A Guide to IMF Stress Testing
Chapter

Chapter 11. Bankers without Borders? Implications of Ring-Fencing for European Cross-Border Banks

Author(s):
Li Ong
Published Date:
December 2014
Share
  • ShareShare
Show Summary Details
Author(s)
Eugenio Cerutti, Anna Ilyina, Yulia Makarova and Christian Schmieder This chapter was previously published as IMF Working Paper 10/247 (Cerutti and others, 2010). It has benefited from comments by Bas Bakker, Jan Brock-meijer, Miroslav Kollar, Inci Otker-Robe, Daniel Hardy, Jérôme Vandenbussche, and Rachel van Elkan, as well as by the participants of a Monetary and Capital Markets Department and European Department seminar.

This chapter presents a stylized analysis of the effects of ring-fencing (i.e., different restrictions on cross-border transfers of excess profits and/or capital between a parent bank and its subsidiaries located in different jurisdictions) on cross-border banks. Using a sample of 25 large European banking groups with subsidiaries in Central, Eastern, and Southern Europe (CESE), we analyze the impact of a CESE credit shock on the capital buffers needed by the sample banking groups under different forms of ring-fencing. Our simulations show that under stricter forms of ring-fencing, sample banking groups have substantially larger needs for capital buffers at the parent and/or subsidiary level than under less strict (or in the absence of any) ring-fencing.

Method Summary

Method Summary
OverviewThe method is an extension of the standard balance sheet stress testing framework that allows explicit modeling of risks associated with different types of constraints on intragroup cross-border transfers of liquidity/capital.
ApplicationThis framework may be used for estimating the liquidity/capital shortfall of a banking group that is restricted in its ability to reallocate excess liquidity/capital across jurisdictions following a shock that affects some of the affiliates within a group.
Nature of approachEconometric and balance sheet-based analyses. Credit shock is modeled as deterioration in macroeconomic conditions leading to an increase in nonperforming loans and a decrease in the return on assets of banks.
Data requirementsAccounting information and data on both the banking group and each subsidiary of the group to be analyzed. Economic and market data (GDP, interest rates).
StrengthsThe framework enables the inclusion of ring-fencing scenarios (i.e., different types of restrictions on cross-border transfers of excess profits and/or capital between a parent bank and its subsidiaries located in different jurisdictions) when stress testing cross-border banks.
Weaknesses
  • The framework does not explicitly capture dependencies across banks and banking systems (as portfolio models may do).

  • This method is data intensive, as it requires up-to-date information on the geographical structure of a banking group, as well as detailed balance sheet data for the group (consolidated), parent bank, and all (or at least all the major) foreign banking subsidiaries.

ToolStandard econometrics package.

The concept of centralized capital and liquidity management by internationally active banks was challenged by the recent crisis, sparking a debate about the desirable organizational and regulatory arrangements for cross-border banking groups. This chapter focuses on the costs for these banking groups that are associated with different restrictions on intragroup cross-border transfers imposed by the host or home country regulators (henceforth referred to as “ring-fencing”). More specifically, it provides a stylized analysis of how much additional capital might be needed if the banking groups are restricted, to different degrees, in their ability to reallocate funds across jurisdictions following a credit shock affecting their lending activities in a given region. The study does not estimate the group-level potential recapitalization needs under an extreme scenario (which is typically done in a stress test) but rather considers the implications of adverse economic conditions for cross-border banking groups under different forms of ring-fencing. The analysis is based on bank-level data for European banking groups and their subsidiaries in Central, Eastern, and Southern Europe (CESE).1

At present, a number of European countries have legal restrictions on intragroup cross-border asset transfers. These limits are aimed at preventing undue influence by a foreign parent on its subsidiaries (e.g., in the form of disproportionate transfers of assets that could potentially trigger solvency or liquidity problems) or aimed at protecting the interests of minority shareholders and creditors of subsidiaries (European Commission, 2010). In order to ring-fence the subsidiary from the rest of the group, the host country can target the subsidiary’s ability to transfer funds abroad directly or indirectly, through measures affecting the entire domestic banking system (e.g., stopping the distribution of dividends by all banks during a crisis).2

During 2008–09, many subsidiaries of European banking groups had to rely on their foreign parents for capital and liquidity support. There is some evidence that, ex ante, the CESE subsidiaries had expected that they could rely on parent banks in case of need (e.g., the average capitalization levels of the foreign-owned subsidiaries in most CESE countries were 1 to 2 percentage points lower than the banking system averages at the outset of the crisis (Appendix I). Ex post, these expectations were validated by the assistance provided by parent banks.3 Yet, in order to maintain financial and economic stability, the regulators in many host countries tightened restrictions on intragroup cross-border transfers, limiting the ability of cross-border banking groups to reallocate funds from subsidiaries with excess capital (liquidity) to those that were in need of capital (liquidity).4

There are arguments both for and against ring-fencing. The arguments in favor of centralized cross-border bank structures and against ring-fencing rely on efficiency and financial stability considerations (e.g., benefits of diversification across country-specific shocks). From a cross-border bank’s perspective, the ability to freely reallocate funds across its affiliates is essential for achieving the most efficient outcome—a point emphasized in the 2010 report prepared by the Institute of International Finance (IIF, 2010). Centralized cross-border bank structures may yield benefits for the host country economies as well. De Haas and van Lelyveld (2010), for example, show that the ability of international banks to attract liquidity and raise capital allows them to operate an internal capital market, which provides their subsidiaries with better access to capital and liquidity than what they would have been able to achieve on a stand-alone basis and hence may help to reduce the pressure to scale back lending during economic downturns. For both home and host authorities, the absence of ring-fencing facilitates diversification and can thus make the group as a whole more stable, for example, against shocks in the home country.

However, there are also arguments in favor of ring-fencing. For a host country regulator, the decision to impose ring-fencing typically would be driven by macro-financial stability considerations, such as the need to protect the domestic banking system from negative spillovers from the rest of the group or, more generally, to increase reserves for the whole domestic banking system during a crisis when the magnitude of the impending output collapse and bank losses are uncertain. The possibility of contagion from a parent bank to subsidiaries in the European context was analyzed by Popov and Udell (2010), who showed that the contraction of banks’ balance sheets caused by losses and/or a deterioration in bank solvency was transmitted across borders to Eastern Europe by Western European banking groups in the early stages of the 2007–08 crisis.

Moreover, the difficulties in resolving cross-border banking groups and the absence of agreements on burden-sharing mechanisms during the crisis triggered a discussion about the desirability of promoting greater self-sufficiency of banking groups’ affiliates in normal and in crisis times. Hoelscher and others (2010) consider the pros and cons of the so-called stand-alone subsidiarization (SAS) approach, according to which a cross-border banking group should be set up as a network of fully self-sufficient national subsidiaries. The authors note that from a banking group’s perspective, the SAS approach may be beneficial if it can provide additional incentives for subsidiaries to better manage liquidity and credit risk. From the host or home country perspective, the key benefits of SAS include limiting intragroup contagion and allowing selective resolution of problem parts of the group with minimal disruption for the rest of the group.

Leaving aside the question of the potential benefits of imposing greater autonomy on the banking group’s affiliates operating in different jurisdictions, this chapter focuses on the costs of ring-fencing for cross-border banking groups under different forms of ring-fencing. The cost is measured in terms of the amount of external capital that is required to cover capital shortfalls faced by the affiliates of these groups as a result of a credit shock. More specifically, this study estimates the amount of additional capital that might be needed if the sample banking groups are restricted in their ability to reallocate excess profits and/or capital across jurisdictions following a credit shock that affects some of the affiliates within these groups.5

It should be noted that the transfers of excess profits/capital are not the only mechanisms through which banking groups could manage the level of capitalization of their affiliates. For example, the latter could also be achieved through capital injections via subordinated debt or by “shifting” assets (instead of capital) between different parts of the group.6 However, the empirical analysis of these alternative mechanisms is constrained by the lack of publicly available banklevel data on intragroup lending and asset transfers. That said, the conclusions that such exercises might yield are likely to be quite similar to the results regarding the transfers of excess capital/profit presented here.

Three different types of ring-fencing are considered in this chapter, ranging from partial ring-fencing to full ring-fencing. Partial ring-fencing assumes that only excess profits of subsidiaries but not their excess capital buffers can be reallocated within a group. Near-complete ring-fencing assumes that only transfers from the parent to a subsidiary are allowed. Full ring-fencing corresponds to the strict stand-alone subsidiarization (SAS) model, where no intragroup transfers are allowed. The analysis presented here takes into account the parent banks’ ownership stakes in their subsidiaries.

The sample of banks included in the analysis consists of 25 European banking groups and their 113 subsidiaries located in 18 countries in CESE. There are several reasons for using this sample: (1) most of these banks have a large network of subsidiaries operating in several countries in the CESE region; and (2) the fact that many countries in the region were severely hit by the crisis allows us to illustrate a range of outcomes under different ring-fencing assumptions, given a severe but realistic credit shock affecting parts of these banking groups. The individual bank-level data on branches are not used in the estimation because branches are not standalone entities, which makes it difficult for the host country authorities to ring-fence them. The CESE exposures via branches are analyzed as part of the total direct cross-border exposures of parent banks.7

Qualitatively, the results of the analysis are fairly intuitive: any type of restrictions on intragroup transfers would entail the need for additional, and possibly significant, capital buffers at the subsidiary and/or the parent bank level of cross-border banking groups. Quantitatively, the sample banks’ capital needs resulting from a simulated credit shock affecting their CESE subsidiaries over the 2009–10 period are 1.5–3 times higher in the ring-fencing/SAS scenarios than those under no ring-fencing. These results are robust to variations in the methodology for computing capital needs, including the postshock adjustment in risk-weighted assets (standardized versus the Basel II Internal Ratings-Based [IRB] approach).

What are the policy implications of this analysis? First, the establishment of a credible framework for the resolution of cross-border banking groups would help to avoid unilateral and likely more costly solutions (in terms of capital requirements). This is because the existence of such a framework could reduce the incentives for and the incidence of ring-fencing by the home or host country authorities. Second, in the absence of such resolution and burden-sharing mechanisms, setting the minimum capital requirements for cross-border banking groups would have to take into account the potential presence of ring-fencing, especially in crisis times. Such a possibility may force cross-border banks to gravitate toward organizational structures that are more immune to ring-fencing (either SAS-type structures or branch structures). Third, should regulators decide to promote an SAS-like approach, its potential benefits would have to be carefully weighed against its potential costs.

The rest of the chapter is organized as follows. Section 1 provides a description of the exercise and the data. Section 2 explains the calibration of the credit shock affecting CESE subsidiaries. Section 3 presents the methodology for calculating capital needs, as well as the main results under different ring-fencing scenarios. Section IV draws conclusions and discusses policy implications.

1. Cross-Border Banking Groups

A. Description of the exercise

Consider a stylized cross-border banking group that has Subsidiaries A, B, and C operating in Countries A, B, and C respectively (Figure 11.1). Suppose that Countries A, B, and C are affected by a regional shock that leads to a significant deterioration in the credit quality of the loan books of subsidiaries operating in these countries. Suppose that losses resulting from this shock are offset by profits and capital buffers held by each of these subsidiaries (as a first line of defense) and by funds transferred from the rest of the group (as a second line of defense).

Figure 11.1A Stylized Example of a Cross-Border Banking Group—Impact of a Regional Credit Shock on Subsidiaries A, B, and C

Source: Authors.

The capital needs resulting from the CESE credit shock are estimated in two steps:

  • For each subsidiary, the capital need is defined as the amount of capital required to bring its post-shock CAR back to either the country-specific (Basel II) regulatory minimum or to the subsidiary-specific preshock level.8 The latter is conservative in that it requires subsidiaries not to run down preshock buffers.

  • At the group level, total capital needs are computed by adding up all the capital needs of individual subsidiaries (and also losses on direct cross-border exposures of parent banks, in some simulations) and offsetting them against any other funds (i.e., excess profits and/or capital) that can be reallocated from other parts of the banking group.

Hence, the resulting total capital needs at the group level depend on the availability of excess profits and/or capital in the subsidiaries and parent bank, as well as on the degree to which these funds (excess profits and/or capital) can be reallocated within a group.

Suppose that as a result of the shock, one of the three subsidiaries (Sub C in Figure 11.2) experiences a capital shortfall (i.e., its regulatory capital falls below the national minimum capital requirement). Then, the extent to which this subsidiary can be recapitalized using the funds transferred from other parts of the group (i.e., without having to raise fresh capital) would depend on the existence of restrictions on such transfers (i.e., on the degree of ring-fencing).

Figure 11.2A Stylized Example of a Cross-Border Banking Group—Reallocation of Funds within a Group to Cover Capital Shortfall at Subsidiary C

Source: Authors.

Four ring-fencing scenarios are analyzed in this chapter and illustrated in Figure 11.2:

  • The no ring-fencing scenario assumes that the parent bank’s profits as well as subsidiaries’ excess profits and excess capital buffers can be used to cover capital shortfall in any of the subsidiaries.

  • The partial ring-fencing scenario assumes that the parent bank’s profits and only subsidiaries’ excess profits but not excess capital can be reallocated within a group.

  • The near-complete ring-fencing scenario assumes that only transfers from the parent to any of the subsidiaries are allowed.

  • The full ring-fencing, that is, stand-alone subsidiarization (SAS), assumes that no transfers between any of the group’s affiliates (including from the parent bank to subsidiaries) can take place.

B. Sample description

The analysis focuses on 25 European cross-border banking groups with parent banks domiciled in Austria, Belgium, Norway, France, Germany, Greece, Italy, the Netherlands, and Sweden that have significant presence in the CESE region (Table 11.1). The CESE subsidiaries of the European banking groups listed in Table 11.1 include the ones operating in Albania, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Russia, Serbia, Slovakia, Slovenia, Turkey, and Ukraine.

Table 11.1Sample Banking Groups and Their CESE Subsidiaries
Parent BankParent bank’s home countrySubsidiaries
Central EuropeSouthern and South-Eastern EuropeSouthern Europe (Balkans)Baltic CountriesCIS-3
SloveniaCzech Rep.SlovakiaPolandTurkeyBulgariaRomaniaHungaryAlbaniaBosniaCroatiaSerbiaEstoniaLatviaLithuaniaRussiaUkraineBelarus
Erste Group BankAustria
RZBAustria
VolksbankAustria
Bank Austria1Austria
Hypo Alpe Adria Group2Austria
DexiaBelgium
KBC BankBelgium
DNB NORNorway
BNP ParibasFrance
Société GénéraleFrance
Crédit AgricoleFrance
Bayern LBGermany
CommerzbankGermany
Deutsche BankGermany
Alpha BankGreece
EFG EurobankGreece
NBGGreece
Piraeus BankGreece
Allied Irish BanksIreland
Intesa SanpaoloItaly
UniCredit SpAItaly
ING BankNetherlands
Nordea BankSweden
SEBSweden
SwedbankSweden
Sources: Bankscope; and individual bank reports.Note: A circle may indicate the presence of more than one subsidiary. See Appendix II for a detailed description of the sample.

Subsidiary of Unicredit.

Subsidiary of BayernLB.

Sources: Bankscope; and individual bank reports.Note: A circle may indicate the presence of more than one subsidiary. See Appendix II for a detailed description of the sample.

Subsidiary of Unicredit.

Subsidiary of BayernLB.

For each banking group, only the CESE subsidiaries with total assets of at least $1 billion, in which the parent bank has an ownership stake of at least 20 percent, are included in the sample.9 The final sample contains 25 banking groups and 113 CESE subsidiaries, covering the majority of European banking groups with a significant presence in the CESE region, as well as a substantial share of the banking sector assets in most CESE host countries. The total assets of the sample subsidiaries represent, on average, about 50 percent of the host country’s total banking system assets, with significant coverage within EU countries (about 60 percent) (Figure 11.3).

Figure 11.3Total Assets of Sample CESE Subsidiaries

(in percent of total banking assets of the host country, December 2008)

Sources: Authors; Bankscope; and national authorities.

Note: CESE = Central, Eastern, and Southern Europe.

That said, the total assets of subsidiaries included in the sample do not necessarily capture the full CESE exposures of these banking groups. This is because the latter could also include the parent banks’ direct cross-border lending to CESE countries, as well as lending by the branches operating in the host countries alongside the sub-sidiaries.10 In order to capture these exposures, we used the aggregate Bank for International Settlements (BIS) data on foreign claims by reporting banks on the CESE countries to impute the residual exposures of the banking groups to the CESE countries, including both direct cross-border exposures and exposures through branches (Figure 11.4). The data used for this purpose come from the BIS consolidated international banking statistics (see Appendix III for details).

Figure 11.4Total Foreign Claims of Sample Banking Groups on the CESE Countries

(in percent of host country GDP, December 2008)

Sources: Authors; Bank for International Settlements; Bankscope; and national authorities.

Note: Direct cross-border and branches claims were calculated as a residual from total foreign claims (see Appendix III).

CESE = Central, Eastern, and Southern Europe.

For each banking group, end-2008 data on total assets, total customer loans, profits, nonperforming loans (NPLs), loan loss provisions, regulatory capital, Tier 1 capital, and risk-weighted assets were collected at the group level, at the parent bank level, and at the level of individual CESE subsidiaries (when available). The main data sources include Bankscope, Bloomberg, and individual bank reports.

2. Calibration of the Cese Regional Shock

A. Country-level data

The CESE credit shock is modeled as deterioration in macroeconomic conditions during 2009–10, leading to an increase in NPLs and a decrease in the returns on assets (ROAs) of the CESE subsidiaries. The simulation of the shock relies largely on the actual data for 2009 and on projections for the CESE country-level NPLs and ROAs for 2010, which assume a slower economic recovery than the one envisaged in the IMF’s April 2010 World Economic Outlook (WEO) forecasts.

Changes in NPLs and ROAs are linked to the changes in macroeconomic conditions via panel regression models. The rationale behind this approach is to use consistent data across countries to come up with a specification that captures historical NPL and ROA patterns in the CESE region rather than fitting country-specific dynamics separately or extrapolating from past crises that occurred in other regions.

The panel regression analysis uses annual data for all CESE countries for the period of 1999–2009. The data for the dependent variable, aggregate NPLs at the country level, comes from Table 24 of the Statistical Appendix in the IMF’s Global Financial Stability Reports (IMF, 2010a, table 24). The data are examined for possible structural breaks and inconsistencies in the NPL definitions across countries.11

The GDP data come from the IMF’s WEO database. Interest rate data come from different sources: when T-bill rates are not available, the Money Market Rates (Serbia, Ukraine, and Estonia) or other comparable interest rates (Bosnia and Herzegovina, Bulgaria, Romania, Slovakia, and Belarus) are used instead (Appendix V). Table 11.2 shows descriptive statistics for the variables used in the regression analysis.

Table 11.2Descriptive Statistics
Variables (all in percent)NMSDMin.Max.
NPLs1857.67.30.235.8
GDP growth1983.04.9−18.512.2
Interest rate18710.512.91.493.2
ROA1791.21.3−6.13.3
Sources: IMF (2010a); and international financial statistics.Note: N denotes the number of observations. NPLs = nonperforming loans; ROA = return on assets.
Sources: IMF (2010a); and international financial statistics.Note: N denotes the number of observations. NPLs = nonperforming loans; ROA = return on assets.

B. Regression analysis

For the NPL model, from a large set of macroeconomic variables that could potentially influence the NPL dynamics, only two variables turned out to have reliable predictive power, namely, real annual GDP growth rates and short-term interest rates. Other potential explanatory variables (e.g., inflation rates, output gaps, private sector credit growth as a share of GDP, long-term nominal interest rates, the real effective exchange rate overvaluation, foreign currency debt as a share of GDP, government debt, and real government revenue) were either not significant or not available for the entire sample period. For the ROA model, the same variables (GDP growth, short-term interest rates) as well as the NPL ratios were statistically significant.

The estimated dynamic panel model for NPLs is as follows:

where npl stands for nonperforming loans (as a share of total loans), ΔGDP for the real GDP growth on an annual basis, and interest for the nominal short-term interest rate; ε is the error term, i = 1,2, … ,N denotes the country, and t= 1,2, … T denotes the time period.

Three specifications of model (11.1)—fixed effects, Arellano and Bond (1991), and Arellano and Bover (1995)—are considered, yielding similar results (Table 11.3). For the simulation of shocks, the Arrellano-Bover dynamic panel specification is chosen, given its better asymptotic properties for small-t and large-N dynamic panels than fixed effects and the fact that it minimizes the data loss compared with the Arrellano-Bond dynamic panel specification.

Table 11.3Dynamic Panel Regression Output for Nonperforming Loans
VariableFixed EffectsArellano-BondArellano-Bover
NPL (t – 1)0.6116**0.6342**0.7198**
GDP (t)−0.2633**−0.2930**−0.3576**
Interest (t)0.1241**0.1437**0.0714*
Constant2.1183**1.8891**2.1161**
Number of observations170143161
Number of groups181818
R20.69N/AN/A
Wald χ2N/A361424
Source: Authors.Note: NPL = nonperforming loan;

p < .05.

p < .01.

Source: Authors.Note: NPL = nonperforming loan;

p < .05.

p < .01.

Unsurprisingly, NPLs display high persistence. The signs of the explanatory variables are also as expected, with a negative sign for GDP growth rates and a positive sign for interest rates. Real GDP growth rates have the most significant impact on NPLs among the macroeconomic variables included in the regression equation. Overall, the model-based NPLs match the historical NPL patterns quite well. The model works worst for Estonia, where the model-based NPLs are substantially higher than the observed ones, which is likely to be partly driven by the definition of NPLs (Appendix IV).

For the ROA model, all explanatory variables are highly significant in all specifications, except for the lagged ROAs in the fixed-effects and the Arellano-Bover specifications (Table 11.4). All signs are as expected and in line with other studies (e.g., Babihuga, 2007). The Arellano-Bover model is the preferred specification.

Table 11.4Dynamic Panel Regression Output for ROAs
VariableFixed EffectsArellano-BondArellano-Bover
ROA (t−1)0.08560.1326*0.0834
GDP (t)0.0723**0.0840**0.0768**
Interest (t)−0.0523**−0.0318**−0.0230**
NPLs (t)−0.0479**−0.0514**−0.0799**
Constant1.7519**1.5055**1.7181**
Number of observations157139157
Number of groups181818
R20.52N/AN/A
Wald χ2N/A146232
Source: Authors.Note: NPLS = nonperforming loans; ROA = return on assets;

p < .05

p < .01

Source: Authors.Note: NPLS = nonperforming loans; ROA = return on assets;

p < .05

p < .01

C. Calibration of NPLs and ROAs

The baseline scenario uses the actual 2009 NPL ratios for the banking sectors of the CESE countries and the 2010 NPL ratios projected using the Arellano-Bover regression estimation, the WEO forecast of GDP, CPI (consumer price index), and LIBOR six-month (in €) for 2010.

The adverse scenario uses the actual NPL data for 2009 (as in the baseline) and the 2010 NPL projections based on the assumption that for each of the CESE countries, the 2010 GDP growth rate is 2 percentage points lower than the 2010 April WEO GDP growth rate forecasts, and the 2010 interest rate is 200 basis points higher than in 2009. Given the dominant role of GDP in the regression specification that is used to calibrate the shock, the adverse scenario features a slow recovery and high NPLs in both years, with most of the NPL increase taking place in 2009. The NPLs estimated under the 2010 baseline scenario are broadly in line with the estimates in the IMF’s Global Financial Stability Report (IMF, 2010a).12 Overall, the adverse scenario can be characterized as relatively “mild” among the plausible adverse scenarios, not necessarily too far away from the baseline (Table 11.5).

Table 11.5Country-Specific NPL Assumptions
Region2008

Median
2009

Median1
2010

Baseline Median2
2010

Adverse Median3
Baltic countries3.616.415.916.8
CEE-33.35.65.97.0
CIS3.89.68.19.1
SEE4.36.47.48.3
Sources: Authors; and IMF (2010a).Note: CEE-3 = 3 Central and Eastern Europe sample countries; CIS = Commonwealth of Independent States; NPL = nonperforming loan; SEE = Southeastern Europe; WEO = World Economic Outlook.

The 2009 provisional data come from Table 24 in the Statistical Appendix of IMF (2010a).

Baseline scenario uses NPLs estimated via dynamic panel regression using CESE 1999–2008 data. WEO assumptions are used for out-of-sample forecasts.

Adverse scenario assumes a slow recovery, that is, 2010 GDP growth is 2 percentage points below the 2010 WEO GDP growth forecasts, and interest rates are 200 basis points higher than in 2009.

Sources: Authors; and IMF (2010a).Note: CEE-3 = 3 Central and Eastern Europe sample countries; CIS = Commonwealth of Independent States; NPL = nonperforming loan; SEE = Southeastern Europe; WEO = World Economic Outlook.

The 2009 provisional data come from Table 24 in the Statistical Appendix of IMF (2010a).

Baseline scenario uses NPLs estimated via dynamic panel regression using CESE 1999–2008 data. WEO assumptions are used for out-of-sample forecasts.

Adverse scenario assumes a slow recovery, that is, 2010 GDP growth is 2 percentage points below the 2010 WEO GDP growth forecasts, and interest rates are 200 basis points higher than in 2009.

For subsidiaries, 2009–10 profits are calculated by taking the actual 2008 preprovision profits of individual subsidiaries as a base and applying the same rate of change as that of the country-level ROAs calculated based on the regression model (Table 11.6). The regression-based estimates of ROAs are adjusted downward for Slovenia, Slovakia, Belarus, and Bosnia and Herzegovina and upward for Romania and Bulgaria.13

Table 11.6Country-Specific ROA Assumptions
Region2008

Median
2009

Median1
2010

Baseline Median2
2010

Adverse Median3
Baltic countries1.2−0.10.0−0.2
CEE-31.21.11.21.0
CIS-31.40.51.00.8
SEE1.70.91.00.8
Sources: Authors; and IMF (2010a).Note: CEE-3 = 3 Central and Eastern Europe sample countries; CIS = Commonwealth of Independent States; ROA = return on assets; SEE = Southeastern Europe; WEO = World Economic Outlook.

The 2009 provisional data come from Table 26 of the Statistical Appendix in IMF (2010a).

Baseline scenario uses ROAs estimated via dynamic panel regression using CESE 1999–2008 data.

Adverse scenario assumes a slow recovery, that is, 2010 GDP growth is 2 percentage points below the 2010 WEO GDP growth forecasts, and interest rates are 200 basis points higher than in 2009.

Sources: Authors; and IMF (2010a).Note: CEE-3 = 3 Central and Eastern Europe sample countries; CIS = Commonwealth of Independent States; ROA = return on assets; SEE = Southeastern Europe; WEO = World Economic Outlook.

The 2009 provisional data come from Table 26 of the Statistical Appendix in IMF (2010a).

Baseline scenario uses ROAs estimated via dynamic panel regression using CESE 1999–2008 data.

Adverse scenario assumes a slow recovery, that is, 2010 GDP growth is 2 percentage points below the 2010 WEO GDP growth forecasts, and interest rates are 200 basis points higher than in 2009.

For parent banks, the 2009 net profits are either the actual numbers or estimates based on market consensus forecasts.14 The 2010 profits are assumed to be equal to the 2009 profits, provided that the latter were positive, and zero otherwise. Although this assumption is ad hoc, it is fairly neutral and is unlikely to introduce an upward bias in the estimates of capital needs.

3. Assessing Bank Capital Needs Under Alternative Ring-Fencing Scenarios

A. Methodology

This section presents the method applied to calculate capital adequacy under stress and capital requirements, respectively.

The loan loss reserve (LLR) for subsidiary k located in a CESE country i following a credit shock is as follows:

where NPLk,i is the postshock NPL ratio15 (nonperforming loans in percent of total exposure), Ek,i is the total exposure (customer loans), and LGDi is the loss given default (assumed to be the same for all subsidiaries operating in country i). Because bank-level end-2008 NPL data are not available for the majority of subsidiaries in the sample, the country-level end-2008 NPL ratios are used to proxy for the preshock bank-level NPL ratios. Country-specific LGDs come from the World Bank’s Doing Business Web page. In order to account for the empirical finding that LGDs tend to increase during economic downturns, a formula proposed by the Board of Governors of the Federal Reserve (2006) to derive downturn LGDs is applied to country-specific LGDs.16

For each subsidiary, the capital need is defined as the amount of capital required to bring its postshock CAR back to the country-specific (Basel II) regulatory minimum level.17

The CAR of subsidiary k located in a CESE country i before the shock is

The postshock CARs of subsidiaries reflect the impact of losses and possible underprovisioning (net of preshock LLR and preprovision profits at the subsidiary level) and an adjustment in risk-weighted assets (RWA) after the shock (denoted ΔRWAk,i). More specifically, the postshock CAR of subsidiary k located in a CESE country i is as follows:

where

Under the standardized approach, ΔRWAk,i is negative and corresponds to written-off losses.18,19

Then, the capital need (CN) at the group level is defined as the total amount of capital required to restore the CARs of all of the group’s affiliates to their regulatory minimums. Clearly, the extent to which this can be done by reallocating excess profits or excess capital within a group would reduce the need for fresh capital at the group level. This, in turn, would depend on the degree of ring-fencing within a group.

The precise definitions of the banking groups’ capital needs related to their CESE subsidiaries under different ring-fencing assumptions are presented in Table 11.7. Excess profit refers to the residual profit of a given subsidiary after it covers its own capital needs, and excess capital refers to the capital cushion above the regulatory minimum. The excess profits/capital and losses of the subsidiaries are computed by taking into account the parent bank’s ownership stake in each of these subsidiaries.

Table 11.7Definitions of Capital Needs under Four Ring-Fencing Scenarios
Degree of Ring-FencingCapital Needs after a CESE Credit Shock (if positive)
No ring-fencingCN(1) = sum of capital needs of all CESE subsidiaries – sum of excess profits and capital of all CESE subsidiaries – profits of the parent bank
Partial ring-fencingCN(2) = sum of capital needs of all CESE subsidiaries – sum of excess profits of all CESE subsidiaries – profits of the parent bank
Near-complete ring-fencingCN(3) = sum of capital needs of all CESE subsidiaries – profits of the parent bank
Stand-alone subsidiarizationCN(4) = sum of capital needs of all CESE subsidiaries
Source: Authors.Note: CESE = Central, Eastern, and Southern Europe; CN = capital needs.
Source: Authors.Note: CESE = Central, Eastern, and Southern Europe; CN = capital needs.

Note that under the SAS scenario, the capital needs of the banking group are equal to the sum of the recapitalization needs of all its CESE subsidiaries. Because the SAS approach presumes that all subsidiaries are self-sufficient, the capital needs under SAS would have to be covered by individual subsidiaries themselves, either by raising funds in the capital market or from other sources (such as the assistance from the local authorities).

As an additional exercise, the capital needs of sample banks are also computed by taking into account the parent bank’s losses from direct cross-border exposures to the CESE countries. This is done in two steps: first, the 2009 cross-border losses are estimated as direct cross-border exposures * additional NPLs * LGD; second, the 2009 net profits of parent banks are adjusted by adding back 50 percent of the 2009 estimated cross-border losses to the actual or estimated 2009 net profits (50 percent provisioning assumption).20 This conservative approach is used because the data on provisions on direct cross-border exposures of parent banks are not available.

B. Results

Using the framework and assumptions described above, we first computed the capital needs of the sample banking groups for their indirect CESE exposures via subsidiaries, assuming that subsidiaries have to restore their postshock CARs back to the regulatory minimum levels. Next, the banking groups’ capital needs are computed for both indirect exposures via subsidiaries and direct cross-border exposures to the CESE region. Finally, the simulations are repeated using a different definition of capital needs for the subsidiaries, namely, the amount of capital required for the subsidiaries to restore the postshock CARs back to the preshock (end-2008) levels; because these precrisis CARs were generally above the regulatory minimums, capital needs computed under this alternative definition are higher.

Focusing only on indirect exposures via subsidiaries, the results are as follows (see Figure 11.5):

Figure 11.5Estimated Capital Needs Resulting from a CESE Shock—Only Indirect Exposures via Subsidiaries in the Sample

(as a percent of the group’s regulatory capital)

Source: Authors.

Note: CESE = Central, Eastern, and Southern Europe; CN = capital need.

  • Eight out of 25 banking groups have no capital needs related to their CESE subsidiaries (i.e., CN(4) = 0).

  • Five out of 25 banking groups have significant capital needs related to their CESE subsidiaries (i.e., CN(4) > 10 percent of the banking group’s regulatory capital). As expected, the capital needs of the banking groups to ensure adequate capitalization of all parts of the group after the shock are higher under near-complete/partial ring-fencing than under no ring-fencing, with the differences being larger for more diversified groups. For example, one of the banking groups (#24), which faces the CESE-related capital needs of over 20 percent of its regulatory capital under the SAS model (CN(4)), has zero capital needs under no ring-fencing (CN(1)). In the cases when the parent banks’ profits are zero/negative (meaning that they cannot provide support for their subsidiaries), CN(3) = CN(4). More generally, in the no ring-fencing scenario (which allows reallocation of both excess profits and capital), only 5 out of 25 banking groups would still face nonzero capital needs after reallocation, compared with 17 in the case of the SAS (where no transfers are allowed within a group).

Figure 11.6 presents the estimated capital needs taking into account direct cross-border exposures and lending via branches, in addition to the exposures via subsidiaries in the sample. Although the CN measures in Figure 11.6 are notably higher than in Figure 11.5, the results are broadly similar, that is, more ring-fencing entails larger capital needs for most banking groups, with 9 banks (in the no ring-fencing scenario) to 22 banks (in the SAS scenario) in need of extra capital.

Figure 11.6Estimated Capital Needs Resulting from a CESE Shock—Indirect Exposures via Subsidiaries and Direct Cross-Border Exposures

(as a percent of the group’s regulatory capital)

Source: Authors.

Note: CESE = Central, Eastern, and Southern Europe; CN = capital need.

Recall that the simulations presented in Figures 11.511.6 use the definition of capital needs, which requires that the postshock CARs of all affiliates of the banking groups in the sample are restored to their respective country regulatory minimum levels. One could argue, however, that it may be prudent to maintain higher-than-regulatory minimum capital buffers, especially in crisis times, when uncertainty about credit quality is particularly high. To account for this possibility, capital needs are recomputed assuming that the postshock CARs of all affiliates of the banking groups have to be restored back to their preshock levels. Although the latter means that the overall capital needs are somewhat higher, the main result—that in the case of ring-fencing/SAS, the sample banks’ aggregate capital needs are much higher than in the case of no ring-fencing—still holds (Figure 11.7).

Figure 11.7Aggregate Capital Needs of Sample Banks Resulting from a CESE Shock

(in billions of U.S. dollars)

Source: Authors.

Note: Top panel shows only indirect exposures via subsidiaries in the sample; bottom panel shows indirect exposures via subsidiaries + direct cross-border exposures.

CAR = capital adequacy ratio; CESE = Central, Eastern, and Southern Europe;

CN = capital needs; REG MIN = regulatory minimum.

To sum up, the results shown in Figures 11.511.6 suggest that in the case of ring-fencing/SAS, the sample banks’ aggregate capital needs resulting from a CESE shock are 1.5–3 times higher than in the case of no ring-fencing (see Figure 11.7). These estimates appear to be within the range of the loss/recapitalization estimates provided by private analysts at end-2008 or early 2009 for different economic downturn scenarios in Eastern Europe during 2009–10.21

4. Conclusion

The simulation of the capital needs of 25 large European banking groups resulting from a credit shock affecting their CESE subsidiaries, under different ring-fencing scenarios, shows that these groups would need to have substantially higher capital buffers at the parent and/or subsidiary level if they face a risk of being unable to transfer capital and/or profits across borders. The extent to which this would have a material impact on individual groups would depend on the significance of the CESE subsidiaries for each group as a whole.

As discussed above, the use of ring-fencing by regulators, as well as proposals, such as the SAS approach, have been motivated largely by the difficulty of resolving cross-border banking groups and the lack of agreed-upon principles on cross-border resolution and burden sharing. This highlights the policy dilemma that emerges from tensions between the increasingly international nature of banking activities and the national nature of regulatory/legal frameworks.

There seem to be two possible alternative paths going forward:

  • A credible international resolution and burden-sharing mechanism would reduce the need for and incidence of ring-fencing and allow greater scope for cross-border banking groups to manage their capital and liquidity in a centralized manner, provided that it is in line with the bank’s business model.

  • In the absence of such burden-sharing mechanisms, regulators would need to consider setting minimum capital requirements for cross-border banking groups at a higher level, taking into account the possibility of ring-fencing, especially in crisis situations. In order to ensure that a banking group is resilient to all types of ring-fencing, all legally independent parts of the group (parent company and subsidiaries) would have to hold capital buffers that ensure self-sufficiency. The latter is likely to have implications for the banking group structures, the scope/scale of their cross-border activities, the supply of credit, and competition between local and foreign banks.

The choice between (1) and (2) would ultimately depend on the balance of potential benefits and costs, including those associated with higher capital requirements, as well as possible political constraints involved in choosing either one of the two paths. In particular, the European experience before the crisis showed that reaching an agreement on a credible resolution and burden-sharing mechanism may be challenging. However, the crisis provided additional impetus toward reaching an international consensus on a cross-border resolution framework, as evidenced by the ongoing work of the Financial Stability Board (on principles for cross-border resolutions), as well as of other international organizations.22

Appendix I. Capital Adequacy Rates by Country and Bank Type
Table 11.8Average CARs of Foreign-Owned Sample Subsidiaries versus the Country-Level Average CARs
CountryCAR (Sample subsidiaries)1CAR (Foreign-owned and domestic banks)2
Albania16.417.2
Belarus14.321.8
Bosnia and Herzegovina14.016.3
Bulgaria13.614.9
Croatia14.415.4
Czech Republic11.012.3
Estonia15.213.3
Hungary9.411.1
Latvia12.311.8
Lithuania11.512.9
Poland10.910.8
Romania13.713.8
Russia15.016.8
Serbia18.621.9
Slovakia10.011.1
Slovenia11.711.7
Turkey15.818.0
Ukraine16.114.0
Cross-country average13.514.7
Sources: Bankscope; Global Financial Stability Report, IMF; and individual bank reports (sample).Note: CAR = capital adequacy ratio.

Authors’ calculations.

Global Financial Stability Report, IMF (2010a)

Sources: Bankscope; Global Financial Stability Report, IMF; and individual bank reports (sample).Note: CAR = capital adequacy ratio.

Authors’ calculations.

Global Financial Stability Report, IMF (2010a)

Appendix II. Sample of Banking Groups and Their Subsidiaries
Table 11.9Sample Description
Parent BankHome CountrySubsidiaryHost Country
1RZBAustriaRaiffeisen Banka DDSlovenia
2Bank AustriaAustriaUnicredit Banka Slovenija DD
3Hypo Alpe Adria GroupAustriaHypo Alpe-Adria-Bank DD
4KBCBelgiumNLB DD-NOVA Ljubljanska Banka
5SocGenFranceSKB Banka DD
6IntesaItalyBanka Koper DD
7Erste GroupAustriaCeska SporitelnaCzech Republic
8RZBAustriaRaiffeisenbank Akciova Spolecnost
9VolksbankAustriaVolksbank CZ
10Bank AustriaAustriaUnicredit Bank Czech Republic
11KBCBelgiumCeskoslovenska Obchodni Banka
12KBCBelgiumCeskomoravska Stavebni Sporitelna
13SocGenFranceKomercni Banka
14Erste GroupAustriaPRVA Stavebna Sporitelna ASSlovakia
15Erste GroupAustriaSlovenska Sporitel’na AS
16RZBAustriaTatra Banka AS
17VolksbankAustriaVolksbank Slovensko, AS
18Bank AustriaAustriaUnicredit Bank Slovakia AS
19KBCBelgiumCeskoslovenska Obchodna Banka
20RZBAustriaRaiffeisen Bank Polska SAPoland
21KBCBelgiumKredyt Bank SA
22DNB NorNorwayDNB Nord
23BNP ParibasFranceFortis Bank Polska SA
24CommerzbankGermanyBre Bank SA
25DeutscheGermanyDeutsche Bank Polska SA
26Allied Irish BksIrelandBank Zachodni WBK SA
27UnicreditItalyBank Polska Kasa Opieki Sa—Bank Pekao SA
28INGNetherlandsING Bank Slaski SA—Capital Group
29NordeaSwedenNordea Bank Polska SA
30Bank AustriaAustriaYapi VE Kredi Bankasi ASTurkey
31DexiaBelgiumDenizbank AS
32BNP ParibasFranceTurk Ekonomi Bankasi AS
33Credit AgricoleFranceCalyon Yatirim Bankasi Türk AŞ
34DeutscheGermanyDeutsche Bank AS
35Eurobank EFGGreeceEurobank Tekfen
36NBGGreeceFinansbank AS
37INGNetherlandsING Bank AS
38RZBAustriaRaiffeisenbank (Bulgaria) EADBulgaria
39Bank AustriaAustriaUnicredit Bulbank
40KBCBelgiumCIBANK PLC
41SocGenFranceSociete Generale Expressbank
42Credit AgricoleFranceEmporiki Bank—Bulgaria EAD
43CommerzbankGermanyProcredit Bank (Bulgaria) AD
44Eurobank EFGGreeceEurobank EFG Bulgaria AD (Postbank)
45NBGGreeceUnited Bulgarian Bank—UBB
46Allied Irish BksIrelandBulgarian-American Credit Bank
47Erste GroupAustriaErste Bank Hungary NYRTHungary
48RZBAustriaRaiffeisen Bank ZRT
49VolksbankAustriaMagyarorszagi Volksbank RT
50Bank AustriaAustriaUnicredit Bank Hungary ZRT
51KBCBelgiumK&H Bank ZRT
52Bayern LBGermanyMKB Bank ZRT
53DeutscheGermanyDeutsche Bank Zártköruen Muködo Részvénytársaság
54IntesaItalyCIB Bank ZRT
55Erste GroupAustriaBanca Comerciala Romana SARomania
56RZBAustriaRaiffeisen Bank SA
57VolksbankAustriaVolksbank Romania
58Bank AustriaAustriaUnicredit Tiriac Bank
59SocGenFranceBRD-Groupe Societe Generale SA
60Credit AgricoleFranceEmporiki Bank—Romania SA
61CommerzbankGermanyProcredit Bank SA
62AlphaGreeceAlpha Bank Romania
63Eurobank EFGGreeceBancpost SA
64NBGGreeceBanca Romaneasca SA
65IntesaItalyIntesa Sanpaolo Romania SA
66RZBAustriaRaiffeisen Bank AlbaniaAlbania
67SocGenFranceBanka Popullore SHA
68CommerzbankGermanyProcredit Bank (Albania) SHA
69PiraeusGreeceTirana Bank SA-Banka e Tiranes SHA
70IntesaItalyIntesa Sanpaolo Bank Albania
71Bank AustriaAustriaUnicredit BankBosnia and Herzegovina
72Hypo Alpe Adria GroupAustriaHypo Alpe—Adria—Bank AD, Mostar
73Hypo Alpe Adria GroupAustriaHypo Alpe—Adria—Bank AD Banja Luka
74Erste GroupAustriaErste & Steiermärkische Bank DDCroatia
75RZBAustriaRaiffeisenbank Austria DD, Zagreb
76VolksbankAustriaVolksbank DD
77Bank AustriaAustriaZagrebacka Banka DD
78Hypo Alpe Adria GroupAustriaHypo Alpe—Adria—Bank DD
79Hypo Alpe Adria GroupAustriaSlavonska Banka DD, Osijek
80SocGenFranceSociete Generale—Splitska Banka
81Erste GroupAustriaErste Bank AD Novi SADSerbia
82RZBAustriaRaiffeisenbank AD
83Bank AustriaAustriaUnicredit Bank Serbia JSC-Unicredit
Bank Srbija AD
84Hypo Alpe Adria GroupAustriaHypo Alpe—Adria—Bank AD Beograd
85SEBSwedenSEB PankEstonia
86SwedbankSwedenSwedbank AS
87Bank AustriaAustriaUnicredit Bank ASLatvia
88DNB NordDenmarkDNB Nord
89SEBSwedenSEB Banka AS
90SwedbankSwedenSwedbank AS
91DNB NordNorwayDNB NordLithuania
92SEBSwedenSEB Bankas
93SwedbankSwedenSwedbank AS
94RZBAustriaZAO RaiffeisenbankRussia
95Bank AustriaAustriaUnicredit Bank ZAO
96KBCBelgiumAbsolut Bank
97BNP ParibasFranceBNP Paribas Vostok
98SocGenFranceBank Société Générale Vostok
99SocGenFranceJSC Rosbank
100CommerzbankGermanyCommerzbank (Eurasija)
101IntesaItalyKMB Bank/ Small Business Credit Bank
102INGNetherlandsING Bank (Eurasia) ZAO
103NordeaSwedenNordea Bank
104SwedbankSwedenSwedbank
105Erste GroupAustriaErste Bank OJSCUkraine
106RZBAustriaRaiffeisen Bank Aval
107Bank AustriaAustriaUkrsotsbank
108BNP ParibasFranceJSIB Ukrsibbank
109IntesaItalyPravex Bank
110INGNetherlandsING Bank Ukraine
111SwedbankSwedenPublic Joint Stock Company Swedbank
112RZBAustriaPriorbankBelarus
113SocGenFranceBelrosbank
Sources: Analysts’ reports; Bankscope; and individual bank reports.
Sources: Analysts’ reports; Bankscope; and individual bank reports.
Appendix III. Using BIS Data to Capture Remaining Banking Groups’ Exposures to the CESE Countries

The portfolio of subsidiaries included in the sample does not necessarily capture the full CESE exposures of the sample banking groups, as it is missing their direct cross-border claims and lending through their branches in the host countries. In order to capture these exposures, the subsidiaries’ balance sheet data are combined with the consolidated Bank for International Settlements (BIS) international banking statistics, which report, at the country level, the consolidated claims of BIS reporting banks on each CESE country.

The BIS total foreign claims data (on the immediate borrower’s basis) include both the cross-border claims of foreign banks and all foreign affiliate—subsidiaries and branches—claims. In order to separate direct cross-border and branches’ claims, we subtract the domestic lending to customers by sample subsidiaries from the total foreign claims (e.g., the customer lending by Greek-controlled subsidiaries operating in Turkey is subtracted from the consolidated foreign claims of the BIS reporting Greek banks on Turkish residents).

A few caveats need to be highlighted:

  • The residual labeled as “direct cross-border and branches claims” could also include nonlending claims by subsidiaries, such as government bonds held by a subsidiary. Nevertheless, the potential recap needs calculated for each subsidiary were based on their lending portfolio, and hence, the treatment of other nonlending claims as claims directly linked to the parent bank seems appropriate.

  • As documented in other papers, there are some discrepancies between the BIS data and other sources. In our case, we found that there would be a potential small downward bias in the BIS total claims of Italian banks on Ukraine and Swedish banks on Latvia, based on the subsidiaries balance sheet data. This is similar to the discrepancies found by Maechler and Ong (2009) when comparing the BIS data with the central bank data sources. These discrepancies could be due to the differences in the consolidation method and the group structure classification used by different internationally active banks.

  • The residual exposure identified at a parent bank’s home-country level is distributed across the sample banking groups of each home country as a proportion of the group’s assets. Because the sample of banking groups in this study was put together with the objective to capture all major European cross-border banking groups with significant presence in the CESE region, it is unlikely that the BIS data include large cross-border banks with operations in the CESE region that are not included already in the sample.

Appendix IV. Panel Regression Analysis

Basic Specification

We estimated the fixed-effects and random-effects specifications of the following panel regression model:

where npl stands for nonperforming loans (as a share of total loans), ΔGDP for the real GDP growth on an annual basis, and interest for the short-term nominal interest rates (T-bill rate); ε is the error term, i = 1,2,..N denotes the country, and t = 1,2,…,T denotes the period. The results are presented in Table 11.10. Based on the Hausman test, the random-effects specification is more appropriate than the fixed-effects specification.

Table 11.10Panel Regression Analysis (Dependent Variable: NPLs)
VariableFixed EffectsRandom Effects
GDP (t)−0.1864**−0.1797**
Interest (t)0.2325**0.2301**
Constant5.8369**5.7222**
Number of observations178178
Number of groups1818
R20.230.23
Source: Authors.Note:

p < .01.

Source: Authors.Note:

p < .01.

Robustness Checks

In order to investigate the differences in NPL definitions across countries, we compared the average level of bank provisions with NPLs over time. For most of the 18 countries, this ratio varies between 60 percent and 90 percent. In the case of Estonia, where the level of provisions ranges between 150 percent and 300 percent of NPLs during 2003–06 (the years for which the data are available), the level of NPLs seems particularly low, suggesting that the definition of NPLs in Estonia may be different from the one used in other countries. Estonia has, therefore, been excluded from some regression specifications as a means of a robustness check. In addition, the ratio of provisions to NPLs in Russia fluctuated between 120 percent and 160 percent during 2003–08, which can again be seen as a sign that the definition of NPLs is relatively narrow (i.e., not as conservative) as in other countries. For Romania and Ukraine, the opposite seems to be true, which indicates that NPLs may be based on a fairly broad definition of potential losses.

For countries with structural breaks (Latvia, Lithuania, Poland, Slovakia, Serbia, Ukraine), we also checked whether there were any major changes in the levels of NPLs observed around the timing of structural breaks in definitions. Because this was not the case, all NPL data from these countries were included in the regression models. Similarly, the regression results remained robust when excluding countries with structural breaks.

Appendix V. Interest Rates Used for Regression Analysis
Table 11.11Overview on Data Sources for Short-Term Interest Rates
CountryExplanationSource
AlbaniaTreasury bill rateIFS
BelarusData are scarce; adjusted deposit rates according to money market rates to deposit rates in UkraineIFS
Bosnia and HerzegovinaInterest rate is constructed as 2/3 times deposit rate + 1/3 times lending rate, starting in 2002 (other data are not available)MBS
BulgariaTreasury bill rate when available (2 years); otherwise inferred from interbank rate (3 months, Bloomberg); otherwise from lending rateMBS and Bloomberg
CroatiaTreasury bill rate from Bloomberg (from 2002 onward); before that, adjust money market rate weighted by the relative portion of T-Bill rates on money market rates from 2002 to 2008IFS (for 1992 to 2001), Bloomberg (from 2002)
Czech RepublicTreasury bill rateMBS
EstoniaMoney market rateIFS
HungaryTreasury bill rateMBS
LatviaTreasury bill rateMBS
LithuaniaTreasury bill rate (12 months) where possible; otherwise government bond yield (12 months, from Bloomberg), recalculated to 12 months level; otherwise interbank rate, recalculated to 12-month T-Bill rateMBS and Bloomberg
PolandWeighted average Treasury bill rateMBS
RomaniaInterbank short-term lending rate, similar to Treasury bill rate (91 days) (which is available only for some years)MBS
RussiaTreasury bill rate for 2000–3; after 2003, average of money market rate and interbank rate (3 months); before 1999, money market rateMBS, Bloomberg
SerbiaMoney market rateIFS
SlovakiaAverage deposit rate (1993–99), government bond yield (2002–6), government bond yield (1 year, from Bloomberg, 2007–9)MBS, Bloomberg
SloveniaTreasury bill rate (from 1999); before that, money market rate adjusted to Treasury bill rateMBS
TurkeyTreasury bill rate, if available; (before 1999: money market rate); after 2007, 1-year government bond yieldIFS, Bloomberg
UkraineMoney market rateIFS
Sources: Bloomberg; International Financial Statistics (IFS); and Money and Banking Statistics (MBS).
Sources: Bloomberg; International Financial Statistics (IFS); and Money and Banking Statistics (MBS).
Appendix VI. Loss-Given-Default Ratios in the CESE Countries

The country-specific loss-given-default ratios (LGDs) are taken from the World Bank Doing Business Web page and are based on work of Djankov and others (2008). The differences in LGDs across countries reflect the differences in bankruptcy codes, the duration of the proceedings, and legal costs. The CESE long-term average LGDs range from 51 percent (Lithuania) to 91 percent (Ukraine). The CESE average of 68 percent is above the LGD for senior unsecured credit under the Basel II Foundation Internal Ratings–Based approach (45 percent), which is often used as a benchmark for a through-the-cycle LGD.

In order to account for the empirical finding that LGDs increase during downturn periods, we use a formula proposed by the Board of Governors of the Federal Reserve (2006) to derive the downturn LGDs:

The downturn LGDs for the CESE countries range from 55 percent (Lithuania) to 92 percent (Ukraine) (Table 11.12). The estimated CESE downturn LGDs are thus substantially higher than those observed for senior bank loans in the OECD countries, which range from 30 percent to 45 percent (see Moody’s 2010; Doing Business database).

Table 11.12Loss-Given-Default Ratios by Country
CountryLong-Term Average LGDsDownturn LGDs
Albania1N/A70.0
Belarus66.669.3
Bosnia and Herzegovina64.167.0
Bulgaria67.970.5
Croatia69.571.9
Czech Republic241.346.0
Estonia62.565.5
Hungary61.664.7
Latvia71.073.3
Lithuania50.654.6
Poland70.272.6
Romania71.573.8
Russia71.874.1
Serbia74.676.6
Slovakia54.157.8
Slovenia54.558.1
Turkey79.881.4
Ukraine90.991.6
Average68.370.7
Sources: http://www.doingbusiness.org and local authorities.Note: LGD = loss given default.

For Albania, the downturn LGD is assumed to be equal to its peer country average (70 percent).

For the Czech Republic, the average LGD is the one for corporate credit (Czech National Bank, 2009, p. 81).

Sources: http://www.doingbusiness.org and local authorities.Note: LGD = loss given default.

For Albania, the downturn LGD is assumed to be equal to its peer country average (70 percent).

For the Czech Republic, the average LGD is the one for corporate credit (Czech National Bank, 2009, p. 81).

Appendix VII. Regulatory Minimum Capital Requirements by Country
Table 11.13Regulatory Minimum Capital Requirements in the CESE Countries
CountryMinimum Capital Adequacy

Requirements (in percent)
Albania12
Belarus10
Bosnia and Herzegovina12
Bulgaria12
Croatia10
Czech Republic8
Estonia10
Hungary8
Latvia8
Lithuania10
Poland8
Romania10
Russia10
Serbia12
Slovakia8
Slovenia8
Turkey12
Ukraine8
Source: IMF staff calculations.Note: CESE = Central, Eastern, and Southern Europe.
Source: IMF staff calculations.Note: CESE = Central, Eastern, and Southern Europe.
Appendix VIII. Analysts’ Estimates of Losses and Capital Needs
Table 11.14Private Analysts’ Estimates of Losses and Capital Needs of Western European Banks on their CESE Exposures
SourcePublication DateNPL Scenario (%)LGD (%)PeriodLoss (billion EUR)Capital Needs
JP MorganMarch 5, 200910 to 301752 years32Average loss: 22 (in percent of equity)
BarclaysFebruary 27, 200910/20/302N/A2 years67/133/200EUR 100–150 billion
Deutsche BankFebruary 23, 20091 scenario, no details availableN/A2 years34Average loss: 17 (in percent of tangible common equity)
Goldman SachsJanuary 29, 200910/20/30/differentiated NPLs1,2N/AN/A19/62/106/59N/A
Merrill LynchOctober 27, 20087/11/20/332N/A3 years43EUR 20 billion
Source: Analysts’ reports.Note: CESE = Central, Eastern, and Southern Europe; LGD = loss given default; NPL = nonperforming loan.

Different NPL assumptions for different asset “buckets.”

Different NPL scenarios with uniform level of NPLs.

Source: Analysts’ reports.Note: CESE = Central, Eastern, and Southern Europe; LGD = loss given default; NPL = nonperforming loan.

Different NPL assumptions for different asset “buckets.”

Different NPL scenarios with uniform level of NPLs.

References

The interbank linkages in this region and their role in transmitting or mitigating shocks have been recently analyzed in Arvai, Driessen, and Otker-Robe (2009); Maechler and Ong (2009); and Hermann and Mihaljek (2010) but these analyses relied mainly on the country-level (not bank-level) data on cross-border bank exposures.

The current initiatives at the European Union (EU) level aim to achieve the dual objective of (1) preventing the risk of insolvency that could potentially be generated by a disproportionate transfer of assets for the credit institution making the transfer; and (2) lifting restrictions on transfers of assets if such transfers can potentially limit the extent of a crisis (European Commission, 2010).

See, for example, the “Statement at the End of the European Bank Coordination Initiative’s Second Full Forum Meeting,” Press Release No.10/106, March 22, 2010. In the context of this initiative, the large bank groups with systemic presence in several CESE countries have committed to maintain their exposure and keep their subsidiaries well capitalized (http://www.imf.org/external/np/sec/pr/2010/pr10106.htm).

To name a few examples, bank regulators in Croatia, Poland, and Turkey recommended the nondistribution of profits by the subsidiaries of foreign banks despite relatively strong bank fundamentals. In the case of Croatia, the Croatian National Bank (CNB) governor, Dr. Željko Rohatinski, at a press conference held on February 18, 2009, said that “the CNB would not look favorably upon attempts to withdraw capital, deposits, or pay out total accumulated profits, because that would destabilize the domestic banking system. In such a case, the CNB would be forced to undertake protective measures, regardless of the connected risks.” In the case of Turkey, the head of the banking regulation agency stated in December 2009 that “it is our natural right to expect those profits generated in this country to be invested and used in credit extension again in this country.” Banks in Turkey were expected to consult the regulator before distributing any dividends in the two years preceding 2009. The IMF Article IV report on Poland (IMF, 2010b) classified the regulator’s recommendation for subsidiaries of foreign banks to refrain from paying out dividends, despite robust capital buffers, as a form of capital control.

The issue of intragroup liquidity transfers is not considered in this study. It is left for future research.

At the onset of the crisis, many European parent banks had direct crossborder loans on their books, which sometimes had been purchased from the subsidiaries in the boom years. There is anecdotal evidence that suggests that the reverse happened during the crisis; that is, in some cases, subsidiaries with large capital buffers bought back loans from the parent banks, thereby reducing their capital adequacy ratios (CARs).

The choice between branches and subsidiaries has been analyzed in the literature. Cerutti, Dell’Ariccia, and Martinez-Peria (2007), for example, found that cross-border banking groups are more likely to set up a branch than a subsidiary in host countries with relatively higher corporate taxes, because this makes it easier to transfer profits across borders. Other considerations in the choice between branches and subsidiaries include the following: (1) branches are more common when foreign operations are smaller in size and do not have a retail orientation; (2) branches are less common in countries with highly risky macroeconomic environments, where parent banks seem to prefer the “hard” shield of limited liability provided by subsidiaries; (3) foreign banks tend to specialize in one organizational form or the other, beyond what is explained by their home-country regulation; and (4) foreign banks are less likely to operate as branches in countries that limit their activities and where regulation makes it difficult to establish new banks.

The postshock CAR is estimated by taking into account actual or projected losses, provisions, capital buffers, and possible increases in risk-weighted assets.

Data availability is another limiting factor. It should also be noted that for the majority of subsidiaries in the sample the ownership stake of the parent bank exceeds 50 percent.

There are some cross-border banking groups that conduct mainly direct cross-border lending instead of lending through subsidiaries/branches (see McCauley, McGuire, and von Peter, 2010).

See Appendix IV for details.

The adverse scenario used in this study is somewhat less severe than the adverse growth scenario in IMF (2010a), with NPLs that are 3–5 percentage points lower for the CEEs (Central and Eastern Europe), SEEs (Southeastern Europe), and the Baltic states. For the CIS, the figures are comparable, as the figures in this study include Belarus (with low NPLs), whereas IMF (2010a) does not.

The adjustment accounts for the fact that the returns have been lower (first group of countries) or higher (Romania and Bulgaria) than on average in the sample in the past, which could be triggered by the level of competition, for example. The adjustment was 0.4 (Bulgaria) and 0.7 (Romania) in positive terms as well as 0.5 (Belarus and Slovakia) and 0.7 (Bosnia and Herzegovina and Slovenia) in negative terms.

Parent banks’ net profits are used because parent bank’ non-CESE-related losses are not explicitly included in the simulations. As explained at the end of the next section, some adjustments are needed to parent banks’ net profits when including parent banks’ direct cross-border CESE losses to avoid double counting.

The stock (rather than the flow) of NPLs is considered in order to account for both possible underprovisioning as well as provisions on additional NPLs.

See Appendix VI for details.

The regulatory minimum CARs for the CESE countries are presented in Appendix VII.

An alternative approach—the Basel II Internal Ratings—Based (IRB) approach—was considered as well. According to the IRB approach, ΔRWAk,i takes into account the overall increase in the riskiness of performing loans that is likely to be associated with the increase in NPLs and hence can be positive despite the write-offs. The results based on the IRB approach are broadly similar to the standardized approach, with the estimated recapitalization needs being slightly higher under the IRB approach. The small difference between the two approaches results from the high level of credit risk. In “normal” times, the difference could be substantial.

Dividends are not explicitly modeled. It is implicitly assumed that most of the profits are retained.

The detailed discussion of how cross-border exposures are imputed from the Bank for International Settlements data is in Appendix III.

See Appendix VIII for details.

See, for example, IMF (2010c).

    Other Resources Citing This Publication