Lending relationships in loan renegotiation; evidence from

Lending relationships in loan renegotiation; evidence from (PDF)

2022 • 62 Pages • 524.39 KB • English
Posted June 30, 2022 • Submitted by pdf.user

Visit PDF download

Download PDF To download page

Summary of Lending relationships in loan renegotiation; evidence from

Working Paper Series Lending relationships in loan renegotiation: evidence from corporate loans Melina Papoutsi Disclaimer: This paper should not be reported as representing the views of the European Central Bank (ECB). The views expressed are those of the authors and do not necessarily reflect those of the ECB. No 2553 / May 2021 Abstract This paper presents evidence that personal relationships between corporate borrowers and bank loan officers improve the outcomes of loan renegotiation. Analysing a bank reorganization in Greece in the mid-2010s, I find that firms that experience an exogenous interruption in their loan officer relationship confront three consequences: one, the firms are less likely to renegotiate their loans; two, conditional on renegotiation, the firms are given tougher loan terms; and three, the firms are more likely to alter their capital structure. These results point to the importance of lending relationships in mitigating the cost of distress for borrowers in loan renegotiations. Keywords: loan renegotiation, bank branch closures, loan officers, corporate credit JEL codes: G21, L14, E44, E58, O16 ECB Working Paper Series No 2553 / May 2021 1 Non-technical summary An important issue for bank governance, risk and credit supply is the management of lending rela- tionships. On the one hand, such relationships can reduce asymmetric information and improve loan monitoring. On the other hand, they can contribute to ever-greening behavior and potentially to an in- crease in non-performing exposures. This paper studies these trade-offs by looking at interruptions in relationships between loan officers and borrowing firms and their implications for loan renegotiations and firms’ sources of financing. Two motivations, among others, are to inform bank managers and su- pervisors about the implications of loan officer rotation as part of regular bank governance as well as resolution authorities about potential implications of bank restructurings, which may lead to the removal of loan officer. The analysis draws on a novel dataset on corporate loans and bank reorganization in Greece in the mid-2010s, which helps to empirically identify the causal effect of interrupted relationships. The main findings are that when the relationship between a loan officer and a firm is interrupted: (1) the firm has a significantly lower probability to renegotiate a loan compared to a firm with a continuous relation- ship; (2) when renegotiation occurs, firms with interrupted loan officer relationships receive tougher loan terms – notably, significantly shorter maturities and higher collateral requirements, whereas interest rates hardly increase –; and (3) firms with interrupted relationships raise more equity, decrease leverage, and substitute partially lending from other banks. An interruption in the loan officer relationship could be caused by different reasons such as by the closure of branches, which is quite frequent nowadays as a consequence of digitization or consolidation, a loan officer rotation scheme, or a bank restructuring. The first and the last often happen in a situation of financial distress. The results thus suggest that bank managers, supervisors, and resolution authorities need to be mind- ful of the potential costs of changed loan officers. For example, in a context of general stress multiple interruptions of bank relationships could have a significant effect on firms’ capital structure and borrow- ing capacity. Overall, while the results indicate net benefits of continued relationships for firms, they do not lead to unambiguous conclusions for bank risk. ECB Working Paper Series No 2553 / May 2021 2 1 Introduction Are personal relationships with bank loan officers consequential for corporate borrowers? Could stronger relationships help a firm secure better loan terms in a renegotiation? Most corporate credit is mediated by a personal relationship between a firm and a loan officer, as this connection helps to mitigate agency prob- lems. Financial intermediation theory suggests that relationships can be beneficial because they reduce information asymmetries and alleviate moral hazard (through monitoring, screening, repeated interac- tion etc.).1 At the same time, strong relationships between borrowers and lenders may give rise to other problems, e.g., soft-budget-constraint problems, hold-up problems, or nepotistic behavior.2 Whether the benefits of these relationships offset the costs is a challenging empirical question. In this paper, I study the impact of lending relationships between loan officers and firms on loan renegotiations and investigate whether these relationships have significant effects on the probability of renegotiation and on the newly agreed loan terms. A renegotiation can be initiated by either the creditor or the borrower prior to or coincident with default on a loan. Classic contract theory suggests that in unanticipated states of the world, renegotiation is Pareto improving due to the debt-overhang.3 Especially when financial distress results from a macroeconomic shock, it is probable that both the bank and the borrower benefit from a successful renegotiation. Renegotiation relaxes inefficient constraints on the borrowers as well as the bank and can increase expected loan repayments. There are two main challenges for accurately estimating the impact of personal relationships on loan renegotiation. The first is the difficulty of quantifying the value of a personal relationship. No direct measure of relationship intensity exists. The length of a given relationship may seem like a straightfor- ward measure, but the endogeneity of the decision to sever an existing relationship will complicate the interpretation of the time span measure. The endogeneity factor creates the second challenge. A bank’s decision to break an existing relationship may reflect its perception of the declining creditworthiness of the borrower. Under some circumstances a successful firm may seek to broaden its access to external finance by weakening its relationship with the particular bank. Such decisions are endogenous and would bias any results estimated by treating relationships as exogenous in a simple OLS framework. To overcome these challenges, I use micro data and a 2013 bank reorganization in Greece. This 1Classic references: Leland and Pyle (1977), Diamond (1984), Ramakrishnan and Thakor (1984), Allen (1990) 2Sharpe (1990), Rajan (1992), Dewatripont and Maskin (1995), Bolton and Scharfstein (1996) 3Hart and Moore (1988), Rajan (1992), Aghion, Dewatripont, and Rey (1994), Bolton and Scharfstein (1996), Hart and Moore (1998), Maskin and Moore (1999), Gorton and Kahn (2000), Garleanu and Zwiebel (2008) ECB Working Paper Series No 2553 / May 2021 3 experiment provides exogenous variation in the length of the relationships between loan officers and firms. Moreover, detailed confidential data on corporate loans allow me to quantify accurately the effect of interrupted relationships, by controlling for region, bank unit, and firm-specific effects. My central finding is that relationships between loan officers and firms have a significant positive impact on loan renegotiation. Firms with interrupted relationships are less likely to renegotiate a loan compared to firms experiencing continuous relationships. In addition, firms with interrupted relationships receive tougher loan terms on the loans that are renegotiated. I also observe that firms alter their capital structure after the relationship with the loan officer is interrupted. The empirical setting is based on the consolidation of a major commercial bank in Greece with business activity throughout the country. Bank network consolidation is a common response of banks to financial distress, as consolidation reduces operating costs and centralizes lending decisions.4 Dur- ing consolidation, some bank units are closed and the loan accounts from those units are merged with accounts in other surviving units. A bank unit closure interrupts personal relationships between loan of- ficers and firms because merged accounts obtain new loan officers. Thus, after consolidation, two types of firms are identified: one, those whose loans were transferred to another unit and whose personal re- lationships were consequently discontinued, and two, those that remained at the same unit for the entire period. The criterion for bank units’ consolidation was geographic location. My identifying assumption is that the decision to close a bank unit is orthogonal to firms’ characteristics. I find support for that assumption in the data by testing for differences in observable characteristics. First, I compare the loan terms and performance, and firm financial variables of borrowers in closed and open units in the pre- unit closure period and I find no statistically significant differences. Second, I provide out-of-sample evidence that the identifying assumption holds by using the Amadeus Bureau van Dijk dataset to test for differences based on the zip codes of bank units. I apply a difference-in-difference methodology to estimate the effect of relationship interruption. I define as treated the firms with exogenously discontinued personal relationships and as control those with continuing relationships. Hard information passed from one loan officer to another as the transfer happened within the same bank. Observed differences between the two groups in the post-consolidation period should be driven by the consequences of interrupted relationships. 4Several banks in Italy, Spain, Portugal have consolidated their network during the recent crisis as well as banks in the United States. ECB Working Paper Series No 2553 / May 2021 4 I find strong evidence for the significant effect of personal relationships on loan renegotiations at both extensive and intensive margins. Loans to firms with interrupted relationships have a 13.4% lower probability to be renegotiated upon their transfer to another unit, compared to loans to firms that re- mained at the same unit. The unconditional probability of renegotiating a loan is 59% and the estimated effect corresponds to a 8% lower probability of renegotiation. Moreover, conditional on renegotiating a loan, affected firms received tougher loan terms on their renegotiated loans. The affected loan terms for firms whose accounts were transferred to another unit include higher interest rates, approximately 170% shorter maturities, and requirements that these firms pledge collateral with 65% higher value compared to firms that remained with their original bank unit for the entire period. The economic magnitudes of the impact on loan maturity and collateral are significant as they correspond, on average, to approximately two and a half years shorter maturity and an additional e 0.78 of collateral for each euro of loan amount. The results remain robust when I exploit a within-firm variation and apply a difference-in-difference-in- differences methodology. This confirms that the effect of the interrupted relationships is not driven by any firm characteristics. I also observe that firms alter their capital structure after their relationship with the bank is inter- rupted. The change in capital structure indicates that firms cannot substitute lending from other banks without cost when the relationship with one bank is exogenously interrupted. In particular, we observe that, when the relationship with one bank is interrupted, firms raise more equity and decrease leverage. Firms only partially substitute loans from other banks to make up for the borrowing reduction from the bank whose relationship was severed. This change in a firm’s sources of financing is likely to have important implications for the firm’s business model and investments. A possible selection bias on renegotiation could change the interpretation of the results for loan term differences between interrupted- and continuing-relationship loans. If the loan officer chooses to renegotiate with firms based on their performance, the intensive margin results will be biased. To address this concern, I conduct two tests. First, I compare the pre-unit-closure period characteristics of the treated and control firms that renegotiated a loan in the post-unit-closure period and I find no statistically significant difference. Second, I consider all the hard information available to both the loan officer and the econometrician to identify the variables that can trigger a renegotiation. This test demonstrates that the loan officer who remained at the same unit for the entire period chose to renegotiate with treated firms with higher profitability prospects. Loan officer behavior, therefore, biases estimated coefficients towards zero, implying that my intensive margin results are conservative. ECB Working Paper Series No 2553 / May 2021 5 To further investigate the explanatory mechanism for the value of a relationship between a loan officer and a firm, I examine closer and more distant relationships separately. In most cases, a closer relationship entails fewer outside financing options, and when a closer relationship is interrupted, it causes a significant informational loss. In contrast, firms with more distant relationships and, hence, greater outside financing options are expected to have stronger bargaining power. To test for the value of a stronger relationship, I construct a measure of relationship strength. This analysis confirms the value of relationships, as a firm with a stronger previous loan-officer relationship displays a significant negative effect on its renegotiated loan terms when this relationship ends. This paper’s main conclusion is that personal relationships mitigate the cost of distress for the firm in a loan renegotiation. The firm is worse off following the interruption of its loan officer relationship, as it is less able to renegotiate, and receives tougher loan terms on renegotiated loans. Moreover, an interruption of the firm-loan officer relationship causes the firm to alter its capital structure and its sources of financing. These results may be driven either by a loss of valuable soft information or by unwarranted favoritism. From the results on loan performance we observe that firms with continuing relationships have similar performance with firms with interrupted relationships. Moreover, a few heterogeneity tests indicate that the impact of an interrupted loan officer relationship on the probability of renegotiating a loan and on the renegotiated loan’s terms is stronger for firms with good repayment histories, high leverage, and positive EBITDA growth rate. These findings therefore support the hypothesis that lending relationships between a loan officer and a firm help to alleviate debt-overhang through the acquisition of information. In the next section, I describe the paper’s contributions to the literature. In section 3, I provide an overview of the institutional background and the dataset structure. In section 4, I present the empirical specification and a detailed comparison of the treated and control groups. The regression results are presented in section 5. A discussion about the underlying mechanism is included in section 6. In section 7, I conclude. 2 Related Literature This paper contributes to the literature by being the first to combine two classic research streams— relationship banking and contract renegotiation—and by estimating the consequences of personal rela- tionships between loan officers and firms for loan renegotiations. Analysing this effect, this paper is the ECB Working Paper Series No 2553 / May 2021 6 first to provide empirical evidence that relationships between a loan officer and a firm help to alleviate debt-overhang. The broader literature to which this study contributes addresses the role of relationships in credit markets. A rich theoretical literature on bank debt highlights the importance of informational asymmetry and moral hazard for financial intermediation.5 Several empirical papers have examined the relationship between banks and borrowers for evidence to determine whether asymmetric information affects lending. Petersen and Rajan (1994) and Berger and Udell (1995) used the repeated interaction between a borrower and a financial institution as a measure of relationship. Mian (2006), Agarwal and Hauswald (2010), Canales and Nanda (2012), and Bolton, Freixas, Gambacorta, and Mistrulli (2016) define relationship lending as a function of geographic distance. Ongena and Smith (2001) analyze the duration of a bank relationship with a firm. Sufi (2007) and Ivashina (2009) explore informational asymmetries in a lending syndicate.6 The empirical literature to which this paper is most closely related identifies the effects of personal relationships between bank employees and borrowers. These studies focus on how loan approvals or performance can be influenced by different factors such as cultural proximity (Fisman, Paravisini, and Vig, 2017), social connections (Haselmann, Schoenherr, and Vig, 2018), hierarchical and geographical distance (Liberti and Mian, 2009), or the loan officer being on leave (Drexler and Schoar, 2014). Several papers examine the effect on interest rates of strong interpersonal connections between a banker and a firm (Engelberg, Gao, and Parsons, 2012), or the effect of a strong relationship as measured by the number of interactions (Herpfer, Working Paper), or the effect of an interruption of a relationship caused by an executive’s death or retirement (Karolyi, 2018). Lastly, Hertzberg, Liberti, and Paravisini (2010) and Cole, Kanz, and Klapper (2015) provide insights on the effects of bank-specific governance policies on the moral hazard behavior of a loan officer. Within the field of literature on relationship banking, this paper is the first to demonstrate how ex- ogenous interruptions in bank-borrower relationships affect loan renegotiation. One important difference between new and renegotiated loans is the bank’s prior exposure to the risk of default on pre-existing loans. The risk of a higher probability of default is magnified in a crisis period, such as in Greece in 2010-2015, when banks faced high delinquency ratios that drove their overall risk assessment. In such 5Jaffee and Russell (1976); Stiglitz and Weiss (1981); Myers and Majluf (1984); Diamond(1984,1991); Ramakrishnan and Thakor (1984);Sharpe (1990); Besanko and Kanatas (1993); Rajan and Winton (1995);Bolton and Freixas (2000) 6Extensive surveys of this literature are provided by Ongena and Smith (2000), Boot (2000), Srinivasan (2014) ECB Working Paper Series No 2553 / May 2021 7 periods, the value of a successful renegotiation surges.7 Moreover, by analysing the impact of lending relationships in loan renegotiations, this paper is the first to provide empirical evidence that relationships help to alleviate debt-overhang. Further, whereas existing literature focuses on estimating relationship effects on lending either at the extensive margin or on the interest rate only, this paper examines the impact on the probability of renegotiation as well as on the three main variables characterizing the loan structure (i.e., interest rate, maturity, and collateral). Given that the main focus of this paper is loan renegotiations, the analysis also contributes to the literature on contract renegotiation. Several influential papers have examined renegotiation in incomplete contracts as an out-of-equilibrium phenomenon.8 Although the existing theory of contact renegotiation has evolved significantly, the empirical evidence on this topic is limited due to data restrictions. This paper investigates renegotiation independent of previous defaults on a loan payment, and for that reason is also related to the work of Roberts and Sufi (2009), who analyze in detail the factors triggering a renegotiation, and those determining its outcomes. Roberts (2015) is the closest to my paper, as he shows that a corporate loan renegotiation happens frequently, modifies significantly the initial loan terms, and is affected significantly by the duration of the lending relationship. Although similar conclusions to those found in these two papers arise in my analysis regarding the frequency and the outcomes of a renegotiation, my paper advances beyond these to estimate the effect of the relationship between a loan officer and a borrower on the probability of renegotiation and the nature of its outcomes. A few empirical papers have considered other factors affecting loan renegotiation. The importance of the liquidation value of collateral (Benmelech and Bergman, 2008) of the mortgage securitization (Piskorski, Seru, and Vig, 2010), and of policy intervention (Agarwal, Amromin, Ben-David, Chom- sisengphet, Piskorski, and Seru, 2017) on loan renegotiation are highlighted. James (1995) focuses on debt restructurings and shows that the financial condition of the firm determines the bank’s incentives to make concessions. Lastly, Chodorow-Reich and Falato (Working Paper) show the importance of the financial institution’s health on contracting credit using covenant violations. This paper also complements literature examining the impact of bank branch consolidation. In this area, the most relevant papers are by Nguyen (2019), as she examines how branch closures in the United States affect local access to credit, and by Bonfim, Nogueira, and Ongena (Working Paper), as they show that branch closures in Portugal cause an increase in the interest rate that firms receive. Lastly, this paper 7Karolyi (2018) highlights also the importance of lending relationships during recessions. 8Hart and Moore(1988 1998, Rajan (1992), Aghion et al. (1994), Von Thadden (1995), Bolton and Scharfstein (1996), Maskin and Moore (1999), Gorton and Kahn (2000), Garleanu and Zwiebel (2008), Tirole (2010) ECB Working Paper Series No 2553 / May 2021 8 is linked to the strain of European debt crisis literature that employs micro–level data to identify the impact on bank lending.9 3 Institutional Background & Data 3.1 Institutional Background The identification setting and the data come from a Greek bank and cover the period 2012–2015. Several important facts characterize the economy and the banking sector of that period. 10 From 2008 and until the end of 2016, the Greek GDP contracted by approximately 25%; unemployment rose to approximately 26%; and investment declined by 75%. The collapse in investment was partially caused by a decreased access to credit. Access to finance was the most pressing concern for small- and medium-size enterprises (SMEs) operating in Greece, as 33% of SME owners consider this their most important problem.11 This issue is critical given that SMEs account for more than 90% of private companies and 87% of total employment. The Greek banking sector suffered during this period from a lack of access to international capital markets, deposit flight, and losses from the sovereign debt restructuring. Several banks were resolved and their deposits as well as a number of their loan portfolios were transferred to the four largest banks, thus causing a significant centralization of the sector. The four largest banks went through three large-scale recapitalizations (July 2013, May 2014, and December 2015) during this period. However, availability of long-term finance remained limited, and cost of credit was very high compared to EU standards. One of the main reasons for the limited credit supply was the deterioration of banking asset quality. In 2016, the nonperforming loans (NPLs) reached 45% of the loan portfolio, and provisions stood at 50% of total NPLs. In particular, in the corporate sector, about 60% of loans to SMEs were nonperforming. The rise of the NPLs ties up bank capital, thereby reducing profitability and increasing funding costs. In this paper, I exploit a major internal reorganization that was implemented in one of Greece’s largest banks, and led to the closure of bank units. It is important to clarify what a bank unit is and distinguish bank units from branch closures. In this case, a bank unit entails a center that manages corporate loans, 9Acharya, Eisert, Eufinger, and Hirsch (2018), Bentolila, Jansen, and Jim´enez (2017), Cingano, Manaresi, and Sette (2016), Popov and Van Horen (2015), De Marco (2019) 10Gourinchas, Philippon, and Vayanos (2017) provide a detailed empirical and theoretical analysis of the Greek crisis. 11OECD, 2016 Financing SMEs Report ECB Working Paper Series No 2553 / May 2021 9 and more centralized centers manage specific loan types. For the analysis that follows, a bank unit can be considered a type of branch, because, as in distinct branches, personal relationships develop between loan officers and the firm whose loans they manage. The main goal of this internal reorganization was increased efficiency regarding NPL management. In response to the significant rise in the NPL ratio for corporate loans, the bank established specialized NPL workout units at the end of 2013. The new units were separate from the units responsible for loan origination and were responsible for monitoring, managing, or liquidating the nonperforming exposures. Approximately half of total loans to small- and medium-size corporations were transferred to these spe- cialized NPL units. The criterion that was used to decide which loans to transfer to the NPL units is very strict and is related with the performance of the firm. In particular, if all the loans that a firm had at the subject bank in 2013 were nonperforming, the management of this firm’s loans is transferred to the NPL unit. Otherwise, if at least one of the firm’s loans had no delays on payments in 2013, the management of all the firm’s loans remained at the bank unit, irrespective of having a nonperforming loan. This strict rule can be seen at the data as if the ratio of a firm’s nonperforming loans per total number of loans is equal to one, the probability to be transferred to the NPL unit is one, while if the ratio is smaller than one, the probability is zero. This transfer of loans to the NPL units caused a significantly reduced workload for the original units assigned to manage corporate loans. Consequently, it became cost effective for the bank to consolidate the original units, by closing several and relocating the loans’ management to the closest unit that re- mained open. Originally there were 112 units that managed corporate loans, and after the mergers, there were 37. The bank intended to retain its network across the country, ensuring clients would remain. For that reason, the main criterion for mergers was geographic location, and mergers would only take place in areas served by two or more units. The unit remaining open in a given area, would be the unit man- aging the largest volume of loans. An important feature of the consolidation was the relocation of the loan officers. Loan officers who had worked in units that closed were transferred to the newly established NPL units. Loan officers who worked in units that remained open continued to manage their old loans and became additionally responsible for the loans transferred from closed units. Because firms whose loans were transferred to the closest unit lost the relation with the loan officer who had managed their loans, the feature of unit mergers, provides a good setting for testing the effect of personal relationships on loan renegotiation. ECB Working Paper Series No 2553 / May 2021 10 3.2 Dataset structure and descriptive statistics The main data used in this study come from one of the largest commercial banks in Greece. The dataset contains detailed annual information on corporate loans for approximately 8,000 small and medium nonretail enterprises (SME) covering four years (2012–2015). The construction of the sample is based on the ECB supervision guidelines for the Asset Quality Review (AQR). Under these guidelines, an SME is defined as a corporation that has annual turnover up to e 50 million and employs fewer than 250 persons. For several reasons, a restricted sample of nonretail SMEs is the most appropriate sample for the proposed analysis. First, it is necessary to exclude large corporations as they have access to other sources of financing, such as international banks and the stock market. Moreover, credit for large corporations is approved at higher level at the bank, and for that reason, relationships with loan officers are expected to be irrelevant. Second, the retail sector is also excluded from the dataset, as a different department at the bank manages this sector. By focusing on nonretail SMEs, I gained access to credible firm financial information, since the majority of the firms in the sample have audited financial statements, which they are required to submit to the bank. Moreover, by excluding very small firms, we can expect sample firms to be unaffected by the narrow local economic environment. Firms in the sample operate either regionally or nationally and their performance is expected to be affected by the economic conditions at the region and industry level. The dataset includes detailed information on the loan terms and performance as well as basic firm financial information. Each firm has multiple loans at this bank, and the loan types vary from the more secure, such as leasing, to the less secure, such as factoring, letters of credit, and revolving credit. More- over, an indicator of the bank unit responsible for each loan is included, which allows for tracking transfers across units. Personal relationships develop between loan officers and firms at the bank unit and changes in the bank unit indicator reflect interruptions of such relationships. As I focus on the merger of the original bank units, I exclude the loans that were transferred to the specialized NPL units. For that reason, the sample included is not representative of the Greek economy during this period, but rather represents the set of firms that performed relatively well during the crisis. The final sample consists of loans to 3,984 firms located across the 9 geographic regions of the country. Following the merger, a single bank unit managed on average 340 loans to 107 firms. Table 1 presents the summary statistics for the main variables in the pre-period (2012, 2013). A median firm in the sample ECB Working Paper Series No 2553 / May 2021 11 has two loans with a total balance of approximately e 536,000 and total collateral cover of e 187,630. The average interest rate is 5.97%, and the average remaining maturity of the loans is approximately one and a half years (528.34 days). The median firm performs well, with no nonperforming exposures, and has not delayed a payment. Regarding the firm’s financial information, the median firm was medium-size with approximately e 6 million in total assets and e 4.25 million in total debt. It has a positive EBITDA of approximately e 270,000, and a high leverage equal to 0.69. The summary statistics confirm the fact that the sample is comprised of firms that performed relatively well during the crisis. 4 Empirical Methodology 4.1 Identification Strategy The identification of relationship interruption is based on a bank’s internal reorganization and the closure of bank units. I employ this exogenous variation, define appropriate treatment and control groups, and apply the difference-in-difference methodology to accurately estimate the effect of interrupted relation- ships between loan officers and firms on loan renegotiation. The first step is to exclude loans that were transferred to the specialized NPL units. The sample is thus constrained to relatively good performing firms. The next step is to identify the firms whose accounts were transferred to another unit because their original unit closed. By using the closure of the original bank unit as the source of exogenous variation, I overcome the selection bias that may arise at the firm level. In particular, treated firms are defined as those whose loan accounts were transferred to another bank unit because their original unit closed. Control firms are defined as those whose loan accounts were managed at a bank unit that remained open during the whole period of the sample. This specification of treatment and control groups ensures that the variation comes only from the bank unit level and not from the firm level. I need to clarify that there is a set of firms whose accounts were transferred to other bank units without their original unit closing. Even though the transfer for these firms is driven by endogenous reasons, I include them in the control group because otherwise the control group would be biased. 12 The estimated coefficients present the intent to treat effect of the exogenous interruption of the relationship between a loan officer and a firm. 12In previous versions of the paper I had excluded this set of observations from the analysis. The results remain robust under both specifications. The estimated results excluding these firms are available upon request. ECB Working Paper Series No 2553 / May 2021 12 The feature that allows me to identify an interruption in relationships between loan officers and firms is the relocation of loan officers. Loan officers who worked in units that closed were transferred to the new NPL units. Loan officers who worked in the units that remained open continued to manage their old loans and became responsible for the loans that were transferred from the closed units. For that reason, firms that were monitored by a unit that remained open would continue to interact with the same loan officers, while firms whose accounts were exogenously transferred had to establish a new relationship with a loan officer. The baseline specification is a difference-in-difference, which allows me to compare the difference on the outcome variables between loans to firms in the treated group and those in the control group in the post-unit closure period (2014, 2015) relative to the difference that the two groups had in the pre-unit closure period (2012, 2013). The baseline regression is: yi jurt = α + αj + αpre−u + αpost−u + αrt + δ(Postt ∗ Treatijur) + εijurt (1) where yijurt stands for the outcome variable for firm j obtaining a loan i from bank unit u and located in region r in year t. Treati jur is a dummy variable equal to one for treated firms and zero for control firms. Postt is a dummy variable equal to zero for the period before the bank units’ closure (2012 – 2013) and one after the closures (2014 – 2015). The baseline specification includes firm fixed effects (α j) to capture any time-invariant firm characteristics. Both pre-period bank unit (αpre−u) and post-period bank- unit (αpost−u) fixed effects are included to capture any time-invariant characteristics of the original bank unit and of the bank unit that the loan was transferred in the post-period (e.g., different lending limits). Region-year fixed effects (αrt) capture any region and time-varying shocks. The coefficient of interest is δ, which measures the difference in the outcome variable for the firms that experienced an interruption in their relationship with the loan officers, relative to the firms that did not, controlling for the pre-period difference. The outcome variables of interest belong to three groups: (i) the probability of renegotiating a loan (extensive margin); (ii) renegotiated loan terms (intensive margin); and (iii), firm level effects. Regarding the loan terms of the renegotiated loans, I use as outcome variables the interest rate, the remaining maturity of the loan, and the collateral value. I also construct two additional variables to capture the effect of collateral. The first is an unsecured loan dummy and the second is a type-of-collateral dummy. To capture firm outcomes from an interrupted relationship, I examine the firm’s equity over total assets, ECB Working Paper Series No 2553 / May 2021 13 total debt over total assets, EBITDA over total assets, and the firm’s total loan balance at the subject bank over its total debt. 4.2 Comparison of treated and control groups in the pre-unit closure period The most important threat to identification is a possible selection bias that arises from the decision to close a bank unit. It is necessary to assume that the bank did not close units where debtors performed, or were expected to perform, worse. The main criterion for unit closure is geographic location: in areas where there were two or more units, the bank kept only one, while in areas with only one unit, it was optimal for the bank to keep it open and retain its clients. A comparison between the two groups in the pre-unit-closure period (2012–2013) provides evidence that there are no statistically significant differences across the two groups. Tables 2 – 3 present this comparison and includes variables related to loan terms, performance, and firms’ financials. Table 2 shows the comparison of all the observations in the sample, including those that were transferred to the NPL units, while in table 3 only the treated and control firms are included. In both tables, Column 1 shows the mean value and the standard deviation for firms in bank units that remained open, Column 2 for firms in bank units that closed, and Column 3 shows the p-value for the difference, with regional fixed effects included. The only variable significantly different across the two groups is the ratio of loans transferred to the NPL units from the originals: this variable is higher for the bank units that remained open. This difference indicates that, to the extent that units were selected for closure based on loan performance, the bank units that closed had better performing loans. For that reason, if a selection bias exists on which units closed, it will bias the results downwards. The fact that we do not observe any statistically significant difference for the loan terms, performance, and firms’ financial information supports the assumption that the two groups shared similar characteristics. The probability is therefore high that the two groups would have continued to look similar if the personal relationships with loan officers had not been interrupted. I report also parallel trend graphs in Figures 1 and 2 for the nonperforming dummy, as higher default ratios could predict the bank-unit closure. These graphs provide evidence against this hypothesis. Figure 1a plots the mean values for the treated and the control groups, and Figure 1b plots the mean values of the residuals from a regression of the nonperforming dummy on region fixed effects. Figure 2 plots the regression coefficients and confidence intervals from a regression of the nonperforming dummy on ECB Working Paper Series No 2553 / May 2021 14