is used to model the cost of deposit insurance Analysis from deposit insurance perspective Strong criticism of utility maximization models due to improper assumptions of: no bankruptcy, no deposit insurance and no risk measures (Kahane 1977, Koehn, Santomero 1980 called KKS model). KKS model referred to risk exposure and default risk G. Gennotte, D. Pyle (1991) R. Shrieves, D. Dahl (1992) Theoretical Default risk Direct answer: an increase in capital reduces bank s default risk if the elasticity ratio of loan portfolio risk decreases, otherwise bank s default risk increases Empirical, 1800 FDIC insured US commercial banks with Risk exposure Direct answer: banks reduce the effect of increased capital by increasing risk exposure; in the case of wellcapitalized banks positive relation between risk and capital shall be explained rather by private incentives of owners and/or managers than by regulations. Authors Authors assumed that bank managers are seeking to maximize the market value of equity. Analysis included also deposit insurance perspective (subsidy) and imperfect regulatory control. The changes in risk and capital levels are determined by both endogenous and exogenous factors

6 K. Jacques, P. Nigro (1997) R. Aggarwal, K. Jacques (2001) B.Rime (2001) F. Gonzales (2005) T. Jokipii, A. Milne (2011) assets higher than 100 M USD Empirical, US banks Empirical, US banks Empirical, Swiss banks Empirical, 251 banks in 36 countries Empirical, US bank holding companies and commercial banks Source: own elaboration Risk exposure Risk exposure Risk exposure Risk exposure Risk exposure analyzed also assets risk taking into account nonperforming loans, but conclusions present impact of macroeconomic factors, rather than the level of capital. Direct answer: risk based standards increased capital ratios and reduced risk. Risk is measured as relation of risk weighted assets to assets. Direct answer: increase in capital and capital ratios did not cause higher credit risk. Credit risk is measured as relation of risk weighted assets to assets. Direct answer: increase in capital did not cause higher risk Direct answer: stricter regulations cause lower charter value (measured by Tobin s Q) and do not provide incentives to reduce risk. Higher charter values (in countries with less regulation) stimulate bank to take risk prudently. Risk is measured as a share of non-performing loans (NPL) for credit risk and the standard deviation of daily bank stock returns for each year for overall risk Direct answer: well capitalized banks maintain capital buffers by increasing risk if capital increases; banks with small buffers increase capital and reduce the risk in order to improve capital buffers. Risk is measured by risk weighted assets to total assets, NPL and share of commercial and industrial loans in loan portfolio. Analysis of the implementation of risk based capital standards; modified Shrieves & Dahl model (1992) in order to incorporate risk based capital standards Prompt corrective action was regarded as a positive factor for bank capital and risk taking; modified Shrieves & Dahl model (1992) Modified Shrieves & Dahl model (1992) Analysis embraced charter value, risk-taking, deposit insurance, quality of contracting environment, and the legal origin of the country Positive and two-way relationship between capital buffers and risk

7 As presented in table 1, the research focused on the risk exposure prevails, since data allowing to calculate ratios or other measures of risk exposures are usually available. As the review of the literature indicates, there are two common measures of risk, i.e. relation of risk-weighted assets to assets and share of non-performing loans (NPL). The first risk measure shall be regarded as quite uniform across countries because of the BCBS methodology, but in case of NPL the accounting standards differ between US (US-GAAP), Europe (IAS/IFRS) and Japan (JP-GAAP). Additionally, not all banks or capital groups decide to publish data on nonperforming loans. Conclusions from the theoretical research, focused on the default risk, shows that new capital reduces the risk of bank s default, especially in case of value-maximizing banks [Keeley, Furlong 1990; Gennotte, Pyle 1991]. Value maximizing banks are those, who have long-term perspective for profit generation, what discourages them from excessive risk-taking, deriving mostly from loan portfolio. If bank managers prefer other goals than value-maximization, the increase in capital levels will cause an increase in default risk. The results of empirical research focused on the risk exposure are mixed. Those based on Shrieves and Dahl s model proves that more capital did not cause higher risk. In case of Gonzales [2005] and Jokipii, Milne [2011] the scope of analysis was broader and empirical results are differentiated respectively for banks with high charter value and high capital buffers. According to Gonzales [2005] banks operating in the environment with restrictive regulations that have high charter value are prudent in their risk taking behavior, what suggests that they have more to lose and are not willing to destruct the value. Jokipii and Milne [2011] took into account the level of capital buffers banks usually hold in excess of regulatory requirements. They concluded that bank with small capital buffers aimed at capital increase and risk reduction in order to improve capital buffers. On the other hand, banks with high capital buffers used capital increase to expand the risk exposure in order to improve profits and thus ROE. Since the empirical results are mixed it may be attributed to the following factors: since the banking sector credit activity, credit portfolio quality (measured by NPL) and thus earnings are procyclical also due to capital regulations (for discussion see e.g. Blum, Hellwig 1995; Altman et al. 2005, Andersen 2011), the models are missing information on the cycle, which affects the bank risk taking behaviour; 7

8 deposit insurance is actually applicable to small and medium sized banks, in the case of any big bank (not even exactly G-SIB) no deposit insurer is capable of paying deposits back to the customers; thus in the case of big banks deposit guarantees shall not be taken into account in the models, since it plays no significant role for bank behavior; psychological and demographic factors are missing risk taking behavior is related to some psychological and demographic features (for discussion see e.g. Pathan 2009, Berger et al. 2013); competition on the market though competition may stimulate aggressive risk-taking (for discussion see e.g. Boyd, De Nicoló 2005, Jimenez et al. 2013) management compensation which may stimulate aggressive risk-taking (for discussion see e.g. John et al. 2000, Hagendorff, Vallascas 2011); as a reaction to the crisis, regulators imposed new approach to management compensation schemes; as well as ownership structure if shareholders are very dispersed they are not able to exercise actual discipline over the management, thus having at least one institutional investor seems to be a good risk reducing factor especially in terms of managerial moral hazard (for discussion see e.g. Saunders et al.1990; Anderson, Fraser 2000, Laeven, Levine 2009, Iannotta et al. 2013). Since in Aggarwal and Jacques [2001] prompt corrective action was regarded as a positive factor for bank capital and risk taking, we found it necessary to include in the future models at least factor called resolution cost or resolution factor due to the fact that both in Europe and globally, policy makers are inclined to implement resolution regime (with living wills) also for big banks in order to reduce moral hazard. So far, the resolution has been used for dealing with small and medium-sized financial institutions. Although bail-out of banks is criticized, this factor shall be also incorporated future models in case of big banks. The first theoretical model with resolution factor and management compensation factor for SIFIs, not G-SIFIs as a special case, was presented by Freixas, Rochet [2013]. The authors suggest to set up systemic risk authority with the above mentioned responsibility, as well as to levy a systemic tax. The first measure has been already implemented in many major economies. 8

9 Since the FSB s decision on assessing a bank as G-SIB is of discretional nature we decided to use all the banks selected by the FSB in 2011 and 2012 which constitute the total sample of 31 banks. Those banks were divided into three groups (list is presented in Annex): G-SIB H financial institutions concentrated mostly on the home country market (7) in terms of assets and/or earnings; G-SIB E financial institutions concentrated on the European market (12) in terms of assets and/or earnings; G-SIB G global financial institutions (12) in terms of assets and/or earnings or activities. In order to analyze the impact of the crisis and changes in regulations we used the basic ratios (see: table 2) like a solvency ratio 2, ROE and leverage (assets/capital). More detailed ratios that might be defined and used for our analysis were skipped due to the fact that not all G-SIB provide the same scope and structure of detailed information. Table 2: Selected ratios for G-SIBs and three groups of G-SIBs G-SIBs Solvency ratio MAX 18,40% 15,20% 26,80% 20,60% 22,00% 21,60% 25,20% MIN 7,40% 7,40% 7,40% 9,80% 10,02% 10,30% 12,50% Median 11,77% 11,01% 12,49% 14,15% 15,20% 14,84% 16,30% ROE* MAX 26,65% 27,10% 17,13% 34,93% 16,22% 17,24% 16,89% MIN 5,28% -10,74% -132,38% -17,07% -5,53% -2,59% -8,24% Median 17,30% 14,34% 3,41% 7,74% 8,58% 7,41% 4,84% Leverage* MAX 47,3 51,9 183,3 52,9 52,8 39,6 37,0 MIN 9,0 6,7 6,7 7,3 7,5 9,3 9,0 Median 23,0 24,4 25,0 19,6 19,7 19,8 17,8 G-SIBs H Solvency ratio MAX 13,59% 13,34% 14,80% 15,02% 16,63% 16,93% 16,68% MIN 11,31% 10,56% 10,53% 10,90% 11,60% 11,60% 12,50% Median 12,39% 11,45% 12,39% 14,66% 15,30% 15,40% 14,71% ROE MAX 15,62% 13,76% 13,39% 34,93% 16,22% 17,24% 16,89% MIN 5,28% -5,51% -132,38% -0,16% -0,98% 0,63% 1,77% Median 12,47% 10,21% 0,30% 8,55% 7,85% 6,19% 10,23% Leverage 2 For years 2006 and 2007 Morgan Stanley and Goldman Sachs did not report their solvency ratios. 9

10 MAX 36,8 46,3 183,3 52,9 44,1 37,2 30,518 MIN 10,8 11,7 10,3 9,6 9,9 9,3 9,327 Median 14,8 17,3 19,2 16,3 18,1 18,6 16,929 G-SIBs E Solvency ratio MAX 12,50% 11,60% 14,10% 16,60% 16,90% 16,40% 20,90% MIN 8,80% 8,60% 9,40% 9,80% 11,50% 10,30% 12,70% Median 10,86% 10,31% 11,70% 13,68% 14,30% 13,90% 16,26% ROE* MAX 25,27% 26,72% 15,01% 17,59% 10,85% 11,60% 11,08% MIN 10,51% 5,13% -57,81% -17,07% -5,53% -2,59% -8,24% Median 16,54% 15,11% 4,18% 5,27% 5,62% 5,45% 0,53% Leverage* MAX 47,3 49,0 115,9 48,2 52,8 39,6 36,985 MIN 19,2 17,7 19,0 15,6 14,5 18,0 14,762 Median 29,8 32,3 37,0 26,4 25,1 25,4 23,120 G-SIBs G Solvency ratio MAX 18,40% 15,20% 26,80% 20,60% 22,00% 21,60% 25,20% MIN 7,40% 7,70% 9,10% 10,08% 10,02% 11,01% 13,00% Median 13,00% 12,45% 15,65% 16,10% 16,45% 17,00% 17,33% ROE MAX 26,65% 27,10% 17,13% 18,93% 15,11% 12,12% 12,19% MIN 13,97% -10,74% -51,13% -12,98% 6,39% 6,16% -4,45% Median 19,86% 14,72% 4,38% 8,01% 10,35% 7,59% 6,04% Leverage MAX 42,1 51,9 49,7 27,6 25,4 25,5 25,1 MIN 9,0 6,7 8,5 7,3 7,5 9,3 9,0 Median 17,0 17,7 15,6 14,8 12,9 13,8 13,2 * without Dexia in due to extreme values of ratios Source: own elaboration We need to keep in mind however that there are differences in accounting standards between Europe (IAS/IFRS) and the US (US-GAAP) 3. This is a very important issue because in some cases differences in leverage may be significant if we use different accounting standards. G- SIBs shall be required to use one set of standards to inform the wide public about their financial standing in order to increase comparability. Without that change the market and regulatory discipline could not be exercised properly. As presented in table 2 in the year 2012 in comparison with 2006 the median solvency ratio for G-SIBs improved by about 4,5 p.p., median ROE dropped by 12.5 p.p. and median 3 There are also Japanese accounting standards (JP-GAAP), but Japanese banks publish also reports according to US-GAAP, which were used in the sample. 10

11 leverage (assets/capital) by 5.2. G-SIBs concentrated in their home country faced the smallest drop in median return on equity and a smallest increase in the level of solvency ratio among 3 distinguished groups of G-SIBs. In the case of the leverage, G-SIBs H increased their leverage, while G-SIBs E and G-SIBs G decreased their leverage to about ¾ of the 2006 year level. At the year-end of 2012 G-SIBs H represented the highest median ROE, the lowest median solvency ratio. The lowest median ROE and the highest leverage was shown by G- SIBs E, while the highest median solvency ratio and the lowest leverage was shown by G- SIBs G. The indicated differences may suggest different managerial reactions to higher capital levels. In order to prove that long time series are required and are not available yet. Table 3: Results of significance test for median values (2012 vs. 2006) Variable test statictics G-SIBs G-SIBs H G-SIBs E G-SIBs G p-value test p-value test p-value test (onesidedsided) statictics (one- statictics (onesided) statictics p-value (onesided) CAR 5,0480 0,0000 2,9692 0,0015 4,7958 0,0000 2,5100 0,0060 ROE -6, ,0000-1, ,0463-3, ,0000-4, ,0000 Leverage -1, ,1269 0,8248 0,2048-1, ,0550-2, ,0082 Source: own elaboration In order to assess statistical significance of observed changes we employ Wilcoxon test for medians which doesn t require any assumptions regarding distributions (see: table 3). All the changes in CAR and ROR values in 2012 compared to 2006 proved to be significant across all types of SIBs (at the 5% significance level). Observed decrease in leverage occurred to be insignificant except for G-SIBs E (however at the higher significance level 5.5%). 4 Table 4: Number of value destructors among G-SIBs Group of G-SIBs Number of G-SIBs value ROE<0 destructors 2006 G-SIBs H 0 2 G-SIBs E 0 0 G-SIBs G G-SIBs H Formally we analyzed the general population of G-SIBs year by year, however treating each year results as empirical realization of the general population allows us to treat year results as a sample and therefore to test statistical significance. 11

12 G-SIBs E 0 2 G-SIBs G G-SIBs H 3 5 G-SIBs E 4 11 G-SIBs G G-SIBs H 1 5 G-SIBs E 3 12 G-SIBs G G-SIBs H 1 6 G-SIBs E 2 12 G-SIBs G G-SIBs H 0 6 G-SIBs E 3 12 G-SIBs G G-SIBs H 0 5 G-SIBs E 5 12 G-SIBs G 1 12 Source: own elaboration In the eyes of the value based approach, ROE shall be at least higher than the cost of equity capital to allow shareholders to create value added. ROE below the cost of equity capital means value destruction. We defined the cost of equity capital as follows: c = r rf + β risk premium, whereas: r rf - is risk free rate calculated with the use of 10 years treasury bonds β - is CAPM parameter Data for calculation of r rf and β where taken from publically available data resources. We assumed flat risk premiums for G-SIBs for each year. For (pre-crisis period) risk premium was set at 5%, for the remaining period at 10%, since market participants risk perception for the banking sector has significantly changed. In the pre-crisis period the number of G-SIBs generating losses and/or destructing value was low. It increased significantly in 2008 and remained high until The last two analyzed 12

13 years ( ) show however that G-SIBs generating losses are concentrated in Europe, which shall be associated with the sovereign debt crisis which started in G-SIBs will still face the need to attract new capital, especially classified as additional Tier 1 or Tier 2 under Basel 3 from 2013 until 2023 when the phase out period ends. Attracting new capital requires the interest of investors. The banking industry shall prove to the market participants that it is able to generate reasonable ROE and increase the value of their investments. The long-time investors shall regard banking sector as attractive, what may be measured by ROE, while the short-time investors are rather focused on stock price. We found that long-time investors are more advisable for G-SIBs to increase their stability and control risk-taking behavior, thus we attached importance to ROE and value creation. The question arises how to improve return on equity in order to avoid value destruction in the regulatory environment, in which banks are requested to hold higher capital ratios, reduce leverage, improve liquidity, bear the costs related to resolution mechanism or new taxes. There are some ways, e.g.: cut the fixed costs such as staff costs used during the crisis to a large extent, enter new geographical markets and attract new customers financial institutions were rather reluctant to do that during the crisis, switch to cheaper in maintenance, modern distribution channels in the case of retail financial services it is rather heavily used by institutions concentrated on the retail market, enter new areas that are unregulated which places higher demands on regulators, supervisors (micro- and macroprudential) and institutional market participants to react on time and not allow the risk to accumulate in the financial sector. Which tools will be used by G-SIBs depends on their financial standing, geographical coverage, strategy, market opportunities, managers and owners approach to risk taking. However, there might be anxiety that the last way is the only solution to meet both regulatory requirements and shareholders expectations. So far, only one theoretical model developed in the literature has concentrated on the large banks [Freixas, Rochet 2013]. Previous models (theoretical and empirical) did not 13

14 differentiate between big, medium-sized and small banks, most of them have overstated the role of deposit insurance which in practice is not applicable to large banks (except paying premiums) and none has taken into account the application of the resolution regime and management compensation issues. These underline the need to develop a new theoretical and empirical framework to analyze the relationship between capital and risk for large, internationally active banks that operate in different environments and under differentiated regulatory and supervisory regimes. 3. BCBS methodology The BCBS has proposed in a consultative document in July 2011 and confirmed in the rules text in November 2011 [BCBS 2011] the use of indicators, which assessment would be completed by qualitative factors in order to set additional capital needs (called additional loss absorbency ). The methodology was presented in up-dated version in July 2013 [BCBS 2013]. According to the proposed framework, additional loss absorbency is supported by global recovery and resolution frameworks. Table 5: Indicators for G-SIBs identification Category and Indicators weighting Crossjurisdictional claims Cross-jurisdictional activity (20%) Cross-jurisdictional liabilities Size (20%) Interconnectedness (20%) Substitutability / Financial Institution Indicator weight Initial data sources and authors comments 10% BIS consolidated banking statistics; very aggregated 10% approach; in order to assess the channels of contagion exposures among G-SIBs and other SIFIs shall also be analyzed Total exposures as 20% Reported by banks; size shall defined in the Basel 3 also be analyzed in relation to leverage ratio GDP and as a share in the market of the home and host countries which is missing Intra-financial system 6,67% Source was not stated; a part assets of this data is covered by BIS Intra-financial system 6,67% statistics; very aggregated liabilities approach; in order to assess Securities outstanding 6,67% the channels of contagion (introduced in 2013, exposures among G-SIBs and before wholesale funding other SIFIs shall also be ratio) analyzed Assets under custody 6,67% GlobalCustody.com, but also collected from banks; from the legal perspective assets under 14

15 Infrastructure (20%) with a cap introduced in 2013 Complexity (20%) Payments cleared and settled through payment systems Underwritten transactions in debt and equity markets OTC derivatives notional value Level 3 assets as defined in Basel 3 liquidity coverage ratio Value of the securities held for trading and available for sale without high-quality liquid assets included in LCR custody shall be separated from banks assets 6,67% Source was not stated, but BIS is collecting some data on the payment system; no information on distinction between gross and net settlements systems; structure of payments is missing (country and financial institutions) 6,67% Bloomberg and Dealogic, but also be collected from banks; structure of underwritten transaction is missing, e.g. on a country level 6,67% Reported by banks; structure of OTC derivatives is missing (country and financial institutions) 6,67% Reported by banks; structure of level 3 assets is missing 6,67% Source was not stated, but probably reported by banks or taken from financial statements; structure of securities is missing (country and financial institutions) Source: BSBS (2013), p. 6 and own comments The indicators cover the following areas: size, cross-jurisdictional activities, interconnectedness in the financial system, substitutability by another institution and complexity (see table 5). On the basis of these indicators a synthetic indicator is calculated (called score ). The value of a single indicator for a given bank is calculated as a relation to the aggregate value of all the banks in the sample (from 2013 on up-dated annually), what has its pros and cons for the results of the analysis. The advantage of this solution is simplicity, but the disadvantage is related to going with the trend in the peer group. If most G-SIBs implemented the same strategy of development in the areas covered by the indicators, the supervisors might miss considerable change in the structure of their activities. Thus we find it useful to introduce some fixed thresholds for indicators that may be reviewed periodically. This shall help to set operational definition of a G-SIB. 15

16 The categories of ratios selected to identify G-SIBs were selected properly, but the level of analysis is too general. As presented in table 5, data are analyzed at a very aggregated level and in many cases an in-depth approach is missing. We divided the G-SIBs into three different groups and some of them are not so global since they are concentrated on their home market or just the European market, which for EU banks might also mean home market. Keeping in mind that the problems of one G-SIB may have a destructive impact on financial markets and in order to better select an actual G-SIBs, we find it useful to take into account more detailed data and/or additional information (e.g. country and groups of institutions). Data sources have been differentiated, which has been explained initially by putting lower informational burden on G-SIBs. This should be questioned, because all interested parties, i.e. G-SIBs, supervisors and regulators, public authorities and market participants shall be well informed from one source with the use of single set of standards. It is better to impose higher informational burden to better know the structure of G-SIBs operations than to put additional loss absorbency in a wrong way. Additionally, supervisors and regulators should be able to verify the data. If they are taken from different sources and are representing different accounting standards, the verification possibilities are rather vague. The initial solution seemed to be a stopgap. In 2013 BCBS [BCBS 2013] announced disclosure requirements from 2013 year end data for G-SIBs and other banks with total exposure exceeding 200 billion EUR (altogether 75 banks). They will be obliged to publish at least 12 indicators following the reporting instructions and template. Some of the indicators that are missing on the main list (table 5), have been put by the BCBS on a list of ancillary indicators that may be used by supervisors in the supportive supervisory judgment, that is expected be used rarely. There are the following ancillary indicators: (1) wholesale funding dependence ratio (this ratio was removed for the list of main indicators in 2013), (2) foreign net revenue, (3) peak equity market capitalization, (3) participation in securities lending, (4) number of jurisdictions, (5) unsecured settlement/clearing lines provided. In our opinion 3 indicators shall be included in the main list of the indicators. Equity market capitalization is a very important market figure which presents the scale of possible losses to shareholders and thus distress on the financial market. The number of jurisdictions shows the potential legal complexity of coordination of G-SIBs management as 16

17 well as the resolutions procedures. Indicator (2) exhibits the share of non-domestic profits, which may help to differentiate between global G-SIBs and not so global G-SIBs in a more effective way. The same applies to the ancillary indicator that has been removed in 2013, i.e. cross-jurisdictional claims and liabilities in relation to total assets. This indicator should be taken into account in case of the assessment of global or non-domestic activities. Therefore it is worth mentioning that IAIS [2013] initial assessment methodology for global systematically important insurers (G-SIIs) is reflecting better the issue of cross-border activities, although the methodology for additional loss absorbency is on an earlier stage than in case of banks. The final outcome of the use of indicators presented in table 5 and when applicable supervisory judgment, supported by ancillary indicators is a score assigned to each G-SIB which will require a minimum additional loss absorbency. Intentionally, bucket 5 (score D) shall be empty to act as the sword of Damocles for G-SIBs. The first results were published in November 2013 with the refreshed list of G-SIBs (see: table 6). Table 6: Required loss absorbency Bucket Score Minimum additional loss absorbency (common equity as a percentage of risk-weighted assets) 5 How many G-SIBs (for Nov list)?* How many G-SIB G (authors analysis)?* How many G-SIB H (authors analysis)?* D - 3,5% (empty) 4 C - D 2,5% 2 (2 ) B C 2,0% 4 (2 ) 1-2 A B 1,5% 8 (-) Cut-off - A 1,0% 14 (-) 6 4 * ICBC (bucket 1) has been not taken into account Source: BSBS (2013), p. 12 and FSB (2012, 2013), p. 3 Only 2 G-SIBs were classified into bucket 4. These are: HSBC (110% relation of consolidated assets to UK GDP; 16.1% solvency ratio for the 2012 year-end) and JP Morgan Chase (17% relation of consolidated assets to US GDP; 15.3% solvency ratio for the 2012 year-end). 5 One of them were treated in our analysis as G-SIB G (HSBC). 5 A year before there also in bucket 4: Citigroup and Deutsche Bank. Both were moved to bucket 3. 17

18 As mentioned in BCBS documents, the methodology will be reviewed periodically, which shall be strongly supported by in-depth, multifaceted G-SIBs analyses and academic research. 4. Data analysis The empirical part of the research was carried out on the basis of panel data set covering all 31 G-SIB s on a consolidated level during the time range of The database was prepared with the usage of publicly disclosed information from the annual financial statements. Since there are two basic dimensions along which a bank s standing should be assessed being risk exposure and profitability, we focused on the basic ratios and indicators related to these dimensions, especially in relative terms, i.e. taking into account risk absorbing capacity and economic capability for profit generation. The data set have passed the basic data quality checks related to the ranges of the values, identification of the missing values and outliers. In the case of the extreme outliers we decided to amend the base with additional variables with outlier value replaced by the missing. Due to differences in currencies we decided to limit our indicators to relative ones. In order to be able to account for information on the cycle, which affects the bank s risk taking behaviour some basic information describing the macroeconomic condition, i.e. GDP, GDP growth, inflation and unemployment were taken into account, as well as additional ratios combining both the description of the G-SIBs and the economy (e.g. assets related to the GDP of the domestic market). We also found that one of the possible important partial indicator of risk exposure is the relation of RWA (risk weighted assets) to assets. Due to the unavailability of RWA for some G-SIBs we decided to use a proxy value based on CAR (or CAR Tier I) and Equity with the full awareness of its approximate nature. On the other hand that proxy indicator is formed by relation of CAR to Leverage and can be interpreted in those terms directly. Finally the variables describing the G-SIBs were organized into groups covering the following categories: 18

19 information of identification nature (name, market, type of financial statement, year, etc.), information from financial statements (e.g. assets, liabilities, equity, off-balance liabilities, P&L, etc.), information related to the loan and deposit activity (loans, deposits, net interest margin, etc.), macroeconomic variables (GDP, GDP growth, inflation, unemployment, current account), geographical location, ratios referring to capital adequacy, financial ratios referring to risk taking, financial ratios referring to profitability. We need to stress that we are fully aware of the limitations embedded in the collected data set, i.e. differences in accounting standards, differences in Basel 2 adoption, missing values, extreme outliers (which prove to be correct values). Since these limitations could potentially weaken the conclusions resulting from the analysis we find it important to continue the research as well as to maintain and develop the database in future. 4.1.Methodology used The method we decided to use allows for the identification of the internal dependences structure for multivariate observations. We employed explanatory factor analysis which examines the underlying patterns of dependencies for financial ratios describing the G-SIBs. We also found a factor model attractive as a starting point in the empirical research since it allows us not to be biased in our research and conclusions by the prior decision on the relationship between variables (naturally, except the obvious, deterministic relations) and their structure, e.g. which of them should be treated as independent and dependent variables. Especially the regression like analysis would need the definition of the independent variables 19

20 to be modeled. Last, but not least we decided to take advantage of factor analysis in order to try to address one of the shortcomings (enumerated in Section 2) identified in the foregoing literature on the relation between the increase in capital and the risk-taking. The issue is that the foregoing literature is missing the influence of sociological and psychological components as one of the factors influencing the perception of the risk and therefore willingness to take or not additional risk along with capital increase. The factor model is one of the multivariate statistical techniques that has been widely used in psychology and sociology in order to model the latent variables and assess their influence on directly observed phenomena. If the phenomena being modeled can be perceived as a result of some psychological or sociological factor, the latent variables are often interpreted as attitudes, motivation, emotions or preferences, all of which are unobserved directly [Anderson, 2003]. We will check if there is any evidence of the latent variables of that kind influencing the bank s situation presented in the financial statement. The factor analysis has also been used in modern finance. Let us recall the Modern Portfolio Theory and Arbitrage Pricing Theory which employed a factor model as a solid foundation for establishing the theory itself [Cha, Chan, 2000]. Within the banking industry it is widely recognized as the statistical foundation of the IRB model within the Basel 2 regime. The philosophy behind the model assumes that there are some latent variables (called factors ), which cannot be observed directly but at the same time they exert influence on the objects being analyzed (i.e. G-SIBs) by having an impact on directly observed variables (e.g. financial ratios). We assume that each of the observed variables Xi is a function of m latent factors (common for all the X s) and a specific factor εi (separate for each variable): X i = f(f 1, F 2,, F m ) + ε i where: Xi represents i-th observable variable X (e.g financial ratio), Fj represents j-th unobservable factor F, εi represents the error term specific for i-th observable variable. 20

21 The factor model assumes that each of the observed variables Xi is a linear combination of the underlying common factors and a specific factor εi. The specific factor represents residual in the model and a unique part of the variable, that does not depend on common factors. Each factor represent an area of conceptual generalization and different factors represent different areas, qualitatively different from each other. Latent factors are both of a systematic and nonsystematic nature. Systematic ones could be possibly related to the qualities of banking and other financial institutions sector or some sociological phenomena behind the risk takingbehaviour (e.g. preferences of the supervisor) and non-systematic factor accounts for the specificity of the observed financial ratio itself. The model can be presented in the matrix notation as: X = ΛF + ε where: X = [X 1 X 2 X p ] T F = [F 1 F 2 F m ] T ε = [ε 1 ε 2 ε p ] T λ 11 λ 12 λ 1m Λ = [ ] λ p1 λ p2 λ pm and: λij the loading of the i-th X variable on the j-th common factor F, m number of common factors F, p number of observed variables X. The loadings represent sensitivities to the underlying factors and therefore allow for interpretation of the latent variables based on the importance of the j-th factor to the i-th X variable. The total variation of the i-th variable is decomposed within the model into a common variation (called communality ) and residual variation (called specifity ) resulting from the error term εi. 21

22 The basic assumptions of the model can be summarized as follows: there are common unobserved factors underlying observed variables, the specific factors are orthogonal to each other, the specific factor is orthogonal to the systematic ones. Additionally, in order to carry out an effective analysis the observed variables should be at least moderately correlated (in terms of the Pearson correlation) which leads to m - the number of factors being much smaller than the number of original observed variables p. That allows the estimated factors to reflect a substantial part of information contained in the original data set. The parameters of the model to be estimated are loadings λij and communalities. There are several methods of estimation that can be used, however the most commonly used are the principal factor method and maximum likelihood. The estimated loadings are not unambiguously determined since the factors can be rotated in the multidimensional space of X s in order to determine the possible clear interpretation. In order to avoid the possible ambiguity in interpretation, we examined different orthogonal rotation techniques in order to check the similarities in the results. Finally we used varimax rotation since it provided (as expected) the most clear picture of dimensions in the data described by the factors. Taking into account the dynamic nature of the phenomena we analyze, formally we should have employed the dynamic factor analysis. Due to the fact that the time range is relatively short compared to the number of variables and the suggestions in the literature [Cha, Chan 2000] that the estimates generated by the static factor model are a good approximation of the dynamic ones we decided to focus on the development of the static factor model as a starting point. The factor model allows for: analysis of the underlying patterns of dependencies within multivariate description of G-SIBs. determination if the information embedded in a set of financial ratios can be summarized within a smaller set of factors without substantial loss of information. summarization of the empirical relationships between the set of partial indicators referring to risk-taking behaviour and profitability. 22

23 reduction of multidimensionality together with the reduction of information redundancy. possibility to identify empirically the sources of risk and profitability within the G- SIBs. 4.2.Results[d1] The usage of the factor model allows us to find answers to the following questions we pose: Are there common latent factors for all the G-SIBs? Is there reasonable economic interpretation for latent variables? How many latent factors should be taken into account? Which of the observable variables are correlated to latent factors? Are there differences between factor models developed separately for G-SIBs active in different geographical locations? What is the nature of the differences, if any? We took the following steps in the analysis: 1. Confirm correlation between the observed variables. 2. Estimate preliminary one size fits all factor model for all the G-SIBs with different methods of estimation and check the number of underlying factors. Decide on the method and number of factors. 3. Use different techniques for the factors rotation in order to check if the results converge to a similar structure of dependences. Decide on the method of rotation to use for the final model with the preference for orthogonal rotation. 4. Estimate the final one size fits all factor model for all the G-SIBs and check the possibility to interpret the factors and 5. Check the common factors distribution for different groups of G-SIBs separately (i.e. G-SIBs E, - G and -H). 6. In the case that there are considerable differences in the factors distributions (step 5) among different types of G-SIBs, we estimate the separate factor models repeating steps 1-4. Based on the Pearson correlation criterion we decided not to limit the set of financial ratios, however we dropped the pure macroeconomic variables. The model estimated with the usage of the principal factor method detected 4 latent variables (based on the Guttman-Kaiser 23

24 rule and Cattell s scree test) accounting for 89.8% coverage of information from the original data set. On the other hand, restraining the number of common factors to 3 allows for 80% coverage of the information, which is really satisfactory result. Table 7. One size fits all factor model with 4 common factors. Factor1 Factor2 Factor3 Factor4 Net Interest Margin RWA to Assets RWA (Tier1) to Assets Provisions to Loans Deposits to Assets CAR (Tier1) CAR Deposits to Loans Interest Income to Provisions Loan to Assets ROE ROA Leverage Deposits to GDP Assets to GDP Source: own elaboration The first latent variable (Factor 1, accounting for 45.1% of the variation) is risk related with focus on credit risk and can be summarized as risk and lending activity. It represents high level of net interest margin together with high RWA to assets ratio and high level of provisions. It can be perceived as relatively high level of credit risk with high level of deposits as substantial source of financing. What s interesting, the factor F1 handles both sides of credit risk: exposure (high RWA to assets) and effects (net interest margins, level of provisions). The second latent factor (covering additional 20.0% of the variation) represent capital adequacy with strong capital base, high relation of deposits to loans (again, deposits as 24

25 substantial source of financing), low level of loans to assets and high levels of interest income to provisions. The third factor represent high effectiveness with low leverage (additional 14.7% of variation) and the forth (9.83% of variation) is size related and can be clearly interpreted as too big to fail. In order to check to what extend the identified structure of latent factors is shared by differently located G-SIBs we use box plots of F1-F4 and 2 dimension distribution of F1 and F2, since those two factors retain ca. 2/3 of information retained from the original data set. Figure 1. Box plots of F1. Source: own elaboration Figure 2. Box plots of F2. 25

26 Source: own elaboration Figure 3. Box plots of F3. 26

27 Source: own elaboration Figure 4. Box plots of F4. 27

28 Source: own elaboration The most striking differences in conditional distributions can be assigned to the 1 and 2 factor, which represent combination of risk&lending and capital adequacy dimensions. In terms of geographical location, globally active G-SIBs are mostly distinguishable in almost all dimensions along identified latent factors. They are more diversified in terms of all but the 3 rd (effectiveness with low leverage) latent factors compared to both other G-SIBs. In the 3 rd latent factor dimension however, they are more effective compared to European and home. Typical European G-SIBs are mostly located both in the region of lower values of risk&lending activity and lower capital adequacy. Typical G-SIBs operating on home market is located within the confidence limits for all but 4 th latent factors. Home G-SIBs are not too big to fail compared to both other groups. We examine additionally two-dimensional distribution of those factors both for all G-SIBs and separately for different geographical locations (Figure 5 Figure 8). Figure 5. Unconditional distribution of F1 and F2. 28

29 Source: own elaboration Figure 6. Distribution of F1 and F2 for G-SIBs operating on the European market. 29

30 Source: own elaboration Figure 7. Distribution of F1 and F2 for G-SIBs operating on the global market. 30

31 Source: own elaboration Figure 8. Distribution of F1 and F2 for G-SIBs operating on the home market. Source: own elaboration Comparison of the conditional distributions of the 1 st and the 2 nd latent variable confirms what was suspected, i.e. there are strong differences between different types of G-SIBs. There is no one size fits all solutions especially related to the risk and capital adequacy dimensions. Comparison of the latent factors distributions suggests that globally active G-SIBs enjoy higher values of all the latent factors compared to both the other groups G-SIBs E and G-SIBs H, i.d. distributions are shifted towards higher values in risk and lending dimension, more capitalized and more effective, and on the top of that all there are even bigger players in the too big to fail league. Analysis of the F1 and F2 dimensions reveals two potential clusters concentrated in the region of less capital with less risk and more capital with average risk. What s interesting there s no distinguishable group of more risk with more capital in case of European G-SIBs, whereas for global ones that type of cluster is 31