Essays on the European Banks Financial Stability, Profitability, and Efficiency

DOCTORAL THESIS ESSAYS ON THE EUROPEAN BANKS FINANCIAL STABILITY, PROFITABILITY, AND EFFICIENCY José Fernando da Silva Neto ESCOLA DE DOUTORAMENTO INTERNACIONAL PROGRAMA DE DOUTORAMENTO EN ECONOMÍA E EMPRESA SANTIAGO DE COMPOSTELA 2021 DECLARACIÓN DEL AUTOR DA TESE D. José Fernando da Silva Neto Título da tese: Essays on the European Banks Financial Stability, Profitability, and Efficiency Presento a miña tese, seguindo o procedemento axeitado ao Regulamento, e declaro que: 1) A tese abarca os resultados da elaboración do meu traballo. 2) De selo caso, na tese faise referencia ás colaboracións que tivo este traballo. 3) Confirmo que a tese non incorre en ningún tipo de plaxio doutros autores nin de traballos presentados por min para a obtención doutros títulos. 4) A tese é a versión definitiva presentada para a súa defensa e coincide coa versión enviada en formato electrónico. E comprométome a presentar o Compromiso Documental de Supervisión no caso de que o orixinal non estea na Escola. En Santiago de Compostela, 06 de Xullo de 2021 AUTORIZACIÓN DO DIRECTOR / TITOR DA TESE Dna. Maria Célia López Penabad En condición de: Titora e directora Título da tese: Essays on the European Banks Financial Stability, Profitability, and Efficiency INFORMA: Que a presente tese, correspóndese co traballo realizado por D. José Fernando da Silva Neto, baixo a miña dirección e titorización, e autorizo a súa presentación, considerando que os requisitos esixidos no Regulamento de Estudos de Doutoramento da USC, e que como directora e titora desta non incorre nas causas de abstención establecidas na Lei 40/2015. De acordo co indicado no Regulamento de Estudos de Doutoramento, declara tamén que a presente tese de doutoramento é idónea para ser defendida en base á modalidade Monográfica con reprodución de publicacións, nos que a participación do doutorando foi decisiva para a súa elaboración e as publicacións se axustan ao Plan de Investigación. En Santiago de Compostela, 06 de Xullo de 2021 AUTORIZACIÓN DO DIRECTOR DA TESE Dna. Ana Iglesias Casal En condición de: Directora Título da tese: Essays on the European Banks Financial Stability, Profitability, and Efficiency INFORMA: Que a presente tese, correspóndese co traballo realizado por D. José Fernando da Silva Neto, baixo a miña dirección, e autorizo a súa presentación, considerando que os requisitos esixidos no Regulamento de Estudos de Doutoramento da USC, e que como directora desta non incorre nas causas de abstención establecidas na Lei 40/2015. De acordo co indicado no Regulamento de Estudos de Doutoramento, declara tamén que a presente tese de doutoramento é idónea para ser defendida en base á modalidade Monográfica con reprodución de publicacións, nos que a participación do doutorando foi decisiva para a súa elaboración e as publicacións se axustan ao Plan de Investigación. En Santiago de Compostela, 06 de Xullo de 2021 ABSTRACT As a result of the financial crisis of 2007-2008, all banking systems in Europe experienced a complete overhaul. These major changes justify the motives for this research, which has been based on a set of essays on the European banking sector. Firstly, we analyse the effects of bank competition on a bank’s financial stability. Using a sample of 117 listed banks from 16 Western European countries from the period 2011 - 2018, the main findings indicate that an excessive increase in competition tends to generate financial instability, especially in countries where banking systems have low financial stability. Secondly, the effect of the implementation of negative interest rate policies on the profitability and risk of banks is evaluated. Considering a sample of 2,596 banks from 29 European countries in the period 20112019, the results obtained have led us to conclude that the implementation of a negative interest rate policy reduces the net interest margin and the profitability of most banks, but it does not lead to the adoption of investment strategies with high-risk exposure. However, these conclusions do not apply to all banks, which differ according to the business models adopted by each bank. Finally, the impact of adopting socially responsible policies on banking efficiency is analysed. Based on a sample of 108 listed banks from 21 European countries during the period 2011 - 2019, it is concluded from the evidence of a U-shaped relationship between corporate social performance and banking efficiency that banks with good performance in the social area and with high-quality governance models are the most efficient. KEY WORDS Financial stability; bank competition; negative interest rates, profitability, corporate social responsibility, bank efficiency tradicional, permitindo ter en conta a persistencia na relación entre a competencia bancaria e a toma de risco e utilizando estimadores máis eficientes. Segundo a literatura existente, existen dúas visións diametralmente opostas sobre a relación existente entre a competencia bancaria e a estabilidade financeira dun banco. A literatura bancaria tradicional que apoia a hipótese da “competencia-fraxilidade”, segundo a cal un aumento da competencia entre bancos reduce a marxe financeira e as comisións cobradas polos bancos nos servizos prestados, facendo presión á baixa sobre os resultados, aumentando a probabilidade de insolvencia dos bancos e, consecuentemente, poñendo en perigo a estabilidade do sistema bancario (Marcus, 1984; Keeley, 1990; Allen & Gale, 2004). Boyd e De Nicoló (2005) presentan argumentos que sustentan a hipótese da “competencia-estabilidade”, segundo a cal nun mercado pouco competitivo os bancos tenderán a cobrar taxas de xuros máis elevadas, o que incentivará aos prestatarios a investir en proxectos de risco máis elevado aumentando a probabilidade de incumprimento destes, deteriorando a calidade da carteira de crédito dos bancos. Por tanto, segundo esta visión, un aumento da competencia, diminuirá a exposición dos bancos ao risco de crédito, aumentando a estabilidade do sistema bancario. Martinez-Miera e Repullo (2010) presentaron un modelo que pretende conciliar as dúas visións opostas sobre a relación entre a competencia e a estabilidade financeira, defendendo unha relación en forma de U. A partir da revisión da literatura foron establecidas as seguintes hipóteses de investigación: H1: A competencia bancaria diminúe a estabilidade bancaria, o que apoia a visión “competencia-fraxilidade”, e H2: Existe unha relación en forma de U entre competencia bancaria e a toma de risco dun banco. Coa finalidade de profundar no tema, analizouse se a relación obxecto de estudo podía diferenciarse segundo o banco actúase nun sistema bancario máis ou menos estábel en conxunto, resultando de aquí a terceira hipótese de investigación H3: A relación entre a competencia bancaria e a toma de risco dun banco diferénciase segundo o banco opere nun sistema bancario máis ou menos estable no seu conxunto. Para cuantificar o risco individual dun banco foron consideradas dúas medidas de mercado, Distance-to-Default e Distance-to-Insolvency, e unha medida contable, o Z-score. Para medir a intensidade competitiva do mercado onde o banco opera considerouse o Índice de Lerner que permite medir a capacidade do banco para manter os seus prezos por enriba do custo marxinal. En termos metodolóxicos, a relación foi estimada empregando un modelo de datos de panel dinámico e o estimador do método dos momentos xeneralizado, proposto por Arellano and Bover (1995) e Blundell and Bond (1998), co fin de controlar eventuais problemas de endoxeneidade. A análise empírica baseouse nunha mostra de 117 bancos cotizados, procedentes de 16 países de Europa Occidental, que abrangue o período comprendido entre 2011 e 2018. Os resultados obtidos permitiron concluír que o poder de mercado, medido polo Índice Lerner, aumenta a estabilidade financeira dun banco, o que corrobora a visión tradicional da "competencia-fraxilidade" e que a relación entre competencia e estabilidade financeira só é significativa en bancos que operan nun país cun sistema bancario menos estable. Tamén a evidencia permite concluír que os bancos con maior dimensión, mellor capitalizados e con fontes de ingresos máis diversificadas son máis estables. As conclusións extraídas neste segundo capitulo permiten establecer algunhas recomendacións para as autoridades políticas e reguladoras do sector bancario en Europa. En primeiro lugar, as políticas públicas deben garantir un certo nivel de competencia bancaria, porque esta é esencial para o incremento do benestar da sociedade en xeral, mais limitando a asunción excesiva de riscos bancarios, especialmente en países con sistemas bancarios menos estables financeiramente. Isto significa que calquera medida que supoña aumentar a competencia na banca europea debe ir acompañada por regulamentación que garanta a estabilidade financeira dos bancos, por exemplo, a través do aumento dos requirimentos de capital e limitando a exposición a varios tipos de risco aos que a actividade bancaria está suxeita. En segundo lugar, o fomento de políticas que promovan a consolidación do sector bancario europeo, permitirá a formación de bancos máis sólidos e resistentes sen comprometer a competencia. Xuntamente coas fusións domésticas, as autoridades europeas e os distintos gobernos nacionais deben promover fusións transfronteirizas para afondar na integración e na construción dun sector bancario verdadeiramente europeo. Para acadar este obxectivo, a Unión Bancaria Europea, iniciada en 2014, pode desempeñar un papel esencial. No terceiro capítulo analízase o efecto da aplicación de políticas de taxas de xuro negativas por parte dos bancos centrais sobre a rendibilidade e o risco da banca europea. Na última década, nun intento de evitar a deflación e estimular o crecemento económico, un número considerable de bancos centrais a nivel mundial aplicaron un conxunto de políticas monetarias expansionistas facendo uso de instrumentos non convencionais entre os cales destacan os programas de compra de activos a gran escala e a cobranza de xuros negativos sobre os excedentes de reservas que as institucións de crédito manteñen depositadas nos seus respectivos bancos centrais. En Europa este tipo de políticas fíxose sentir con elevada intensidade xa que seis bancos centrais situaron as súas taxas de xuro oficiais en valores negativos, xustificando desta forma, a investigación do efecto que estas políticas poden ter sobre a rendibilidade e o risco da actividade bancaria. A principal contribución deste ensaio á literatura existente reside no feito de investigar se os efectos das taxas de xuros negativas sobre a rendibilidade e a asunción de riscos dos bancos son diferentes segundo o modelo de negocio adoptado polo banco. Segundo a literatura revisada, a adopción de políticas de taxas de xuro negativas pode ter efectos contrarios sobre a rendibilidade e a estabilidade financeira dun banco. Taxas de xuro baixas ou menos negativas teñen un impacto positivo sobre a rendibilidade da banca vía ganancias de capital e redución das provisións para crédito vencido, dada a menor probabilidade de incumprimentos dos prestatarios (Boungou, 2019). Con todo, taxas de xuro baixas ou negativas, sobre todo por un prazo longo, poden tamén provocar unha baixada na marxe financeira afectando, desa forma, negativamente á rendibilidade do banco. Iso acontece porque a actividade de intermediación financeira está baseada na marxe financeira que ven dada pola diferenza entre as taxas de xuro dos préstamos e dos depósitos dos clientes. Cando as taxas de xuro se aproximan a cero, os bancos poden ter que axustar á baixa as taxas dos préstamos por cuestións de competencia, pero poden ser reacios a levar as taxas de xuro dos depósitos a valores negativos, sobre todo as dos depósitos minoristas, resultando de aí unha diminución da marxe financeira (Claessens et. al., 2018)). Estes argumentos permitiron formular as seguintes hipóteses de investigación; H1: A implementación de políticas de taxas de xuro negativas levou a un descenso na marxe financeira e da rendibilidade dos bancos na Europa, e H2: O efecto dunha variación das taxas de xuro sobre a marxe financeira e sobre a rendibilidade é máis pronunciado cando as políticas de taxas de xuro negativas están implementadas. En canto aos efectos que a aplicación de políticas de taxas de xuro negativas poden ter sobre a estabilidade financeira dun banco, a literatura tamén presenta argumentos que apuntan en sentidos opostos. Por un lado, unha baixada das taxas de xuro de referencia afecta á percepción e/ou á tolerancia ao risco por parte dos xestores bancarios, aumenta o valor dos activos e das garantías asociadas aos préstamos, aumentando a capacidade dos bancos de asumir riscos (Borio & Zhu, 2008). Por outro lado, se as taxas de xuros permanecen baixas ou negativas por un longo período, será grande a probabilidade dunha forte expansión do crédito, levando aos bancos a relaxar os seus estándares de concesión de préstamos e a aumentar o crédito a clientes con máis risco (Chen et al., 2017). Nun contexto de taxas de xuros baixas ou negativas, se os obxectivos de rendibilidade dos bancos son ríxidos, isto podería levar aos xestores bancarios a investir en activos de maior risco na "procura de rendemento" (Rajan, 2005). Estes dous últimos argumentos levaron á formulación das seguintes dúas hipóteses de investigación; H3: A implementación de políticas de taxas de xuro negativas levou a unha maior asunción de risco, e H4: O efecto dunha variación negativa das taxas de xuro sobre o aumento da asunción de risco é máis pronunciado cando as políticas de taxas de xuro negativas están implementadas. Varios autores tamén suxiren que o efecto da implementación dunha política de taxas de xuro negativas na rendibilidade e na asunción de riscos por parte dun banco depende de características específicas do banco, como a súa dimensión, a súa estrutura de financiamento e investimento e liña de produtos e/ou servizos ofrecidos aos seus clientes. Así, é natural que un entorno de taxas de xuro baixas ou negativas teña afectado aos bancos de forma diferenciada segundo o seu modelo de negocio (Molyneux et al., 2019; Boungou, 2019). Este argumento permitiu, por último, formular as seguintes hipóteses de investigacións adicionais; H5: Os efectos da implementación de políticas de taxas de xuro negativas sobre a rendibilidade dun banco dependen do modelo de negocio adoptado, e H6: Os efectos da implementación de políticas de taxas de xuro negativas sobre a asunción de riscos dun banco dependen do seu modelo de negocio. Para examinar as hipóteses de investigación enunciadas, a rendibilidade dos bancos foi medida mediante a marxe financeira e a rendibilidade do activo dos bancos, mentres que para cuantificar a asunción de riscos consideráronse tres indicadores: o Z-score, a ratio de incumprimento nos préstamos concedidos polo banco e finalmente a ratio entre os activos ponderados polo risco e o activo líquido do banco. O entorno das taxas de xuro foi recollido mediante unha taxa de xuro de curto prazo, o descenso da curva de rendementos e unha variable dummy para reflexar a implementación ou non dunha política de taxas de xuro negativas. Empregouse unha mostra de datos de panel composta por 2.596 bancos, de 29 países europeos, no período de 2011 a 2019, na que as variables endóxenas, rendibilidade e risco, foron regresadas, usando un estimador de efectos fixos, contra as variables que recollen o entorno das taxas de xuro e outras variables de control. Para mitigar un posible nesgo causado pola endoxeneidade, foi considerado un modelo no que todas as variables explicativas foron desfasadas un período e incluídos efectos fixos para cada banco e para cada ano da mostra. A identificación dos modelos de negocio existentes na banca europea foi feita mediante a análise de clusters, realizándose agrupacións en función da estrutura de activos e do financiamento de cada banco. Esta análise permitiu identificar catro modelos de negocio diferenciados na banca europea: un primeiro grupo de bancos con actividade orientada cara ao por menor, un segundo grupo con actividade baseada no mercado monetario interbancario e os dous últimos grupos orientados cara a actividade típica da banca de investimento. As principais conclusións deste terceiro capítulo indican que a implementación dunha política de taxas de xuro negativas reduciu a marxe financeira e a rendibilidade global, medida pola rendibilidade do activo, da maioría dos bancos europeos. Tamén foi encontrada evidencia de que unha diminución das taxas de xuro de curto prazo provoca unha caída máis pronunciada na rendibilidade dos bancos cando as taxas de xuro xa se encontran en valores negativos. A análise permitiu igualmente concluír que a implementación dese tipo de políticas non levou á adopción de estratexias de investimento cunha maior exposición ao risco. Non obstante, estas conclusións non son transversais a todos os bancos, diferindo segundo o modelo de negocio utilizado polo banco. Así, comprobouse que a marxe financeira dos bancos, cuxo modelo de negocio baséase no financiamento a través de depósitos captados no mercado minorista, viuse afectada mais negativamente pola aplicación de políticas de taxas de xuro negativas que nos restantes casos. En canto á toma de risco, concluíuse que os bancos pertencentes ao grupo que se financia no mercado monetario interbancario e a un dos grupos centrados na banca de investimento, adoptaron estratexias de investimento máis arriscadas, mentres que os bancos pertencentes ao grupo orientado ao por menor adoptaron estratexias de investimento menos arriscadas. Non obstante, a adopción de estratexias de investimento máis arriscadas non ten repercusións en termos de risco de crédito e estabilidade financeira en ningún dos modelos de negocio identificados. Os resultados obtidos suxiren que as entidades de supervisión e regulación do sector bancario europeo controlen de preto os efectos dun entorno caracterizado por taxas de xuro negativas que, a longo prazo, parece deprimir a rendibilidade dos bancos poñendo en risco a súa estabilidade financeira. O cuarto capítulo analiza, no contexto europeo, o impacto da adopción de políticas socialmente responsables na eficiencia bancaria. Nun mercado global e competitivo, os bancos, como outras organizacións, buscan presentarse como organizacións socialmente responsables. Nun contexto de globalización empresarial, caracterizado pola contaminación ambiental nalgunhas partes do planeta e cada vez máis marcado pola escaseza de recursos, os bancos, como outras grandes empresas, son presionados para que xestionen os seus negocios de forma máis responsable socialmente (Gao, 2009). A crise financeira de 2007-2008 provocou que moitos gobernos de Europa inxectaran fondos dos contribuíntes para reforzar a solvencia dos bancos e conter a propagación do risco sistémico. Este feito fixo que a opinión pública pasase a examinar a actividade desenvolvida polos bancos dun xeito mais pormenorizado, xustificando pola súa banda maiores esforzos para recuperar a credibilidade corporativa e a confianza dos seus clientes (Pérez et al., 2013). A responsabilidade social corporativa converteuse nunha ferramenta esencial para que os bancos restablezan a súa reputación na sociedade en xeral. Como resultado, os bancos aumentaron as súas prácticas de responsabilidade social, reforzando a súa credibilidade e a confianza que seus stakeholders depositan neles (Coulson, 2009). Esta crecente preocupación polo desenvolvemento de organizacións sostibles levou a moitos académicos a investigar se o desempeño social corporativo, como medida da responsabilidade social corporativa, ten un impacto positivo no desempeño financeiro da empresa. No sector bancario, os estudos existentes demostran que o desempeño social corporativo ten impacto no desempeño financeiro dos bancos (Esteban-Sanchez et al., 2017; Bätae et al., 2021), no valor de mercado das súas accións (Miralles-Quirós et al., 2019; Azmi et al., 2021) e no seu risco financeiro (Neitzert & Petras, 2019). Esta investigación, distínguese da maioría das anteriores, analizando o efecto do desempeño social corporativo, e de cada unha das súas dimensións (ambiental, social e governanza) considerada illadamente, sobre a eficiencia na banca europea. A revisión da literatura levada a cabo permitiu concluír que existen dúas visións antagónicas sobre o efecto que o investimento en políticas socialmente responsábeis ten sobre o desempeño financeiro dunha organización. A Teoría Neoclásica defende que a empresa debe aplicar os seus recursos escasos en actividades que teñen como obxectivo a maximización do beneficio, actuando de acordo coas regras básicas, leis e costumes xeralmente aceptadas pola sociedade (Friedman, 1970). Segundo esta visión, os accionistas son vistos como os principais stakeholders da empresa e, por iso, os recursos deben ser asignados para satisfacer a este grupo. A posición Neoclásica, por tanto, sostén que a xestión da empresa débese preocupar só dos intereses de seus propietarios ou accionistas (Miralles-Quirós et al., 2019). Segundo a Teoría da Axencia, a participación en actividades ligadas á responsabilidade social corporativa é unha responsabilidade da xestión sendo o custo soportado polos accionistas. Desde esta perspectiva, o investimento en actividades asociadas á responsabilidade social corporativa ten un impacto negativo no desempeño financeiro da empresa. No campo oposto encóntrase a Teoría dos Stakeholders, desenvolvida por Freeman (1984), segundo a cal unha empresa non pertence só aos propietarios/accionistas, debéndose ter en conta tamén os intereses dos restantes axentes que gravitan na súa esfera. Neste sentido, o obxectivo da empresa non debe ser maximizar o valor para os accionistas, senón crear valor para todas as partes interesadas incluíndo os empregados, os consumidores, as comunidades locais, os recursos naturais ou ambientais (Post et al., 2002). Con base nestas dúas visións antagónicas estableceuse a primeira hipótese de investigación do cuarto capítulo; H1: un bo desempeño social corporativo aumenta a eficiencia bancaria. Como referido por Xie et al. (2019), as actividades ligadas á responsabilidade social corporativa son resultado das políticas de xestión e de obrigacións legais e comprenden distintas dimensións: a dimensión ambiental, a dimensión social e a dimensión asociada ao modelo de governanza. Naturalmente, estas tres dimensións teñen unha contribución distinta para o desempeño social corporativo e poden impactar de forma diferenciada sobre a eficiencia bancaria. Esta observación levounos a estabelecer as seguintes 3 hipóteses de investigación; H2: un bo desempeño na dimensión ambiental da responsabilidade social corporativa aumenta a eficiencia bancaria; H3: un bo desempeño na dimensión social da responsabilidade social corporativa aumenta a eficiencia bancaria; e H4: Os bancos con bo desempeño en termos de modelo de governanza son máis eficientes. Nunha tentativa de reconciliación das dúas visións sobre o tema e en liña co defendido por Nollet et al. (2016) e Shabbir et al. (2020) foi formulada unha última hipótese de investigación a saber; H5: a relación entre o desempeño social corporativo (e cada un dos seus compoñentes considerado illadamente) e a eficiencia bancaria non é lineal. Para analizar as hipóteses de investigación formuladas considerase unha mostra de datos de panel que contén 108 bancos cotizados, de 21 países europeos, durante o período de 2011 a 2019. En termos metodolóxicos, para medir o desempeño social corporativo, e cada unha das súas dimensións illadamente, construíronse catro índices empregando un modelo de análise envolvente de datos (DEA, Data Envelopment Analysis) sen inputs explícitos. No estudo da relación entre o desempeño social corporativo e a eficiencia bancaria, considerouse o método de estimación en dúas etapas proposto por Simar e Wilson (2007). Nunha primeira etapa, a técnica non paramétrica DEA foi utilizada para estimar as puntuacións de eficiencia, considerando un modelo con dous inputs (activos fixos e custo medio da man de obra) e dous outputs (préstamos e ingresos distintos da marxe financeira), asumindo que os bancos operan baixo a hipótese de rendementos variables a escala. Na segunda etapa, un modelo de regresión truncado foi estimado por medio do algoritmo II proposto por Simar & Wilson (2007), onde a puntuación de eficiencia, da primeira etapa, é regresada contra un conxunto de variables que potencialmente poderían explicar a eficiencia do banco, incluíndo a variable relativa ao desempeño social corporativo como un todo ou relativa a cada unha das súas tres dimensións consideradas illadamente. As principais conclusións do cuarto capítulo permiten dicir que, en xeral, os bancos europeos presentan baixos niveis de eficiencia, cun valor de eficiencia técnica pura en torno ao 50%. Os resultados tamén nos permiten concluír que hai evidencia dunha relación en forma de U entre o desempeño social corporativo e a eficiencia dos bancos en Europa. En particular, constátase que os bancos con niveis intermedios de desempeño social corporativo son menos eficientes, mentres que os bancos con niveis baixos ou altos de actividades ligadas á responsabilidade social corporativa presentan mellores niveis de eficiencia. Considerando o efecto illado de cada dimensión do desempeño social corporativo, constátase que os bancos con boas performances nas áreas social e de governanza son máis eficientes. A dimensión ambiental das actividades ligadas a responsabilidade social corporativa non se revela estatisticamente significativa na explicación da eficiencia bancaria. Os resultados obtidos teñen dúas implicacións para a xestión bancaria: (i) a primeira é que a adopción de prácticas ambientais para corrixir externalidades ou disfuncións que o mercado non consegue resolver debe limitarse, unha vez que non parecen influenciar a eficiencia dos bancos; (ii) a segunda é que as boas prácticas nas dimensións social e na asociada ao modelo de governanza das actividades ligadas á responsabilidade social corporativa teñen un impacto positivo na eficiencia do banco, mais só no longo prazo. O presente traballo remata coa presentación dunha síntese das principais conclusións dos ensaios realizados, limitacións presentes e futuras liñas de investigación sobre as temáticas abordadas. À minha mulher, Paula Aos meus filhos, Lucinda e Nelson iv Table 17 - Bank financial stability and competition (sub-sample of the most stable banking systems) – RE Model ............................................................................................................... 49 Table 18 - Negative interest rate policy (NIRP) announcements ............................................ 66 Table 19 - Descriptive statistics ............................................................................................... 68 Table 20 - Effect of interest rates and NIRP on net interest margin and return on assets ....... 70 Table 21 - Effect of interest rates and NIRP on bank risk-taking ............................................ 73 Table 22 - Business model identification based on bank assets structures and their sources of financing................................................................................................................................... 75 Table 23 - Effect of interest rates and NIRP on net interest margin and return on assets by business model ......................................................................................................................... 77 Table 24 - Effect of interest rates and NIRP on bank risk-taking by business model ............. 79 Table 25 - Variables definition and data source ...................................................................... 83 Table 26 - Effect of interest rates and NIRP on net fee & commission income and net trading income ...................................................................................................................................... 84 Table 27 - Effect of interest rates and NIRP on other operating revenues and loan loss provisions ................................................................................................................................. 85 Table 28 - Descriptive statistics of potential inputs and outputs for the efficiency model .... 102 Table 29 - Spearman rank correlation coefficients ................................................................ 102 Table 30 - Categories from Thomson Reuters Eikon Asset 4 ESG database used in the estimation of CSP and each of its dimensions indexes .......................................................... 104 Table 31 - Descriptive statistics of ESG scores used in the estimation of DEA-WEI model for CPS and each of its dimensions indexes ................................................................................ 104 Table 32 - Descriptive statistics of control variables of banks’ efficiency ............................ 105 Table 33 - Bank’s efficiency score and bias-corrected efficiency score (means) by year based on the BCC model, with output orientation, using DEA ....................................................... 106 Table 34 - Descriptive statistics of banks’ efficiency and CSP and each of its dimensions .. 108 Table 35 - Results of bootstrap truncated regressions for determinants of bank’s efficiency (linear relationship assumed between CSP and bank’s efficiency) ....................................... 109 Table 36 - Results of bootstrap truncated regressions for determinants of bank’s efficiency (non-linear relationship assumed between CSP and bank’s efficiency) ................................ 111 Table 37 - Robustness test to non-linear relationship assumed between CSP and bank’s efficiency ................................................................................................................................ 116 v Table 38 - Robustness test to linear relationship assumed between CSP and bank’s efficiency ................................................................................................................................................ 117 Table 39 - Bank’s ESG activity and each of its dimensions indexes and bias-corrected indexes (means) by year based on DEA-WEI model .......................................................................... 120 vii INDEX OF FIGURES Figure 1 – Number of credit institutions in the EU-28 (2008-2018) .......................................... 6 Figure 2 – Total Assets of the EU-28 banks, in € trillions (2008-2018) .................................... 7 Figure 3 – Deposits and loans in EU-28 banks as a share of total banking assets (%) (20082018) ........................................................................................................................................... 7 Figure 4 – Bank Regulatory Capital to RWA in European banks (%) ....................................... 8 Figure 5 – Return on Equity in European banks (%) ................................................................. 8 Figure 6 - Distance-to-default, Ln Z-score and Lerner index (average) 2011-2018 ................ 28 Figure 7 - Distance-to-default and Ln Z-score against Lerner index (average) by country ..... 28 Figure 8 - European Union structural financial indicators ....................................................... 35 Figure 9 - Evolution of the central bank's policy rate in European countries that have adopted NIRP ......................................................................................................................................... 62 Figure 10 - Bank’s ESG activity and each of its dimensions bias-corrected indexes (means) ................................................................................................................................................ 107 Figure 11 - Relationship between Bank’s inefficiency and ESG, Social and Governance indexes for a representative bank ........................................................................................... 113 ix LIST OF ACRONYMS/ABBREVIATIONS CSP: Corporate Social Performance CSR: Corporate Social Responsibility DD: Distance-to-Default DEA: Data Envelopment Analysis DI: Distance-to-Insolvency DMU: Decision Making Unit ECB: European Central Bank ESG: Environmental, Social and Governance EU: European Union GDP: Gross Domestic Product GMM: Generalized Method of Moments IMF: International Monetary Fund NIRP: Negative Interest Rate Policy NPL: Non-Performing Loans ROA: Return on Assets ROE: Return on Equity RWA: Risk-Weighted Assets VRS: Variable Returns to Scale INTRODUCTION The process of financial deregulation, which began mainly in the 1980s, caused changes in the financial system of the more developed economic blocs, which include Europe. This process changed how financial institutions act, expanding financial disintermediation, which contributed to an expansion of financial innovations and the change in the architecture of the global financial system. With this, according to some authors, the process of financial deregulation increased the vulnerability of the US financial system, constituting itself as a structural cause of the subprime crisis. The 2007-2008 financial crisis and later the Eurozone sovereign debt crisis shook the stability of the European banking system, leading to the disappearance of many credit institutions and numerous public interventions by the different European governments to contain the spread systemic risk. At the same time, in the post-financial crisis, there is a strengthening of prudential regulation, a good example of which is the implementation of the Basel III Accord, which obliges banks to increase their capital reserves to face future crises. In the European Union, in 2014, we witnessed the creation of the European Banking Union based on two essential pillars: the Single Supervisory Mechanism and the Single Resolution Mechanism. The changes that have taken place will have had a positive impact on the financial stability of the European banking system, but they will also have changed the competitive conditions in which banks operate. Based on this idea, emerges the first essay of this thesis, which revisits the study of the relationship between competition in the banking sector and the financial stability of banks in Europe. The changes underwent by the European banking sector in the last decade, the fact that it plays a vital role in the good functioning of the economy, and facing new challenges, such as technological disruption, maintain interest in studying the relationship between competition and financial stability, justifying this research. In the last decade, to combat the deflation scenario and simultaneously stimulate economic growth, some central banks worldwide have implemented a series of unconventional monetary policies that have brought interest rates to negative ground. In Europe, this phenomenon has been and continues to be particularly felt, with interest rates on the interbank money market and JOSÉ FERNANDO DA SILVA NETO 2 public debt issued by most European countries remaining negative for a long period. This circumstance motivated the second investigation carried out in the present thesis. In the context of the European banking sector, we study the impact that a negative interest rate policy has on profitability and risk-taking in banks. As already mentioned, the 2007-2008 financial crisis leads many European governments to provide public aids to contain the spread of a systemic crisis in the banking sector in their countries. These public aids and the need to recover corporate credibility and customers’ trust led bank administrations to choose the adoption of socially and environmentally responsible policies as an essential tool for the development of their business. This raises the question of how it is the relationship between corporate social responsibility and banks' financial performance. In the context of the European banking sector, some studies have investigated the impact of corporate social performance on traditional financial measures such as profitability and the market value of banks. However, studies that assess the impact of corporate social performance on banking efficiency are scarce. This scarcity for the European banking sector motivated the third investigation of this thesis. To carry out the essays identified about the European banking system, this document is structured as follows. The present introduction, followed by a chapter that characterizes the European banking sector after the financial crisis, defines the objectives of the thesis and the methodology employed. The next three chapters constitute the main body of the thesis, each corresponding to an essay. The thesis ends finally with the conclusion. The first chapter presents a characterization of the European banking sector after the financial crisis, covering the period 2008-2018. In particular, aspects related to the market structure, the typology of assets and the sources of financing used by banks in Europe are analyzed. The levels of financial stability and profitability of these institutions are also addressed. This chapter includes the objectives pursued and the methodology that was used in the present research work. In chapter 2 is analysed the relationship between competition and bank risk-taking in Europe. Based on the literature review, the competition-stability and competition-fragility views are hypothesized. It is also investigated the hypothesis that the relationship between competition and risk-taking in banking is given by a U-shaped relationship. This essay extends Introduction 3 the existing literature by investigating if that nexus is differentiated depending on whether the bank operates in a weaker or more stable banking system as a whole. To measure competition is considered the Lerner index, a measure of the bank’s market power. Bank’s risk-taking is proxied by distance-to-default, a market risk’ measure, and Z-score, an accounting measure of risk. Because a market measure is used to quantify the bank's risk-taking, the sample utilized is made up of 117 listed banks from 16 European countries covering the period 2011-2018. To address the endogeneity problem between the bank competition measure and bank risk measures, the relationship between competition and bank’s risk-taking is estimated considering a dynamic panel model with a 2-step GMM estimator. To control the effects of other variables in the bank’s risk-taking is considered a set of variables at bank-level and some macroeconomic variables. This is followed by Chapter 3, where the effect of negative interest rate policies on the profitability and risk-taking of European banks is investigated. After reviewing the literature, it is investigated (i) the effects of negative interest rates on the bank’s net interest margin and the remaining components of banks' profitability; (ii) the effects of negative interest rates in bank’s risk-taking; (iii) and lastly if the referred effects are differentiated according to the bank's business model. The main contribution to the related literature of this essay lies in the fact that it is investigated whether the effects of negative interest rates on banks' profitability and risktaking are differentiated according to the business model adopted by the bank. Bank’s margin and overall profitability are proxied by net interest margin and return on assets, respectively. To measure the bank’s risk-taking three measures are considered: Z-score, as a measure of the overall bank risk, non-performing loans ratio as a measure of credit risk, and finally the riskweighted assets to total assets ratio as a measure of the risk associated with the bank's investment strategy. The interest rate environment is characterized by a short-term interest rate, 3‐month interbank money market interest rate, and the slope of the yield curve measured by the difference between 10-year Treasury yield and 3‐month interbank money market interest rate. A sample of 2596 banks, from 29 European countries, over the period 2011-2019 is considered in the study. To test the research hypotheses, static panel models are estimated using a fixedeffect estimator, where all explanatory variables are lagged one period and bank and time fixed effects are included to mitigate a possible endogeneity bias. To identify the different business models existing in European banking are used k-medians clustering based on the asset and funding structure of each bank. The cluster analysis allowed us to identify four different JOSÉ FERNANDO DA SILVA NETO 10 The last objective of this research work is to study the determinants of banking efficiency in European banks. More specifically, it seeks to investigate the effect of adopting socially responsible policies, in its environmental, social and governance dimensions, on bank efficiency levels. 1.3. RESEARCH METHODOLOGY To achieve the proposed objectives, the hypothetical-deductive method approach was applied by performing a set of essays in the European banking sector, that constitute the main body of this thesis. In this analysis, longitudinal samples from European banks, covering the period from 2011 to 2019 and parametric and non-parametric methodologies were utilized. In almost all of the essays about European banking performed in the thesis, panel data models were employed to test the research hypotheses formulated in the body of the research’s work. To achieving the first objective, about the relationship between financial stability and competitive conditions, the proposed model was estimated using the two-step “system GMM estimator” developed by Arellano and Bover (1995) and Blundell and Bond (1998), which proves to be suitable in situations of the possible presence of endogeneity. In robustness tests, the model was also estimated using a fixed and a random effect estimator with robust standard errors to account for the possible existence of autocorrelation and/or heteroscedasticity. To estimate the proposed models, with the objective of analysing the impact of negative interest rate policies in the bank’s profitability and risk-taking, a fixed effect estimator with robust standard errors was used to account for the possible existence of autocorrelation and/or heteroscedasticity. To identify the different business models adopted by European banks, cluster analysis was used. Namely, it was used k-medians clustering to assign each bank to a specific banking business model according to its asset and funding structure. Lastly, to estimate the bank’s efficiency scores a non-parametric technique was used, namely, data envelopment analysis (DEA), considering the BCC (Banker, Charnes and Cooper) model. We also used a DEA model without explicit inputs (DEA-WEI model) to estimate indices related to corporate social responsibility and each of its dimensions. The truncated regression model was estimated using algorithm II proposed by Simar & Wilson (2007). In Competition and Financial Stability in the European Listed Banks 11 robustness tests, the model also was estimated using the two-step “system GMM estimator” of Arellano and Bover (1995) and Blundell and Bond (1998). Further details on the computation of some variables, databases and the estimation methods used will be presented in the chapters of the three essays, which we will develop from now. 2. COMPETITION AND FINANCIAL STABILITY IN THE EUROPEAN LISTED BANKS Published in SAGE Open, ISSN: 2158-2440, Volume 11(3), pages 1–13, with the title: Competition and Financial Stability in the European Listed Banks Authors: Maria Celia López-Penabad (University of Santiago de Compostela, Spain); Ana Iglesias-Casal (University of Santiago de Compostela, Spain) and José Fernando Silva Neto (University of Santiago de Compostela, Spain) DOI: 10.1177/21582440211032645 2.1. INTRODUCTION The impact of bank competition on financial stability has been widely discussed in the academic and political communities over the last two decades and particularly since the 20072008 global financial crisis (Fu et al., 2014; Clark et al., 2018). During the decades of the seventies and eighties in the last century, there was an intensification of financial deregulation that promoted the globalization of financial markets and the financial innovation, which in turn led banks to adopt much more aggressive policies, increasing the degree of competition (Danisman & Demirel, 2019; Cuestas et al., 2020). For many, this excessive risk-taking behaviour by the banks was the key to the 2007-2008 crisis. This has led in Europe, as in the worldwide, in the past few years, a strengthening of prudential regulation via increased capital requirements and other obligations that incorporate aspects that can affect competition in the banking sector. Also, there was a reduction in the number of banks operating in most countries, with the troubled banks being bailed out by national governments or absorbed by other banks. These two phenomena may have modified the competitive conditions in which banks operate, relaunching the discussion about the relationship between competition in the banking sector and its financial stability in the scientific community. While it is agreed at an academic level that greater competition in the banking sector leads to greater innovation and efficiency (Schaeck & Čihák, 2010; Turk Ariss, 2010), there is still no consensus as to whether the impact of competition on the banking sector will lead to greater JOSÉ FERNANDO DA SILVA NETO 14 or lesser financial stability. The traditional banking literature supports a “competition-fragility” nexus. Under this hypothesis, bank competition will lower the net interest margin, eroding bank’s profits, which will lead to an increased probability of bankruptcy, and consequently, the overall disruption of the financial system (Marcus, 1984; Keeley, 1990; Allen & Gale, 2004). More recently, Boyd and De Nicoló (2005) present arguments that support the competitionstability hypothesis, which states that banks with more market power tend to charge higher interest rates, which provides an incentive to borrowers to engage in riskier activities. So, under this theory, competition increases financial stability. Martinez-Miera and Repullo (2010) present a model that tries to reconcile the two opposing views on the relationship between competition and financial stability of banks. Although this topic has already been investigated in the European context, this research is of particular interest because it analyses a sector in constant change and which is essential for the good functioning of the economy. The changes that took place in the different European banking sectors due to the 2007-2008 financial crisis and the regulatory changes to stabilize them have led in recent years to great restructuring that has altered the conditions of competition. This reason justifies our work, which presents the following distinctive aspects from those previously carried out. First, we emphasize the fact that the relationship between bank competition and risk-taking can be differentiated depending on whether the bank operates in a more or less stable banking system. Second, to measure the bank risk-taking, we considered a new market measure, computed with market data, and not obtained from data provider services. Finally, to account for the persistence in the relationship between banking competition and risk-taking, we consider a dynamic panel data model, instead of the traditional static model, estimated by a method that allows us to obtain more efficient estimators. Initially, as a proxy for individual bank risk, two alternative measures are considered. These measures, which have been intensively used in previous empirical investigations, are the Zscore, an accounting measure, and the distance-to-default, a market measure. In robustness tests, we also considered a third measure, distance-to-insolvency, which, to the best of our knowledge, has never been used in previous empirical research to measure the bank’s risk. To measure banking competition, we consider the Lerner index, which measures the bank's ability to keep its prices above its marginal costs. Competition and Financial Stability in the European Listed Banks 15 Using a dynamic panel data model with a Generalized Method of Moments (GMM) estimator, to control for endogeneity, the empirical analysis is carried out for 117 banks, in 16 Western European countries, between 2011-2018. The findings indicate that market power increases the bank’s financial stability, which corroborates the traditional “competitionfragility” view, and that relationship is only significant for countries with a less stable banking system. We also find evidence that banks with greater dimension, more well-capitalized and with more diversified earnings sources are more stable. The remainder of the chapter is organized as follows. Subchapter 2.2. provides a review of the literature on competition and stability in banking and formulates the research hypotheses. Subchapter 2.3. describes the econometric methodology and the data used in the econometric tests. The results are reported and discussed in Subchapter 2.4. A set of robustness tests are conducted in Subchapter 2.5. and Subchapter 2.6. concludes. 2.2. LITERATURE REVIEW AND RESEARCH HYPOTHESIS The literature on the study of the relationship between competition and stability in the banking sector is based on two different views: the competition-fragility view and the competition-stability view. According to the traditional competition-fragility hypothesis, banks become more fragile when they operate in more competitive banking systems. Over time, several arguments have been suggested to support this hypothesis. The first is based on the well-known "charter/franchise value" paradigm for bank risktaking, which states that banks limit risk-taking to protect the quasi-monopoly rents granted by their governments' charters. Marcus (1984) and Keeley (1990) provide a theoretical framework that suggests in more competitive banking systems, due to lower charter/franchise value, the bankruptcy costs are lower, leading banks to adopt riskier investment strategies deteriorating thereby the quality of the bank's assets and the financial stability. Another argument of the competition-fragility view rests on the market structure in which banks operate. More concentrated banking systems are composed by large banks that benefit from economies of scale and/or scope and have more diversified portfolios, lowering that way JOSÉ FERNANDO DA SILVA NETO 16 the risk exposure (Williamson, 1986). This argument should be taken with caution as greater banking concentration does not necessarily mean less competition in the sector. The competition-fragility hypothesis is also supported by the borrower-bank relationship. Several authors argue that in more competitive banking environments, the economic rents from intermediation decrease considerably, leading banks to reduce their screening of potential borrowers and, thus, overall portfolio credit quality declines (Chan et al., 1986; Marquez, 2002). The competition-fragility hypothesis also finds support in the fact that the existence of deposit guarantee systems to mitigate liquidity risk introduces moral hazard by providing incentives to banks to engage in riskier activities, in more competitive banking environments (Matutes & Vives, 1996). A last argument that supports the competition-fragility view is based on the fact that the stability of the banking system can also be affected by contagion. In a perfectly competitive market, banks are price takers and have no incentive to provide liquidity to troubled banks. If banks in difficulty eventually fail, this could have negative repercussions on the whole sector increasing the instability. In a more concentrated banking system, with a small number of large institutions, it is relatively easier to monitor banking activity by the supervisory authority and to obtain an agreement to rescue troubled banks, thus preventing contagion and increasing financial stability (Allen & Gale, 2000; Sáez & Shi, 2004). The alternative and more recent competition-stability hypothesis states that more competitive and/or less concentrated banking systems are more stable. The main argument of this view is based on the risk-shifting effect introduced by Boyd and De Nicoló (BDN, 2005). They developed a model based on the argument that banks operating in markets with uncompetitive banking systems tend to charge higher interest rates on loans granted. This may encourage borrowers to invest in high-risk projects, increasing the probability of default on loans. Consequently, the volume of non-performing loans may increase, resulting in a higher probability of the bank’s bankruptcy. Another argument presented by proponents of the competition-stability hypothesis is related to the doctrine "too-big-to-fail". Mishkin (1999) and Barth et al. (2012) argue that in highly concentrated banking systems, largely made up of large banks, policymakers are more Competition and Financial Stability in the European Listed Banks 17 likely to "save" these banks in case of bankruptcy. This creates a moral hazard problem, encouraging risk-taking behaviour by the bank managers and increasing financial fragility (Rosenblum, 2011; Demirgüç-Kunt & Huizinga, 2013). Most of the empirical investigations found evidence supporting the view of competitionfragility. Beck et al. (2006), in a cross-country study of 69 countries over the period 1980-1997, using concentration ratio as a measure of competition and a dummy variable indicative of a systemic crisis, found evidence that in economies with more concentrated banking systems, crises are less likely, which supports the competition-fragility view. Using data at the bank level for 23 developed countries, over the period 1999-2005, Berger et al. (2009) conclude that banks with more market power, measured by the Lerner index, present riskier loan portfolios but the overall bank risk, measured by the Z-score index, is more reduced, which supports the competition-fragility view. Evidence of this view also can be found in more recent studies (Beck et al., 2013; Leroy & Lucotte, 2017; Danisman & Demirel, 2019). Some but relatively fewer studies, using new measures for the competition, such as the Boone indicator, found evidence of the competition-stability view. Schaeck and Čihák (2010), using a panel data sample of banks from 10 European countries (covering the period 19952005) and a cross-section sample of U.S. local banks (for the year 2005), concluded that promoting competition improves banks' financial stability via efficiency channel, which supports the competition-stability view. Similar results were found by Clark et al. (2018) for a bank’s panel data set from 10 Commonwealth of Independent States countries in the period 2005-2013. They concluded that there was a statistically significant negative relationship between the Lerner index and the Z-score, which supports the competition-stability view. According to the European Banking Federation (EBF), since the financial crisis in 2008 until 2018, more than a quarter of credit institutions in the European Union have disappeared2. This downward trend gave rise to considerable bank consolidation processes in countries such as Spain, Italy and Greece. Consistent with this trend and the apparent stabilization of most banking systems in Europe in recent years, we expect, in line with the most recent empirical 2 According to the EBF (2019), the number of credit institutions in the EU-28 decreased by 28.6% from 8525 in 2008 to 6088 in 2018. See Fig. 3, in Appendix of this chapter, for detailed information by country about change in the number of credit institutions between 2008 and 2018. JOSÉ FERNANDO DA SILVA NETO 18 studies, to find results that support the competition-fragility view to the detriment of the competition-stability hypothesis. Based on this we formulate Hypothesis I below: Hypothesis I: Bank competition decreases the stability in banking, indicating the competition-fragility view. More recently Martinez-Miera and Repullo (MMR, 2010) developed a model, that assumes an imperfect correlation in the loan’s probability of default, to demonstrate the existence of a U-shaped relationship between competition and risk. Increased competition in the banking sector leads to a decrease in loans interest rates which potentially has two opposite effects on financial stability. The first is the already mentioned risk-shifting effect of the BDN model that decreases the loan portfolio risk. The second effect, defined as a "margin" effect, leads to a decrease in banks' revenues, given the reduction of interest payments by firms, which potentially increase the bank risk. MMR demonstrated that the “risk-shifting” effect dominates in markets with greater banking concentration (monopolistic markets) so that the entry of new banks in the sector can improve bank risk measures. In already highly competitive banking markets, the "margin" effect dominates in such a way that the entry of new banking entities into the sector tends to worsen bank risk. This leads the authors to conclude that the lowest degrees of bank risk occur at moderate levels of competition and so a U-shaped relationship between competition and the risk of bank failure generally obtains. In the context of European banking, despite the increase in banking concentration, quite different market structures still coexist. Countries such as Germany, Austria, Italy and France whose share of total assets of the five largest credit institutions does not exceed 50%, at the end of 2018, and countries like Greece, Netherlands and Finland where that value is greater than 80% (see Figure 8 presented in the appendix of this chapter). This diversity of market structures in European banking makes it possible to admit that both approaches, competition-fragility and competition-stability, may be appropriate, depending on the level of concentration and competition. On the other hand, a nonlinear investigation could be useful from a policy point of view, as it allows an optimal threshold to be identified beyond which bank competition, or inversely a lack of competition, becomes dangerous for the stability of the banking sector. Competition and Financial Stability in the European Listed Banks 19 Based on those arguments, we formulated the following hypothesis: Hypothesis II: There is a U-shaped relationship between competition and bank risktaking. Some recent empirical studies found evidence of Hypothesis II. Jiménez et al. (2013), using a panel data sample of commercial and savings banks from Spain, in the period 1988-2003, concluded by a nonlinear relationship between competition in the loan market and bank risktaking as in the MMR model. Empirical evidence of the U-shaped relationship between bank competition and risk-taking can also be found in the study of Cuestas et al. (2020) for banks operating in the Baltic countries over the period 2000–2014. As previously mentioned, the 2007-2008 financial crisis and the eurozone sovereign debt crisis put many European banks under severe financial stress. This led the different European governments to adopt a set of measures to stabilize their countries' banking systems. Those set of measures can be grouped into three categories and, in general, they were implemented sequentially as the crisis worsened: i) guarantees, ii) capital injections and iii) asset restructuring/resolution. In countries with strong budgetary constraints and excessive levels of public debt, such as Italy, Portugal, and Greece, the implementation of steps ii) and iii) was avoided or delayed as much as possible, resulting in an even less stable banking system. In those countries, banks will tend to have poorer results, lower capitalization levels and a lower charter value. According to charter value hypothesis, banks with a lower charter value could be encouraged to take on more risk to benefit from the deposit insurance put option (Bakkar et al., 2020). So, in countries with less stable banking systems, we could expect that, increased competition may lead management to invest in riskier assets, amplifying the level of risk. In countries with more stable banking systems, where banks present higher charter value, more competition will not increase the incentive for risk-taking as much given the higher bankruptcy costs that banks can endure if they fail. Motivated by these differences, in terms of financial stability, we investigate the influence of the stability of the banking system as a whole on the relationship between market power and bank risk-taking. In particular, we analyse the hypothesis that the relationship between competition and bank risk-taking is influenced by the fact that the bank operates in a more or JOSÉ FERNANDO DA SILVA NETO 26 authors, a higher inflation rate makes banks achieve a high return on assets but also carries a high level of risk. So, we expect that a higher inflation rate reduces the bank’s stability. Examining whether market power influences the bank’s risk-taking raises the question of endogeneity bias. Indeed, Schaeck and Čihák (2010) argued that the level of risk-taking could affect competition between banks, which could then influence our measures of market power. When banks face a high probability of default, they might have an incentive to change the price of their products to access new financial resources and attract new customers, affecting the existing market power. To address the endogeneity problem between the bank competition measure and bank risk measures, as well as capitalization levels, we estimate the Eq. (2.8) using the Generalized Method of the Moments (GMM). Namely, we use the System GMM method who estimates two equations, one in differences and one in levels, including lagged values in levels as instruments of the endogenous variables in first differences and additionally lags of the first differences in the endogenous variables as instruments for the equations in levels. In this research, we use the two-step “system GMM estimator” developed by Arellano and Bover (1995) and Blundell and Bond (1998), with Windmeijer (2005) corrected standard errors6. 2.3.4. Sample Description and Data Statistics To evaluate the effects of bank competition on financial stability in Europe after the global financial crisis, we considered an unbalanced panel data set constituted by listed European banks, that covers the period from 2011 to 2018, from the following countries: Austria(AU), Belgium(BE), Denmark(DK), Finland(FI), France(FR), Germany(GE), Greece(GR), Ireland(IR), Italy(IT), Netherlands(NL), Norway(NO), Portugal(PT), Spain(SP), Switzerland(CH), Sweden(SE) and the United Kingdom(UK)7. Accounting and stock market information of the banks is obtained from BankFocus Database by Bureau van Dijk and Datastream, respectively (all monetary data has been converted into euros). Real GDP growth and inflation rate are obtained from the World Economic Outlook Database of International Monetary Fund. After excluding banks with (1) missing, negative or zero values for the cost 6 We use the two-step GMM estimator, instead one-step GMM estimator, because is more efficient. 7 We considered Western European listed banks for which balance-sheet and market data are available over the period of study. We only considered commercial banks, savings banks, cooperative banks, and bank holdings & holding companies, with at least 3 of years consolidated accounts available. Competition and Financial Stability in the European Listed Banks 27 function needed to calculate the Lerner index, (2) missing data to estimate distance-to-default, and (3) missing Z-score values, we obtain a final sample that includes unbalanced panel data for 117 banks, with 860 bank/year observations. In the estimation of the Lerner Index (and efficiency-adjusted Lerner index) the sample used was different. As discussed previously, the translog cost function was estimated for each country to better address differences in technology. To increment the number of observations that allowed the estimation of Eq. (2.5) by country, we extended our sample to all listed and non-listed European banks for which we have consolidated data. The statistics of the variables that are used in the main regression are reported in Table 1. Table 1 - Sample descriptive statistics of variables used in the main model (2011-2018) Variable Obs. Mean Std. Dev. Min Max Dependent variables: Distance-to-default 860 3.837 3.030 -2.280 16.669 Ln Z-score 860 3.515 0.927 1.215 5.278 Distance-to-insolvency 860 3.502 2.422 -1.038 15.888 Independent variables: Lerner index 860 0.140 0.211 -0.959 0.440 E-Lerner index 860 0.236 0.212 -0.746 0.605 Capitalization 860 0.083 0.034 0.017 0.174 Size 860 10.425 2.224 5.788 14.529 Non-interest income share 860 0.428 0.167 0.060 0.919 Share of wholesale funding 860 0.380 0.192 0.007 0.763 Liquidity 860 0.185 0.128 0.023 0.607 Asset composition 860 0.608 0.186 0.133 0.875 Inefficiency 860 0.628 0.157 0.314 1.216 Real GDP growth 860 0.012 0.015 -0.040 0.045 Inflation 860 0.013 0.011 -0.011 0.036 Source: Own production Considering the results obtained for the financial stability measures, it is verified that banks included in our sample present average values of 3.837 and 3.515 for the distance-to-default and the Ln Z-score indicators, respectively. These values are substantially higher than those obtained by Leroy and Lucotte (2017) which is explained by the fact that their sample period was characterized by the occurrence of subprime crises. The sample average value of distanceto-insolvency is slightly lower than the value of the distance-to-default. Regarding the measures of bank competition, we observed mean values of 0.140 for the conventional Lerner index and JOSÉ FERNANDO DA SILVA NETO 28 0.236 for the efficiency-adjusted Lerner index, which indicates relatively low market power by the banks included in our sample. Between 2011 and 2018, the financial stability of the banks included in our sample improved considerably, with the average value of distance-to-default increasing, during the entire period, about 109% from 1.56 to 3.26, while the average value of Z-score registered a more modest growth (see Figure 6). Regarding the evolution of the bank’s market power, the trend for the conventional Lerner index is ascending suggesting an increase in pricing power of the European banks of our sample. Figure 6 - Distance-to-default, Ln Z-score and Lerner index (average) 2011-2018 Source: Own production Figure 7 - Distance-to-default and Ln Z-score against Lerner index (average) by country Source: Own production Comparing bank financial stability by country, using distance-to-default, we concluded that on average, listed banks of our sample that operate in Greece, Ireland and Portugal are the most fragile, while the Austrian and Swedish banks are the most stable (see Figure 7). If we considered Z-score the French and the Swedish banks are the most stable. Looking for bank’s market power, comparisons lead to conclude that Greek, Irish and Competition and Financial Stability in the European Listed Banks 29 Portuguese banks present the lowest Lerner indexes, while at the opposite extreme, with the highest pricing power, are the banks of Sweden, Belgium and Norway. We also observe a positive relationship between the average values of the measures of financial stability and market power. 2.4. EMPIRICAL RESULTS AND DISCUSSION Columns (2) and (4) of Table 2 present the estimation results of Eq. (2.8), by alternatively considering distance-to-default and Ln Z-score as measures of financial stability. Results, for both the financial stability measures, do not support, for our sample, the U-shaped relationship between competition and risk (Hypothesis II) of MMR (2010). Although the signs of coefficient estimates associated with the Lerner index and its square indicate the possibility of an inverse U-shaped relationship between the Lerner Index and the measures of financial stability, the Ushape test of Lind and Mehlum (2010) does not allow us to reject the null hypothesis of a monotone relationship in the model of column (2) and indicate a turning point outside the sample range of the Lerner index for the model of column (4). Given these results, we re-estimated Eq. (2.8) excluding the quadratic term of the Lerner index [see columns (1) and (3) of Table 2]. For both market and accounting-based stability measures, we find a positive and significant relationship with the Lerner index, which confirms the competition-fragility view (Hypothesis I). The obtained results allow us to conclude that an increase in the competition encourages individual bank risk-taking of European listed banks, which confirms our Hypothesis I and the evidence found in the recent studies for European banking systems (Leroy & Lucotte, 2017). Discussing now the impact of the other control variables on bank stability, we found a positive and statistically significant relationship between the Ln Z-score and the levels of capitalization, size, non-interest income share and asset composition [see columns (3) and (4)]. These results indicate that largest banks, best-capitalized, with higher loans-to-assets ratio, and with more diversification of their sources of income, are more financially stable. For the distance-to-default model [see columns (1) and (2)] only non-interest income share and real GDP growth are significant at 5% level of significance. We highlight the negative sign and the magnitude of the estimate of the coefficient associated with real GDP growth, indicating that JOSÉ FERNANDO DA SILVA NETO 30 the economic growth encourages banks to reduce financial restrictions to increase lending, generating more risk and consequently less stability. Table 2 - Bank financial stability and competition (whole sample) Dependent variable Distance-to-default (DD) Ln Zscore (1) (2) (3) (4) Dependent variable (t-1) 0.401*** 0.417*** 0.914*** 0.908*** (0.053) (0.056) (0.019) (0.018) Dependent variable (t-2) 0.248*** 0.274*** (0.047) (0.043) Lerner index 3.036*** 1.791* 0.249*** 0.146 (0.927) (0.976) (0.093) (0.113) Lerner index squared -2.659** -0.010 (1.246) (0.143) Capitalization 6.087 2.966 0.750* 0.720** (4.964) (4.591) (0.382) (0.343) Size 0.051 -0.012 0.015*** 0.012* (0.051) (0.051) (0.005) (0.006) Non-interest income share 1.444** 1.411** 0.147* 0.182** (0.630) (0.626) (0.080) (0.083) Share of wholesale funding 0,096 0.564 -0.013 0.027 (0.441) (0.472) (0.052) (0.057) Liquidity 0.663 0.415 0.067 0.127** (1.110) (1.043) (0.050) (0.060) Asset composition 1.448* 0.894 0.186** 0.194** (0.849) (0.799) (0.080) (0.097) Inefficiency -0.000 -0.775 -0.109 -0.193* (0.913) (0.897) (0.084) (0.105) Real GDP growth -29.388*** -24.878*** -0.409 -0.198 (8.814) (9.118) (0.696) (0.749) Inflation 13.489 13.241 0.752 0.603 (10.098) (9.241) (0.744) (0.759) Constant -0.972 0.747 -0.019 0.049 (1.381) (1.439) (0.111) (0.160) U-Shape test 0.300 Extremum outside interval p-value [U-Shape test] [0.384] Turning point 0.337 Number of observations 619 619 737 737 Number of banks 115 115 117 117 Number of instrumental variables 80 102 87 107 F-Test 58,71*** 55.84*** 1512.00*** 1349.86*** RHO(1) Test -5,643 -5.699 -3.741 -3.692 p-value [RHO(1) Test] 0.000 0.000 0.000 0.000 RHO(2) Test 1.195 1.045 1.427 1.199 p-value [RHO(2) Test] 0.232 0.296 0.154 0.230 Hansen's J Test 70.328 91.774 72.820 95.272 p-value [Hansen's J Test] 0.219 0.239 0.353 0.280 Note: The table reports the dynamic panel regression results. The two-step system GMM estimator (Arellano & Bover, 1995) is used with Windmeijer (2005) corrected standard errors. Some regressions include an additional lag of dependent variable as explanatory variable to remove second-order autocorrelation. Time Dummies are included. Robust standard errors are reported in parentheses below their coefficient estimates. The U-shape test is based on Competition and Financial Stability in the European Listed Banks 31 Lind and Mehlum (2010) and “Extremum outside interval” means that the extremum point (i.e. the turning point) is outside the interval, then we cannot reject the null hypothesis of a monotone relationship. Arellano-Bond test is used to test serial correlation, where RHO(1) and RHO(2) are the estimated coefficients of firstand second order correlation and apply to residuals in differences. To analyse the validity of instruments we used the Hansen's (1982) J Test for overidentification. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production To test Hypothesis III, we divided our sample into two sub-samples, one containing banks that are based in countries with less stable banking systems and the other with banks that belong to countries with more stable banking systems. For this purpose, using Z-score data extracted from the World Bank's Global Financial Development Database, we calculated the average of that indicator, for each country, in the period 2011-20178. Then the countries were ranked in ascending order of the Z-score and split into two sub-samples: the first group, of the countries with less stable banking systems, which includes banks from the 8 countries with a lower average Z-score (Italy, Portugal, Finland, United Kingdom, Netherlands, Norway, Ireland, Greece) and a second group, of countries with more stable banking systems, which includes banks from the 8 countries with a higher average Z-score (Austria, Germany, France, Spain, Denmark, Belgium, Switzerland, Sweden). It is interesting to note that countries such as Finland, the United Kingdom, the Netherlands, and Norway, which traditionally have sustainable public finances, are part of the group of countries with less stable banking sectors, based on Z-score. On the contrary, countries like Spain and Belgium, which in the recent past had some problems with the sustainability of public finances, are part of the group of countries with more stable banking sectors. This finding allows us to conclude that the stability of a country's banking sector is not necessarily influenced by that country's public finances, reinforcing the hypothesis that the relationship between competition and risk-taking can be differentiated depending on whether the bank operates in a banking system more or less stable. The estimation results of Eq. (2.8), with and without the Lerner Index quadratic term, for the two sub-samples, are reported in Table 39. The results confirm a linear and positive relationship between market power and bank's financial stability, confirming the competitionfragility view (Hypothesis I), in the countries with less stable banking systems. The same conclusion cannot be drawn for countries with more stable banking systems, where the 8 2018 was not considered, because at the time of the investigation, that year was not yet available. 9 The complete estimation results can be consulted in the Table 4 and Table 5 of the appendix of this chapter. JOSÉ FERNANDO DA SILVA NETO 32 relationship between market power and financial stability is not statistically significant at a 5% level of significance. In both sub-samples, there was no evidence of the U-shaped relationship between competition and risk. Table 3 - Bank financial stability and competition: less vs more stable banking systems Dependent variable Distance-to-default (DD) Ln Zscore (1) (2) (3) (4) Less stable banking systems: Lerner index 5.700*** 2.078 1.098*** 0.486 (2.031) (1.446) (0.182) (0.360) Lerner index squared -0.955 -0.994 (1.620) (0.665) Number of observations 364 364 305 305 Number of banks 58 58 57 57 U-Shape test Extremum outside interval 0.420 p-value [U-Shape test] [0.338] Turning point 0.244 More stable banking systems: Lerner index 2.062 10.785* -0.456 -0.721* (4.245) (5.449) (0.319) (0.405) Lerner index squared 9.509 -0.279 (6.239) (0.636) Number of observations 314 314 373 373 Number of banks 58 58 59 59 U-Shape test 0.730 Extremum outside interval p-value [U-Shape test] [0.235] Turning point -0.567 Source: Own production 2.5. ROBUSTNESS TESTS We test the robustness of our results in several ways. First, we considered an alternative measure of the Lerner index: the efficiency-adjusted Lerner index as outlined in section 2.3.2. The estimation results, for the whole sample and sub-samples of less and more stable banking systems, are reported in Table 6, Table 7 and Table 8 of the appendix of this chapter, respectively. Second, we used distance-to-insolvency, described in section 2.3.1, as a measure of a bank's financial stability in estimating Eq. (2.8). The estimation results, for the whole sample and sub-samples of less and more stable banking systems, are reported in Table 9, Table 10 and Table 11 of the appendix of this chapter, respectively. Finally, we estimate a static version of Eq. (2.8) using the fixed effects model and the random effects model. In these models, to consider the endogeneity issue, all explanatory variables are lagged one period. Estimation results for fixed effects model, for the whole sample and sub-samples of less and Competition and Financial Stability in the European Listed Banks 33 more stable banking systems, are reported in Table 12, Table 13 and Table 14 of the appendix of this chapter, respectively. For the random effects model, the results are reported, similarly, in Table 15, Table 16 and Table 17. In general terms, the results obtained were the same, supporting the “competition-fragility” view for the whole sample and for the sub-sample of the banks that belongs to countries with less stable banking systems. There is no evidence of the U-shaped relationship between competition and risk in the whole sample and in the two sub-samples considered. For banks based in countries with more stable banking systems, market power does not appear to influence risk-taking. 2.6. CONCLUSIONS The beginning of the 21st century was marked by serious financial crises, such as the global financial crises and Eurozone sovereign debt crises, which severely decreased the financial stability of banks worldwide. This forced the governments of several countries to adopt measures to rescue the banks, and thus, prevent the propagation of a systemic risk crisis. This set of public interventions has probably changed the relationship between competition and financial stability, which motivated this study. This work investigated the competition-stability nexus in the European banking systems using a sample of listed banks. We extend the existing literature by investigating if that nexus is differentiated depending on whether the bank operates in a more stable or less stable banking system as a whole. We proxy competition with the Lerner index and focused on overall risk measures, such as distance-to-default, distance-to-insolvency and Z-score, for bank risk-taking. To deal with the persistence of bank risk-taking over time, we used a dynamic panel data model, which was estimated by a 2-step GMM estimator to address the endogeneity problem between the bank competition measure and capitalization levels and the bank risk-takings measures. The results obtained do not confirm the U-shaped relationship between competition and bank risk-taking as predicted by MMR (2010). We find support for the competition-fragility view in European banking as a whole, indicating that additional market power decreases the individual risk-taking behaviour of a bank. Perhaps because the competitive environment in European banking systems is already high, the “margin” effect dominates the risk-shifting effect. However, the JOSÉ FERNANDO DA SILVA NETO 34 competition-fragility view appears only to be valid in countries with less stable banking systems. In countries with more stables banking systems, the relationship between market power and financial stability did not prove to be statistically significant. These results remained unchanged even when we considered the efficiency-adjusted Lerner index as a measure of competition, distance-to-insolvency as a measure of bank risktaking or when we estimated a static panel data model with fixed effects or random effects. Our findings highlight several issues for policymakers and regulators. Public policies must guarantee banking competition, for welfare reasons, but limiting excessive bank risk-taking, especially in countries with less financially sound banking systems. This means that any attempt to increase competition in European banking should be accompanied by regulation that guarantees bank stability, for example by increases in capital standards and limiting the risk exposure. Consolidation of the European banking industry can lead to stronger and more resilient banks without compromising competition. However, this process of consolidation in Europe has a significant number of obstacles due to political, economic, regulatory, and cultural factors. Although the European Banking Union was created in 2014 to stimulate this integration, it remains unfinished and European banks - especially retail banks - still mostly operate on a national basis. Along with domestic mergers, European authorities and different national governments should promote cross-border mergers to deepen the integration and construction of a truly European banking sector. Cross-border banks would be able to offset losses in one country with income from other countries and would be better prepared to face the challenges posed by technological disruption. Competition and Financial Stability in the European Listed Banks 35 2.7. APPENDIX Figure 8 - European Union structural financial indicators Herfindahl index (total assets) for credit institutions (Index ranging from 0 to 10.000) Share of total assets of five largest credit institutions (%) Change in the number of credit institutions between 2008 and 2018 (%) Source: Own production 0 500 1000 1500 2000 2500 3000 3500 2008 2018 0 10 20 30 40 50 60 70 80 90 100 2008 2018 -80 -70 -60 -50 -40 -30 -20 -10 0 JOSÉ FERNANDO DA SILVA NETO 42 Table 10 - Bank financial stability, measured by distance-to-insolvency, and competition (subsample of the less stable banking systems) Dependent variable Distance-to-insolvency (DI) (1) (2) Dependent variable (t-1) 0.308** 0.346* (0.127) (0.202) Lerner index 3.923*** 3.332 (1.079) (3.172) Lerner index squared 1.865 (3.581) Capitalization -3.714 -1.292 (10.012) (6.204) Size 0.031 0.056 (0.079) (0.065) Non-interest income share 0.417 1.234** (0.575) (0.555) Share of wholesale funding -0.211 -0.255 (0.634) (0.591) Liquidity -1.069 -0.982 (1.054) (1.136) Asset composition -0.247 -0.020 (0.574) (0.705) Inefficiency 1.874 1.289 (1.170) (2.352) Real GDP growth 0.885 4.506 (7.110) (8.774) Inflation 25.904*** 28.517** (7.245) (11.520) Constant -0.143 -0.848 (2.248) (2.301) U-Shape test 0.060 p-value [U-Shape test] [0.475] Turning point -0.893 Number of observations 364 364 Number of banks 58 58 Number of instrumental variables 24 27 F-Test 27.89*** 50.06*** RHO(1) Test -4.227 -3.334 p-value [RHO(1) Test] 0.000 0.001 RHO(2) Test 1.511 1.481 p-value [RHO(2) Test] 0.131 0.139 Hansen's J Test 8.244 10.126 p-value [Hansen's J Test] 0.221 0.256 Note: The table reports the dynamic panel regression results. The two-step system GMM estimator (Arellano & Bover, 1995) is used with Windmeijer (2005) corrected standard errors. Time Dummies are included. Robust standard errors are reported in parentheses below their coefficient estimates. The U-shape test is based on Lind and Mehlum (2010) and “Extremum outside interval” means that the extremum point (i.e. the turning point) is outside the interval, then we cannot reject the null hypothesis of a monotone relationship. Arellano-Bond test is used to test serial correlation, where RHO(1) and RHO(2) are the estimated coefficients of firstand second order correlation and apply to residuals in differences. To analyse the validity of instruments we used the Hansen's (1982) J Test for overidentification. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production Competition and Financial Stability in the European Listed Banks 43 Table 11 - Bank financial stability, measured by distance-to-insolvency, and competition (subsample of the most stable banking systems) Dependent variable Distance-to-insolvency (DI) (1) (2) Dependent variable (t-1) 0.529*** 0.597*** (0.070) (0.073) Dependent variable (t-2) 0.327*** 0.301*** (0.086) (0.062) Lerner index 0.719 2.314 (1.977) (1.935) Lerner index squared 1.348 (2.389) Capitalization -1.615 -5.068 (12.366) (12.763) Size -0.097 -0.079 (0.164) (0.154) Non-interest income share -0.603 -1.204 (1.012) (1.263) Share of wholesale funding 0.666 0.554 (0.886) (0.909) Liquidity -0.810 -0.046 (1.448) (1.554) Asset composition -0.960 -0.102 (1.276) (1.448) Inefficiency -0.538 1.022 (1.523) (1.650) Real GDP growth -40.473*** -39.712*** (12.045) (14.682) Inflation 16.902 6.930 (22.298) (25.153) Constant 4.113 2.376 (3.861) (3.680) U-Shape test 0.080 p-value [U-Shape test] [0.469] Turning point -0.858 Number of observations 314 314 Number of banks 58 58 Number of instrumental variables 56 46 F-Test 60.45*** 91.20*** RHO(1) Test -3.800 -4.028 p-value [RHO(1) Test] 0.000 0.000 RHO(2) Test -0.540 -0.269 p-value [RHO(2) Test] 0.590 0.788 Hansen's J Test 48.008 36.441 p-value [Hansen's J Test] 0.128 0.106 Note: The table reports the dynamic panel regression results. The two-step system GMM estimator (Arellano & Bover, 1995) is used with Windmeijer (2005) corrected standard errors. Some regressions include an additional lag of dependent variable as explanatory variable to remove second order autocorrelation. Time Dummies are included. Robust standard errors are reported in parentheses below their coefficient estimates. The U-shape test is based on Lind and Mehlum (2010) and “Extremum outside interval” means that the extremum point (i.e. the turning point) is outside the interval, then we cannot reject the null hypothesis of a monotone relationship. Arellano-Bond test is used to test serial correlation, where RHO(1) and RHO(2) are the estimated coefficients of firstand second order correlation and apply to residuals in differences. To analyse the validity of instruments we used the Hansen's (1982) J Test for overidentification. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production JOSÉ FERNANDO DA SILVA NETO 44 Table 12 - Bank financial stability and competition (whole sample) – Fixed Effects (FE) Model Dependent variable Distance-to-default (DD) Ln Zscore (1) (2) (3) (4) Lerner index 0.482 0.733 0.138* 0.165*** (0.589) (0.779) (0.074) (0.055) Lerner index squared 0.441 0.047 (1.052) (0.113) Capitalization 2.648 2.775 2.653*** 2.666*** (7.037) (7.121) (0.671) (0.675) Size -1.277** -1.268** -0.177*** -0.176*** (0.610) (0.610) (0.049) (0.049) Non-interest income share -0.399 -0.401 -0.064 -0.064 (1.212) (1.213) (0.107) (0.107) Share of wholesale funding 1.858 1.898 -0.018 -0.014 (1.696) (1.696) (0.115) (0.112) Liquidity -0.809 -0.815 -0.067 -0.068 (2.043) (2.050) (0.138) (0.138) Asset composition -6.886*** -6.888*** -0.068 -0.068 (1.575) (1.584) (0.142) (0.141) Inefficiency -0.517 -0.378 0.058 0.073 (0.683) (0.822) (0.093) (0.090) Real GDP growth -14.348** -14.265** -0.368 -0.359 (7.149) (7.089) (0.624) (0.624) Inflation 26.994*** 26.592*** -0.325 -0.368 (9.132) (9.021) (0.662) (0.651) Constant 19.230*** 18.980*** 5.119*** 5.092*** (6.676) (6.751) (0.545) (0.551) U-Shape test 0.070 Extremum outside interval p-value [U-Shape test] [0.473] Turning point -0.831 Number of observations 737 737 737 737 Number of banks 117 117 117 117 F-Test 27.945*** 26.451*** 21.069*** 24.873*** Note: The table reports the static panel regression results for the fixed effects model. To consider the endogeneity issue, all explanatory variables are lagged one period. Robust standard errors clustered at the bank level are reported in parentheses below their coefficient estimates. Time Dummies are included. The U-shape test is based on Lind and Mehlum (2010) and “Extremum outside interval” means that the extremum point (i.e. the turning point) is outside the interval, then we cannot reject the null hypothesis of a monotone relationship. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production Competition and Financial Stability in the European Listed Banks 45 Table 13 - Bank financial stability and competition (whole sample) – Random Effects (RE) Model Dependent variable Distance-to-default (DD) Ln Zscore (1) (2) (3) (4) Lerner index 1.356*** 2.194*** 0.171** 0.229*** (0.488) (0.842) (0.079) (0.055) Lerner index squared 1.570 0.099 (1.150) (0.114) Capitalization 7.118 7.199 3.129*** 3.173*** (5.126) (5.042) (0.735) (0.732) Size -0.257** -0.244** -0.072** -0.069** (0.102) (0.101) (0.031) (0.031) Non-interest income share 0.126 0.115 -0.025 -0.024 (0.994) (1.005) (0.116) (0.117) Share of wholesale funding 2.582*** 2.529*** 0.153 0.172 (0.865) (0.859) (0.111) (0.109) Liquidity 0.028 -0.029 -0.024 -0.023 (1.739) (1.738) (0.153) (0.153) Asset composition -3.481*** -3.454*** 0.006 0.010 (1.239) (1.251) (0.156) (0.158) Inefficiency -0.294 0.199 0.075 0.106 (0.624) (0.782) (0.099) (0.093) Real GDP growth -6.148 -6.239 -0.042 -0.009 (6.376) (6.348) (0.658) (0.662) Inflation 34.077*** 32.565*** -0.085 -0.171 (8.995) (8.691) (0.618) (0.628) Constant 5.023** 4.463** 3.787*** 3.708*** (2.162) (2.236) (0.377) (0.379) U-Shape test 0.500 Extremum outside interval p-value [U-Shape test] 0.309 Turning point -0.699 Number of observations 737 737 737 737 Number of banks 117 117 117 117 Chi2-Test 488.861*** 479.766*** 319.756*** 423.175*** Note: The table reports the static panel regression results for the random effects model. To consider the endogeneity issue, all explanatory variables are lagged one period. Robust standard errors clustered at the bank level are reported in parentheses below their coefficient estimates. Time Dummies are included. The U-shape test is based on Lind and Mehlum (2010) and “Extremum outside interval” means that the extremum point (i.e. the turning point) is outside the interval, then we cannot reject the null hypothesis of a monotone relationship. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production JOSÉ FERNANDO DA SILVA NETO 46 Table 14 - Bank financial stability and competition (sub-sample of the less stable banking systems) – FE Model Dependent variable Distance-to-default (DD) Ln Zscore (1) (2) (3) (4) Lerner index -0.182 0.511 0.187** 0.249*** (0.437) (0.762) (0.089) (0.072) Lerner index squared 1.122 0.100 (0.751) (0.145) Capitalization -0.086 0.053 2.109** 2.121** (6.808) (7.035) (0.904) (0.900) Size -1.323** -1.291** -0.176*** -0.173*** (0.629) (0.626) (0.057) (0.056) Non-interest income share 0.402 0.488 -0.142 -0.135 (0.767) (0.775) (0.146) (0.150) Share of wholesale funding -0.067 0.078 0.093 0.106 (2.034) (2.052) (0.160) (0.153) Liquidity 2.152 2.084 -0.228 -0.234 (2.284) (2.303) (0.153) (0.153) Asset composition -7.421*** -7.342*** 0.065 0.072 (1.995) (2.026) (0.190) (0.191) Inefficiency -1.069 -0.693 0.087 0.120 (0.734) (0.751) (0.107) (0.109) Real GDP growth 2.271 2.627 0.890 0.922 (7.859) (7.755) (0.904) (0.899) Inflation 47.985*** 46.203*** -0.095 -0.254 (11.791) (11.738) (0.807) (0.786) Constant 19.407*** 18.613*** 4.703*** 4.632*** (6.948) (6.968) (0.629) (0.630) U-Shape test 1.110 Extremum outside interval p-value [U-Shape test] 0.136 Turning point -0.227 Number of observations 364 364 364 364 Number of banks 58 58 58 58 F-Test 19.177*** 18.204*** 18.648*** 18.064*** Note: The table reports the static panel regression results for the fixed effects model. To consider the endogeneity issue, all explanatory variables are lagged one period. Robust standard errors clustered at the bank level are reported in parentheses below their coefficient estimates. Time Dummies are included. The U-shape test is based on Lind and Mehlum (2010) and “Extremum outside interval” means that the extremum point (i.e. the turning point) is outside the interval, then we cannot reject the null hypothesis of a monotone relationship. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production Competition and Financial Stability in the European Listed Banks 47 Table 15 - Bank financial stability and competition (sub-sample of the less stable banking systems) – RE Model Dependent variable Distance-to-default (DD) Ln Zscore (1) (2) (3) (4) Lerner index 0.899** 1.835** 0.218** 0.296*** (0.455) (0.870) (0.093) (0.072) Lerner index squared 1.687* 0.125 (0.949) (0.148) Capitalization -3.198 -2.740 2.138** 2.154** (5.825) (5.954) (0.929) (0.921) Size -0.236** -0.217** -0.130*** -0.127*** (0.109) (0.108) (0.044) (0.043) Non-interest income share 1.327** 1.429** -0.087 -0.075 (0.652) (0.664) (0.145) (0.150) Share of wholesale funding 0.434 0.474 0.152 0.171 (1.320) (1.332) (0.152) (0.144) Liquidity 1.780 1.736 -0.133 -0.138 (2.518) (2.536) (0.177) (0.176) Asset composition -2.264** -2.089* 0.054 0.063 (1.128) (1.113) (0.207) (0.209) Inefficiency -1.037 -0.456 0.116 0.159 (0.674) (0.734) (0.109) (0.110) Real GDP growth 16.720** 15.756** 1.249 1.294 (7.831) (7.717) (0.930) (0.928) Inflation 59.842*** 56.144*** 0.414 0.222 (9.773) (9.823) (0.757) (0.767) Constant 4.350** 3.502 4.116*** 4.030*** (2.111) (2.210) (0.526) (0.532) U-Shape test 1.240 Extremum outside interval p-value [U-Shape test] 0.108 Turning point -0.544 Number of observations 364 364 364 364 Number of banks 58 58 58 58 Chi2-Test 485.355*** 513.712*** 279.345*** 310.345*** Note: The table reports the static panel regression results for the random effects model. To consider the endogeneity issue, all explanatory variables are lagged one period. Robust standard errors clustered at the bank level are reported in parentheses below their coefficient estimates. Time Dummies are included. The U-shape test is based on Lind and Mehlum (2010) and “Extremum outside interval” means that the extremum point (i.e. the turning point) is outside the interval, then we cannot reject the null hypothesis of a monotone relationship. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production JOSÉ FERNANDO DA SILVA NETO 48 Table 16 - Bank financial stability and competition (sub-sample of the most stable banking systems) – FE Model Dependent variable Distance-to-default (DD) Ln Zscore (1) (2) (3) (4) Lerner index 0.796 -0.703 0.004 -0.002 (1.973) (1.731) (0.092) (0.080) Lerner index squared -3.504 (0.016) (2.347) (0.125) Capitalization 44.001** 42.792** 3.180*** 3.174*** (18.109) (17.456) (1.014) (1.013) Size 0.241 0.254 -0.164* -0.164* (1.023) (0.988) (0.092) (0.092) Non-interest income share -5.133* -4.683 0.058 0.060 (2.941) (2.919) (0.137) (0.138) Share of wholesale funding 4.718 4.380 -0.136 -0.138 (3.106) (3.106) (0.155) (0.156) Liquidity -7.992** -7.937** 0.087 0.087 (3.057) (2.995) (0.250) (0.250) Asset composition -13.547*** -13.204*** -0.101 -0.100 (3.400) (3.374) (0.212) (0.215) Inefficiency 1.001 0.365 -0.134 -0.136 (1.955) (1.957) (0.167) (0.166) Real GDP growth -25.556** -25.578** -1.679** -1.680** (10.348) (10.576) (0.818) (0.819) Inflation 14.625 11.125 -1.925* -1.941* (17.937) (17.517) (1.144) (1.135) Constant 6.020 6.525 5.455*** 5.458*** (11.994) (11.583) (1.054) (1.050) U-Shape test 1.230 0,110 p-value [U-Shape test] 0,112 0.455 Turning point -0.100 -0.073 Number of observations 373 373 373 373 Number of banks 59 59 59 59 F-Test 23.844*** 23.128*** 19.733*** 18.932*** Note: The table reports the static panel regression results for the fixed effects model. To consider the endogeneity issue, all explanatory variables are lagged one period. Robust standard errors clustered at the bank level are reported in parentheses below their coefficient estimates. Time Dummies are included. The U-shape test is based on Lind and Mehlum (2010) and “Extremum outside interval” means that the extremum point (i.e. the turning point) is outside the interval, then we cannot reject the null hypothesis of a monotone relationship. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production Competition and Financial Stability in the European Listed Banks 49 Table 17 - Bank financial stability and competition (sub-sample of the most stable banking systems) – RE Model Dependent variable Distance-to-default (DD) Ln Zscore (1) (2) (3) (4) Lerner index 2.352 1.693 0.006 0.026 (1.546) (1.385) (0.094) (0.086) Lerner index squared -1.561 0.047 (2.623) (0.118) Capitalization 19.905** 20.482** 4.936*** 5.058*** (9.334) (9.458) (0.772) (0.767) Size -0.328 -0.330 -0.007 -0.002 (0.252) (0.254) (0.043) (0.042) Non-interest income share -3.571 -3.491 0.016 0.007 (2.349) (2.341) (0.156) (0.160) Share of wholesale funding 2.485** 2.566** 0.001 0.021 (1.213) (1.221) (0.127) (0.129) Liquidity -6.760* -6.732* 0.050 0.048 (3.470) (3.474) (0.288) (0.289) Asset composition -8.703*** -8.669*** -0.061 -0.058 (3.120) (3.133) (0.244) (0.249) Inefficiency 1.179 0.856 -0.140 -0.133 (1.664) (1.702) (0.174) (0.171) Real GDP growth -24.882*** -24.806*** -1.636** -1.626** (8.835) (9.051) (0.796) (0.795) Inflation 21.420 19.782 -1.652 -1.599 (21.006) (21.043) (1.225) (1.201) Constant 10.713* 10.974* 3.531*** 3.450*** (5.893) (5.855) (0.557) (0.548) U-Shape test Extremum outside interval 0.270 p-value [U-Shape test] 0.392 Turning point -0.282 Number of observations 373 373 373 373 Number of banks 59 59 59 59 Chi2-Test 355.893*** 356.670*** 320.634*** 335.307*** Note: The table reports the static panel regression results for the random effects model. To consider the endogeneity issue, all explanatory variables are lagged one period. Robust standard errors clustered at the bank level are reported in parentheses below their coefficient estimates. Time Dummies are included. The U-shape test is based on Lind and Mehlum (2010) and “Extremum outside interval” means that the extremum point (i.e. the turning point) is outside the interval, then we cannot reject the null hypothesis of a monotone relationship. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production 3. EFFECTS OF NEGATIVE INTEREST RATE POLICY IN BANK PROFITABILITY AND RISK-TAKING: EVIDENCE FROM THE EUROPEAN BANKS 3.1. INTRODUCTION In the last decade, some central banks around the world, in an attempt to avoid low inflation rates and stimulate economic growth, resorted to a set of unconventional monetary policy instruments, such as a large scale asset purchase in the form of quantitative easing, the implementation of negative interest rate policies (NIRP’s) as well as policy rate forward guidance. After, in July 2012, Danmarks Nationalbank lowered, for the first time, its policy rate to negative values, several central banks from other countries (Euro Area, Hungary, Norway, Sweden, Switzerland, Bulgaria, and Japan) adopted the same behaviour. The implementation of NIRP means that central banks are now charging interests (instead of paying) on the excess reserves that commercial banks have deposited there, encouraging them to take them back on the balance sheet. This is expected to have positive effects on economic activity through the increased supply and demand for loans due to the decline in funding costs for both banks and borrowers (Cœuré, 2016). However, the effect that NIRP can have on banks' profitability is not so clear (Boungou, 2019). Low or negative interest rates help to improve banks' balance sheets and performance, leading to capital gains and a reduction in loan loss provisions. However, low or negative interest rates also can mean lower net interest margins. This is because bank intermediation is a spread business, based on the difference between interest rates on loans and deposits. When market interest rates fall, becoming low or even negative, banks may have to adjust loan interest rates down but are very reluctant to lower deposit rates for negative levels, especially for retail depositors, compressing the net interest margin (Claessens et al., 2018). Banks know that a negative deposit interest rate would lead their customers to withdraw their deposits for other banks. Refusing to pass negative interest rates on customer deposits, bank’s profitability related to maturity transformation will be negatively affected eroding their equity capital and deteriorating their financial stability (Zimmermann, 2019). Banks can compensate the margin JOSÉ FERNANDO DA SILVA NETO 58 Altunbas et al. (2014) states, in addition, that bank risk may also be influenced by communication policies, with a moral hazard problem. Ease monetary policy perception in bad economic outcomes could lower the expectations of large downside risks and encourage liquidity risk-taking. This leads to the low-interest rates paradox introduced by Maddaloni & Peydró (2011) according to which when interest rates are low, credit risk and liquidity risk increase and so do the likelihood of a financial crisis. The interaction between finance, behavioural finance and macroeconomics associated to the risk-taking channel of monetary policy, have been justified because it captures the measurement and managing risk, the effects of monetary policy on bank’s risk perceptions and incentives and because excessive bank risktaking has effects on the general equilibrium, respectively (Andries et al., 2015). In complement of the moral hazard problem, monetary policy would impact on the adverse selection problem, considering the reducing incentive to screen and monitor loan applicants by weakening banks (Dell’Ariccia et al., 2011; DellʼAriccia et al., 2014). Considering the empirical studies, Jiménez et al. (2014) test the existence of a risk-taking channel for Spain. According to the authors, low interest rates affect the risk of the loan Spanish Bank’s portfolio, as follows: i) in short term, low interest rates reduce the probability of default of the outstanding loans; ii) in the medium term banks tend to take more risk, softening their lending standards, and lending to borrowers with bad credit history. Ioannidou et al. (2009) have reached the same conclusion, investigating the impact of changes in the monetary policy rates on loan pricing in Bolivia. They state a negative relationship between the interest rate and loans’ risk. They also conclude that banks increase the number of new risky loans and reduce the rates they charge to riskier borrowers, relative to less risky ones. Heider et al., (2018), using loan-level information for Europe, covering the period from January 2013 to December 2015, conclude that the introduction of negative policy rates by the European Central Bank in mid2014 leads to more risk-taking and less lending by euro-area banks, which can pose a risk to financial stability if lending is done by high-deposit banks. Basten & Mariathasan (2018) analyse the effect of negative monetary policy rates on banks, using detailed supervisory information from Switzerland, namely, comparing changes in the behaviour of banks that had different fractions of their central bank reserves exempt from negative rates. They conclude that more affected banks reduce costly reserves and bond financing while maintaining non-negative deposit rates and larger deposit ratios. With higher fee and interest income, banks compensate for squeezed liability margins but increase credit and interest rate risk. Boungou (2020) analyse Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 59 for the first time that it is known, the effects of negative rates on the risk-taking of banks operating on the 28 member countries of the European Union. The author concluded that negative interest rates contributed to a reduction in banks’ risk-taking. During the implementation period of negative rates, banks took less risk, in particular by reducing the share of non-performing loans. Boungou (2019) does not find encourage from banks to take more risk, despite a reduction in interest margins. Several studies in the risk-taking literature tend to explain how the interest rate structure would encourage excessive risk-taking by banks. For some authors, the effects of interest rates on risk-taking depend on the profitability level of banks (Repullo, 2004; Martynova et al., 2019), and for others, on the bank's capitalization level (Jiménez et al., 2014; Dell’ariccia et al., 2017). Studies that investigate the effects of the NIRP on bank risk-taking are still limited and the results obtained are contradictory. Those contradictory results justify the present investigation and considering the explained above, the following hypotheses are formulated to test: Hypothesis III: The NIRP lead to a greater bank risk-taking. Hypothesis IV: The effect of a decrease in interest rates on bank risk-taking increase is most pronounced when a NIRP is implemented. 3.2.3. Negative Interest Rates and Bank’s Business Model Some authors emphasize the fact that the impact of low (negative) interest rates on profitability varies depending on a set of bank’s characteristics. Molyneux et al., (2020) reinforce the existence of specific characteristics that significantly influence the relationship between the negative interest rate and bank margins. The characteristics are presented as follows: bank’s size, funding structure and the business model, including assets repricing, and product-line specialization. The bank’s size could explain the reduced elasticity of net interest margin to interest rate volatility. The funding structure is important, because, when policy rates turn negative, banks that rely on deposit funding are reluctant to reduce deposit rates, trying to keep their funding base, avoiding passing negative rates onto depositors. The business model can provide different degrees of sensitivity to interest rate risk. This risk is different from a real estate mortgage specialist bank, comparing to a bank that holds mostly floating interest rate JOSÉ FERNANDO DA SILVA NETO 60 loans. Because the evidence shows that the effect of adopting a NIRP on profitability can be differentiated depending on the bank's business model, we still formulated the following hypothesis: Hypothesis V: The effect of a NIRP on profitability depends on the business model adopted by the bank. Lucas et al., (2019) in an empirical study of 208 European banks between 2008Q1– 2015Q4, identify six distinct business model and conclude that changes in the slope in the yield curve changes in average business model characteristics. So it is expected that, depending on the business model adopted by a bank, its risk response will be differentiated (Schwaab, 2017). This led us to formulate the last research hypothesis: Hypothesis VI: The effect of a NIRP on bank risk-taking depends on the business model adopted by the bank. 3.3. METHODOLOGY AND DATA In this subchapter, we introduce the methodology and empirical models that allow analysing the impacts of the adoption of NIRP’s by some central banks on profitability and risktaking in European banks. For this purpose, profitability and risk-taking measures are defined and discussed. The variables used to characterize the interest rate environment are also specified. To investigate whether these impacts are differentiated according to the business model adopted by the bank, we also describe the methodology that allows us to allocate banks to different business models. 3.3.1. Profitability and Risk-Taking Measures To measure banks' profitability two main indicators were considered: net interest margin, defined as the difference between interest-earning assets and interest-bearing liabilities divided by total earning assets, and return on assets (ROA), a commonly used performance measure, defined by the ratio of net income over total assets10. Because monetary policy also has an 10 Similar measures are considered by Borio et al. (2017) and Boungou (2019). Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 61 impact on other components of profit (Borio et al. 2017), in addition to net interest margin, we also study the effect of NIRP on net fee & commission, net trading income, other operating revenues and finally on loan loss provisions11. Considering the different risk channels of the monetary policy identified in the literature review, we consider three measures of risk (e.g. Boungou, 2020). The non-performing loans (NPL) ratio, defined as the non-performing loans divided by gross loans, as a measure of credit risk. A riskier loan portfolio increases the credit risk and the need to constitute more provisions. The ratio of risk-weighted assets (RWA) over total assets. In a context of reduced interest rates, the “search for yield” effect, leads the banks to adopt more risky investment strategies, increasing their exposure to risk and consequently this ratio. Lastly, we consider the ratio Zscore12 as a measure of overall banking risk. This ratio is estimated as the sum of current ROA with equity-to-asset (E/A) ratio divided by the standard deviation of ROA (𝜎𝑅𝑂𝐴)13. The idea behind the Z-score is that a bank becomes insolvent when its current losses exhaust all bank's equity. Thus, a lower bank's Z-score implies a greater risk of insolvency (Ngambou Djatche, 2019). 3.3.2. Interest Rate Environment Measures This study is concerned with the impact that the adoption of NIRP by some central banks in Europe has on banks' profitability and risk-taking. To this end, the following variables are considered: a short-term interest rate, the slope of the yield curve and a dummy variable reflecting the adoption or not of a NIRP. As a short‐term interest rate, we take the 3‐month interbank money market interest rate (e.g. Delis & Kouretas, 2011; Bikker & Vervliet, 2018). We prefer an interbank money market interest rate to the central bank's policy rate because the former reflects more appropriately the adoption of unconventional monetary policy measures. Making the same assumption as in the literature that the short‐term interest rate reflects the general interest rate level, we expect that 11 All those variables are considered in our models as a percentage of total assets. 12 Because literature indicates that the Z-score is highly skewed, we use a natural logarithm transformation. 13 Because the sample period covered by the present investigation is relatively short, we assumed that 𝜎𝑅𝑂𝐴 is constant and given by the standard deviation of the return on assets in the period under analysis. JOSÉ FERNANDO DA SILVA NETO 62 lower interest rates impair the bank’s net margin and increase risk exposure and that this impact is stronger when interest rates are already low or negative. The slope of the yield curve, that also helps control the effects of unconventional monetary policy measures, is approximated by the difference between 10-year treasury yield and 3‐month interbank money market interest rate (e.g. Borio et al., 2017; Claessens et al., 2018). As a result of the maturity transformation function performed by banks, a positive correlation is expected between the banks’ profits or net interest margins and the slope of the yield curve (Alessandri & Nelson, 2015). Finally, to characterize the interest rate environment, we consider a dummy variable to reflect if the central bank of the country where the bank is based adopted or not a NIPR. Figure 9 shows the evolution of the central bank's policy rate14 in European countries that have adopted NIRP’s. Figure 9 - Evolution of the central bank's policy rate in European countries that have adopted NIRP Source: Own production 14 Refers to the main deposit policy rate in most cases, and the main refinancing rate for the Riksbank. No historical data is reported for Bulgaria due to its unavailability. -2,0 -1,0 0,0 1,0 2,0 3,0 4,0 5,0 6,0 7,0 2011 2012 2013 2014 2015 2016 2017 2018 2019 Policy Rate Denmark Eurozone Sweden Switzerland Hungary Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 63 3.3.3. Model Specifications and Estimation Method To study the effects of interest rates on banks' profitability and risk-taking as a result of the adoption of NIRP’s, we consider the following models: 𝑦𝑖𝑡=𝛽0+𝛽1∗𝑖𝑟𝑡−1+𝛽2∗𝑠𝑙𝑜𝑝𝑒𝑡−1+𝛽3∗𝑁𝐼𝑅𝑃𝑡−1+𝛾1∗𝑋𝑖𝑡−1+𝛾2∗𝑊𝑡−1 +𝜇𝑖+𝛿𝑡+𝜀𝑖𝑡 (3.1) 𝑦𝑖𝑡=𝛼0+𝛼1∗𝑖𝑟𝑡−1+𝛼2∗𝑠𝑙𝑜𝑝𝑒𝑡−1+𝛼3∗𝑁𝐼𝑅𝑃𝑡−1∗𝑖𝑟𝑡−1+𝛼4 ∗𝑁𝐼𝑅𝑃𝑡−1∗𝑠𝑙𝑜𝑝𝑒𝑡−1+𝛾1∗𝑋𝑖𝑡−1+𝛾2∗𝑊𝑡−1+𝜇𝑖+𝛿𝑡+𝜉𝑖𝑡 (3.2) where 𝑖 and 𝑡 are bank and time indicators, respectively, 𝑦𝑖𝑡 represents alternatively one of the bank’s profitability or risk-taking measures defined above, 𝑖𝑟𝑡 represents the short‐term interest rate, 𝑠𝑙𝑜𝑝𝑒𝑡 represents the slope of the yield curve and 𝑁𝐼𝑅𝑃𝑡 takes the value of 1 if in the country where the bank is based adopted a NIRP in year 𝑡 and 0 otherwise. 𝑋𝑖𝑡 and 𝑊𝑡 represents a vector of bank‐specific and macroeconomic variables, respectively. 𝜇𝑖 and 𝛿𝑡 represent a bank-specific effect and time fixed effect, respectively. In all regressions, we follow Borio et al. (2017) and Leroy & Lucotte (2017), and explanatory variables are lagged one period and we include bank and time fixed effects in order to mitigate a possible endogeneity bias. Both equations are estimated using a fixed-effect estimator and in the statistical inference robust standard errors clustered at the bank level are used to consider the existence of autocorrelation and/or heteroscedasticity. For profitability models, we expect that 𝛽3<0, meaning that the adoption of NIRP's will put pressure on banks' net interest margin and profitability (Hypothesis I). According to Hypothesis II, it expected 𝛼3−𝛼4>0, meaning that a decrease in the short-term interest rates lead to a negative change in banks' net interest margin and profitability more pronounced when a NIRP is implemented. For risk-taking models, according to Hypothesis III, it is expected that 𝛽3<0 if the risk measure used is Z-score, and 𝛽3>0 if the risk is measured by the NPL ratio or the ratio of RWA over total assets. According to Hypothesis IV, we expect that 𝛼3−𝛼4>0 if the risk measure used is Z-score, and 𝛼3−𝛼4<0 if the risk is measured by the NPL ratio or the ratio of RWA over total assets. To control possible effects of other determinants of the bank’s profitability and risk-taking, we include the following bank‐specific variables in vector 𝑋𝑖𝑡. First, we consider the bank’s size, measured by the natural logarithm of the total assets. According to Goddard et al. (2004), JOSÉ FERNANDO DA SILVA NETO 64 the bank’s size influences positively its profitability through the realisation of economies of scale. However, as suggested by Demirgüç-Kunt et al. (2004), large efficient banks could apply lower spreads to customers through increasing returns to scale. So, the effect of the bank’s size on profitability is unclear. The same conclusion can be drawn regarding the relationship between the bank’s size and risk. On one hand, managers of large banks may be tempted to adopt higher-risk policies in the case that governments are prepared to bail-out large problematic banks (Demirgüç-Kunt & Huizinga, 2013) and, on the other hand, larger banks can achieve economies of scale that allow them to be more stable than small banks (Williamson, 1986). We employ several variables to control for bank risk aversion, credit risk and bank operating efficiency. We use capitalization, measured by the ratio of equity over total assets, to proxy bank risk aversion. Given their risk aversion, we expect that better-capitalized banks will require higher margins and take less risk (Berger, 1995; Bikker & Vervliet, 2018). Credit risk is measured by the non-performing loans (NPL) ratio. We expect that banks with higher credit risk apply a premium to margins (Philip Molyneux et al., 2019) and shows a higher overall risk15. Bank’s management inefficiency is measured using the cost-to-income ratio. As referred by Molyneux et al. (2019), inefficient management of a bank translates into lower margins and profits and consequently more risk. In order to control for the impact of bank business models, we also consider as a determinant of a bank's profitability and risk-taking the asset composition, measured by the loans-to-asset ratio, its funding structure, measured by the share of wholesale funding and the diversity of its incomes, measured by the non-interest income share (Bikker & Vervliet, 2018 and Molyneux et al., 2019). The banking literature suggests that the macroeconomic environment in which banks operate may have effects on their behaviour. Thus, both the structure of the banking sector and the economic environment can affect banks' profitability and risk-taking. Like Boungou (2019) we considered real GDP growth and the inflation rate to characterize the macroeconomic conditions. To measure the impact of market structure on the bank’s profitability and risktaking, we use the Herfindahl-Hirschman Index (HHI) (Chen et al., 2017), which is measured as the sum of the squares of individual bank's market share in total banking assets, to proxy the 15 In the risk-taking model in which the dependent variable is the NPL ratio, for methodological reasons, this variable was not considered as an explanatory variable. Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 65 average concentration level of the banking sector. A Herfindahl-Hirschman Index close to one indicates more concentration. In Table 25 of the Appendix of this chapter, we present a detailed description of all the variables used in the present investigation as well as the different sources of information used. To analyse research hypotheses V and VI, we need, first, to identify the different business models existing in European banking and, second, allocate each bank in our sample to one of the previously identified business models. For this purpose, following the methodology adopted by Hryckiewicz & Kozłowski (2017) and Roengpitya et al. (2017), we use k-medians16 clustering to assign each bank to a specific banking business model given its asset and funding structure. The objective of this k-medians clustering is to group banks with similar asset and funding structures into the same cluster and those with different characteristics into different clusters. The k-medians approach identifies a cluster by minimizing the differences between the individual financial variables of different banks using Manhattan distance: 𝑆=∑∑|𝑥𝑖𝑗−𝑚𝑒𝑑𝑘𝑗| 𝑥𝑖∈𝐶𝑘 𝐾 𝑘=1 (3.3) where 𝐾 is the number of clusters, 𝑥𝑖𝑗 are the observation of the 𝑗 financial variable for the bank 𝑖 used in cluster analysis, 𝐶𝑘 is the cluster 𝑘 and 𝑚𝑒𝑑𝑘𝑗 is the median on cluster 𝑘. For our analysis, we perform the grouping based on earning assets structures and liability sources. Among bank asset structures, we distinguish the following positions: loans to customers, loans and advances to banks and trading securities, all scaled by bank total assets. Among bank funding sources, we distinguish between customer deposits and wholesale funding17. Additionally, because the NIRP’s adoption may have caused a bank to change its business model, we allow our sample banks to modify their banking business models throughout the sample period. To ensure a good compromise between the homogeneity within each cluster and the number of clusters selected, the pseudo F-index, proposed by Calinski & Harabasz (1974), is 16 We prefer k-medians clustering to k-means clustering because medians are less sensitive to outliers than means. 17 This item includes bank deposits, debt securities, repurchase agreements and subordinated liabilities. JOSÉ FERNANDO DA SILVA NETO 66 used to help us decide18. To evaluate the goodness of clustering by considering how well the clusters are separated and how compact the clusters are, we use the silhouette coefficient. This measure ranges from -1 to +1, where a high value indicates that the bank is well matched to its cluster and poorly matched to neighbouring clusters (Rousseeuw, 1987). 3.3.4. Sample Description and Data Statistics In our investigation, an unbalanced panel data of European banks is used, covering the period from 2011 to 2019, from the 29 following countries: 18 countries that, at the end of 2019, belonged to the Eurozone19, Bulgaria, Croatia, the Republic Czech, Denmark, Hungary, Norway, Poland, Romania, Sweden, Switzerland, and the United Kingdom. In those countries, 6 central banks have adopted NIRP’s. Table 18 summarizes information on the date and level of the policy rate at which the 6 central banks first adopted a NIRP. Table 18 - Negative interest rate policy (NIRP) announcements Country Central Bank Policy rate Date Rate Bulgaria Central Bank of Hungary Overnight deposit rate Jan 2016 -0.30% Denmark Danmarks Nationalbank 1-week certificate of deposit rate July 2012 -0.20% Eurozone European Central Bank Overnight deposit facility rate June 2014 -0.10% Hungary Magyar Nemzeti Bank Overnight deposit rate March 2016 -0.05% Sweden Sveriges Riksbank 1-week repo rate February 2015 -0.10% Switzerland Swiss National Bank Overnight sight deposit rate December 2014 -0.25% Source: Own production and based on information collected from Central Banks Specific information about bank variables is obtained from Moody's Analytics BankFocus, with all data converted to euros20. Historical information about the short-term interest rate, the 10-year treasury yield, the GDP growth rate and the inflation rate has been taken from Thomson Datastream. Finally, the Herfindahl-Hirschman Index is computed using data of the bank’s total assets available from Moody's Analytics BankFocus database. After excluding banks with missing data or with unplausible data for the variables used, we obtain a final sample that includes an unbalanced panel data sample for 2596 banks21, with 18 We should select the number of clusters that maximize the pseudo F-index. 19 We do not include the Banks of Estonia because some macroeconomic data is not available. 20 We considered all commercial banks, savings banks, real estate & mortgage banks, cooperative banks, and bank holdings & holding companies, with at least 2 years of information. We considered consolidated accounts when available and unconsolidated accounts for individual banks. We excluded all domestic bank subsidiaries (to avoid duplication of data) and holding companies with residual bank activity. 21 A note to mention that the sample used is dominated by German banks that represent 48.57% of the total number of banks, followed by Italian banks with 16.80%. Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 67 15119 bank/year observations, where 8743 correspond to the period after the implementation of the NIRP by central banks. The descriptive statistics of the variables that are used in the main regressions are reported in Table 19 , distinguishing the pre-NIRP period from the NIRP period. As we can see, as a result of the adoption of NIRP's by several central banks in Europe, there has been a considerable decrease in short-term interest rates. Namely, on average, short-term interest rates fell by 143.1 b.p. from 1.142% to -0.289%. There was also a sharp decline in long-term interest rates from the pre-NIRP period to the NIRP period, smoothing the yield curve, with its slope’s mean value decreasing from 2.456 to 1.375. This latter movement can be explained by the various asset purchase programs implemented by several central banks during the period under analysis. As a result of the decrease in short-term interest rates and the slope of the yield curve, the mean value of net interest margin decreased by 39.9 b.p. from the pre-NIRP period to the NIRP period from 2.289% to 1.890%. The mean value of the ROA registered, from one period to the other, only a slight decrease from 0.386% to 0.342%. This less pronounced decrease in ROA may be explained in part by a less severe loan loss provisions policy and an increase in the mean value of other operating revenues: the mean value of loan loss provisions relatively to total assets decreased from 0.365% in the period pre-NIRP to 0.204% in the NIRP period, while the other operating revenues in total assets increased from 0.223% to 0.360%. Looking at risk-taking measures, we can conclude that banks, in the NIRP period, took less risk. From the pre-NIRP period to NIRP period the mean value of the natural logarithm of Zscore, a proxy for the overall bank risk, increased from 4.283 to 4.737. The mean value of the NPL ratio, a proxy of credit risk, decreased from 6.759% in the pre-NIRP period to 5.343% in NIRP period. Lastly, the mean value of the ratio RWA/Total Assets also decreased which indicates that, on average, banks adopted less risky investment strategies, in the NIPRP period. Thus, the preliminary evidence does not seem to allow us to conclude that banks' risk-taking has increased with the implementation of NIRP's. JOSÉ FERNANDO DA SILVA NETO 74 About the exposure to credit risk, as measured by the NPL ratio, it can be concluded that the most efficient banks have lower credit risk, while banks with a higher share of wholesale funding and non-interest income are more exposed to default’s risk. Economic growth, inflation and increased banking concentration have a positive, statistically significant effect on credit risk, decreasing the NPL ratio. Lastly, we can conclude that banks better capitalized, more exposed to credit risk, with high loan-to-asset ratios follow riskier investment strategies, measured by RWA/TA ratio. Higher economic growth and greater banking concentration also lead to greater bank’s risk-taking. 3.6. EFFECTS OF A NEGATIVE INTEREST RATE POLICY ON DIFFERENT BANKS' BUSINESS MODELS To study whether the effect of implementing a NIRP on profitability and risk-taking depends on the business model adopted by the bank, we use the methodology described in section 3.3.3. Based on the bank assets structures and their sources of financing, and using cluster analysis, it was concluded that the optimal number of business models to use in our sample was four. These four business models are characterized in Table 22 by a set of characteristics listed there. We have designated the four business models as follows: − Investment-oriented banks type I (Model I): Large banks, with funding sources diversified, having substantial trading activities and trading income has a relatively high weight on operational revenues; − Retail-oriented banks (Model II): Midsize banks, whose main source of financing is customer deposits, and which are highly oriented for lending to customers. Its major sources of operational revenues are net interest margin and net fees and commissions; − Investment-oriented banks type II (Model III): Small banks, whose main source of financing is customer deposits, having substantial trading activities. Its major sources of operational revenues are net interest margin and net fees and commissions; − Interbank lending-oriented banks (Model IV): Midsize banks, whose main source of financing is customer deposits, and which are highly oriented for lending to other banks. Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 75 Its major sources of operational revenues are net interest margin and net fees and commissions. Table 22 - Business model identification based on bank assets structures and their sources of financing Variable Business Model Model I Model II Model III Model IV Variables used in the cluster analysis (% of total assets): Loans to customers 63.62 76.60 53.03 51.41 Loans and advances to banks 8.40 5.58 7.26 31.45 Trading securities 22.11 13.45 35.25 12.04 Customer deposits 43.40 72.06 76.89 77.46 Wholesale funding 41.69 15.58 10.80 9.74 Other variables (% of operational revenues except for total assets): Total Assets (in millions €) 71320 8419 3715 6744 Net interest margin 61.10 67.96 64.03 60.94 Net fees and commissions 22.49 21.30 22.70 26.17 Net trading income 10.54 1.67 3.17 3.18 Other operacional revenues 5.87 9.07 10.11 9.71 Number of banks 628 1288 1180 355 Note: This table shows the mean values of the listed variables for each business model. Source: Own production After identifying the different bank business models in our sample, we study the effects of the implementation of the NIRP on the profitability and risk-taking of each one. Regarding the effects of short-term interest rates and NIRP on profitability, Table 23 reports the (partial) estimation results of equations (3.1) and (3.2) for the net interest margin [columns (1) and (2)] and return on assets [columns (3) and (4)]. Looking at the effects of the implementation of the NIRP in some European countries, we can conclude that, except for investment-oriented banks (type I), all other banks see their net interest margin decrease. In particular, it appears that the banks where customer deposits have a greater weight in their financing, greater is the negative impact on banks' net interest margin. In the interbank lendingoriented banks, where the customer deposits represent, on average, 77.46% of funding sources and the loans and advances to banks represent 31,45% of the investments, the net interest margin reduced 32.2 b.p. as a result of the implementation of the NIRP. This happens because banks that adopt this business model are forced to lower interest rates on bank loans without being able to lower interest rates on customer deposits, given the reluctance of banks to lower JOSÉ FERNANDO DA SILVA NETO 76 the latter to negative values. It can also be seen that, when interest rates are already on the negative ground, an additional fall in them puts greater pressure on the net interest margin of the retail-oriented and interbank lending-oriented banks. Analysing the effect of interest rates and NIRP on overall profitability, measured by ROA, there seem to be no differences between the pre-NIRP period and the NIRP period across different bank business models. Table 24 reports the (partial) estimation results of equations (3.1) and (3.2) for the effects of short-term interest rates and NIRP on the bank’s risk-taking. Looking at columns (1) and (2) of Table 24, it can be concluded that the adoption of NIRP's did not affect banks' financial stability, measured by the Z-score, regardless of their business model. Only for interbank lending-oriented banks, we conclude that when interest rates are already on the negative ground, an additional fall in them leads to a decrease in financial stability higher than that which would occur if interest rates were in positive territory. We also conclude that the adoption of NIRP's did not have a different impact on the credit risk of the different bank business models identified [see columns (3) and (4) of Table 24]. Lastly, looking at columns (5) and (6), we can conclude that in the NIRP period, the investment-oriented banks (type I) and interbank lending-oriented banks adopted more risky investment strategies, which is in line with the idea of “search for yield” presented by Rajan (2005). On the contrary, in the NIRP period, retail-oriented banks implemented less risky strategies. We also verify that when interest rates are already on the negative ground, an additional fall in them leads investment-oriented banks (type I) to increase their risk exposure. From the analysis carried out, we can conclude for research hypotheses V and VI that the effect of a NIRP in the net interest margin and the investment strategies adopted depends on the business model adopted by the bank. Regarding to overall profitability, credit risk and financial stability, that dependency is not so evident. Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 77 (continued) Table 23 - Effect of interest rates and NIRP on net interest margin and return on assets by business model Bank Business Explanatory variables Net interest margin Return on assets Model (1) (2) (3) (4) Model I Short-term interest rate 0.129* 0.118 0.203** 0.223** (0.070) (0.074) (0.099) (0.106) Slope of the yield curve 0.017 0.011 0.155*** 0.148*** (0.018) (0.019) (0.044) (0.038) NIRP 0.095 -0.011 (0.058) (0.082) NIRP * Short-term interest rate -0.066 -0.049 (0.142) (0.211) NIRP * Slope of the yield curve 0.029** 0.083** (0.014) (0.038) Research hypothesis: H1 [𝛽3=0] 1.639 -0.135 H2 [𝛼3−𝛼4=0] -0.661 -0.594 Number of observations 2481 2481 2481 2481 Model II Short-term interest rate -0.041 -0.061 0.050 0.044 (0.047) (0.049) (0.079) (0.088) Slope of the yield curve -0.016 -0.012 0.036 0.045 (0.021) (0.021) (0.062) (0.060) NIRP -0.163*** -0.055 (0.032) (0.052) NIRP * Short-term interest rate 0.155*** 0.162 (0.063) (0.121) NIRP * Slope of the yield curve -0.033* 0.139 (0.017) (0.103) Research hypothesis: H1 [𝛽3=0] -5.110*** -1.070 H2 [𝛼3−𝛼4=0] 2.817*** 0.188 Number of observations 4313 4313 4313 4313 Model III Short-term interest rate 0.042 0.013 -0.218 -0.301 (0.066) (0.069) (0.197) (0.224) Slope of the yield curve -0.048*** -0.031 0.028 -0.010 (0.018) (0.019) (0.047) (0.049) NIRP -0.226*** -0.279 (0.205) NIRP * Short-term interest rate 0.062 0.488 (0.181) (0.367) NIRP * Slope of the yield curve 0.040 -0.150*** (0.027) (0.048) Research hypothesis: H1 [𝛽3=0] -2.729*** -1.631 H2 [𝛼3−𝛼4=0] 0.118 1.702* Number of observations 4305 4305 4305 4305 JOSÉ FERNANDO DA SILVA NETO 78 Source: Own production Table 23 (continued) - Effect of interest rates and NIRP on net interest margin and return on assets by business model Bank Business Explanatory variables Net interest margin Return on assets Model (1) (2) (3) (4) Model IV Short-term interest rate 0.207*** 0.176*** -0.211 -0.302 (0.057) (0.067) (0.256) (0.253) Slope of the yield curve 0.017 0.051* 0.124 0.156** (0.029) (0.030) (0.075) (0.075) NIRP -0.322*** -0.326 (0.114) (0.740) NIRP * Short-term interest rate 0.596*** 1.045 (0.244) (0.655) NIRP * Slope of the yield curve -0.103*** -0.040 (0.020) (0.057) Research hypothesis: H1 [𝛽3=0] -2.825*** -0.441 H2 [𝛼3−𝛼4=0] 2.870*** 1.638 Number of observations 951 951 951 951 Note: This table shows the (partial) results of the effects of the adoption of NIRP and interest rates on bank net interest margin and return on assets by business model. In all regressions, explanatory variables are lagged one period and we include bank and time fixed effects to soften eventual endogeneity issues. Section “Research hypothesis” report the t-statistics for the respective hypothesis. Robust standard errors clustered at the bank level are reported below their coefficient estimates. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 79 Table 24 - Effect of interest rates and NIRP on bank risk-taking by business model (1) (2) (3) (4) (5) (6) Short-term interest rate 0.014 0.014 -1.032 -1.291 0.414 0.438 (0.020) (0.020) (0.725) (0.799) -1.002 -1.000 Slope of the yield curve 0.023*0.024*-1.811*** -1.803*** 0.383 0.324 (0.012) (0.013) (0.273) (0.259) (0.474) (0.501) NIRP -0.007 -0.066 2.780*** (0.031) (0.618) -1.008 NIRP * Short-term interest rate -0.001 2.245 -5.882*** (0.078) -1.752 -2.579 NIRP * Slope of the yield curve -0.014 0.142 -0.681 (0.014) (0.274) (0.477) Research hypothesis: -0.220 -0.108 2.757*** 0.162 1.188 -1.997** Number of observations (Banks) 2481 2481 2481 2481 2481 2481 Short-term interest rate 0.026 0.027 0.088 0.013 1.311*** 1.125* (0.018) (0.021) (0.418) (0.493) (0.641) (0.682) Slope of the yield curve 0.022*0.021*-0.028 -0.041 0.835*** 0.759** (0.013) (0.012) (0.325) (0.329) (0.360) (0.363) NIRP 0.008 -0.136 -0.960* (0.012) (0.254) (0.531) NIRP * Short-term interest rate -0.029 0.282 -0.064 (0.025) (0.640) -1.035 NIRP * Slope of the yield curve -0.022*-0.269 -1.983*** (0.013) (0.252) (0.436) Research hypothesis: 0.710 -0.536 -1.807* -0.253 0.816 1.700* Number of observations 4313 4313 4313 4313 4313 4313 (continued) Table 24 - Effect of interest rates and NIRP on bank risk-taking by business model Business model Explanatory variables Ln Zscore NPL ratio RWA/TA Model I Model II JOSÉ FERNANDO DA SILVA NETO 80 (1) (2) (3) (4) (5) (6) Short-term interest rate -0.021 -0.042 1.089 1.307* -0.488 -0.639 (0.050) (0.060) (0.681) (0.740) (0.707) (0.684) Slope of the yield curve 0.016 -0.004 -0.217 -0.265 1.421*** 0.874** (0.012) (0.012) (0.305) (0.321) (0.404) (0.409) NIRP -0.070 0.896 -0.019 (0.058) -1.107 (0.677) NIRP * Short-term interest rate 0.031 -2.164 -2.150 (0.099) -2.086 -1.837 NIRP * Slope of the yield curve -0.073*** -0.121 -1.941*** (0.013) (0.284) (0.431) Research hypothesis: -1.202 0.809 -0.028 1.014 -0.969 -0.118 Number of observations (Banks) 4305 4305 4305 4305 4305 4305 Short-term interest rate -0.016 -0.029 3.942*** 4.266*** 0.501 2.091 (0.053) (0.056) -1.937 -2.312 -1.379 -1.843 Slope of the yield curve 0.057** 0.068** -0.533 -0.468 2.170* 2.122* (0.026) (0.027) (0.488) (0.489) -1.121 -1.171 NIRP -0.107 -0.782 4.865** (0.113) -2.824 -2.246 NIRP * Short-term interest rate 0.207* -2.387 -9.640 (0.112) -6.293 -6.132 NIRP * Slope of the yield curve -0.029** -0.540 -1.471*** (0.014) (0.349) (0.543) Research hypothesis: -0.953 -0.277 2.166** 1.994** -0.294 1.338 Number of observations 951 951 951 951 951 951 Explanatory variables Ln Zscore NPL ratio RWA/TA Source: Own production Model III Model IV Note: This table shows the partial results of the effects of the adoption of NIRP and interest rates on bank risk-taking by business model. In all regressions, explanatory variables are lagged one period and we include bank and time fixed effects to soften eventual endogeneity issues. Section “Research hypothesis” report the t-statistics for the respective hypothesis. Robust standard errors clustered at the bank level are reported below their coefficient estimates. *, ** and *** indicate statistical sign. at 10%, 5% and 1% levels. Table 24 (continued) - Effect of interest rates and NIRP on bank risk-taking by business model Business model Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 81 3.7. CONCLUSIONS Since 2012, several central banks implemented NIRP’s intending to boost economic activity and fight low inflation rates by facilitating an increase in the supply of bank loans. These policies have generated controversy with the most sceptics pointing to several factors that might affect the bank’s financial stability and raising doubts about the transmission mechanism from negative policy rates to higher bank lending. This investigation studies the effect of negative interest rates on bank’s profitability and risktaking. Using a sample of 2596 banks from European countries, over the period 2011-2019, we conclude that NIRP implementation lowered the net interest margin of a representative bank by 14.5 b.p., in average. This finding combined with the results of chapter 2 allows to conclude that the increase in market power, measured by the Lerner index, of European banks, in the period analysed, cannot be explained by the behaviour of the banks' net interest margin, as it decreased in the period. The increase in market power may have been explained by the rise in fees and commissions charged and the cut in operating costs. Despite the rise in the weight of fees and commissions charged by banks, the decrease of the weight of net trading income and the increase of the weight of loan loss provisions led the overall profitability of a representative bank, measured by ROA, to decrease 18.5 b.p. It is interesting to note that despite the decrease in ROA, banks' solvency risk, measured by the Zscore, decreased in a scenario of negative interest rates. This can be explained by the reinforcement of prudential rules, namely, in terms of capital requirements and/or a lower volatility in the returns. We also conclude that a decrease in short-term interest rates lower net interest margin and the ROA in a more pronounced way when interest rates are already negative than when, they are positive. In a scenario of negative interest rates, banks try to compensate for the decrease in the net interest margin with a more pronounced increase in the net fees and commissions. Despite the negative effect that the implementation of NIRP had on the net interest margin and the ROA, we do not find that European banks, on average, increased risk-taking. We also conclude that in an environment of negative interest rates, additional decreases in short-term interest rates do not lead banks to take more risk, that is, there is no evidence of the "search for JOSÉ FERNANDO DA SILVA NETO 82 yield" effect. For this result, once again, the tightening of prudential regulation rules will have been decisive. Based on the bank assets structures and their sources of financing and using cluster analysis, we identify four different business models: investment-oriented banks (type I), retail-oriented banks, investment-oriented banks (type II) and interbank lending-oriented banks. Our study leads us to conclude that, except for investment-oriented banks (type I), all other banks see their net interest margin decreased with the implementation of the NIRP. This is particularly true for those banks whose main source of finance is retail deposits. We also conclude that, when interest rates are already on the negative ground, an additional fall in them puts greater pressure on the net interest margin of the retail-oriented and interbank lending-oriented banks. Looking at the effect of interest rates and NIRP on ROA, we do not find differences between the preNIRP period and the NIRP period across different bank business models. The analysis carried out also make it possible to conclude that the implementation of NIRP did not affect banks' financial stability and credit risk, regardless of their business model. Lastly, we conclude that investment-oriented banks (type I) and interbank lending-oriented banks adopted more risky investment strategies, while retail-oriented banks have adopted less risky investment strategies. The empirical results obtained in this study require that special emphasis be given, by the regulatory and supervisory entities of European banking systems, to the monitoring of the bank’s profitability and risk-taking that were most affected by the introduction of NIRP's. Effects of Negative Interest Rate Policy in Bank Profitability and Risk-taking: Evidence from the European Banks 83 3.8. APPENDIX Table 25 - Variables definition and data source Variable Units Description Source Bank profitability: Net interest margin percentage Difference between interestearning assets and interestbearing liabilities divided by total earning assets BankFocus Database Return on assets percentage Net income divided by total assets BankFocus Database Net fee & commission percentage Net fee & commission divided by total assets BankFocus Database Net trading income percentage Net trading income divided by total assets BankFocus Database Other operating revenues percentage Other operating revenues divided by total assets BankFocus Database Loan loss provisions percentage Loan loss provisions divided by total assets BankFocus Database Bank risk-taking: Ln Z-score logarithm Z-score is computed as the ratio between the sum of the expected return on assets and the equity to total assets ratio and the standard deviation of the return on assets BankFocus Database and own calculations NPL Ratio percentage Non-performing loans divided by gross loans BankFocus Database RWA / Assets percentage Risk-weighted assets divided by total assets BankFocus Database Bank-specific variables: Size logarithm Natural logarithm of total assets BankFocus Database Capitalization percentage Equity divided by total assets BankFocus Database Inefficiency percentage Cost-to-income ratio computed as the ratio of operating expenses on net operating income BankFocus Database Liquidity percentage Liquid assets divided by total assets BankFocus Database Share of wholesale funding percentage Wholesale funding divided by total funding BankFocus Database Asset composition percentage Loans and advance to customers divided by total assets BankFocus Database Non-interest income share percentage Non-interest income divided by operating revenues BankFocus Database Interest rate environment measures: Short-term interest rate percentage 3-month interbank rate Thompson Datastream Slope of yield curve percentage Difference between 10-year Treasury yield and 3-month interbank rate Thompson Datastream NIRP Dummy variable Takes the value of 1 if a NIRP was adopted and 0 otherwise Central Banks Country variables: Real GDP growth percentage Yearly growth rate of real GDP Thompson Datastream Inflation percentage Yearly growth rate of the consumer price index Thompson Datastream Herfindahl-Hirschman Index (Assets) units Measure of market concentration BankFocus Database and own calculations Source: Own production JOSÉ FERNANDO DA SILVA NETO 90 According to the agency theory, when the interests of managers are not aligned with those of shareholders, the former invest in CRS activities in their own interest (Jensen & Meckling, 1976). In particular, because investment in this type of activity is well regarded by society in general, managers build a good image and reputation at the expense of the company. Jiraporn & Chintrakarn (2013) demonstrate that managers who are less entrenched are more likely to increase CSR activity than other CEOs. This is likely due to the private benefits and reputational benefits afforded by CRS activity. In disagreement with the views presented, we find the stakeholder theory developed by Freeman (1984). This theory states that a company does not belong just to owners or shareholders, meaning that it must be considered the mass of agents involved in it. In this sense, the objective of the company should not be to maximize value for the shareholder, but instead, the company should create value for all the stakeholders including employees, consumers, local communities, natural or environmental resources. Some authors, such as Post et al. (2002), argue that companies should apply those social, environmental, and corporate governance aspects that are necessary, regardless of the costs incurred or the income they produce. Stakeholder theory suggests that environmental, social and governance practices are important issues for stakeholders. The conceptualization made by Carroll (1991) and Wood (1991) about CSR includes a stakeholder approach in which any party, including employees, customers, shareholders, environment, society, and investors, who might be affected by the business activities of organizations, should be considered as a stakeholder of an organization. The resource-based view sees investment in CRS activities as being strategic that allow the company to gain competitive advantages by acquiring additional skills that are difficult to replicate (Russo & Fouts, 1997). This means that an increment in corporate social performance improves financial performance. On the other hand, authors like Porter & Kramer (2011) claim that the company’s objective as to be the maximization of shareholder value while, at the same time, trying to incorporate social, environmental and corporate governance measures into management, as a way to create shared value for the company and society. In other words, it is important to know whether these measures are profitable for the company, in the sense that they allow shareholder value to be maximized. Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 91 Miralles-Quirós et al., (2019), in their study about the relationship between ESG performance and shareholder value creation in the banking industry, presents a summary of the recent studies about this controversy, underlining that previous empirical evidence for the banking industry presents inconclusive results. Simpson & Kohers (2002) provided evidence of a positive and significant relationship between CSR activities and the performance of the company. Later, Soana (2011) showed that there is no statistically significant link between the two measures of performance for a sample of Italian banks. Wu & Shen (2013), with a sample of 162 banks from 22 countries covering the period 2003–2009, observed that CSR is positively associated with financial performance in terms of return on assets, return on equity, net interest income, and non-interest income. On the contrary, CSR is negatively associated with unproductive loans. Meanwhile, Cornett et al. (2016) analysed the relationship between CSR and financial performance in US banks during the financial crisis. Their results indicate that larger banks perform significantly more CSR activities than smaller banks. Mixed results were obtained by Esteban-Sanchez et al. (2017) when analysing the effect of different CSR dimensions on the financial performance of 154 banks in 22 countries, before and during the years of the financial crisis. Belasri et al. (2020), using an international sample of 184 banks in 41 countries over the 2009-2015 period, founds evidence that CSR has a positive impact on bank efficiency in developed countries, in countries where investor protection is high and in countries featuring a high degree of stakeholder orientation. Shah et al. (2019) using a sample of 45 banks from 14 countries for a period of nine years (2010-2018) founded evidence that sustainable banks are more efficient and productive. In a more recent study for 39 European banks, for the period from 2010–2019, Bătae et al. (2021) find mixed results for the relationship between the corporate financial performance and the different dimensions of CSR activities. They concluded by a positive relationship between emissions reductions and financial performance. The same cannot be concluded for its product quality and social responsibility policies. Regarding the corporate governance dimension, they concluded that an increase in its quality negatively affects the bank's financial performance. Belasri et al. (2020) suggest that various reasons are pointing that ESG activities could have an impact on bank’s inputs and outputs, and as a result on bank efficiency: CSR activities can help firms build a strong reputation (Branco & Rodrigues, 2006; Hillman & Keim, 2001) which can, in turn, provide many benefits such as an increased ability to attract and retain valuable employees (Branco & Rodrigues, 2006; Bătae et al. (2021). Increased employee JOSÉ FERNANDO DA SILVA NETO 92 productivity and loyalty are associated with better management of human capital resources or, from an efficiency perspective, a better use (processing) of inputs. On the other hand, customers may be willing to accept a lower rate on their deposits if it comes from a bank with strong CSP (Wu & Shen, 2013). In the case of banks, a good reputation could therefore increase profit by enabling banks to attract new customers and charge higher interests on their loans. Also, a strong CSR-induced reputation can provide banks with the ability to charge higher fees and commissions on other services (Wu & Shen, 2013). This expected positive impact of CSR on both interest and non-interest income indicates that CSP could increase a bank’s outputs. Consistent with these arguments and in line with stakeholder theory, we can formulate the following hypothesis: Hypothesis I: CSP has a positive impact on banking efficiency. 4.2.1. CSP Dimensions and Bank’s Efficiency As noted by Xie et al. (2019), ESG activities are the result of management policies and legal obligations and comprise different dimensions. Naturally, these different dimensions have a different contribution to the CSP, depending on the activity of the company. The environmental dimension of CSP is a highly researched subject, but the relationship between environmental practices and corporate efficiency remains inconclusive (Ambec et al., 2013). The neoclassical traditional view argues that environmental regulations represent an additional cost to the company that reduce profitability and lead to low efficiency (Friedman, 1970). In contrast, Porter & van der Linde (1995) argue that environmentally friendly regulation promotes technological innovation in companies, creating efficiencies that more than offset additional costs. Although banks are not seen as polluters in comparison, for example, with chemical or oil companies, banks use a considerable amount of resources such as energy and paper and generate indirect carbon emissions (Bătae et al., 2021). By investing in renewable energy for office buildings, offering eco-friendly services such as e-banking apps, switching paper by electronic documents, banks can reduce the operational costs improving their environmental performance. According to the resource-based view on environmental practices, pollution prevention and product stewardship can become a source of competitive advantage, through differentiation or Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 93 cost savings (Hart, 1995). However, Finger et al. (2018) consider that in the banks of developed countries environmental management is a form of window-dressing in the sense that banks have already optimized their processes to the point that more environmental measures do not bring significant improvements in their sustainability performance. In line with stakeholder theory, banks that implement environmentally responsible practices are more likely to create positive stakeholder perceptions, resulting in improved economic performance (Sila & Cek, 2017). Although some studies (e.g., Wagner et al., 2002) reported a negative relationship between these two variables, others (e.g., Bătae et al., 2021) found a positive relationship. In this study, we posit that environmental performance is positively related to bank’s efficiency: Hypothesis II: Environmental performance is positively related to banking efficiency. Social performance refers to how the organization treats its employees, the community and the customers, through responsibility in their products and services (Miralles-Quirós et al., 2019). According to Rhouma et al. (2014), stakeholders greatly appreciate the implementation of different social practices by organizations. These practices are, among others, those related to employees' rights, their training and career development, issues related to customers and the support of social causes. Starting from within the organization, stable and fair relationships between employees and management will lead to higher personal satisfaction and loyalty (Birindelli et al., 2015), contributing to an increase in corporate efficiency. In line with the Equator Principles, a socially responsible bank must optimize its credit portfolio to finance socially responsible investments. Wu & Shen (2013) argue that a bank that engages in CSR activities builds a strong loyalty with its customers that allows it to pay a lower interest rate on deposits, charge a higher interest rate on loans and higher fees and commissions on other services, improving the financial performance and efficiency. In this sense, we can highlight the works of Simpson & Kohers (2002) that observed that banks that are more involved with the community in which they operate achieve greater financial performance. Fombrun (2005) also refers those social practices can serve as a marketing tool for companies to increase demand for their products and services. Based on the referred arguments, we hypothesize that JOSÉ FERNANDO DA SILVA NETO 94 Hypothesis III: Social performance has a positive impact on banking efficiency. Corporate governance is defined as the organisation’s code of conduct to ensure whether board members and executives actions are compatible with the stakeholder’s interests (EstebanSanchez et al., 2017). Miralles-Quirós et al. (2019) refer to corporate governance as how the power is exercised and how decisions are made in a bank that guarantees that members of its board of directors and executives act in the best interest of their long-term shareholders. The scope of corporate governance also embraces business ethics, disclosure and accountability (Shakil et al., 2019). Strong corporate governance may influence the financial performance of banks. Esteban-Sanchez et al. (2017) find a significant positive relationship between corporate governance and bank financial performance in an international sample that includes most developed country banks. Besides, Soana (2011) also find a positive link between corporate governance and the performance of the assets of the Italian banks analysed. Based on the agency's theory, it is expected that in banks with better governance models, shareholders and managers interests are better aligned, resulting in higher levels of efficiency. This leads us to formulate the following research hypothesis: Hypothesis IV: The relationship between corporate governance quality and banking efficiency is positive. 4.2.2. A Non-Linear Relationship Between CSP (and each of its dimensions) and Bank’s Efficiency In an attempt to reconcile the two opposing views on the relationship between CSP (and each of its dimensions) and banking efficiency and in line with the studies of Nollet et al. (2016) and Shabbir et al. (2020), we also test whether the relationship between those two variables it is non-linear. It seems reasonable to admit that for low levels of CSR activity, as it increases, bank efficiency decreases because CSR activity costs do not yet cover its benefits. However, it is expected that after a certain level of CSR activity, an increase in CSR activity will have a positive impact on bank efficiency. This means that the most efficient banks will have low or high levels of CSP. Banks with an intermediate level of CSP will be the least efficient. Based on this idea, we formulate the following research hypothesis: Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 95 Hypothesis V: The relationship between CSP (and each its dimensions) and banking efficiency is non-linear. 4.3. METHODOLOGY To investigate the formulated hypotheses, we will have to measure bank efficiency. Over time, in operational research, several techniques, both parametric and nonparametric, have been used to measure corporate efficiency. Among non-parametric techniques, Data Envelopment Analysis (DEA) has been extensively used for the efficiency evaluation of banks. Radojicic et al. (2018) present an excellent review of research that uses the DEA technique in the study of bank efficiency. The efficiency measurement indicates whether a bank maximizes the output quantity by using the given quantity of inputs or minimize the quantity of inputs used to produce a given output quantity. We apply Simar & Wilson (2007) method in a two-stage procedure to estimate bank efficiency and study its relationship with CSP and its components. In general, two major problems arise when the analysis is based on a conventional two-step procedure: (i) the lack of a well-defined data generating process (e.g., inappropriate censored regression) and (ii) misleading inference. To overcome these problems, Simar & Wilson (2007) proposed a doublebootstrap DEA approach that is grounded on a statistical theory. In the first stage, it combines the classical DEA model with the bootstrap procedure to estimate the relative efficiency scores and confidence intervals. In the second stage, efficiency estimates are regressed on a set of explanatory variables, including ESG variables, using the truncated regression with bootstrap. The authors proposed two algorithms to implement the two-stage procedure described. We use algorithm II, which is more involved and rests on bias-corrected DEA scores as the left-handside variable of the truncated regression from the second stage. Stage 1: Estimation of Efficiency Scores Using linear programming, the DEA technique allows to estimate the production frontier and calculate the efficiency score of a DMU (Decision Making Unit) to homogeneous entities. Our study focuses on European banking and assumes that the banks considered have similar characteristics and have a common production frontier: (i) common economic objective JOSÉ FERNANDO DA SILVA NETO 96 (maximize the shareholder wealth), (ii) similar activities (most perform the typical activities of commercial banking), (iii) similar regulatory environment and (iv) similar legal form. Since the original work of Charnes et al. (1978) many DEA models have been proposed in the literature (static or dynamic, with constant or variable returns to scale). The most popular are the CCR (Charnes, Cooper and Rhode, 1978) and the BCC (Banker, Charnes and Cooper, 1984) models. Both are based on radial efficiency measurements and can be carried out from both orientations (either input or output). The CCR model is based on the assumption of constant returns to scale (CRS) and the BCC model assumes that the evaluated entity may be operating under the variable returns to scale (VRS) hypothesis, implying that the relative efficiency of each DMU is obtained by comparing that DMU with those that are efficient and possess similar operational dimensions. The CRS assumption is only justifiable when all DMUs are operating at an optimal scale. However, banks or DMUs in practice might face either economies or diseconomies to scale, so in this work, following Grmanová & Ivanová (2018), we used the BCC model. The VRS assumption provides the measurement of pure technical efficiency, which is the measurement of technical efficiency devoid of scale efficiency effects (Řepková, 2014). We considered the output orientation since banks usually aim to maximize profits with an adequate combination of productive factors (inputs). For each period t (t = 1, 2, …,T), consider that exists nt DMUs (i = 1, 2, …, nt), for which we considered a set of q outputs (r = 1, 2,…,q) that produce 𝑌𝑖𝑡={𝑦𝑟𝑖𝑡} and p inputs (s = 1, 2,…,p) that consume 𝑋𝑖𝑡={𝑥𝑠𝑖𝑡}. The BCC model, with output orientation, maximizes the output keeping unchanged the inputs and can be mathematically represented as: 𝑀𝑎𝑥 𝜃0𝑡 (4.1) subject to: 𝑥𝑠0𝑡−∑𝜆𝑖𝑡𝑥𝑠𝑖𝑡 𝑛𝑡 𝑖=1 ≥0 𝑠=1,2,…,𝑝 (4.2) ∑𝜆𝑖𝑡𝑦𝑟𝑖𝑡 𝑛𝑡 𝑖=1 −𝜃0𝑡𝑦𝑟0𝑡≥0 𝑟=1,2,…,𝑞 (4.3) Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 97 ∑𝜆𝑖𝑡 𝑛𝑡 𝑖=1 =1 (4.4) 𝜆𝑖𝑡≥0 𝑖=1,2,…,𝑛𝑡 (4.5) where 𝜃0 is the efficiency score of DMU0 and 𝜆 is the weight. More precisely, 𝜃0 represents how much all outputs must be multiplied, keeping inputs unchanged, for the DMU0 to reach the efficient frontier. If 𝜃0 is equal to 1, the DMU0 is efficient, if 𝜃0 is greater than 1, the DMU0 is inefficient and higher values mean more inefficiency. Because the empirical study was carried out on panel data, we estimate the value of 𝜃 for each bank using a one-year window, as suggested by Charnes et al. (1994). One of the weaknesses of the DEA methodology is that it tends to generate biased estimates of 𝜃. To correct this weakness, we use the procedure proposed by Simar & Wilson (2000) bootstrapping the initial efficiency scores and obtaining bias-corrected efficiency estimations 𝜃𝑖𝑡. Stage 2: Estimation of Truncated Regression Next, to determine the effect of CSP and each of its dimensions on the bank’s efficiency, we estimate a truncated regression model using algorithm II proposed by Simar & Wilson (2007) where the efficiency score, from the first stage, is regressed against a set of variables that could potentially explain the bank’s efficiency, including the CSP variable and its three dimensions. The second stage regression is given by: 𝜃𝑖𝑡=𝛿𝐶𝑆𝑃𝑖𝑡+𝛽𝑍𝑖𝑡+𝜂𝐷𝑡+𝜀𝑖𝑡 (4.6) where 𝜃𝑖𝑡 is the dependent variable, the bootstrapped bias-corrected efficiency score of bank i in year t; 𝐶𝑆𝑃𝑖𝑡 is a variable that measures CSR of bank i in year t or one of each of its dimensions; 𝑍𝑖𝑡 is a vector of control variables that are expected to explain bank efficiency; 𝐷𝑡 is a vector of year dummies; 𝛿, 𝛽 and 𝜂 are the parameters to be estimated in the second stage; JOSÉ FERNANDO DA SILVA NETO 98 𝜀𝑖𝑡 is an independent error that follows the normal distribution with a zero mean and 𝜎𝜀2 variance 𝑁(0,𝜎𝜀2) with left-tail truncation (1−𝛿𝐶𝑆𝑃𝑖𝑡−𝛽𝑍𝑖𝑡−𝜂𝐷𝑡). To implement algorithm II proposed by Simar & Wilson (2007), we have to carry out the following steps: 1. For each year 𝑡=1,2,…,𝑇, using original data of outputs, 𝑌𝑖𝑡, and inputs, 𝑋𝑖𝑡 (that are all positive), estimate DEA efficiency scores for each bank, 𝜃𝑖𝑡; 2. Use the method of maximum likelihood to obtain estimates 𝛿󰆹, 𝛽󰆹 and 𝜂 of 𝛿, 𝛽 and 𝜂, respectively, as well an estimate 𝜎𝜀 of 𝜎𝜀 in the truncated regression of 𝜃𝑖𝑡 on 𝐶𝑆𝑃𝑖𝑡, 𝑍𝑖𝑡 and 𝐷𝑡 in (4.6) using the observations when 𝜃𝑖𝑡>1; 3. For each 𝑖=1,2,…,𝑛𝑡 and 𝑡=1,2,…,𝑇, loop over the next four ([3.1.]-[3.4.]) steps 𝐿1 times to obtain a set of bootstrap estimates 𝔅={𝜃𝑖𝑡 𝑏}𝑏=1 𝐿1: 3.1. Generate the residual 𝜀𝑖𝑡 from the normal distribution 𝑁(0,𝜎𝜀2) with left-truncation at (1−𝛿󰆹𝐶𝑆𝑃𝑖𝑡−𝛽󰆹𝑍𝑖𝑡−𝜂𝐷𝑡) 3.2. Compute 𝜃𝑖𝑡=𝛿󰆹𝐶𝑆𝑃𝑖𝑡+𝛽󰆹𝑍𝑖𝑡+𝜂𝐷𝑡+𝜀𝑖𝑡 3.3. Set 𝑋𝑖𝑡 ∗=𝑋𝑖𝑡 and 𝑌𝑖𝑡∗=𝑌𝑖𝑡(𝜃𝑖𝑡/𝜃𝑖𝑡) 3.4. Use 𝑋𝑖𝑡 ∗ and 𝑌𝑖𝑡∗ to estimate the pseudo-DEA efficiency scores 𝜃𝑖𝑡 𝑏; 4. For each 𝑖=1,2,…,𝑛𝑡 and 𝑡=1,2,…,𝑇, compute the bias-corrected efficiency as: 𝜃𝑖𝑡=𝜃𝑖𝑡−𝑏𝑖𝑎𝑠 𝑖𝑡 𝑏𝑖𝑎𝑠 𝑖𝑡 is the bootstrap estimator of bias, according to Simar & Wilson (1998): 𝑏𝑖𝑎𝑠 𝑖𝑡=(1 𝐿1∑𝜃𝑖𝑡 𝑏 𝐿1 𝑏=1 )−𝜃𝑖𝑡 5. Use the method of maximum likelihood to estimate the truncated regression of 𝜃𝑖𝑡 on 𝐶𝑆𝑃𝑖𝑡, 𝑍𝑖𝑡 and 𝐷𝑡 to obtain estimates 𝛿󰆹󰆹, 𝛽󰆹󰆹 and 𝜂󰆹 of 𝛿, 𝛽 and 𝜂, respectively, as well an estimate 𝜎𝜀 of 𝜎𝜀; Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 99 6. For each 𝑖=1,2,…,𝑛𝑡 and 𝑡=1,2,…,𝑇, loop over the next three ([6.1.]-[6.3.]) steps 𝐿2 times to obtain a set of bootstrap estimates 𝔇={𝛿󰆹󰆹𝑏, 𝛽󰆹󰆹𝑏,𝜂󰆹𝑏,𝜎𝜀𝑏}𝑏=1 𝐿2: 6.1. Generate the residual 𝜀󰆻𝑖𝑡 from the normal distribution 𝑁(0,𝜎𝜀2) with left-truncation at (1−𝛿󰆹󰆹𝐶𝑆𝑃𝑖𝑡−𝛽󰆹󰆹𝑍𝑖𝑡−𝜂󰆹𝐷𝑡) 6.2. Compute 𝜃𝑖𝑡=𝛿󰆹󰆹𝐶𝑆𝑃𝑖𝑡+𝛽󰆹󰆹𝑍𝑖𝑡+𝜂󰆹𝐷𝑡+𝜀󰆻𝑖𝑡 6.3. Use the maximum likelihood method to estimate the truncated regression of 𝜃𝑖𝑡 on 𝐶𝑆𝑃𝑖𝑡, 𝑍𝑖𝑡 and 𝐷𝑡 to obtain estimates bootstrap estimates 𝛿󰆹󰆹𝑏, 𝛽󰆹󰆹𝑏 and 𝜂󰆹𝑏 of 𝛿, 𝛽 and 𝜂, respectively, and 𝜎𝜀𝑏 of 𝜎𝜀; 7. Calculate confidence intervals and standard errors for 𝛿󰆹󰆹, 𝛽󰆹󰆹, 𝜂󰆹 and 𝜎𝜀 from the bootstrap distribution of 𝛿󰆹󰆹𝑏, 𝛽󰆹󰆹𝑏,𝜂󰆹𝑏 and 𝜎𝜀𝑏. In the empirical investigation, a truncated regression including also a quadratic term of 𝐶𝑆𝑃𝑖𝑡 was estimated to investigate the hypothesis of a non-linear relationship between CSP and bank efficiency (Hypothesis V). To estimate regression (4.6), we need to obtain a measure for the bank’s CSP and each of its dimensions. This concept is nowadays widely recognized in the academic and professional world as a multidimensional construct that essentially covers three aspects related to environmental, social and governance issues. This multidimensionality implies that a unidimensional quantitative index is needed to account for the simultaneous organizational aspects when assessing CSP (Belu & Manescu, 2013). In the past, many empirical studies frequently employed the Kinder, Lydenberg, Domini (KLD) data set which has become the standard measure of CSP in academic research (Mattingly, 2017). However, many researchers have questioned the weighting system used by the KLD and other indexes provided by CRS rating agencies in aggregating the different CSP dimensions into one single measure (Crane et al., 2017). According to Capelle-Blancard & Petit (2017), this aggregation process suffers from some inaccuracy and subjectivity and should not be the same across sectors. JOSÉ FERNANDO DA SILVA NETO 106 4.5. RESULTS AND DISCUSSION 4.5.1. Bank’s Efficiency Scores and ESG Indexes We apply the methodology described in subchapter 4.3., and estimate the BCC model, with output orientation, to obtain the estimates of the bank’s efficiency score (𝜃) and the biascorrected efficiency score (𝜃) for models 1 and 2 proposed in section 4.4.1. Table 33 presents the mean values of the estimates of those coefficients, as well as the number of banks used in the estimation and the percentage of fully efficient banks for the two models. Table 33 - Bank’s efficiency score and bias-corrected efficiency score (means) by year based on the BCC model, with output orientation, using DEA Model 1 Model 2 Inputs: I(DEP) and I(ACL) Inputs: I(FA) and I(ACL) Outputs: O(L) and O(NII) Outputs: O(L) and O(NII) Year Efficiency Score Biascorrected efficiency score % Efficient DMUs' Efficiency Score Biascorrected efficiency score % Efficient DMUs' Number of banks 2011 2.177 2.569 13.89 2.022 2.395 12.50 72 2012 1.681 1.898 18.06 1.668 1.903 12.50 72 2013 1.701 1.925 13.70 1.585 1.789 13.70 73 2014 2.265 2.714 13.51 2.254 2.764 16.22 74 2015 1.843 2.135 13.92 1.614 1.817 17.72 79 2016 1.860 2.153 13.25 1.515 1.673 22.89 83 2017 2.231 2.667 11.36 2.161 2.581 22.73 88 2018 2.182 2.580 9.26 1.715 1.960 21.30 108 2019 2.053 2.410 8.79 1.550 1.761 21.98 91 2011-2019 2.011 2.355 12.57 1.781 2.062 18.38 740 Source: Own production In general, in average terms, the level of the bank’s efficiency in Europe of the period 20112019 was low. Considering the efficiency scores of models 1 and 2, we can conclude that, keeping the inputs unchanged, the outputs would have been multiplied, on average, by 2.011 and 1.781 times, respectively, for a given bank to reach the efficient frontier. These results are in line with those obtained by Neves et al. (2020) and Christopoulos et al. (2020) who reported low levels of efficiency in European banking in the periods of 2011-2016 and 2009-2015, respectively. These low levels of efficiency could be explained by activity restrictions and high Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 107 capital requirements imposed by the regulatory authorities during the European sovereign debt crisis (Bace & Ferreira, 2020). Bias-corrected efficiency scores reported shows that the rankings do not change substantially, however, efficiency scores are generally augmented. Comparing the results obtained for models 1 and 2, we conclude that the average efficiency level and the percentage of fully efficient banks are higher in model 225, and for this reason, the bias-corrected efficiency scores of this model are used in estimating the truncated regression of the second stage. To measure CSP, ESG indexes were estimated using the DEA-WEI model described in subchapter 4.3. and section 4.4.2. Figure 10 presents the evolution of the mean value of the bias-corrected index of ESG activity and each of its dimensions (𝜙)26. As we can see, in the period 2011-2019, European banking presents good levels of performance at the level of ESG activity with the bias-corrected efficiency score to present an average value of 1.260, that is, very close to unity. Figure 10 - Bank’s ESG activity and each of its dimensions bias-corrected indexes (means) Source: Own production 25 Remember that a fully efficient bank will have a coefficient 𝜃 equal to one and a higher coefficient means that bank is less efficient. 26 In Table 39 of the appendix of this chapter is presented detailed information about the evolution of ESG activity and its dimensions indexes, by year, using DEA-WEI model. 0,0 0,5 1,0 1,5 2,0 2,5 3,0 3,5 4,0 4,5 2011 2012 2013 2014 2015 2016 2017 2018 2019 ESG Index Environmental Index Social Index Governance Index JOSÉ FERNANDO DA SILVA NETO 108 In the Social and Governance dimensions of the CSP, it can be concluded that European banking also has good levels of performance. The mean value of the corporate governance index ranged between a minimum value of 1.495 in 2019 and a maximum value of 1.702 in 2013. In the activities related to the social pillar, there is even an increase in performance, with the respective index decreasing from 1.731 in 2011 to 1.393 in 2019. The dimension of the CSP in which European banking has the worst performance is the environmental dimension, with an average index of 3.140 in the period 2011-2019. This result was somewhat expected as banking activity has, at least directly, little environmental impact. However, over the period studied, there was a significant increase in the performance of this dimension, with the Environmental index falling from 3.332 in 2011 to 2.292 in 2019. In Table 34, we present the descriptive statistics of the variables of interest in our study and that will be used to test the hypotheses formulated in subchapter 4.2.: the bias-corrected efficiency score (𝜃) and the bias-corrected index of ESG activity and each of its dimensions (𝜙). Table 34 - Descriptive statistics of banks’ efficiency and CSP and each of its dimensions Obs. Mean Median Std.Dev. Min Max Bank Efficiency Score 740 2.997 2.457 1.798 1.086 13.860 ESG Index 740 1.260 1.090 0.537 1.013 9.267 Environmental Index 740 3.140 1.256 4.419 1.010 20.878 Social Index 740 1.668 1.172 1.726 1.012 17.822 Governance Index 740 1.565 1.246 1.114 1.011 19.761 Source: Own production 4.5.2. Truncated Regression Analysis for Bank’s Efficiency Table 35 presents the results for the truncated regression that allows us to analyse the effect of different variables, including those related to CSP, on the bank’s efficiency. The results presented assume that the bank’s efficiency is linearly related to the CSP and each of its dimensions. As already mentioned, to measure the bank’s efficiency, we used the bootstrapped bias-corrected efficiency score (𝜃), and the CSP and each of its dimensions were measured by the ESG Index, Environmental Index, Social Index and Governance Index described in section 4.5.1. Column (1) presents the estimation results of Eq. (4.6), including only the bank-specific Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 109 and the macroeconomic control variables. In column (2)-(4), CSP’s variables were included to investigate the effect of CSP and each of its dimensions on the bank’s efficiency. Table 35 - Results of bootstrap truncated regressions for determinants of bank’s efficiency (linear relationship assumed between CSP and bank’s efficiency) (1) (2) (3) (4) (5) Size -1.299*** -1.224*** -1.323*** -1.333*** -1.248*** [-1.480,-1.087] [-1.405,-1.009] [-1.512,-1.107] [-1.533,-1.117] [-1.428,-1.033] Revenue Diversificat. -0.043*** -0.043*** -0.043*** -0.043*** -0.043*** [-0.056,-0.030] [-0.056,-0.031] [-0.057,-0.030] [-0.056,-0.030] [-0.056,-0.030] Liquidity -0.017* -0.019** -0.017* -0.016* -0.019** [-0.035,0.002] [-0.037,-0.001] [-0.035,0.001] [-0.036,0.003] [-0.037,0.000] Return on assets -0.336*** -0.327*** -0.336*** -0.340*** -0.323*** [-0.466,-0.184] [-0.459,-0.178] [-0.468,-0.184] [-0.469,-0.193] [-0.455,-0.184] Equity to assets ratio -0.076*** -0.072*** -0.077*** -0.078*** -0.071*** [-0.139,-0.028] [-0.132,-0.024] [-0.142,-0.025] [-0.141,-0.027] [-0.130,-0.023] Board independence 0.027*** 0.027*** 0.027*** 0.026*** 0.027*** [0.018,0.036] [0.017,0.036] [0.017,0.036] [0.017,0.036] [0.018,0.036] Board gender diversity 0.008 0.009 0.008 0.007 0.009 [-0.010,0.025] [-0.007,0.027] [-0.010,0.025] [-0.010,0.025] [-0.009,0.026] HHI 4.162** 3.751** 4.075** 4.107** 4.006** [0.638,7.506] [0.212,7.109] [0.468,7.504] [0.478,7.601] [0.352,7.172] Real GDP growth -0.110*** -0.116*** -0.106*** -0.107*** -0.116*** [-0.193,-0.032] [-0.201,-0.041] [-0.195,-0.029] [-0.195,-0.029] [-0.198,-0.041] Inflation -0.564*** -0.552*** -0.563*** -0.583*** -0.560*** [-0.841,-0.281] [-0.819,-0.257] [-0.853,-0.293] [-0.853,-0.284] [-0.859,-0.281] ESG Index 0.547*** [0.229,0.856] Environmental index -0.019 [-0.063,0.025] Social index -0.093 [-0.224,0.026] Governance index 0.213*** [0.058,0.353] Constant 19.295*** 17.892*** 19.625*** 19.897*** 18.453*** [16.857,21.552] [15.412,20.005] [17.026,22.015] [17.244,22.42] [15.96,20.77] Sigma 1.898*** 1.879*** 1.899*** 1.908*** 1.881*** [1.683,2.040] [1.663,2.015] [1.703,2.038] [1.695,2.039] [1.688,2.022] Note: The table reports the estimation results for the truncated regression using algorithm II proposed by Simar & Wilson (2007). The dependent variable of all regressions is the bootstrapped bias-corrected efficiency score obtained considering in the first stage the BCC model, with output orientation, with two inputs [(I)FA, (I)ACL] and two outputs [(O)L, O(NII)]. All regressions were estimated with 740 observations. Time Dummies are included. The number of bootstrap replications for bias correction of DEA scores and for estimating confidence intervals (CI) for the regression coefficients was 2000. The 95% CI are reported in the squared brackets. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production Looking to column (1), we conclude that: (i) the coefficient associated with the bank’s size is statistically significant, showing that a larger size implies higher levels of efficiency27; (ii) banks with more revenue diversification present more high levels of efficiency; (iii) better27 Remember that a higher value of 𝜃 means that bank is more inefficient. JOSÉ FERNANDO DA SILVA NETO 110 capitalized and more profitable banks are more efficient; (iv) banks that present a greater percentage of independents directors on the board are more inefficient; (v) banks that are based in countries with more concentrated banking sectors, with a higher HHI, are less efficient; and finally, (vi) in a context of economic and inflationary expansion, banks are more efficient. The gender diversity of the board seems does not to influence the bank’s efficiency and the variable liquidity only are significant to a significance level of 10%. Considering now the effect of ESG activities, measured by the ESG Index, on the bank’s efficiency, we conclude that an increase in the ESG index increases the efficiency score (the sign of the estimate of the coefficient associated with the ESG Index variable in column (2) is positive and significant for 1% of significance level). This means that the statistical evidence supports Hypothesis I which states that banks with worse CSP are less efficient. These results are in line with those obtained by Belasri et al., (2020), who find evidence that CSR has a positive impact on bank efficiency in developed countries, and Shah et al. (2019) who found support in their study that sustainable banks are more efficient and productive. Looking at the results presented in columns (3) and (4), we do not find statistical evidence that banks with better social and environmental practices are more efficient since the coefficients associated with the Environmental Index and Social Index variables are not statistically significant. This means that our results do not support Hypotheses II and III of our study. However, we found evidence that supports Hypothesis IV. Looking at column (5), we could see that the estimate of the coefficient associated with the Governance Index is positive and statistically significant. This means that banks with good governance practices are more efficient. These first results seem to support what is advocated by the stakeholder theory, according to which banks with the best CSP will be the ones with the best levels of efficiency. However, the dimension linked to the bank's governance model seems to be the only one that contributes to that positive relationship. As suggested by agency theory, the banks that adopt governance practices that best align the interests of shareholders and managers will be the ones that will be most efficient. Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 111 Table 36 - Results of bootstrap truncated regressions for determinants of bank’s efficiency (nonlinear relationship assumed between CSP and bank’s efficiency) (1) (2) (3) (4) Size -1.178*** -1.283*** -1.274*** -1.224*** [-1.365, -0.974] [-1.467, -1.059] [-1.456, -1.049] [-1.402, -1.017] Revenue Diversification -0.044*** -0.043*** -0.044*** -0.044*** [-0.056, -0.031] [-0.057, -0.030] [-0.057, -0.031] [-0.056, -0.030] Liquidity -0.022** -0.017* -0.017* -0.022** [-0.042, -0.003] [-0.037, 0.001] [-0.036, 0.001] [-0.040, 0.001] Return on assets -0.319*** -0.337*** -0.342*** -0.308*** [-0.457, -0.174] [-0.477, -0.191] [-0.482, -0.198] [-0.444, -0.168] Equity to assets ratio -0.072*** -0.076*** -0.078*** -0.073*** [-0.129, -0.027] [-0.136, -0.025] [-0.140, -0.027] [-0.134, -0.025] Board independence 0.027*** 0.027*** 0.026*** 0.028*** [0.018, 0.036] [0.017, 0.036] [0.016, 0.036] [0.018, 0.037] Board gender diversity 0.008 0.008 0.007 0.008 [-0.009, 0.025] [-0.010, 0.025] [-0.010, 0.026] [-0.010, 0.025] HHI 3.590** 4.072** 3.994** 4.036** [0.008, 6.911] [0.236, 7.401] [0.425, 7.423] [0.596, 7.386] Real GDP growth -0.122*** -0.111*** -0.110*** -0.124*** [-0.210, -0.046] [-0.198, -0.029] [-0.193, -0.035] [-0.207, -0.046] Inflation -0.549*** -0.561*** -0.574*** -0.567*** [-0.828, -0.268] [-0.841, -0.293] [-0.859, -0.282] [-0.846, -0.276] ESG Index 1.375*** [0.496, 2.181] ESG Index squared -0.111** [-0.220, -0.003] Environmental index 0.142 [-0.069, 0.333] Environmental index squared -0.009 [-0.019, 0.002] Social index 0.334 [-0.057, 0.771] Social index squared -0.034** [-0.073, -0.007] Governance index 0.507*** [0.157, 0.813] Governance index squared -0.021* [-0.043, 0.002] Constant 16.662*** 18.976*** 18.822*** 18.875*** [13.966, 19.058] [16.269, 21.48] [16.01, 21.26] [15.301, 20.05] Sigma 1.888*** 1.904*** 1.906*** 1.894*** [1.674, 2.016] [1.690, 2.029] [1.700, 2.036] [1.692, 2.017] Note: The table reports the estimation results for the truncated regression using algorithm II proposed by Simar & Wilson (2007). The dependent variable of all regressions is the bootstrapped bias-corrected efficiency score obtained considering in the first stage the BCC model, with output orientation, with two inputs [(I)FA, (I)ACL] and two outputs [(O)L, O(NII)]. All regressions were estimated with 740 observations. Time Dummies are included. The number of bootstrap replications for bias correction of DEA scores and for estimating confidence intervals (CI) for the regression coefficients was 2000. The 95% CI are reported in the squared brackets. *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. Source: Own production JOSÉ FERNANDO DA SILVA NETO 112 To analyse the validity of the hypothesis of a non-linear relationship between bank efficiency and CSP (and each of its three dimensions) [Hypothesis V], equation (4.6) was reestimated by additionally including the term 𝐶𝑆𝑃𝑖𝑡 2. The estimation results are shown in Table 36. Looking at column (1) of that table, we concluded that the coefficients associated with the ESG index and ESG index squared variables are both significant for a 5% significance level, which allows us to conclude by a non-linear relationship between the CSP and banks efficiency. Namely, considering the signals obtain for the estimates of the coefficients, we can conclude by a U-shaped relationship between the CSP and the efficiency of the banks. That is, banks with low or high CSP are the most efficient. Banks with intermediate levels of CSP are the most inefficient. These results are in line with those obtained by Nollet et al. (2016) and Shabbir et al. (2020) who also find a U-shaped relationship between CSP and financial performance. These results allow us to reconcile the two opposing theoretical views on the relationship between CSP and banking efficiency. In favour of the trade-off view of ESG activities we can point to the fact that if banks with low levels of CSP invest more money in CSR activities they will see their efficiency levels go down. In support of stakeholder theory, evidence is found that for banks with high levels of CSP, additional investment in ESG activities tends to improve efficiency levels. The results of columns (2)-(4) of Table 36 allow us to draw the same conclusion for the social and governance dimensions of ESG activities, that is, banks with low or high performance in those two dimensions are the most efficient. Banks with intermediate performance levels in social and governance dimensions of ESG activities are the least efficient. These results imply that if a bank decides to invest in socially responsible practices, it will have to do so in a sustained way to obtain high levels of performance, as only for these levels the investment is transformed into efficiency gains. As referred by Birindelli et al. (2015), stable and fair relationships between employees and management will lead to higher personal satisfaction and loyalty, contributing to an increase in corporate efficiency. Banks that sustainably engage with the local communities in which they operate can build an image of a good reputation that results in increased demand for their products and services increasing the bank’s efficiency. The environmental dimension of ESG activities continues to prove insignificant in explaining the efficiency of banks. These results corroborate the arguments of Finger et al. Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 113 (2018) according to which banks of developed country have already optimized their processes in such a way that additional environmental measures do not result in efficiency gains. Figure 11 shows the relationship between a bank's inefficiency and ESG performance and its social and governance dimensions for a bank representative of our sample (average values). We can conclude that a bank's inefficiency is maximum when the ESG index takes the value of 6.2, the Social index takes the value of 12.1 and the Governance index takes the value of 4.9. It is also noted that the U-shaped curve is less pronounced when the Social index is related to the bank’s efficiency. Figure 11 - Relationship between Bank’s inefficiency and ESG, Social and Governance indexes for a representative bank 0 1 2 3 4 5 6 0 2 4 6 8 10 12 Bank Ineffciency ESG Index 0 1 2 3 4 5 6 7 0 5 10 15 20 25 Bank Inefficiency Governance index JOSÉ FERNANDO DA SILVA NETO 114 Source: Own production 4.6. ROBUSTNESS ANALYSIS Several studies have highlighted that financial factors are crucial in explaining the adoption of CSR practices. The most efficient banks will be, a priori, better able to have these financial resources. Based on this idea, it is reasonable to assume that bank efficiency can itself influence CSP, resulting in a possible bidirectional relationship between CSP and bank efficiency. This means that an endogeneity problem arises when we estimate Eq. (4.6), given the simultaneity between the bank's efficiency and the CSP. To overcome the problem of endogeneity motivated by the simultaneity between bank efficiency and CSP and by possible correlation between the delayed endogenous variable and the unobservable effects, we use the System Generalized Method of Moments (GMM) proposed by Arellano & Bover (1995) and Blundell and Bond (1998), described in the chapter 2, to re-estimate the Eq. (4.6) including the term 𝐶𝑆𝑃𝑖𝑡 2 given the evidence of a non-linear relationship. This method combines the first differences in our regression equation with the level form, reducing any biases and imprecision associated with the first-difference GMM. We use the two-step GMM estimator, instead one-step GMM estimator, with Windmeijer (2005) corrected standard errors, because is more efficient. To satisfy the instruments' validity, we test for over-identifying restrictions using Hansen's (1982) J test and Arellano-Bond test to guarantee that second-order autocorrelation coefficient, RHO(2), is null. The results presented in Table 37 are free from any endogeneity issues and using System GMM also allow us to control for persistence. The results obtained only confirm the non-linear relationship between the social dimension of ESG activities and the bank’s efficiency. Given 0 0,5 1 1,5 2 2,5 3 3,5 4 0 2 4 6 8 10 Bank inefficiency Social index Does Corporate Social Performance improve the Bank’s Efficiency? Evidence from the European Banking 115 these results, we re-estimate the model excluding the quadratic terms from the ESG index, Environmental index, Social index and Governance index. Results are presented in Table 38. The results indicate the existence of a significant and positive linear relationship between the efficiency of a bank and the CSP. At the disaggregated level, a good performance in the social and governance components improves the bank's efficiency.