新冠疫情与银行分行贷款：数字化的调节作用.pdf

Journal of Banking and Finance 152(2023)106869 Contents lists available at ScienceDirect Journal of Banking and Finance journal homepage: COVID-19 and bank branch lending:The moderating effect of digitalization?Thiago Christiano Silva a,b,Sergio Rubens Stancato de Souza a,Solange Maria Guerra a,Benjamin Miranda Tabak c a Research Department,Banco Central do Brasil,Braslia,Brazil b Universidade Catlica de Braslia,Distrito Federal,Brazil c FGV/EPPG Escola de Polticas Pblicas e Governo,Fundao Getlio Vargas,Distrito Federal,Brazil a r t i c l e i n f o Article history:Received 12 October 2022 Accepted 28 April 2023 Available online 4 May 2023 JEL classification:C58 D22 D40 G21 I19 O31 Keywords:COVID-19 Digitalization Bank branch Marginal cost Bank lending Credit type a b s t r a c t We examine how COVID-19 and digitalization have changed bank lending behavior.Using microdata from Brazil,we investigate the determinants of these changes at the bank branch level and by credit type.Branches in areas more affected by COVID-19 reduced loan issuances and experienced lower credit rev-enues.These branches could not adjust their costs in the short term due to this decline in lending,result-ing in increased marginal costs.We also find that branches of more digitalized banks were less sensitive to local borrowers conditions and could expand their clientele.These branches extended credit to bor-rowers in remote localities less affected by COVID-19,positioning themselves better than branches of less digitalized banks.Our research highlights the critical role of digitalization in distressed periods,as it enables banks to respond more swiftly and effectively,favoring financial stability.2023 Elsevier B.V.All rights reserved.1.Introduction The COVID-19 pandemic affected the global economy,causing recessions,business failures,and increased unemployment.How-ever,the impact was heterogeneous among economic agents,sec-tors,and regions(Demirg-Kunt et al.,2021;Hasan et al.,2021;Muggenthaler et al.,2021).1 The high level of uncertainty had di-?We thank the two anonymous referees and Vasso Ioannidou for their insightful comments.The views expressed in this paper are those of the authors and do not necessarily reflect those of the Banco Central do Brasil.Corresponding author at:Banco Central do Brasil,Setor Bancrio Sul(SBS)Quadra 3 Bloco B-Ed.Sede,Braslia,DF CEP:70 074-90 0,Brazil.E-mail addresses:thiago.silvabcb.gov.br(T.C.Silva),sergio.souzabcb.gov.br(S.R.S.de Souza),solange.guerrabcb.gov.br(S.M.Guerra),benjamin.tabakfgv.br(B.M.Tabak).1 For instance,the entertainment,restaurants,and tourism sectors,which require face-to-face interaction,suffered severe losses.In contrast,information,communi-cation,and delivery experienced substantial growth(Rio-Chanona et al.,2020).The economic consequences of COVID-19 also varied significantly between countries,depending on their economy s pre-pandemic conditions,the extent of public con-verse consequences on banks due to changes in depositors behav-ior and the supply and demand for loans(Levine et al.,2021;olak and zde ztekin,2021;Fuster et al.,2021;Acharya and Steffen,2020).Depending on their location,sectoral lending exposure,and level of digitalization,the pandemic may have affected banks dif-ferently.This paper investigates this hypothesis by examining the role of bank digitalization on lending behavior during the first year of the pandemic at the branch level.Innovation is a key factor that could have influenced COVID-19 s impact on bank lending behavior.Like other pandemics,COVID-19 transformed and accelerated certain trends,2 including digital-tainment measures,and the quality of institutional settings(Baumeister et al.,2022;Muggenthaler et al.,2021).2 We highlight the following structural changes in previous pandemics.The Black Death destroyed a large portion of the world s workforce and resources,contribut-ing to the shift from labor-based to capital-based production and significantly increasing rural-urban migration(Clark,2016).As of 2014,the Spanish flu was the fourth largest economic shock to income and consumption after WWII,WWI,and the Great Depression(Barro and Ursa,2008).Additionally,pandemics have https:/doi.org/10.1016/j.jbankfin.2023.106869 0378-4266/2023 Elsevier B.V.All rights reserved.T.C.Silva,S.R.S.de Souza,S.M.Guerra et al.Journal of Banking and Finance 152(2023)106869 ization(Ceylan et al.,2020).Before the COVID-19 outbreak,finan-cial systems were undergoing a heavy process of digitalization.3 With the introduction of public health measures that discouraged person-to-person contacts,this process accelerated in both the fi-nancial and real sectors(Saka et al.,2022;OECD,2020).One important question is whether this financial dig-italization alleviated the COVID-19 impact on banks.Dadoukis et al.(2021)reported that banks with higher levels of digitalization issued more loans and charged lower interest rates at the onset of the pandemic.Consequently,we would expect that the distribution of bank technologies(digitalization levels)would result in differential effects of the COVID-19 shock on banks.The degree of internal technology influences the or-ganization of bank branch networks.Banks with more advanced technological resources can close branches because they can collect and process hard data more efficiently,allowing them to extend credit to consumers over greater distances.However,it may occur a possible adverse effect that greater(technology-enabled)distance may deteriorate loan quality if lending becomes riskier(Granja et al.,2022).In contrast,banks with lower levels of technology rely more on soft information collected by their branches(Keil and Ongena,2020;Hauswald and Marquez,2003).We extend the analysis of Dadoukis et al.(2021)and examine the heterogeneous impact of the pandemic on banks based on their levels of digitalization.We also investigate the impact of bank digitalization on branches lending behavior,including credit income,cost,funding,provisions,and geographic lending patterns.We address these empirical questions by employing a difference-in-differences approach with several matched datasets from Brazil,a country with interesting features that allow us to assess the pandemic s impact on the operations of bank branches.First,Brazil is a strongly bank-oriented economy:SMEs mostly rely on bank credit for external funding,and credit for households is provided mostly by the banking system.Second,according to the Survey of Banking Technology2020 and 2021 editionsfrom the Brazilian Federation of Banks(Febraban),Brazilian banks modern-ized their technology park in 2019.These surveys showed that the budget for investment in technology from the 21 banks(represent-ing 87%of the financial system s total assets in 2020)increased 48%in 2019 when compared to 2018.This led to an increase in the number of banking operations carried out through digital channels,further intensified by the pandemic.In 2020,7 out of 10 banking operations and 9 out of 10 credit issuances were carried out us-ing digital channels.Third,bank credit in Brazil increased substan-tially during COVID-19 due to government programs designed to combat the economic effects of the pandemic.Despite being per-vasive across all Brazilian regions,credit growth rates were hetero-geneous across municipalities in 2020.Fourth,Brazil has a consid-erable variety of economic development,climate,and demograph-ics across municipalities,leading to very distinct settings where bank branches operate.Fifth,during COVID-19,the federal govern-ment took the majority of economic measures to combat the pan-demic,such as providing financial support to families and small businesses,deferring loan payments,modifying banking system re-serve requirements to increase credit capacity,and easing mone-medium and long-term economic effects by leading to labor shortages or precau-tionary savings(Jord et al.,2022)and increasing income inequality(Galletta and Giommoni,2022).3 Philippon(2015)examines data up to the global financial crisis in 2008 and finds an interesting puzzle.Despite significant advances and investments in com-puter and communications technologies,the unit cost of financial intermediation has remained close to 200 basis points for over a century.Philippon(2020)reruns the model with data after the global financial crisis and finds that the unit cost of financial intermediation declined from 2010 to 2020.He attributes the structural change to the rapid growth of fintechs,which have benefited from digital innova-tions that can disrupt industry structures.tary policy.The diversity of characteristics of the localities and the different ways the pandemic affected them led to a great diversity of local economic developments,even in the context of a similar set of government measures.This diversity provides the conditions for examining the impact of COVID-19 on branches lending behav-ior.In addition,Brazil has detailed datasets that make these analy-ses feasible.Despite the expansion of digital services,physical bank branches remain essential in Brazil.The majority of customers of non-digital banks have a contractual connection with a physical branch due to the pre-digitalization legacy.Particularly in 2020,according to Febraban,the physical presence of the customer in the branches is still important for more complex operations,such as credit renegotiation.Additionally,the opening of new accounts in these banks can still be done through physical channels in branches.In a sample of eight banks,the 2021 edition of the Febraban Survey of Banking Technology reveals that the number of accounts opened via digital channels increased more than those opened via physical channels between 2020 and 2019.However,the total number of accounts opened via physical channels was still greater than those opened via digital channels.Finally,not all traditional banks have the same resources as large banks to invest rapidly in technology.These elements keep the physical location of bank branches relevant.Our empirical identification exploits the fact that municipali-ties in Brazil faced substantially different levels of COVID-19 sever-ity.Brazil experienced a spread pattern of COVID-19 similar to the United States(Bollyky et al.,2022;Siddique et al.,2022).Initially,the virus heavily impacted state capitals,which are the most pop-ulous areas in Brazil and home to major airports.The virus subse-quently spread to inland municipalities.Neighboring Brazilian mu-nicipalities have strong social,economic,and financial intercon-nections.These interconnections form networks,typically with a large urban center surrounded by small towns.These geographical units are more suitable for our model than those based on mu-nicipal boundaries,as banks can obtain the majority of their lo-cal resources,such as personnel,from the same geographical unit.Based on this concept of urban networks,the Brazilian Institute of Geography and Statistics(IBGE)defines the geographical unit known as the Immediate Geographic Region.This unit is compara-ble to counties in the US,both in size and by being formed by neighboring municipalities.We use these geographical units in our analysis,which we refer to as a locality for simplicity.Similar to Levine et al.(2021),we exploit the cross-sectional heterogeneity of COVID-19 shocks across localities.We measure the locality s COVID-19 intensity as the number of COVID-19 cases as a share of the locality s population.4 In terms of identification,one potential concern is that the timing and intensity of COVID-19 in a locality are likely not exogenous to its economic and geographic conditions,such as population,economic development,distance to state capitals,and economic structure.The economy s structure may affect COVID-19 transmission rates as agriculture-related ac-tivities are geographically sparser than services and industrial ac-tivities.We show that once we compare localities within the same macrolocality 5 a group of contiguous and economically dependent 4 A natural measure to assess the COVID-19 spread is mobility.However,the available databases for Brazil either have a higher level of aggregation than is re-quired for analyzing the local credit markets,or they do not provide broad geo-graphic coverage.The Google(Community Mobility Reports)is only available at the state level,the So Paulo Government Intelligent Monitoring System includes data only for 104 cities in the state of So Paulo with more than 70 thousand inhabi-tants,and the BID/WAZE has data for large cities only.5 Macrolocality refers to Intermediate Geographical Region defined by IBGE.All municipalities within the same Intermediate Geographical Region belong to the same state.To clarify,we have the following geographical hierarchy in Brazil:mu-nicipalities(5570 municipalities in 2021)Immediate Geographical Region(510 2 T.C.Silva,S.R.S.de Souza,S.M.Guerra et al.Journal of Banking and Finance 152(2023)106869 localitiesand with similar wealth levels,the locality s COVID-19 intensity becomes unrelated to many locality-specific correlates.This fairly exogenous variation of COVID-19 intensity across locali-ties within the same macrolocality and with similar wealth levels is essential to support the causal interpretation of the results.6 To assess branch lending behavior,we aggregate all branches of a specific bank located in the same locality to form an aggre-gate branch of the bank.We refer to this entity as the bank branch for ease of readability.The term“locality”refers to the geographi-cal location of the bank branch that extends credit.Borrowers can be located anywhere.This approach accommodates bank branches lending both to local and remote borrowers,which is important given the increasing bank digitalization and adoption of online banking(see Fig.S1 in the Supplementary Material).We exam-ine changes in effective pricesi.e.,credit revenues over granted creditand costs of the loans issued during the pandemic.If the cost of a credit type decreases or its revenue increases relative to the amount granted,branches have incentives to channel resources to this type of credit.We hypothesize that effective price and costs,as well as digitalization,play a significant role in the lending prac-tices of branches.We have an extensive dataset at the loan level,which enables us to measure effective prices by credit type for each branch.Unfortunately,we lack the same level of detail re-garding costs,as banks only report aggregated costs.We overcome this limitation by estimating the marginal costs of credit issuances for each credit type and branch using a granular branch-level pro-duction function.This approach allows us to perform a wide range of cost-related analyses.We resort to a within-bank and across-locality empirical strat-egy to analyze how COVID-19 affected branch lending behavior and the role of bank digitalization.This strategy enables us to iso-late bank-specific credit supply shocks while allo