REGULATORY ARBITRAGE IN CROSS-BORDER MERGERS OF EU BANKS

Transcription

1 1 REGULATORY ARBITRAGE IN CROSS-BORDER MERGERS OF EU BANKS Santiago Carbó-Valverde (University of Granada and Federal Reserve Bank of Chicago*) Edward Kane (Boston College) Francisco Rodríguez-Fernández (University of Granada) * The views expressed are those of the author and do not represent the views of the Federal Reserve Bank of Chicago or the Federal Reserve System.

2 Motivation 2 Thesis: The Current Crisis shows that undersupervision of systemically important financial institutions (SIFIs) is a problem. If so, shouldn t this weakness be exploited in cross-border merger activity which is not specifically examined for safetynet consequences? Safety nets displace some degree of private market discipline. They subsidize hard-to-monitor risk in good times. This is partly because the accounting frameworks used by banks and government officials make no one directly accountable for reporting, monitoring, or controlling subsidies until and unless markets become turbulent.

3 3 The costs and benefits a country receives from its safety net depend on how much market discipline the net displaces and how successfully safety-net managers substitute oversight for the market discipline they displace: dangerous gaps exist in safety-net managers vision, tools, and incentives. By engaging in regulation-induced innovation, building clout and exerting lobbying pressure, a country s SIFIs can keep tail risk from being adequately disciplined. No reason to expect industry clout or safety-net benefits to be the same in all countries.

4 4 Because authorities can observe leverage much better than asset-return volatility, an increase in a bank s exposure to deep tail risk can almost always increase the value of its safety-net benefits. This incentivizes SIFIs to search out and to exploit weaknesses (i.e., loopholes) in informational flow & risk-control arrangements. The per-period flow of safety-net benefits that a particular institution enjoys can be defined as a fair insurance premium percentage (IPP) expressed per euro of a bank s deposits.

5 5 This paper indicates that, in years leading up to the crisis, differences across countries --either in rules or in their enforcement --led to statistically significant differences in IPP. A second issue is whether these differences might have encouraged cross-border mergers by EU banks --at least in part-- as a way to expand their access to safety-net subsidies.

6 6 How, why, and for whom individual mergers and acquisitions generate net economic benefits is an increasingly important policy issue. In the banking industry, besides the usual suspects of efficiency & market power, the existence of safetynet subsidies introduces two special sources for concern: Opportunity costs generated by individual-country policies of entry or exit resistance. The possibility that the merger or acquisition represents a form of regulatory arbitrage that leads to more efficient exploitation of national safety nets.

7 7 Opportunities for regulatory arbitrage occur when, by changing the geographic footprint of their activities, financial institutions (and some of their counterparties) can shift poorly monitored risk exposures to taxpayers in one or another country on advantageous terms (Kane, 2000; Carbo, Kane, and Rodriguez, 2008; Campa and Hernando, 2008). Previous evidence as to whether regulatory arbitrage is a motive in cross-border mergers comes mainly from event studies that follow two empirical steps: Researchers first use one or more forms of marketmodel regression to identify significant shifts in parametric measures of value or risk-taking at partner banks during or after merger events. When significant parameter shifts are observed, the estimated shift is regressed on various characteristics of one or the other merger partner and on structural characteristics of the markets, economies, or regulatory systems within which these firms operate.

8 8 Most of these papers (Amihud, DeLong, and Saunders, 2002; Buch and DeLong, 2008) conclude that regulatory arbitrage has posed little problem for EU authorities Although these results are comforting, they are less than fully convincing. This is because they leave open some critical loose ends: The second stage of these event studies has limited power to reject the hypotheses of no effect. Neither paper directly estimates or controls for differences in safetynet benefits across countries. While these papers incorporate indirect measures based on differences across countries in the scope of regulatory and supervisory powers, the models used do not and cannot control for variation in the intensity with which authorities monitor individualbank risk exposures or exercise their enforcement authority when excessive leverage or other forms of inappropriate risk-taking is observed. The possibility that merger partners differ from other banks with respect to the second-stage regressors (i.e., the issue of sampleselection bias) is not explored.

9 9 MAIN AIM OF THIS PAPER: this paper examines whether and how EU banks that engage in crossborder mergers (CBM banks) differ from other EU banks with respect to the safety-net benefits they extract or how effectively risk-shifting controls restrain their incremental risk-taking. Carbo, Kane, and Rodriguez (2008) synthetically estimate differences in safety-net benefits and in supervisory effectiveness for EU-15 countries excluding Greece. These estimates use Hovakimian and Kane s (2000) adaptation of the two-equation model of capital discipline and safety-net control devised by Duan, Moreau, and Sealey (1992). Our methods and models generate a metric that can be used to determine the extent to which safety-net benefits attach to merger deals.

10 10 1. Three synthetic variables are analyzed in our regressions: V, σ V and IPP 2. These variables are constructed from four observable variables: B : total debt: computed as the difference between the book values of total assets and common equity. E : the market value of a bank s equity: computed as the end-ofperiod stock-market capitalization. σ E ; standard deviation of the return on equity: computed as the standard deviation of deleveraged quarterly holding-period returns on stock for commercial banks. δ: fraction of bank assets distributed yearly as dividends to stockholders. These variables are taken directly from the Bankscope database, provided by BureauVan Dijk.

11 11 For the observables, we use the same Bankscope dataset as in Carbo et al (2008). Duan (1994) offer an improved maximum-likelihood method of estimation that provides an statistical (maximum-likelihood) estimation of the value of assets. Applying an improved maximum-likelihood method of estimation (Duan 1994; Duan and Simonato, 2002) to the same model and the same Bankscope dataset, this paper shows that -- both within and across countries -- significant differences exist in risk-taking and access to safety-net subsidies between CBM and other commercial banks. The new method yields smaller, but similar differences between banks that Carbo, Kane and Rodriguez (2008) designate as country-champion banks and other banks in the sample. In this paper, we label these champion banks as too difficult to fail and unwind (TDFU) on the grounds that they are large and complex enough to compete in international markets and politically and administratively difficult to force into receivership. TDFU banks are defined as those whose total assets fall in the first size decile of EU banks.

12 12 I. Modeling Safety-Net Benefits as a Function of Asset-Return Volatility and Capital Controls The procedure follows Merton (1977) in portraying deposit insurance as a single-period European put option on an insured bank s assets. Ronn and Verma (1986) adapt Merton s model to account for the likelihood that safety-net managers would forbear from exercising their right to call the put when their claim is only slightly in the money (The RV model scales down the effective exercise price of the put for all banks and all dates by a factor of ρ=0.97).

13 13 Merton (1977, 1978) shows that the IPP increases both with a bank s leverage and with the volatility of its returns. In Merton s model, leverage is measured as the ratio of the market value (D) of deposits and other debt to the market value of a bank s assets (V). Volatility is defined as the standard deviation of the return on bank assets (σ V ). Risk-shifting occurs when creditors or guarantors are exposed to loss without receiving adequate compensation. Our model linearizes Merton s model of deposit insurance (1977, 1978).

14 14 The variable IPP expresses the fair premium for safety-net support per Euro of debt as an increasing function of a bank s asset risk (σ v ) and leverage. Leverage is measured as the ratio of the face value of an institution s debt (B) to the estimated market value of its assets (V). We Reframe the Merton model to endogenize the debt-to-asset ratio, B/V. We conceive of leverage and IPP as the outcome of a bargaining game between the bank, its counterparties, and its regulators as establishing two loci of potential marketequilibrium points.

15 15 ASSUMED ASYMMETRIES IN INFORMATION PLAY A KEY PART IN THE BARGAINING GAME 1. Bank Managers Know; a. the true values of B/V, σ v, and IPP and set them and other structural variables (such as off balance sheet activity and geographic footprint) to maximize V-B. b. regulators can be made to underestimate leverage, volatility, and IPP. 2. Counterparties have unbiased estimates of these variables and are willing to substitute safety-net support for private support from the bank s capital. 3. Because of incentive conflicts, staffing limitations, and blockages in their access to information, in their enforcement activity regulators may be maneuvered into using underestimates of the values of all three variables.

16 THE LOCI 16 B/V = α 0 + α 1 σ V + ε 2 IPP = β 0 + β 1 σ V + ε 3 (1) (2) Equation (1) expresses the hypothesis that in each round of action and response, outside monitoring constrains banks to choose points that lie on a locus of mutually acceptable leverage and volatility pairs. We focus on bank changes in σ V and regulator and market disciplinary responses to them.

17 17 The slope coefficients in (1) and (2) may be interpreted as follows: d B V α ( / ) 1 dσ, (3) v IPP IPP β 1 + α1 = γ 1 + γ 2α1. (4) σ ( B / V ) v By themselves, the positive partial derivatives that are shown in equation (4) tell us how much value bank stockholders could extract from the safety net if managers were free to make unconstrained adjustments in volatility and leverage, respectively. However, in practice, safety-net officials and important private counterparties insist on having at least some power to monitor and constrain bank risk taking.

18 18 In portraying volatility as an exogenous variable, the model assumes that σ V represents a value-maximizing decision made in response to rational expectation of returns on a vector of unobserved bank-specific profit generators. The underlying intuition is that of a dynamic rational-expectations game: each bank sets its leverage and asset volatility jointly, with the understanding that creditors and safety-net managers monitor these values and restrain leverage by raising funding costs and standing ready to take over the franchise in the event of insolvency. In turn, creditors and regulators expect banks to react ex-ante to the discipline they are expected to exert.

19 19 Given the external discipline a bank faces, the sign of β 1 in equation indicates whether a bank s covenanted contracting environment allows increases in asset volatility to increase the value of its safety-net guarantees. Empirically, the total derivative β 1 must be non-positive for risk-shifting incentives in a given country to be fully neutralized. A negative α 1 only implies that risk-sensitive capital regulation and complementary market discipline partially constrain realizable safety-net benefits.

20 20 Thus, for market and regulatory pressure consistently to discipline -- and potentially to neutralize -- incremental risk-shifting incentives, two conditions must be met: Capital must increase with volatility: α 1 < 0 Guarantee value must not rise with volatility: β 1 0.

21 II. A Preliminary Look at the Focal Variables 21 To identify cross-border merging banks, we use the Thompson One-Banker M&A database for the European Union. This source permits us to identify target and acquirer banks. We also require that the selected mergers be registered as completed deals in the European Central Bank registry of banks.

22 22 The total sample of European banks consists of panel observations from 1993 to There is also a sub-sample of pre-merger banks (292 obs.) and post-merger banks (155 obs.) and a subsample that distinguishes between acquiring banks (282 obs.) and target banks (165 obs.). Acquiring banks are labelled in the Thomson database as those that initiated and led the M&A process. Pre-merger banks are considered as a pro-forma combination of the values or partner merging banks in the pre-merger period.

23 23 Table 1 compares mean values for B/V, IPP, and σv for other banks in a country with those for CBM banks. Because no Danish or Finnish bank engaged in a crossborder merger during , these countries join Greece in dropping out of our analysis. Except for Spain and Germany, CBM banks extract higher mean benefits from country EU safety nets than other banks do. Leverage is higher for CBM banks in three fourths of the cases, while increases and decreases in asset volatility divide almost equally.

24 TABLE 1 MEAN LEVERAGE RATIOS (B/V), MEAN FAIR PREMIUM (IPP), AND VOLATILITY OF RETURN ON ASSETS (σ V ): ALL BANKS VS. CROSS- BORDER MERGING BANKS Value of fair premiums generated by the procedure of Ronn and Verma-RV (JF, 1986) and Duan s Maximum-Likelihood (MF, 1994) 24 All banks (excluding cross-border Cross-border merging banks merging banks) Country B/V (%) IPP (%) σ V (%) B/V (%) IPP (%) σ V (%) RV ML RV ML RV ML RV ML RV ML RV ML Austria Belgium Denmark Finland Luxembourg Netherlands Portugal Sweden Ireland United Kingdom Spain France Italy Germany All estimated parameters are significant at the 1% level The differences between the RV and the ML estimated parameters where found to be statistically significant in all cases according to the mean-difference tests.

25 25 Table 2 (not shown for simplicity) indicates separately for all banks and for CBM banks that leverage, fair premiums, and asset volatility differ significantly for most country pairs. This supports the hypothesis that selectively extending a bank s operations into another EU country can indeed lower the firm s overall regulatory burden. For example, a bank can book risk exposures on which a home country enforces a high effective capital requirement in subsidiaries located in countries that treat these particular exposures less onerously.

26 26 Tables 3 and 4 aggregate the data globally. Table 3 shows a negative correlation between B/V and IPP. Table 4 establishes that on average CBM banks achieve slightly and insignificantly higher leverage and safety-net benefits than TDFU banks do, while other banks trail significantly in both respects. It also shows that, after a cross-border merger, leverage and safety-net benefits increase substantially.

27 27 TABLE 3. CORRELATIONS BETWEEN B/V AND IPP ACROSS COUNTRIES FOR THREE GROUPS OF BANKS RV ML TDFU BANKS CROSS-BORDER MERGING BANKS OTHER BANKS (excluding TDFU and cross-border merging banks)

28 TABLE 4 MEAN LEVERAGE RATIOS (B/V), MEAN FAIR PREMIUM (IPP), AND VOLATILITY OF RETURN ON ASSETS (σ V ): ALL BANKS, TDFU BANKS AND CROSS-BORDER MERGING BANKS 28 Country B/V (%) IPP (%) σ V (%) RV ML RV ML RV ML OTHER BANKS (excluding TDFU and cross-border merging banks) TDFU BANKS CROSS-BORDER MERGING BANKS Pre-merger Post-merger Mean difference tests: OTHER BANKS vs. TDFU BANKS Mean difference tests: OTHER BANKS vs. CROSS BORDER MERGING BANKS Mean difference tests: TDFU BANKS vs. CROSS-BORDER MERGING BANKS All estimated parameters are significant at the 1% level The test statistics report the p value of a one tailed t test of the hypothesis that the means are equal for the indicated groups.

29 III. Evidence that Cross-Border Mergers Offer Partner Institutions Incremental Regulatory Relief and Safety-Net Benefits 29 We undertake a series of difference-on-difference regression equations (since we compare differences in the groups of banks we are interested in with other control groups of banks) in which errors are clustered at the individual-bank level. The first column of Table 5 shows that across the 12 sample countries, accounting capital is subject to less and less discipline as asset size increases. However, although CBM banks receive more discipline, the second column shows that this discipline does not prevent them from extracting incremental safety-net benefits. At the margin, CBM banks find ways to expand their portfolio risk that extract safety-net subsidies.

30 TABLE 5 SINGLE-EQUATION ESTIMATES OF THE EFFECTIVENESS OF SAFETY- NET CONTROL IN THE EU-12 INCLUDING ASSET SIZE AS A REGRESSOR Fixed-effects panel regressions relating changes in a bank s leverage, ( B/V), and 30 changes in its fair deposit insurance premium, IPP, to the riskiness of its assets, σ V. B is the face value of bank s debt, including deposits. V is the market value of bank assets. The second and third columns report the value of α 1 and β 1, respectively. The errors are clustered at the firm level (B/V) IPP RV ML RV ML σ V ** (-32.15) ** (-44.85) 0.005** (24.19) ** (-36.71) Size 0.015** (26.18) 0.012** (31.01) ** (-19.37) ** (22.23) σ V X cross-border M&A dummy ** (-4.17) ** (6.33) 0.013** (3.81) 0.018** (5.88) Observations R * Statistically significant at 5% level ** Statistically significant at 1% level

31 31 Table 6 contrasts CBM banks pre-merger and postmerger experience, suppressing the size term. It shows that, although accounting capital is policed roughly twice as closely after a cross-border merger, CBM banks incremental access to safety-net benefits doubles. Wald tests confirm that these differences are statistically significant.

32 TABLE 6 PRE- AND POST-MERGER RISK-SHIFTING BEHAVIOUR AT CROSS- BORDER MERGING BANKS Fixed-effects panel regressions relating changes in a bank s leverage, ( B/V), and changes in its fair deposit insurance premium, IPP, to the riskiness of its assets, σ V. B is the face value of bank s debt, including deposits. V is the market value of bank assets. The first entries of the second and third columns report the value of α 1 and β 1, respectively. 32 Pre-Merger (B/V) IPP RV ML RV ML σ V ** (-3.93) ** (5.95) 0.008** (6.16) 0.010** (8.27) Observations R Post-Merger (B/V) IPP σ V * (-2.41) ** (3.58) 0.016** (7.74) 0.015** (7.87) Observations R TEST OF THE DIFFERENCES IN σ V BETWEEN PRE AND POST-MERGER PERIODS (p-value) * Statistically significant at 5% level ** Statistically significant at 1% level

33 33 Table 7 indicates that discipline and benefits accrue differently at target and acquiring banks. Although acquiring banks (who presumably initiate cross-border deals) face significantly more capital discipline, they extract significantly more safety-net benefit at the margin than targets do. These findings strongly support the hypothesis that the pursuit of safety-net benefits help to motivate cross-border merger activity.

34 TABLE 7 PRE-MERGER RISK-SHIFTING AT CROSS-BORDER MERGING BANKS: ACQUIRING VS. ACQUIRED BANKS Fixed-effects panel regressions relating changes in a bank s leverage, ( B/V), and changes in its fair deposit insurance premium, IPP, to the riskiness of its assets, σ V. B is the face value of bank s debt, including deposits. V is the market value of bank assets. The second and third columns report the value of α 1 and β 1, respectively. Errors are clustered at the firm level 34 (B/V) IPP RV ML RV ML σ V ** (-7.18) ** (11.35) 0.003** (2.96) 0.005** (2.99) σ V X acquiring banks dummy ** (-5.92) ** (-8.35) 0.008* (2.14) 0.005** (3.23) Size 0.012** (24.15) 0.016** (30.18) ** (13.53) * (2.30) Observations R * Statistically significant at 5% level ** Statistically significant at 1% level

35 35 A growing empirical literature emphasizes the role of size and relative profit performance as motives for banks to merge. This leads us to hypothesize that targets or acquirers might be especially large and might be responding to changes in their safety-net benefits when a cross-border deal is initiated. In Table 8 we propose a selection equation to control for potential selection bias.

36 TABLE 8 (SEE COMPLETE TABLE IN THE PAPER): SELECTION EQUATION FOR CBM BANKS: FIXED-EFFECTS PROBIT REGRESSIONS FOR EACH SAMPLE YEAR AND EXPLAINING THE CBM-BANK DUMMY (1=CROSS-BORDER MERGING BANK; 0=NON-MERGING BANK) AS A FUNCTION OF SELECTED BANK CHARACTERISTICS. σv, IPP and B/V correspond to Duan s ML estimations IPP 1.719** (4.96) B/V ** (-8.55) Bank size 0.091** (6.18) Bank inefficiency (0.74) Bank capitalization ** (-4.51) Bank liquidity (1.40) Intangible capital ratio 17.62** (4.22) Non-deposit debt ratio 1.38** (5.16) Ownership concentration (0.50) Observations Log-likelihood Fraction of correct predictions 0.99 * Statistically significant at 5% level ** Statistically significant at 1% level

37 37 We do this by introducing the Mills odds ratio from time-series or cross section Heckman selection equations into our baseline models. This ratio can be added in our main estimated equations and lets us sort out the effects of a bank s leverage and access to safety-net benefits from the influence of asset size and other potential M&A determinants on a bank s decision to participate in a cross-border deal. Table 8 (see the paper): high values of IPP, size, nondeposit debt, and intangible assets consistently encourage CBM activity, while leverage and tangible capital restrain it. In contrast to studies that examine within country mergers, measures of operating inefficiency, liquidity, and ownership concentration are never significant.

38 38 To account for the potential endogeneity of any classificatory variable, we adopt Heckman s procedure (1976, 1978). This introduces into our previous models a variable Heckman calls Lambda. This variable is also known as Mill s inverse odds ratio ( Mills ratio ). It measures the covariance between the error terms of the single-equation regression for an endogenous variable with the residuals from the selection equation. In our tables, the coefficient assigned to the Mills ratio measures how surprising it is to learn that a particular bank is either engaging in a cross-border merger or (in Tables 11 and 13) acquiring a bank in another country.

39 39 In Table 9, Lambda proves significantly negative in both panels. This indicates that incremental leverage and safety-net benefits are algebraically larger, the less surprising it seems for a particular bank to be engaging in a cross-border M&A. Table 10 reports year-by-year and pooled equations (not shown for simplicity) for selecting acquirers from targets. Our findings contrast with the literature on strictly domestic M&As in that leverage, size, inefficiency, nondeposit debt, and ownership concentration are never significant. Instead, safety-net benefits, intangible capital, and liquidity prove to be positive predictors for being an acquirer.

40 TABLE 9 PRE- AND POST-MERGER RISK-SHIFTING AT CROSS-BORDER MERGING BANKS Fixed-effects panel regressions relating changes in a bank s leverage, ( B/V), and changes in its fair deposit insurance premium, IPP, to the riskiness of its assets, σ V. B is the face value of bank s debt, including deposits. V is the market value of bank assets. The second and third columns report the value of α 1 and β 1, respectively. σ V, IPP and B/V correspond to Duan s ML estimations Errors are clustered at the firm level 40 (B/V) IPP σ V ** (-5.62) 0.017** (4.72) Lambda (Mills ratio) ** (-3.18) ** (-5.77) σ V X pre-merger dummy ** (18.01) 0.003** (7.93) Size 0.011** (10.88) ** (-8.89) Observations R * Statistically significant at 5% level ** Statistically significant at 1% level

41 41 Table 11 expands on the experiment reported in Table 7. It introduces the Mills ratio that emerges from using the pooled equation in Table 10. While other coefficients are not much affected, the more likely (i.e., the less surprising) it is for a particular bank to be the acquirer, the less incremental capital discipline it faces and the more safety-net benefits it can extract. We interpret this to mean that investors and creditors recognize that EU banks with an established crossborder acquisition program are adept at creating value through regulatory arbitrage.

42 TABLE 11 DIFFERENCES IN PRE-MERGER RISK-SHIFTING AT ACQUIRING VS. ACQUIRED CBM BANKS Fixed-effects panel regressions relating changes in a bank s leverage, ( B/V), and changes in its fair deposit insurance premium, IPP, to the riskiness of its assets, σ V. B is the face value of bank s debt, including deposits. V is the market value of bank assets. The second and third columns report the value of α 1 and β 1, respectively. σ V, IPP and B/V correspond to Duan s ML estimations Errors are clustered at the firm level 42 (B/V) IPP σ V ** (-5.16) 0.006** (3.06) Lambda (Mills ratio) ** (-5.99) ** (-3.58) σ V X acquiring banks dummy ** (-4.58) 0.009** (2.79) Bank Size 0.009** (24.33) ** (18.45) Observations R * Statistically significant at 5% level ** Statistically significant at 1% level

43 43 Allowing for sample-selection bias, Table 12 investigates whether and how risk-shifting behavior at CBM banks varies before and after a cross-border merger. The coefficient of the Mills ratio is always negative, but becomes much larger and more significant after the transaction than it was before. Unsurprising combinations attract less capital and supervisory discipline than surprising ones. Although, other things equal, post-merger discipline grows with the size of the resulting conglomerate, incremental benefits from expanding asset risk increase as well.

44 TABLE 12 PRE- AND POST-MERGER RISK-SHIFTING AT CROSS-BORDER MERGING BANKS WITH HECKMAN S CORRCTION FOR SELECTION BIAS Second-step panel regressions relating changes in a bank s leverage, ( B/V), and changes in its fair premium, IPP, to the riskiness of its assets, σ V. and to the Lambda parameter (inverse Mills ratio estimated from the selection equation shown at the bottom of the table). B is the face value of bank s debt, including deposits. V is the market value of bank assets. The second and third columns report the value of α 1 and β 1, respectively. σ V, IPP and B/V correspond to Duan s ML estimations 44 Errors are clustered at the firm level Pre-Merger (B/V) IPP σ V ** (-4.88) 0.004** (6.99) Lambda (Mills ratio) * (-2.96) ** (-6.86) Bank Size 0.028** (31.05) ** (10.12) Observations R Post-Merger (B/V) IPP σ V ** (-25.37) 0.013** (20.02) Lambda (Mills ratio) ** ** (-9.15) Bank Size 0.022** (20.03) Observations R (-18.16) ** (12.89) SELECTION EQUATION NOT SHOWN FOR SIMPLICITY

45 45 Table 13 contrasts the behavior of leverage and safetynet benefits at acquirers and targets prior to the crossborder transaction using Heckman s two-equation framework. Other things equal, target-bank access to incremental safety-net benefits is twice that of acquirers. Taken together with our other results, this suggests that CBM acquirers identify targets that possess unexploited opportunities for extracting safetynet benefits.

46 TABLE 13 PRE-MERGER RISK-SHIFTING AT CROSS-BORDER MERGING BANKS WITH HECKMAN S CORRECTION FOR SELECTION BIAS: ACQUIRING VS. TARGET BANKS Second-step panel data estimations relating changes in a bank s leverage, ( B/V), and changes in its fair deposit insurance premium, IPP, to the riskiness of its assets, σ V. and to the Lambda parameter (inverse Mills ratio estimated from the selection equation shown at the bottom of the table). B is the face value of bank s debt, including deposits. V is the market value of bank assets. The 46 second and third columns report the value of α 1 and β 1, respectively. σ V, IPP and B/V correspond to Duan s ML estimations Errors are clustered at the firm level Acquiring bank (B/V) IPP σ V ** (-2.96) 0.018* (1.99) Lambda (Mills ratio) ** (-16.44) ** (-3.74) Bank size 0.012** (16.94) ** (3.85) Observations R Target bank (B/V) IPP σ V ** (-18.46) 0.028** (14.43) Lambda (Mills ratio) ** ** (-7.19) Bank size 0.004** (5.96) Observations R (-18.68) ** (7.33) SELECTION EQUATION NOT SHOWN FOR SIMPLICITY

47 V. Summary Implications 47 This paper confirms two complementary and potentially worrisome hypotheses during our sample period: Regression evidence suggests first that CBM banks were not responding to opportunities for increasing their operating efficiency, at least as measured conventionally by their expense ratios. Instead, statistical analysis indicates that these banks were responding principally to opportunities for shifting risk onto EU safety nets. What makes this form of arbitrage hard to supervise is that safety-net benefits appear to strengthen a country s banks in the short run.

48 48 Policies that do not adequately monitor and discipline merger-created safety-net benefits could end up subsidizing risk taking and increasing the fragility of country s banking system to disruptive movements in the prices of important bank assets. Our findings have implications for the future of macroprudential supervision.