Private and public risk sharing in the euro area

European Economic Review 121 (2020) 103347

Contents lists available at ScienceDirect

European Economic Review journal homepage: www.elsevier.com/locate/euroecorev

Private and public risk sharing in the euro area ✩ Jacopo Cimadomo a,∗, Gabriele Ciminelli b, Oana Furtuna c, Massimo Giuliodori d a

European Central Bank, Fiscal Policies Division, Sonnemannstr. 20, 60314, Frankfurt am Main, Germany Asia School of Business, 2 Jalan Dato Onn, 50480, Kuala Lumpur, Malaysia European Central Bank, Sonnemannstr. 20, 60314, Frankfurt am Main, Germany d University of Amsterdam (Amsterdam School of Economics) and Tinbergen Institute, University of Amsterdam, Roetersstraat 11, 1018 WB Amsterdam, the Netherlands b c

a r t i c l e

i n f o

Article history: Received 14 March 2018 Accepted 31 October 2019 Available online 21 November 2019 JEL codes: C23 E62 G11 G15 Keywords: Risk sharing Time-variation Euro area Financial integration Official financial assistance

a b s t r a c t This paper investigates the contribution of private and public channels for consumption risk sharing in the euro area. In particular, it explores the role of financial integration versus official financial assistance for consumption smoothing. In addition, it presents a timevarying test which allows estimating how risk sharing has evolved since the start of the euro, including the recent great recession and European sovereign debt crisis. Our results suggest that, whereas in the early years of the euro only about a third of country-specific output shocks were smoothed, in the aftermath of the crisis almost 60% of these shocks were absorbed, therefore reducing consumption growth differentials across countries. This improvement was mostly due to a higher degree of financial integration, as reflected in particular in cross-border portfolio holdings of corporate and government bonds. Importantly, the provision of official loans to distressed governments in the wake of the crisis considerably improved risk sharing since 2010. © 2019 Elsevier B.V. All rights reserved.

1. Introduction The architecture and the functioning of the European Economic and Monetary Union (EMU) have been severely challenged in the context of the recent global financial crisis and in particular during the 2010–2012 European sovereign debt crisis. Many commentators have argued that the lack of appropriate risk sharing mechanisms at the euro area level may have contributed to aggravate the severity of the economic downturn in the eurozone periphery and delay the recovery in ✩ We would like to thank Cinzia Alcidi, Roel Beetsma, Cristina Checherita-Westphal, Luca Dedola, Silvia Delrio, Francesco Drudi, Mardi Dungey, Ekkehard Ernst, Davide Furceri, Malin Gardberg, Michele Lenza, Lukasz Goczek, Sebastian Hauptmeier, Fédéric Holm-Hadulla, Christophe Kamps, Hans-Joachim Klöckers, Gerrit Koester, Stefano Maiani, Klaus Masuch, Georg Mueller, Joan Paredes, Beatrice Pierluigi, Javier Pérez, Doris Prammer, Alexander Rathke, Philipp Rother, Martin Schmitz, Frank Smets, Pietro Tommasino, Gianluca Violante and Thomas Warmedinger, for their helpful suggestions and input. We are also grateful for comments from participants at the 21st Annual ICMAIF conference, the Computing in Economics and Finance 2017 conference, the Banca d’Italia “Public Finance” 2018 workshop, the first annual workshop of ESCB Research Cluster 2, the EEA 2019 annual congress in Manchester, the KOF-ETH conference on ‘the euro area at 20’; and from seminar participants at the ECB and the ESM. Oana Furtuna gratefully acknowledges the Fiscal Policies Division of the ECB for its hospitality. The views expressed in this paper are those of the authors and do not necessarily represent those of the ECB and the Eurosystem. ∗ Corresponding author. E-mail addresses: [email protected] (J. Cimadomo), [email protected] (G. Ciminelli), [email protected] (O. Furtuna), [email protected] (M. Giuliodori).

https://doi.org/10.1016/j.euroecorev.2019.103347 0014-2921/© 2019 Elsevier B.V. All rights reserved.

2

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

the aftermath of the crisis (see, e.g., Allard, 2011). Against this background, the Five Presidents Report highlighted that euro area countries have to take steps, both individually and collectively, to compensate for the lack of national adjustment tools that they gave up on entry in the EMU.1 First, when economic shocks occur, each country should be able to respond effectively at the domestic level. Second, countries may also smooth the impact of output shocks through risk sharing within the EMU. Such risk sharing mechanisms would facilitate consumption smoothing, thus decoupling consumption from output growth fluctuations.2 Risk sharing can be achieved through integrated financial and capital markets, which is generally referred to as “private risk sharing”. Private risk sharing operates through two main channels. First, internationally diversified investment portfolios can generate income flows that are unrelated to fluctuations in the domestic economy. This is the case when the return on foreign assets is highly correlated with output growth in the issuer economy and weakly correlated with output growth in the domestic (holder) economy. Second, integrated credit markets could reinforce risk sharing: the supply of credit to the domestic economy is expected to be less affected by domestic shocks when international banks – which are in principle less exposed to country-specific shocks – operate in that economy. At the same time, however, more integration in the banking sector and financial markets may also amplify idiosyncratic shocks if their effects spill over more rapidly in an interconnected economic environment. In addition to private risk sharing channels, public policies at the supra-national level may also contribute to enhance risk sharing (see, e.g., Farhi and Werning, 2017). Traditionally, “public risk sharing” has been thought to work through fiscal transfers: when the domestic economy is affected by an idiosyncratic negative shock, foreign countries can design fiscal transfers towards the domestic economy to smooth the impact of the shock. While a fully-fledged fiscal stabilization mechanism has been discussed, but not yet introduced in the euro area, in recent years the EMU architecture has benefitted from the introduction of several official facilities which have provided financial assistance to governments under stress.3 Although the main motivation of these facilities was to safeguard financial stability, the provision of financial assistance might have also contributed to increase risk sharing, ex post. Indeed, without financial assistance, distressed governments would have likely faced default, implying drastic cuts in public expenditures and increases in taxes. Instead, although accompanied by tough conditionality and adjustment programmes, official financial assistance allowed them to smooth out cuts in public spending and hikes in taxes, thereby helping to sustain private consumption relative to a hypothetical no bailout and sovereign default scenario.4 Thus, our testable hypothesis is that public official assistance may have helped consumption risk sharing in the eurozone, on top of private risk sharing channels. This paper makes three main contributions. First, based on a sample of 11 euro area countries over the period 2001– 2017, we explore the role of financial integration and official financial assistance for consumption risk sharing. Second, we propose a time-varying framework which allows estimating how risk sharing and the relative importance of the individual private and public risk sharing channels have evolved in the eurozone throughout this period. Third, we analyze the degree and evolution of risk sharing between “Core” and “Periphery” euro area countries. To measure financial integration, we use data from the Coordinated Portfolio Investment Survey (CPIS) of the International Monetary Fund (IMF), which records bilateral cross-border holdings of portfolio investment securities, as well as their breakdown into debt and equity assets. Debt securities include both corporate and government bonds issued in a eurozone country and belonging to the portfolio of agents of another eurozone country. This is, to our knowledge, the first use of the CPIS database in a panel framework for the analysis of cross-country risk sharing.5 We augment this information by using the restricted-access version of the dataset of the Bank for International Settlements (BIS) on cross-border bank loans. This dataset reports the outstanding cross-border positions of the banking sector of several eurozone countries against residents of each other eurozone counterparty country.6 As regards official financial assistance, we use data on official loans disbursed to some EMU countries (Greece, Ireland, Portugal and Spain) since 2010 through the Greek Loan Facility, the European Financial Stability Facility (EFSF), the European Financial Stability Mechanism (EFSM) and the European Stability Mechanism (ESM). To measure risk sharing, we focus on the deviation of real per capita household consumption growth with respect to real per capita output growth across EMU countries, as suggested by the reference literature in this field (see Asdrubali et al., 1996; Sørensen and Yosha, 1998). More specifically, we follow Fratzscher and Imbs (2009) and, based on our still largely unexplored dataset of bilateral financial holdings and bilateral official assistance, we estimate bilateral risk sharing spec-

1

See https://ec.europa.eu/priorities/sites/beta- political/files/5- presidents- report_en.pdf. In general, perfect or full income risk sharing, through both private and public channels, characterizes a situation where consumption growth rates are equalized across all countries (Mace, 1991). 3 These official financing facilities are the Greek Loan Facility, the European Financial Stability Facility (EFSF), the European Financial Stability Mechanism (EFSM) and the European Stability Mechanism (ESM), which were progressively activated from 2010 onwards in the euro area. See Section 3 for a detailed description of these facilities. 4 In some cases (e.g. ESM loans in Spain), official financial assistance was also used by governments to recapitalize banks under stress. This might have also indirectly contributed to sustain private consumption through the provision of credit to the private sector. 5 Previous papers have generally focused on specific waves of the CPIS survey, mainly due to limited availability of data at the time of the study (see, e.g., Fratszcher and Imbs, 2009). Our analysis builds on different waves of this dataset, which expands considerably the number of observations used in the analysis. 6 Cyprus began reporting in 2008 and it thus not included in our sample. 2

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

3

ifications which allow us to take full advantage of the time-series and cross-country-pairs information contained in our dataset. There are three main results from our analysis. First, in the early years of the EMU only about a third of country-specific (i.e., idiosyncratic) output shocks were smoothed. However, in the aftermath of the European sovereign debt crisis, this share increased to almost 60%, therefore contributing to decrease consumption growth differentials across countries. Second, the progressive improvement of the shock absorption capacity was due to higher financial integration, but also to the provision of official financial assistance to countries under stress since 2010. Third, among private risk sharing channels, cross-border holdings of (corporate and government) debt were the most effective in smoothing consumption. Instead, our results do not offer strong evidence for a role of cross-border bank loans in shock absorption. We then focus on the links between “Core” and “Periphery” euro area countries and find that not only cross-border holdings of debt, but also of equity, were effective in absorbing output shocks across these two groups of countries. Our results are robust to a number of variants, including one that deals with potential reverse causality between countries’ growth differentials and financial integration, an issue raised by Fratzscher and Imbs (2009) and Kalemli-Ozcan et al. (2013). The remainder of the paper is organized as follows. Section 2 presents a short review of the related literature. Section 3 offers a brief account of the main initiatives put in place to rescue countries under financial stress during the European sovereign debt crisis. Section 4 describes the empirical methodology, while Section 5 illustrates the construction of the dataset used in the analysis. Section 6 presents and comments the results, Section 7 reports a number of robustness checks and Section 8 concludes.

2. Related literature The literature on income and consumption risk sharing has expanded considerably in the last three decades, reflecting stronger interest among academics, commentators, and policy-makers on how countries (or states within a federation) may better insulate themselves against idiosyncratic shocks hitting their economies (see Ioannou and Schäfer, 2017, for a survey). Empirical studies of cross-country consumption risk sharing are motivated by a testable prediction derived in international real business cycle models with complete markets. In a world with a single internationally-traded contingent bond, the Euler equations for the asset holdings indicate that the marginal rates of substitution between current and statecontingent future consumption should be equal across countries at each point in time. Consequently, consumption growth in any given country is only affected by global – uninsurable – shocks. Thus, in an equilibrium characterized by perfect risk sharing, countries should exhibit the same relative growth rate of consumption at each point in time, irrespective of relative output shocks (Mace, 1991; Canova and Ravn, 1996).7 However, the hypothesis of full international risk sharing has been largely rejected in the empirical literature. Indeed, as first highlighted by Backus et al. (1992), cross-country correlations in consumption growth are smaller than correlations in income growth. One possible reason for the lack of full international risk sharing might be related to the presence of outright capital account restrictions or of cross-country differences in regulation and accounting standards, which might generate home bias in asset holdings. Confirming this intuition, Lewis (1996) investigates the role of financial markets and shows that these restrictions partly account for the lack of observed cross-country consumption risk sharing. This indicates that financial liberalization could help boosting shocks absorption. Related to this, Sørensen et al. (2007) show that international risk sharing increased over the period 1993–2003, a time in which home bias had significantly decreased. More recently, a number of other empirical studies have documented that greater financial globalization tends to increase risk sharing, at least among industrial countries. The underlying intuition is that more internationally diversified investment portfolios generate income changes that are unrelated to fluctuations in domestic income, therefore better insulating agents from idiosyncratic shocks that hit their economies (see Kose et al., 2007; Demyanyk et al., 2008; Pierucci and Ventura, 2010, Rangvid et al., 2016). However, there might be caveats regarding the effects of higher financial integration for international risk sharing. In particular, the papers cited above are generally based on periods of financial upturns, while the effects of more financial market integration may be reversed during downturns.8 In addition, if globalization leads to stronger comovements between international stock markets, the benefits of cross-border holdings of financial assets might be limited (see, e.g., Beine et al., 2010). This is sometimes referred to as the “knife-edge” property of financial markets: interconnections work as a shocks absorber (i.e., leading to risk sharing) in certain states of the world, while in others they tend to amplify shocks, i.e., risk-spreading (see Tasca and Battiston, 2014; Balli et al., 2013). Turning to the role of public policies, Canova and Ravn (1996) suggest that setting up public institutions which operate through taxes and transfers, aid or lending agreements, could be another way to achieve higher consumption risk sharing. A positive role for fiscal transfers is indeed backed by several empirical studies. For instance, Mélitz and Zumer (2002) examine risk sharing within the United States, Canada, France and the United Kingdom and find that approximately 20% of regional income shocks are stabilized through the central government budget in these countries. Similarly,

7 Farhi and Werning (2017) show that, even in presence of complete markets, some degree of public intervention allowing insurance against idiosyncratic shocks is welfare improving. In fact, private agents do not fully internalize the benefits from public risk sharing channels when forming their decisions. Therefore, the authors make a strong theoretical case for fiscal insurance as a necessary complement to private risk sharing. 8 See also Albertazzi and Bottero (2014) on the pro-cyclicality of lending from cross-border banks.

4

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

Hepp and Von Hagen (2013) suggest that the fiscal channel absorbed about 10% of regional shocks in Germany in the postreunification period (1995–2006). The literature has also tested the joint role of financial markets and transfer schemes for risk sharing. Asdrubali et al. (1996) propose a framework based on a cross-sectional variance decomposition of shocks to GDP to quantify the amount of interstate risk sharing in the United States over the period 1963–1990. They find that 39% of shocks to gross state product were smoothed by capital markets, 13% were smoothed by the federal government (via taxes, transfers, and grants to states), 23% were smoothed by credit markets, while the remaining 25% were unsmoothed.9 As regards the euro area, Delrio et al. (2018) follow the approach of Asdrubali et al. (1996) and study the role of the current account, and in particular the TARGET balances via the ECB, in influencing risk sharing in the EMU. Their findings highlight a reduction of risk sharing during and after the crisis and point to the current-account channel as the main driver of this reduction (see also ECB, 2017). Milano (2017) revisits the Asdrubali et al. (1996)’s approach to explore the role of European institutions for risk sharing. She finds that shocks absorption in the eurozone increased from 23% in the period 1999–2006 to 31% in the period 2007–2014. The potential role of fiscal transfers for cross-country consumption insurance within the euro area is analyzed by Furceri and Zdzienicka (2015) and Beetsma et al. (2018), among others. On the basis of a counterfactual experiment introducing a fictitious supranational redistribution mechanism, Furceri and Zdzienicka (2015) suggest there could be considerable insurance gains from setting up a fiscal stabilization mechanism. Instead, Beetsma et al. (2018) propose a stabilization scheme which would compensate negative export shocks hitting single sectors. In terms of methodology, our paper connects in particular with Fratzscher and Imbs (2009), which extends the conventional tests of international consumption risk sharing introduced by Lewis (1996), using cross-sectional information on bilateral asset holdings for 23 lending economies and 54 borrowing economies.10 Here, however, we focus exclusively on the eurozone and, while Fratzscher and Imbs (2009) only consider private risk sharing channels, we also analyze the role of public channels. Moreover, we also exploit the time-series information on bilateral holdings of financial assets among euro area countries. Through this framework, we are able to gauge the relative contribution of official lending channels as well as capital and credit markets instruments to the time variation in consumption risk sharing from the early years of the EMU to the aftermath of the European sovereign debt crisis. 3. Official financial assistance during the European debt crisis In this section we briefly review the official financial assistance initiatives set up in the eurozone since 2010. The large fiscal imbalances emerged during the global financial crisis of 20 07–20 09 and the European sovereign debt crisis of 2010– 2012 contributed to erode financial markets’ confidence on the sustainability of public finances in some euro area countries. Tensions in financial markets were so acute that, by early 2010, Greece had de facto lost market access. In order to preserve financial stability, in May 2010 euro area governments agreed to provide loans to Greece on a bilateral basis. At the same time, the European Council further set up several mechanisms to provide official financial assistance in case other countries were to be cut-off from financial markets. Following Greece, also Ireland, Portugal, Cyrpus and Spain received financial assistance. Ireland was bailed-out in 2010, while Portugal in 2011. Spain, instead, received loans to recapitalize its banking sector in 2012 and 2013.11 More specifically, in May 2010, euro area governments launched the Greek Loan Facility to provide financial assistance to Greece. Loans were provided on a bilateral basis, with the amount disbursed by each country being proportional to its share in the equity of the European Central Bank (ECB). A total of €52.9 billion worth of loans was disbursed over a twoyear period. At the same time, EMU countries further established the European Financial Stability Facility (EFSF). This was a temporary special purpose vehicle aimed at providing financial assistance to distressed eurozone member states. The EFSF was backed-up by capital provided by each eurozone member country, with shares based on national GDP and population. Hence, differently from the Greek Loan Facility, the EFSF had its own borrowing capacity. Overall, the EFSF provided loans worth €18.2 billion to Ireland, €26.9 billion to Portugal and €141.8 billion to Greece over the 2011–2014 period. In parallel to the EFSF, the European Council set up the European Financial Stability Mechanism (EFSM). Rather than being restricted to EMU countries, the EFSM was designed to provide assistance to any EU government in financial distress. Like the EFSF, the EFSM had its own borrowing capacity, but differently from it, it was guaranteed by the EU budget. The EFSM provided loans worth €22.5 billion to Ireland and €24.3 billion to Portugal. 9 These results have been challenged by del Negro (2002) who shows that, once measurement error in income and consumption is taken into account, the actual amount of risk sharing across U.S. states may be significantly lower than what suggested by Asdrubali et al. (1996). 10 Fratzcher and Imbs (2009) provide a theoretical model in which agents can hold different classes of foreign assets to achieve risk sharing and choose those with lower ex-ante transaction costs. The nature of official loans we model in our paper differs substantially from other assets like bonds, bank loans, foreign direct investment and equity. When a country is close to default, the priority is to safeguard solvency and financial stability, and considerations of transaction costs are likely to have a negligible role. In such a framework, cross-country risk sharing via official financial assistance should be intended as a consequence of the crisis management policies put in place to deal with the crisis, rather than as the result of deliberate decisions of countries to apply for these loans, as substitute for other type of assets. However, although it was not its ex-ante objective, the provision of financial assistance might have also contributed to enhance risk sharing ex post. 11 Cyprus also received financial assistance in 2012. However, in our empirical analysis, we do not consider this country since data on private financial integration are only available for half of our sample.

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

5

Finally, in September 2012, EMU governments established the European Stability Mechanism (ESM), a special purpose vehicle with permanent nature. The ESM is structured similarly to the EFSF and it effectively took over all of its operations. The ESM has provided loans worth €38.2 billion to Greece, €41.3 billion to Spain and €6.3 billion to Cyprus thus far. In our empirical analysis, we include loans from these four facilities together and label them as ‘official financial assistance’. Tables A1-A3 in Appendix A report country-specific capital shares of these facilities and the amounts of loans disbursed.12 4. Methodology 4.1. Baseline empirical setup Most tests of consumption risk sharing are based on the difference between real per capita consumption growth in country i and real per capita consumption growth observed in the same currency area, federation or in the rest of the world (depending on the relative importance of links between countries in a certain area).13 Such tests are based on the following simple model,

(Ci,t − Ct ) = α + β (Yi,t − Yt ) + εi,t where the log-growth of variable X is denoted as Xi, t , Ci, t denotes real per capita household consumption and Yi, t stands for real per capita output in country i at time t. Ct and Yt denote aggregate consumption and output in a certain reference area (e.g. the EMU). Under the null hypothesis of perfect risk sharing, differences between the country-specific and the aggregate consumption growth (i.e., idiosyncratic consumption growth), should be decoupled from the differences in output growth (i.e., idiosyncratic output growth), thus yielding a β coefficient equal to zero. Under the alternative hypothesis, a β coefficient statistically different from zero indicates imperfect risk sharing, and its magnitude reflects the extent of the deviation from the theoretical benchmark. Fratzscher and Imbs (2009) extend this benchmark setup by exploiting the information which is available in a twodimensional (bilateral) dataset and study the degree of risk sharing between country i and country j (for a given year t). We contribute to their framework by estimating the risk sharing relationship over time. Namely, we test the relationship between consumption and output growth differentials between country i and country j at time t. In this setting, we define as ‘country-specific’ or ‘idiosyncratic’ a shock hitting country i but not country j. It can be shown that this corresponds to estimating the coefficient β in the equation above (see Section 2.2. in Fratzscher and Imbs, 2009). The advantage of this setup is that we can exploit a much larger dataset and therefore increase considerably the efficiency of the estimates. In our three-dimensional panel, the basic risk sharing test then becomes:

(Ci,t − C j,t ) = α + β (Yi,t − Y j,t ) + γ Zi j,t−1 + ηt + μi j + εi j,t

(1)

Eq. (1) includes time-fixed effects ηt to control for unobserved common euro area-wide factors (e.g., the sovereign debt crisis), country-pair fixed effects μij to account for unobserved time-invariant characteristics (such as distance and similarity in language), and a set of control variables Zij, t that vary across pairs (ij) and over time (t). In particular, following Epstein et al. (2016), the Z matrix of controls includes country-pair differentials of the growth rate of the value added tax rate, computed as VATi, t -VATj, t , and of the tax rate on dividend income, PITi, t -PITj, t , where the  operator again denotes log-growth.14 We favor the use of statutory tax rates as opposed to measures of effective taxation derived from national accounts in order to alleviate the concerns about endogeneity of tax revenues to the dynamics of consumption and income. In addition to tax rate growth differentials, Z also includes the inflation differential, πi,t − π j,t , the 10-year sovereign bond real yield differential, Y iel di,t − Y iel d j,t , and the real domestic credit growth differential, C rediti,t − C redit j,t , where credit is defined as the real total credit by domestic banks to the private non-financial sector (see Appendix A for description of data and sources). The inclusion of the inflation differentials is theoretically justified by the link between the relative growth rates of consumption and the dynamics of the real exchange rate (Backus and Smith, 1993; Galí and Monacelli, 2005). As EMU countries are characterized by invariant nominal exchange rates vis-à-vis other eurozone countries, we account for real exchange rate differentials by including the relative dynamics of prices across countries. From a theoretical perspective, in a New Keynesian framework, cross-country inflation differentials impact relative consumption growth rates (see, for example, Galí and Monacelli, 2005). We also include differentials in 10-year sovereign bond yields to capture sovereign default risk. Indeed, based on a New Keynesian model featuring a sovereign risk channel, Corsetti et al. (2013) show that a larger default risk premium translates into higher relative borrowing costs, thus exerting downward pressure on the relative growth rate of 12 Loans to countries under financial stress were also provided by the IMF. These loans are not included in our analysis, as they do not influence risk sharing among European countries. 13 See, e.g., Kose et al. (2007). 14 Epstein et al. (2016) account for the risk sharing wedge generated by international differences in taxation. The authors augment a business cycle model with distortionary taxes and find that an increase in the relative consumption tax or capital income tax growth leads to lower relative consumption growth. They find that across country pairs, accounting for the distortionary effect of the capital tax wedge on the relative consumption growth rates contributes to revealing a positive link between insurance and financial integration.

6

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

consumption. Finally, we also control for credit by domestic banks, given that this is a main source of financing for the domestic private sector, and therefore can substantially affect private consumption growth. Controlling for domestic credit availability is important for the period we analyze, given the documented increase of home bias in bank lending during the sovereign debt crisis (Saka, 2017 and Ongena et al., 2016) To mitigate potential concerns about reverse causality in annual data, we use lagged values of all the covariates.15 4.2. Extended model including integration measures Following Fratzscher and Imbs (2009), we use an interaction terms model to explore the role of financial integration as a source of time-varying heterogeneity in risk sharing within the EMU. We further extend the analysis by also accounting for the possibility of a public risk sharing channel operating through official financial assistance among euro area countries. Given our focus on the time variation in risk sharing, we enrich the specification by allowing the interacting variables to vary not only across country pairs, but also over time. Allowing for the risk sharing coefficient to be a linear function of the financial and fiscal integration measures, the full model takes the following form:



    Ci,t − C j,t = Yi,t − Y j,t β0 + β1 F Ai j,t−1 + β2 LOANi j,t−1 + β3 EQUIT Yi j,t−1 + β4 DEBTi j,t−1

+ δ1 F Ai j,t−1 + δ2 LOANi j,t−1 + δ3 EQUIT Yi j,t−1 + δ4 DEBTi j,t−1 + γ Zi j,t−1 + ηt + μi j + εi j,t

(2)

where FAij, t is our integration measure based on bilateral official financial assistance, LOANij, t is the integration variable based on cross-border bank loans, DEBTij, t and EQUITYij, t are the integration measures based on, respectively, cross-border holdings of corporate and government debt and cross-border holdings of equity. Their sum is labelled as cross-border portfolio holdings (PORTij, t ). The construction of these integration measures is described in Section 5.2. Given this setup, the full extent of “shock absorption” between country i and j will be equal to 1 minus the sum of the income growth differential coefficient (β 0 ) and the components that capture how risk sharing is related to the four integration measures. Formally, we define shock absorption as:



γt = 1 − β0 +

K 



β

k k INTi j,t−1

(3)

k=1

This measure is time-varying to the extent that the underlying K measures of integration (INTij, t ) change over time. The null hypothesis of perfect shock absorption amounts to testing whether the γ t coefficient in Eq. (3) is not statistically different from one. For positive values of β 0 , negative (positive) coefficients of β k indicate that integration improves (worsens) cross-country consumption risk sharing. A γ t equal to 1 indicates full shock absorption (i.e., full risk sharing), whereas γ t equal to 0 indicates no shock absorption (i.e., no risk sharing). 5. Data Our sample includes 11 eurozone countries: Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Portugal and Spain.16 The dataset is collected at the yearly frequency and covers the period 1999–2017. Due to some missing data for the early years of the sample, we restrict the analysis to the 2001–2017 period. 5.1. Data sources To assemble the dataset we combine information from multiple sources. As a first building block we use bilateral data on portfolio investment from the IMF’s Coordinated Portfolio Investment Survey (CPIS). The CPIS provides data on crossborder holdings of equity and debt securities, collected from holders by means of a survey and classified according to the residence of the issuer. We combine the CPIS data with the BIS International Locational Banking Statistics dataset (restricted access version), which reports bilateral positions of the banking sector in country i against each counterparty country j. The data is recorded using the residence principle. To minimize the overlap with portfolio investment data, we restrict our attention to cross-border loans provided by a creditor banking system to the economy of a given debtor country. To measure integration through official financial assistance, we use information contained in the websites of the ESM and the European Commission, on the magnitude and timing of official loans disbursements, as well as on the capital shares of the EFSF and the ESM. In addition to financial and official assistance data, we source real final household consumption expenditures, real GDP and nominal GDP from Eurostat and the IMF’s World Economic Outlook (WEO). We deflate real consumption and real output by total population to express them in real per capita terms. Regarding the set of controls, we source the reduced VAT rates from the European Commission and the overall (corporate plus personal) statutory tax rate on distributed profit from 15

Our estimates are robust to the use of contemporaneous values of the covariates. Results using contemporaneous values are available upon request. Although our sample initially comprised the 12 eurozone countries for which cross-border bank loans from the Bank for International Settlements (BIS) are available, we exclude Luxembourg given its status as financial hub and the observed cross-border exposures which indicate that this country is a clear outlier. 16

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

7

the OECD Tax Database. Both series enter the specification as country-differences in growth rates.17 Data on the 10-year sovereign bond yield and the inflation rate is collected from the IMF’s WEO. Finally, data on credit supplied by domestic banks to the private non-financial sector comes from the BIS total credit statistics. Tables A4 and A5 in Appendix A contain the data sources, descriptions and summary statistics of all variables used in the analysis.

5.2. Construction of the integration measures The financial integration measures are computed following (among others) Epstein et al. (2016), as the sum of claims of country i over country j and claims of country j over country i, scaled by the sum of the nominal GDP of country i and of country j:18

IN Ti j,t =

Ai→ j,t + A j→i,t Yi,t + Y j,t

(4)

First, we compute a measure of overall financial integration, FINij, t , where Ai → j, t is the sum of cross-border bilateral bank loans (LOANij, t ) and cross-border portfolio investment (PORTij, t ). Then, we study these two variables, LOANij, t and PORTij, t , separately. Lastly, we further differentiate between debt and equity within the category of portfolio investment, and compute measures of integration for the corresponding assets, labelled respectively as DEBTij, t and EQUITYij, t . We further use Eq. (4) to compute a measure of bilateral official financial assistance, which we label FAij, t . In this case, Ai → j, t represents the financial assistance provided by country i to country j at a given point in time and channeled through, either, the Greek Loan Facility, the EFSF, the EFSM, or the ESM. To obtain Ai → j, t for EFSF and ESM loans, we multiply the capital share of each country i with the loan disbursed to country j. As an example, the ESM had provided loans worth €41.4 billion to Spain in 2013. Since Germany’s capital share in the ESM is 27.1% and the sum of Germany’s and Spain’s GDP was €3851.9 billion in 2013, our financial assistance variable takes value equal to 0.29% in 2013 for the Spain-Germany country pair. For the loans disbursed through the EFSM we follow a similar approach. Recall that the EFSM was a financing facility guaranteed by the budget of the European Commission. Hence, we first derive the relative contribution of each country i in the overall Commission’s budget. To do so, we divide the contribution of country i by the Commission’s total revenues. For each country i, we then multiply its relative contribution with the loan disbursed by the EFSM to country j. Finally, we divide the resulting amount by the sum of country i and country j GDP to obtain a measure of EFSM-integration between these two countries. As for the assistance provided through the Greek Loan Facility we simply take the value of the bilateral loans disbursed from the other euro area countries to Greece through this facility. We then add up the measures derived for financial assistance through the EFSF, EFSM, ESM and the Greek Loan Facility to construct our FAij, t variable. Consistently with the financial integration measures, this variable is based on stocks of outstanding loans. Fig. 1 shows the evolution of our integration measures, averaged across all country pairs, for the 2001–2017 period. Cross-border bilateral debt holdings constituted by far the largest component, peaking at around 3.6% of the euro area GDP in 2009. They declined during the 2010–2012 sovereign debt crisis and remained stagnant thereafter. Cross-border bilateral bank loans were substantially larger than equity holdings for all of the sample, except the last three years. Both these variables displayed a similar upward trend up to the global financial crisis of 20 07–20 09, when they experienced a decline. While cross-border bank loans kept falling even after the crisis, signaling fragmentation in the eurozone credit market, bilateral equity holdings recovered swiftly and increased for the rest of the sample, up to about 1.5% of euro area GDP in 2017. The financial assistance variable was, by construction, zero up to 2009 given that official loans were activated only in 2010. As of 2010, it started increasing, although it remained smaller than the other integration measures. Our approach also allows to compute integration measures at the country level, i.e., capturing the interlinkages of a single country vis-à-vis all other countries in the sample. Fig. A1 in Appendix A shows these country-specific integration measures, where country i is fixed, and the integration measure is constructed by averaging the bilateral integration measure across the ten remaining j countries. All countries display a higher level of integration through portfolio debt instruments than through equity or bank loans.19 Ireland, the Netherlands and France exhibit the highest levels of financial integration, while Finland, Greece and Portugal are the least financially integrated. As expected, the fall in integration through debt holdings around the sovereign debt crisis is particularly marked for Ireland, Spain, Greece, Portugal (and to a lesser extent Italy and France). Again not surprisingly, the official financial assistance variable reaches the highest values for the countries receiving assistance, and it is particularly high in the case of Greece.

17

The results are robust to using the level rather than growth rates of the tax rates and to using the standard rather than the reduced VAT rate. Consistently with the related empirical literature (e.g., Lane and Milesi-Ferretti (2007), Frazscher and Imbs (2009), Kalemli-Ozcan, et al. (2013) and Epstein et al. (2016)), we view “integration” as a concept that should be reflected in a variable taking a positive value when two countries are integrated and a value equal to 0 when two countries are fully disconnected. As a result, an integration measure cannot be negative. Reflecting this idea, the financial integration variables are based on stocks of the underlying financial variables and loans. Note that these measures can be declining. For the FA measure, this occurs when a certain loan is repaid (e.g., Spain started to repay its ESM loans as of 2014). This is also reflected in Figs. 1 and A1, which show slightly declining values for FA at the end of the sample (indeed due to loan repayment). 19 Only for Belgium, and in a few years, the cross-border bank integration measure is higher than the one based on cross-border debt holdings. 18

8

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

Fig. 1. Financial integration and official financial assistance in the euro area. Notes: Annual country-pair averages in percent of GDP. “Official financial assistance”, “Bank loans”, “Portfolio equity” and “Portfolio debt” are defined as the sum of the relevant bilateral exposure of country i in country j and the bilateral exposure of country j in country i over the sum of the GDP of countries i and j. “Official assistance” refers to the official financial assistance through bilateral loans (Greek Loan Facility), as well as the EFSF, EFSM and ESM. “Bank loans” refers to data on cross-border bank lending from the Bank of International Settlements. “Portfolio equity” and “Portfolio debt” mark the corresponding components of the IMF Coordinated Portfolio Investment Survey.

Table 1 Simple risk sharing regression model. (1)

Yi,t −  Yj,t

(2) ∗∗∗

0.654 (0.115)

V ATi,t−1 −  V ATj,t−1 P ITi,t−1 −  P ITj,t−1

(3) ∗∗∗

0.565 (0.088) −0.116∗ ∗ ∗ (0.029) −0.041∗ ∗ (0.014)

 C rediti,t−1 −  C redit j,t−1

0.656 (0.095)

(5) ∗∗∗

−0.139∗ ∗ ∗ (0.038) −0.041 (0.069)

0.558∗ ∗ ∗ (0.068) −0.094∗ ∗ (0.030) −0.041∗ ∗ (0.015) 0.058∗ ∗ ∗ (0.011) −0.048 (0.028) −0.073 (0.059)

0.614

0.679

0.618 (0.108)

0.065∗ ∗ ∗ (0.011)

Y iel di,t−1 − Y iel d j,t−1

πi,t−1 − π j,t−1 R-squared

(4) ∗∗∗

0.597

0.644

0.638

Notes: OLS estimation results from Eq. (1). The sample comprises 816 observations, 55 country pairs and 11 countries. All specifications include time- fixed effects and country-pair fixed effects. Standard errors, clustered at the country i and at the country j levels, are in parenthesis. ∗ ∗ ∗ , ∗ ∗ and ∗ indicate statistical significance at the 99%, 95% and 90% confidence level respectively.

6. Empirical results 6.1. Simple risk sharing regression In Table 1 we report the results of the simple bilateral risk sharing regression as in Eq. (1), linking consumption growth differentials to output growth differentials and a set of controls. In this first regression, we do not include any measure

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

9

of financial integration and official assistance. Estimation is based on OLS with two-way clustered standard errors (at the country i and country j levels), as in Fratzscher and Imbs (2009).20 A coefficient on the output differential term (Yi,t − Y j,t ) equal to zero would signal perfect risk sharing, given that output growth differentials would not be reflected in consumption growth differentials. A coefficient equal to one would indicate no risk sharing. Column (1) shows the results of the simple consumption-output regression with no controls added. Hence, this parsimonious specification only includes the (Yi,t − Y j,t ) term as well as country-pair and year fixed effects. The other columns show the results when we include our controls, namely V ATi,t−1 − V AT j,t−1 and P ITi,t−1 − P IT j,t−1 (Column 2), the domestic credit growth differential C rediti,t−1 − C redit j,t−1 (Column 3), the 10-year real sovereign bond yield Y iel di,t−1 − Y iel d j,t−1 and inflation differentials πi,t−1 − π j,t−1 (Column 4). Finally, Column (5) reports the regression results with all controls included at the same time. The coefficient on the output differential is rather stable across all specifications and lies in the 0.56–0.66 interval. This indicates that, on average over the full sample, only about 40% of idiosyncratic output shocks were smoothed in the eurozone, while the remaining 60% was unsmoothed. This risk sharing coefficient appears to be slightly larger than other estimates in the literature. For example, using the Asdrubali et al. (1996) approach in a sample of eurozone countries, Alcidi et al. (2017) obtain risk sharing estimates of 42% between 1998 and 2007, 55% between 2008 and 2009 and 16% between 2010–2013. With the same methodology, Milano (2017) places the estimate around 23% for the period 1999–2006 and 31% for the period 2007–2014. As expected, the VAT and PIT growth rate differentials enter with a negative coefficient, suggesting that an increase in these tax rates in country i leads to a decrease in consumption in that country relative to country j. As regards the other controls, the coefficient on the bond yield differentials is negative, indicating a depressive effect of a higher risk premium on consumption. If the sovereign risk premium (as captured by the yield differential) increases in country i relative to country j, then in country i the cost of borrowing will increase compared to country j. Ceteris paribus, this will exert downward pressure on the growth of consumption in country i relative to j. Domestic credit growth differentials are positively associated with diverging relative consumption growth rates. In other words, an increase in credit growth in country i relative to country j is associated with an increase in consumption of country i relative to country j. The variable capturing inflation differentials is not statistically significant. Finally, once we include all controls at once, the bond yield differential loses statistical significance. In the remainder of the paper, we use the most general specification including all controls (Column 5) as the reference model. 6.2. The effects of financial integration and official financial assistance on risk sharing Table 2 reports the results from richer specifications based on Eq. (2). Specifically, we interact the output growth differential with (i) the official financial assistance variable and (ii) the terms representing private financial integration (i.e., bilateral cross-border bank loans and portfolio holdings). The underlying intuition behind this interacted variable regression is that idiosyncratic output shocks may affect consumption depending on the level of the integration variables. For example, output shocks may have smaller (bigger) effects on consumption if financial integration is higher (weaker). In Column (1) we augment the baseline specification with our international financial assistance variable (F Ai j,t−1 ), entering both as an interaction with the growth differential, Yi,t − Y j,t , and in level. The coefficient estimated for the interaction term is negative and highly significant.21 This indicates that since 2010, when it was first activated, official financial assistance has contributed in a statistically significant way to enhance risk sharing in the eurozone. In Column (2), we include the overall private financial integration measure F INi j,t−1 . The regression results suggest a statistically significant effect of financial integration in sharing income shock risks across countries. The finding that both private financial integration and official financial assistance contribute to risk sharing survives when F Ai j,t−1 and F INi j,t−1 enter the regression jointly (Column 3). We then differentiate between credit and capital markets integration by including the LOANi j,t−1 and P ORTi j,t−1 separately (Column 4). Finally, we also distinguish among equity and bond integration by including both the EQUIT Yi j,t−1 and DEBTi j,t−1 variables (Column 5). The interaction term for portfolio holdings has the expected negative sign, which indicates that capital market integration brings about more risk sharing. As shown in Column (5), this result is entirely driven by debt holdings. We also find that bilateral bank loans (credit market integration) do not have any effect on consumption risk sharing, as reflected by the almost null and statistically insignificant interaction coefficient in Columns (4) and (5). The coefficients of the control variables are broadly in line with those of Table 1. PIT and VAT growth rate differentials are always negative and significant. The contribution of bond yield differentials is also negative and significant. At the same time, the differentials in domestic credit growth rates still enter the regression with a positive sign. Next, in Fig. 2 we show the time-varying measure of the degree of shock absorption in the EMU, defined as γ t (see Eq. (3)). To derive it, we use the βˆ0 − βˆ4 coefficients estimated in Column (5) of Table 2 and the annual country-pair averages of their respective financial integration and official assistance measures. A value of 1 corresponds to perfect shock absorption (full-risk sharing) while a value of 0 indicates no shock absorption. The figure indicates that the degree of shock absorption 20

In Appendix B1 we show that the results are robust to different assumptions regarding the structure of the standard errors. The level of the private financial integration and official financial assistance measures are always statistically insignificant across the different specifications. As a result, here and in the rest of the paper we do not report them, although we always include them as controls. 21

10

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347 Table 2 The role of private financial integration and official financial assistance on risk sharing. (1)

Yi,t −  Yj,t (Yi,t −  Yj,t )F Ai j,t−1

(2) ∗∗∗

0.582 (0.056) −0.487∗ ∗ (0.189)

(Yi,t −  Yj,t )F INi j,t−1

(3) ∗∗∗

0.710 (0.060)

−0.029∗ ∗ ∗ (0.007)

(4) ∗∗∗

0.724 (0.054) −0.396∗ ∗ (0.160) −0.028∗ ∗ ∗ (0.007)

(Yi,t −  Yj,t )LOANi j,t−1 (Yi,t −  Yj,t )PORTi j,t−1

(5) ∗∗∗

0.733 (0.057) −0.360∗ ∗ (0.134)

0.769∗ ∗ ∗ (0.049) −0.392∗ ∗ ∗ (0.115)

−0.008 (0.012) −0.038∗ (0.017)

0.004 (0.006)

(Yi,t −  Yj,t )DEBTi j,t−1 (Yi,t −  Yj,t )EQUIT Yi j,t−1 V ATi,t−1 −  V ATj,t−1 P ITi,t−1 −  P ITj,t−1  C rediti,t−1 −  C redit j,t−1 Y iel di,t−1 − Y iel d j,t−1

πi,t−1 − π j,t−1 R-squared

−0.100∗ ∗ (0.032) −0.042∗ ∗ (0.016) 0.056∗ ∗ ∗ (0.011) −0.094∗ ∗ (0.034) −0.097 (0.058) 0.684

−0.085∗ ∗ (0.028) −0.035∗ ∗ (0.012) 0.050∗ ∗ ∗ (0.011) −0.031 (0.027) −0.057 (0.056) 0.698

−0.090∗ ∗ (0.029) −0.037∗ ∗ (0.013) 0.049∗ ∗ ∗ (0.011) −0.069∗ ∗ (0.029) −0.078 (0.058) 0.701

−0.088∗ ∗ (0.029) −0.036∗ ∗ (0.013) 0.048∗ ∗ ∗ (0.011) −0.064∗ ∗ (0.024) −0.085 (0.057) 0.703

−0.070∗ ∗ ∗ (0.007) −0.005 (0.032) −0.086∗ ∗ (0.028) −0.037∗ ∗ (0.013) 0.049∗ ∗ ∗ (0.012) −0.061∗ ∗ (0.025) −0.083 (0.061) 0.705

Notes: OLS estimation results from Eq. (3). The sample comprises 816 observations, 55 country pairs and 11 countries. All specifications include time fixed effects and country-pair fixed effects. Standard errors, clustered at the country i and at the country j levels, are in parenthesis. Note that interacted variables are always included as direct controls in all regressions, but coefficients are not reported in the table as they are always statistically insignificant. ∗ ∗ ∗ , ∗ ∗ and ∗ indicate statistical significance at the 99%, 95% and 90% confidence level respectively.

Fig. 2. Degree of shock absorption and contribution of financial integration and financial assistance in the EMU. Notes: The figure plots the degree of shock absorption (solid line) defined as γ t from Eq. (4) and relative 90% confidence bands (dotted lines) as well as the contribution of each risk sharing channels considered (vertical bars), based on the estimates in Column (5) of Table 2. The difference between the bars and the solid line reflects the coefficient [beta_0] of Eq. (4), therefore the component which cannot be explained on the basis of the integration channel considered here. A value of one corresponds to full-risk sharing (full shock absorption of idiosyncratic output shocks), a value of zero indicates no shock absorption. The interaction terms are evaluated at their annual country-pair averages (see Fig. 1). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

11

Table 3 Risk sharing between “Core” and “Periphery”.

Yi,t −  Yj,t (Yi,t −  Yj,t )F Ai j,t−1 (Yi,t −  Yj,t )LOANi j,t−1 (Yi,t −  Yj,t )DEBTi j,t−1 (Yi,t −  Yj,t )EQUIT Yi j,t−1 V ATi,t−1 −  V ATj,t−1 P ITi,t−1 −  P ITj,t−1  C rediti,t−1 −  C redit j,t−1 Y iel di,t−1 − Y iel d j,t−1

πi,t−1 − π j,t−1 Observations Unique country-pairs R-squared

Full sample (1)

Core-Periphery sample (2)

0.769∗ ∗ ∗ (0.049) −0.392∗ ∗ ∗ (0.115) 0.004 (0.006) −0.070∗ ∗ ∗ (0.007) −0.005 (0.032) −0.086∗ ∗ (0.028) −0.037∗ ∗ (0.013) 0.049∗ ∗ ∗ (0.012) −0.061∗ ∗ (0.025) −0.083 (0.061) 816 55 0.705

0.816∗ ∗ ∗ (0.056) −0.535∗ ∗ ∗ (0.089) 0.009 (0.006) −0.065∗ ∗ ∗ (0.015) −0.053∗ ∗ (0.018) −0.081∗ ∗ (0.029) −0.036∗ (0.014) 0.051∗ ∗ (0.014) −0.074∗ ∗ (0.023) −0.094 (0.084) 450 30 0.790

Notes: OLS estimation of Eq. (3). The sample includes 11 countries. Standard errors, clustered at the country i and at the country j levels, are in parenthesis. ∗ ∗ ∗ , ∗ ∗ and ∗ indicate statistical significance at the 99%, 95% and 90% confidence level respectively. “All sample” refers to all pairs across EMU countries and “Core-Periphery” refers to all unique country pairs consisting of “Core” and “Periphery” countries. “Periphery” refers to vulnerable countries in the euro area (Greece, Spain, Italy, Portugal and Ireland) and “Core” refers to resilient countries in the euro area (Germany, the Netherlands, Belgium, Austria and Finland. All regressions include country-pair and year fixed effects. Note that interacted variables are always included as direct controls in all regressions, but coefficients are not reported in the table as they are always statistically insignificant.

increased from around 33% (of country-specific output shocks) in the early 20 0 0s, to around 56% the end of the sample. This reveals that risk sharing has progressively improved in the EMU, and also during the recent crisis period. Based on Eq. (3), we are able to derive the contribution of each of the risk sharing channels considered to the overall risk sharing dynamics. Such contributions are reported as bars in Fig. 2. Portfolio debt integration had been increasingly important to absorb shocks in the early period of the EMU, with its relative contribution to overall risk sharing growing from less than a third to about half during the 20 02–20 08 period. Although in absolute terms the amount of risk sharing achieved through portfolio debt integration was broadly stable around 20% since 2008, its relative contribution gradually declined during the 2009–2017 period, to reach about a third in 2017. On the other hand, financial assistance channeled through the facilities set up during the debt crisis had been an increasingly important shock-absorber mechanism since 2011, both in absolute and relative terms. The amount of risk sharing achieved through official financial assistance reached about 14% in 2017 and contributed to around one fourth to overall risk sharing. Finally, the contribution of cross-border loans and portfolio equity holdings was negligible over the entire sample. 6.3. Risk sharing links between “Core” and “Periphery” In this section, we zoom into the risk sharing links between “Periphery” and “Core” countries within the EMU. The first group includes those eurozone countries that have been often defined as “vulnerable”, i.e., the ones most hit by the recent crisis: Greece, Portugal, Ireland, Spain and Italy. The second group includes Austria, Belgium, Germany, Finland, France and the Netherlands. We present the results in Table 3. For convenience, Column (1) reports the baseline results obtained estimating Eq. (3) on the full sample. Column (2) presents the results for a panel model in which country pairs consist of one core and one periphery country. The coefficient of the financial assistance interaction term (FA) remains negative and significant, and it is larger in absolute value relative to the full sample baseline specification. The coefficient on the bank loans and portfolio debt integration are very similar to the baseline, while that for portfolio equity holdings is negative and statistically significant, thus suggesting that cross-border holdings of equity may have led to stronger shock absorption in the Core-Periphery sample than in the full sample, which includes country pairs of more similar countries (i.e., Core-Core and Periphery-Periphery). In the same vein of Fig. 2, Fig. 3 below reports the full extent of risk sharing (γ t ) and the importance of its different channels in the Core-Periphery sample. For comparison purposes, we also report the extent of risk sharing for the full sample (dotted black line). Risk sharing in the Core-Periphery sample doubled over the period considered, from about 30%

12

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

Fig. 3. Evolution of risk sharing between “Core” and “Periphery”. Notes: See Notes of Fig. 2. We evaluate the non-linear effect in each subsample by using the relevant column of Table 3. The integration terms are evaluated at their bilateral annual averages in the corresponding sub-sample of country pairs.

in 2002 to about 60% in 2017. We notice that while at the beginning of the sample risk sharing was somewhat higher in the full sample, the opposite was true at the end of the sample, suggesting that risk sharing might have increased more across less homogenous countries. The point estimate is however always within the error bands of the full sample point estimate. Hence, we cannot derive a strong conclusion from this analysis. As expected, the larger increase in risk sharing in the Core-Periphery sample was achieved mostly through the activation of official financial assistance. Our estimates suggest that the relative contribution of this channel amounted to about one third in 2017. Differently from the full sample, bilateral portfolio equity holdings also positively contributed to risk sharing, although their relative importance was still marginal.

7. Robustness In this section, we assess the sensitivity of our results to different specifications and estimation assumptions. So far, we have used two-way clustered standard errors (at the country i and the country j levels). In Table B1 of Appendix B we show that the coefficients we estimate are statistically significant also using different standard error corrections. For sake of comparison, in Column (1) we report our baseline results (Column 5 of Table 2). We then present results obtained using OLS with country-pair clustering (Column 2), OLS with the dyadic-clustering standard error correction proposed by Cameron and Miller (2014) (Column 3), OLS with Driscoll-Kraay standard errors (Column 4) and feasible GLS with panelspecific AR(1) autocorrelation in the error term (Column 5). The Driscoll and Kraay (1998) standard error correction accounts for general forms of both cross-sectional and time correlation, whereas the feasible GLS estimation allows for panel-specific autocorrelation of order one in the errors and the dyadic correction proposed by Cameron and Miller (2014) models the possibility that standard errors of all country-pairs that have one country in common might be cross-sectionally correlated. The results are consistent with the baseline estimates. In Table B2 we assess the sensitivity of our baseline results to different estimation specifications. To account for intertemporal smoothing via domestic saving, we augment the set of controls with the lagged differential in savings growth (Column 2). Indeed, private agents may decide to smooth their consumption via their domestic savings (intertemporal channel), on top or as substitute of international channels. More specifically, in case of an aggregate shock, which hits all countries equally, only the intertemporal channel would operate. In case of an asymmetric shock, both the intertemporal and the intratemporal (international) channel are at play. Adding domestic savings as an additional independent variable in our regression equation allows capturing the smoothing effect of the intertemporal channel (following both aggregate and country-specific shocks) and estimating the effects of the international channel, which is the focus of our paper. Table B2 shows that the coefficient associated with the lagged differential saving growth rates is positive and statistically significant. This is consis-

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

13

tent with the idea that saving growth differentials lead to consumption growth differentials across countries.22 This also reflects a consumption-smoothing behavior: ceteris paribus, if agents save in a given period t, they will use these savings to increase consumption in the next period t + +1 (and vice versa). Importantly, controlling for domestic saving does not affect the coefficients on the other variables. We also run additional robustness checks. First, since official financial assistance was activated only in 2010, we check that our results are also valid in the restricted sample in which the official assistance variable takes non-zero values (Column 3). Second, there might a possibility that financial holdings (therefore financial integration) are influenced by growth differentials. Namely, it could be the case that countries exhibiting different growth patterns have incentives to promote financial investments between them (see, for instance, Fratzscher and Imbs, 2009, and Kalemli-Ozcan et al., 2013). In our setup this problem is mitigated by the fact that we used lagged values of financial integration. Yet, we check if our results are robust to using more exogenous measures of financial integration. To derive those, we take the fitted values from four regressions having, as dependent variables, in turn, each of our financial integration variables and, as independent variables, one lag of the dependent variable and the lag of a de-jure measure of bilateral financial integration, derived following the approach of Kalemli-Ozcan et al. (2013). The latter, captures the degree to which two countries have transposed into national law the package of EU directives belonging to the Financial Services Action Plan, aimed at unifying credit and capital markets in the EU. Since the timing of the transposition into national law of EU directives varies on a country basis due to domestic reasons and it is unlikely to have been influenced by growth differentials, we can reasonably consider it as an exogenous instrument for financial integration. The results from this variant are reported in Column (4) of Table B2 and show that our baseline findings are robust. In Column (5) we implement this IV-approach also to the Core-Periphery sample. Results are again very similar. Finally, we check whether our results are robust to the exclusion of single countries. In Fig. B1 of Appendix B we show the overall shock absorption indicator (γ t ) when we exclude from the estimation one country at the time, as well as when we use the full sample. All lines fall within the confidence bands of the full sample, thus reassuring us that the results are not driven by a single country. Admittedly, however, two lines fall very close to the confidence bands of the full sample estimation. These are the ones obtained when excluding Greece (upper dashed line) and Ireland (lower dotted line). We can therefore argue that the presence of Greece in the euro area resulted in a slightly lower level of consumption insurance, while the presence of Ireland induced a higher one after 2011. This may be related to the different levels of financial integration observed in Greece and Ireland throughout the period, with Greece being less integrated than Ireland and most other countries in the sample.

8. Conclusions Many commentators have argued that the effects of both the financial and the European sovereign debt crisis have been aggravated by the absence of appropriate risk sharing mechanisms within the EMU. In this paper, we propose a novel approach aimed at gauging the extent of consumption risk sharing, and its main drivers, among euro area member countries. In particular, based on a sample of 11 eurozone countries for the period 2001–2017, we explore the role of private channels (i.e., cross-border loans and holdings of financial assets) versus public channels (i.e., official financial assistance to distressed euro area countries) for consumption risk sharing. Our results suggest that the shock-absorption capacity generated by international (private and public) channels has increased since the start of the euro area: in the early years of the EMU only a third of idiosyncratic output shocks were smoothed, while in the aftermath of the eurozone’s sovereign debt crisis about 60% of idiosyncratic output shocks were absorbed. Both financial integration and international official assistance play an important role in explaining this improvement. Despite the tough adjustments required in the context of financial assistance programmes, official loans likely helped countries under stress to sustain a certain level of private consumption, which would have otherwise decreased more severely in case of a disorderly sovereign default. As regards private financial integration, we show that cross-border holdings of (corporate and sovereign) debt are a powerful channel in insulating households against country-specific shocks, while cross-border holdings of equity seem to be less effective. In addition, when we focus on the links between “Core” and “Periphery” EMU countries, cross-border holdings of equity also turn out to be powerful instruments in better smoothing consumption across countries. Our results also show that banking integration (via cross-border loans) tended to be ineffective in improving risk sharing. Looking ahead, the introduction of a well-functioning Banking Union for the EMU may reinforce risk sharing via this channel, as banks operating transnationally are likely to sustain credit in the countries affected by negative idiosyncratic shocks. All in all, the finding that risk sharing has improved over time in the eurozone, also during the recent crisis, is to some extent surprising. Yet, this result does not imply that the severity of the crisis would have not been attenuated even further by a fully-fledged centralized fiscal capacity at the eurozone level, providing transfers to countries more severely hit by the recession.

22 We also added the saving growth rate differential interacted with the real GDP growth rate differential. The related coefficient is close to zero and statistically insignificant. The other coefficients in our regression are hardly affected. Results are available upon request.

14

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

Appendix A: Data Table A1 Cumulative loan disbursements to Greece under the Greek Loan Facility. 2010

2011

2012

2013

2014

2015

2016

2017

Belgium Germany Estonia Ireland Spain France Italy Cyprus Latvia Lithuania Luxembourg Malta Netherlands Austria Portugal Slovenia Slovakia Finland

0.8 5.9 0.0 0.3 2.6 4.4 3.9 0.0 0.0 0.0 0.1 0.0 1.2 0.6 0.5 0.1 0.0 0.4

2.0 15.1 0.0 0.3 6.6 11.3 10.0 0.1 0.0 0.0 0.2 0.0 3.1 1.5 1.1 0.3 0.0 1.0

2.0 15.1 0.0 0.3 6.6 11.3 10.0 0.1 0.0 0.0 0.2 0.0 3.1 1.5 1.1 0.3 0.0 1.0

2.0 15.1 0.0 0.3 6.6 11.3 10.0 0.1 0.0 0.0 0.2 0.0 3.1 1.5 1.1 0.3 0.0 1.0

2.0 15.1 0.0 0.3 6.6 11.3 10.0 0.1 0.0 0.0 0.2 0.0 3.1 1.5 1.1 0.3 0.0 1.0

2.0 15.1 0.0 0.3 6.6 11.3 10.0 0.1 0.0 0.0 0.2 0.0 3.1 1.5 1.1 0.3 0.0 1.0

2.0 15.1 0.0 0.3 6.6 11.3 10.0 0.1 0.0 0.0 0.2 0.0 3.1 1.5 1.1 0.3 0.0 1.0

2.0 15.1 0.0 0.3 6.6 11.3 10.0 0.1 0.0 0.0 0.2 0.0 3.1 1.5 1.1 0.3 0.0 1.0

Total

20.8

52.6

52.6

52.6

52.6

52.6

52.6

52.6

Notes: Numbers are billions of euros and are obtained dividing the total amount of the funds lent to Greece through the Greek loan facility by the ECB capital shares. Estonia, Latvia, Lithuania and Slovakia have 0 s since these countries were not part of the Euro area in 2010–2011. Ireland has 0 for 2011 since it was exempted as it received financial assistance itself in late 2010. Data available at https://www.esm.europa.eu/assistance/greece.

Table A2 Cumulative loan disbursements under the EFSF, ESM and EFSM. 2010

2011

2012

2013

2014

2015

2016

2017

EFSF Greece Ireland Portugal

0.0 0.0 0.0

0.0 7.6 6.9

108.2 12.0 18.2

133.5 17.7 24.8

141.5 17.7 26.0

130.6 17.7 26.0

130.6 17.7 26.0

130.6 17.7 26.0

ESM Greece Spain

0.0 0.0

0.0 0.0

0.0 39.5

0.0 41.3

0.0 39.7

21.4 35.7

31.7 34.7

38.2 31.7

EFSM Ireland Portugal

0.0 0.0

13.9 14.1

21.7 22.1

21.7 22.1

22.5 24.3

22.5 24.3

22.5 24.3

22.5 24.3

Notes: Numbers are billions of euros. Decreasing values reflect repayments. Data available from the websites of the ESM and the European Commission.

Table A3 Capital keys of the EFSF, ESM and EFSM. Country

Belgium Germany Estonia Ireland Greece Spain France Italy Cyprus

EFSF

2.8 27.1 0.3 1.6 2.8 11.9 20.3 17.9 0.2

ESM

3.5 27.1 0.2 1.6 2.8 11.9 20.4 17.9 0.2

EFSM 2010

2011

2012

2013

2014

2015

2016

2017

2.6 16.2 1.0 1.6 7.0 14.2 10.7 0.1 0.1

2.6 15.1 0.9 1.4 7.6 13.9 11.0 0.1 0.1

2.6 16.4 0.9 1.2 6.9 14.2 10.7 0.1 0.2

2.6 17.5 1.0 1.2 6.9 14.6 10.5 0.1 0.2

2.5 17.9 1.0 1.3 6.9 13.6 10.0 0.1 0.2

2.5 16.6 1.1 0.8 6.0 13.0 9.8 0.2 0.1

2.8 14.8 1.3 1.1 7.0 14.3 10.3 0.1 0.2

2.1 14.1 1.3 0.9 5.8 11.7 8.6 0.1 0.1

(continued on next page)

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

15

Table A3 (continued) Country

Luxembourg Malta Netherlands Austria Portugal Slovenia Slovakia Finland

EFSF

0.3 0.1 5.7 2.8 2.5 0.5 1.0 1.8

ESM

0.3 0.1 5.7 2.8 2.5 0.4 0.8 1.8

EFSM 2010

2011

2012

2013

2014

2015

2016

2017

0.2 3.0 3.0 1.3 0.3 0.4 0.4 2.6

0.2 3.0 1.9 1.2 0.3 0.4 1.2 2.6

0.2 3.0 1.9 1.2 0.2 0.5 1.4 2.6

0.2 3.2 2.0 1.1 0.3 0.5 1.3 2.6

0.2 4.4 2.0 1.1 0.2 0.4 1.4 2.5

0.2 3.9 1.9 1.1 0.2 0.4 1.2 2.5

0.2 1.8 1.7 1.2 0.3 0.5 1.2 2.8

0.2 2.4 2.0 1.0 0.2 0.4 1.4 2.1

Notes: Numbers are in percent. Data for the EFSF and ESM are constructed using information contained in their respective websites. Data for the EFSM are constructed dividing each country’s contribution to the European Commission’s budget by the total European Commission revenues. Both these variables are taken from the website of the European Commission. Capital keys for the EFSM are time varying since both member countries’ contributions and the European Commission’s total revenues change every year and do not add to 100 across all euro area countries since also other non-euro area EU countries are EFSM guarantors.

Fig. A1. Financial integration and official financial assistance at country level. Notes: Country-level financial integration measures across all partners, in percent of GDP. “Official financial assistance”, “Bank loans”, “Portfolio equity” and “Portfolio debt” are defined as the sum of the relevant bilateral exposure of country i in country j and the bilateral exposure of country j in country i over the sum of the GDP of countries i and j. “Official assistance” is official financial assistance through bilateral loans (Greek Loan Facility), as well as the EFSF, EFSM and ESM. “Bank loans” refers to data on cross-border bank lending from the Bank of International Settlements. “Portfolio equity” and “Portfolio debt” mark the corresponding components of the IMF Coordinated Portfolio Investment Survey.

16

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

Table A4 Data sources and description. Variable

Source

Bilateral loan stocks

International Locational Banking Statistics - Bank for International Settlements (BIS) Coordinated Portfolio Survey (CPIS) – International Monetary Fund (IMF) European Stability Mechanism (ESM), European Commission and European Central Bank

Bilateral portfolio equity and debt stocks Official financial assistance

Household consumption

Eurostat

Gross domestic product

International Monetary Fund, World Economic Outlook

Population Value added tax

Eurostat European Commission, Vat rates applied in the Member States of the European Union. Situation at 1st July 2018. OECD Tax Database Table II.4.

Statutory tax on dividend income (PIT) Domestic credit

Long-term sovereign bond yield Inflation Gross national savings

Bank for International Settlements (Long series on total credit to the non-financial sectors) International Monetary Fund, World Economic Outlook International Monetary Fund, World Economic Outlook International Monetary Fund, World Economic Outlook

De-jure financial integration measure

European Commission website, following Kalemli-Ozcan et al. (2013).

USD/EUR exchange rate

International Monetary Fund, World Economic Outlook

Definition Outstanding loans of banks in reporting countries vis-à-vis counterparty countries (banking and non-banking sectors) in US dollars. Values outstanding at the end of the year. Cross-border holdings of equities and debt securities self-reported by holder economies and classified by the economy of residence of the issuer, in US dollars. Greek Loan Facility, EFSF, EFSM and ESM loan disbursements and country contribution keys, in euro. We first convert all values into US dollars. To construct the bilateral flows through the EFSF and ESM, we multiply the amount withdrawn by each country with the capital keys of all contributors. To construct bilateral flows through the EFSM, we multiply the amount withdrawn by each country with the share of contributions of member states to the European Commission budget (which vary across time). To construct bilateral flows through the Greek Loan Facility, we simply take the values of bilateral loans from other eurozone member countries to Greece. When the year of payment into the fund is different from the year of withdrawal, we record the bilateral flow at the time when a given recipient (GR, ES) withdraws some funds. For the period before 2010 we set all values to zero. Final consumption of households. Chain-linked volumes, in millions of euros. We derive its per capita growth rate by dividing it by population and taking the log-growth rate. We download two GDP series: (a) the gross domestic product at market prices in billions of US dollars, and (b) the growth rate of per capita gross domestic product, chained-linked volumes, in euros. Note: In 2016 GDP for Ireland was revised upwards by 32% due to a change in accounting rules for profits made by multinational corporations. To deal with the level-change in (a) we (i) derive its growth rate, (ii) linearly interpolate its value for 2016, and (iii) splice the nominal GDP series for the years 2016 and 2017. To deal with the jump in (b) we linearly interpolate its value for 2016. Total population, national concept, thousand persons. Standard and reduced Value Added Tax (VAT) rates.

Overall statutory tax rate on dividend income (sum of the rate on distributed profit and the rate on grossed-up dividend). Total credit by domestic banks to the private non-financial sector. Billions of euros. Nominal 10-year sovereign bond yield. Growth rate in Consumer Price Index, All items. Gross disposable income less final consumption expenditure after taking account of an adjustment for pension funds. Current prices. Billions of euro. We obtain its per capita growth rate by dividing it by total population as well as the consumer price index and taking its log-growth. We construct a variable of de-jure bilateral financial integration according to the following approach. Construct one time- and country-pair-varying series for each directive (and successive repealing directive) belonging to the Financial Services Action Plan. Code 0 if country both countries i and j have adopted the directive and 0 otherwise. Construct a variable being the sum of all series and take the log of one plus such variable. ECB reference exchange rate, USD/EUR. We use it to convert official financial assistance data from euros to US dollars.

Notes: All variables are collected at the yearly frequency and span over the 2001–2017 period.

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

17

Table A5 Descriptive statistics. Variable

Obs.

Mean

Std. Dev.

Min.

Max.

Ci, t Yi, t

177 198 187 198 187 198 198 935 885 917 911

0.467 0.855 5.504 46.172 1.461 1.804 3.637 0.129 1.308 2.816 0.800

2.186 2.473 2.385 7.432 7.023 1.284 2.514 0.363 2.067 2.520 1.387

−10.61 −9.166 1.833 25 −21.613 −1.676 0.150 0.000 0.000 0.026 0.000

4.794 7.890 10.333 69.817 24.05 5.113 24.006 2.224 19.140 12.490 10.838

VATi, t PITi, t  Crediti, t

π i, t

Yieldi, t FAij, t LOANij, t DEBTij, t EQUITYij, t

Notes: This table reports descriptive statistics for the variables used in the analysis. The sample ranges from 2001 to 2017. Obs., Mean, Std. Dev., Min. and Max. shows respectively the number of observations, the mean, the standard deviation, the minimum and the maximum value. Ci, t , Yi, t and Si, t denote, respectively, real per capita consumption, GDP and savings growth in country i at time t. VATi, t and PITi, t are the reduced value added tax and the statutory tax on dividend income rates. Crediti, t is the growth rate of credit of domestic banks to the non-financial sector. π i, t and Yieldi, t are the inflation rate and the 10-year sovereign bond yield. FAij, t denotes the level of financial assistance integration between country i and country j, calculated as the sum of official financial assistance from country i to country j and that from country j to country i, divided by the sum of the GDP of these two countries. Similarly, LOANij, t , DEBTij, t , and EQUITYij, t measure the level of banking loan, portfolio debt and portfolio equity integration between country i and country j at time t. All variables are in percent.

Appendix B: Additional results and robustness analysis

Table B1 Robustness checks on standard errors correction and estimation method.

Yi,t −  Yj,t (Yi,t −  Yj,t )F Ai j,t−1 (Yi,t −  Yj,t )LOANi j,t−1 (Yi,t −  Yj,t )DEBTi j,t−1 (Yi,t −  Yj,t )EQUIT Yi j,t−1 V ATi,t−1 −  V ATj,t−1 P ITi,t−1 −  P ITj,t−1  C rediti,t−1 −  C redit j,t−1 Y iel di,t−1 − Y iel d j,t−1

πi,t−1 − π j,t−1

Baseline (1)

Country-pair clustering (2)

Dyadic correction (3)

DK correction (4)

GLS (5)

0.769∗∗∗ (0.049) −0.392∗∗∗ (0.115) 0.004 (0.006) −0.070∗ ∗ ∗ (0.007) −0.005 (0.032) −0.086∗ ∗ (0.028) −0.037∗ ∗ (0.013) 0.049∗∗∗ (0.012) −0.061∗ ∗ (0.025) −0.083 (0.061)

0.769∗∗∗ (0.045) −0.392∗∗ (0.170) 0.004 (0.006) −0.070∗ ∗ ∗ (0.015) −0.005 (0.033) −0.086∗∗∗ (0.017) −0.037∗∗∗ (0.008) 0.049∗∗∗ (0.007) −0.061∗ ∗ (0.024) −0.083 (0.050)

0.769∗∗∗ (0.108) −0.392∗∗∗ (0.111) 0.004 (0.011) −0.070∗ ∗ ∗ (0.021) −0.005 (0.020) −0.086∗∗∗ (0.024) −0.037∗ (0.019) 0.049∗ (0.024) −0.061∗ (0.029) −0.083∗ ∗ (0.039)

0.773∗∗∗ (0.034) −0.394∗∗∗ (0.136) 0.0 0 0 (0.011) −0.078∗∗∗ (0.014) −0.004 (0.014) −0.089∗∗∗ (0.013) −0.036∗∗∗ (0.006) 0.050∗∗∗ (0.007) −0.049∗ (0.025) −0.106∗∗ (0.048)

0.769∗∗∗ (0.049) −0.392∗∗∗ (0.115) 0.004 (0.006) −0.070∗ ∗ ∗ (0.007) −0.005 (0.032) −0.086∗∗ (0.028) −0.037∗ ∗ (0.013) 0.049∗∗∗ (0.012) −0.061∗ ∗ (0.025) −0.083 (0.061)

Notes: The table reports robustness checks regarding the estimation method and standard error correction. Column (1) reports the baseline results (standard errors clustered at the country i and the country j levels). Column (2) reports estimates with standard errors corrected for country-pair clustering. Column (3) reports estimates with standard errors corrected to account for dyadic correlation, as proposed by Cameron and Miller (2014). Column (4) reports estimates with Driscoll-Kraay corrected standard errors (for general cross-sectional correlation and autocorrelation up to lag 4). Column (5) report estimates based GLS estimation with country pair-specific autocorrelation in the standard errors. Note that interacted variables are always included as direct controls in all regressions, but coefficients are not reported in the table as they are always statistically insignificant. The sample comprises 816 observations, 55 country pairs and 11 countries. All specifications include time fixed effects and country-pair fixed effects. The R-squared is 0.705.

18

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347 Table B2 Additional robustness checks.

Yi,t −  Yj,t (Yi,t −  Yj,t )F Ai j,t−1 (Yi,t −  Yj,t )LOANi j,t−1 (Yi,t −  Yj,t )DEBTi j,t−1 (Yi,t −  Yj,t )EQUIT Yi j,t−1 V ATi,t−1 −  V ATj,t−1 P ITi,t−1 −  P ITj,t−1  C rediti,t−1 −  C redit j,t−1 Y iel di,t−1 − Y iel d j,t−1

πi,t−1 − π j,t−1

Baseline

Savings growth

Post-2009 sample

Financial integration instrument

(1)

(2)

(3)

(4)

0.769∗ ∗ ∗ (0.049) −0.392∗ ∗ ∗ (0.115) 0.004 (0.006) −0.070∗ ∗ ∗ (0.007) −0.005 (0.032) −0.086∗ ∗ (0.028) −0.037∗ ∗ (0.013) 0.049∗ ∗ ∗ (0.012) −0.061∗ ∗ (0.025) −0.083 (0.061)

0.722∗ ∗ ∗ (0.050) −0.425∗ ∗ ∗ (0.126) 0.002 (0.008) −0.067∗ ∗ ∗ (0.007) −0.012 (0.034) −0.071∗ ∗ (0.023) −0.034∗ ∗ (0.013) 0.050∗ ∗ ∗ (0.013) −0.125∗ ∗ ∗ (0.022) −0.078 (0.059) 0.030∗ ∗ ∗ (0.006) 816 0.720

0.866∗ ∗ ∗ (0.055) −0.548∗ ∗ ∗ (0.168) 0.002 (0.023) −0.079∗ ∗ ∗ (0.016) −0.001 (0.026) −0.075∗ ∗ (0.025) −0.043∗ ∗ (0.015) −0.052∗ ∗ ∗ (0.012) −0.065∗ (0.030) −0.058 (0.176)

0.807∗ ∗ ∗ (0.054) −0.344∗ ∗ (0.141) −0.009 (0.015) −0.080∗ ∗ ∗ (0.020) −0.010 (0.051) −0.089∗ ∗ (0.029) −0.037∗ ∗ (0.015) 0.052∗ ∗ ∗ (0.013) −0.065∗ (0.030) −0.090 (0.069)

0.875∗ ∗ ∗ (0.042) −0.478∗ ∗ (0.134) 0.004 (0.009) −0.067∗ ∗ (0.024) −0.078∗ ∗ (0.026) −0.087∗ ∗ (0.029) −0.035∗ (0.017) 0.057∗ ∗ (0.016) −0.066∗ (0.032) −0.087 (0.097)

400 0.869

745 0.707

412 0.790

Si,t−1 −  Sj,t−1 Observations R-squared

816 0.705

Financial integration instrument Core-Periphery (5)

Notes: The table reports robustness checks on the baseline specification (Eq. (2)). Standard errors, clustered at the country i and at the country j levels, are in parenthesis. ∗ ∗ ∗ , ∗ ∗ and ∗ indicate statistical significance at the 99%, 95% and 90% confidence level respectively. Column (1) reports results using the contemporaneous rather than lagged values of FAij, t , LOANij, t , DEBTij, t , EQUITYij, t , V ATi,t − V AT j,t , P ITi,t − P IT j,t , C rediti,t − C redit j,t , Y iel di,t − Y iel d j,t , πi,t − π j,t . Column (2) reports estimates obtained controlling for lagged savings growth differentials Si,t − S j,t , where Si, t indicates real per capita savings. Column (3) reports estimates obtained on the restricted 2010–2017 sample. Column (4) reports estimates obtained using fitted values for the FAij, t , LOANij, t , DEBTij, t , EQUITYij, t variables, where fitted values are obtained after regressing them onto their lag and the lagged de-jure bilateral integration measure proposed by Kalemly-Ozcan et al. (2013). Column (5) reports the same IV approach for the Core-Periphery sample. Note that interacted variables are always included as direct controls in all regressions, but coefficients are not reported in the table as they are always statistically insignificant. The sample comprises 55 country pairs and 11 countries. All specifications include time fixed effects and country-pair fixed effects.

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

19

Fig. B1. Baseline estimation excluding one country at the time. Notes: the figure shows the degree of shock absorption (black thick solid line) defined as γ t from Eq. (4) and relative 90% confidence bands (thick dotted lines) in the full sample, as well as in restricted samples in which, in turn, one single country is excluded. The integration terms are evaluated at their bilateral annual averages in the corresponding sub-sample of country pairs.

Supplementary materials Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.euroecorev.2019. 103347. References Albertazzi, U., Bottero, M., 2014. The procyclicality of foreign bank lending: evidence from the global financial crisis. J. Int. Econ. 92. Alcidi, C. and G. Thirion, (2017), “Fiscal risk sharing and resilience to shocks: lessons for the euro area from the US”, CEPS Working Paper No. 2007/17. Allard, C., 2011. More Europe, not less. Int. Monetary Fund Finance Dev. 48, 52–54. Asdrubali, P., Sørensen, B.E., Yosha, O., 1996. Channels of interstate risk sharing: United States 1963–1990. Q. J. Econ. 111, 1081–1110. Balli, F., Basher, S.A., Balli, H.O., 2013. International income risk sharing and the global financial crisis of 20 08–20 09. J. Bank. Finance 37, 2303–2313. Backus, D.K., Kehoe, P.J., Kydland, F.E., 1992. International Real Business Cycles. J. Political Econ. 100 (4), 745–775. Backus, D.K., Smith, G.W., 1993. Consumption and real exchange rates in dynamic economies with non-traded goods. J. Int. Econ. 35 (3–4), 297–316. Beine, M., Cosma, A., Vermeulen, R., 2010. The dark side of global integration: increasing tail dependence. J. Bank. Finance 34, 184–192. Beetsma, R. Cima, S., Cimadomo, J. (2018), “A minimal moral hazard central stabilisation capacity for the EMU based on world trade”, ECB Working Paper No 2137. Canova, F., Ravn, M., 1996. International consumption risk sharing. Int. Econ. Rev. 37 (3), 573–601. Cameron, C., Miller, D.L., 2014. Robust Inference for Dyadic Data. Mimeo. Corsetti, G., Kuester, K., Meier, A., Müller, G., 2013. Sovereign risk, fiscal policy, and macroeconomic stability. Econ. J. 123 (566), F99–F132. Del Negro, M., 2002. Asymmetric shocks among U.S. States. J. Int. Econ. 56 (2), 273–297. Delrio, S., Mueller, G., Kamps, C., Koester, G. (2018), “Risk sharing in EMU”, European Central Bank, manuscript. Demyanyk, Y., Ostergaard, C. and Sorensen, B. (2008), “Risk sharing and portfolio allocation in EMU”, European Economy - Economic Papers 2008–2015, No 334, European Commission. Driscoll, J., Kraay, A., 1998. Consistent covariance matrix estimation with spatially dependent panel data. Rev. Econ. Stat. 80 (4), 549–560. European Central Bank (2017), “Financial integration in Europe”, Chapter 1: Recent developments in financial integration in the euro area. Epstein, B., Mukherjee, R., Ramnath, S., 2016. Taxes and international risk sharing. J. Int. Econ. 102 (issue C), 310–326. Farhi, E., Werning, I., 2017. Fiscal unions. Am. Econ. Rev. 107 (12), 3788–3834. Fratzscher, M., Imbs, J., 2009. Risk sharing, finance, and institutions in international portfolios. J. Financ. Econ. 94, 428–447. Furceri, D., Zdzienicka, A., 2015. The euro area crisis: need for a supranational fiscal risk sharing mechanism? Open Econ. Rev. 26, 683–710. Galí, J., Monacelli, T., 2005. Monetary policy and exchange rate volatility in a small open economy. Rev. Econ. Stud. 72 (3), 707–734. Hepp, R., von Hagen, J., 2013. Interstate risk sharing in Germany: 1970–2006. Oxf. Econ. Pap. 65 (1), 1–24. Ioannou, D. and D. Schäfer (2017), “Risk sharing in EMU: key insights from a literature review”, SUERF Policy Note, Issue No 21. Kalemli-Ozcan, S., Papaioannou, E., Peydro, J.-L., 2013. Financial regulation, financial globalization, and the synchronization of economic activity. J. Finance LXVIII (3), 1179–1228. Kose, M.A., Prasad, E.S., Terrones, M.E. (2007), “How does financial globalization affect risk sharing? Patterns and channels”, IMF Working Paper, 07/228. Lane, P.R., Milesi-Ferretti, G.M., 2007. The external wealth of nations mark II: revised and extended estimates of foreign assets and liabilities, 1970–2004. J. Int. Econ. 73 (2), 223–250.

20

J. Cimadomo, G. Ciminelli and O. Furtuna et al. / European Economic Review 121 (2020) 103347

Lewis, K.K., 1996. What can explain the apparent lack of international consumption risk sharing? J. Political Econ. 104 (2), 267–297. Mace, B.J., 1991. Full insurance in the presence of aggregate uncertainty. J. Polit. Econ. 99 (5), 928–956. Mélitz, J., Zumer, F., 2002. Regional redistribution and stabilization by the center in Canada, France, the UK and the US: a reassessment and new tests. J. Public Econ. 86 (2), 263–286. Milano, V. (2017), “Risk sharing in the eurozone: the role of European institutions, center for labor and economic growth, Department of Economics and Finance, LUISS Guido Carli Working Paper, No. 01/17. Ongena, S. and Popov, A.A. and van Horen, N. (2016), “The invisible hand of the government: moral suasion’ during the European sovereign debt crisis”, ECB Working Paper No. 1937. Pierucci, E., Ventura, L., 2010. Risk sharing: a long run issue? Open Econ. Rev. 21 (5), 705–730. Rangvid, J., Santa-Clara, P., Schmeling, M., 2016. Capital market integration and consumption risk sharing over the long run. J. Int. Econ. 103, 27–43. Saka, O. (2017), “Domestic banks as lightning rods? Home bias during the eurozone crisis”, LEQS Paper No. 122. Sørensen, B.E., Yosha, O., 1998. International risk sharing and European monetary reunification. J. Int. Econ. 45, 211–238. Sørensen, B.E., Wu, Y.T., Yosha, O., Zhu, Y., 2007. Home bias and international risk sharing: twin puzzles separated at birth. J. Int. Money Finance 26, 587–605. Tasca, P., Battiston, S., 2014. Diversification and Financial Stability”. The London School of Economics and Political Science SRC Discussion Paper, No 10. Systemic Risk Centre.