Euro Area Policies: Financial Sector Assessment Program Technical Note—Systemic Risk Analysis
Author:
International Monetary Fund. Monetary and Capital Markets Department
Search for other papers by International Monetary Fund. Monetary and Capital Markets Department in
Current site
Google ScholarClose
Publication Date:
19 Jul 2018
eISBN:
9781484369586
Language:
English
Keywords:
ISCR; CR; bank; OFI residual; other financial institution; pds of bank; Pd estimate; MCS-GVAR model; CCA-MCS-GVAR sample; forecasts of bank; Debt default; Solvency stress testing; Global
Download PDF (1.2 MB)

Financial Sector Assessment Program-Technical Note-Systemic Risk Analysis

Abstract

Financial Sector Assessment Program-Technical Note-Systemic Risk Analysis

Contingent Claims Analysis42

91. This chapter presents an overview of the contingent claims analysis (CCA). An integrated contingent claims, mixed cross-section global vector autoregressive (CCA-MCS-GVAR) model is developed which combines a large scale empirical (MCS-GVAR) framework with CCA indicators and satellite modules.43 The model was used to develop forecasts of bank and insurance companies’ probability of defaults (PDs), conditional on the FSAP macrofinancial scenario assumptions.44 In addition, these conditional forecasts are complemented with historical decompositions. A concise summary of the MCS-GVAR model structure is presented in Appendix I.

92. The framework encompasses banks, insurers, sovereigns, and the nonfinancial corporate sector. In particular, the CCA-MCS-GVAR combines PD estimates for financial institutions (30 in the model), insurers (11) and non-financial corporate (NFC) sectors (19), as well as sovereign credit spreads (for 19 sovereigns) in multi-country model, which includes supervisory ECB data on exposures of all banks and insurers relative to each other and to sovereigns and the nonfinancial sectors across countries. The need for market price data implies that the sample of banks is not the sample as in the case of the solvency and liquidity stress tests (see Appendix II for the list of banks, insurers, and sovereigns). The PD estimates are derived using CCA. The model is estimated based on data spanning the 1999Q1–2017Q4 period and used to produce scenario conditional forecasts for all model variables based on the FSAP macro-financial scenarios. The model contains the variables summarized in Table 18.

Table 18.

Euro Area: Sectors and Model Variables

article image
Source: Historical PDs are derived from Moody’s CreditEdge database.

93. The scenario conditional forecasts for banks and insurers are shown in Figure 21. The groups A and B for insurers correspond to a group of “global” and “domestic” insurers, respectively (see Appendix II Table 2 for the list and group assignment). Figure 21 also shows the corresponding ratios of the PDs under the adverse and the baseline scenario.

Figure 21.
Figure 21.

Euro Area: Scenario Conditional PD Forecasts

Citation: IMF Staff Country Reports 2018, 231; 10.5089/9781484369586.002.A004

Source: IMF staff estimates.

94. Bank PDs are higher under the stress scenario relative to those of insurers. Under the adverse scenario, the G-SIB in the sample experience nearly a 5-fold increase of their baseline PDs compared to an approximate 3-fold increase of PDs for all other banks—which broadly corroborates the findings of the balance sheet-based solvency stress tests. One consistent result is that a minority of banks is much more vulnerable to an adverse shock than others. The increase in insurer PDs is from a low base

95. The adverse scenario would not only increase the average fair value CDS (FVCDS), but also lead to much greater dispersion (Figure 22). The box plots for the starting point (2017Q4) and the baseline (horizon average) and adverse scenario (horizon maximum) reflect the distribution of the underlying 30 banks’ and 11 insurers’ estimated level FVCDS. The mean increases by about 100 basis points for the banks, and for a substantial tail of banks the increase is much greater. A similar pattern is apparent among insurers. The FVCDS for some banks reach levels of about 350 basis points, that is, remain below 400 basis points which is deemed to be a critical value at which strong nonlinear effects in relation to the firms’ wholesale funding costs have historically been observed to materialize (lenders to banks may be reluctant to roll over their debt, hence possibly implying liquidity shortages for banks). For insurers, the FVCDS stay a more comfortable margin below the 400-basis points threshold than do those of the banks.

Figure 22.
Figure 22.

Euro Area: Fair Value CDS Estimates for Banks and Insurers

Citation: IMF Staff Country Reports 2018, 231; 10.5089/9781484369586.002.A004

Source: IMF staff estimates.Note: The box plots depict the distribution of fair value CDS (FVCDS) level spreads in the cross-sections of banks and insurers, respectively. The red lines indicate the median; the upper and lower edges of the boxes mark the 25th and 75th percentiles; the whiskers extend to the data points farthest out of the distributions that are not considered technical outliers yet. The red crosses mark “outliers” in a statistical sense as being farther away from the median.

96. The historical contributions of the model variables toward the PDs of banks and insurers are presented in Figure 23. Historical decompositions depict how the underlying contributions from other sectors explain PD dynamics banks and insurers. Some of the model variables’ contributions are combined to not overload the charts with too much information: the sum of long-term rates and CDS contributions is referred to as the contribution of the “sovereign”; the sum of the contributions from nonfinancial corporate PDs, GDP growth and stock prices is referred to as the contribution of ‘macro’ in the charts. The figure illustrates how the sharp (bank and insurer) PD spikes are attributable to different variables over time.

Figure 23.
Figure 23.

Euro Area: Historical Contributions to the Dynamics of PDs of Banks and Insurers

Citation: IMF Staff Country Reports 2018, 231; 10.5089/9781484369586.002.A004

Source: IMF staff estimates.

97. Table 19 summarizes the historical contribution estimates of different groups of model variables toward bank PDs, insurers’ PDs and sovereign CDS. Four time windows are considered over which the average contributions were computed (consistent with the contributions visualized in Figure 23 for banks and insurers): a pre-crisis period from 1999Q1–2008Q3, an initial financial crisis period covering 2008Q4–2009Q4, the sovereign debt crisis from 2010Q1–2012Q4, and the remainder of the sample from 2013Q1–2017Q4.

Table 19.

Euro Area: Historical Contributions to Changes in Variation of Bank, Insurer, and Sovereign PDs

(Percent)

article image
Source: Authors’ calculations.

Tabulates on the CDS component of “sovereign” in basis points.

98. The following observations can be made based on the analysis:

  • Regarding the average PDs, the G-SIBs’ PDs peaked during the sovereign debt crisis period and have then decline relatively steeply. Non-G-SIB banks experienced their peak in default risk in the early phase of the crisis, and their PDs have declined less in absolute and relative terms.

  • The predominant role in determining bank PDs is played by the macroeconomic (which is an aggregation of three variables) and the sovereign variables. In contrast, banks’ contribution to banks’ PD dynamics amongst each other are comparably small across groupings as well as over time compared to other factors. This may reflect banks’ interconnectedness with other sectors: banks are susceptible to factors outside the banking system, mirroring in turn their role in creating credit for the real sector, investing in sovereign bonds, as well as providing funds to insurers.

  • At the onset of the initial phase of the crisis in 2008Q4, non-G-SIBs’ rising default risk stemmed from a larger contribution from insurers as their PD rose to 22 percent during the financial crisis. The appreciable contribution of the residual hints at the role of (non-euro area) global factors (not captured by the modeling framework).

  • Concerning the contributing factors to changes in insurers’ default risk, the two groups (global Group A vs. domestic oriented Group B) exhibit a notable difference with a view to the insurers’ own contribution and the residual category: the group of insurers exerts a close to 0 percent contribution to the sub-group of global insurers, and a sizable contribution to domestic insurers (10-17 percent). The residual contribution on the other hand is more sizable for global insurers. The latter finding can be interpreted as meaning that global factors contribute to changes in risk of global insurers.

  • Regarding the contributions to sovereign risk (here with respect to CDS only; see last column in Table 19), the bank to sovereign contribution increased markedly in the financial crisis and remained relatively high. The cross-sovereign risk contribution increased markedly with the move into the sovereign debt crisis period (up to 24 percent). The contribution of macroeconomic factors has steadily abated over time.

99. A number of caveats are in order. The CCA- MCS-GVAR model does not fully capture structural changes in risk exposures which have changed for some banks after the crisis. At the same time, because of data limitations, proxies were used when calibrating some sectoral interlinkages (see Appendix I Table 1). The basis for the contribution analysis was a generalized impulse response function concept, that is, the estimates were not based on a structural shock identification method. In comparing the results from the CCA-MCS-GVAR analysis with the findings from the solvency stress tests, it should be noted that market data is not available for some banks, and therefore the solvency stress testing and CCA-MCS-GVAR samples differ. In addition, the former summarizes results using weighted averages, while the CCA- MCS-GVAR uses simple averages.

Appendix I. An Overview of the Structure of the CCA-MCS-GVAR Model

The CCA-MCS-GVAR model system comprises four cross-sections: N banks, M insurers, L sovereigns and a nonfinancial private (“macro”) sector covering F countries. The variables corresponding to these four cross-sections are denoted xi, yj, zl, and wf in the following. All equations contain weighted variable vectors (denoted by an asterisk in the following equations) of the respective own and other cross-sections on the right hand-side of the equations, which are constructed using time-varying weights of different kinds.

x i t = a i + ∑ p 1 = 1 P 1 Φ i p 1 x x i , t − p 1 + ∑ p 2 = 0 P 2 Λ i , 0 , p 2 x i , t − p 2 * , B − B + ∑ p 3 = 0 P 3 Λ i , 1 , p 3 y i , t − p 3 * , B − I + ∑ p 4 = 0 P 4 Λ i , 2 , p 4 z i , t − p 4 * , B − S + ∑ p 5 = 0 P 5 Λ i , 3 , p 5 w i , t − p 5 * , B − M + ε i t x
y i t = b j + ∑ q 1 = 1 Q 1 Φ j q 1 y y j , t − q 1 + ∑ q 2 = 0 Q 2 Ξ j , 0 , q 2 x j , t − q 2 * , I − B + ∑ q 3 = 0 Q 3 Ξ j , 1 , q 3 y j , t − q 3 * , I − I + ∑ q 4 = 0 Q 4 Ξ j , 2 , q 4 z j , t − q 4 * , I − S + ∑ q 5 = 0 Q 5 Ξ j , 3 , q 5 w j , t − q 5 * , I − M + ε j t y
z l t = c l + ∑ r 1 = 0 R 1 Φ l r 1 z z l , t − r 1 + ∑ r 2 = 0 R 2 Ψ l , 0 , r 2 x l , t − r 2 * , S − B + ∑ r 3 = 0 R 3 Ψ l , 1 , r 3 y l , t − r 3 * , S − I + ∑ r 4 = 0 R 4 Ψ l , 2 , r 4 z l , t − r 4 * , S − S + ∑ r 5 = 0 R 5 Ξ l , 3 , r 5 w l , t − r 5 * , S − M + ε l t z
w f t = d f + ∑ u 1 = 0 U 1 Φ f u 1 w z f , t − u 1 + ∑ u 2 = 0 U 2 Ψ f , 0 , u 2 x f , t − u 2 * , M − B + ∑ u 3 = 0 U 3 Ψ f , 1 , u 3 y f , t − u 3 * , M − I + ∑ u 4 = 0 U 4 Ψ f , 2 , u 4 z f , t − u 4 * , M − S + ∑ u 5 = 0 U 5 Ξ f , 3 , u 5 w f , t − u 5 * , M − M + ε f t z

The model in this form has time-contemporaneous relationships, which means that it has to be “solved”, for the solved form to not contain such contemporaneous dependence anymore, for the model in turn to be usable for forecasting and impulse response simulations, forecast error variance decompositions, etc.45

Chart A summarizes the data that was employed to capture the exposure profiles among banks, insurers, sovereigns and the non-financial private sector, based on which weights are derived to inform the structure of the MCS-GVAR model. All weights are time-varying at a quarterly frequency over the 1999Q1–2017Q4 sample period, except for the Stress Test 2016 (ST2016) database which contains cross-country loan exposure profiles as of end-2015.

Appendix I Table 1.

Euro Area: Sources of Exposure Data for Banks, Insurers, Sovereigns, and the Private Sector

article image
Note: SHSG abbreviates Securities Holdings Statistics (Group level, available for individual banks). SHSS abbreviates the Securities Holdings Statistics (Sector level). Macro (M): this label denotes the nonfinancial private sector, i.e. nonfinancial corporations and households.

Appendix II. List of Banks, Insurers, and Countries

Appendix II Table 1.

Banks

article image
Appendix II Table 2.

Insurers

article image
Appendix II Table 3.

Countries

article image

42

This chapter was prepared by Dale Gray (Monetary and Capital Markets, IMF) and Marco Gross (ECB).

43

The model structure employed for the analysis presented here combines some elements of different GVAR variants developed in the past. The inclusion of CCA-based indicators in a GVAR setting, as one element, has been pursued in Gray, Gross, Paredes, Sydow (2013), “Modeling banking, sovereign, and macro risk in a CCA Global VAR”, IMF WP/13/218. The Mixed-Cross-Section feature of the GVAR, as a second core feature, has been developed in Gross and Kok (2013), “Measuring contagion potential among sovereigns and banks using a Mixed-Cross-Section GVAR”, ECB WP No. 1570. It has been further extended to a semi-structural model set up and including more cross section types in Gross, Kok, Zochowski (2016), “The impact of bank capital on economic activity – Evidence from a Mixed-Cross-Section GVAR model”, ECB WP No. 1888.

44

Please see the EA FSAP Technical Note on Stress Testing the Banking Sector for details on the scenarios.

45

The details concerning the solution method can be found in Gross, M., Kok, C. and D. Zochowski, 2016, “The impact of bank capital on economic activity – Evidence from a Mixed-Cross-Section GVAR model”. ECB Working Paper No. 1888.

Same Series

Other IMF Content

Other Publishers

  • Collapse
  • Expand
Cover IMF Staff Country Reports
Euro Area Policies: Financial Sector Assessment Program-Technical Note-Systemic Risk Analysis
Author:
International Monetary Fund. Monetary and Capital Markets Department
Volume/Issue:
Volume 2018: Issue 231
Publisher:
International Monetary Fund
ISBN:
9781484369586
ISSN:
1934-7685
Pages:
94
DOI:
https://doi.org/10.5089/9781484369586.002
  • Figure 21.
    View in gallery
    Figure 21.

    Euro Area: Scenario Conditional PD Forecasts

  • Figure 22.
    View in gallery
    Figure 22.

    Euro Area: Fair Value CDS Estimates for Banks and Insurers

  • Figure 23.
    View in gallery
    Figure 23.

    Euro Area: Historical Contributions to the Dynamics of PDs of Banks and Insurers

Figure 21.

Euro Area: Scenario Conditional PD Forecasts
Figure 22.

Euro Area: Fair Value CDS Estimates for Banks and Insurers
Figure 23.

Euro Area: Historical Contributions to the Dynamics of PDs of Banks and Insurers