The bank-sovereign nexus across borders H.-Johannes Breckenfelder, Bernd Schwaab ∗ European Central Bank, Financial Research
April 24, 2015 Preliminary; comments welcome.
Abstract We quantify the risk contagion from the banking to the sovereign sector within and across borders in the euro area. Our empirical findings are based on difference and difference-in-differences specifications around the European Central Bank’s (ECB) release of the outcome of its Comprehensive Assessment (CA) of the 130 most significant banks in the euro area, on 26 October 2014. The associated information shock led to a reassessment of bank risk in some countries, and, as a consequence, of sovereign risk. We find that risk spillovers from domestic banks to the respective sovereign are weaker or insignificant within countries that are more likely to face sovereign distress. Instead, the additional bank risk is borne in part by other euro area sovereigns. The well-known bank-sovereign nexus has an important cross-border component.
JEL classification: C68, G15, F34. Keywords: ECB, Comprehensive Assessment, Contagion, Stress test, Asset Quality Review.
∗
Corresponding author: Johannes Breckenfelder, European Central Bank, Sonnemannstraße 22, 60314 Frankfurt, Germany, email:
[email protected]. Bernd Schwaab, European Central Bank, Sonnemannstraße 22, 60314 Frankfurt, Germany, email:
[email protected]. We thank Geert Bekaert, Philipp Hartmann, and Simone Manganelli for valuable comments. Francesca Barbiero, Gabriele Beneduci, and Tim van Ark provided excellent research assistance. The views expressed in this paper are those of the authors’ and do not necessarily reflect those of the European Central Bank or the Eurosystem.
1
Introduction
The sovereign debt crisis, which erupted in 2010 in the euro area, put a strain on the global financial system and highlighted that credit risk is transmitted through several important links within an economy. The foremost apparent one is the close connection between sovereigns and banks; see, for example, Cooper and Nikolov (2014), Acharya et al. (2014), Farhi and Tirole (2014), and Acharya and Steffen (2015). This strong relation may cause, through multiple feedback loops, a “deadly embrace”, or “doom loop”, through which both banks and their sovereign can end up in crisis simultaneously. A second important link exists “horizontally” between financial firms across countries, for example due to counterparty credit risk and information contagion; see Lang and Stulz (1992), Jorion and Zhang (2007, 2009), and Helwege and Zhang (2012).1 Finally, contagion may be observed between sovereigns as well, and was arguably present during the most severe phases of the euro area sovereign debt crisis from 2011-2012; see, for instance, Kallestrup et al. (2013), Lucas et al. (2014), and Benzoni et al. (2015). In this paper, we show that the risk spillover from banks to their own sovereign, widely referred to as the bank-sovereign nexus, is weaker or insignificant within a group of euro area countries that is more likely to face sovereign distress. In such countries, surprise increases in banking sector risk are borne at least in part by other euro area sovereigns. As a result, the bank-sovereign nexus in the euro area is not necessarily – and not even primarily – a within-country phenomenon. We demonstrate that there exists an important cross-sector cross-border dimension through which bank risk surprises in one country affect the credit risk and refinancing cost of other sovereigns. We quantify these spillovers and discuss the relevant transmission channels. Our study exploits exogenous variation from an information shock associated with the 1
When companies file for bankruptcy, other firms in the same industry often suffer as a result. Lang and Stulz (1992) conclude that rivals’ stocks drop in response to the news because investors learn about future industry cash flows from the filing. Consistent with this result, Jorion and Zhang (2007) report that credit default swap (CDS) premiums typically rise for firms in the same industry after a default. Theocharides (2007), Hertzel and Officer (2012), and Boissay and Gropp (2012) present evidence of similar patterns for corporate bonds, bank loans, and trade credit, respectively. In addition, Helwege and Zhang (2012) find that both counterparty contagion and information contagion have significant effects on other financial firms’ stock prices.
1
publication of the ECB’s Comprehensive Assessment (CA) results on 26 October 2014. The change in the assessment of banks’ health in response to the information shock allows us to estimate the causal effect from bank credit risk to sovereign risk both within and across countries in the euro area. The CA was a one year-long rigorous examination of the resilience of the 130 largest banks in the euro area, and consisted of a backward-looking Asset Quality Review (AQR) and a forward-looking supervisory stress test (ST) of the examined banks. The CA covered bank assets of e22 trillion and represented more than 80% of total banking assets in the euro area. It was carried out by the ECB together with the national supervisors from November 2013 to October 2014. 26 national supervisors were involved, with a total of approximately 6000 people active during its course. We show that the release of the CA results led to a substantial reassessment of banking sector risks within a subset of euro area countries. For example, comparing bank equity prices two weeks post-CA to an earlier two week period pre-CA, average bank equity market values dropped by approximately 12 percent in Italy, and also by more than 10 percent on average across banks in Spain, Ireland, Greece, and Portugal. At the same time, bank equity values remained approximately unchanged in other euro area countries that had suffered less during the peak of the euro area sovereign debt crisis from 2010–2012.2 Strikingly, the CDS premia of banks that exhibit the biggest negative equity surprises increased only marginally, or not at all, indicating that bank debt appears to be insured to at least some extent. Instead, the covariation between the equity returns of negatively affected banks and the CDS of other sovereigns which did not face an adverse information shock on their banks, becomes positive and significant. In addition to serving as a source of exogenous variation, the completion of the CA was a major milestone in the ECB’s preparation for the Single Supervisory Mechanism (SSM), the euro area’s newly created cross-border banking supervisor. The SSM became operational just two weeks after the announcement of the CA results, on 4 November 2014. The SSM, in turn, is a key pillar of the European ‘Banking Union’, a set of legislation that was ratified by the European Council and the European Parliament in successive steps between 2012 and 2
Throughout the paper, we refer to Greece, Ireland, Italy, Portugal, and Spain as stressed countries, and to other euro area countries such as Austria, Belgium, Finland, France, Germany, and the Netherlands, as non-stressed countries. Cyprus and Slovenia are stressed countries, but we do not have banks that are referenced by a liquid CDS contract. For a similar grouping of countries see, for example, Acharya and Steffen (2015) and Eser and Schwaab (2015).
2
2014, with the main objective to break, or at least minimize, the ‘deadly embrace’ between the credit risk of national banking sectors and that of the respective sovereign; see, for instance EC (2012) and Constˆancio (2014). To which extent the Banking Union legislation was able to achieve this aim is currently an open question. Establishing a causal (contagion) link from banks to sovereigns is particularly challenging in the euro area setting. The main identification problem is the pronounced risk dependence between the two sectors. Banks depend on its own (and other) sovereigns because they hold large amounts of sovereign debt for investment and liquidity insurance purposes. Vice versa, sovereigns depend on their own (and other) banks because they provide a fiscal backstop in times of crisis; in addition, bank lending matters for economic growth and thus for future sovereign tax revenue; see, for instance, Farhi and Tirole (2014), Acharya et al. (2014), and Fratzscher and Rieth (2015). Fortunately, the announcement of the CA results can serve as a quasi-experimental setup that allows us to identify the risk spillovers from banks to sovereigns. Changes in sovereign CDS premia before and after the announcement of the CA results are, plausibly, due to the revised outlook on banks’ health, and not the other way around. Our main empirical approach relies on simple differential effects and difference-in-differences estimates to quantify and test the risk spillovers from banks to sovereigns. We use these panel data regression results to, first, investigate how bank-sovereign risk sensitivities change within countries from before to after the ECB’s announcement of the CA results. In addition, we study the cross-sectional differences in the risk transmission within countries. Finally, we relate sovereign risk to the risk of both domestic and foreign banks, and investigate to what extent risk sensitivities change in light of the ECB’s CA announcement. As a complementary approach, we use a time-varying parameter model in state space form to estimate the time-variation in risk sensitivity parameters directly. This approach allows us to cover a longer period of time, particularly focusing on market developments after the start of the CA in late 2013 to after its conclusion in October 2014. Both regression and state space modeling approaches provide approximately similar estimates of differential effects around the time of the CA.
3
We focus on three main empirical findings. First, in terms of the market impact of the CA announcement, we find that the overall level of bank risk (CDS) moved up on average at the announcement of the CA results, with essentially a one-to-one pass-through from banks to sovereigns in stressed countries. In non-stressed countries, instead, only sovereign CDS spreads increased, while, surprisingly, bank CDS spreads remained approximately flat. Bank equity prices in stressed markets dropped sharply after the CA, by approximately 12 percent, and in line with the outcome of the CA that found that the majority of banks that failed the ECB’s CA was located in stressed countries. Strikingly, the corresponding bank CDS premia moved relatively little. This ‘non-response’ of CDS premia compared to the pronounced drops in bank’s equity valuations and stress test outcomes suggests that bank bonds are insured against the impact of the additional losses to at least some extent. Despite a pronounced drop in Italian bank equity prices after the CA-announcement, Italian banks’ CDS actually declined. Even more starkly, the CDS data suggests that banks were considered less risky, on average, than the Italian sovereign, after the announcement of the CA results (and not before). This violation of the ‘sovereign ceiling’, is a strong tell-tale sign in line with cross border risk sharing. Given the higher CDS, it is highly unlikely that it is the Italian sovereign that serves as the ‘guarantor of last resort’ to its banking system at that time. Second, we study bank-sovereign risk transmission within and across borders. We find that that the risk nexus strengthened in non-stressed countries, but not in stressed ones, around the time of the CA. This is counter-intuitive at first glance, given that the CA revealed risks in stressed countries. In the two weeks before the CA announcement, a plausibly exogenous increase in bank CDS of 1 percent leads to an average increase in its respective sovereign CDS of approximately 0.3–0.5 percent in stressed countries, while there is no evidence for such a bank-sovereign link in non-stressed countries; the respective point estimates are between 0.0–0.2 percent. The ordering reverses after the announcement of the CA results, when the within-country elasticity is higher in non-stressed countries (0.3 percent) than in stressed countries. When seen in conjunction with the development of equity prices, the time differences in risk elasticities, again, strongly point towards significant risk sharing across borders.
4
Third, we quantify and test for cross-border risk transmission directly. Our quantitative analysis suggests that an exogenous decrease in average bank equity prices in stressed countries of 10 percent would lead to an average increase in sovereign CDS of non-stressed countries of up to 7 percent (elasticity parameter) from its current levels, implying an increase in average sovereign CDS in non-stressed countries from approximately 52 basis points (bps) post-CA to approximately 56 bps. While such an increase is not extremely large, it is economically significant given the substantial amounts of sovereign debt that are refinanced each year in the euro area. The amount of public sector risk sharing is therefore both statistically and economically significant. Based on our empirical findings, we conclude that financial markets expect at least a partial transfer of bank risks from stressed countries to non-stressed ones should potentially large bank losses materialize in the future. In terms of transmission channels, conditional insurance is provided by new institutions that were created during the course of the euro area sovereign debt crisis from 2010 to 2012, such as the European Stability Mechanism (ESM), with a total size of more than e700 bn. In addition, a subset of central bank non-standard monetary policy measures include features of partial and conditional insurance, such as the European Central Bank’s (ECB) Outright Monetary Transactions programme (OMT), and to a minor extent also its private sector purchase programme (PSPP), as discussed in the media in 2014 and formally announced in January 2015. Finally, events observed during the euro area sovereign debt crisis may also have contributed to the impression that novel and unprecedented rescue policies can be created over the course of a weekend, if necessary. Our findings lead to a relatively straightforward policy implication. Shortly after the conclusion of the CA, ECB president Draghi (2014) observed that ”... economies [in the euro area] will never be so flexible that adjustment [to economic shocks] happens as quickly as if they had their own exchange rate. ... [W]hen a shock hits ..., we need other ways to help spread those costs. In a monetary union like ours, there is a particular onus on private risk-sharing to play this role. Indeed, the less public sector risk sharing we want, the more private sector risk sharing we need.” Our results suggest that risk sharing arrangements in the euro area are, at least in part, of the “public sector” kind. In late 2014, sovereigns (taxpayers) were still affected from banking sector woes, both domestically and across national 5
borders. The risk transfer is potentially problematic owing to its effects on risk taking incentives and moral hazard. In this regard we conjecture that endowing the euro area banking supervisor with a sizable bank resolution fund, funded by risk-sensitive contributions from banks, would support shifting risk dependence away from the public sector to the private sector (i.e., banks). According to current plans, the bank resolution fund is scheduled to (only, eventually) achieve a total size of approximately e55 bn, eight years after its inception in 2014.3 Instead, the size and leverage capabilities of the bank resolution fund could be made more similar to the ESM, a rescue facility which is backed by European sovereigns. We proceed as follows. Section 2 explains why the setting is a valid quasi-natural experiment to study risk transfers from banks to sovereigns. It also discusses the main aims, outcomes, and communication timeline of the ECB’s CA. Section 4 explains the empirical methodology. We present our data and main empirical findings in Section 5. Section 6 concludes.
2
The ECB’s Comprehensive Assessment
The CA was a financial health check of 130 banks in the euro area, covering bank assets of e22 trillion and representing 82% of total banking assets in the euro area. It was carried out by the ECB together with the national supervisors between November 2013 and October 2014. 26 national supervisors were involved, with a total of approximately 6000 people active during its course. The CA concluded with an aggregate disclosure of the overall outcomes as well as bank-level data, along with recommendations for supervisory measures. All results and data were published on the ECB’s website on 26 October 2014. This section provides a timeline of the ECB’s communication to market participants, before discussing the outcome of the ECB’s CA. The reaction of financial markets is studied in detail in Section 5.1. 3
To put this into perspective, the 2012 losses of a single Spanish bank (Bankia) alone amounted to approximately e19 bn, see WSJ (2013).
6
2.1
The CA as a quasi-natural experiment
The chronology of official communication and news reports leading up to the 26 October 2014 is important for our interpretation of that event as an information shock to the health of the banking sector. It also serves to motivate some of our empirical choices. We argue the CA offers a promising quasi-experimental setup through which to identify the spillover effects from bank risk to the risk of sovereigns. To this purpose we distinguish three periods throughout our analysis: i) a pre-announcement period (Pre-CA, 29 Sep - 10 Oct); ii) a soft information period in which information is gradually becoming available to the news media and financial markets (Soft Info, 13- 24 Oct); and iii) a post-announcement period for which the hard information is available to all market participants (Post-CA, 27 Oct - 04 Nov). Figure 1 plots the chronology of ECB communication up to the announcement of the CA results. The ECB announced on 10 October 2014 that the final CA results will be published in about two weeks time, on 26 October 2014. After that initial announcement, media attention began to focus on the CA. Soon afterwards, the information transmission to the market occurred to a significant extent through various news reports, and in part through official communication by the ECB and national central banks. Much less likely is that CA-related information potentially affected markets through the ongoing ‘supervisory dialogue meetings’ that the ECB held with all participating banks; these had commenced on October 6, 2014 and were scheduled to take place within two weeks time. News coverage of the upcoming CA was fairly intense during the Soft Info period. Indeed, news reports were so frequent, and resulting market movements so volatile, that the ECB offered a press release on 22 October that “any media reports on the outcome of the tests are ... highly speculative”. Arguably, three examples of key news annoucements stand out. First, a Bloomberg News report on Tuesday November 14 appears to have had a significant impact on financial markets. This report offered little more than general guidance, but alluded to some of the ECB’s (strictly confidential) dark room numbers. Second, a Bloomberg News report on Thursday November 23 named a few banks that had failed the stress test, and
7
Figure 1: Timeline of news around the ECB’s Comprehensive Assessment A summary of major news about the CA from 29 September 2014 to 7 November 2014. The period between the two dashed vertical lines denotes the Soft Info period which commences two weeks prior to the announcement of the CA results on 26 October 2014.
OCTOBER 13: Communications October 10: ECB announces CA results to be published on 26 October 2014
between ECB and NCB over CA results get started
OCTOBER 22:
OCTOBER 23:
ECB press release: “Until that time (October 26, Edit) any media reports on the outcome of the tests are by their nature highly speculative.”
Bloomberg News reports that Italian MPS and Carige, jointly with Irish Permanet TBS, had failed the stress test, whereas Deutsche Bank had not
October 26:
time
OCTOBER 13 to 21:
OCTOBER 22:
media spread information on potential outcomes of the CA – mainly reports of investment banks
First leak by Spanish press-agency EFE .
8
OCTOBER 24: Bloomberg News reports that exactly 25 banks had failed the stress test. Other leaks from National press anticipate the reaction of stressed banks to the stress tests
CA results are released to the public by the ECB
also some that had not failed. Finally, a third Bloomberg News report appeared on Friday November 24, based on a leaked document, which stated that exactly 25 banks had failed the stress test (the correct number). To ensure a reliable quantification of the differential effects, we take the Soft Info period to start on Monday 13 October. This is after the ECB’s initial announcement on Friday 10 October that the final CA results would be communicated on 26 October, and before the first Bloomberg News report on Tuesday 14 October.
2.2
Micro-level information shocks to banks’ risk
The CA consisted of a backward looking Asset Quality Review (AQR) and a forward looking supervisory stress test (ST) of the banks. The key objectives of the two elements were to: strengthen banks’ balance sheets by repairing the uncovered problems, enhance transparency by improving the quality of information available on the health of the banks, and to build confidence by assuring that, on completion of the required remedial actions, all banks are soundly capitalized, see ECB (2014). As the main result, a total capital shortfall of e25 billion was identified at 25 participant banks. Twelve of these 25 banks had already covered their capital shortfall by increasing their capital by e15 billion during 2014, leaving 13 banks that fell short and “really failed” the CA. These 13 banks needed to prepare capital plans within two weeks of the announcement of the results, and were given up to nine months to cover the capital shortfall. If the required new equity could not be raised in private markets, the respective sovereign was called upon to provide a public sector fiscal backstop. Table 1 contains the names of the banks that failed and nearly failed the CA. The AQR part of the CA documented that, as of end-2013, the book values of banks’ assets needed to be adjusted downwards by e48 billion. The capital shortfall of e25 billion and asset value adjustments of e37 billion implies an overall impact, or additional risk, of e62 billion on banks. In addition, the review found that banks’ non-performing exposures needed to be adjusted upwards by e136 billion, to a total of e879 billion.
9
Table 1: Bank sample Banks are sorted by Equity Surprise (in ascending order), or, when the latter is missing, by CDS Change (in descending order). Stressed countries are highlighted. Equity Surprise and CDS Change are defined as the log-return between the 2-week average price preceding October 10, 2014 and the 2-week average price following October 27, 2014. Beta is the coefficient obtained regressing daily CDS log-return on daily equity log-return. The CA outcomes are codified as follows: P (Pass) if the bank meets the 5.5% CET1 ratio requirement under the adverse scenario and meets the 8% ratio under the baseline scenario; NP (Near-pass) if the bank does not meet either of the required ratios, but has already covered its capital shortfall; NF (Near-fail) if the bank does not meet the ratios, has not covered the shortfall, but its plans to raise capital have been deemed adequate; F (Fail) if the capital ratios are not met and none of the repairing measures has been implemented.
Bank Name
Country
Equity Surprise
CDS Change
Beta
Banca Monte dei Paschi di Siena SpA Banca Carige SpA Permanent TSB plc Alpha Bank SA Banco Comercial Portuguˆ es SA National Bank of Greece SA Eurobank Ergasias SA Piraeus Bank SA Deutsche Bank AG Liberbank SA Banca Popolare Di Milano Scarl Banco Santander SA UniCredit SpA Unione Di Banche Italiane Scpa Banco Bilbao Vizcaya Argentaria SA Banca Popolare di Sondrio Scpa Soci´ et´ e G´ en´ erale Banca Popolare Dell’Emilia Romagna SC Banca Piccolo Credito Valtellinese SC Banco de Sabadell SA Banco Popular Espa˜ nol SA BNP Paribas Bankinter SA Intesa Sanpaolo SpA KBC Group NV Banco BPI SA Groupe Cr´ edit Agricole Raiffeisen Zentralbank AG Banco Popolare SC IKB Deutsche Industriebank AG ING Bank NV Aareal Bank AG The Governor and Company of the Bank of Ireland Mediobanca - Banca di Credito Finanziario SpA Commerzbank AG Erste Group Bank AG
IT IT IE GR PT GR GR GR DE ES IT ES IT IT ES IT FR IT IT ES ES FR ES IT BE PT FR AT IT DE NL DE IE IT DE AT
-38.60% -35.23% -15.30% -13.33% -12.22% -12.13% -10.90% -8.59% -8.21% -8.18% -7.14% -7.03% -6.77% -6.33% -5.62% -4.77% -3.78% -3.60% -3.47% -3.36% -3.14% -2.96% -2.83% -1.93% -1.05% -0.77% -0.19% 0.61% 0.73% 0.76% 0.88% 1.62% 2.09% 2.80% 5.93% 10.07%
3.79%
-0.34
-1.44% 20.98% 6.89% 20.98% 20.96% 11.33% 10.67%
0.02 -0.23 -0.52 -0.08 -0.11 -0.02 -1.43
-4.03% 6.85% 4.78% 2.24% 6.78%
-0.25 -1.56 -1.17 -0.34 -1.93
6.20%
-1.03
-12.70% -16.70% 7.66% -9.85% 0.09% 5.59% 14.78% 8.16% -11.67% -13.13% -7.28% -3.58%
-0.03 0.06 -1.42 -0.02 -1.13 -0.35 -0.05 -0.78 -0.61 -0.52 0.05 -0.69
-4.68% 4.50% -2.39% -13.95%
-0.03 -1.43 -0.90 -0.69
Dexia NV Caixa Geral de Dep´ ositos SA C.R.H. Allied Irish Banks plc Banque PSA Finance Landesbank Baden-W¨ urttemberg ABN AMRO Bank NV Co¨ operatieve Centrale Raiffeisen-Boerenleenbank B.A. Landesbank Hessen-Th¨ uringen Girozentrale Bayerische Landesbank Norddeutsche Landesbank Girozentrale BAWAG P.S.K. RCI Banque HSH Nordbank AG SNS Bank NV AXA Bank Europe SA The Royal Bank of Scotland NV DZ Bank AG
BE PT FR IE FR DE NL NL DE DE DE AT FR DE NL BE NL DE
10
13.01% 6.79% 0.19% -1.04% -1.46% -1.94% -2.07% -2.10% -2.14% -2.61% -2.83% -2.84% -2.86% -2.89% -2.95% -3.72% -14.24% -16.46%
CA Outcome
F F F P F NF NF NP P NP F P P P P NP P NP NP P P P P P P P P P NP P P P P P P P NF P NP P P P P P P P P P P P P NP P P
Taking the AQR outcome as a basis, the ST part of the CA found that a severe macrofinancial stress scenario would deplete the banks’ top-quality, loss-absorbing Common Equity Tier 1 (CET 1) capital – the measure of a bank’ financial strength – by about e263 billion. If the severe stress scenario were to realize, the median CET1 ratio taken over all participating banks was predicted to decrease by 4 percentage points from 12.4% to 8.3%. This reduction was substantially higher than in previous similar exercises, such as the stress tests previously undertaken by the European Banking Authority (EBA) between 2010-2013. This harshness may have contributed to a perception that, this time, the results were sufficiently bad to be credible. Importantly, a vast majority of the banks that did badly in the CA turned out to be located in countries that had experienced above-average sovereign stress during the euro area sovereign debt crisis. For example, 9 out of the 25 banks that failed the CA were located in Italy. Other banks with capital shortfalls were located in Spain, Portugal, and Greece. Banks located in countries that experienced less sovereign stress during the euro area sovereign debt crisis, and less of the associated economic woes, received a relatively clean bill of health. Table 1 lists the banks for which we have either liquid CDS data or equity data available, and which we use in our empirical analysis below.
3
Data
We consider CDS data and bank equity data at the daily frequency, as well as CA outcomes and other balance sheet items for banks covered by the CA. CDS spreads are obtained from the Credit Market Analysis (CMA) database through Thomson/Reuters Datastream. All CDS are at the 5 year maturity. We choose the fullrestructuring credit event clause, as this is the standard contract documentation for Western European sovereign reference entities. Unlike sovereign bonds or bank bonds, CDS are standardized products with pre-determined and comparable contractual agreements allowing for a consistent comparison of credit risk across banks and sovereigns. 11
Bank equity data are taken from Bloomberg for the subset of CA banks that are listed at a stock exchange. Bank stock returns are based on daily closing prices. Finally, data on the CA outcomes are available from the ECB’s public website. Additional bank balance sheet items are obtained from Bureau van Dijk’s Bankscope database. Table 2 reports the relevant summary statistics. Table 2: Summary statistics We report summary statistics for bank CDS, sovereign CDS, equity data, and a control covariate. The horizontal axis distinguishes the three time periods as explained in Section 2. Pre-CA
Soft Info
Post-CA
Mean
5p
95p
Std Dev
N
Mean
5p
95p
Std Dev
N
Mean
5p
95p
Std Dev
N
Bank CDS Non-stressed Stressed (no IT) IT All
104.50 228.80 139.70 147.58
48.31 72.69 79.72 54.03
188.26 390.30 238.59 388.49
54.86 112.25 55.71 94.82
270 150 70 490
108.93 253.99 159.09 161.03
52.33 79.00 84.20 55.81
205.31 469.95 273.84 425.89
55.13 130.67 67.42 108.29
265 150 70 485
104.73 251.30 138.38 155.44
52.45 77.36 81.66 54.22
207.40 489.31 250.06 470.03
54.56 147.17 54.62 114.42
260 150 70 480
Sovereign CDS Non-stressed Stressed (no IT) IT All
30.69 208.38 112.33 109.93
18.28 52.38 109.88 18.38
45.99 539.50 117.80 531.11
11.65 193.15 2.22 147.83
50 40 10 100
35.58 259.25 134.93 134.98
19.32 58.36 115.26 20.03
57.31 730.49 146.91 713.18
15.08 251.84 10.37 190.62
50 40 10 100
36.15 281.97 139.21 144.78
19.48 61.78 133.37 19.94
56.69 757.71 142.86 747.17
15.48 276.97 3.03 209.54
50 40 10 100
Equity Non-stressed Stressed (no IT) IT All
23.72 2.68 4.07 9.41
0.92 0.09 0.11 0.10
50.53 9.17 11.23 41.44
15.16 2.96 3.37 12.79
107 140 110 357
22.55 2.58 3.96 9.10
0.85 0.07 0.09 0.09
48.29 8.79 11.50 39.24
14.11 2.86 3.38 12.13
110 140 110 360
23.51 2.58 3.95 9.47
0.93 0.07 0.08 0.08
48.89 8.69 11.33 41.09
14.47 2.84 3.40 12.63
110 136 110 356
VIX
16.75
14.55
21.24
1.96
10
20.60
16.08
26.25
4.01
10
14.47
13.12
16.04
0.81
10
4
Empirical methodology
Our empirical analysis proceeds in four consecutive steps. We first need to demonstrate that the ECB’s announcement of the CA results constitutes a significant information shock that led to a pronounced reassessment of banks’ default risk and which, consequently, also affected sovereign risk in the euro area. Taking this as a given, Section 4.1 shows how we relate the variation in sovereign risk to the risk of the respective domestic banks in a difference and difference-in-differences setting. The former specification allows us to test for (time) differences in the within-country risk sensitivities around the time of the CA announcement, while the latter specification allows us to test whether the time differences in risk sensitivities are different within stressed and non-stressed countries. Section 4.2 demonstrates how we quantify and test for cross-border contagion effects by relating sovereign risk to the risk of both domestic and foreign banks. Finally, Section 4.3 presents a time-varying parameter 12
model allows us to efficiently estimate the time variation in cross-country risk sensitivities directly over longer periods of time such as 2009–2014.
4.1
Sovereign-bank risk sensitivities within countries
We fist study the within-country sensitivity of sovereigns with respect to their own domestic banks, and how that risk sensitivity changes over time. To this purpose we specify a panel regression as ∆cdssj,t = α0 + α1 × Pt × ∆cdsbi,j,t + α2 × ∆cdsbi,j,t + α3 × Pt + α4 ∆vixt + δi + εi,j,t ,
(1)
where ∆cdsbi,j,t is the daily log change in CDS spread for bank i in country j at time t, ∆cdssj,t is the daily log change in CDS spread for sovereign j, P is a dummy variable that takes the value of one either during the Soft Info period or during the Post-CA period, and zero during the Pre-CA period. Bank fixed effects δi eliminate the influence of unobserved firm specific characteristics on the bank-sovereign risk sensitivity. ∆vixt is the log change in the volatility index (VIX). The VIX is included because it may be a common factor to both bank and sovereign CDS. The main coefficient of interest is α1 , and is expected to be positive if the perception of additional bank risk moves sovereign risk perceptions as well. Regression specification (1) contains repeated CDS values on the left hand side for country j if there are more than one bank i located in that country. This implies cross-sectionally dependent error terms εi,j,t at the country level. While the repeated values do not affect moment-based estimation, they do affect inference. We take this issue into account by clustering all standard errors at the country level in addition to the bank level, and use standard errors from a non-parametric (wild) bootstrapping procedure. Equation (1) implicitly supposes that within-country risk transmission is approximately similar in magnitude across euro area countries. However, there is no a-priori reason to expect that this is the case. In particular, risk transmission may be less pronounced in relatively more stressed countries, as these countries are perceived to be less likely to be able 13
to provide a comprehensive public sector fiscal backstop should additional banking sector risk materialize. The following difference-in-differences panel regression specification allows us to study cross-sectional differences in sovereign-bank risk sensitivities before and after the ECB’s CA announcement.
∆cdssj,t = α0 + α1 × Pt × ∆cdsbi,j,t × Sj + α2 × Pt × ∆cdsbi,j,t + α3 × Pt × Sj + α4 × ∆cdsbi,j,t × Sj + α5 × ∆cdsbi,j,t + δi + γt + εi,j,t ,
(2)
where ∆cdssj,t , ∆cdsbi,j,t , and Pt are as in (1), Sj is a new cross-sectional dummy variable that distinguishes banks in stressed and non-stressed countries. Firm and time fixed effects are given by δi and γt , respectively. We stress that daily time fixed effects are a strong set of controls that absorb the influence of common macroeconomic and financial factors. In addition, the bank fixed effects eliminate the impact of unobserved bank heterogeneity.
4.2
Risk sensitivities across countries
A bank risk shock may propagate across borders to affect other sovereigns, in addition to the one it is located in. The following regression specification relates sovereign risk in nonstressed (n-s) countries in the euro area to domestic and foreign bank risk. ∆cdss,ns = α0 + α1 × Pt × ∆equityb,st + α2 × ∆equityb,st t t j,t + α3 × ∆cdsbi,j,t + κ0 X(i),j,t + δi + γt,week + εi,j,t ,
(3)
where ∆cdss,ns j,t refers to the daily log change in the CDS of a non-stressed sovereign j at time t, ∆equityb,st is the average daily equity return of banks located in stressed countries, t ∆cdsbi,j,t are daily log CDS changes of domestic banks in non-stressed countries, κ0 X(i),j,t are additional control variables, and γt,week are weekly time fixed effects. CDS in non-stressed countries may also be affected by implicit cross-border government 14
guarantees, although arguably to a much smaller extent. To take this issue into account, we also estimate b,st ∆cdss,ns + α2 × ∆equityb,st t j,t = α0 + α1 × Pt × ∆equityt 0 + α3 × ∆equityb,ns i,j,t + κ X(i),j,t + δi + γt,week + εi,j,t ,
(30 )
where ∆equityb,ns i,j,t refers to the daily change in log equity of a bank that is located in a non-stressed sovereign j at time t. Finally, we use both CDS and equity returns to proxy banking sector risk in stressed countries. This specification is useful since our bank samples for which we have CDS and bank equity data do not coincide, see Table 1. The resulting panel data specification is b,st + α2 × ∆equityb,st ∆cdss,ns t j,t = α0 + α1 × Pt × ∆equityt
+ α3 × Pt × ∆cdsb,st + α4 × ∆cdsb,st t t + α5 × ∆cdsbi,j,t + κ0 X(i),j,t + δi + γt,week + εi,j,t ,
(300 )
where equity and CDS covariates are defined as in (3) and (30 ).
4.3
A medium term perspective
Finally, we seek to to study medium term developments in the risk sensitivities of interest. To this purpose we introduce a time-varying parameter model that allows us to efficiently estimate the time-variation in risk sensitivities of sovereign CDS in non-stressed countries with respect to their own domestic as well as foreign banks. We will fit the model to weekly panel data from January 2009 to November 2014. We consider the panel regression model with time-varying parameters b,ns b,st ∆cdss,ns + it , j,t = γt + δi + βt ∆cdsi,j,t + κt ∆equityt
(4)
where ∆cdss,ns denotes the weekly difference in log CDS of non-stressed (ns) sovereigns, it
15
γt ∼ NID(0, σγ2 ) is a serially uncorrelated time effect, δi is a bank fixed effect, ∆cdsb,ns it is the difference in log CDS of banks located in non-stressed countries, ∆equityb,st is the t average weekly returns of banks located in stressed countries, and it is an idiosyncratic error term. Indexes i, j, t are as before. We are most interested in estimating the time-varying cross-country effect κt , controlling for domestic banks’ risks βt . To allow for time-variation in the measurement error between 2009-2014, we specify t = (1t , . . . , N t ) ∼ NID (0, Ht ) ,
(5)
where the covariance matrix is specified as Ht = diag (h1t , . . . , hN t ), where hit = σ2 · s,ns 2 CDSs,ns i,t−1 ≥ 0, σ a parameter to be estimated, and CDSi,t−1 the lagged CDS spread of
the respective non-stressed sovereign. As a result, error variances are serially correlated and higher during more stressful times; see Feldh¨ utter and Lando (2008) and Krishnamurthy et al. (2014) who point out the connection with a square-root interest rate process. While somewhat restrictive, the volatility specification (5) is parsimonious and sufficiently flexible to estimate the time-varying parameters and to test the key economic hypotheses at hand. In addition to the panel specification (4), we also consider the alternative specifications b,ns b,st ∆cdss,ns j,t = γt + δi + βt ∆cdsi,j,t + κt ∆cdsi,j,t + it ,
(40 )
b,ns b,st ∆cdss,ns + it , j,t = γt + δi + βt ∆equityi,j,t + κt ∆equityt
(400 )
as well as
which interchange CDS and equity data as average risk measurements. The time-varying within-country effect βt and cross-country effect κt capture the elasticity of sovereign CDS in non-stressed countries in the euro area with respect to risks to domestic and foreign banks, respectively. The time-varying parameters evolve over time as αt = (βt , κt )0 = I · αt−1 + ηt ;
ηt ∼ NID (0, Q) ,
(6)
where I is the identity matrix, ηt is the two-dimensional state equation error term, and Q 16
is a covariance matrix. The off-diagonal elements of Q are not necessarily zero, allowing the two time-varying parameters to be correlated. The time-varying coefficients are initialized as uninformative α1 ∼ N(0, κI), with κ → ∞; see Durbin and Koopman (2012, Chapter 5). Model (4) – (6), together with its initial condition, is a standard linear Gaussian model in state space form. The log-likelihood is easily obtained by a single run of the Kalman Filter (Hamilton (1994)). Similarly, (filtered) estimates of the time varying parameters and standard errors are also provided by the Kalman Filter. Inference on model parameters is straightforward as a result.
5
5.1
Main empirical results
The information shock due to the ECB’s CA announcement
This section studies the impact of the release of the CA results. We document that the information provided by the CA led to a substantial reassessment of banks’ risks, particularly the probability of requiring an equity recapitalization under stress. We also document that changes in bank risk perceptions changed the perception of sovereign risk. Figure 2 plots the cumulative log changes in equity prices on average over the period from September 29, 2014 to November 7, 2014. The dashed vertical line marks the begin of our Soft Info period, while the solid vertical line marks the announcement of the actual results on October 26, 2014. The figure distinguishes non-stressed countries (solid line), stressed countries without Italy (dashed line), and Italy (dotted line). Equity prices for banks located in stressed countries dropped sharply after the CA, by approximately 8–10 percent, compared to the Pre-CA period. Italian banks’ equity suffered the most, followed by banks located in Spain, Ireland, Portugal, and Greece. This is intuitive, given that the majority of banks that failed the ECB’s stress test were located in Italy and other stressed countries. By contrast, bank equity prices remained approximately unchanged in non-stressed countries. The overall level of risk in stressed countries moved up on average around the announcement of the CA results, with essentially a one to one pass-through from banks to sovereigns.
17
Figure 2: Cumulative bank equity returns in the euro area
-15
Equity Return (cumulative) -10 -5 0
5
Equity prices are from 29 September 2014 to 7 November 2014. In each panel, the dashed vertical line marks the beginning of the Soft Info period, while the solid vertical line marks the release of the CA results. The top panel plots the cumulative log changes in equity closing prices for non-stressed countries’ bank equities (solid line), stressed countries’ bank equities without Italy (dashed line), and Italian bank equities (dotted line). The bottom panel plots the cumulative log changes in equity of banks located in stress countries relative to the cumulative log changes in equity of banks located in non-stressed countries.
01oct2014
08oct2014
15oct2014
22oct2014
29oct2014
05nov2014
date Equity (Non-Stressed Countries)
Equity (Stressed Countries without Italy)
-10
Relative Equity Return (cumulative) -5 0
5
Equity (Italy)
01oct2014
08oct2014
15oct2014
22oct2014
29oct2014
date Relative Equity (Stressed without Italy)
18
Relative Equity (Italy)
05nov2014
Figure 3: Bank CDS and sovereign CDS in the euro area
220
140
160
180 200 Sov CDS Spread
Stressed Countries' (without IT) Bank CDS 230 260 240 250 270
220
CDS levels for both banks and sovereigns within the euro area (except Italy) from September 29, 2014 to November 7, 2014. In either panel the dashed vertical line marks the start of the Soft Info period, while the solid vertical line marks the announcement of the CA results on October 26, 2014. The top panel plots average CDS levels for stressed countries’ banks’ CDS (solid line), and stressed countries’ sovereign CDS (dashed line, without Italy). The middle panel plots average CDS levels for non-stressed countries’ banks’ CDS (solid line) and non-stressed countries’ sovereign CDS (dashed line). The bottom panel refers to Italy.
01oct2014
08oct2014
15oct2014
22oct2014
29oct2014
05nov2014
date Stressed (without IT) Sovereign CDS
50
100
60 55 Sov CDS Spread
Non-Stressed Countries' Bank CDS 105 110
65
115
Stressed Countries' (without IT) Bank CDS
01oct2014
08oct2014
15oct2014
22oct2014
29oct2014
05nov2014
date
120
140
160
180
Non-Stressed Sovereign CDS
100
100
120
CDS Spread 140Spread 160 CDS
180
Non-Stressed Countries' Bank CDS
01oct2014
08oct2014
15oct2014
22oct2014
29oct2014
05nov2014
date
01oct2014
08oct2014
15oct2014 Italian Bank CDS
22oct2014 29oct2014 Italian Sovereign CDS date
Italian Bank CDS
Italian Sovereign CDS
19
05nov2014
Figure 3 plots CDS levels for both banks and sovereign within the euro area except Italy from September 29, 2014 to November 7, 2014. Again, the dashed vertical line marks the Soft Info period, while the solid vertical line marks the announcement of the CA results on October 26, 2014. The top panel plots average CDS levels for banks (solid line) and sovereigns (dashed line) for stressed countries except Italy. The bottom panel refers to non-stressed countries, such as Germany and France. While both bank and sovereign CDS increase in most stressed countries (top panel), only sovereign risk increased in non-stressed countries. Bank CDS in non-stressed countries remained approximately flat, in line with the equity developments from Figure 2. As a result, the pronounced increase in sovereign CDS in non-stressed countries around the CA is not associated with an increase with the risk of their own banks, and needs to come from somewhere else. Figure 3 also plots the development of Italian bank and sovereign CDS. Strikingly, Italian banks’ average CDS are approximately unchanged post-CA compared to their pre-CA values, despite the stark 10% drop in equity prices post-CA and the observation that the majority of banks that failed the ECB’s stress test were headquartered in Italy. Italian sovereign CDS increased in line with that of other euro area sovereigns, indicating that overall bank risk perceptions in the euro area had increased.
5.2
The bank-sovereign nexus within countries
This section presents our empirical findings on the bank-sovereign risk nexus within countries. I.e., we focus on how sovereign risk depends on the changes in risk perceptions regarding their own domestic banks. The sovereign-bank nexus exists in the euro area, on average. Table 3 presents the results from relating the log changes in the sovereign CDS spreads in country j, ∆cdssj,t , to log changes in the CDS of their respective domestic banks ∆cdsbi,j,t . All banks and countries are included in the panel regression. We observe that risk sensitivities with respect to a sovereign’s own banks α1 and α2 are positive and significant across specifications. Comparing estimates across the three time periods suggests that coefficients are higher in the Soft Info and Post-CA than in the Pre-CA period. The coefficient estimates are robust to controlling 20
Table 3: Changes in bank-sovereign risk sensitivities This table reports the results from regressing the log changes in the sovereign CDS spreads in country j, ∆cdssj,t , on log changes in the bank CDS spread of the same country ∆cdsbi,j,t . Columns 1–2, 3–4, and 5–6 include observations from the Pre-CA, Soft Info, and Post-CA period, respectively. Columns 7 and 8 include all observations and report difference estimators: column 7 reports the the time differential effects between the Pre-CA and Soft Info periods; column 8 reports the time differential effects between the Pre-CA and Post-CA periods. Columns 2, 4, 6, 7, and 8 include bank fixed effects and changes in the volatility index (VIX) as controls. Standard errors are clustered at the bank level. Each column indicates whether the regression contains time (Time FE) and firm fixed effects (Firm FE).
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Pre-CA
Pre-CA
Soft Info
Soft Info
Post-CA
Post-CA
Diff (4)-(2)
Diff (6)-(2)
0.337*** (0.124) 0.019*** (0.001)
0.322*** (0.089)
∆Log(Bank CDS) x Post Soft Info FE Post-CA FE ∆Log(Bank CDS)
0.071** (0.027)
∆Log(VIX (US))
Observations R-squared Bank FE Cluster Bank
422 0.0100 NO YES
0.089*** (0.029) -0.030** (0.012)
0.406*** (0.129)
0.356** (0.134) 0.094*** (0.019)
0.395*** (0.082)
422 470 470 235 0.0613 0.1004 0.1356 0.1015 YES NO YES NO YES YES YES YES *** p<0.01, ** p<0.05, * p<0.1
21
0.419*** (0.103) -0.038 (0.086)
0.050 (0.032) 0.037*** (0.011)
-0.007*** (0.002) 0.084*** (0.027) -0.031* (0.015)
235 0.1399 YES YES
892 0.1290 YES YES
657 0.0970 YES YES
for bank fixed effects and including additional controls such as the U.S. VIX (or alternatively time fixed effects, not reported). The finding that bank and sovereign risk are correlated on average in the euro area is in line with Leonello (2014), Farhi and Tirole (2014), and Acharya et al. (2014). Additional risk in the banking sector, as uncovered by CA, lead to an increase in sovereign risk, with an elasticity parameter of approximately 0.3%. Comparing columns 2, 4, and 6, the average association between bank and sovereign risk goes up in the Soft Info period, and then remains at the higher level in the Post-CA period. This is in line with our earlier assessment that a substantial share of news regarding the CA results trickled into the market already during the Soft Info period. The time difference in risk sensitivity is approximately 0.3% compared to the Pre-CA value (Column 8). This is a causal effect on the risk sensitivity parameter due to the information shock. It implies that, on average, the additional risk borne by sovereigns from their own banking sectors, due to the risks uncovered by the CA, can be quantified as approx 0.3% times the change in the respective country’s banks’ CDS. Table 1 and Figure 2 indicates that the information shock was not uniform in magnitude across euro area countries. In particular, the differential equity price developments suggest that the additional risk uncovered by the CA was most pronounced for banks located in stressed countries, and much less pronounced for banks in non-stressed countries. It is therefore unlikely that the impact of the information shock is uniform across euro area countries. Table 4 separates stressed from non-stressed countries. Panel A (top) compares nonstressed countries with four stressed countries in the euro area (i.e., Spain, Portugal, Ireland, and Greece; without Italy). In the Pre-CA period, we observe no differences in risk transmission between the two groups of countries. In the Soft Info period, however, the risk sensitivity coefficient increases by 0.45% for stressed countries compared to non-stressed countries (column 7). In the Post-CA period, the sensitivity coefficient is lower than in non-stressed countries, and the risk sensitivity increased for the latter group (column 8). Panel B (bottom) of Table 4 compares non-stressed countries to Italy (i.e., stressed countries without Spain, Portugal, Ireland, and Greece). It is instructive to study Italy in 22
Table 4: Differential changes in within-country sensitivities This table reports the results from difference-in-differences regressions (2) which allow for country differences in the risk transmission from banks to their respective sovereigns. Panel A compares non-stressed countries with four stressed countries in the euro area (without Italy). Panel B compares non-stressed countries in the euro area with Italy. Columns 1–2, 3–4, 5–6 include only observations from the Pre-CA (29 September to 12 October 2014), Soft Info (13 to 26 October 2014), and Post-CA (27 October to 7 November 2014) period, respectively. Columns 7–8 report the difference-in-differences estimators. Columns 7 and 8 refer to time differences between the Pre-CA on the one hand and Soft Info and Post-CA periods, respectively. Soft Info FE and Post-CA FE are dummy variables that take the value 1 in the respective period, and zero in the Pre-CA period. Standard errors are clustered at the bank level. Each column indicates whether the regression contains time (Time FE) and firm fixed effects (Firm FE).
Dependent Variable: ∆Log(Sovereign CDS)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Pre-CA
Pre-CA
Soft Info
Soft Info
Post-CA
Post-CA
Diff-Diff (4)-(2)
Diff-Diff (6)-(2)
-0.263** (0.120) 0.083 (0.079) 0.292*** (0.097) -0.002 (0.003) -0.037 (0.076)
Panel A: Stressed Countries (without Italy) versus non-Stressed Countries Stressed (no IT) x Post x ∆Log(Bank CDS) Stressed (no IT) x ∆Log(Bank CDS)
-0.015 (0.073)
0.088 (0.084)
0.570*** (0.155)
0.528*** (0.147)
-0.428*** (0.124)
-0.190** (0.080)
0.063 (0.069) 0.006*** (0.002)
-0.038 (0.082)
0.315*** (0.102) -0.001 (0.002)
-0.198 (0.136)
0.593*** (0.078) 0.005** (0.002)
0.261*** (0.057)
0.455*** (0.155) 0.079 (0.080) -0.158 (0.141) -0.004 (0.003) -0.026 (0.078)
378 0.0130
378 0.2141
415 0.1411
415 0.5569
410 0.1006
410 0.5466
793 0.5051
788 0.4344
-0.421*** (0.113) 0.119 (0.081) 0.288** (0.107) -0.000 (0.003) -0.042 (0.084)
636 0.4877 YES YES YES
Post x ∆Log(Bank CDS) Stressed (no IT) x Post ∆Log(Bank CDS) Stressed (no IT) FE
Observations R-squared
Panel B: Italy versus non-Stressed Countries Italy x Post x ∆Log(Bank CDS) Italy x ∆Log(Bank CDS)
-0.002 (0.073)
0.122 (0.087)
-0.131 (0.202)
0.040 (0.153)
-0.582*** (0.084)
-0.309*** (0.066)
0.063 (0.069) -0.002 (0.001)
-0.043 (0.089)
0.315*** (0.103) 0.005*** (0.002)
-0.048 (0.115)
0.593*** (0.078) -0.002 (0.001)
0.253*** (0.060)
-0.080 (0.166) 0.111 (0.083) -0.006 (0.141) 0.007*** (0.001) -0.032 (0.085)
330 0.1091 NO NO YES
330 0.5888 YES YES YES
641 0.4055 YES YES YES
Post x ∆Log(Bank CDS) Italy x Post ∆Log(Bank CDS) Italy FE
Observations R-squared Bank FE daily Time FE Cluster Bank
306 0.0079 NO NO YES
306 335 335 0.2533 0.0417 0.4131 YES NO YES YES NO YES YES YES YES *** p<0.01, ** p<0.05, * p<0.1
23
isolation, for several reasons. First, 9 out of 25 failures (or near failures) were located in Italy. Second, CDS and equity prices decoupled around the announcement of the CA results, and to a larger extent than in the other four countries. Third, the violation of the sovereign ceiling in Italy on the day after the CA suggests that risk transfers play a particularly important role in this case. Finally, Italy is the largest of the five stressed countries, and potentially systemically important for the euro area as a whole. As the main lesson from Panel B of Table 4, there is no association between bank and sovereign risk in Italy. This is the case in each of the three periods: Pre-CA, Soft Info, and Post-CA. Italy is special in this regard: Risk sensitivity coefficients are different from the four other stressed countries in Panel A in the Soft Info period, and also significantly different from the non-stressed countries in Panel B in the Post-CA period. Column 8 suggests that, Post-CA, the bank-sovereign risk sensitivity has become significantly weaker in Italy than in non-stressed euro area countries. Post-CA, only the non-stressed countries experience significant bank-sovereign risk transfers, of about 0.3%. The uncovered pattern is consistent with the expectation that, if banking risk were to materialize on a large scale in Italy, it would be ESM funds (or the bank resolution fund) that would be used for bank equity recapitalization. This cross-border cross-sector risk transfer also explains the violation of the sovereign ceiling in Italy observed post-CA, as well as the higher risk sensitivities observed in non-stressed countries.
5.3
The bank-sovereign nexus across countries
This section formally tests for the presence of cross-country cross-sector risk spillovers in the euro area around the time of the ECB’s CA. We find that risk transfers across borders and sectors are statistically and economically significant. As a result, they can explain the somewhat counter-intuitive finding that the bank-to-sovereign risk transfers appear to be strongest in precisely those countries for which no additional banking sector risks were uncovered by the ECB’s CA. Table 5 reports the estimation results from estimating the panel data regression (3).
24
Coefficient α3 is always positive across specifications, suggesting that controlling for changes in the risk of a country’s own banks is important. Controlling for the direct effect, we find significantly negative loadings on foreign banks’ equity returns. I.e., a decline in the market value of banks located in stressed countries increases the sovereign risk of non-stressed countries. The final column reports the time differential effect in cross country and sector risk transfer around the ECB’s CA. Both the loading on the CDS of domestic banks and the equity returns of foreign banks are significant and economically large. Our estimates suggest that an exogenous decrease in average bank equity prices in stressed countries of 10 percent would lead to an average increase in sovereign CDS of non-stressed countries, with an elasticity of up to 7 percent, implying an increase in average sovereign CDS in non-stressed countries from approximately 52 basis points (bps) post-CA to approximately 56 bps. While such an increase is not extremely large, it is economically significant given the substantial amounts of sovereign debt that are refinanced each year in the euro area. Table 6 reports the estimation results from estimating the panel data regression (300 ). Most differential effects in risk sensitivities come from the change banking sector risk in stressed countries as measured by drops in market equity. The respective coefficient is robust to including the CDS data as an additional risk control and remains at approximately 0.6%.
25
Table 5: Changes in sensitivities from bank equity to sovereign CDS across countries The table reports the results from regressing log changes in the sovereign CDS spreads in country j, ∆cdssj,t , on log changes in average bank equity prices b of stressed countries ∆equityb,st t , controlling for domestic banks’ log changes in CDS ∆cdsi,j,t . Columns 1–3, 4–6, and 7–9 include only observations from the Pre-CA (29 September to 12 October 2014), Soft Info (13 to 26 October 2014), and Post-CA (27 October to 7 November 2014) periods, respectively. Columns 10 and 11 report the difference estimators that compare the Pre-CA to Soft Info and Pre-CA to Post-CA periods, respectively. Standard errors are block bootstrapped within bank clusters (Cluster Bank). Each column indicates whether the regression contains time (Time FE) and bank fixed effects (Bank FE).
VARIABLES
(1) Pre-CA
(2) Pre-CA
(3) Pre-CA
(4) Soft Info
(5) Soft Info
(6) Soft Info
(7) Post-CA
(8) Post-CA
(9) Post-CA
(10) Diff (6)-(3)
(11) Diff (9)-(3) -0.687*** (0.184) 0.046 (0.176) 0.282*** (0.083) -0.080*** (0.017) 503 0.1372 YES YES YES
∆Log Stressed Equity x Post
26
∆Log(Stressed Equity) ∆Log(Bank CDS)
-0.047 (0.161) 0.035 (0.088)
-0.144 (0.162) 0.029 (0.068) -0.074*** (0.014)
-0.149 (0.195) 0.023 (0.071) -0.074*** (0.015)
243 0.0964 NO YES YES
243 0.1528 NO YES YES
243 0.1868 YES YES YES
∆Log(VIX (US))
Observations R-squared Bank FE Weekly Time FE Bank Block Bootstrapped SE
0.005 (0.097) 0.225 (0.139)
0.593*** (0.168) 0.213 (0.133) 0.269*** (0.033)
0.590*** (0.148) 0.205 (0.140) 0.268*** (0.031)
265 265 265 0.1314 0.2398 0.2534 NO NO YES YES YES YES YES YES YES Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
-0.507*** (0.147) 0.528*** (0.099)
-0.626*** (0.084) 0.502*** (0.080) -0.107 (0.108)
-0.621*** (0.086) 0.514*** (0.073) -0.104 (0.102)
-0.098 (0.187) 0.193 (0.167) 0.137* (0.081) 0.071*** (0.016)
260 0.1588 NO YES YES
260 0.1622 NO YES YES
260 0.1790 YES YES YES
508 0.1738 YES YES YES
Table 6: Changes in sensitivities from bank equity to sovereign CDS across countries This table reports the results from regressing the log changes in the sovereign CDS spreads in country j, ∆cdssj,t on log changes in average bank equity b,st prices, ∆equityb,st of stressed countries and on log changes in the bank CDS spread of the same country ∆cdsbi,j,t . t , and average CDS prices, ∆cdst Columns 1–3, 4–6, and 7–9 include only observations from the Pre-CA (29 September to 12 October 2014), Soft Info (13 to 26 October 2014), and Post-CA (27 October to 7 November 2014) periods, respectively. Columns 10 and 11 report the difference estimators that compare the Pre-CA to Soft Info and Pre-CA to Post-CA periods, respectively. Standard errors are block bootstrapped within bank clusters (Cluster Bank). Each column indicates whether the regression contains time (Time FE) and bank fixed effects (Bank FE).
VARIABLES
(1) Pre-CA
(2) Pre-CA
(3) Pre-CA
(4) Soft Info
(5) Soft Info
(6) Soft Info
(7) Post-CA
(8) Post-CA
(9) Post-CA
(10) Diff (6)-(3)
(11) Diff (9)-(3) -0.639*** (0.203) -0.177 (0.253) 0.019 (0.158) 0.352 (0.255) 0.261*** (0.064) -0.122** (0.052) 503 0.1412 YES YES YES
∆Log Stressed Equity x Post
-0.434** (0.175) 0.195 (0.127) 0.498*** (0.084)
-0.606*** (0.196) 0.028 (0.238) 0.500*** (0.092) -0.098 (0.191)
-0.608*** (0.220) 0.018 (0.261) 0.513*** (0.099) -0.099 (0.198)
2.409*** (0.237) 3.730*** (0.190) -0.094 (0.177) -1.780*** (0.107) 0.062 (0.061) 0.273*** (0.023)
260 0.1609 NO YES YES
260 0.1622 NO YES YES
260 0.1790 YES YES YES
508 0.4250 YES YES YES
∆Log Stressed CDS x Post
27
∆Log(Stressed Equity) ∆Log(Stressed CDS) ∆Log(Bank CDS)
-0.094 (0.191) -0.388*** (0.067) 0.033 (0.086)
-0.175 (0.175) 0.459* (0.239) 0.025 (0.064) -0.140*** (0.046)
-0.180 (0.146) 0.460** (0.229) 0.019 (0.068) -0.140*** (0.046)
243 0.1325 NO YES YES
243 0.1596 NO YES YES
243 0.1937 YES YES YES
∆Log(VIX (US))
Observations R-squared Bank FE Weekly Time FE Bank Block Bootstrapped SE
1.401*** (0.160) 1.602*** (0.231) 0.103 (0.106)
2.730*** (0.136) 2.127*** (0.219) 0.045 (0.115) 0.398*** (0.023)
2.723*** (0.141) 2.130*** (0.229) 0.029 (0.112) 0.398*** (0.030)
265 265 265 0.3090 0.5276 0.5416 NO NO YES YES YES YES YES YES YES Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
We conclude that the amount of public sector risk sharing is both statistically and economically significant. The well-known bank-sovereign nexus has an important cross-border component.
5.4
Medium-term developments
This section presents our main results from applying a time-varying parameter approach. The top panel of Figure 4 plots estimates of the time-varying coefficients βt and κt , measuring the elasticity of sovereign CDS in non-stressed euro area countries with respect to changes in their own banks’ CDS and the mean equity returns of foreign banks, respectively, see (4). The bottom panel of Figure 4 plots the same time-varying estimates, but now replacing the foreign banks’ mean equity returns with the corresponding CDS measure, average weekly differences in log CDS. The top and bottom panels plot the weekly timeseries variation from 01 January 2014 onwards until 30 November 2014, five weeks after the end of the CA. We formulate three main findings from the medium-term analysis. First, the time variation in both parameters is approximately stable during almost all weeks during 2014, and usually not statistically different from zero. Importantly, our two-weeks Pre-CA period in Section 5 appears to be approximately representative of the medium term bank-sovereign relationship for both domestic and foreign banks during 2014. Second, risk sensitivity parameters change significantly during our Soft Info period, and stay approximately unchanged thereafter. The differential effects in the (maximum likelihood, weekly) time series estimates of elasticities are in line with the (moment-based, daily) difference-in-differences estimates as reported in Section 5. Our main findings are robust to very different econometric approaches, as well as to variations in the data sample. The (time difference in) cross-elasticity of non-stressed sovereigns with respect to average equity returns of banks in non-stressed countries increases in absolute value to about -0.5% during the post-CA period, controlling for bank CDS in non-stressed countries as well as additional effects. Our finding implies that an unexpected 10% decline in the market value of bank
28
Figure 4: Time-varying parameter estimates The top and bottom panels plot filtered estimates of the time-varying parameters βt and κt in panel regression equations (4) and (40 ), respectively. Standard error bands are reported at a 95% confidence level. Estimation sample is weekly data from January 2009 to November 2014. Reported values are from March 2014 to November 2014. Vertical lines distinguish the Pre-CA, Soft Info, and Post-CA period, respectively.
elasticity in pct
0.4
Filtered β t, loading on domestic banks’ CDS 0.95 SE band
0.2
0.0 PreCA
-0.2 03/2014
05/2014
06/2014
07/2014
08/2014
09/2014
PostCA 10/2014
11/2014
Filtered κ t, loading on foreign banks’ equity 0.95 SE band Pre- Soft PostCA Info CA
elasticity in pct
0.5
04/2014
Soft Info
0.0
-0.5 03/2014
0.4
04/2014
05/2014
06/2014
07/2014
08/2014
09/2014
10/2014
11/2014
Filtered β t, loading on domestic banks’ CDS 0.95 SE band
elasticity in pct
0.2
0.0
PreCA
-0.2 03/2014
elasticity in pct
1.5
04/2014
05/2014
06/2014
07/2014
08/2014
Filtered κ t, loading on foreign banks’ CDS 0.95 SE band
1.0
Soft Info
09/2014
PostCA 10/2014
Pre- Soft CA Info
PostCA
09/2014
10/2014
11/2014
0.5 0.0
-0.5 03/2014
04/2014
05/2014
06/2014
07/2014
29
08/2014
11/2014
equity in stressed countries would raise sovereign CDS in non-stressed countries by about 5%, from, say, a CDS level of 52 basis points (bps) on average to a level of 54 bps on average. Third, the cross-elasticity estimate differs substantially depending on whether relative changes in bank CDS or bank equity returns are used to approximate banking sector risk. This is intuitive, as both instruments respond differently to different economic risks. Bank CDS respond to the probability of a default on bank debt, while equity returns primarily reflect the risk of a bank (equity) recapitalization under stress and the accompanying dilution of existing shareholders. Given the current bail-in rules in the EU, as stipulated by the EU BRRD from December 2013, bank “risk” in stressed countries appears to be better captured by developments in equity prices, at least in our sample, rather than changes in CDS.
6
Conclusion
We documented the extent of cross-border risk sharing between the banking and sovereign sectors in the euro area. Based on difference and difference-in-difference estimates around the ECB’s announcement of the outcome of its Comprehensive Assessment on 26 October 2014, we show that the information shock to bank risk affected some domestic sovereigns, but also propagated across borders to impact the CDS of other sovereigns. We conclude that the often mentioned bank-sovereign nexus in the euro area has an important cross-border component.
References Acharya, V. and S. Steffen (2015). The “greatest” carry trade ever? Understanding eurozone bank risks. Journal of Financial Economics 115 (2), 215 – 236. Acharya, V. V., I. Drechsler, and P. Schnabl (2014). A pyrrhic victory? bank bailouts and sovereign credit risk. The Journal of Finance 69 (6), 2689–2739. Benzoni, L., P. Collin-Dufresne, R. S. Goldstein, and J. Helwege (2015). Modeling credit
30
contagion via the updating of fragile beliefs: An investigation of the european sovereign crisis. The Review of Financial Studies, forthcoming. Boissay, F. and R. Gropp (2012). Payment defaults and interfirm liquidity provision. Review of Finance 17(6), 1853–1894. Constˆancio, V. (2014). The role of the banking union in achieving financial stability. Keynote lecture at the FT Banking Summit “Ensuring Future Growth”, London, 26 November 2014 . Cooper, R. and K. Nikolov (2014). Government debt and banking fragility: The spreading of strategic uncertainty. Unpublished working paper . Draghi, M. (2014). Stability and Prosperity in Monetary Union. Speech at the University of Helsinki, 27 November 2014 . Durbin, J. and S. J. Koopman (2012). Time series analysis by state space methods, 2nd edition. Oxford: Oxford University Press. EC (2012). A roadmap towards a Banking Union. Communication from the Commission to the European Parliament and the Council, Brussels, 12 September 2012.. ECB (2014).
Aggregate report on the Comprehensive Assessment.
available at
https://www.ecb.europa.eu/pub/pdf/other/ aggregatereportonthecomprehensiveassessment201410.en.pdf.. Eser, F. and B. Schwaab (2015). Evaluating the impact of unconventional monetary policy measures: Empirical evidence from the ECB’s Securities Markets Programme. Journal of Financial Economics, forthcoming.. Farhi, E. and J. Tirole (2014). Deadly embrace: Sovereign and financial balance sheets doom loops. Unpublished working paper, Harvard University (wp164191). Feldh¨ utter, P. and D. Lando (2008). Decomposing swap spreads. Journal of Financial Economics 88 (2), 375–405.
31
Fratzscher, M. and M. Rieth (2015). Monetary policy, bank bailouts and the sovereign-bank risk nexus in the euro area. DIW Discussion Paper 1448. Hamilton, J. D. (1994). Time Series Analysis. Princeton University Press. Helwege, J. and G. Zhang (2012). Financial firm bankruptcy and contagion. Unpublished working paper . Hertzel, M. and M. S. Officer (2012). Industry contagion in loan spreads. Journal of Financial Economics 103, 493–506. Jorion, P. and G. Zhang (2007). Good and bad credit contagion: Evidence from credit default swaps. Journal of Financial Economics 84 (3), 860–83. Jorion, P. and G. Zhang (2009). Credit contagion from counterparty credit risk. Journal of Finance 64 (5), 2053–2087. Kallestrup, R., D. Lando, and A. Murgoci (2013). Financial sector linkages and the dynamics of bank and sovereign credit spreads. Unpublished working paper. Krishnamurthy, A., S. Nagel, and A. Vissing-Jorgensen (2014). ECB policies involving government bond purchases: Impact and channels. Unpublished working paper . Lang, L. and R. M. Stulz (1992). Contagion and competitive intra-industry effects of bankruptcy announcements. Journal of Financial Economics 8, 45–60. Leonello, A. (2014). Government guarantees and the two-way feedback between banking and sovereign debt crises. Unpublished working paper. Lucas, A., B. Schwaab, and X. Zhang (2014). Conditional euro area sovereign default risk. Journal of Business and Economics Statistics 32 (2), 271–284. Theocharides, G. (2007). Contagion: Evidence from the bond market. Unpublished working paper . WSJ (2013). Spain pours billions into bank. Wall Street Journal, 25 May 2013.
32
A
Additional results
Table 7: Changes in sensitivities from bank equity to sovereign CDS This table reports the results from regressing the log changes in the sovereign CDS spreads in country j, ∆cdssj,t , on log changes in the bank equity prices of the same country ∆equitysi,j,t . Columns 1–2, 3–4, and 5–6 include only observations from the Pre-CA (29 September to 12 October 2014), Soft Info (13 to 26 October 2014), and Post-CA (27 October to 7 November 2014) periods, respectively. Column 7 refers to the time differences between the Pre-CA and the Soft Info periods, with Soft Info FE a dummy variable that takes the value 1 within the Soft Info period, and zero for the Pre-CA period. Column 8 refers to the differences between the Pre-CA and the Post-CA periods, with Post-CA FE a dummy variable that takes the value 1 within the Post-CA period, and zero for the Pre-CA period. Bank fixed effects and changes in the volatility index (VIX) are added sequentially. Standard errors are clustered by bank (Cluster Bank). Each column indicates whether the regression contains time (Time FE) and firm fixed effects (Firm FE).
VARIABLES
(1) Pre-CA
(2) Pre-CA
(3) Soft Info
(4) Soft Info
(5) Post-CA
(6) Post-CA
∆Log(Stock Price) x Post Soft Info FE
(7) Diff (4)-(2)
(8) Diff (6)-(2)
-0.358** (0.138) 0.011*** (0.002)
0.010 (0.167)
Post-CA FE ∆Log(Stock Price)
-0.231* (0.125)
∆Log(VIX (US))
Observations R-squared Bank FE Cluster Bank
266 0.0126 NO YES
-0.197 (0.133) 0.126*** (0.035)
-0.671*** (0.130)
-0.644*** (0.138) 0.020 (0.023)
-0.267** (0.112)
266 300 300 142 0.1296 0.1724 0.1932 0.0254 YES NO YES NO YES YES YES YES *** p<0.01, ** p<0.05, * p<0.1
33
-0.316* (0.159) -0.003 (0.066)
-0.205 (0.127) 0.075*** (0.015)
0.001 (0.004) -0.180 (0.131) 0.117*** (0.034)
142 0.1226 YES YES
566 0.1380 YES YES
408 0.0957 YES YES
