Book
Share
Chapter

Chapter 12 How to Capture Macro-Financial Spillover Effects in Stress Tests

Author(s):
Li Lian Ong, and Andreas A. Jobst
Published Date:
September 2020
Share
  • ShareShare
Show Summary Details
Author(s)
Heiko Hesse Christian Schmieder and Ferhan Salman

This chapter is based on IMF Working Paper 14/103 (Hesse, Salman, and Schmieder 2014). The authors thank Eugenio Cerruti, Martin Čihák, Amadou Sy, and participants at a Banque de France conference (September 2012) as well as at a meeting in May 2013 of the Liquidity Stress Testing Subgroup of the Basel Committee on Banking Supervision’s Research Task Force for helpful comments and suggestions.

One of the challenges of financial stability analysis and bank stress testing is how to establish scenarios with meaningful macro-financial linkages, that is, taking into account spillover effects and other forms of contagion. This chapter presents an approach to simulate the potential impact of spillover effects based on the “traditional” design of macroeconomic stress tests. Specifically, the chapter examines spillover effects observed during the financial crisis and simulates their impact on banks’ liquidity and capital positions. The outcome suggests that spillover effects have a highly nonlinear impact on bank soundness, both in terms of liquidity and solvency.

1. Introduction

Stress testing has garnered broad attention during recent years, which has spurred numerous conceptual developments.1 Yet, overarching approaches to establish macro-financial linkages, and explicitly capture the nonlinearity of shocks (originating from spillover effects and other types of contagion) are still evolving. Such linkages have seen a particularly significant growth during the last decade (for example, Frank, González-Hermosillo, and Hesse 2008) and are therefore an important dimension to be captured by meaningful empirical analysis. This chapter focuses on the design of stress tests to capture spillover effects and demonstrates the potential impact based on a case study.

The first part of the chapter deals with the establishment of macro-financial scenarios that are explicitly informed by spillover effects. Scenario design for macroeconomic stress tests is typically based on an “indirect approach” (Jobst, Ong, and Schmieder 2013, Figure 12.1): (1) first, economic and financial variables are estimated conditional on a macroeconomic scenario; (2) in the second step, the trajectories of the economic and financial variables are translated into bank solvency and liquidity2 measures based on so-called satellite or auxiliary models. Tree approaches have commonly been used to predict economic and financial variables under stress (see Foglia 2009): first, a structural econometric model; second, vector autoregressive methods; and third, pure statistical approaches. The satellite models commonly take the form of (panel) regression models. The “direct approach” is based on projections of the actual solvency and liquidity parameters without an explicit link to the state of economic and financial variables. While this approach could be equally meaningful in terms of the outcome of stress tests, it does not allow for a detailed storytelling and can underestimate the importance of nonlinear macro-financial factors for bank-specific stress tests.

Stylized Design of Stress Tests

Source: Authors.

Modeling contagion effects and their impact typically constitutes a challenge (see Jobst, Ong, and Schmieder 2013, for example). By definition, spillover effects and other dynamic contagion effects are implicitly captured in past data but not necessarily if one uses structural econometric models— usually perceived as being a “best practice.” Even if potential spillover events are captured in past data, this data might not be representative for a future scenario if, for example, linkages between economies and banks have become gradually more intense over time. This study focuses on spillover effects originating from the recent sovereign debt crisis. Other spillover catalysts could be, for instance, a macroeconomic downturn in a major world economy as well as the failure of a large financial institution such as in the case of Lehman Brothers.

The chapter aims to come up with a stress testing approach that captures spillover effects in detail. The solution is an amended version of the indirect approach: the starting point is to establish a macroeconomic scenario, typically not informed by potential spillover effects— at least not explicitly. In the second step, the potential marginal increase of stress due to spillover effects is estimated by translating the spillover effects into reduced output paths, that is, an adverse macroeconomic scenario.

The stylized design of macroeconomic stress tests (Figure 12.1) implicitly incorporates a quasi-feedback loop into the linear design of traditional stress tests— through a sensitivity-type approach.3 The approach could also include a test for interbank bank contagion, as shown in Figure 12.1. The chapter builds on previous IMF work to establish an explicitly iterative process, that is, establish a scenario informed by initial spillover effects based on a structural econometric approach, compute the impact on banks’ solvency parameters, recompute the resulting spillover effects and feed them back to the structural model, and so on, until an equilibrium is reached.4 The approach presented here uses proxies for the “ultimate” impact of spillovers for different advanced and emerging economies conditional on the evolution of sovereign spreads in the euro area periphery (that serves as the stress catalyst). Dynamic effects can also be captured via “direct” approaches, as done by Jobst and Gray 2013, for example, but renders the outcome a reduced-form type.5

Specifically, the chapter infers from market data the magnitude of sovereign spread spillovers effects resulting from an increase in peripheral EU sovereign debt spreads, while controlling for changes in the market sentiment (that is, risk aversion) and macroeconomic factors. Using market data, the chapter seeks to capture point-in-time and dynamic time series’ effects, while recognizing the limitations of using market data, that is, that they might not necessarily “only” reflect underlying vulnerabilities and risks. The translation of sovereign spread spillovers into a loss of output is based on recent work at the IMF (Vitek and Bayoumi 2011).

Two approaches are used to capture the spillover effects in sovereign debt markets: panel regressions and a generalized autoregressive conditional heteroskedasticity (GARCH) model. The panel regressions, which are used to establish an “average” impact of spillover effects during periods of stress on countries with advanced markets (AMs) and those with emerging markets (EMs), respectively, suggest that increasing sovereign risk in the euro periphery was a major driving force behind spillover effects. As expected, risk aversion, measured through changes in the VIX and high-yield spreads, is found to increase during periods of financial stress, exhibiting a nonlinear pattern. Country-specific macroeconomic factors also matter, but to a lesser degree, and their impact does not appear to change significantly under periods of stress.

GARCH models were run to obtain more granular spillover effects, such as the country-specific comovements between peripheral European GIIPS6 sovereign debt spreads and the corresponding spreads in the banks’ home countries (that is, the 25 most systemically important financial systems, the “S25” sample) for specific points in time. The study reveals significant differences in terms of the spillovers across countries, with a higher impact observed for most core euro area countries (in particular during peak periods of the crisis) than for Scandinavian countries, Switzerland, the United Kingdom and most non-European countries. The findings also show a fight-to-quality element, that is, a negative comovement of GIIPS spreads with German bunds and US Treasury bonds.

The second part of the chapter illustrates how the established spillover effects would feed through to banks based on a case study for 154 large international banks from the “S25” country sample. The impact of different degrees of spillover on banks’ solvency and liquidity positions is compared with baseline-type conditions (which correspond to realized stress scenarios in recent years, unlike in “normal” times). Stress at the bank level is simulated based on a recently developed IMF stress testing framework for liquidity (Schmieder and others 2012) and benefits from work on solvency (Schmie-der, Puhr, and Hasan 2011; Hardy and Schmieder 2013), which together allow running integrated solvency and liquidity scenarios.7

The outcome suggests that spillover effects have a highly nonlinear impact on bank soundness, both in terms of liquidity and solvency. It is thereby shown (once more) that the design of stress scenarios is a highly crucial element of stress testing, and is sensitive with respect to the outcome of stress tests.8 The magnitude of the impact on bank solvency and liquidity could serve as a benchmark for other studies, while recognizing that future spillover channels could be highly different, both in terms of direction and magnitude. In this sense, this study could help to identify potential systemic vulnerabilities ex ante, a role that stress tests have not necessarily played in the past for a number of reasons (see Borio, Drehmann, and Tsatsaronis 2012, for example).

The chapter is organized as follows. Section 2 investigates financial spillovers at the sovereign and bank level, based on panel regressions and a GARCH model framework. Section 3 provides a brief overview of the stress testing framework used to simulate the impact of spillover effects on bank liquidity and solvency. Section 4 shows the impact of different degrees of spillover based on a case study. Finally, Section 5 concludes and offers some avenues for future research.

2. Financial Spillovers from the Euro Periphery to the Rest of the World

Panel Approach

Financial market linkages across economies have grown significantly in recent decades, which was felt strongly when the financial crisis started in 2008 with the failure of Lehman Brothers, and later, when it continued to become a sovereign debt crisis, especially in the European periphery. AM financial spillovers have been a dominant determinant of AM and EM financial soundness during the previous years.

Recent studies identified three important factors for spillover effects (see, for example, Caceres and Unsal 2011): (1) a stress spillover catalyst— in this study AM sovereign debt yields; (2) risk aversion in global markets; and (3) country-specific risk factors.

This chapter seeks to establish benchmark parameters to simulate spillover effects at the bank level. Initially, a risk premium variable for the sample of 35 countries was constructed.9 The risk premium is the spread between 10-year domestic treasuries to US Treasury bonds for non-European AM countries, to German bunds for AM countries in Europe, and to the J.P. Morgan Emerging Markets Bond Index

for the EM countries.10

Based on random effects’ panel regressions the sovereign spreads are regressed on three sets of peripheral spreads: average spreads for (1) the European peripheral countries (GIIPS); (2) for the GIP (Greece, Ireland, Portugal); and (3) for IT- ES (Italy and Spain). Risk aversion is identified by two variables, high-yield spreads and the VIX. The former is the difference between yields to maturity of Moody’s Aaa-rated and Baa-rated US corporate bonds. The latter is the implied volatility for S&P 500 index options. Trade openness, liquidity (proxied by M2 to GDP and the level of reserves to GDP), inflation rates, GDP growth, the current account, the level of public debt and deficits-to-GDP ratios are used as macroeconomic control variables to capture country-specific cyclical effects.

The regressions are estimated for two time periods based on quarterly data: (1) 2006–12 and (2) 2008–12. The choice of the two sample periods is meant to capture the impact of the systemic stress.

The results (displayed in Appendix 12.1) present various model specifications considered useful to identify drivers of spillover stress and their actual impact, respectively. Using the sovereign debt spreads of the 35 sample countries as the dependent variable, Appendix Table 12.1.1 shows the outcome for 2006–12 and Appendix Table 12.1.2 for 2008–12.

The results confirm previous studies in that all three factors, that is, a catalyst, risk aversion, and country-specific factors, are actually important to explain financial stress (measured in terms of sovereign spreads), at least for the global financial crisis. Specifically:

  • Increasing sovereign risk in the euro periphery was found to be a catalyst for spillover effects.
  • The global perception of risk magnifes stress conditions as do expected future interest rates.
  • Country-specific macroeconomic factors also matter, but to a lesser degree.
  • While the impact11 of country-specific factors does not appear to change significantly under stress, the impact of the former two factors is higher during 2008–12, that is, in the period covering only the crisis years (compared to the full sample period).

For the longer sample period (that is, 2006–12) a 1-percentage-point change in euro periphery sovereign spreads (that is, GIIPS and GIP) translates into a 0.2- to 0.3-percentage-point change of sovereign debt spreads in the 35 sample countries (Appendix Table 12.1.1). Global risk aversion (measured by changes in high-yield spreads) has an even higher impact—a 1-percentage-point change in high-yield spreads translates into about a 0.6-percentage-point change in sovereign spreads. As global risk aversion and high-yield spreads are highly correlated during episodes of stress, the joint impact on the peripheral spreads is exacerbated, which is illustrated in a comparison of the coefficients in Appendix Tables 12.1.1 and 12.1.2. The transmission of risk premium shocks from Italy and Spain to the countries in the sample is more pronounced than for the GIPs. Depending on the model specification, the availability of domestic liquidity and trade openness also contribute to some degree to spillovers.12

The outcome for the crisis period only (covering the years from 2008 to 2012, Appendix Table 12.1.2) indicates that the coefficients for all three major drivers, that is, European periphery shocks and global risk aversion, as well as the slope of the US yield curve, are higher than for the period including precrisis years (Appendix Table 12.1.1). A 1 percent shock to euro periphery spreads translates into a 0.5- percentage-point increase in the risk premium of the 35 sample countries if the shock originates in the GIPs and a 1-percentage-point increase in spreads if it originates in Italy and Spain. Hence, it seems that the size of the peripheral European country determines the size of spillovers, as expected. Moreover, global r isk-aversion shocks also translate almost one-to-one into spreads.

DCC GARCH Approach

The panel regression approach provided the average spillover effect on countries’ sovereign spreads. Later in this chapter, the previous work is complemented by estimating country-specific daily comovements, in order to differentiate more between countries, and to come up with the range of the potential spillover impact observed over time. A multivariate GARCH framework is used for the estimation, which allows for hetero-skedasticity of the data and a time-varying correlation in the conditional variance. Specifically, the dynamic conditional correlation (DCC) specification by Engle 2002 is adopted, which provides a generalization of the constant conditional correlation model by Bollerslev 1990.13 The DCC GARCH models are estimated in first differences to account for the non-stationarity of the variables in the crisis period.

These econometric techniques allow us to analyze the daily comovement of the GIIPS spreads and the sovereign bond spreads of the sample of AMs and EMs. The GIIPS spreads are included in the model as a conditioning variable, as is the VIX. The methodology is therefore closely aligned to the one of the panel regression and further explained in Appendix 12.2.

The sample period chosen was daily data from 2007 to the end August 2012, with a view to cover the full crisis period. As before, for the European AMs the risk premium of 10-year instruments was measured as the difference between the average GIIPS spread as well as those of the domestic treasuries to German bunds. For the non-European countries, the spread to the 10-year US Treasury bonds is calculated and for EM countries the J.P. Morgan Emerging Markets Bond Index global spread and the HSBC Asian US dollar spread for Asian countries are used.

As expected, our findings suggest that the spread between GIIPS to German bunds exhibits a higher degree of comove-ment with the risk premia for European countries than non-European countries (Figures 12.212.5). In particular, implied DCC GARCH correlations with the GIIPS spread were as high as 0.7–0.8 for Austria, Belgium, France, and the Netherlands during episodes of systemic stress (Figure 12.2, panels 1 and 2). In contrast, the GIIPS comovement with the United Kingdom spread to German bunds is relatively low and oscillates between 0 and 0.2, while the model-implied correlation with the Swiss spreads reaches a maximum of 0.4 (Figure 12.2, panel 3). The results also show that the spreads of the Scandinavian countries, namely Denmark, Norway, Sweden, and Finland (though with higher average levels),14 on average exhibit a lower comovement with the GIIPS spread than do their continental European peers (Figure 12.2, panel 4). The outcome also suggests a constant level of stress, with some easing toward the end of the observation period, a finding which also applies to the non-European sovereigns.

Estimated GARCH Correlations GIIPS with European Countries

Sources: Bloomberg; and authors’ calculations.

Note: GARCH = generalized autoregressive conditional heteroskedasticity; GIIPS = Greece, Ireland, Italy, Portugal, and Spain.

Estimated GARCH Correlations GIIPS with Non-European Countries

Sources: Bloomberg; and authors’ calculations.

Note: GARCH = generalized autoregressive conditional heteroskedasticity; GIIPS = Greece, Ireland, Italy, Portugal, and Spain; Hong Kong SAR = Hong Kong Special Administrative Region.

Estimated GARCH Correlations GIIPS with EM Countries and Korea

Sources: Bloomberg; and authors’ calculations.

Note: EM = emerging market; GARCH = generalized autoregressive conditional heteroskedasticity; GIIPS = Greece, Ireland, Italy, Portugal, and Spain.

Estimated GARCH Correlations GIIPS with Germany and the United States

Sources: Bloomberg; and authors’ calculations.

Note: Unlike the other GARCH models, the average GIIPS interest rates are taken and not the GIIPS spread to German bunds. GARCH = generalized autoregressive conditional heteroskedasticity; GIIPS = Greece, Ireland, Italy, Portugal, and Spain.

Comovements of the GIIPS spread with Australian and Canadian spreads (relative to US Treasury bonds) are rather low, with implied correlations up to 0.2 (Figure 12.3, panel 1). Looking at the Asian countries Hong Kong, Japan, and Singapore shows a somewhat higher correlation with the GIIPS spread of up to 0.3 and with one jump to 0.4 (Figure 12.3, panel 2). In terms of EM countries, results suggest that China’s comovement with the GIIPS spread is rather subdued compared to the other EM countries: Brazil, Mexico, Russia, and Turkey (Figure 12.4). Out of this EM sample, Turkey has the highest implied correlation with the GIIPS during episodes of system stress at up to 0.6.

Since the onset of sovereign debt crisis by 2009, the average GIIPS interest rates exhibit a negative correlation with both the German bund and US Treasury bond interest rates (Figure 12.5). Since 2009, the implied correlation has turned negative for both countries, with lows at -0.4 (United States) and -0.6 (Germany), indicating a sudden fight to safety, in line with other recent studies (IMF 2011, for example).

3. Liquidity and Solvency Stress Testing

The area of stress testing has seen a number of advances during recent years. This study uses a recently developed IMF liquidity stress testing framework to run integrated solvency and liquidity stress tests. The liquidity stress testing frame-

work presented in Schmieder and others 2012 was developed in the context of recent FSAPs15 and IMF technical assistance, extending the seminal work of Čihák 2007 and drawing upon work at the Austrian National Bank.16 An overview of recent academic and policy research on integrating liquidity and solvency stress testing is given in Box 12.1.

Integrating Liquidity, Solvency Risks, and Bank Reactions in Stress Tests

Banks have numerous and overlapping ways to react to credit and funding shocks. High-quality capital and profits are usually the first line of defense, and retained earnings can help buffer banks’ capital levels. In terms of liquidity, banks have an inherent counterbalancing capacity to generate liquid assets by using high-quality eligible securities as collateral to generate market funding, or, if interbank markets freeze entirely, central bank funding. As seen post-Lehman, fire sales of securities can also be an option to generate liquidity, but at a considerable cost in an environment of sharply declining asset prices. Deleveraging, especially targeted at assets with higher risk weights, is also a way to raise capital adequacy ratios by reducing risk-weighted assets. In practice, banks have been using a combination of these, as well as other hybrid measures, ranging from debt-to-equity conversions, to issuance of convertible bonds, to optimizing risk-weighted assets, to react to shocks.

Incorporating banks’ reactions to shocks is a critical component for the design of informative stress tests, especially over longer time horizons. This, however, requires modeling solvency and liquidity shocks in a coherent manner because first, when banks react to financial stress, the source of the shock (solvency or liquidity) is not always clear; and second, the measures banks take in reaction to these shocks have both capital and liquidity aspects that are not easy to disentangle.

Recently, a number of analytical approaches have attempted to integrate solvency and liquidity more systematically.

  • Empirical work includes Van den End 20081 at the Dutch Central Bank and Wong and Hui 2009 from the Hong Kong Monetary Authority,2 for example. Barnhill and Schumacher (2011) developed a more general empirical model, incorporating the previous two approaches that attempt to be more comprehensive in terms of the source of the solvency shocks and compute the longer term impact of funding shocks.
  • Schmieder and others (2012) provide an Excel-based framework that allows running liquidity tests informed by banks’ solvency conditions, and to simulate the increase in funding costs resulting from a change in solvency.
  • An integrated approach to model funding liquidity risks and solvency risk is the Risk Assessment Model for Systemic Institutions developed by the Bank of England (Aikman and others 2009). The framework simulates banks’ liquidity positions conditional on their capitalization under stress and other relevant dimensions such as a decrease in confidence among market participants under stress. A recent attempt by the Austrian National Bank to come up with an integrated framework and to overcome operational challenges identified with previous work on integrated models, the Applied Risk, Network and Impact Assessment Engine, should also be mentioned (OeNB 2013).

For an overview of liquidity stress tests, including the link to solvency, see also BCBS 2013. IMF 2013 examines the European Banking Authority stress tests.

Source: Oura and Schumacher 2012.1Van den End (2008) developed a stress testing model that tries to endogenize market and funding liquidity risk by including feedback effects that capture both behavioral and reputational effects. A number of central banks and bank supervisors have been successfully using the Monte Carlo framework of Van den End 2008.2 The authors sought to explicitly capture the link between default risk and deposit outflows. Their framework allows simulating the impact of mark-to-market losses on banks’ solvency position leading to deposit outflows; asset fire sales by banks are evaporating and contingent liquidity risk sharply increases.

In this study, the focus is on scenario design, namely building integrated scenarios for solvency and liquidity risks that take into account spillover effects and feedback loops.17 The central question becomes how the findings established in Section 2 can be used to inform bank-level stress tests.

Nevertheless, while this chapter attempts to condense a wealth of information and assumptions to establish integrated scenarios this should not, in any sense, give a false sense of precision. Instead, it is recommended that a whole range of scenarios be run, which can build upon the ones established in the study, with varying degrees of severity. Reverse stress tests can be also included.18 This is an important way forward to obtain a better understanding of key solvency and liquidity risks faced by banks, and to gain a more comprehensive view on their respective risk tolerances.

Liquidity Stress Testing Approach

An implied cash-flow approach is applied to simulate the impact of a bank-run-type stress scenario (Appendix 12.3). The banks’ liabilities are broken down into demand and term deposits, short-term wholesale funding (including bank and secured funding), derivative funding as well as long-term funding such as senior debt or subordinated debt. On the asset side, a range of potentially liquid asset positions is included, such as cash, government, trading, and investment (both available-for-sale and held-to-maturity) securities, loans and advances to banks, reverse repos, and cash collateral. Given European periphery banks’ increasing collateral use of pools of loans (such as covered bonds) for liquidity, a crude definition of banks’ loan level as a portion of their total assets is also included.

Solvency Stress Testing Approach

Rules of thumb for solvency stress testing are used, as proposed by Hardy and Schmieder 2013, and thereby a simplified solvency test.19 Credit losses, banks’ preimpairment income, and the trajectories of RWAs for a two-year horizon were simulated based on the GDP trajectories, with and without spillover effects. The capital shortfall was measured against a Tier 1 capital ratio (Tier 1 capital/RWAs) of 6 percent, below which a bank is considered undercapitalized.20

4. Integration of the Financial Spillover Analysis with the Stress Testing Approach

This integrated approach to simulate stress at the bank level is illustratively shown in Figure 12.6:

  • 1. Scenario design: GDP trajectories of a specific macroeconomic scenario were used, the World Economic Outlook (WEO) baseline scenario for 2013–14 as of April 2012, and the spillover stress component was added.
  • 2. Spillover analysis: The outcome of the spillover analysis (see previous sections of this chapter), measured through a widening of sovereign spreads, worsens the macroeconomic scenario, and is used as a sensitivity analysis. The translation of the spillover effects into the revised macroeconomic trajectories is based on recent IMF work.
  • 3. Soundness of banks: The scenario is translated into bank-level stress parameters to simulate both the banks’ solvency and liquidity positions, drawing on work by Hardy and Schmieder 2013 and Schmieder and others 2012, respectively.

Overview of the Concept to Simulate Stress at the Bank Level

Source: Authors.

Note: GARCH = generalized autoregressive conditional heteroskedasticity; P/L = profit/loss; WEO = World Economic Outlook.

Bank-level data from BankFocus (from the end of June 2012) was used for large systematically important banks. In total, the sample includes 154 large banks from the following 26 countries: Austria, Australia, Belgium, Brazil, Canada, Switzerland, China, Germany, Denmark, Finland, France, United Kingdom, Hong Kong Special Administrative Region, India, Japan, Korea, Luxembourg, Mexico, Netherlands, Norway, Poland, Russia, Sweden, Singapore, Turkey, and the United States.

The sample comprises almost the full European Banking Authority sample for the European banks (except for the banks in the GIIPS countries) and includes the largest banks in the non-European countries. In total, it captures $84 trillion of bank assets (that is, about 50 percent of the assets held by banks worldwide), $39 trillion nonbank deposits, and about $7 trillion of government securities held by banks.

Scenarios

Four different scenarios are referred to: The April 2012 WEO baseline scenario for 2013–14 (Scenario 1); and three spillover scenarios (referred to as Scenarios 2.X) conditional on Scenario 1—scenarios that banks could potentially face in case increasing degrees of spillovers affect the general growth trend.

Specifically, Scenario 1 is adjusted for an increase of GIIPS spreads by 100 (Scenario 2a), 200 (Scenario 2b), and 300 (Scenario 2c) basis points, respectively. The study further distinguishes between the spillover impact observed during periods of substantial financial stress (using the panel regression for 2008–12 and the GARCH model for 2010–12) and during periods of less significant stress (using the panel regression for 2006–12 and the GARCH model for 2008–12), that is, refer to a total of six spillover scenarios (2a/1, 2a/2, 2b/1, 2b/2, 2c/1, 2c/2).

For the banks’ solvency, their Tier 1 capital ratios are stimulated by end-2014, based on the evolution of the main solvency dimensions (banks’ income and losses). For liquidity, the impact is determined of a worst-case idiosyncratic shock to the bank’s liquidity profile on top of the impact on liquidity resulting from the macroeconomic/spillover scenarios. Illustrative examples are provided in Appendix 12.4 (solvency) and Appendix 12.5 (liquidity).

Impact on Bank Solvency

As outlined earlier in this chapter, the study uses the outcome of the IMF’s 2012 Spillover Report (IMF 2012b), which simulates the impact of a 300-basis-point increase in peripheral countries’ spreads (including a lower yield increase for core countries) on European countries’ GDP paths based on the IMF G35 model (drawing upon Vitek and Bayoumi 2011).

Appendix 12.4 provides an illustrative example for a stylized Austrian bank. In the first step, the increase of Austrian sovereign debt spreads is simulated, using the evidence established in Section 2. A 100-basis-point shock of GIIPS spreads (Scenario 2a) would thereby result in an increase of Austrian spreads by 24 basis points for less significant spillover stress (Scenario 2a/1) and 50 basis points (2a/2) for more substantial spillover stress. Measured relative to the April 2012 WEO baseline scenario for Austria, suggesting real GDP growth rates of 1.8 percent (2013) and 2.2 percent (2014), spillover analysis carried out at the IMF (2012b) would predict a drop of real GDP growth by about 0.45 percentage point for Scenario 2a/1 (less significant spillover stress), whereby the GDP trajectory becomes 1.4 percent (2013) and 1.8 percent (2014). For a period with more significant spillover (Scenario 2a/2), the impact is about twice (0.9 percentage point), whereby the GDP trajectory is 0.9 percent (2013) and 1.3 percent (2014). For a 200-basis-point shock (Scenario 2b), growth drops by 1.7 percentage point and for 300 basis points (Scenario 2c) by 2.6 percentage points (per year) under substantial spillover conditions (Appendix 12.4).

The satellite models by Hardy and Schmieder (2013) are then used to determine banks’ loan impairment levels and preimpairment income for 2013 and 2014.21 For a stylized bank with loss impairment rates of 0.5 percent and a preimpairment return on capital of 10 percent in 2012,22 loan impairment rates are simulated to decrease slightly under the baseline scenario and mild spillover conditions, while they would increase (nonlinearly) under increasing levels of spillover stress. The same pattern holds for preimpairment income. This input is used to simulate the bank’s capital, risk-weighted, and capital ratio. Again, the same pattern holds, with a decrease of the stylized banks’ assets,23 capital ratio to 7.5 percent under the most severe scenario, which is above the hurdle rate in terms of Tier 1 capital to pass the stress test (6 percent).

The outcome of this solvency stress test applied to the 154 banks presented in Figure 12.7 shows that the large international banks would be in a position to digest the baseline scenario plus some level of spillover stress, while additional stress in the euro area periphery results would have a highly nonlinear impact on potential capital needs. The nonlinearity results from two factors: (1) the nonlinearity in the satellite models for loan impairment rates and preimpairment income, and (2) the effect of the kick-in of capital needs for banks that fall below the hurdle rate.

Outcome of Solvency Stress Tests

Source: Authors.

Note: bps = basis points; Sc = Scenario.

Impact on Bank Liquidity

For the liquidity stress test, the study simulates the impact of stress on both banks’ market liquidity (that is, their ability to fire sale assets) and funding liquidity (that is, the potential outflow of funding).24 Again, it was assumed that the bank is affected by the shock in its home country.25

The link between the level of stress and bank liquidity is established based on empirical work of Schmieder and others 2012. The study links the GDP trajectories implied by the changes of sovereign spreads to funding shocks experienced by the most affected banks during the Lehman crisis. In other words, the study simulates highly adverse idiosyncratic liquidity shocks conditional upon macroeconomic conditions.

In line with (very limited) empirical evidence, it was expected that the relationship between the shock and the potential adverse impact on the bank level would be highly nonlinear (as implied by the scenarios in Appendix 12.3, and in addition to the nonlinearity for the banks hitting the hurdle rate, as for capital). Under a worst case scenario, banks would experience a shock equal to a “Lehman Brothers– type” scenario, the “severe stress scenario” in Appendix 12.3. This shock level represents how the stress at the time of the Lehman Brothers event affected the banks that were most severely hit, that is, overlays a market shock with an idiosyncratic liquidity shock. The stress level relative to the one experienced by banks at the time of the Lehman Brothers crisis is established via the cumulative GDP trajectory under stress compared to the long-term average. For the stylized example presented in Appendix 12.5, the stress level is at 0.65, that is, the benchmark funding stress parameters (for the “severe stress scenario”) in Appendix 12.3 have to be multiplied by 0.65. The funding available for the specific banks under the European Central Bank’s Long Term Refinancing Operations is inferred from country-level data and used as a cushion for the relevant European banks.

Figure 12.8 shows the outcome of this liquidity stress test. Under the baseline scenario all banks have sufficient liquidity, as expected. Adding spillover stress triggers a nonlinear increase of liquidity needs (which occur in case the liquidity needs exceed the available liquidity generated via fire sales), and more substantial spillover stress makes the stress highly nonlinear. Measured against Tier 1 capital rather than total assets, the substantial spillover stress leads to a maximum liquidity shortfall of 20 percent for the entire bank sample for Scenario 2c/2 (300-basis-point spread shock, significant spillover stress) and close to 6 percent for Scenario 2b/2 (200-basis-point spread shock), compared to 0.3 percent and 1 percent if measured against total assets.26

Outcome of Liquidity Tests in Terms of Assets

Source: Authors.

Note: bps = basis points; Sc = Scenario.

5. Conclusion

This study attempts to address the challenge faced by current financial stability analysis, namely to capture spillover effects and other types of contagion that ultimately determine macro-financial stress at the bank level.

By integrating recent IMF work on financial spillover analysis and stress testing, the study uses a novel framework that allows shedding some light on the potential impact of spillover effects on bank-level solvency and liquidity. Nevertheless, it is recognized that significant additional effort and evidence are needed to make the modeling of dynamic macro-financial linkages more robust, not least due the many potential channels of spillover and contagion, the fact that the use of crude data available for stress tests is subject to uncertainty, and other factors that contribute to uncertainty (such as mixed evidence for the use of market data).

The outcome of the stress tests suggests that spillover effects observed for the sovereign debt markets in recent years have a highly nonlinear impact on bank soundness, both in terms of liquidity and solvency. This implies (once more) that the design of stress scenarios is a crucial element of stress testing, and is very sensitive with respect to the outcome of stress tests. The approach used in this chapter is meant to be a menu for future analyses of the impact of potential spillovers. Sensitivity analysis and reverse stress tests appear to be an important complement in this context.

Appendix 12.1. Outcome of Panel Regressions Assessing Spillover Risks
Appendix Table 12.1.1Panel Regressions, 2006:Q1–2012:Q2(Dependent variable: sovereign spreads of 35 sample countries)(Quarterly data)
Explanatory Variables1(1)(2)(3)(4)(5)(6)
GIIPS spread0.237***

(0.045)
0.244***

(0.047)
GIP spread0.288***

(0.046)
0.289***

(0.047)
Italy/Spain spread0.611***

(0.09)
0.653***

(0.094)
High-yield spread0.666***

(0.242)
0.621***

(0.229)
0.357

(0.30)
VIX0.348

(0.238)
0.342

(0.229)
-0.070

(0.291)
Openness0.015

(0.017)
0.015

(0.017)
0.031*

(0.016)
0.030*

(0.016)
0.025

(0.020)
0.025

(0.021)
M2/GDP0.080***

(0.017)
0.078***

(0.017)
0.061***

(0.016)
0.060***

(0.016)
0.053***

(0.020)
0.051**

(0.020)
Constant0.297**

(0.131)
-0.632

(0.744)
0.256*

(0.136)
-0.660

(0.718)
0.700***

(0.166)
0.997

(0.912)
R2 (within)0.770.700.790.730.790.78
Observations415415435435454454
T252523232626
Source: Authors.Note: Standard errors in parentheses. GIIPS = Greece, Ireland, Italy, Portugal, and Spain; GIP = Greece, Ireland, and Portugal; M2 = M2 money supply;T = Number of quarters covered by the regressions; VIX = CBOE Volatility Index.*p < 0.1; **p < 0.05; ***p < 0.01.

Right-hand-side variables are in logs.

Source: Authors.Note: Standard errors in parentheses. GIIPS = Greece, Ireland, Italy, Portugal, and Spain; GIP = Greece, Ireland, and Portugal; M2 = M2 money supply;T = Number of quarters covered by the regressions; VIX = CBOE Volatility Index.*p < 0.1; **p < 0.05; ***p < 0.01.

Right-hand-side variables are in logs.

Appendix Table 12.1.2Panel Regressions, 2008:Q1–2012:Q2(Dependent variable: sovereign spreads of 35 sample countries)(Quarterly data)
Explanatory Variables1(1)(2)(3)(4)(5)(6)
GIIPS spread0.492***

(0.105)
0.463***

(0.106)
GIP spread0.511***

(0.090)
0.479***

(0.090)
Italy/Spain spread1.002***

(0.173)
0.998***

(0.175)
High-yield spread1.042***

(0.299)
1.033***

(0.279)
0.735**

(0.366)
VIX0.823**

(0.322)
0.813***

(0.301)
0.517

(0.397)
Openness0.018

(0.021)
0.017

(0.021)
0.034*

(0.019)
0.033*

(0.019)
0.033

(0.027)
0.032

(0.027)
M2/GDP0.078***

(0.020)
0.075***

(0.020)
0.057***

(0.018)
0.056***

(0.018)
0.045*

(0.025)
0.043*

(0.025)
Constant-0.133

(0.222)
-2.418**

(1.084)
-0.216

(0.222)
-2.459**

(1.022)
0.308

(0.246)
-1.117

(1.307)
R2 (within)0.930.780.910.780.910.85
Observations321321357357341341
T181818181818
Source: Authors.Note: Standard errors in parentheses. GIIPS = Greece, Ireland, Italy, Portugal, and Spain; GIP = Greece, Ireland, and Portugal; M2 = M2 money supply; T = Number of quarters covered by the regressions; VIX = CBOE Volatility Index.*p < 0.1; **p < 0.05; ***p < 0.01.

Right-hand-side variables are in logs.

Source: Authors.Note: Standard errors in parentheses. GIIPS = Greece, Ireland, Italy, Portugal, and Spain; GIP = Greece, Ireland, and Portugal; M2 = M2 money supply; T = Number of quarters covered by the regressions; VIX = CBOE Volatility Index.*p < 0.1; **p < 0.05; ***p < 0.01.

Right-hand-side variables are in logs.

Appendix Table 12.1.3Main Explanatory Variables
FactorVariableDescription
Sovereign riskGIIPS spreadAverage of euro periphery sovereign spreads to German bunds
GIP spreadAverage of Greece, Ireland, and Portugal sovereign spreads to German bunds
Italy/Spain spread (IS spread)Average of Italy and Spain sovereign spreads to German bunds
Risk aversionHigh-yield spreadDifference between yields to maturity of AAA-rated and BAA-rated corporate US bonds
Macroeconomic environmentVIX

Openness

M2/GDP
Implied volatility of S&P 500 index options Sum of imports and exports to GDP ratio Broad money to GDP ratio
Source: Authors.Note: GIIPS = Greece, Ireland, Italy, Portugal, and Spain; GIP = Greece, Ireland, and Portugal; Italy-Spain; M2 = M2 money supply; S&P = Standard & Poor’s; VIX = CBOE Volatility Index.
Source: Authors.Note: GIIPS = Greece, Ireland, Italy, Portugal, and Spain; GIP = Greece, Ireland, and Portugal; Italy-Spain; M2 = M2 money supply; S&P = Standard & Poor’s; VIX = CBOE Volatility Index.
Appendix 12.2. Outline of the DCC GARCH Method

The dynamic conditional correlation (DCC) model is estimated in a three-stage procedure. Let rt denote an n × 1 vector of asset returns, exhibiting a mean of zero and the following time-varying covariance:

Here, Rt is made up from the time-dependent correlations, and Dt is defined as a diagonal matrix comprised of the standard deviations implied by the estimation of univariate generalized autoregressive conditional heteroskedasticity (GARCH) models, which are computed separately, whereby the ith element is denoted as hit. In other words, in this first stage of the DCC estimation, univariate GARCH models are ft for each of the five variables in the specification. In the second stage, the intercept parameters are obtained from the transformed asset returns, and in the third stage, the coefficients governing the dynamics of the conditional correlations are estimated. Overall, the DCC model is characterized by the following set of equations (see En-gle 2002 for details):

Here, S is defined as the unconditional correlation matrix of the residuals εt of the asset returns rt. As defined above, Rt is the time-varying correlation matrix and is a function of Qt, which is the covariance matrix. In the matrix Qt,l is a vector of ones, A and B are square and symmetrical, and o is the Hadamard product. Finally, Qt is a weight parameter with the contributions of Dt12 declining over time, while Ki is the parameter associated with the squared lagged asset returns. The estimation framework is the same as in Frank, Gonzalez-Hermosillo, and Hesse 2008 and Frank and Hesse 2009.

Appendix 12.3. Benchmark Liquidity Stress Scenarios
Appendix Table 12.3.1Benchmark Liquidity Stress Scenarios
ScenarioModerate Stress ScenarioMedium Stress ScenarioSevere Stress ScenarioVery Severe Stress Scenario
Severity (x times Lehman1)0.250.512
Liquidity Outflows
Customer Deposits
Customer deposits (term)2.5 percent5 percent10 percent20 percent
Customer deposits (demand)5 percent10 percent20 percent40 percent
Wholesale Funding
Short term (secured)5 percent10 percent20 percent40 percent
Short term (unsecured)25 percent50 percent100 percent100 percent
Contingent liabilities0 percent need funding5 percent need funding10 percent need funding20 percent need funding
Liquidity Inflows
Haircut for cash0 percent0 percent0 percent0 percent
Haircut for government securities21 percent2 percent5 percent10 percent
Haircut for trading assets33 percent6 percent30 percent100 percent
Proxies, specific assetsEquities: 3; Bonds: 3Equities: 4–6; Bonds: 3–8Equity: 10–15; Bonds (only LCR eligible ones): 5–10Not liquid
Haircut for other securities10 percent30 percent75 percent100 percent
Proxies, specific assetsEquities: 10; Bonds: 10Equities: 25;
Bonds: 20 (some not liquid)Equity: 30; Bonds (only LCR eligible ones): 20–30Not liquid
Percent of liquid assets encumbered410 percent (or actual figure)20 percent (or actual figure plus 10 ppt)30 percent (or actual figures plus 20 ppt)40 percent (or actual figures plus 30 ppt)
Source: Schmieder and others 2012.Note: LCR = liquidity coverage ratio; ppt = percentage points.

The Lehman-type scenario would correspond to a scenario encountered by banks that were hit severely during the 30-day period after the Lehman collapse, that is, a stress situation within a stress period rather than an average. The scenario has been put together based on expert judgment, using evidence as available.

The haircut highly depends on the specific features of the government debt held (rating, maturity, market depth) and can be higher or lower. The figures displayed herein are meant for high-quality investment-grade bonds, taking into account recent market conditions. The same applies for the remainder of the liquid assets. For the securities in the trading book, it is assumed that they are liquidated earlier, resulting in lower haircuts.

A haircut of 100 percent means that the asset is illiquid, that is, the market has closed.

The figures account for a downgrade of the bank, which triggers margin calls, and higher collateral requirements more generally. Please note that the unencumbered portion applies to a gradually narrower definition of liquid assets.

Source: Schmieder and others 2012.Note: LCR = liquidity coverage ratio; ppt = percentage points.

The Lehman-type scenario would correspond to a scenario encountered by banks that were hit severely during the 30-day period after the Lehman collapse, that is, a stress situation within a stress period rather than an average. The scenario has been put together based on expert judgment, using evidence as available.

The haircut highly depends on the specific features of the government debt held (rating, maturity, market depth) and can be higher or lower. The figures displayed herein are meant for high-quality investment-grade bonds, taking into account recent market conditions. The same applies for the remainder of the liquid assets. For the securities in the trading book, it is assumed that they are liquidated earlier, resulting in lower haircuts.

A haircut of 100 percent means that the asset is illiquid, that is, the market has closed.

The figures account for a downgrade of the bank, which triggers margin calls, and higher collateral requirements more generally. Please note that the unencumbered portion applies to a gradually narrower definition of liquid assets.

Appendix 12.4. Illustrative Example for the Solvency Test

Table 12.4.1 provides an illustrative example for a hypothetical bank in Austria.

Appendix Table 12.4.1Solvency: Illustrative Example for a Hypothetical Austrian Bank
Step 1.1. Spillover Impact in Sovereign Debt Markets Observed for Austria
Scenario (increase of GIIPS sovereign debt spreads by . . .)Impact on Austria (Average for 2006–12)Impact on Austria during Peak Spillover Stress (2008–12)
Increase of spreads (bps)SourceIncrease of spreads (bps)Source
100 bps (2a)24.41 (=24*1.017)Appendix Table 12.1.1, spec (22), Figure 12.2, panel 249.83 (=49*1.017)Appendix Table 12.1.2, spec (1), Figure 12.2, panel 2
200 bps (2b)48.8Same, linear increase assumed99.6Same, linear increase assumed
300 bps (2c)73.2Same, linear increase assumed149.4Same, linear increase assumed
Step 1.2. GDP Trajectory for Austria, Adjusted for the Impact of Spillovers
(GDP elasticity of widening of spreads for Austria estimated for two-year period from 2013–14: 3.5 [based on IMF 2012b])4
Trajectory based on evidence for 2006–12 (less significant spillovers)
Scenario201220132014Cumulative deviation of output from 2012 real GDP growth level (2013–14), ppts
Baseline (1)0.91.82.22.2
(2 a /1)0.91.4 (= 1.8–0.5*3.5*0.244)1.8 (=2.2–0.5*3.5*0.244)1.3
(2b/1)0.90.9 (= 1.8–0.5*3.5*0.488)1.3 (=2.2–0.5*3.5*0.488)0.4
(2c/1)0.90.5 (= 1.8–0.5*3.5*0.732)0.9 (=2.2–0.5*3.5*0.732)-0.4
Trajectory based on evidence for 2008–12 (more significant spillovers)
Scenario201220132014Cumulative deviation of output from 2012 real GDP growth level (2013–14), ppts
Baseline (1)0.91.82.22.2
(2a/2)0.90.9 (=1.8–0.5*3.5*0.498)1.3 (=2.2–0.5*3.5*0.498)0.4
(2b/2)0.90.1 (= 1.8–0.5*3.5*0.996)0.5 (=2.2–0.5*3.5*0.996)-1.2
(2c/2)0.9-0.8 (=1.8–0.5*3.5*1.494)-0.4 (=2.2–0.5*3.5*1.494)-3
Source: Authors.Note: bps = basis points; GIIPS = Greece, Ireland, Italy, Portugal, and Spain; RWAs = risk-weighted assets; spec = specification.

The average impact of stress (in terms of GIIPS spreads) on euro area countries is 24 basis points (based on the panel analysis; see Appendix Table 12.1.1) and for Austria the relative severity of this impact approximately matches the impact observed for the EU, that is, it is 1.0 times of this level (GARCH analysis, average impact from 2008–12 based on Figure 12.2, panel 2, relative to average of the average impact for other EU countries).

The study used the higher impact on the GIIPS spreads from Appendix Table 12.1.1 and Appendix Table 12.1.2—that is, specification 2 (Appendix Table 12.1.1) and 1 (Appendix Table 12.1.2), respectively, that is, 24 bps and 49 bps, respectively.

The average impact of stress on euro area countries is 49 basis points (based on the panel analysis, Appendix Table 12.1.2), and for Austria the impact is again estimated to be at a similar level (GARCH analysis, average impact from 2008–12, Figure 12.2, panel 2, relative to average of the average impact for other EU countries).

The GDP elasticities of sovereign debt spreads vary between 0.5 (for example, Brazil) and 3.5.

See Hardy and Schmieder 2013 for further information.

Credit growth is assumed to be constant for simplification.

For simplification, RWA elasticity to credit losses is assumed to be 0.5—that is, for a 1-percentage-point change of credit loss rates RWAs will change by 0.5 percentage points.

Step 2: Simulation of the impact at the bank level (example for stylized bank)5
Change of key solvency parameters6
ScenarioLoan impairment rates (Percent of credit exposure)Preimpairment income (Percent of total capital)
201220132014201220132014
Baseline0.50.40.41010.310.5
2a/10.50.450.41010.1510.3
2b/10.50.50.45101010.1
2c/10.50.550.5109.810
2a/20.50.50.45101010.1
2b/20.50.70.6109.79.8
2c/20.50.90.8109.29.5
Evolution of risk-weighted assets and capital7
ScenarioRWAs (Indexed)Capital
201220132014201220132014
Baseline10090901010.5811.21
2a/110095901010.5711.18
2b/1100100951010.5611.15
2c/11001051001010.5411.12
2a/2100100951010.5611.15
2b/21001201111010.5211.08
2c/21001401321010.4710.99
Source: Authors.Note: bps = basis points; GIIPS = Greece, Ireland, Italy, Portugal, and Spain; RWAs = risk-weighted assets; spec = specification.

The average impact of stress (in terms of GIIPS spreads) on euro area countries is 24 basis points (based on the panel analysis; see Appendix Table 12.1.1) and for Austria the relative severity of this impact approximately matches the impact observed for the EU, that is, it is 1.0 times of this level (GARCH analysis, average impact from 2008–12 based on Figure 12.2, panel 2, relative to average of the average impact for other EU countries).

The study used the higher impact on the GIIPS spreads from Appendix Table 12.1.1 and Appendix Table 12.1.2—that is, specification 2 (Appendix Table 12.1.1) and 1 (Appendix Table 12.1.2), respectively, that is, 24 bps and 49 bps, respectively.

The average impact of stress on euro area countries is 49 basis points (based on the panel analysis, Appendix Table 12.1.2), and for Austria the impact is again estimated to be at a similar level (GARCH analysis, average impact from 2008–12, Figure 12.2, panel 2, relative to average of the average impact for other EU countries).

The GDP elasticities of sovereign debt spreads vary between 0.5 (for example, Brazil) and 3.5.

See Hardy and Schmieder 2013 for further information.

Credit growth is assumed to be constant for simplification.

For simplification, RWA elasticity to credit losses is assumed to be 0.5—that is, for a 1-percentage-point change of credit loss rates RWAs will change by 0.5 percentage points.

Evolution of the bank’s capital ratio
ScenarioCapital Ratio (= capital/RWA, percent)
201220132014
Baseline10.011.812.5
2a/110.011.112.5
2b/110.010.611.7
2c/110.010.011.1
2a/210.010.611.7
2b/210.08.89.9
2c/210.07.58.3
Source: Authors.Note: bps = basis points; GIIPS = Greece, Ireland, Italy, Portugal, and Spain; RWAs = risk-weighted assets; spec = specification.

The average impact of stress (in terms of GIIPS spreads) on euro area countries is 24 basis points (based on the panel analysis; see Appendix Table 12.1.1) and for Austria the relative severity of this impact approximately matches the impact observed for the EU, that is, it is 1.0 times of this level (GARCH analysis, average impact from 2008–12 based on Figure 12.2, panel 2, relative to average of the average impact for other EU countries).

The study used the higher impact on the GIIPS spreads from Appendix Table 12.1.1 and Appendix Table 12.1.2—that is, specification 2 (Appendix Table 12.1.1) and 1 (Appendix Table 12.1.2), respectively, that is, 24 bps and 49 bps, respectively.

The average impact of stress on euro area countries is 49 basis points (based on the panel analysis, Appendix Table 12.1.2), and for Austria the impact is again estimated to be at a similar level (GARCH analysis, average impact from 2008–12, Figure 12.2, panel 2, relative to average of the average impact for other EU countries).

The GDP elasticities of sovereign debt spreads vary between 0.5 (for example, Brazil) and 3.5.

See Hardy and Schmieder 2013 for further information.

Credit growth is assumed to be constant for simplification.

For simplification, RWA elasticity to credit losses is assumed to be 0.5—that is, for a 1-percentage-point change of credit loss rates RWAs will change by 0.5 percentage points.

Source: Authors.Note: bps = basis points; GIIPS = Greece, Ireland, Italy, Portugal, and Spain; RWAs = risk-weighted assets; spec = specification.

The average impact of stress (in terms of GIIPS spreads) on euro area countries is 24 basis points (based on the panel analysis; see Appendix Table 12.1.1) and for Austria the relative severity of this impact approximately matches the impact observed for the EU, that is, it is 1.0 times of this level (GARCH analysis, average impact from 2008–12 based on Figure 12.2, panel 2, relative to average of the average impact for other EU countries).

The study used the higher impact on the GIIPS spreads from Appendix Table 12.1.1 and Appendix Table 12.1.2—that is, specification 2 (Appendix Table 12.1.1) and 1 (Appendix Table 12.1.2), respectively, that is, 24 bps and 49 bps, respectively.

The average impact of stress on euro area countries is 49 basis points (based on the panel analysis, Appendix Table 12.1.2), and for Austria the impact is again estimated to be at a similar level (GARCH analysis, average impact from 2008–12, Figure 12.2, panel 2, relative to average of the average impact for other EU countries).

The GDP elasticities of sovereign debt spreads vary between 0.5 (for example, Brazil) and 3.5.

See Hardy and Schmieder 2013 for further information.

Credit growth is assumed to be constant for simplification.

For simplification, RWA elasticity to credit losses is assumed to be 0.5—that is, for a 1-percentage-point change of credit loss rates RWAs will change by 0.5 percentage points.

Appendix 12.5. Illustrative Example for Liquidity

This appendix provides an illustrative example for a hypothetical bank in Austria.

Step 1. GDP trajectory for Austria, adjusted for the impact of spillovers.

The first steps uses the same GDP trajectories as for solvency (see Appendix 12.4). Accordingly, the severity of the liquidity shock is simulated relative to the Lehman Brothers benchmark scenario in Appendix 12.4. Specifically, based on the observation that the cumulative US real GDP growth deviated by about 8 percentage points from the long-term average, the corresponding figures are computed for each of the scenarios. For Austria (and for the other European countries), the baseline growth rates for 2013–14 (that is, 2 percent) are (for simplicity) used as a proxy for the long-term trend. For Scenario 2c/2, the cumulative deviation from the baseline is 5.2 percentage points. For the severity of the liquidity test, the study therefore used the stress parameters for the severe scenario in Appendix 12.3 multiplied by a factor of 0.65 (= 5.2/8).

Step 2. Simulation of the impact at the bank level (example for stylized bank).27

Relevant asset and liability balance sheet items are shocked based on the severity of each scenario, that is, the stress factor (such as 0.65) multiplied by the respective stress parameters. The balance sheet items are taken from BankFocus. For the long-term refinancing operation, the available total funding was assigned to the single banks based on their size, using the available evidence for the total at the country level.

In the table below, Scenario 2c/2 is simulated for a stylized bank based on Austria. The composition of the banks’ asset and liabilities resemble those of an average Organisation for Economic Co-operation and Development (OECD) bank.28 The stress factor reduces the haircuts and outflows of the benchmark scenario. In the example, the bank is able to generate an inflow of 21.5 units of assets, compared to a required level of 13.7 units, whereby the bank remains liquid.

Appendix Table 12.5.1Liquidity: Illustrative Example for a Hypothetical Austrian Bank
Assets (of stylized bank)
Portion of Total1Haircut, Percent (Appendix 12.3)Haircut Scenario 2c/2Available Assets (Fire sales)
Cash and cash-like4004.0
Government securities6535.8
Trading securities530204.0
Other securities1575497.7
Loans60NANA
Other10NANA
Liabilities (of stylized bank)
Portion of Total2Outflow, Percent (Appendix 12.3)Outflow Scenario 2c/2Required Funding
Customer term deposits30106.52
Customer demand deposits2020132.6
Secured short-term wholesale funding1020131.3
Unsecured short-term wholesale funding10100656.5
Long-term funding20000
Equity-based funding10000
Contingent liabilities20106.51.3
Source: Authors.

Aligned to the average composition of Organisation for Economic Co-operation and Development banks’ balance sheets. See Schmieder and others 2012, p. 38.

Aligned to the average composition of Organisation for Economic Co-operation and Development banks’ balance sheets. See Schmieder and others 2012, p. 38.

Source: Authors.

Aligned to the average composition of Organisation for Economic Co-operation and Development banks’ balance sheets. See Schmieder and others 2012, p. 38.

Aligned to the average composition of Organisation for Economic Co-operation and Development banks’ balance sheets. See Schmieder and others 2012, p. 38.

References

    AikmanDavidPiergiorgioAlessandriBrunoEklundPrasannaGaiSujitKapadiaElizabethMartinNadaMoraGabrielSterne andMatthewWillison. 2009. “Funding Liquidity Risk in a Quantitative Model of Systemic Stability.” Bank of England Working Paper 372Bank of EnglandLondon, United Kingdom. https://www.bankofengland.co.uk/working-paper/2009/funding-liquidity-risk-in-a-quantitative-model-of-systemic-stability.

    • Search Google Scholar
    • Export Citation

    Austrian National Bank (OeNB). 2013. “ARNIE in Action: The 2013 FSAP Stress Tests for the Austrian Banking System.” Financial Stability Report 26Vienna, Austria.

    • Search Google Scholar
    • Export Citation

    BarnhillTeodoreJr. andLilianaSchumacher. 2011. “Modeling Correlated Systemic Liquidity and Solvency Risks in a Financial Environment with Incomplete Information.” IMF Working Paper 11/263International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Modeling-Correlated-Systemic-Liquidity-and-Solvency-Risks-in-a-Financial-Environment-with-25356.

    • Search Google Scholar
    • Export Citation

    Basel Committee on Banking Supervision (BCBS). 2008. Principles for Sound Liquidity Risk Management and Supervision. Basel: Bank for International Settlements. https://www.bis.org/publ/bcbs138.htm.

    • Search Google Scholar
    • Export Citation

    Basel Committee on Banking Supervision (BCBS). 2013. “Liquidity Stress Testing: A Survey of Theory, Empirics, and Current Industry and Supervisory Practices.” BCBS Working Papers 24OctoberBank for International SettlementsBasel, Switzerland. https://www.bis.org/publ/bcbs_wp24.htm.

    • Search Google Scholar
    • Export Citation

    BollershevTim. 1990. “Modelling the Coherence in Short-Run Nominal Exchange Rates: A Multivariate Generalized ARCH Approach.” Review of Economics and Statistics72 (3): 498505.

    • Search Google Scholar
    • Export Citation

    BorioClaudioMatthiasDrehmann andKostasTsatsaronis. 2012. “ Stress-Testing Macro Stress Testing: Does It Live up to Expectations?BIS Working Paper 369Bank for International SettlementsBasel, Switzerland. https://www.bis.org/publ/work369.htm.

    • Search Google Scholar
    • Export Citation

    CaceresCarlos andD.Filiz Unsal. 2011. “Sovereign Spreads and Contagion Risks in Asia.” IMF Working Paper 11/134International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Sovereign-Spreads-and-Contagion-Risks-in-Asia-24910.

    • Search Google Scholar
    • Export Citation

    ČihákMartin. 2007. “Introduction to Applied Stress Testing.” IMF Working Paper 07/59International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Introduction-to-Applied-Stress-Testing-20222.

    • Search Google Scholar
    • Export Citation

    EngleRobert. 2002. “Dynamic Conditional Correlation: A Simple Class of Multivariate Generalized Autoregressive Conditional Heteroskedasticity Models.” Journal of Business & Economic Statistics20 (3): 33950.

    • Search Google Scholar
    • Export Citation

    FogliaAntonella. 2009. “Stress Testing Credit Risk: A Survey of Authorities’ Approaches.” International Journal of Central Banking (September). http://www.ijcb.org/journal/ijcb09q3a1.htm.

    • Search Google Scholar
    • Export Citation

    FrankNathanielBrendaGonzález-Hermosillo andHeikoHesse. 2008. “Transmission of Liquidity Shocks: Evidence from the 2007 Subprime Crisis.” IMF Working Paper 08/200International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Transmission-of-Liquidity-Shocks-Evidence-from-the-2007-Subprime-Crisis-22238.

    • Search Google Scholar
    • Export Citation

    FrankNathaniel andHeikoHesse. 2009. “Financial Spillovers to Emerging Markets during the Global Financial Crisis.” IMF Working Paper 09/104International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Financial-Spillovers-to-Emerging-Markets-During-the-Global-Financial-Crisis-22936.

    • Search Google Scholar
    • Export Citation

    HardyDanielC. andChristianSchmieder. 2013. “Rules of Tumb for Bank Solvency Stress Testing.” IMF Working Paper 13/232International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Rules-of-Thumb-for-Bank-Solvency-Stress-Testing-41047.

    • Search Google Scholar
    • Export Citation

    HesseHeikoFerhanSalman andChristianSchmieder. 2014. “How to Capture Macro-Financial Spillover Effects in Stress Tests?IMF Working Paper 14/103International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/How-to-Capture-Macro-Financial-Spillover-Effects-in-Stress-Tests-41644.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund (IMF). 2011. “Euro Area Policies: Spillover Report for the 2011 Article IV Consultation and Selected Issues.” IMF Country Report 11/185Washington, DC. http://www.imf.org/external/pubs/cat/longres.aspx?sk=25056.0.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund (IMF). 2012a. Global Financial Stability Report— The Quest for Lasting Stability Chapter 3. Washington, DCApril. https://www.imf.org/external/pubs/ft/gfsr/2012/01/.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund (IMF). 2012b. “2012 Spillover Report.” IMF Policy PaperWashington, DC. https://www.imf.org/en/Publications/Policy-Papers/Issues/2016/12/31/2012-Spillover-Report-PP4678.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund (IMF). 2013. “European Union: Stress Testing of Banks— Technical Note.” IMF Country Report 13/68Washington, DC. https://www.imf.org/en/Publications/CR/Issues/2016/12/31/European-Union-Publication-of-Financial-Sector-Assessment-Program-Documentation-Technical-40396.

    • Search Google Scholar
    • Export Citation

    JobstAndreasA. andDale F.Gray. 2013. “Systemic Contingent Claims Analysis—Estimating Market-Implied Systemic Solvency Risk.” IMF Working Paper 13/54International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Systemic-Contingent-Claims-Analysis-Estimating-Market-Implied-Systemic-Risk-40356.

    • Search Google Scholar
    • Export Citation

    JobstAndreasLiLian Ong andChristianSchmieder. 2013. “A Framework for Macroprudential Bank Solvency Stress Testing: Application to S-25 and Other G20 Country FSAPs.” IMF Working Paper 13/68International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/A-Framework-for-Macroprudential-Bank-Solvency-Stress-Testing-Application-to-S-25-and-Other-G-40390.

    • Search Google Scholar
    • Export Citation

    OuraHiroko andLilianaSchumacher. 2012. “ Macro-Financial Stress Testing— Principles and Practices.” IMF Policy PaperWashington, DCAugust. https://www.imf.org/en/Publications/Policy-Papers/Issues/2016/12/31/Macrofinancial-Stress-Testing-Principles-and-Practices-PP4702.

    • Search Google Scholar
    • Export Citation

    SchmiederChristianHeikoHesseBenjaminNeudorferClausPuhr andStefan W.Schmitz. 2012. “Next Generation System-Wide Liquidity Stress Testing.” IMF Working Paper 03/12International Monetary FundWashington, DC. https://www.imf.org/external/pubs/cat/longres.aspx?sk=25509.0.

    • Search Google Scholar
    • Export Citation

    SchmiederChristianClausPuhr andMaherHasan2011Next Generation Balance Sheet Stress Testing.” IMF Working Paper 11/83International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Next-Generation-Balance-Sheet-Stress-Testing-24798.

    • Search Google Scholar
    • Export Citation

    TalebNassimN.ElieCanettiTidianeKindaElenaLou-koianova andChristianSchmieder. 2012. “A New Heuristic Measure of Fragility and Tail Risks: Application to Stress Testing.” IMF Working Paper 216/2012International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/A-New-Heuristic-Measure-of-Fragility-and-Tail-Risks-Application-to-Stress-Testing-26222.

    • Search Google Scholar
    • Export Citation

    Van den EndJanWillem. 2008. “Liquidity Stress Tester: A Macro Model for Stress-Testing Banks’ Liquidity Risk.” DNB Working Paper 175/2008Dutch National BankAmsterdam, Netherlands. https://www.dnb.nl/en/news/dnb-publications/dnb-working-papers-series/dnb-working-papers/auto175528.jsp.

    • Search Google Scholar
    • Export Citation

    VitekFrancis andTamimBayoumi. 2011. “Spillovers from the Euro Area Sovereign Debt Crisis: A Macroeconometric Model Based Analysis.” CEPR Discussion Paper 8497Center for Economic and Policy ResearchWashington, DC. https://cepr.org/active/publications/discussion_papers/dp.php?dpno=8497.

    • Search Google Scholar
    • Export Citation

    WongEric andCho-HoiHui. 2009. “A Liquidity Risk Stress Testing Framework with Interaction between Market and Credit Risks.” HKMA Working Paper 06/2009Hong Kong Monetary AuthorityHong Kong SAR. https://www.hkma.gov.hk/eng/publications-and-research/research/working-papers/2009/.

    • Search Google Scholar
    • Export Citation
1

For work on stress testing at the IMF, for example, see Jobst, Ong, and Schmieder 2013.

2

For liquidity stress tests, most tests have typically relied on the “direct” approach.

3

Further information on macroeconomic scenarios used for financial sector assessment programs (FSAPs) can be found in Jobst, Ong, and Schmieder 2013.

4

At the IMF, such analyses were carried out by combining the work of Schmieder, Puhr, and Hasan (2011) and Vitek and Bayoumi (2011) as part of early-warning analysis and vulnerability exercises. It should be noted that running such an approach requires close cooperation between staff running macroeconomic forecasts and staff simulating the impact of stress at the bank level (typically done by financial stability departments).

5

See also IMF 2011 and 2012a for further information on related work.

6

GIIPS refers to Greece, Ireland, Italy, Portugal, and Spain.

7

The frameworks were developed in the context of recent FSAPs and IMF technical assistance, extending the seminal work of Čihák 2007, and drawing upon work at the Austrian National Bank (OeNB 2013).

8

See also Taleb and others 2012 on how to test the sensitivity (that is, nonlinearity) of the outcome of stress tests.

9

The sample of countries includes Australia, Austria, Belgium, Brazil, Canada, China, Cyprus, Denmark, Finland, France, Germany, Greece, Hong Kong SAR, Hungary, India, Ireland, Italy, Japan, Korea, Luxembourg, Malta, Mexico, Netherlands, Norway, Poland, Portugal, Russia, Singapore, Slovenia, Spain, Sweden, Switzerland, Turkey, the United Kingdom, and the United States.

10

The panel regressions adjust for exchange rate changes.

11

Measured in terms of the R- squared and the actual coefficients.

12

For a robustness check, a separate set of regressions were run to estimate the impact of expectations of higher interest rates, represented by the slope of the US Treasury yield curve on the global risk premium. Results indicate that a steepening of the curve implies higher costs of borrowing for the periphery countries.

13

Given the high volatility movements during the global financial crisis, the assumption of constant conditional correlation among the variables in the constant conditional correlation model is not very realistic, especially in times of stress where correlations can rapidly change. Therefore, the DCC model is a better choice, since correlations are time-varying.

14

Finland is the only euro-area country within the sample, which seems to explain the higher level of correlations.

15

Examples include Chile, Germany, India, Spain, Turkey, and the United Kingdom.

16

It is complemented by a previously developed solvency stress testing tool by Schmieder, Puhr, and Hasan 2011. While developing the solvency and liquidity stress testing frameworks, four key facts were accounted for, which constitute key challenges of contemporaneous financial stability analysis: (1) the availability of data varies widely, and lack of data is common; (2) both solvency and liquidity risk have various dimensions, which requires multidimensional analysis, thereby integrating risks; (3) designing and calibrating scenarios is challenging, even more so for liquidity risk than for solvency risk (mainly as liquidity crises are relatively rare and originate from different sources); and (4) communication of stress test results is a key integral part of the exercise. The answer to these multiple dimensions is Excel-based balance-sheet-type frameworks.

17

The exercise thereby reflects key principles for liquidity stress testing put forward by the Basel Committee in the aftermath of the first wave of shocks following the default of Lehman Brothers (BCBS 2008).

18

The work by Taleb and others 2012 and Hardy and Schmieder 2013, for example, could be useful to consider in this context.

19

However, it should be noted that the evidence is based on a comprehensive set of data from 16,000 banks during the last 15 years (as available).

20

Please note that this specific choice is meant for illustration only— through a similar level as used for the European stress tests conducted in 2010 and 2011, for example.

21

For simplification, it was assumed that banks are affected according to their domestic scenarios, that is, that their businesses are predominantly based in their home countries.

22

In a few cases, the latest available figures were from 2011.

23

The risk-weighted assets are simulated based on work by Schmieder, Puhr, and Hasan (2011), assuming point-in-time credit risk parameters.

24

Unlike for the solvency scenario, the study does not simulate stress for a specific point in time; rather, the simulated stress conditions reflect a worst-case situation resulting from the general macroeconomic conditions as well as an idiosyncratic shock to the bank conditional.

25

In other words, it is assumed that all of its assets are based in the home country, which is a crude simplification.

26

The study did not explicitly model a central bank response as the Lender of Last Resort to mitigate the estimated liquidity shortfall. In reality and as seen during the crisis period, central banks would provide large liquidity support to solvent banks, subject to an appropriate haircut.

27

See Schmieder and others 2012 for further information.

28

See Schmieder and others 2012, p. 38, for more information.

    Other Resources Citing This Publication