Volume 23 Issue 4 Thematic Issue: The Economic and Social Impacts of Covid-19 Article 1 12-2021 Tightening and Loosening of Macroprudential Policy, Its Effects Tightening and Loosening of Macroprudential Policy, Its Effects on Credit Growth and Implications for the COVID-19 Crisis on Credit Growth and Implications for the COVID-19 Crisis Aida Ć ehajić University of Ljubljana, School of Economics and Business, PhD Student, Ljubljana, Slovenia, aida.mujnovic@gmail.com Marko Koš ak University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia, marko.kosak@ef.uni-lj.si Follow this and additional works at: https://www.ebrjournal.net/home Part of the Finance and Financial Management Commons Recommended Citation Recommended Citation Ć ehajić , A., & Koš ak, M. (2021). Tightening and Loosening of Macroprudential Policy, Its Effects on Credit Growth and Implications for the COVID-19 Crisis. Economic and Business Review, 23(4), 207-233. https://doi.org/10.15458/2335-4216.1293 This Original Article is brought to you for free and open access by Economic and Business Review. It has been accepted for inclusion in Economic and Business Review by an authorized editor of Economic and Business Review. ORIGINAL ARTICLE TighteningandLooseningofMacroprudentialPolicy, its Effects on Credit Growth and Implications for the COVID-19 Crisis Aida Cehajic a, *, Marko Kosak b a University of Ljubljana, School of Economics and Business, PhD Student, Ljubljana, Slovenia b University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia Abstract In this study, we analyze the effects of macroprudential measures on bank lending in the European Union. We develop several dedicated macroprudential policy indices reflecting different policy actions taken by the authorities in individual member countries, with the aim to affect credit activity in national banking sectors. In our empirical model, wemeasureresponsiveness ofgrossloans inbankstoselectedmacroprudential policyindices,takingintoaccountaset of bank level and macroeconomic control variables. We use the Fitch Connect bank level dataset with financial state- mentsfor3434Europeanbankswith18,616observationsandmacroeconomicdataprovidedbytheWorldBankandIMF statistics covering the period between 2000 and 2017. Information on the use of macroprudential instruments is taken from a new macroprudential policy database, MaPPED, gathered and published by European Central Bank, where we were able to extract the information on both timing and the direction of use of the macroprudential policy instruments. Ourfindingsshowthatmacroprudentialinstrumentscanbeusedeffectivelyforregulatorymodulationofcreditactivity in banks, with some fluctuations in the level of the effectiveness through the business cycles. Therefore, in loosening cycles, macroprudential measures are found to be strongly and positively associated with bank lending. On the other side, tightening actions are found to have a downward effect on bank lending, while these effects are less pronounced. These results are of great importance in the current crisis arising from the impact of COVID-19, as policymakers are trying to support the economy by easing macroprudential regulatory constraints to ensure lending to the real sector. Keywords: Macroprudential policy, Bank lending, Credit growth, Financial stability, Credit cycles JEL classification: E58, G21, G28 Introduction F inancial stability concerns have been the domi- nantthemeofmanyresearchpapersandregula- tory discussions, particularly in the years following the2008globalfinancialcrisis.Thefocushasbeenon establishingsystemicregulatorymechanismstopro- tect the entire financial sector from the risks arising from the interconnectedness of financial institutions and their procyclical behavior. The role of systemic surveillance and stability has been assigned to mac- roprudential policy. In the current period of the Covid-19 pandemic crisis, macroprudential policy is onceagainatthecenterofacademicandinstitutional discussions regarding its effectiveness in mitigating the worsening effects of the pandemic-related crisis on investments and economic growth. The recent crisis has led to a significant decline in bank share prices,yetthecrisishasnotdestabilizedthefinancial sector. This could be a result of a shift in financial regulation, characterized by the introduction of strictercapitaldemandsandvariousmacroprudential tools following the 2008 financial crisis, (Borri & Giorgio,2021). In the period leading up to the 2008 financial crisis, it was believed that microprudential mea- Received 29 June 2020; accepted 7 May 2021. Available online 23 December 2021. * Corresponding author. E-mail addresses: aida.mujnovic@gmail.com (A. Cehajic), marko.kosak@ef.uni-lj.si (M. Kosak). https://doi.org/10.15458/85451.1293 2335-4216/© 2021 School of Economics and Business University of Ljubljana. This is an open access article under the CC-BY-NC-ND license (http://creativecommons. org/licenses/by-nc-nd/4.0/). sures were sufficient to ensure financial stability of the financial sector. Microprudential measures aim to ensure the safety of individual financial in- stitutions and counter idiosyncratic risks, while macroprudential measures aim to protect the entire financial system by taking into account the in- teractions and interdependencies of individual financial institutions and systemic vulnerabilities, while limiting the collective exposure to system- wide risks (Aikman et al., 2015). The shift from the micro to the macro perspective called for a calibra- tion of macroprudential instruments through two dimensions: cross-sectional and time dimension. The cross-sectional dimension of macroprudential policy requires a broader regulatory scope and the assessment of the system-wide importance of some institutions. On the other hand, the time dimension requires that macroprudential buffers be tightened in good times and relaxed in downturns, which represents the so-called “countercyclical” approach. This approach could increase the probability of survival of institutions as well as access to credit for the real economy, while reducing procyclical behavior in the financial system (Borio, 2003). Macroprudential measures aim to address vul- nerabilities in the financial sector that could trigger the development of systemic risk. As suggested by Yellen (2011), the risks manifest themselves in cyclical and structural forms, either through “too big to fail” concerns or through credit booms and soar- ing asset prices. Yellen (2011) also argues that some policies address certain risks that are present in all economic conditions, while cyclical risks should be addressedwithcountercyclicaltools,withthegoalof maintaining financial stability while ensuring that economic development can continue unimpeded. Countercyclicalcapitalrequirementsimposedbythe macroprudential regulator in good times can serve as an additional safety buffer in bad times. The dy- namic nature of macroprudential instruments could contain banks’ procyclicality more efficiently and reduce government spending in times of crisis (Jimenez et al., 2017). Given the current crisis that has arisen as a result of the Covid-19 pandemic, the effectiveness of mac- roprudential policy is of momentous importance. The crisis should demonstrate the effectiveness of countercyclical design, as policymakers loosen built- up macroprudentialbuffers toallow credittoflowto therealeconomy(Araujoetal.,2020;Nakatani,2020). Thiscallsforfurther evidenceontheeffectivenessof macroprudential policy, in particular on its easing cycles and their impact on the availability of credit. The aim of this paper is to investigate the impact of macroprudential policy on bank lending, using a sample of 3434 European banks spanning over 28 EU countries and covering the period between 2000 and 2017. This study contributes to the literature on financial regulation in several ways. First, to our knowledge, this is the first study to examine the effects of macroprudential policy on bank lending, using the recently collected and published MaPPED database covering all EU member countries. 1 This database is different with respect to the existing datasets provided by Cerutti et al. (2017), which captures only presence of the instrument, or Lim et al. (2011), which captures the direction of tools, but with limited number of macroprudential in- struments covered. MaPPED contains the informa- tion on a larger number of tools by covering more than 50 policy instruments and recording their activation, tightening, loosening or deactivation, thus providing complete information on the life of the instrument. Second, we study these effects by using bank-level data that allow us to control for bank heterogeneity and the response of bank sub- sidiaries operating in different countries. By using bank-leveldata,wealsoreducethesensitivityofour analysis to endogeneity biases associated with the introduction of macroprudential measures. 2 Third, wearealsoabletoanalyzetighteningandloosening actions of macroprudential policy separately, as the macroprudential policy database provides this in- formation in full. This allows us to observe the life cycle of macroprudential policy instruments, the effects of activation, tightening, easing, and deacti- vation of the instruments, and how the instruments work over the cycle. Fourth, given the information onthe objectiveofmacroprudential policies andthe intention of their activation, we also examine the effectiveness of different groups of macroprudential instruments based on their respective objective. Following Altunbas et al. (2018), we design our main macroprudential measures by summing all policy changes over time, both tightening and easing. This allows us to capture the overall mac- roprudential stance in a given country and time period. A higher value of the index indicates a tighter stance, while a lower value of the index in- dicates a looser macroprudential policy stance. To investigate whether there is an asymmetric effect of tightening and easing measures,we form additional 1 Macroprudential Policies Evaluation Database, available at: https://www.ecb.europa.eu/pub/research/working-papers/html/mapped.en.html. 2 See Morgan et al. (2018); Claessens et al. (2013). 208 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 indices that capture only tightening or only loos- ening actions. Furthermore, we measure the effec- tiveness of different types of macroprudential instruments based on their initial target and objec- tive. This approach has also been followed by other studies in the literature (see Cerutti et al., 2017; Meuleman & Vander Vennet, 2020; Olszak et al., 2018). In our section on robustness checks, we additionally perform several tests by examining the effectiveness of groups of macroprudential in- struments partitioned by their economic objective. To observe the relationship between macro- prudential instruments and bank lending in different periods of the financial cycle, we run several estimations for the pre-crisis, crisis, and post-crisis periods. Finally, we also test whether macroprudential measures have stronger effects on a subsample of listed banks. To account for endogeneity biases associated with the introduction of macroprudential policies, we conduct the empirical analysis using the Arellano- Bond generalized method of moments estimator corresponding to our small T and long N panel. 3 This estimator has become established in the liter- ature as a standard tool for the analysis of macro- prudential measures (see Altunbas et al., 2018; Cerutti et al.,2017; Morgan et al., 2018; Olszak et al., 2018; Zhang & Zoli, 2016). Our results show that macroprudential measures are significantly associated with credit cycle move- ments. Macroprudential indices are significantly and negatively associated with bank lending when controlling for different bank and macroeconomic variables. This indicates that the macroprudential policy framework is successful in curbing credit growth and limiting excessive bank lending. When we test tightening and loosening actions separately, we find that macroprudential loosening actions have a stronger effect on bank behavior, i.e. when macroprudential measures are relaxed or deacti- vated,wecan expectbanks toincreasetheirlending activity. Testing various indices of macroprudential policy, we find similar results, finding that credit- related macroprudential measures are most effec- tive in curbing excessive lending. When we test macroprudential easing measures, we find that capital- and credit-related instruments are posi- tivelyassociatedandhavethestrongestrelationship withcreditactivity.Theresultssuggestthatrelaxing macroprudential tools during downturns, such as the current one caused by the Covid-19 pandemic, can successfully improve market liquidity and sup- port monetary policy efforts to stabilize credit sup- ply in order to promote investment and economic growth. Nevertheless, various macroprudential measures still in place could alleviate concerns about deteriorating bank asset quality and keep bank capital from falling to dangerous levels. The rest of the paper is structured as follows. The nextpartrepresents the overviewofexistingstudies in the field. In section 2, we describe data, meth- odological approach in the paper and its findings. The section 3 represents extensions and additional robustness analysis, while the last, 4th, section concludes the paper. 1 Existing literature and hypotheses development The literature on macroprudential policy can be divided into three distinct categories: cross-country macro studies and single-country studies, both of which cover most of the existing macroprudential literature, and, to a lesser extent, those studies that combine macro- and firm-level data. Some of the most important contributions to the fieldhavebeenmacro-levelstudies,whichhavealso provided the first datasets on macroprudential in- struments in different countries. Such is the study by Lim et al. (2011), which uses a sample of 49 countries between 2000 and 2010 to show the effectiveness of macroprudential instruments in containing credit and debt financing and thus limiting the manifestation of systemic risks. In a study by Cerutti et al. (2017), which employs the extensive macroprudential database for 119 countries during the period 2000e2013, the authors analyzed the effectiveness of macroprudential in- struments in reducing credit growth and house prices. The paper shows that loan-to-value and debt-to-income limits, dynamic provisioning, and leverage ratio limits are most effective in curbing excessive lending. The effects were found to be stronger in upturns, while macroprudential tools were found to be less effective in downturns. As the authors pointed out, the impact of macroprudential measures depends on the level of economic devel- opment and capital openness of countries, and macroprudential measures were found to have a weaker relationship with credit growth in more open and financially developed economies than in developing countries. 3 Countriesmighttakemacroprudentialmeasuresinresponsetoexcessivecredit activityorsystemicriskconcerns.Thismayleadtoreversecausalityin our dependent variable. However, using bank-level data makes macroprudential policy studies less prone to this bias, as it is unlikely that regulators' decisions to apply macroprudential tools depend on the determinants of individual banks. It is more likely that these decisions depend on aggregate macroeconomic conditions (see Cerutti et al., 2017; Morgan et al., 2018). ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 209 Vandenbussche et al. (2013) found that reserve requirements associated with credit growth and restrictions on foreign-based financing have a strong impact on house price appreciation. Crowe et al. (2013) established that macroprudential tools such as loan-to-value limits or dynamic provision- ing are effective in dampening house prices. A paper by Bruno et al. (2017), which investigates the effects of macroprudential instruments and capital flowmanagementmeasuresin12countriesoverthe period 2004e2013, shows that capital flow manage- ment measures are associated with reductions in bankinflows,whilealsoprovidingsomeevidenceof increased effectiveness of macroprudential policy when implemented in a complementary manner to monetary policy measures. Dell' Ariccia et al. (2012) foundthatmacroprudentialpolicycouldbeeffective inreducingtheimpactandcontainingcreditbooms, while the authors also caution against possible circumvention and avoidance of these policies and the need for coordination with other macroeco- nomic policies. Some of the studies combining both macro and micro data have examined the impact of macro- prudential measures on overall bank risk. For example, a study by Claessens et al. (2013) confirmed that macroprudential measures are effective in curbing bank leverage and asset and non-core liability growth during the upswings. In a study by Altunbas et al. (2018), it is shown that macroprudential measures are significantly associ- atedwithbanks'riskexposureandthattheeffectsof these measures strongly depend on banks’ charac- teristics. The authors also suggest that a tightening ofmacroprudentialmeasureshasstrongereffectson bank risk. Zhang and Zoli (2016) examine whether macroprudential measures have an impact on financialstabilityin13Asianeconomies.Theresults show that borrower-based macroprudential mea- sures and tax measures are effective in stabilizing property prices, lending, and bank leverage. Morgan et al. (2018) analyze the effects of a spe- cific macroprudential measure, the loan-to-value ratio, introduced in many countries over the period 2000e2013, using a panel of 4000 banks. The results suggest that loan-to-value ratios strongly affect banks’ mortgage lending, while the impact is less pronounced for larger banks and banks with a high share of non-performing loans. They also suggest that the introduction of LTV should be followed by othermacroprudentialmeasuresthatcouldenhance and complement its effects. Igan and Kang (2011) analyze whether loan-to-value and debt-to-income limits dampen house prices and stabilize the hous- ing market. Their results show that after the activation of borrower-targeting macroprudential measures, a significant stabilization effect on prices is observed, while their activation, especially loan- to-value limits, also curbs expectations and specu- lative incentives. Jimenez et al. (2017) examine the effects of mac- roprudentialmeasuresontheprocyclicalityofcredit and the supply of credit to firms. Over the period 2000e2013, the authors study dynamic provisioning in Spain and find that this instrument constrains credit procyclicality and positively affects the credit supply offirms in bad times. The effects are weaker in good times and may also cause an increase in bank risk induced by the demand for higher profits. The paper shows the beneficial effects of increasing capital requirements in good times, which stabilizes credit supply in bad times. InastudybyAkinciandOlmstead-Rumsey(2018), covering 57 countries, macroprudential policies are found to be effective in mitigating excessive lending and house prices. The most effective tools are those that target borrowers, such as loan-to-value and debt-to-income limits. In a recent paper by Gamba- cortaandMurcia(2019),theauthorsusecentralbank lending data from five Latin American countries to examine the impact of macroprudential policies on banklendingandfindthatmacroprudentialpolicies are effective in mitigating the procyclicality of lending, while the impact is amplified when com- binedwithtightmonetarypolicy. In summary, previous literature has shown that macroprudential measures are associated with restrained credit growth (Akinci & Olmstead-Rum- sey,2018;Ceruttietal.,2017;Gambacorta& Murcia, 2019; Lim et al., 2011; Morgan et al., 2018). To test whether macroprudential instruments have a miti- gating effect on lending in the banking sector in European Union, we develop our first hypothesis using the macroprudential policy database (Budnik & Kleibl, 2018) with extensive information on over 50 policy instruments. H1. Macroprudential policy measures are significantly and negatively associated with bank lending volumes. We test this hypothesis by examining the overall impact of macroprudential policy stance in a coun- try on banking sector lending, as captured by the aggregate macroprudential indices. Since the database captures all activations, tighten- ings and loosenings of all macroprudential in- struments, we can test the effectiveness of these policy actions separately. Since macroprudential instruments were mostly introduced as tightening 210 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 measures during the period under consideration, requiring banks to either raise more capital, restrict lendingtoborrowers,orholdmoreliquidassets,we expect a negative relationship between the macro- prudential policy index and lending. On the other hand,ifmacroprudentialmeasuresareloosened,we expect a positive relationship with credit growth, as in Poghosyan (2020). Accordingly, we develop the following hypotheses. H2a. Tightening policy actions of macropr- udential policy are negatively associated with bank lending. H2b. Bank lending activity increases in response to macroprudential policy easing. We test these hypotheses by studying the relation- ship between indices that capture only tightening actions or only loosening actions and bank lending. Following previous evidence (Cerutti et al., 2017; Morgan et al., 2018) and considering the goal of macroprudential tools to directly target the credit market and limit excessive lending and overheating in the mortgage market, we expect credit related instruments to be most effective in curbing credit growth. We thus develop the fourth hypothesis. H3. Credit-related macroprudential measures have the strongest effects in curbing bank lending. 2 Data and methodology To conduct our analysis, we combine bank-level and country-level data obtained from different da- tabases. For macroprudential data, we use the recently published database, MaPPED, collected by ECB researchers and national authorities (Budnik & Kleibl, 2018). The database is based on a survey conductedwiththehelpofnationalauthoritiesinEU countries.MaPPEDprovidesdataonprudentialand macroprudentialmeasurescovering1700actionsand 53prudentialinstrumentsgroupedinto11categories accordingtotheirobjective.Thedatabaseshowsthe life span of the instrument: its introduction, tight- ening or loosening actions, and deactivation of the instrument. The database also captures some policy measures that are ambiguous in nature, which we exclude from our analysis because we want to cap- ture the accurate direction of the policy. We also exclude policies that were introduced as recom- mendationsbytheregulator,sothatouranalysisonly capturesbindingmeasuresforbanks.Weareableto construct different indices for prudential measures, eithertheoveralldirectionofmacroprudentialpolicy inacountryinagivenperiodorindicesbasedonthe direction of the measures, i.e. whether the macro- prudential measure was introduced as a tightening or easing tool. 4 In addition, the database offers the possibility to measure the effectiveness of macro- prudential instrumentsbasedontheirobjective. 2.1 Macroprudential policy measurement design The MaPPED database is based on a survey designed to capture the majority of regulatory ac- tions over time. It was conducted with the help of the central banks of the participating countries and other regulatory and supervisory institutions. The policy instruments included in the database are macroprudential or have a macroprudential char- acter, meaning that they affect the entire financial sector.MaPPEDdescribesacompletelifeofapolicy instrument, distinguishing three possible responses to the question on changes and policy actions: a) tightening, b) loosening, and c) other and with ambiguous effect. We capture these changes numerically by assigning 1 to tightening policy ac- tions, 1 to loosening policy actions, and 0 to no change or actions with other and ambiguous effects (see Altunbas et al., 2018). This information is captured in the database on a monthly and quar- terlybasis,andsinceourbank-leveldataisavailable on an annual basis, we capture policy changes across quarters and then sum them to obtain annualized information. Using this approach, we obtain an annual policy stance with a positive or negative sign, while we can also obtain 0, if the summed measures cancel each other out or if there were no changes in the respective period. Given the design and construction ofthe MaPPED database, we form six indices of macroprudential stance. First, we sum all annualized indicators of macroprudentialpolicyactionsacrossallcountriesin the sample consisting of all macroprudential mea- sures and form the index MPP. We construct addi- tional indices as an alternative measure of the macroprudential stance or a measure specifically designedtocaptureeitherperiodsofpolicytightening or easing. Finally, we also construct various indices basedonthetargetofthesemeasures. 4 To develop these indices, we follow several approaches from the literature (see Akinci & Olmstead-Rumsey, 2018; Altunbas et al., 2018; Garcia Revelo et al., 2020; Lim et al., 2011). ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 211 The MPP index represents the sum of all policy actions over macroprudential instruments recorded in the database. The index can take values from 4 to9withinoursampleandtimecoverage.Thelower value of the index indicates a looser macro- prudential policy stance, or simply put, more mac- roprudential policy easing actions were activated during this period. A higher value of the index in- dicates tighter macroprudential policy, with more tightening policy actions. The design of the index is given below: MPP k;t ¼ X j X j;k;t ð1Þ X j;k;t represents each policy action attached to the respective macroprudential instrument in country k and time t. As this index represents the sum of all policy actions: tightening and easing, higher value of this measurement represents macroprudential policy with a tightening stance. On the contrary, a lower or negative value indicates looser macro- prudential policy stance within a given period. In order to test the robustness of our results to alternative kind of measurement, we design our second index MPP2 similarly as the former, but in this case the index isbounded and can take value of 1, 0 and 1 (for an application see Garcia Revelo et al., 2020). It is constructed based on the overall macroprudentialstanceandorientationofthepolicy incountriesoverdifferentperiods.Theconstruction of the index is shown below: MPP2 k;t ¼ 8 > > > > > < > > > > > : 1 if X j X j;k;t >0 0 if X j X j;k;t ¼0 1 if X j X j;k;t <0 ð2Þ where MPP2 k;t ¼ {-1,1}. It is equal to 1, if the stance of macroprudential policy in the respective country kandyeartisoftighteningnature,whichmeansthe P j X j;k;t is higher than 0 and indicates restrictive macroprudential stance. From this, it follows that the number of tightenings exceeds the number of loosenings. Otherwise, when macroprudential stance in a country k and time t is of loosening na- tureandmorerelaxed, P j X j;k;t islowerthan0,which means there have been more loosening macro- prudential actions in the respective period. Additionally, we construct several indices which serve as an additional measurement in order to pro- videaccuratenessandrobustnessofourresults,while minimizing the limitations of our initial indices. The limitationsofMPPandMPP2arereflectedinthefact that these measures take into account all policy ac- tions, tightenings and loosenings of all instruments, while they do not distinguish the direction of the policies. Meanwhile, due to design of the indices, some of these policy actions, when summed up, cancel each other out, and their effect is neutralized. To account for separate effects of tightening and loosening actions of macroprudential policy is rele- vantandrepresentsamajorcontributionofthepaper. With thenext four indices, we aim to measure the effects of tightenings and loosenings of macro- prudential policy separately and to distinguish be- tween these two directions of policy actions. Our third macroprudential index, MPP3, takes into ac- count the macroprudential stance, and is equal to 1, if the sum of all policy actions is higher than 0, which reflects a tightened macroprudential stance. The index design is as follows: MPP3 k;t ¼ 8 > < > : 1 if X j X j;k;t >0 0 if X j X j;k;t 0 ð3Þ where MPP3 k;t ¼{0,1}.Itisequalto1,ifthestanceof macroprudential policy in the respective country k and year t is of tightening nature, which means the P j X j;k;t is higher than 0 and indicates restrictive macroprudential stance. If P j X j;k;t is equal to 0, or lower than 0, the index takes the value of 0. Our next index, MPP4, takes into account loos- ening stance of macroprudential policy in the respective period. MPP4 k;t ¼ 8 > < > : 1 if X j X j;k;t <0 0 if X j X j;k;t 0 ð4Þ where MPP4 k;t ¼{0,1}.Itisequalto1,ifthestanceof macroprudential policy in the respective country k and year t is of loosening nature, which means the P j X j;k;t is lower than 0 and indicates loosened mac- roprudentialstance.If P j X j;k;t isequalto0,orhigher than 0, the index takes the value of 0. However, the potential limitation of our previous indices, MPP3 and MPP4, whose values lie between 0 and 1, is that they do not weight the macro- prudential stance by considering the number of tightening or loosening policy actions in a certain 212 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 period. One possibility to measure how restrictive macroprudential policy is during a given period in timeistotakethesumofalltighteningpolicyactions, leaving out 0 for all loosening policy actions and if thereisnochange.Theindexdesignisgivenbelow: MPP5 k;t ¼ X r X r;k;t ð5Þ where MPP5 k;t corresponds to the sum of all restrictive (tightening) policy actions in the respec- tive country k and year t. The higher the value of this index, the more restrictive macroprudential stance. On the contrary, a lower value indicates less restrictive macroprudential policy stance within a certain period. We do the same to measure loosening macro- prudential policy stance, as we take into account only loosening policy actions with our last index, MPP6, whose design is as follows: MPP6 k;t ¼ X l X l;k;t ð6Þ where MPP6 k;t equals the sum of all loosenings in country k in year t. The higher the value of this index, the more relaxed macroprudential policy stance in a given period. 5 To show the design and structure of our macro- prudential measures, we review the exemplary country case studies of Cyprus and Lithuania for 2011 (see Table 1). The table shows the path of our indices design, from recording the policy actions over quarter, annualizing them, and finally, putting together macroprudential indices. 2.2 Bank and country level data Wecollectbank-leveldatafromtheplatformFitch Connect.Thesampleincludesdatafrom3434banks, with balance sheet and income statement data over theperiodbetween2000and2017.Wecoverdatafor all EU member states, including commercial, sav- ings and cooperative banks and both inactive and active banks. 6 We exclude negative values for total assets and gross loans and winsorize all other vari- ablesatthe1%levelatbothendsofthedistribution. The summary statistics of the bank variables can be found in Table 2. The database was checked for doublecountingandallduplicateobservationswere manually removed. All bank statements are annual and denominated in US dollars. In the unbalanced panel, we include all banks with available data on our dependent variable for two consecutive years, with the average bank observed for five periods. Since most macroprudential measures are intro- duced by national authorities, we analyze uncon- solidated bank data to capture the impact on individual bank subsidiaries. To control for macroeconomic differences across countries in the sample, we include several macro- economic variables by drawing on IMF data plat- forms International Financial Statistics, World Bank and national central bank databases. Summary sta- tisticsforcountryvariablescanbefoundinTable3. 7 Table 1. Exemplary design of macroprudential indices based on country cases for Cyprus and Lithuania in 2011. Cyprus Lithuania Macroprudential instruments Tax on assets/ liabilities Single client exposure limits Sector and market segment exposure limits Loan to value Debt Service to Income Maturity and amortization restrictions Policy actions in quarters: Q1:0 Q1:-1 Q1:1 Q1:0 Q1:0 Q1:0 Tightening: (þ1) Q2:1 Q2:0 Q2:0 Q2:0 Q2:0 Q2:0 Loosening: ( 1) Q3:0 Q3:0 Q3:0 Q3:0 Q3:0 Q3:0 Q4:0 Q4:0 Q4:0 Q4:1 Q4:1 Q4:1 Policy actions annualized 1 11 111 Number of tightening actions 2 3 Number of loosening actions 1 0 MPP 1 3 MPP2 1 1 MPP3 1 1 MPP4 0 0 MPP5 2 3 MPP6 1 0 Source: Budnik & Kleibl, 2018: Macroprudential policy Evaluation Database (MaPPED), based on authors' elaboration. 5 Since we code loosening actions with 1 in the process of construction of MPP6 index, we multiply the index with 1 to have a comparable mea- surement similar to MPP5 index. 6 Weincludedallbanksintheperiodcovered,eveniftheywithdrewfromthemarketinagivenyearforvariousreasonssuchasmergerandacquisition activity, or if the banks were liquidated or went bankrupt. This also means that our sample does not suffer from a survival bias (Kosak et al., 2015). 7 For a description of all variables used in the regression analysis and their sources, please see Appendix, Table A2. ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 213 2.3 Empirical model and findings Toexaminetheeffectsofmacroprudentialpolicies on bank loans, we specify the following model. Y j;k;t ¼aY j;k; t 1 þbMaPru k;t 1 þgX j;k;t þdZ k;t 1 þl j þ q t þ3 j;k;t ð7Þ where Y j,k,t is the dependent variable, measured by logarithm of gross loans. Y j;k; t 1 represents the lagged dependent variable included in the right- hand side to account for underlying autoregressive process. MaPru k;t represents the macroprudential index, our variable of interest, which is captured by different measurements. We expect that macro- prudential measures are negatively associated to Table 3. Summary statistics of country variables. Variable Observations Mean Std. Dev. Min Max Macroeconomic variables D Policy rate 18,616 0.182 0.608 10.46 4.5 GDP growth (%) 18,616 0.741 2.167 6.600 10 Credit to private sector (%) 18,554 89.959 28.038 26.280 191.189 Gross capital formation growth (%) 18,616 0.125 6.887 19.465 20.058 NPL gross (%) 15,020 6.348 5.013 0.379 18.064 Inflation rate (%) 18,616 1.707 1.253 0.791 21.458 Macroprudential data Indicese all actions MPP 18,616 0.704 1.216 49 MPP2 18,616 0.323 0.621 11 MPP3 18,616 0.406 0.491 0 1 MPP4 18,616 0.084 0.277 0 1 MPP5 18,616 0.868 1.185 0 9 MPP6 18,616 0.164 0.458 0 6 Indices by target Credit 18,616 0.103 0.513 44 Market liquidity 18,616 0.031 0.338 15 Concentration 18,616 0.044 0.395 33 Resilience 18,616 0.526 0.789 24 Sub-indices by target Credit growth 18,616 0.002 0.140 34 Lending caps 18,616 0.042 0.293 34 Risk weights 18,616 0.059 0.400 22 Liquidity measures 18,616 0.031 0.338 15 Exposures 18,616 0.044 0.395 33 MCR 18,616 0.336 0.695 13 Capital buffers 18,616 0.089 0.348 13 Taxes 18,616 0.027 0.173 12 Provisioning 18,616 0.000 0.118 22 Other requirements 18,616 0.072 0.271 32 Leverage ratio 18,616 0.003 0.050 0 1 Note: Summary statistics is for the period 2000e2017. All variables are lagged for one period, except policy rate which was transformed with first difference. All variables apart from indicator variables are winsorized 1% on both tails of the distribution. Table 2. Summary statistics of bank variables. Variable Observations Mean Std. Dev. Min Max Gross loans (USD, millions) 18,616 2, 830 9720 0.93 75,400 Gross loans (logs) 18,616 19.871 1.779 13.743 25.045 Size (total assets, logs) 18,616 20.431 1.721 15.115 25.746 Liquidity ratio (%) 16,186 15.081 20.151 0 94.52 Tier 1 ratio (%) 18,616 16.452 8.532 5.73 66.4 Loans to deposits (logs) 18,430 4.640 0.787 1.188 10.589 Loan loss provisioning (%) 18,207 0.666 1.464 4.07 15.99 Market share (%) 18,616 0.761 3.848 0.0000142 81.985 Commercial (0e1) 18,616 0.237 0.425 0 1 Savings (0e1) 18,616 0.188 0.391 0 1 Cooperative (0e1) 18,616 0.575 0.494 0 1 Note: Summary statistics reflects the period 2000e2017. All variables, except the categorical ones, are winsorized 1% on both tails of the distribution. 214 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 developments in lending activity (Akinci & Olm- stead-Rumsey,2018;Ceruttietal.,2017;Gambacorta & Murcia, 2019), although some studies have shown positive association of some macroprudential tools (Lim et al., 2011). X j;k;t representsthevectorofbankvariables.Inthe main regressions, bank size, measured by the log- arithm of total assets, and tier 1 capital ratio are included as bank controls. At various stages of the analysis, theliquidityratio,measuredbytheratioof liquid assets to total assets, and the loans to deposit ratio, which shows the funding positionofthe bank, are also included. As an important factor of the bank's credit activity and risk assessment, we addi- tionally test for the relevance of loan loss provisions (LLP), measured by the ratio of loan loss provisions to gross loans. To control for the bank's market power, we include market share, measured by the size of the bank relative to the total assets of the entire banking sector in a country. Finally, we also test for the relevance of the bank's specialization by using dummies indicating whether it is a commer- cial, savings or cooperative bank. Z k;t 1 is avector of macroeconomic variables. All regressions include the change in the policy rate, while at different stages we include the real GDP growth rate, private sector credit as a percentage of GDP, gross capital formation growth, the gross NPL rate, the level of gross non-performing loans as a percentage of a country's GDP, and the inflation rate. l j and q t are bank and time fixed effects, while 3 j;k;t is the error term. Theintroductionofthelaggeddependentvariable on the right-hand side triggers endogeneity prob- lems. According to Nickell (1981), the lagged dependent variableiscorrelatedwith theerrorterm for small T-panel data and in the presence of fixed effects, leading to general biases associated with dynamic panel models. To control for endogeneity, the literature suggests the use of the dynamic generalized method of moments. The choice of this methodisalsoconfirmedbyoursmallTandlargeN sample. Another endogeneity problem arises from the introduction of macroprudential measures in countries with elevated credit activity and systemic risk concerns. Namely, when countries experience an unsteady increase in overall credit activity that is accompanied by financial stability concerns, they are more likely to enforce macroprudential mea- sures. Since we expect a negative association of the implementation of macroprudential measures with credit growth, especially since most macro- prudential measures are introduced as tightening measures, we are concerned that this effect may arise as a result of reverse causality. The GMM estimation method should reduce such endogeneity concerns. Moreover, by using bank-level data, our estimatesarelesspronetoendogeneitythanmacro- Table 4. Comparison of different estimation methods of our initial model. (1) (2) (3) (4) (5) OLS FE GMM one-step GMM two-step GMM two-step Dependent variable (lag) 0.847*** 0.380*** 0.730*** 0.765*** 0.765*** (27.01) (4.36) (7.96) (13.66) (16.20) Size (total assets, logs) 0.142*** 0.648*** 0.255*** 0.226*** 0.226*** (4.45) (4.70) (2.82) (4.03) (4.75) Tier 1 ratio 0.00556*** 0.000537 0.00803*** 0.00698*** 0.00699*** (-4.36) (-0.18) (-3.67) (-5.10) (-5.57) D Policy rate 0.00908 0.00390 0.0103 0.00885 0.00828 (-1.23) (0.60) (-1.59) (-1.55) (-1.40) MPP (lag) ¡0.00759** (-2.01) Observations 18,616 18,616 18,616 18,616 18,616 Instruments 77 77 78 AR (1) 0.039 0.070 0.068 AR (2) 0.166 0.172 0.186 Hansen 0.260 0.260 0.226 Note: The dependent variable is gross loans in logs. OLS (column 1) stands for ordinary least squares, while FE (column 2) stands for panelfixedeffectsestimationmethod.Estimationmethodincolumns3e5isdynamicone-step(3)ortwo-step(4e5)systemgeneralized methodofmoments(GMM)estimatorwithrobuststandarderrorsandWindmeijer'scorrection.Laggeddependentvariableistreatedas endogenous, and all other variables as exogenous. All regressions include year fixed effects. T statistics is reported in the parentheses. All regressions include weights based on the number of observations of each country. Macroeconomic variables are lagged one period, while all variables, apart from indicators, are winsorized 1% on both tails of the distribution. The following are p-values which indicate the significance level of coefficients: *p < 0.10, **p < 0.05, ***p < 0.01. Bold is used as to highlight our variable of interest. ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 215 level studies, as the credit levels of individual banks are less relevant to policymakers when selecting regulatory instruments (see Claessens et al., 2013; Morgan et al., 2018). Finally, we lag our country variables by one period to further reduce endoge- neity problems. Since the summary statistics presented in Table 2 show large variability and differences in our bank variables,such asthe liquidity ratioandtheloansto deposits ratio, which reveal different liquidity and funding strategies of banks within our sample, we are concerned about possible effects of outliers. To addressconcernsaboutoutliers,allvariables,except indicator variables, are winsorized by 1% at both tails ofthedistribution.Sinceour sample consistsof 28 EU countries, with different numbers of banks and observations in the database, we also include weights that give each country equal importance in all our regression estimates. The weights are con- structed as the inverse of the sum of each country's observations. 2.4 Baseline results We start our analysis with the baseline model consisting of four variables on the right-hand side: the lagged dependent variable, the bank's gross loans transformed into the natural logarithm, and other variables at the bank level: size, measured by total assets and transformed into the natural loga- rithm, tier 1 capital ratio, and finally, the change in the policy rate as a country-level determinant. Bank size should show whether large banks lend more due to their size and access to many different forms of funding, while the tier 1 capital ratio shows the bank's overall capital position. The central bank Table 5. The impact of macroprudential policy on bank lendingebank controls. MPP (1) (2) (3) (4) (5) (6) (7) Dependent variable (lag) 0.735*** 0.692*** 0.783*** 0.697*** 0.688*** 0.693*** 0.689*** (12.42) (7.69) (10.47) (7.45) (7.38) (7.63) (7.49) MPP (lag) ¡0.00568 ¡0.00782** ¡0.00886** ¡0.00819** ¡0.00767** ¡0.00812** ¡0.00773** (-1.43) (-2.05) (-2.23) (-2.11) (-2.03) (-2.09) (-2.05) Size (total assets, logs) 0.253*** 0.286*** 0.202*** 0.268*** 0.293*** 0.287*** 0.290*** (4.37) (3.32) (2.80) (3.10) (3.23) (3.28) (3.27) Tier 1 ratio 0.00762*** 0.00827*** 0.00570*** 0.00802*** 0.00826*** 0.00821*** 0.00832*** (-4.53) (-4.43) (-3.21) (-4.05) (-4.34) (-4.39) (-4.40) Liquidity ratio 0.00145** 0.00111* 0.000549 0.00102* 0.001000* 0.00105* 0.00110* (-2.00) (-1.73) (-0.74) (-1.90) (-1.66) (-1.67) (-1.75) D Policy rate 0.00700 0.00837 0.00665 0.00485 0.00905 0.00829 0.00866 (-1.13) (-1.41) (-1.12) (-0.74) (-1.48) (-1.38) (-1.43) Loans to deposits (logs) 0.109*** 0.0687** 0.109*** 0.110*** 0.109*** 0.109*** (2.93) (2.52) (2.80) (2.87) (2.91) (2.91) LLP 0.0119*** (-3.54) Market share 0.00472*** (3.16) Commercial 0.0269 (-1.20) Savings 0.0347* (1.78) Cooperative 0.00596 (0.33) Observations 16,186 16,065 15,771 16,065 16,065 16,065 16,065 Instruments 79 80 81 81 81 81 81 AR (1) 0.108 0.108 0.000201 0.106 0.108 0.108 0.108 AR (2) 0.344 0.465 0.247 0.484 0.461 0.465 0.463 Hansen 0.324 0.582 0.117 0.461 0.595 0.577 0.586 Note: The dependentvariable is gross loans in logs. Estimation method is dynamic twostep system generalized method of moments (GMM) estimator with robust standard errors and Windmeijer's correction. Lagged dependent variable is treated as endogenous, and all other variables as exogenous. Allregressionsincludeyearfixedeffects.Tstatisticsisreportedinparentheses.Allregressionsincludeweightsbasedonthenumberofobservationsof eachcountry.Macroeconomic variables arelaggedoneperiod, whileall variables, apartfromindicatorsvariables,arewinsorized1%onbothtailsof the distribution. The following are p-values which indicate the significance level of coefficients: *p < 0.10, **p < 0.05, ***p < 0.01. (Continued on next page) 216 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 policyrateisanimportantindicatorofcreditmarket developments associated with monetary policy decisions. The following dynamic panel data model is specified as: Y j;k;t ¼aY j;k; t 1 þgX j;k;t þdZ k;t 1 þl j þ q t þ3 j;k;t ð8Þ where the dependent variable is gross loans trans- formedinnaturallogarithmofbankj,inacountryk and at a time t, Y j;k;t , as a function of lagged dependent variable Y j;k; t 1 , and vector of bank (X) and lagged macroeconomic (Z) variables.l j and q t are bank and time fixed effects, while 3 j;k;t is the disturbance term. Although we have already chosen the system GMM as the primary estimation method, as in Morgan et al. (2018), we decide to also estimate this model using ordinary least squares (OLS) and a fixed effects panel estimator (FE) to compare the magnitude of the coefficients of the lagged depen- dent variable. When the coefficient of the lagged dependent variable is greater than zero, its OLS estimate is biased upward, while the fixed effects estimate is biased downward due to the correlation between the lagged dependent variable and the error term (Nickell, 1981). Since the system GMM estimator uses instruments in levels and first dif- ferences and reduces endogeneity bias, we would expect the GMM estimate of the lagged coefficient of the dependent variable to lie between the OLS and FE estimates. Table 4 presents the estimation results of equation (8). Looking at the coefficient of the lagged dependent variable, we can see that the OLS estimate (0.847) is biased upward, while the FE estimate (0.380) is biased downward. The GMM estimates are within the interval 0.380e0.847 with coefficients around 0.7, as expected. This reinforces ourdecisiontochoosethesystemGMMasthemain estimator. Since the GMM twostep estimator is applied together with the Windmeijer correction for small samples and robust standard errors (Windmeijer, 2005), we estimate all our subsequent regressions with this estimator. We use up to 5 lags of the MPP2 (8) (9) (10) (11) (12) (13) (14) Dependent variable (lag) 0.736*** 0.688*** 0.768*** 0.693*** 0.685*** 0.688*** 0.685*** (13.18) (8.26) (10.20) (7.48) (8.00) (8.14) (8.12) MPP2 (lag) ¡0.0232** ¡0.0287*** ¡0.0226** ¡0.0298*** ¡0.0296*** ¡0.0299*** ¡0.0288*** (-2.23) (-2.80) (-2.31) (-2.87) (-2.82) (-2.79) (-2.82) Size (total assets, logs) 0.251*** 0.290*** 0.217*** 0.271*** 0.296*** 0.291*** 0.293*** (4.66) (3.64) (3.01) (3.15) (3.56) (3.58) (3.61) Tier 1 ratio 0.00737*** 0.00830*** 0.00596*** 0.00800*** 0.00832*** 0.00827*** 0.00836*** (-4.19) (-4.48) (-3.42) (-4.05) (-4.32) (-4.39) (-4.46) Liquidity ratio 0.00149** 0.00115* 0.000682 0.00103** 0.00102* 0.00109* 0.00113* (-2.26) (-1.87) (-0.94) (-1.99) (-1.74) (-1.80) (-1.87) D Policy rate 0.00686 0.00950 0.00781 0.00610 0.0102* 0.00941 0.00981* (-1.07) (-1.64) (-1.33) (-0.95) (-1.70) (-1.61) (-1.67) Loans to deposits (logs) 0.110*** 0.0732** 0.109*** 0.111*** 0.110*** 0.111*** (3.15) (2.54) (2.94) (3.09) (3.11) (3.14) LLP 0.0128*** (-3.62) Market share 0.00476*** (3.38) Commercial 0.0282 (-1.39) Savings 0.0355* (1.93) Cooperative 0.00730 (0.43) Observations 16,186 16,065 15,771 16,065 16,065 16,065 16,065 Instruments 79 80 81 81 81 81 81 AR (1) 0.104 0.102 0.000102 0.101 0.102 0.102 0.102 AR (2) 0.356 0.450 0.272 0.478 0.442 0.449 0.446 Hansen 0.334 0.589 0.0979 0.483 0.612 0.582 0.598 Table 5. (Continued). ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 217 Table 6. The impact of macroprudential policy on gross loansemacroeconomic variables. MPP MPP2 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Dependent variable (lag) 0.697*** 0.673*** 0.684*** 0.698*** 0.685*** Dependent variable (lag) 0.693*** 0.668*** 0.674*** 0.703*** 0.682*** (8.09) (6.99) (7.63) (3.92) (8.20) (8.24) (7.67) (7.66) (3.64) (8.58) MPP (lag) ¡0.00991***¡0.00734** ¡0.00804** ¡0.00669 ¡0.00871** MPP2 (lag) ¡0.0322*** ¡0.0295*** ¡0.0301*** ¡0.0299**¡0.0309*** (-2.85) (-1.97) (-2.05) (-1.61) (-2.21) (-3.51) (-2.85) (-2.98) (-2.45) (-2.91) Size (total assets, logs) 0.281*** 0.304*** 0.293*** 0.284 0.294*** Size (total assets, logs) 0.285*** 0.308*** 0.302*** 0.279 0.296*** (3.41) (3.33) (3.42) (1.64) (3.69) (3.53) (3.73) (3.59) (1.48) (3.90) Tier 1 ratio 0.00802*** 0.00799*** 0.00836*** 0.00825* 0.00859***Tier 1 ratio 0.00808*** 0.00801*** 0.00863*** 0.00803* 0.00861*** (-4.16) (-4.10) (-4.26) (-1.95) (-4.85) (-4.12) (-4.58) (-4.31) (-1.70) (-4.90) Liquidity ratio 0.00121** 0.00123** 0.00134** 0.00191 0.00106* Liquidity ratio 0.00123** 0.00131** 0.00138** 0.00182 0.00107* (-1.96) (-1.98) (-2.11) (-1.34) (-1.65) (-2.10) (-2.13) (-2.22) (-1.20) (-1.73) Loans to deposits (logs) 0.108*** 0.114*** 0.109*** 0.109 0.115*** Loans to deposits (logs) 0.108*** 0.117*** 0.114*** 0.106 0.115*** (2.97) (2.65) (3.00) (1.53) (3.13) (3.15) (2.87) (3.18) (1.38) (3.35) D Policy rate 0.00780 0.00461 0.00981 0.00337 0.00601 D Policy rate 0.00911 0.00609 0.0112* 0.00148 0.00704 (-1.30) (-0.81) (-1.56) (0.49) (-0.98) (-1.55) (-1.21) (-1.85) (-0.21) (-1.14) GDP growth (lag) 0.00714* GDP growth (lag) 0.00726* (1.74) (1.86) Private credit to GDP (lag) 0.000252 Private credit to GDP (lag) 0.000301 (-1.41) (-1.62) Gross capital formation growth (lag) 0.00243** Gross capital formation growth (lag) 0.00259** (2.06) (2.19) NPL gross (lag) 0.00105 NPL gross (lag) 0.00170 (-0.46) (-0.75) Inflation rate (lag) 0.00859** Inflation rate (lag) 0.00911*** (2.54) (2.70) Observations 16,065 16,005 16,065 12,609 16,065 Observations 16,065 16,005 16,065 12,609 16,065 Instruments 81 81 81 69 81 Instruments 81 81 81 69 81 AR (1) 0.105 0.121 0.107 0.206 0.106 AR (1) 0.0992 0.115 0.101 0.204 0.0998 AR (2) 0.431 0.646 0.436 0.485 0.395 AR (2) 0.409 0.638 0.401 0.549 0.374 Hansen 0.589 0.644 0.494 0.471 0.641 Hansen 0.597 0.639 0.522 0.531 0.648 Note:Thedependentvariableisgrossloansinlogs.Estimationmethodisdynamictwostepsystemgeneralized methodofmoments(GMM)estimatorwith robuststandard errorsand Windmeijer'scorrection.Laggeddependentvariableistreatedasendogenous,andallothervariablesasexogenous.Allregressionsincludeyearfixedeffects.Tstatisticsisreportedin parentheses.Allregressionsincludeweightsbasedonthenumberofobservationsofeachcountry.Macroeconomicvariablesarelaggedoneperiod,whileallvariables,apartfromin- dicatorvariables,arewinsorized1%onbothtailsofthedistribution.Thefollowingarep-valueswhichindicatethesignificancelevelofcoefficients:*p<0.10,**p<0.05,***p<0.01. 218 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 dependent variable as instruments to address the endogeneity problem, while instrumenting our exogenous variables with their lags in levels. The lagged dependent variable is treated as endogenous variable and all other variables as exogenous. In all regressions, we include year dummies to control for unobserved heterogeneity over time. To check whether our estimates are consistent, we need to confirm whether the chosen instruments are valid and whether the residuals show serial correlation. WecheckthechoiceofinstrumentswiththeHansen test for overidentifying restrictions and the autocor- relation of the residuals with first and second order autocorrelation tests (see Arellano & Bond, 1991; Baum, 2013; Roodman, 2009). In the case of the first- order autocorrelation test, the error term is expected to be correlated, and in the case of the second-order serial correlation test, we should not reject the null hypothesis of the absence of second-order autocorrelation. When it comes to Hansen'stest,we should not reject the null hypothesis that the over- identification restrictions are valid. Table 4 shows these tests, namely first, AR (1) and second order autocorrelation tests, AR (2) and overidentification test of the restrictions, Hansen's J-test, which con- firms that our estimates are consistent. In column 5, we introduce the MPP index, which measures the overall stance of macroprudential policy. The index is lagged by one period, and its coefficient, obtained using the GMM twostep sys- tem estimator, is statistically significant at the 5% level and has a negative sign, as expected. Consis- tent with previous findings on macroprudential policy, our empirical results show that tighter mac- roprudential policy is associated with a decrease in credit supply. For a one-unit increase in the MPP index indicating a tightening of policy, we find a negative effect of 0.76% on credit supply. The Table 7. The impact of tightening and loosening stance of macroprudential policy on bank lending. (1) (2) (3) (4) MPP3 MPP4 MPP5 MPP6 Dependent variable (lag) 0.698*** 0.728*** 0.704*** 0.698*** (7.36) (7.60) (6.98) (6.94) MPP3 (lag) ¡0.0196* (-1.68) MPP4 (lag) 0.0791*** (4.14) MPP5 (lag) 0.00423 (1.12) MPP6 (lag) 0.0254*** (4.24) Size (total assets, logs) 0.266*** 0.240*** 0.261*** 0.268*** (3.02) (2.70) (2.81) (2.88) Liquidity ratio 0.00116** 0.000886* 0.00112** 0.00104** (-2.27) (-1.86) (-2.13) (-2.08) Tier 1 ratio 0.00789*** 0.00716*** 0.00793*** 0.00787*** (-3.72) (-3.01) (-3.39) (-3.55) Loans to deposits (logs) 0.108*** 0.0994** 0.107*** 0.112*** (2.85) (2.54) (2.67) (2.62) Market share 0.00460*** 0.00390** 0.00449*** 0.00414*** (2.95) (2.57) (2.61) (2.67) D Policy rate 0.00579 0.00415 0.00464 0.00111 (-0.92) (-0.67) (-0.74) (0.17) GDP growth (lag) 0.00401 0.00600 0.00321 0.00494 (0.84) (1.32) (0.58) (0.97) Observations 16,065 16,065 16,065 16,065 Instruments 82 82 82 82 AR (1) 0.104 0.0980 0.108 0.104 AR (2) 0.475 0.446 0.484 0.468 Hansen 0.480 0.471 0.457 0.565 Note: The dependent variable is gross loans in logs. Estimation method is dynamic twostep system generalized method of moments (GMM)estimatorwithrobuststandarderrorsandWindmeijer'scorrection.Laggeddependentvariableistreatedasendogenous,andall other variables as exogenous. All regressions include year fixed effects. T statistics is reported in parentheses. All regressions include weightsbasedonthenumberofobservationsofeachcountry.Macroeconomicvariablesarelaggedoneperiod,whileallvariables,apart from indicator variables, are winsorized 1% on both tails of the distribution. The following are p-values which indicate the significance level of coefficients: *p < 0.10, **p < 0.05, ***p < 0.01. ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 219 coefficientonbanksizeisstatisticallysignificantand positively associated with bank lending, while the tier 1 ratio is statistically significant and negatively associated with bank lending. The policy rate is not significantly associated with the dependent variable in this specification. This may indicate a lower effectiveness of monetary policy in dampening credit growth during the period under consider- ation, which is characterized by expansionary monetary policy and historically low interest rates (Borio & Gambacorta, 2017). Table 5 shows the estimates of equation (7), with an extended set of variables designed to control for various bank characteristics. These results were obtained using two alternative macroprudential measures, the MPP index (right) and the MPP2 index (left). To allow for a better comparison of the results, we report the estimates in the same table. Banks'controlvariableswereintroducedatdifferent stagesoftheregressionanalysis.Inadditiontobank size and tier 1 capital ratio, which were already introduced in our baseline model, we also include liquidity ratio to capture the quality of short-term asset management in an individual bank and the overall liquidity position. To control for the funding structure of banks, we enter the ratio of loans to deposits,transformedintologarithms.Tocontrolfor asset quality and banks' assessment of credit risk based on estimated loan losses, we also introduce the ratioofloanlossprovisionstogross loans(LLP). To control for the market power of banks, we include market share, a variable formed by the size of the bank's total assets standardized to the total size of the banking sector for a given country. Finally,wealso include dummyvariables indicating the bank's specialization: Commercial, Cooperative orSavings.These variables werechosen totakeinto account the overall performance and health of the bank. 8 In six out of seven model specifications, the coef- ficient of the MPP index remains statistically sig- nificant and of negative sign, supporting our first hypothesis. For a one-unit increase in the MPP index, we find a negative impact of 0.5% to 0.8% on credit supply. The results obtained with our alternative measure, the MPP2 index, show similar results, with our second measure showing a stron- ger relationship with bank lending. A one-unit in- crease in the MPP2 index is associated with a decrease in credit supply from 2.26% to 2.99%. These results suggest the effectiveness of macro- prudential policy in mitigating credit booms, while supporting the view that macroprudential policy should be given the primary mandate in promoting financial stability (see Martinez-Miera & Repullo, 2019). The results also warn of the importance of comparison of different measurements of the mac- roprudential stance in related studies reflected in the difference in the coefficients obtained. Banksizeisstatisticallysignificantat1% leveland shows positive association with credit supply in all estimations. On the other hand, tier 1 capital ratio remains statistically significant and with negative Table 8. Macroprudential indices and subindices based on the target of the measures. (1) (2) (3) MPP 0.00991*** (-2.85) Credit (lag) 0.00497 (-0.75) Credit growth (lag) 0.0276** (2.55) Lending caps (lag) 0.0260** (-1.99) Risk weights (lag) 0.00919 (-0.65) Market liquidity (lag) 0.00793 (-0.55) Liquidity measures 0.00837 (-0.59) Concentration (lag) 0.0196** (-2.35) Exposures (lag) 0.0198** (-2.36) Resilience (lag) 0.00594 (-1.12) MCR (lag) 0.00457 (0.60) Capital buffers (lag) 0.0120 (-1.61) Taxes (lag) 0.0114 (-0.56) Provisioning (lag) 0.0218 (-1.42) Leverage ratio (lag) 0.103 (1.21) Other requirements (lag) 0.00908 (-0.64) Note: The dependent variable is gross loans in logs. Estimation method is dynamic twostep system generalized method of mo- ments (GMM) estimator with robust standard errors and Wind- meijer's correction. Lagged dependent variable is treated as endogenous,andallothervariablesasexogenous.Allregressions include year fixed effects. T statistics is reported in parentheses. All regressions include weights based on the number of obser- vationsof eachcountry.Macroeconomicvariablesare laggedone period, while all variables, apart from indicator variables, are winsorized 1% on both tails of the distribution.The following are p-values which indicate the significance level of coefficients: *p < 0.10, **p < 0.05, ***p < 0.01. 8 For a similar variables selection approach see Hasan et al. (2016), Demirgüç-Kunt and Huizinga (2010) and Garcia Revelo et al. (2020). 220 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 sign in all estimations. Liquidity ratio is statistically significant at 5% and 10% level and with negative coefficient in most specifications, indicating that an increase in liquid assets is associated with a decrease in credit activity. Loans to deposits ratio is positively associated with lending activity and sta- tistically significant at 1% level. LLP, entered in columns 3 and 10, is statistically significant and negatively associated with lending. These estimates suggest that an increase in loan-loss provisioning hinders credit activity in banks. Policy rate is not significantly associated with bank lending in these specifications. To summarize, bank controls co- efficients show that larger banks and better performing banks with stable sources of funding tend to lend more. The estimations also show that liquidity needs or capital demands can hinder bank lendingactivities.Ontheotherside,increaseinloan lossprovisioningcouldindicateanincreaseincredit risk assessments, which can also prompt banks to reduce their loan supply. This can come as a result of procyclicality of loan loss provisioning, caused by a delay in provisioning during economic downturn (see Laeven & Majnoni, 2003). Our results show a negative relationship between loan loss provision- ing and bank lending, in line with Pool et al. (2015) and Bouvatier and Lepetit (2013).Wefind market share is significantly and positively associated to bank lending. Numerous studies have shown that banks with higher market power have access to many alternative sources of financing, along with easier access tofinancial markets, thusthey areable to lend more than banks with smaller market shares. These effects were found for normal times and for the period of economic downturns (see Cubillas & Suarez, 2018; Fungacova et al., 2014). Of specialization dummies, only the coefficient receivedforsavingsbanksissignificantlyassociated to bank lending, but only at the 10 percent level. This may be the result of savings banks business models characteristics, which are to a large extent focused on lending business. Table 6 shows estimations with different macro- economic variables introduced in the model. The most relevant bank variables from previous analysis remaininthesetofcontrolsfortheseregressions.To complement our bank controls, we include addi- tionalmacroeconomicvariablesindifferentstagesof the analysis. First, we include GDP real growth rate, as our proxy for credit demand and economic development. We find that an increase in GDP growth is significantly associated with bank lending, but only at a 10% level. This association is expected, ashighercreditdemandandeconomicdevelopment stimulates heightened credit activity in banks. We alsocontrolforprivatecreditgrowthasapercentage of GDP, as an additional measure of financial development (Morgan et al., 2018); gross capital formationgrowth,inordertoaccountforthelevelof investments and the contribution to the economic activityinacountry(Festic etal.,2011);NPLgrossas a percentage of GDP, as a financial soundness indi- cator; and inflation rate, as an additional measure of credit demand. Domestic bank credit to private sector to GDP and country level NPL ratio are not significantlyassociatedtoindividualbanklendingin our specifications. As expected, gross capital forma- tion growth is significantly and positively associated to bank lending, while the inflation rate is also Table 9. Loosening cycles of different macroprudential indices and subindices based on the target of the measures. (1) (2) (3) MPP4 (lag) 0.0791*** (4.14) MPP6 (lag) 0.0254*** (4.24) Credit (lag) 0.0370** (2.53) Credit growth (lag) 0.00114 (-0.06) Lending caps (lag) 0.104*** (3.09) Risk weights (lag) 0.0206 (0.90) Market liquidity (lag) 0.0294 (1.57) Liquidity measures 0.0295 (1.56) Concentration (lag) 0.0798*** (3.30) Exposures (lag) 0.0801*** (3.32) Resilience (lag) 0.0933*** (3.58) MCR (lag) 0.0113 (0.26) Capital buffers (lag) 0.107*** (3.83) Taxes (lag) 0.0712 (1.58) Provisioning (lag) 0.0717** (2.07) Other requirements (lag) 0.0693* (1.89) Note: The dependent variable is gross loans in logs. Estimation method is dynamic twostep system generalized method of mo- ments (GMM) estimator with robust standard errors and Wind- meijer's correction. Lagged dependent variable is treated as endogenous,andallothervariablesasexogenous.Allregressions include year fixed effects. T statistics is reported in parentheses. All regressions include weights based on the number of obser- vationsof eachcountry.Macroeconomicvariablesare laggedone period, while all variables, apart from indicator variables, are winsorized 1% on both tails of the distribution.The following are p-values which indicate the significance level of coefficients: *p < 0.10, **p < 0.05, ***p < 0.01. ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 221 significantly and positively associated to bank lending (Beutler et al., 2020). Our macroprudential indices remain significantly and negatively associated to credit supply in most specifications. For a unit increase in MPP, bank lending exhibits a decrease from 0.7 to 1%. A unit increase in MPP2 index, is significantly associated with a decline in bank lending by around 3%. 2.5 Tightening and loosening of macroprudential policy In order to analyze the existence of asymmetric effects in macroprudential response, we investigate theimpactoftighteningandlooseningpolicyactions separately.Theresultsoftheseestimationsaregiven in Table 7. In these estimations, we include several important bank and macroeconomic determinants (seecolumn1inTable6),alongwithourfourindices measuringthepolicystance:1)tightening:MPP3and MPP5, and 2) loosening: MPP4 and MPP6. Tight- ening actions of macroprudential policy are significantlyassociatedtobanklendingat10%,when measuredwithMPP3index.Theseresultsareinline withPoghosyan(2020),wheretheauthorinvestigates the effectiveness of lending restrictions on credit supplyandfindsonlyweakassociationoftightening policy actions with lending activity in banks. On the other side, when we analyze loosenings of macro- prudential policy, we find that these actions are strongly associated with the increase in lending ac- tivity of banks. We find statistically significant coef- ficientwith positivesigns for both, MPP4and MPP6 indices, which are followed by an increase in bank lending by 8% (MPP4) and 2.5% (MPP6). The differ- ence in the effects lies in the measurement design and suggest its importance when comparing and determining the effects of the policy. This analysis suggests that deactivation or loosening of macro- prudential policy has stronger effects on bank lending and is associated with increase in credit ac- tivities. The asymmetry of results when analyzing tighteningsandlooseningsseparatelycouldcomeas a consequence of regulatory leakages due to Table 10. Macroprudential indices vs. crisis. Pre-crisis Crisis Post-crisis (1) (2) (3) (4) (5) (6) (7) (8) (9) Dependent variable (lag) 0.787*** 0.789*** 0.798*** 0.697*** 0.626*** 0.642*** 0.448 0.509*** 0.533*** (15.94) (15.72) (16.44) (4.90) (5.46) (5.11) (1.08) (2.82) (2.97) MPP (lag) 0.00120 ¡0.0123** ¡0.00283 (0.08) (-1.99) (-0.41) MPP3 (lag) 0.0112 ¡0.0508** 0.00547 (0.22) (-1.99) (0.32) MPP4 (lag) 0.0708** 0.0507* 0.103** (2.35) (1.89) (2.30) Size (total assets, logs) 0.173*** 0.171*** 0.164*** 0.258** 0.321*** 0.308*** 0.173*** 0.171*** 0.164*** (3.90) (3.78) (3.77) (1.98) (3.09) (2.68) (3.90) (3.78) (3.77) Tier 1 ratio 0.00570*** 0.00557*** 0.00556*** 0.0128** 0.0158*** 0.0146*** 0.00570*** 0.00557*** 0.00556*** (-3.63) (-3.65) (-3.60) (-2.37) (-3.30) (-2.91) (-3.63) (-3.65) (-3.60) Liquidity ratio 0.000984* 0.000984* 0.00109** 0.00177* 0.00238** 0.00203** 0.000984* 0.000984* 0.00109** (1.96) (1.91) (2.32) (-1.88) (-2.38) (-2.04) (1.96) (1.91) (2.32) Loans to deposits (logs) 0.0715*** 0.0729*** 0.0695*** 0.114* 0.137*** 0.134*** 0.0715*** 0.0729*** 0.0695*** (2.70) (2.71) (2.69) (1.92) (2.81) (3.02) (2.70) (2.71) (2.69) Market share 0.00124 0.00120 0.000961 0.00569*** 0.00620*** 0.00640*** 0.00564 0.00546** 0.00467* (0.78) (0.75) (0.62) (3.08) (3.07) (2.97) (0.87) (2.07) (1.85) D Policy rate 0.0180** 0.0182** 0.0164** 0.00708 0.0133* 0.0141 0.0180** 0.0182** 0.0164** (-2.38) (-2.31) (-2.13) (0.78) (-1.83) (-1.50) (-2.38) (-2.31) (-2.13) GDP growth (lag) 0.0297*** 0.0291*** 0.0296*** 0.00217 0.00533** 0.00374 0.0297*** 0.0291*** 0.0296*** (4.26) (3.85) (4.36) (0.49) (1.99) (1.16) (4.26) (3.85) (4.36) Observations 3889 3889 3889 4090 4090 4090 8086 8086 8086 Instruments 28 28 28 38 34 34 32 29 29 AR (1) 0.00126 0.00127 0.00141 0.0646 0.0726 0.0827 0.326 0.273 0.265 AR (2) 0.419 0.420 0.441 0.657 0.421 0.375 0.741 0.784 0.846 Hansen 0.156 0.151 0.124 0.390 0.030 0.004 0.253 0.000332 0.000167 Note: The dependent variable is gross loans in logs. Estimation method is dynamic twostep system generalized method of moments (GMM)estimatorwithrobuststandarderrorsandWindmeijer'scorrection.Laggeddependentvariableistreatedasendogenous,andall other variables as exogenous. All regressions include year fixed effects. T statistics is reported in parentheses. All regressions include weightsbasedonthenumberofobservationsofeachcountry.Macroeconomicvariablesarelaggedoneperiod,whileallvariables,apart from indicator variables, are winsorized 1% on both tails of the distribution. The following are p-values which indicate the significance level of coefficients: *p < 0.10, **p < 0.05, ***p < 0.01. 222 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 circumvention of regulatory measures (e.g. through window dressing). On the other hand, some bor- rowers might turn to non-bank institutions, while there is also a question on national jurisdiction for many macroprudential policies which could cause circumvention of the policies (see Aiyar et al., 2015; Poghosyan,2020;Reinhardt&Sowerbutts,2015). 3 Extensions and robustness tests As we are interested in providing additional analysis which would extend and provide further insights into understanding the relationship be- tween bank lending behavior and macroprudential policy, we perform additional tests which consider measurement, time period and sample of banks. First,wedivideourmainindextomeasuresdirected at certain regulatory targets. Second, we inspect the impact of time period and the state of the financial cycle by dividing the sample in respect to the 2008 financial crisis outbreak. Lastly, we measure the impactofmacroprudentialpolicyonasubsampleof listed banks, as we are interested in examining if there is a stronger association between macro- prudential instruments activation and institutions which are part of this subsample. 3.1 Different types of macroprudential instruments Macroprudential data based on MaPPED consists of different macroprudential tools with numerous objectives. By following Meuleman and Vander Vennet (2020), we group macroprudential tools basedontheirtargetintofourindices:1)Credit,tools directed at curbing excessive lending; 2) Market liquidity,toolsaimedtoimproveliquiditypositionof banks; 3) Concentration, policy tools aimed to decreasedifferentbankexposurestocertaintypesof loans or lenders and 4) Resilience, macroprudential measures directed at capital position of banks, as well as specific macroprudential measures aimed at banking sector resilience. First column of Table 8 represents the baseline results for MPP index, for the purpose of compara- bility. 9 Column 2 shows results obtained with four groups of indices. We find statistically significant results only for Concentration index, which indicates that macroprudential measures targeted to limit bankexposurestodifferenttypesofloansorlenders areeffectiveinreducingbanklendingforaround2%. Additionally,wefurtherexaminethesubdivisionof the aforementioned indices, and we form different subgroups of these indices, to test macroprudential indices with more granular approach to their objec- tives. This is evident in column 3. Among the credit targetingmacroprudentialmeasures,wefindlending caps to be most effective. Following the activation or tightening of existing lending caps, bank lending de- clinesfor2.6%,afindingsimilartoPoghosyan(2020). This is expected as these tools specifically target excessive credit growth and overheating in credit market. These results are in line with Cerutti et al. (2017) and Morgan et al. (2018),whosepaperspecif- ically investigate loan to value measures, and other studieswhosefindingsshowthatthesemeasuresare also effective in limiting bank risk (Altunbas et al., 2018; Claessens et al., 2013; Meuleman & Vander Vennet,2020).Surprisingly,wereceivesignificantand positive coefficient for credit growth tools, which are consisted of reserve requirements. Similar results were found by Akinci and Olmstead-Rumsey (2018), whosuggestthatthismightbecausedbythepresence of euro area countries in the sample prone to the ef- fectsofECBactions,whichcouldhavesomecounter- cyclicaleffectsdependingoncountrycharacteristics. Due to the ever-growing interest in the loosening effects of macroprudential policies and in order to assess the beneficial effects of relaxation of some macroprudential measures to counter the effects of the current crisis and to support credit supply, we decide to examine in particular the loosening ac- tions of different groups of macroprudential in- struments.TheresultsfromTable9indicatethatthe strongest effects on bank lending have capital buffers and lending caps. After the relaxation of capital buffers, bank lending increases by about 10.7%,whilethelooseningoflendingcapsleadstoa 10.4% increase in credit supply. The first column shows results received for MPP4 and MPP6 indices which we include for comparability of coefficients. We also obtain statistically significant and positive coefficients for provisioning and exposures related indices. These results confirm the effectiveness of macroprudential policies in easing cycles, which also underpins their use in the current Covid-19 crisis. However, we stress that the use of these policies should be accompanied by a cautious approach that should maintain the ultimate macro- prudential objective of financial stability. This means that policymakers should still have other tools attheirdisposaltopreventdeteriorationinthe quality of bank assets and the riskiness of the borrower base. We also stress the importance of coordinating different macroprudential policies at the national and supranational levels (e.g. in the 9 The results inTables 8and9wereminimizedforbrevity.Complete estimationresults, with allvariablescoefficients,areavailable inAppendix(Tables A3eA6). ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 223 euro area) to maximize the positive effects of easing some instruments that may lead to an increase in credit supply and economic growth. 10 3.2 Macroprudential policy over different periods In order to test the effects of macroprudential in- struments on bank lending in different periods reflecting the impact of the global financial crisis of 2008, we decide to split our sample over different timeperiods:pre-crisis,crisisandpost-crisisperiod. We analyze the effects of MPP index, our base macroprudential index and initial two indices formulated to measure tightening and loosening policy actions separately, MPP3 and MPP4. The analysis shows that macroprudential in- struments are most effective in easing cycle in pre- crisis period, with the coefficient indicating when macroprudential measures are loosened, the credit supply increases by 7%. Other macroprudential indices are not significant in this setting. When we moveontothecrisis period,weobservethatmacro- prudential measures are effective in curbing credit growth during the economic downturn. For the one- unit increase in MPP and MPP3 index, bank lending declines by 1.2% and 5%. By looking at MPP4 index, the loosening effect of macroprudential policy is somewhat weaker indicating that if macroprudential measuresareloosened,creditsupplyincreasesby5%, with 10% level of significance. In the post-crisis Table 11. The impact of macroprudential policy on listed banks. (1) (2) (3) (4) (5) (6) MPP MPP2 MPP3 MPP4 MPP5 MPP6 Dependent variable (lag) 0.604*** 0.651*** 0.667*** 0.668*** 0.649*** 0.649*** (5.65) (6.15) (6.28) (6.85) (6.25) (5.46) MPP (lag) ¡0.0186*** (-4.21) MPP2 (lag) ¡0.0369*** (-3.08) MPP3 (lag) ¡0.0509*** (-3.05) MPP4 (lag) 0.0502* (1.93) MPP5 (lag) ¡0.0116** (-2.20) MPP6 (lag) 0.0218** (2.48) Size (total assets, logs) 0.370*** 0.328*** 0.313*** 0.312*** 0.327*** 0.331*** (3.76) (3.43) (3.29) (3.54) (3.42) (3.02) Tier 1 ratio 0.00368 0.00268 0.00257 0.00290 0.00327 0.00412 (-1.16) (-0.98) (-0.90) (-1.15) (-1.12) (-1.20) Liquidity ratio 0.000889 0.000874 0.000775 0.000901 0.000746 0.000947 (1.59) (1.46) (1.30) (1.32) (1.17) (1.39) Loans to deposits (logs) 0.294*** 0.263*** 0.251*** 0.250*** 0.258*** 0.256*** (3.29) (2.88) (2.66) (2.87) (3.10) (2.63) Market share 0.0000616 0.000262 0.000390 0.000503 0.00000114 0.000815 (0.05) (-0.19) (-0.28) (-0.33) (-0.00) (-0.49) D Policy rate 0.0114 0.00560 0.00470 0.00729 0.00676 0.0122 (1.28) (0.59) (0.49) (0.73) (0.66) (1.36) GDP growth (lag) 0.00300 0.00221 0.00194 0.00130 0.000819 0.00312 (0.88) (0.63) (0.52) (0.36) (0.22) (0.78) Observations 876 876 876 876 876 876 Instruments 82 82 82 82 82 82 AR (1) 0.096 0.081 0.088 0.071 0.094 0.075 AR (2) 0.375 0.575 0.784 0.388 0.586 0.283 Hansen 0.244 0.243 0.251 0.175 0.225 0.197 Note: The dependent variable is gross loans in logs. Estimation method is dynamic twostep system generalized method of moments (GMM)estimatorwithrobuststandarderrorsandWindmeijer'scorrection.Laggeddependentvariableistreatedasendogenous,andall other variables as exogenous. All regressions include year fixed effects. T statistics is reported in parentheses. All regressions include weightsbasedonthenumberofobservationsofeachcountry.Macroeconomicvariablesarelaggedoneperiod,whileallvariables,apart from indicator variables, are winsorized 1% on both tails of the distribution. The following are p-values which indicate the significance level of coefficients: *p < 0.10, **p < 0.05, ***p < 0.01. 10 ThecoefficientfortheleverageratiolimitisnotreportedinTable9asthedatadonotreportanylooseningsoftheleverageratiolimitwithinoursample and time coverage. 224 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 period, we find only MPP4 index to be significantly associated to bank lending. For a unit increase in MPP4 index, we find that bank lending increases by more than 10%. This relationship shows that loos- ening policy actions have stronger effect on bank lending,asinPoghosyan(2020).Thisfindingisespe- cially important in the current state of affairs when most countries are relaxing regulatory demands imposed on banks in order to combat the negative economic effects of the Covid-19 pandemic and further corroborate our previous analysis. Interest- ingly,wealsofindthatthecoefficientofthepolicyrate is strongly and negatively associated with bank lending in the pre-crisis and post-crisis periods, but theestimationsrestrictedtothecrisisperiodshowthat the coefficient of the policy rate is not statistically significant.Thiscanbeasignoflowereffectivenessof monetary policy on credit activity in the period of distress,however,thisinvestigationgoesbeyondthis paper(seeTable10). 3.3 Subsample of listed banks To examine whether our results are robust to a limitedsetofinstitutions,weanalyzethesubsample of 112 listed banks available in the Fitch Connect database for EU countries. Majority of these banks are also the largest EU banks, and of systemic importance. This gives us an additional incentive to measure the impact of macroprudential measures onlendingbehavioroftheseinstitutions.Assomeof macroprudential tools are specifically aimed at systematically large banks, we expected to find sig- nificant relationship between macroprudential pol- icy and lending behavior in this sample setting. We estimate the effects using all of our six macro- prudential indices, as we want to account for both theoverallmacroprudentialstanceandtheeffectsof tightening and easing cycles of policy. The results obtained show that, as in our previous analysis of the full sample of banks, the overall macro- prudential policies and the MPP and MPP2 indices are significantly and negatively associated with bank lending at the 1% level, as expected. Tight- ening of macroprudential measures is followed by a decline in lending activity by 1.86% (MPP) and 3.7% (MPP2). When we analyze tightenings only, we see that tightening effects are even stronger for listed banks, with both indices significantly associated with bank credit at the 1% (MPP3) and 5% (MPP5) levels. A tightening of macroprudential instruments isfollowedbyareductioninbankcreditofabout5% (MPP3) and 1.16% (MPP5). These estimates suggest thatlistedbanksaremoreresponsivetoatightening of macroprudential measures. Finally, we consider the results obtained for the MPP4 and MPP6 easing indices. The obtained statistically significant co- efficients at 10% (MPP4) and 5% (MPP6) levels indicate that bank credit is elevated by 5% (MPP4) and 2.18% (MPP6) following a relaxation of macro- prudential instruments (see Table 11). 4 Conclusion In this study, we investigate whether macro- prudential measures are effective in curtailing bank lending on a sample of 28 EU countries and an over 17-year period horizon between 2000 and 2017. By employingtheArellano-Bondgeneralizedmethodof moments,whichenabledustominimizeendogeneity issues related with the introduction of these mea- sures, we find that macroprudential measures are negatively associated with bank lending. We assess alternativemeasuresofmacroprudentialpolicy.First, weanalyzetwomacroprudentialindices,whichtake intoaccountallinstrumentsavailableinthedatabase and all policy actions: tightenings, and loosenings. We find that these measures are significantly and negativelyassociatedtocreditsupply,whencontrol- lingfordifferentbankandmacroeconomicvariables. Inanattempttodifferentiatetheeffectoftightening and loosening measures, we construct four different indices capturing only tightening or loosening ac- tions. We receive weak association of tightening indices and credit activity of banks. On the contrary, analysis shows that the effects on bank lending are stronger when macroprudential measures are loos- enedordeactivated.Thecoefficientforlooseningac- tions is significantly and positively associated with bank lending when introducing different bank and macroeconomicvariableswhichmighthaveaneffect on bank lending activity. Thesefindings are particu- larlyimportantinthecurrentsituationandeconomic consequences of the Covid-19 pandemic, as policy- makers and governments relax prudential re- quirements to increase market liquidity and support investment and economic growth. The unloosing of macroprudential measures during economic down- turns can also support monetary policy efforts to promote market liquidity without jeopardizing financial stability. In addition, the use of macro- prudential measures in relation to specific sectors, such as SME borrowers or specific industries, may also take into account the special needs of the given country's sector. More efforts could also be made to bettercoordinateregulatoryactionsatbothsuprana- tional and national levels (Guindos, 2021). As a robustness test, we further analyze which macro- prudential measures have the strongest relationship withbanklending.Wefind,asexpected,thatlending ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 225 constraintsaresignificantlyandnegativelyassociated with lending, while credit growth measures, such as reserverequirementsontheliabilityandassetsideof thebankbalancesheet,showsomeprocyclicaleffect, as we obtained a positive coefficient. Moreover, we find that capital buffers and lending measures have the strongest relationship with credit in macro- prudentialpolicyeasingcycles. Toshowwhethertherearedifferencesintheeffects of macroprudential indices in different periods of the financial cycle, we testthe effects of three indices, the overall macroprudential stance, tightenings, and looseningsseparately,inthepre-crisis,crisis,andpost- crisisperiods.Wefindthatmacroprudentialmeasures are most efficient in controlling credit activity in the crisisperiod,whiletheeffectofeasingpolicyactionsis somewhat limited. In the pre-crisis and post-crisis periods,thelooseningcycleofmacroprudentialpolicy hasastrongerimpactonbanklending.Inaddition,we check our results with the subsample of listed banks. Theresultsforlistedbanksareevenstronger,asallof ourmacroprudentialindicesarestatisticallysignificant andexhibittheexpectedrelationship.Whenanalyzing policy instruments, policymakers should evaluate all measures introduced in a country in a given period: both macroprudential and microprudential in- struments that could have some macroprudential ef- fect, analyzing also their interaction with monetary policy. We also support the proposition that macro- prudential policy should be a primary tool to curb creditboomsandfinancialsystemvulnerabilities,asit effectivelylimitsexcessivebanklending.Ontheother hand, the effectiveness of macroprudential policy is significantly challenged by regulatory leakages and national responsibility for many macroprudential measuresin the internationalized and interconnected Europeanbankingsector. Acknowledgement Aida Cehajic thankfully acknowledges the fund- ing provided by the Erasmus Mundus Project Green-Tech-WB: Smart and Green technologies for innovative and sustainable societies in Western Balkans (Grant number: 551984-EM-1-2014-1-ES- ERA MUNDUS-EMA21). The views and opinions expressed in this paper are entirely those of the authors. All errors remain our own. References Aikman, D., Nelson, B., & Tanaka, M. (2015). Reputation, risk- taking, and macroprudential policy. Journal of Banking & Finance, 50, 428e439. Available at: https://doi.org/10.1016/ j.jbankfin.2014.06.014. Aiyar, S., Calomiris, C. W., & Wieladek, T. (2015). Bank Capital regulation:Theory,empirics,andpolicy.IMF EconomicReview, 63(4), 955e983. Available at: https://doi.org/10.1057/ imfer.2015.18. Akinci, O., & Olmstead-Rumsey, J. (2018). How effective are macroprudential policies? An empirical investigation. Journal of Financial Intermediation, 33(1136), 33e57. Available at: https://doi.org/10.1016/j.jfi.2017.04.001. Altunbas, Y., Binici, M., & Gambacorta, L. (2018). Macro- prudential policy and bank risk. Journal of International Money and Finance, 81, 203e220. Available at: https://doi.org/10.1016/ j.jimonfin.2017.11.012. Araujo, J. D., Patnam, M., Popescu, A., Valencia, F., & Yao, W. (2020). Effects of macroprudential policy: Evidence from over 6,000 estimates. IMF Working Papers. Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. The Review of Economic Studies, 58(2), 277e297. Available at: https://www.jstor.org/stable/2297968. Baum, C. F. (2013). Dynamic panel data estimators, EC 823: Applied econometrics. Boston: Boston college. Available at: http:// fmwww.bc.edu/EC-C/S2013/823/EC823.S2013.nn05.slides.pdf. Beutler,T.,Bichsel,R.,Bruhin,A.,&Danton,J.(2020).Theimpact of interest rate risk on bank lending. Journal of Banking & Finance, 115. Available at: https://doi.org/10.1016/ j.jbankfin.2020.105797. Borio,C.(2003).Towardsamacroprudentialframeworkforfinancial supervision and regulation? CESifo Economic Studies, 49(2), 181e215.Availableat:https://doi.org/10.1093/cesifo/49.2.181. Borio, C., & Gambacorta, L. (2017). Monetary policy and bank lending in a low interest rate environment: Diminishing effectiveness? Journal of Macroeconomics, 54, 217e231. Available at: https://doi.org/10.1016/j.jmacro.2017.02.005. Borri, N., & Giorgio, G. (2021). Systemic risk and the COVID chal- lengeintheeuropeanbankingsector.JournalofBanking&Finance. Availableat:https://doiorg/10.1016/j.jbankfin.2021.106073. Bouvatier, V., & Lepetit, L. (2013). Effects of loan loss provisions on growth in bank lending: Some international comparisons. International Economics, 132(4), 91e116. Available at: https:// doi.org/10.1016/s2110-7017(13)60059-1. Bruno, V., Shim, I., & Shin, H. S. (2017). Comparative assess- ment of macroprudential policies. Journal of Financial Sta- bility, 28,1 8 3 e202. Available at: https://doi.org/10.1016/ j.jfs.2016.04.001. Budnik, K., & Kleibl, J. (2018). Macroprudential regulation in the European union in 1995e 2014: Introducing a new data set on policy actions of a macroprudential nature. ECB Working Paper. Cerutti, E., Claessens, S., & Laeven, L. (2017). The use and effectiveness of macroprudential policies: New evidence. Journal of Financial Stability, 28, 203e224. Available at: https:// doi.org/10.1016/j.jfs.2015.10.004. Claessens, S., Ghosh, S. R., & Mihet, R. (2013). Macro-prudential policies to mitigate financial system vulnerabilities. Journal of International Money and Finance, 39, 153e185. Available at: https://doi.org/10.1016/j.jimonfin.2013.06.023. Crowe, C., Dell'Ariccia, G., Igan, D., & Rabanal, P. (2013). How to deal with real estate booms: Lessons from country experi- ences. Journal of Financial Stability, 9(3), 300e319. Available at: https://doi.org/10.1016/j.jfs.2013.05.003. Cubillas,E., & Suarez, N. (2018).Bank marketpowerandlending duringthe globalfinancialcrisis. Journal of International Money and Finance, 89,1 e22. Available at: https://doi.org/10.1016/ j.jimonfin.2018.08.003. Dell'Ariccia, G., Bakker, B. B., Laeven, L., Vandenbussche, J., Igan, O. D., & Tong, H. (2012). Policies for macrofinancial sta- bility: How to deal with credit booms. IMF Staff Discussion Note. Demirgüç-Kunt, A., & Huizinga, H. (2010). Bank activity and funding strategies: The impact on risk and returns. Journal of Financial Economics, 98(3), 626e650. Available at: https:// doi.org/10.1016/j.jfineco.2010.06.004. Festic, M., Kavkler, A., & Repina, S. (2011). The macroeconomic sources of systemic risk in the banking sectors offive new EU 226 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 member states. Journal of Banking & Finance, 35(2), 310e322. Available at: https://doi.org/10.1016/j.jbankfin.2010.08.007. Fungacova, Z., Solanko, L., & Weill, L. (2014). Does competition influencethebanklendingchannelintheeuroarea?Journalof Banking & Finance, 49, 356e366. Available at: https://doi.org/ 10.1016/j.jbankfin.2014.06.018. Gambacorta, L., & Murcia, A. (2019). The impact of macro- prudential policies in Latin America: An empirical analysis using credit registry data. Journal of Financial Intermediation, 100828. Available at: https://doi.org/10.1016/j.jfi.2019.04.004. GarciaRevelo,J.D.,Lucotte,Y.,&Pradines-Jobet,F.(2020).Macro- prudentialandmonetarypolicies:TheneedtodancetheTango inharmony.JournalofInternationalMoneyandFinance.Available at:https://doiorg/10.1016/j.jimonfin.2020.102156. Guindos, L. (2021). Macroprudential policy after the COVID-19 pandemic. [Speech]. Banque de France/Sciences Po Financial Stability Review Conference 2021 “Is macroprudential policy resilient to the pandemic?”, Web conference. Hasan, I., Liu, L., & Zhang, G. (2016). The determinants of global BankCredit-Default-Swapspreads.JournalofFinancialServices Research, 50(3), 275e309. Available at: https://doi.org/10.1007/ s10693-015-0232-z. Igan,D.,& Kang,H. (2011).Doloan-to-value anddebt-to-income limits work? Evidence from Korea. IMF Working Paper, SSRN. Available at: https://doiorg/10.2139/ssrn.1915127. Jimenez, G., Ongena, S., Peydro, J.-L., & Saurina, J. (2017). Mac- roprudential policy, countercyclical Bank Capital buffers, and credit supply: Evidence from the Spanish dynamic provi- sioning experiments. Journal of Political Economy. Available at: https://doiorg/10.1086/694289. Kosak,M.,Li,S.,Loncarski,I.,&Marinc,M.(2015).Qualityofbank capital and bank lending behavior during the global financial crisis. International Review of Financial Analysis, 37, 168e183. Availableat:https://doi.org/10.1016/j.irfa.2014.11.008. Laeven, L., & Majnoni, G. (2003). Loan loss provisioning and economic slowdowns: Too much, too late? Journal of Financial Intermediation, 12(2), 178e197. Available at: https://doi.org/ 10.1016/S1042-9573(03)00016-0. Lim, C. H., Columba, F., Costa, A., Kongsamut, P., Otani, A., Saiyid,M.,Wezel,T.,&Wu,X.(2011).Macroprudentialpolicy: What instruments and how to use them? Lessons from country experiences. IMF Working Papers, 85. Available at: https://www.imf.org/external/pubs/ft/wp/2011/wp11238.pdf. Martinez-Miera, D., & Repullo, R. (2019). Monetary policy, mac- roprudential policy, and financial stability. Annual Review of Economics,11(1),809e832.Availableat:https://doi.org/10.1146/ annurev-economics-080218-025625. Meuleman, E., & Vander Vennet, R. (2020). Macroprudential policy and bank systemic risk. Journal of Financial Stability, 47, 100724. Available at: https://doi.org/10.1016/j.jfs.2020.100724. Morgan, P. J., Regis, P. J., & Salike, N. (2018). LTV policy as a macroprudential tool and its effects on residential mortgage loans. Journal of Financial Intermediation,1 e15. Available at: https://doi.org/10.1016/j.jfi.2018.10.001. Nakatani,R.(2020).Macroprudentialpolicyandtheprobabilityof a banking crisis. Journal of Policy Modeling. Available at: https://doiorg/10.1016/j.jpolmod.2020.05.007. Nickell, S. (1981). Biases in dynamic models with fixed effects. Econometrica, 49(6),1417e1426.Availableat: https://www.jstor. org/stable/1911408. Olszak, M., Roszkowska, S., & Kowalska, I. (2018). Macro- prudential policy instruments and procyclicality of loan-loss provisions e cross-country evidence. Journal of International FinancialMarkets,InstitutionsandMoney,54,228e257.Available at: https://doi.org/10.1016/j.intfin.2018.01.001. Poghosyan, T. (2020). How effective is macroprudential policy? Evidence from lending restriction measures in EU countries. Journal of Housing Economics, 49(May), 101694. Available at: https://doi.org/10.1016/j.jhe.2020.101694. Pool, S., de Haan, L., & Jacobs, J. P. A. M. (2015). Loan loss pro- visioning, bank credit and the real economy. Journal of Mac- roeconomics, 45, 124e136. Available at: https://doi.org/10.1016/ j.jmacro.2015.04.006. Reinhardt, D., & Sowerbutts, R. (2015). Regulatory arbitrage in action: Evidence from banking flows and macroprudential policy. Bank of England Staff Working Paper, 546. Roodman, D. (2009). How to do xtabond2: An introduction to difference and system GMM in Stata. The Stata Journal, 9(1), 86e136. Available at: https://doi.org/10.1177/1536867X09009 00106. Vandenbussche, J., Vogel, U., & Detragiache, E. (2013). Macro- prudential policies and housing prices: A new database and empirical evidence for central, eastern and southeastern europe. Journal of Money, Credit, and Banking, 47(1), 1e35. Available at: https://doi.org/10.1111/jmcb.12206. Windmeijer, F. (2005). A finite sample correction for the variance of linear efficient two-step GMM estimators. Journal of Econometrics, 126(1), 25e51. Available at: https://doi.org/ 10.1016/j.jeconom.2004.02.005. Yellen, J. L. (2011). Pursuing financial stability at the federal reserve [Speech]. In Fourteenth annual international banking conference. Chicago, Illinois: Federal Reserve Bank of Chicago. Zhang, L., & Zoli, E. (2016). Leaning against the wind: Macro- prudential policy in Asia. Journal of Asian Economics, 42, 33e52. Available at: https://doi.org/10.1016/j.asieco.2015.11. 001. Appendix Table A1. Macroprudential instruments in MaPPED. Instrument group Instruments Minimum capital requirements (MCR) Capital adequacy ratio (CAR) Tier 1 capital ratio CommonEquityTier1capitalratio(CET1) Core Tier 1 capital ratio Capital buffers Countercyclical capital buffer Capital conservation buffer Systemic risk buffer G-SII capital buffer OeSII capital buffer Other capital requirements targeting most important institutions Other capital surcharges and own funds requirements Profit distribution restrictions Risk weights Risk weights for loans backed by residen- tial property Risk weights for loans backed by com- mercial property Other sectoral risk weights Leverage ratio Leverage ratio limit Loan loss provisioning Loan classification rules Minimum specific provisioning General provisioning Capital treatment of loan loss reserve Lending standards restrictions Loan-to-value (LTV) limits Loan-to-income (LTI) limits Debt-to-income (DTI) limits Debt-service-to-income (DSTI) limits incl. interest rate stress testing Limits on interest rates on loans Maturity and amortization restrictions Other income requirements for loan eligibility Limits on the volume of personal loans Other restrictions on lending standards (continued on next page) ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 227 Table A2. Description of variables. Variable Description Source Bank specific variables Gross loans (logs) The natural logarithm of gross loans. Fitch Connect, Own calculations Size (total assets, logs) The natural logarithm of total assets. Fitch Connect, Own calculations Liquidity ratio (%) Liquid assets/Total assets ratio. Fitch Connect Tier 1 ratio (%) Tier 1 capital ratio/Total risk-weighted assets ratio. Fitch Connect Loans to deposits (logs) Gross loans/Total customer deposits ratio. Fitch Connect, Own calculations Loan loss provisioning (%) The ratio of loan-loss provisioning/Gross loans. Fitch Connect Market share (%) Total assets of bank j in country k and year t over total assets of the banking sector in country k at year t. Fitch Connect, Own calculations Commercial (0e1) Dummy variable equal to 1 if bank specialization is commercial, and 0 otherwise. Fitch Connect, Own calculations Savings (0e1) Dummy variable equal to 1 if bank specialization is savings, and 0 otherwise. Fitch Connect, Own calculations Cooperative (0e1) Dummy variable equal to 1 if bank specialization is cooperative, and 0 otherwise. Fitch Connect, Own calculations Macroeconomic variables D Policy rate Theyearly change in central bank policy rate, calculated by taking the first difference. IMF, European Central Bank, own calculations GDP growth (%) Lag of annual real growth rate of GDP, lagged. World Bank Credit to private sector (%) Domestic private credit as percentage of annual GDP growth rate, lagged. World Bank Gross capital formation growth (%) Gross capital formation as percentage of GDP growth rate, lagged. World Bank NPL gross (%) Bank non-performing loans to total gross loans, lagged. World Bank Inflation rate (%) Inflation rate, based on CPI, lagged. World Bank Macroprudential data MPP The sum of all macroprudential policy actions in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations MPP2 The sum of all macroprudential policy actions in the database over country k in a year t bounded to interval 1to1 , lagged. MaPPED, ECB, Own calculations (continued on next page) Table A1. (continued) Instrument group Instruments Limits on credit growth and volume Reserve requirements related to banks' liabilities Asset-based reserve requirements Levies/taxes on financial institutions Tax on assets/liabilities Tax on financial activities Liquidity re- quirements and limits on cur- rency and matu- rity mismatch Loan to deposit (LTD) ratios Other stable funding req. incl. Net Stable Funding Requirement Short-term liquidity coverage ratios incl. Liquidity Coverage Ratio Liquidityratiosanddepositcoverageratios Limits on FX mismatches Other liquidity requirements Limits on large ex- posures and concentration Single client exposure limits Intragroup exposures limits Sectorandmarketsegmentexposurelimits Funding concentration limits Limits on qualified holdings outside financial sector Other exposure and concentration limits (continued on next page) Table A1. (continued) Instrument group Instruments Other measures Structural measures Margin requirements Other regulatory restrictions on financial activities Limits on deposit rates Debt resolution policies Crisis management tools Changes in regulatory framework Other Source:Budnik,K.& Kleibl,J.(2018). Macroprudential regulation in the European Union in 1995e2014: introducing a new data set on policy actions of a macroprudential nature. Macroprudential Policies Evaluation Database (MaPPED) 228 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 Table A2. (continued) Variable Description Source MPP3 Thesumofallmacroprudentialtighteningpolicyactions in the database over country k in a year t bounded to interval 0 to 1, lagged. MaPPED, ECB, Own calculations MPP4 The sum of all macroprudential loosening policy actions in the database over country k in a year t bounded to interval 0 to 1, lagged. MaPPED, ECB, Own calculations MPP5 Thesumofallmacroprudentialtighteningpolicyactions in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations MPP6 The sum of all macroprudential loosening policy actions in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations Credit The sum of all macroprudential credit related policy actionsinthedatabaseovercountrykinayear t,lagged. MaPPED, ECB, Own calculations Market Liquidity The sum of all macroprudential liquidity related policy actionsinthedatabaseovercountrykinayear t,lagged. MaPPED, ECB, Own calculations Concentration The sum of all macroprudential policy actions aiming to limit different bank exposures in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations Resilience The sum of all macroprudential policy actions aimed to strengthen the resilience of financial sector in the data- base over country k in a year t, lagged. MaPPED, ECB, Own calculations Credit growth The sum of all macroprudential policy actions related with reserve requirements in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations Lending caps The sum of all macroprudential policy actions aimed to limit excessive lending in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations Risk weights The sum of all macroprudential policy actions related withriskweighinginbanksinthedatabaseovercountry k in a year t, lagged. MaPPED, ECB, Own calculations Liquidity The sum of all macroprudential policy actions aimed to strengthentheliquiditypositionofbanksinthedatabase over country k in a year t, lagged. MaPPED, ECB, Own calculations Exposures The sum of all macroprudential policy actions aimed to limit different bank exposures in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations MCR The sum of all macroprudential policy actions aimed at capital position of banks in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations Capital buffers The sum of all countercyclical macroprudential policy actions aimed at capital positions of banks in the data- base over country k in a year t, lagged. MaPPED, ECB, Own calculations Taxes The sum of all macroprudential policy actions related with taxation of bank assets/liabilities or other financial activities in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations Provisioning The sum of all macroprudential policy actions related with provisioning in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations Leverage ratio The leverage ratio policy actions in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations Other requirements The sum of all other macroprudential policy actions in the database over country k in a year t, lagged. MaPPED, ECB, Own calculations ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 229 Table A3. Macroprudential indices based on the target of the measures. (1) (2) (3) (4) Credit Liquidity measures Concentration Resilience Dependent variable (lag) 0.723*** 0.720*** 0.727*** 0.723*** (6.81) (7.51) (7.03) (7.04) Index (lag) ¡0.00497 ¡0.00793 ¡0.0196** ¡0.00594 (-0.75) (-0.55) (-2.35) (-1.12) Size (total assets, logs) 0.246** 0.247*** 0.243** 0.246*** (2.52) (2.82) (2.54) (2.61) Liquidity ratio 0.00106** 0.00111** 0.00112** 0.00104* (-1.96) (-2.07) (-2.19) (-1.92) Tier 1 ratio 0.00778*** 0.00787*** 0.00768*** 0.00772*** (-2.87) (-3.29) (-2.88) (-3.00) Loans to deposits (logs) 0.106** 0.107** 0.105** 0.105** (2.31) (2.57) (2.35) (2.40) Market share 0.00343* 0.00372** 0.00339** 0.00344** (1.96) (2.20) (2.01) (1.97) D Policy rate 0.0360** 0.0369** 0.0283* 0.0348** (-2.09) (-2.27) (-1.72) (-2.14) GDP growth (lag) 0.00612 0.00593 0.00660 0.00596 (1.34) (1.41) (1.53) (1.34) Observations 16,065 16,065 16,065 16,065 Instruments 82 82 82 82 AR (1) 0.109 0.105 0.107 0.107 AR (2) 0.378 0.357 0.385 0.394 Hansen 0.644 0.656 0.604 0.678 Note: The dependent variable is gross loans in logs. Estimation method is dynamic twostep system generalized method of moments (GMM)estimatorwithrobuststandarderrorsandWindmeijer'scorrection. Laggeddependentvariableistreatedas endogenous,andall other variables as exogenous. All regressions include year fixed effects. T statistics is reported in parentheses. All regressions include weightsbasedonthenumberofobservationsof eachcountry. Macroeconomicvariablesarelaggedone period,whileallvariables,apart from indicator variables, are winsorized 1% on both tails of the distribution. The following are p-values which indicate the significance level of coefficients: *p < 0.10, **p < 0.05, ***p < 0.01. 230 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 Table A4. Macroprudential subindices based on the target of the measures. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Credit growth Lending caps Risk weights Liquidity measures Concentration MCR Capital buffers Taxes Provisioning LEV Other requirements Dependent variable (lag) 0.727*** 0.730*** 0.712*** 0.718*** 0.724*** 0.719*** 0.717*** 0.722*** 0.720*** 0.721*** 0.722*** (7.11) (6.61) (5.44) (7.41) (6.80) (6.75) (6.88) (6.82) (6.79) (6.89) (7.34) Index (lag) 0.0276** ¡0.0260** ¡0.00919 ¡0.00837 ¡0.0198** 0.00457 ¡0.0120 ¡0.0114 ¡0.0218 0.103 ¡0.00908 (-0.64) (2.55) (-1.99) (-0.65) (-0.59) (-2.36) (0.60) (-1.61) (-0.56) (-1.42) (1.21) Size (total assets, logs) 0.240** 0.239** 0.256** 0.248*** 0.244** 0.248** 0.250*** 0.245** 0.247** 0.247** 0.245*** (2.55) (2.34) (2.15) (2.79) (2.47) (2.53) (2.59) (2.51) (2.53) (2.55) (2.70) Liquidity ratio 0.00105** 0.000984* 0.00104* 0.00113** 0.00114** 0.00109** 0.00110** 0.00108** 0.00102* 0.00109** 0.00109** (-1.98) (-1.87) (-1.76) (-2.15) (-2.25) (-2.07) (-2.08) (-2.04) (-1.88) (-2.05) (-2.05) Tier 1 ratio 0.00768*** 0.00774*** 0.00828** 0.00791*** 0.00772*** 0.00788*** 0.00788*** 0.00772*** 0.00769*** 0.00784*** 0.00779*** 0.240** (-2.76) (-2.34) (-3.25) (-2.78) (-2.91) (-3.01) (-2.94) (-2.97) (-3.00) (-3.14) Loans to deposits (logs) 0.105** 0.103** 0.108* 0.108** 0.106** 0.107** 0.107** 0.106** 0.107** 0.107** 0.106** (2.41) (2.18) (1.94) (2.56) (2.30) (2.33) (2.38) (2.34) (2.35) (2.37) (2.51) Market share 0.00401** 0.00380** 0.00367* 0.00424*** 0.00389** 0.00402*** 0.00408*** 0.00398** 0.00397** 0.00401*** 0.00406*** (2.55) (2.44) (1.79) (2.80) (2.53) (2.58) (2.63) (2.52) (2.57) (2.59) (2.62) D Policy rate 0.0358** 0.0380** 0.0365* 0.0366** 0.0279* 0.0375** 0.0357** 0.0359** 0.0386** 0.0366** 0.0357** (-2.01) (-2.10) (-1.74) (-2.22) (-1.65) (-2.10) (-2.06) (-2.08) (-2.39) (-2.11) (-2.20) GDP growth (lag) 0.00440 0.00597 0.00562 0.00563 0.00632 0.00508 0.00533 0.00545 0.00530 0.00523 0.00545 (0.87) (1.29) (1.08) (1.35) (1.46) (1.07) (1.17) (1.21) (1.17) (1.16) (1.23) Observations 16,065 16,065 16,065 16,065 16,065 16,065 16,065 16,065 16,065 16,065 16,065 Instruments 82 82 82 82 82 82 82 82 82 82 82 AR (1) 0.109 0.112 0.112 0.105 0.108 0.108 0.108 0.108 0.106 0.108 0.107 AR (2) 0.354 0.320 0.383 0.359 0.389 0.380 0.385 0.386 0.417 0.375 0.379 Hansen 0.497 0.637 0.589 0.660 0.604 0.618 0.604 0.643 0.727 0.650 0.682 Note:Thedependentvariableisgrossloansinlogs.Estimationmethodisdynamictwostepsystemgeneralized methodofmoments(GMM)estimatorwith robuststandard errorsand Windmeijer'scorrection.Laggeddependentvariableistreatedasendogenous,andallothervariablesasexogenous.Allregressionsincludeyearfixedeffects.Tstatisticsisreportedin parentheses.Allregressionsincludeweightsbasedonthenumberofobservationsofeachcountry.Macroeconomicvariablesarelaggedoneperiod,whileallvariables,apartfromin- dicatorvariables,arewinsorized1%onbothtailsofthedistribution.Thefollowingarep-valueswhichindicatethesignificancelevelofcoefficients:*p<0.10,**p<0.05,***p<0.01. ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 231 Table A5. Loosening cycles of different macroprudential indices based on the target of the measures. (1) (2) (3) (4) Credit Liquidity measures Concentration Resilience Dependent variable (lag) 0.731*** 0.726*** 0.718*** 0.732*** (7.34) (7.39) (6.61) (8.25) Index (lag) 0.0370** 0.0294 0.0798*** 0.0933*** (2.53) (1.57) (3.30) (3.58) Size (total assets, logs) 0.251*** 0.254*** 0.262** 0.249*** (2.65) (2.74) (2.56) (2.97) Liquidity ratio 0.00102* 0.00106* 0.00115* 0.00100* (-1.66) (-1.71) (-1.80) (-1.81) Tier 1 ratio 0.00749*** 0.00772*** 0.00784*** 0.00753*** (-2.99) (-3.12) (-2.78) (-3.47) Loans to deposits (logs) 0.100** 0.102** 0.107** 0.0981*** (2.46) (2.54) (2.34) (2.72) Market share 0.000249 0.000118 0.0000429 0.000210 (-0.49) (-0.24) (-0.08) (-0.47) D Policy rate 0.0372** 0.0392** 0.0316* 0.0336** (-2.04) (-2.27) (-1.68) (-1.98) GDP growth (lag) 0.00800** 0.00693 0.00726 0.00839** (2.01) (1.59) (1.63) (2.39) Observations 16,065 16,065 16,065 16,065 Instruments 82 82 82 82 AR (1) 0.105 0.107 0.109 0.101 AR (2) 0.331 0.355 0.390 0.352 Hansen 0.622 0.653 0.745 0.788 Note: The dependent variable is gross loans in logs. Estimation method is dynamic twostep system generalized method of moments (GMM)estimatorwithrobuststandarderrorsandWindmeijer'scorrection.Laggeddependentvariableistreatedasendogenous,andall other variables as exogenous. All regressions include year fixed effects. T statistics is reported in parentheses. All regressions include weightsbasedonthenumberofobservationsofeachcountry.Macroeconomicvariablesarelaggedoneperiod,whileallvariables,apart from indicator variables, are winsorized 1% on both tails of the distribution. The following are p-values which indicate the significance level of coefficients: *p < 0.10, **p < 0.05, ***p < 0.01. 232 ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 Table A6. Loosening cycles of different macroprudential subindices based on the target of the measures. (1) (2) (3) (4) (5) (6) (7) (8) (9) (11) Credit growth Lending caps Risk weights Liquidity measures Concentration MCR Capital buffers Taxes Provisioning Other requirements Dependent variable (lag) 0.727*** 0.720*** 0.723*** 0.723*** 0.715*** 0.727*** 0.720*** 0.727*** 0.726*** 0.740*** (7.20) (7.07) (7.38) (7.24) (6.48) (7.75) (7.70) (7.26) (7.06) (8.58) Index (lag) ¡0.00114 0.104*** 0.0206 0.0295 0.0801*** 0.0113 0.107*** 0.0712 0.0717** 0.0693* (-0.06) (3.09) (0.90) (1.56) (3.32) (0.26) (3.83) (1.58) (2.07) (1.89) Size (total assets, logs) 0.253*** 0.260*** 0.257*** 0.256*** 0.264** 0.254*** 0.259*** 0.253*** 0.254*** 0.240*** (2.64) (2.69) (2.76) (2.70) (2.52) (2.85) (2.92) (2.66) (2.61) (2.95) Liquidity ratio 0.00107* 0.00112* 0.00113* 0.00107* 0.00116* 0.00109* 0.00110* 0.00109* 0.000937 0.00100* (-1.72) (-1.74) (-1.74) (-1.72) (-1.82) (-1.79) (-1.83) (-1.75) (-1.51) (-1.74) Tier 1 ratio 0.00771*** 0.00786*** 0.00786*** 0.00780*** 0.00791*** 0.00773*** 0.00784*** 0.00770*** 0.00766*** 0.00738*** (-3.05) (-3.25) (-3.14) (-3.09) (-2.75) (-3.35) (-3.39) (-3.08) (-3.00) (-3.47) Loans to deposits (logs) 0.102** 0.108** 0.106** 0.103** 0.108** 0.102*** 0.104*** 0.102** 0.102** 0.0978*** (2.48) (2.40) (2.55) (2.53) (2.33) (2.60) (2.64) (2.49) (2.46) (2.73) Market share 0.0000289 0.0000109 0.000105 0.0000495 0.0000446 0.0000328 0.0000555 0.0000380 0.0000609 0.0000644 (-0.07) (-0.02) (-0.23) (-0.11) (0.10) (-0.08) (0.12) (-0.09) (-0.14) (-0.16) D Policy rate 0.0391** 0.0459*** 0.0383** 0.0392** 0.0316* 0.0356* 0.0308* 0.0398** 0.0398** 0.0402** (-2.16) (-2.63) (-2.14) (-2.23) (-1.65) (-1.89) (-1.84) (-2.20) (-2.13) (-2.45) GDP growth (lag) 0.00683 0.00641 0.00748* 0.00673 0.00710 0.00686 0.00795** 0.00712* 0.00624 0.00821** (1.59) (1.50) (1.76) (1.53) (1.58) (1.64) (2.01) (1.71) (1.42) (2.26) Observations 16,065 16,065 16,065 16,065 16,065 16,065 16,065 16,065 16,065 16,065 Instruments 82 82 82 82 82 82 82 82 82 82 AR (1) 0.108 0.105 0.107 0.107 0.109 0.108 0.108 0.106 0.108 0.104 AR (2) 0.349 0.297 0.357 0.355 0.390 0.352 0.353 0.360 0.333 0.320 Hansen 0.666 0.659 0.553 0.666 0.757 0.630 0.750 0.673 0.760 0.675 Note:Thedependentvariableisgrossloansinlogs.Estimationmethodisdynamictwostepsystemgeneralized methodofmoments(GMM)estimatorwith robuststandard errorsand Windmeijer'scorrection.Laggeddependentvariableistreatedasendogenous,andallothervariablesasexogenous.Allregressionsincludeyearfixedeffects.Tstatisticsisreportedin parentheses.Allregressionsincludeweightsbasedonthenumberofobservationsofeachcountry.Macroeconomicvariablesarelaggedoneperiod,whileallvariables,apartfromin- dicatorvariables,arewinsorized1%onbothtailsofthedistribution.Thefollowingarep-valueswhichindicatethesignificancelevelofcoefficients:*p<0.10,**p<0.05,***p<0.01. ECONOMIC AND BUSINESS REVIEW 2021;23:207e233 233