1. Introduction
Research in the field of cross-border and international banking is increasingly becoming important as the scale and scope of financial globalization has been intensified in the previous decades and will presumably continue to do so in the decades to come. Some advantages of financial globalization and presence of foreign banks come in the form of increased competition in the banking sector and increased banking efficiency (Bonin, Hasan, & Wachtel, 2005; Havrylchyk, 2006; Semih Yildirim & Philippatos, 2007). Additionally the presence of foreign banks in an economy can contribute to local financial stability in emerging markets (Crystal, Dages, & Goldberg, 2001). However, some of the negative effects of an increasingly globalized financial system only became apparent when a large shock occurred in this financial system.
Negative consequences of financial globalization can be found in the international transmission of financial shocks through internal liquidity markets of global banks (Cetorelli & Goldberg, 2011). Evidence for emerging countries also shows that the real economy is impacted by loan supply decreases for foreign bank affiliates (Cetorelli & Goldberg, 2012). As a result the presence of foreign banks can both increase the stability of local banking markets but also import foreign shocks into the financial system of a country, which can be especially harmful for emerging economies (De Haas & Van Lelyveld, 2014).
As international banking results in a more and more entangled system with larger capacity for shock propagation national regulators cannot keep up. This is where the need for internationally coordinated regulation comes into play (Claessens, Herring, Schoenmaker, & Summe, 2010). The Basel Committee on Banking Supervision (BCBS) provides ground for such internationally coordinated banking regulation and aims to increase the efficiency and effectivity of regulation and supervision on a worldwide scale (BCBS, 2016). Recommendations of this committee, consisting of regulators from various countries, translate into regulation on both national and supranational scale such as in the latest European Union capital requirement directives (CRD IV) that went into effect on January 1st, 2014.
The effect of capital regulation for financial institutions has been the topic of extensive research. Prior regulation in the form of Basel I and II have shown both pro-cyclical effects but also resulted in lower social costs in the form of credit crunch in a financial crisis (Repullo & Suarez, 2013). New and higher capital requirements can also cause a temporary decrease in lending while banks rebuild buffers. This effect is most pronounced in commercial real estate and other corporates while the effect on household lending is less severe (Bridges et al., 2014). Consecutively, regulation restricting bank activities affects banks negatively in terms of efficiency, while increased capital requirements can have a slight positive effect (Barth, Lin, Ma, Seade, & Song, 2013).
2013) and potentially influence banks in a disproportionate manner in the execution of the CRD policies. With the implementation of the CRD both domestic- and foreign-owned banks are requested to follow new regulation and adapt its structure to accommodate. The high foreign bank concentration in some EU countries poses the question of whether the single rulebook will actually affect all banks equally.
Literature on shock propagation asserts that financial shocks propagate through global banks’ inner capital markets where liquidity is moved to the headquarter when in need resulting in contraction of foreign bank cross border lending, contraction in lending supply of developing country affiliates, and a contraction in domestic bank lending supply (Cetorelli & Goldberg, 2011, 2012). For banks to adjust to new financial regulation a similar effect might occur where affiliates are used to create faster convergence to new regulation of its parent firm by drawing resources from subsidiary to headquarters. The main question that this paper will try to answer is whether foreign-owned banks react differently to new capital requirements than domestic-owned banks. To answer this, bank risk is used to assess the degree to, and the speed at which banks adjust to new regulation. Loan growth is used to gauge the impact that converging to capital requirements has on loan supply (and therefore the impact on the real economy) of banks. A decrease in loans signals an apparent draw of resources for increasing capital. This paper focusses on the introduction of CRD in 2014 and will utilize data on European, US, and Chinese banks between 2011 – 2016 to provide a pre- and post-CRD situation. The use of the synthetic control method provides an accurate method to compare foreign- and domestic-owned banks.
This study finds that foreign- and domestic-owned banks behave differently after new financial regulation is introduced. Domestic-owned banks adjust faster and stronger to financial regulation than foreign-owned banks. Domestic-owned banks decrease risk more and more quickly than foreign-owned banks after the introduction of CRD. The CRD has an unbalanced effect between these banks and regulators should accommodate to this in the future.
The rest of the paper is structured as follows. Section 2 gives a literature review and provides insight in the necessity for prudent, supranational financial regulation, using existing theory on cross-border banking to build a theoretical framework; it also develops hypotheses. Section 3 develops the methodological background for this research and the strategy to answer the research question. Section 4 expands on the data that is used in this paper. In section 5 the analysis and in section 6 the robustness tests are performed. Section 7 concludes this paper and provides an overview of the results of the paper and the discussion in section 8 additionally lists limitations, prospects for future research, and an alternative explanation for the obtained results.
2. Theoretical Background
2.1 The Crisis in Europe, 2007-2013
of how it spread from the US mortgage market to the real economy and political playing field of the EU in 5 ‘waves’ (Liikanen et al., 2012). This section highlights why the global financial system needs the supranational regulation that has taken shape in recent years.
The first wave takes place in 2007 in the US. US housing prices boomed over the previous decade1, this was mainly funded by increases in credit (mortgages) and the so-called mortgage backed securities (MBS) of which a large part (20%) was considered sub-prime. These sub-prime loans were packaged into tranches and again traded between banks, increasing leverage ratios of banks to unprecedented heights. When investors became aware of the underlying sub-prime loans, the value of MBSs dropped and institutions that had heavily invested in the US mortgage market were severely affected by the loss of value.
The banking crisis of 2007 went systemic in 2008. The crash of Lehman caused severe turmoil on financial markets that manifested itself in rise of volatility, interest rate drops, and currency depreciation. The cross-border and financially integrated nature of the banking system meant that national deposit guarantee schemes in Europe were no longer suitable and required additional intervention, adding to the financial gridlock of 2007 where the ECB had to inject additional liquidity into the European financial system to compensate for the eroded trust between banks. As a whole, liquidity dried up and bank risk hedging was deemed insufficient.
The third wave saw the impact of the banking crisis on the real economy and public debt. With world trade having declined during the crisis there were large decreases in employment opportunities and tax revenue. Spending on stimulus packages and automatic stabilizers caused substantial growth of public debt levels.
In the fourth wave the debt of the Euro area had risen to heights not seen before at 87% of GDP in 2011. However, this statistic hides considerable variation in debt levels among the member states. The financial crisis that started in the banking sector had now also turned into a crisis of sovereign debt.
The final wave essentially consists of the consequences of the waves before it and is currently still influencing Europe. The loss of confidence in Europe halted further financial integration. There has been a large focus on nationalism by both governments and banks (Liikanen et al., 2012) while the shortcomings of existing regulation at the national level were painfully highlighted in the years before. As a result, not only state budgets, but also democratic legitimacy and accountability took a hit in the EU (Lane, 2012). European institutions have since been aiming for more comprehensive regulation of financial markets and institutions to prevent a second financial crisis and promote further financial integration. The CRD IV can be seen as a direct result of the shortcomings of previous financial regulation.
2.2 Cross-Border Banking
The escalation from a national housing bubble into a global financial crisis highlights the scale of financial globalization, its pitfalls, but also the need for unified financial regulation. For example, growth in banking activity (assets) in countries such as Ireland, Greece, and Spain was multiple times higher that economic (GDP) growth during 2001 – 2008, a significant
portion of this growth has been in the expansion of international activities. In some countries, non-domestic banks hold over 80% of assets in the banking sector (Claessens et al., 2010; Liikanen et al., 2012).
Cross border banking and global banks have gained considerable traction over the past decades as international financial integration between nations and financial institutions is increasing. Both in terms of assets and liabilities cross border activity has increased spectacularly after 2000. Claessens et al. (2010) describe two trends in the financial industry; (1) the consolidation and increasing size of financial institutions in countries to end up with only a handful of large financial firms and (2) increasing international activities undertaken by these large firms. Truly global banks are characterized by “their global operations and […]
networks or physical branches and subsidiaries in foreign countries” (Cetorelli & Goldberg,
2012), however, many banks tend to operate regionally, for example within Europe (Claessens & van Horen, 2014).
The trend of financial globalization has caused banks to increase in size and as a result there are now many more systemically important financial institutions (SIFIs) then two decades ago. Claessens et al. (2010) show that between 1998 and 2009 the amount of assets held by the 5 largest global banks doubled from 8% to 16% and that by 2009 47% of global cross-border assets and 50% of global cross border liabilities were held by a selection of 6 countries. This illustrates the growing importance of cross-border banking and should be cause for regulation to follow the same path of globalization.
Furthermore, banks that are deemed too-big-to-fail are increasingly common with the increasing size and international scope of financial institutions. Failure of these institutions would result in catastrophic economic consequences to both the domestic and global economy (Claessens et al., 2010). The Financial Stability Board publishes a list of 30 global systemically important banks (G-SIBs) annually. In 2016, out of these 30 G-SIBs 8 were located in the US, 7 in Asia, 2 in Switzerland, and 13 in the EU2. The large presence of such banks in Europe and the potential international spillover in case of bank-failures reinforce the need for well-functioning regulation on a supranational scale.
Several examples of problematic resolutions of bank failure are available to provide further argumentation for supranational financial regulation. The resolution of Dexia, Fortis, and several Icelandic banks in the financial crisis would have benefited greatly from more international coordination. Coordination and clarity would have decreased uncertainty, decreased potential costs, and would have led to faster resolution (Beck, Todorov, & Wagner, 2013; Claessens et al., 2010).
2.3 Balance Sheet Contagion
Global banks, as described in the previous sections, have played a significant role in the transmission of the financial crisis from the US to the rest of the world. Global banks have a variety of characteristics that can cause transmission of financial shocks.
The use of internal capital markets of global banks can be a source of stability in times of domestic financial instability but can also destabilize markets in which foreign affiliates operate. Capital flowing from foreign affiliates to the head office of global banks in times of liquidity constraints decreases the loan supply in the countries where affiliates are located (Cetorelli & Goldberg, 2011, 2012). Not only intra-bank channels play a role in financial contagion, interbank lending has a similar effect when non-optimal in and between regions. Intraregional claims should, theoretically, protect against local liquidity shocks. However, when intraregional claims are suboptimal and incomplete between regions, liquidity preference can be a source of contagion between banking regions (Allen & Gale, 2000).
require a reaction at the asset side to balance out liabilities and assets, as a result not only the foreign affiliate cuts lending, the domestic owned bank is also influenced through interbank lending channels (Cetorelli & Goldberg, 2011). Hence, because balance sheets of parent and affiliates are intertwined, an event in one country can have consequences in another location. Likewise, the cause of a bank failure does not have to occur within the country it operates in.
With the many globally and regionally systemic banks that are located in Europe the importance of preventing such shock transmission becomes apparent. Although foreign owned banks may improve regional stability in the 2007 – 2009 financial crisis and following recession, foreign subsidiaries of multinational banks had to cut lending (De Haas & Van Lelyveld, 2014). This is in line with other extant research on transmission of financial shock transmission (Cetorelli & Goldberg, 2011, 2012; Claessens & van Horen, 2014).
2.4 The Importance of Capital Regulation
Bank risk-taking behavior has been extensively documented upon in extant literature in the past decades and has been the subject of multiple regulatory projects. Two overall developments in banking have increased risk-taking behavior of banks: competition and moral hazard induced by government policies. The former is a result of new entrants in the banking market and thus increased competition, the latter is mainly concerned with the impact of government policies on the cost-benefit functions of financial institutions.
An increase in banking competition (the result of financial liberalization) overall results in a decrease of potential future rents of the bank, a natural reaction in a competitive environment where profit margins thin as the threat of potential entrants increases. The rationale of this relationship between competition and non-prudential behavior runs through the profit margins and franchise values (capitalized future profit). With lower (future) profits as a direct result of an increase in competition the incentives for making safe loans decreases (Hellmann, Murdock, & Stiglitz, 2000; Repullo, 2004).
The other side of this coin relates to the cost-benefit function of financial institutions and the social and financial externalities related to bank failure. The decision between a safe but low-return investment and risky high-return investment is based on the potential profit (benefit) and losses (cost) of such an investment. Prudent behavior is encouraged if costs are internalized in the bank, an investment loss and potential bank failure will have high, direct consequences for creditors and shareholders of the bank. However, because of the externalities of bank failure such as social costs to depositors and consequences for other financial institutions, governments impose a policy safety net to increase stability and soundness of the banking sector (Berger, Herring, & Szegö, 1995; Rime, 2001). Policies to decrease the externalities of financial distress or bank default involve deposit insurance in case of bank default, central bank support in the form of special discount rates, or even a bailout. However, this safety-net alters the cost-benefit function of banks where the costs are borne by the government and indirectly the tax-payer while the benefits accrue to bank ownership and management, increasing the propensity for risky investments.
times and incentives for more prudent investment. It aims to find a “balance between the
marginal social benefit of reducing the risk of the negative externalities from bank failures and the marginal social cost of diminishing intermediation” (Berger et al., 1995). The latter part
refers to the higher costs accompanying capital that are transferred to the customers, a lower lending supply, and a decrease in financial intermediation by financial institutions under regulatory capital requirements (Berger et al., 1995; Furfine, 2000). The decrease in lending supply is especially relevant for this research.
2.5 Regulation Basel III and CRD IV/CRR
The previous sections establish the importance of prudential banking regulation to prevent events such as the global financial crisis, this section will briefly introduce recent capital requirement regulation by the Basel Committee for Banking Supervision (BCBS) and the adoption of Basel III in the EU under the CRD IV/CRR package.
The BCBS proposes reforms to “strengthen global capital and liquidity rules with the
goal of promoting a more resilient banking sector” (BCBS, 2011). It sees excessive leverage
build-up in financial institutions and the abrasion of both quality and quantity of capital as a prime reason for the financial crisis and recession of 2007 – 2009. The reforms that the Basel Committee proposed for Basel III are meant to prevent such crises, reduce systemic contagion, and reduce the transmission of financial crises to the real economy and the public sector. Basel III consists of policies to reduce the impact of financial shocks by increasing the ability of banks to absorb these shocks, to make banks more transparent, and cross-border banks less systemic (BCBS, 2011).
Capital requirements in Basel III take two forms; Tier 1 and Tier 2 capital. Tier 1 capital mainly consists of common shares and retained earnings and is required to be 6% of risk weighted assets (RWA) of a financial institution. Of this 6% requirement, the dominant capital component is common shares and retained earnings (4,5% ratio to RWA) labelled as Common Equity Tier 1 (CET-1) capital. Additionally, the remaining 1,5% ratio of additional Tier 1 capital to RWA consists of supplementary capital paying non-cumulative dividends, or coupons without a maturity date or incentive to redeem. Tier 2 capital is required to be 2% of RWA and contains instruments that are not Tier 1 capital but are still considered
stable capital and can be used as a risk-buffer (BCBS, 2011). The minimum build-up of required capital of financial institutions determined by Basel III regulations can be seen in figure 2. Previously, Basel I and Basel II also required an 8% capital requirement but with lower Tier 1 capital of 4% to RWA and lacked risk assessment tools compared to Basel III.
Additional to capital requirements, Basel III includes 4 other policies: An increased risk coverage that includes stress testing of financial institutions, leverage ratio constraints, the
4.50% 1.50% 2% 0.00% 1.00% 2.00% 3.00% 4.00% 5.00% 6.00% 7.00% 8.00% 9.00% Ratio to Risk Weighted Assets Tier 1 CET-1 Tier 2
promotion of counter-cyclical buffers and the reduction of pro-cyclical amplification of financial shocks, and reducing the systemic nature of global banks and interconnectedness in the global financial market. Although this paper focusses mainly on the impact of capital requirement regulation the Basel capital requirements and the other policies cannot be seen as separate from each other as they jointly form a comprehensive regulatory package.
The implementation of Basel III in the EU has been performed under the CRD IV/CRR package that includes a directive to be included in national legislation (CRD IV) and a EU wide regulatory component (CRR), the combination on which is referred to as CRD. It is also implemented as the single rulebook for all financial institutions in the Eurozone under the Single Supervisory Mechanism (SSM). The CRD follows Basel III in the definition and required minimum ratio of total capital to RWA of 8% that is seen in figure 2. It also includes ECB regular stress tests of financial institutions to determine the resilience of the European banking sector.
2.6 Regulation and Risk
There has been much debate about the consequences of introducing capital requirements for banks. Arguably there is a need for such requirements to internalize the externalities of bank failure that originate when the market capital requirements are reduced by a government safety net that insulates the bank from any market discipline. In this situation, potential losses are absorbed by governments (and consequentially the tax-payer) and capital requirements exist to prevent such cost (Berger et al., 1995). Other findings reflect on capital requirements and find that for an efficient outcome the use of only (flat rate) capital requirements is not beneficial and may even encourage risk-taking (Hellmann et al., 2000; Repullo, 2004). Finally, another potential consequence of capital regulation is on loan growth. A BIS paper theorizes that a 1% increase in capital requirements is followed by a 4,68% (leverage based level) or 5,46% (risk based level) reduction in loan growth in an estimated model of an US bank (Furfine, 2000).
2.7 Regulation and Lending
As for the effect of capital requirements on lending by banks there is no general consensus. An increase in required capital could reduce the amount of lending in the short term, substituting with other asset types. In the long run the effects are unclear and might yield increased lending (VanHoose, 2007). Cohen (2013) remarks in the BIS quarterly that overall banks have not cut back on lending in adjustment to new Basel 3 capital standards, however, there has been stronger lending growth in banks with high capital ratios compared to other banks and a “shortfall” in lending growth in Europe. In European banks, it can be observed that lending is substituted by other forms of assets such as cash and securities (Cohen, 2013). It is also argued that banks that cannot increase equity by retaining earnings or selling shares (increasing equity) are unhealthy and inherently high riskiness. Correspondingly, banks with more equity have more funding available to increase lending (Admati & Hellwig, 2013).
In a study conducted on the banking system of the United Kingdom over 1990 – 2011 it was found that banks affect lending differently over various economic sectors and that there was an overall cut in loan growth after capital regulation was amended that was restored over a time period of 3 years (Bridges et al., 2014). Evidence for emerging economies shows not only the effect of capital requirements on credit supply but also a contagion effect across countries not enforcing regulation and a decreased effect on foreign owned banks in their sample (Chiuri, Ferri, & Majnoni, 2002).
2.8 Effects Across Banks
In the previous sections, it was established that capital regulation has an effect on both risk and lending of banks and that different banks are affected heterogeneously by capital requirement. Banks with generally high capital buffers can increase risk and lending while the adverse effect is true for banks with a lower capital base (Berger et al., 1995; Calem & Rob, 1999; Cohen, 2013), consequently it can be established that banks seek the optimum capital/risk ratio.
Considering the effects of a financial shock on banks, it can be discerned that the effect on domestic-owned and foreign-owned (subsidiary) banks is different. Foreign bank subsidiaries have to decrease lending more in comparison to domestic banks and interbank contagion takes place via the balance sheets of global banks and interbank capital flows (Cetorelli & Goldberg, 2011, 2012; Claessens & van Horen, 2014; De Haas & Van Lelyveld, 2014). The consequences of adjustment to capital requirements, regulated decrease in risk-weighted assets, or decreasing liquidity by implementation of the CRD IV package in the EU could also transfer between the balance sheets of different banks. This is similar to the causes and consequences of financial shocks being in different locations.
for the bank, illustrated by the blue arrow. As a response, more equity is issued to increase capital investment in the bank, it increases its low-risk assets, and decreases its loans (higher-risk assets). A bank that needs to increase capital has the options to decrease loan growth, decrease riskiness of asset composition, or increase equity until compliant with the new capital requirements.
For a multi-national bank consisting of a parent bank and a subsidiary bank the effect of new financial regulation is more complicated, this is illustrated in figure 4 below. Following the logic of Cetorelli and Goldberg (2011, 2012) that a financial shock can be transmitted though the balance sheets of parent and affiliate banks the impact of new financial regulation can have a similar effect. With both entities under the same regulation, both banks (individually and consolidated) require an increase in capital. For the domestic bank, the likely effect is similar as for a stand-alone bank, an increase in issued equity and RWA decrease. The foreign-owned bank also requires to raise its capital ratio. However, the post-crisis narrative has seen a decrease in bank exposure to other EU countries, signaling divestment from foreign loans. The consequent risk of retrenchment and home bias for banks have even led to initiatives3 that
3 Participating in the Vienna 2.0 initiative are “key International Financial Institutions, the
European Commission and relevant EU institutions, the principal cross-border banking groups, and home and host country authorities” (European Bank Coordination Initiative,
2012).
Figure 3. Stand-alone bank, effect of financial regulation.
aim to ensure cross-border financial stability for countries with high foreign bank presence in light of decreasing cross border loans (Liikanen et al., 2012). In this case, new capital regulation results in a decrease of cross-border loans and with a decrease in funding from abroad, foreign-owned banks can be impacted more severely by new regulation. Contrary to the domestic parent, foreign affiliates would need more extreme measures to meet with CRD. Rather than increasing equity it would need to decrease lending more than the parent bank to adjust its capital ratio. It would take longer for a foreign-owned bank to adjust risk to new capital requirements than for a domestic (parent) bank and the risk of foreign-owned banks would initially remain higher after the introduction of new regulation.
2.9 Hypotheses
The theoretical section explored the need for a new set of financial regulations both globally and regionally in the EU, the workings behind financial regulation and Basel 3, contagion through changes in the balance sheet of a parent bank influencing the balance sheets of affiliate banks, and finally, how financial regulation impacts the balance sheets of domestic and foreign owned affiliates. In extant literature, there is agreement that banks adjusting to capital regulation can increase equity, decrease loans and shift to more risk-free assets. As argued in previous sections, foreign-owned banks can see a decrease in funding through the banks’ internal liquidity market and have difficulty to increase capital through retained earnings or shares and would need to cut loans more severely than domestic-owned banks. The process of adjusting to new capital requirements does not only potentially require a larger cut in loans by foreign-owned banks, the risk of foreign-owned bank remains relatively higher after the introduction of financial regulation as the process of adjusting takes longer. Sources of internal funds are limited by the parent bank and a more disruptive process of increasing capital must be followed. Domestic-owned (parent) banks will therefore be able to adjust faster to new regulation and decrease risk faster and more than foreign-owned affiliates can. This leads to three hypotheses:
Hypothesis 1a: Under new financial regulation, domestic-owned banks decrease risk more
than foreign-owned banks.
Hypothesis 1b: Under new financial regulation, domestic-owned banks decrease risk faster
than foreign-owned banks.
Hypothesis 2: Under new financial regulation, foreign-owned banks decrease loans more
3. Methodology
3.1 Sample Banks
The focus of this research is on the implementation of Basel 3 in the form of CRD in the European Union. The use of banks in the European Union has a number of advantages; Firstly, all banks use similar reporting standards mandated by the “single rulebook”, this makes data more comparable. Secondly, the uniform introduction of Basel 3 as the CRD package in multiple countries allows this study to use a after treatment state of banks. This before-after scenario makes it suitable to use the synthetic control method. Thirdly, most banking activity takes place regionally, not globally (Claessens & van Horen, 2014). For a suitable sample of interrelated banks that operate under the “single rulebook” the prime groups consist of EU banks. A further sample of US and Chines banks is used to verify that the introduction of CRD has seen a significant effect over the banking sector of the EU.
The banks that are selected are required to meet a number of criteria for use in the sample. Firstly, the banks must have sufficient historical data available for this research to be accurate. Although Orbis bank focus is a relatively new database, with the discontinuation of Bankscope, it is the only viable alternative for financial data available at this time. The SCM required pre- and post-treatment periods, the introduction of CRD in 2014 leads to a sample period 2011 – 2016. In this sample, pre- and post-treatment periods are balanced. Secondly, banks must be listed as active and alive during the total sample period. Thirdly, the sample allows for a number of different types of financial institutions: commercial banks, savings banks, cooperative banks, real estate & mortgage banks, and investment banks. Sample banks that exhibit extreme asset fluctuations, loan growth or decline, CAR, or lack data for multiple years are omitted from the sample. Prior to analysis the data is Winsorized to remove further outliers.
3.1.1 Foreign-owned Banks
consists of 41 European, foreign owned banks on which Orbis has sufficient data in the sample period, for 3 of these banks4 CDS data5 is also available on Datastream.
3.1.2 Domestic-owned Banks
To compare the effects of regulation between foreign and domestic owned banks the group that is used to create the synthetic domestic-owned banks is built from European banks that are domestically owned, compared to the foreign owned group. In selecting banks for the domestic-ownership group, it is again important to have sufficient data on the required levels of analysis for the duration of our sample. Financial institutions are selected on domestic ownership using Orbis bank focus, institutions for which bank focus does not list an ultimate owner are further checked in the Claessens & van Horen (2015) database. Banks are considered as domestic when >50% of shares is held by a domestic owner.
This results in a sample of 159 domestically owned European financial institutions. For 28 of 159 financial institutions CDS data is available for a market risk analysis.
3.1.3 Control Group
Additional to foreign- and domestic-owned groups of European banks, a control group of banks not under treatment of CRD is used to verify the method proposed in the next section to use and the final results. This sample contains a sample of 218 banks from the United States and China, both countries had not introduced Basel 3 regulation by 2014 with the implementation of most regulation phasing in from 2016 onward6 (BCBS, 2015). This lack of exposure to the treatment makes this group of banks ideal for use with the SCM as it was originally intended. This group consists of the 250 largest banks in the US and China that meet sample criteria, 14 of these banks are Chinese. Observations with extreme fluctuations in the dependent and independent variables are omitted, resulting in a sample of 218 banks over 6 years.
4 Santander Cards UK Limited (UK), UniCredit Bank AG (Germany), UniCredit Bank Austria AG (Austria)
5 The CDS data is used to measure market risk. This further verifies the results obtained using accounting risk data.
3.2 Levels of Analysis/measures
3.2.1 Bank Risk
Bank risk allows itself to be measured in two general ways, accounting risk and market determined risk. A method used for accounting bank riskiness is the z-score of a bank, using several ratios it determines the distance between a bank’s current financial position and insolvency by measuring the amount of variation in income is necessary to deplete the capital and returns of a bank, a higher Z-score indicates a more stable bank. As the Z-score is highly skewed the natural logarithm of the Z-score is used to normalize data (Laeven & Levine, 2009). The z-score of a bank is calculated by adding the capital-asset ratio (CAR) and the return on assets (ROA) and dividing this by the standard deviation of the return on assets of the bank:
𝑍 =(𝐶𝐴𝑅 + 𝑅𝑂𝐴) 𝜎𝑅𝑂𝐴
The Z-score is determined with firm level data from Orbis, as mentioned above this data is highly skewed and henceforth normalized by natural logarithm. Further mentions of the Z-score thus regard the natural log of the Z-Z-score.
A second method to measure bank risk is to measure the risk the market attributes to the bank. This can be done in two ways, either by using volatility of equity returns or by credit default swap (CDS) spreads. The former method measures the weekly volatility in the daily returns of equity and is available as the total return index on Datastream. The second measure for bank risk are CDS spreads. A CDS is credit derivative that protects the buyer against credit events such as bankruptcy, missed payments, or defaults among others. The premium paid on a CDS represents the likelihood of such an event happening, CDS spreads represent the insurance against a credit event and therefor the default risk of the bank. A higher CDS spread is the direct result of a bank deemed more risky by the market (Beck et al., 2013). CDS spreads are available on Datastream. Data on total market returns and CDS spreads are limited to a small number of very large banks, this makes a broad analysis with market risk levels difficult. For this reason, this paper uses data on CDS to verify the results obtained with Z-scores.
3.2.2 Loan Growth
Loan growth is an indication of the growth on the supply side of credit by banks that expand to new markets or into new products, take over activities of other banks, or loosened standards for new borrowers (Foos, Norden, & Weber, 2010). Foos et al. (2010) find that loan growth is positively correlated to future loan losses, a decline in interest income, and negatively correlated to bank solvency. Furthermore, capital ratios decrease with an increase in loans. In this paper loan growth can be measured by the percentage change in the loans given out by a bank (at point t) compared to the previous year (t-1).
3.2.3 Control Variables
be explained by these controls, while the remainder is explained by our independent variables of interest. In extant literature, there is a widespread use of a number of variables for bank analysis; Size, profit, and liquidity on the micro level and economic size and development of the operating country on the macro level.
On a bank level the control variables are used for the implied differences in bank risk and loan growth for banks with different characteristics. Firstly, the natural logarithm of the total assets of the banks is used to capture size effects, larger banks are implied to be less risky and are able to have an increased loan growth through the reconfiguring of their asset base (Laeven & Levine, 2009; Rime, 2001). Profit is used as profits can positively affect a bank’s capital base. Secondly, an already profitable bank does not need to take high risks to gain more profits (Rime, 2001). Return on assets (ROA) is used as this ratio adjusts profits to bank size. A third control variable on a micro level is the liquidity of banks as a ratio of liquid-assets-to-total-assets. A bank with a higher percentage of liquid assets can adjust faster to new regulations and is inherently less prone to experience assets that negatively affect its liquidity.
On a macro level, economic size is used as the natural logarithm of country GDP, while development is measured as the natural logarithm of GDP per capita. Larger and more developed countries can be seen as more stable environments for banks to operate in with better opportunities for loan growth and a lower risk.
3.3 Synthetic Control Method
This paper applies the synthetic control method (SCM) to study whether the introduction of a new package of financial regulation has a different effect foreign owned banks compared to domestic owned banks. The SCM derives from the methods used in science where the specific effect of a treatment is tested with the use of a treatment and a control group. This method was first used in a research on the economic effects terrorism in the Basque conflict between 1960 and 2000 (Abadie & Gardeazabal, 2003) and further motivated in a paper highlighting the impact of a tobacco control program in the California (US) (Abadie, Diamond, & Hainmueller, 2010).
The SCM is used in this paper to compare the effect of new financial regulation on a group of foreign-owned banks to a situation in which these banks would have been domestically owned. The domestic-ownership group is a (donor) sample of banks similar to the group of foreign owned banks that is used to build a synthetic domestic-owned group. This synthetic group is the counterfactual situation in which the banks in the foreign-ownership sample would have been domestically owned. The synthetic control group allows for a better comparison in the situation where no direct comparison treatment group is available and with a smaller sample than other comparable methods (Abadie et al., 2010). The outcome of comparing such two groups directly would only demonstrate the disparities between the two, not the effect it aims to find (Abadie, Diamond, & Hainmueller, 2015). Ordinarily the “treatment” sample undergoes treatment while the donor group is non-subjective to the same or likewise treatment. The alternative use in this research is discussed in the next section.
construct the synthetic sample provide a better comparison group for the treatment sample. By weighing the individual contributions of the donor sample, it gives transparency about the relative contribution any donor unit has on the counterfactual situation as well as the similarities between the two. Secondly, the nature of the Basel 3 based CRD does not allow us to build two groups of banks either exposed and not-exposed, nor two group of sufficiently comparable domestic- and foreign-owned banks in the EU. Thirdly, the SCM allows us to work with a relatively small sample of banks (Abadie et al., 2010). Next, the SCM is introduced following Abadie et al. (2010) and Kreif et al (2015) to illustrate how the SCM is applied to the effect of foreign ownership on the introduction of new capital regulation.
For a sample of J+1 units (in this case banks) the first unit is considered domestically owned (j=1) while the other units are foreign owned (donor group). Observations take place over T periods where the effect of treatment by new capital regulation starts at T0+1. The observed outcome vector of each bank over time is seen as (Yj = Yj1 …YjT0 …YjT)’. The observation 𝑌-. can be seen as:
𝑌-.= 𝑌-. / + 𝛼 -.𝐷-.
Equation 1
Here 𝑌-. / is the observed outcome for the bank if it were domestically owned plus the
treatment effect of 𝛼-.. 𝐷-. is the dummy variable that takes the value of 1 if the bank is foreign owned, a foreign-owned bank therefor sees an effect of being foreign. A domestic bank does not see this effect. In a sample where one unit is treated, 𝛼-.𝐷-. is the effect of treatment compared to non-treated units. The first bank in this hypothetical sample is under consideration, therefor the estimation of 𝛼2. is:
𝛼2. = 𝑌2. − 𝑌2. /
Equation 2
𝑌2.is observed as in the sample, 𝑌2. / needs to be estimated. This estimation is done using
the weighted combination of bank characteristics in the donor group to form a counterfactual outcome for the subject group denoted as 𝑌2. /. The counterfactual estimation is given by:
𝑌-./= 𝛿
.+ θ.Z-+ λ.µ- + 𝜀-.
Equation 3
In this equation 𝛿. represents the fixed effect at time t; θ.represents the time variant coefficient of which Z- is the factor of time-invariant predictors measured; µ- is the vector of time-independent unobserved predictor variables with the time varying coefficient of λ.; 𝜀-. are unobserved factors with a mean of 0.
The outcome of 𝑌2. / prior to the subject period of new capital regulation is observed,
𝑤 -A?2 -BC 𝑌-. = 𝛿.+ 𝜃. 𝑤-𝑍 -A?2 -BC + 𝜆. 𝑤-𝜇- A?2 -BC + 𝑤-𝜀-. A?2 -BC Equation 4
The counterfactual estimator is then constructed as the combination of the observed outcomes in the control banks with 𝑌-. and 𝑍- as:
𝑌2./ = 𝑤 -A?2 -BC 𝑌-. Equation 5
The treatment effect between foreign and domestic ownership can be estimated by: 𝛼2. = 𝑌2.− 𝑌2./
Equation 6
If the number of pre-CRD observations is large relative to the amount of idiosyncratic errors the prediction of pre-intervention outcomes make time-varying responses of the unobserved predictor variables similar between treated and control group. In this case 𝛼2.
would be an approximate and unbiased estimator of 𝛼2. (Abadie et al., 2010).
In choosing the optimum donor group the task is set to determine W*. W* is obtained by minimizing the distance between the predictor measures prior to the introduction of CRD, in the analysis this can be seen as a measure of fit and denoted as the root mean squared predictor error (MRSPE):
𝑚𝑖𝑛 𝑋2− 𝑋K𝑊 L𝑉 𝑋
2 − 𝑋K𝑊
Equation 7
Here, 𝑋2 is a vector (K * 1) of pre-CRD values of covariates K and the outcomes for the treated bank. 𝑋K is a matrix (K * J) with the variables of 𝑋2 for the combination the number (J) of control banks. V is a matrix (K * K) that assigns weight according to the relative importance of the variables K but can also be determined via a subjective view on the importance of variables.
As noted before, the illustration above is a case where the treatment group consists of 1 bank, while the foreign-owned group in the sample consists of a multitude of banks. Some changes to the method are needed to accommodate the larger treatment group where cases are individually matched to the pre-CRD situation. This should ensure a more statistically significant treatment effect (Kreif et al., 2015). Following Kreif et al. (2015) we have a sample of N1 treated banks and N2 control banks. Modifying the first equation as below:
𝑌. = 𝑌./+ 𝛼 .𝐷.
Here, 𝑌. = RSOTSNOPQOP QOP RS OTS , 𝑌./ = NOP RQ OP RS OTS QOP RS OTS , 𝛼. = RSOTSUOPQOP QOP RS OTS
, and 𝐷. is the indicator for domestic/foreign banks. 𝑠-. represents either the risk or the loan growth for bank j at time-period t. The effect 𝛼-. is of our interest as this is the effect on our foreign-owned group of banks. 𝑌./ can be estimated using the weighted outcomes of banks in the donor group:
𝑌./ = 𝑤W𝑌W.
/X
WB2
Equation 9
𝑤W is chosen to minimize the difference between the predictors for the treatment group and
donor groups in equation 9. As 𝑌. is observed and 𝑌./ is estimated the difference between the
two is the effect of foreignness.
3.3.1 Alternate Use
The synthetic control method is originally designed for analysis of two specific groups; one under a certain treatment (new regulation) and the control group that is not under treatment. The SCM is used to construct a counterfactual situation from the original treatment group with data obtained through the control group. In addition to the original methodological use of the SCM this paper will use the SCM to build a control group under the same treatment as the treatment group but with one very specific different characteristic. Theorizing that a control group sample is equal in all things but treatment, substituting this treatment for an alternate characteristic (foreign or domestic ownership) should also provide an equal statistically significant outcome. Thus, rather than using the SCM to construct a hypothetical non-treated group the SCM is used to construct a group of synthetic, counterfactual domestic-owned banks from a sample of foreign owned banks hereafter known as: synthetic domestic(-owned) banks. One previous SCM research also seemingly violated assumption of non-parallel treatment by comparing multiple treated units for the evaluation of health care policies (Kreif et al., 2015). It also directly compares the effects of treated groups compared to their synthetic controls.
Similarly, this method is a new use of the SCM that has previously been used to construct a counterfactual non-treated synthetic control group. Consequently, the experimental approach used in this paper ought to be verified in its effectiveness, for this reason a control group of banks not affected by CRD or Basel 3 regulation is used on both foreign- and domestic-owned samples and the results compared to the results obtained with the synthetic domestic-owned bank.
4 Descriptive Statistics and Trends
For both the European bank samples as well as the US/Chinese control group data is collected from Orbis on CAR, ROA, loan growth, assets, liquid assets, GDP, and GDP per capita, from Datastream on CDS spreads, and from the bank ownership database of Cleassens & van Horen (2015) on ownership data. Bank level data on CAR and ROA are used to construct bank Z-score, a variable that measures bank risk on the basis of accounting data. Furthermore, assets and liquid assets are used to construct the ratio of liquid assets of banks. This section will give a brief and descriptive overview of the data and will highlight some trends and differences between the sample groups.
4.1 Descriptive Statistics
Table 1 below lists the main descriptive statistics on bank level data on a total of 418 banks with yearly data over a time period of 6 consecutive years between 2011 – 2016. There are still some outliers to be considered in this data where a minor part of observations lies beyond the 1st or 99th percentile of data. As an example, loan growth where the value of the 1st percentile is -20.158% but the smallest value is -76.217%. These extreme values possibly bias further analysis positively or negatively. To prevent biased the observed variables will be winsorized, values beyond the 1st or 99th percentiles are clipped and replaced by the respective values of the 1st and 99th percentile. The only variable with an unacceptable level of skewness is assets, this is common and the solution is to use the natural logarithm. The log of assets does display a normal distribution.
CDS data is available on daily basis but averaged weekly and imported from Datastream for the years 2012 – 2016 for 30 banks. As can be seen in table 1 below CDS spreads are highly skewed and the natural logarithm of CDS must be utilized for further statistical analysis. Furthermore, some banks display extreme CDS values, close inspection reveals that these banks are located in Ireland and Greece. These banks are omitted from the sample prior to analysis as extreme numbers can bias results and limit the statistical significance of this research. Greece and Ireland have suffered from the financial (and
4.2 Trends
To construct a preliminary picture of the development of banks over the sample period this section will give a brief overview of trends that can be observed in the sample. Between 2011 and 2016 the average European bank has seen a decrease in size of 23.7%. The consequences of the financial crisis and the intentions of Basel 3 to decrease the systemic importance and size of banks are a possible cause. The development of banks size can be seen in figure 5. The graph shows the sizes of the foreign, domestic, and total banks in the European sample. In line with theory on foreign bank importance (Claessens & van Horen, 2014) foreign banks are significantly smaller than domestically owned banks. The average size of the domestic banks in our sample is 2.7x larger than that of foreign banks.
The overall trend is a decrease in size but between 2011 and 2012 there has been a minor increase in asset size for domestic banks. The average domestic bank has declined 24% over the sample period while for foreign banks this decrease is slightly smaller at 21.6%. This difference might be due to the fact that entry of banks into foreign markets is selectively determined by market size and potential growth of the host economy.
Variable Obs. Mean Std. Dev. Min. Max. Skewness Z-score 2508 3.711 1.179 -4.693 6.888 -0.878
Loan growth 2508 7.561 14.930 -76.217 100.299 1.888
CAR 2508 9.977 4.110 -2.145 38.687 0.953
ROA 2508 0.736 1.237 -13.519 11.558 -1.041
Assets 2508 1.29e+08 3.46e+08 139764.9 2.80e+09 4.195
log Assets 2508 16.604 1.968 11.848 21.753 0.497
Liquidity 2508 0.134 0.133 0 0.945 1.905
CDS 7830 286.100 415.619 35.460 6289 2.284
log CDS 7830 5.082 0.957 3.568 8.746 0.965 Table 1. Descriptive statistics.
0 50000 100000 150000 200000 250000 300000 2011 2012 2013 2014 2015 2016
Assets in Millions
Size EU Foreign Size EU Domestic Size EU Average
Figure 6 shows the development of the bank Z-scores over the period between 2011 – 2016. An interesting development can be seen in the data for foreign banks, the average Z-score increases until 2014, but decreases in the consecutive years. For domestic banks the annual increase in Z-score is significantly higher between 2014 – 2015. This does point towards an event in 2014 that changed previous trends in bank risk, possibly the introduction of CRD. The Z-scores for both bank types converge after 2014.
5. Empirical Results
5.1 Impact of CRD on EU Banking Sector
To assess if CRD has had a positive impact on the European banking sector the first SCM analysis is performed to gauge the effect of the introduction of the new regulation on the EU banking sector as a whole. A simple comparison between the European average and a control group of US/Chinese banks is inappropriate, average Z-scores for the control group are on higher (1.3x in 2011) than EU bank aggregate and more stable over the sample period, there is however convergence of EU banks’ average risk to that of the control group. Figure (7) plots the data Z-scores for the treatment group of banks under CRD, the control group of banks not under CRD, and the synthetic group that is constructed via the SCM. The latter is the counterfactual situation of European banks would the CRD not have been introduced. As data is based on accounting years, the 2013 bar signals the beginning of the treatment on the 1st of January 2014, thus already having an impact on accounting year 2014. The synthetic control group trend closely matches the EU banks in pre-CRD period between 2011 and 2013 with a maximum gap of 0.0043 and a root mean squared predictor error7 close to 0. After the introduction of CRD the gap increases to 0.139 in 2014 and 0.246 at end-of-year 2016. The higher level of risk post-treatment in the non-CRD control group seems to confirm that CRD
7 The root mean squared predictor error (RMSPE) is a measurement of the SCM that represents the fit of pre-treatment data between treatment and control groups.
2.9 3 3.1 3.2 3.3 3.4 3.5 2011 2012 2013 2014 2015 2016
Z-score
Z-score EU Foreign Z-score EU Domestic Z-score EU Average
has had a negative effect on accounting risk levels of European financial institutions, suggesting that introducing Basel 3 based regulation does make banks safer.
5.2 Analysis of European Banks (Risk)
To answer the question of whether foreign-owned banks react differently than domestically-owned banks to new financial regulation, a second analysis is performed with data on European foreign- and domestically owned banks. Again, for a clear comparison foreign owned banks’ Z-scores must be compared to a synthetic group to determine the effect of foreign ownership on the introduction of CRD. The base group in this analysis is foreign-owned banks falling under CRD, domestic-foreign-owned banks are used to construct the synthetic domestic-owned group. This reference group is constructed with the SCM to represent the
2.5 2.7 2.9 3.1 3.3 3.5 3.7 3.9 4.1 4.3 4.5 2011 2012 2013 2014 2015 2016
Z-scores: EU and US banks
EU Banks Synthetic Control US/Chinese Banks Figure 7. Effect of CRD on European banks overall.
2.9 3 3.1 3.2 3.3 3.4 3.5 2011 2012 2013 2014 2015 2016
Z-scores: Foreign and Domestic Banks
counterfactual situation where our treatment sample is domestically owned. The outcome of this analysis can be seen in figure (8) below. For the pre-CRD period the maximum gap between the synthetic domestic-owned and foreign owned banks’ Z-scores is 0.00638, again with a RMSPE close to 0 (RMSPE = 1.75e-14). In the post-CRD accounting year 2014 Z-scores for foreign-owned banks increase, however, in 2015 and 2016 there is a strong decline. In the counterfactual synthetic domestic-owned group the increase of 2015 is less pronounced, however, the increase is stable throughout the post-CRD period. The results obtained with accounting risk as a measurement for bank risk suggest that the introduction of CRD has led to a significant decrease in risk for domestic-owned banks, but a lesser decrease in risk for foreign-owned banks. The ultimate post-treatment increase for our synthetic domestic-owned group of banks is 2.2% while foreign bank average Z-score decreased 2.39% in 2014 and 2016. The results for the years 2015 and 2016 are significant to a 1% level based on placebo-tests, this is further discussed in section 5.4.1. These results confirm hypothesis 1a that domestic-owned banks are able to decrease risk more in response to new capital regulation than foreign-owned banks can. On the basis of yearly Z-score data we cannot however, confirm hypothesis 1b that domestic banks decrease risk faster in response to capital requirements than foreign owned banks. Furthermore, in our sample foreign-owned banks decreased risk faster in accounting year 2014 than domestic banks, contrasting hypothesis 1b.
5.2.1 Synthetic Domestic Ownership Method Test
To test if the results from section 5.2 are methodologically valid the SCM on bank risk will be repeated in an alternate set-up. This set-up will use the original methodology of the SCM to test the effects of CRD treatment on treatment groups of foreign- and domestic-owned banks. This comparison will be conducted against a synthetic control group that is constructed using our non-CRD sample of Chinese and US banks. When measuring the different effects that CRD has on our samples of foreign- and domestic-owned banks the difference in effects indicates if risk has decreased more for our domestic-owned sample compared to the foreign-owned sample, verifying the results from section 5.2. This approach is broadly in line with Kreif et al. (2015) that also compares effects across treated samples.
behaviour of foreign-owned banks if we use a sample of Chinese and US banks as our control group.
The effects and results of the method in section 5.2 are in line with results of the traditional method displayed in this section, this is graphically observed in figure 9 and in a comparison of the post-CRD effects. This comparison test verifies that alternative use of the SCM to create a synthetic group based on domestic ownership (as explained in section 3.3.1) with a SCM analysis using a non-treated control group of Chinese and US banks. Using the SCM to construct a sample of synthetic domestic-owned banks from a sample of foreign-owned banks yield results consistent with results obtained through a direct comparison of effects on both treated groups using a control group of non-treated units.
5.2.2 Market Risk
For increased accuracy of our results on the effect of CRD on bank risk a second risk based test is performed using CDS, the risk the market attributes to a bank. CDS as a measure of risk is formerly used by Beck et al. (2013), a higher CDS indicates a higher risk bank. A time period of 200 weeks is used due to data restrictions in Stata, CRD introduction takes place in week 105. Results for the foreign-owned group of banks, the domestic-owned group of banks, and the synthetic domestic-owned group are shown in figure 10, the natural logarithm of CDS is used as the dependent variable. In the pre-CRD period the maximum gap is relatively small at 0.17 and the data matches well with a MRSPE of 0.065, this can also be observed in figure 10. In the initial 40 weeks after the introduction of CRD, CDS spreads for the synthetic domestic-owned group are significantly lower than the rates for foreign-owned banks, indicating that the risk attributed to domestic banks by the market is rated lower than the risk on foreign banks in response to CRD. In contrast to the Z-score test, this result is also able to indicate a time-element to the risk adjustment of banks and suggesting a result in line with hypothesis 1b that domestic-owned banks are able to decrease risk faster than foreign-owned banks in the 40 weeks after the introduction of CRD. The results for the initial 40 weeks are
2.9 3 3.1 3.2 3.3 3.4 3.5 2011 2012 2013 2014 2015 2016
Synthetic domestic-ownership verification
CRD Domestic Synthetic Control Domestic CRD Foreign Synthetic Control Foreignmostly significant to a 1% or 5% level with p-values below 0,05. There are outliers in the first 10 weeks after CRD introduction and between week 38 and 42 with p-values above 0,20. The significance of the majority of results indicates that foreign-owned banks are deemed riskier by the market directly after the introduction of new financial regulation. The weekly data used in this analysis allows us to confirm hypothesis 1b that domestic-owned banks can decrease risk faster than foreign-owned banks when looking at market risk.
5.3 Analysis of European Banks (Loan Growth)
The analysis on loan growth is performed using the average for European foreign-owned banks, a group of domestic-foreign-owned European banks, and a SCM constructed synthetic domestic-owned group. The purpose of this analysis is to see if foreign banks cut loans more than domestic banks in the periods after the introduction of new capital regulation. The results of this analysis are shown in figure 11. Prior to the introduction of CRD there are some notable differences between foreign- and domestic-owned banks. Pre-CRD in 2011 and 2012 loan growth is significantly higher for domestic-owned banks compared to foreign-owned banks. For the final pre-CRD year (2013) the degree of growth is comparable between the two groups but a direct comparison between the two is insufficient. The counterfactual synthetic domestic-owned banks match the foreign-domestic-owned bank average for the pre-CRD period accurately with a maximum gap of 0.065 and a MRSPE of 5.05e-15.
4 4.5 5 5.5 6 6.5 2012w 1 2012w 8 2012w 15 2012w 22 2012w 29 2012w 36 2012w 43 2012w 50 2013w 5 2013w 12 2013w 19 2013w 26 2013w 33 2013w 40 2013w 47 2014w 2 2014w 9 2014w 16 2014w 23 2014w 30 2014w 37 2014w 44 2014w 51 2015w 6 2015w 13 2015w 20 2015w 27 2015w 34 2015w 41
CDS: Foreign-owned and Domestic-owned
Foreign-owned Synthetic Control Domestic-owned
Following the introduction of CRD the gap between foreign-owned banks and the synthetic domestic-owned increase to 1.98% points in accounting year 2014. It seems that while domestic-owned banks are able to increase loans significantly in this period, loan growth of banks under foreign ownership stagnates at a 1.54% level. However, after this initial drop growth is restored in 2015. Both the foreign-owned bank sample and the synthetic sample see a strong decline in growth rates over 2016. The initial drop in loan growth over 2014 suggests that foreign owned banks do indeed react differently to new capital regulation by decreasing new loans more than domestic-owned banks to adjust to new capital requirements seemingly confirming hypothesis 2. The results are however highly driven by chance and no post-CRD results indicate sufficient significance to reject the null-hypothesis that foreign-ownership affects lending behavior in banks, this is further discussed in section 5.4.2.
6. Robustness Testing
To verify the results obtained in the previous section the studies are repeated using individual data of banks rather than averages and a placebo study is performed. The individual synthetic method assigns domestic ownership to individual control banks to make results more accurate. The placebo study estimate if the results are driven by chance and is used in various other studies to evaluate the significance of results obtained with the SCM (Abadie et al., 2010, 2015; Abadie & Gardeazabal, 2003; Kreif et al., 2015). To perform a placebo study synthetic estimates are obtained for individual banks in the donor sample of domestic banks and then subject to the SCM by the remaining banks in the donor sample. In each placebo estimate the synthetic domestic ownership under CRD is reiterated on one of the donor sample units. In other words, we treat each of the other banks as if it were the foreign-owned observation of interest and estimate a synthetic version. The distribution of placebo effects can subsequently be compared to the effect of our sample of European foreign-owned banks relative to the synthetic domestic-owned banks.
-0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 2011 2012 2013 2014 2015 2016
Loan Growth: Foreign and Domestic Banks
When evaluating the effect of foreign-ownership under CRD a small cumulative difference between the placebo group effect distribution and the synthetic domestic effect indicates that the estimated effect falls inside the distribution of placebos. This then indicates that the estimated effect of foreign-ownership under CRD is insignificant. Estimations well outside of the placebo effect distribution indicate a more significant effect. The p-value operationalizes this fit within the placebo distribution and is constructed by calculating the fraction of placebo effects that is greater or equal to the effect of synthetic domestic ownership under CRD (Abadie et al., 2015). A higher p-value signals a lower significance of results as the probability of them driven by chance increases.
The next sections not only build the significance of the results but also uses individual treatment of the foreign-owned sample banks rather than averages. This multi-treatment of units does not allow for the graphic display of individual placebo estimates but does generate the p-values representing the probability of results driven by chance. Furthermore, the average effect between the individual units’ outcome and its synthetic domestic-ownership situation is displayed for each post CRD time period. The final section also employs a leave-one-out robustness test to test if synthetic results are driven particularly by particular banks from the donor sample.
5.4.1 Bank Risk
Figure 12 shows the effect of CRD on bank Z-score as the average effect per individual foreign bank and the p-value off the effect taking place by chance. The effect of CRD is in line with the effect seen in section 5.2 where there is a slight initial positive effect in the first year of treatment, followed by a more severe negative effect in the final two years. However, the corresponding 2014 p-value signals that it is likely (31%) that the effect is not driven by the introduction of CRD but by other factors. For the other treatment years, the negative effect on foreign-owned banks is significant with p-levels signaling a very small segment of placebo effects is as large as the main effect. This confirms that there is a negative impact of CRD on the Z-scores of foreign-owned banks compared to the synthetic domestic-owned group. There is a 30% probability that the initial increase in 2014 is driven by chance rather than CRD.
5.4.2 Bank Market Risk
Figure 13 shows the effect of foreign-ownership under CRD on the CDS scores of banks after the introduction of CRD. The effect of foreign-ownership is consistent with the results from section 5.2.2. Overall foreign-owned banks have higher a CDS than domestic-owned banks in the initial 40 weeks after CRD introduction. The p-values denote the degree of chance involved in the obtained results. This shows that the significance of the results between weeks 10 and 40 is very high with most values at a 1% or 5% confidence interval. Moving further into the post-CRD period of the analysis results become less significant, this could be explained by a higher degree of divergence in the donor group or unobserved drivers of CDS scores. To conclude, the results of the analysis in section 5.2.2 are robust for the initial 40 weeks where domestic-owned banks decrease risk faster than foreign-owned banks. This is in line with hypothesis 1b.
5.4.3 Loan Growth
Figure 13 shows the treatment effect of CRD on loan growth as the average treatment per individual foreign bank and the p-value off the effect taking place by chance. The effects measured in the multiple-treated unit analysis are not in line with the effects found in section 5.3 where the average banks’ loan growth is used as the treated unit. The findings of this analysis suggest that the effect on loan growth is positive for foreign-owned banks between 0.40% points and 1.12% points over domestic-owned banks. The p-values signal that the chance that this effect is not caused by CRD is extremely high for all post-treatment years with
Figure 10. CRD effect on market risk (CDS).
values between 68% – 93%. This suggests that the results obtained for loan growth are driven by chance, not CRD and that we cannot confirm hypothesis 2.
5.4.4 Leave-one-out test
As the synthetic domestic-owned bank group is constructed with a sample of domestic “control” banks there is a risk that effects in previous tests are driven by a minority of banks in our donor pool of domestic-owned banks. To assess this effect a number of SCM tests are performed, each time excluding a bank with high weight from the sample (Abadie et al., 2015; Kreif et al., 2015).
To examine the results for foreign-owned bank risk the leave-one-out test is performed on the 10 domestic-owned banks with the highest weights in our sample. The results of these tests are seen in figure 14 below. There are two sample banks that overestimate results for the synthetic domestic-ownership relative to our original sample of 160 domestic-owned banks where Z-score increases to 3,49 and 3,47 respectively. Leaving out one of the remaining 8 banks results in a slight Z-score decrease in the synthetic domestic-owned banks. The absence of dramatic in- or decreases in the leave-one-out test indicate that the results are not driven by one or a small number of donor banks. This suggests that the results in section 5.2 are robust to absence of individual domestic-owned banks.
