I. Introduction
It’s not hard to make decisions when you know what your values are. (Roy E. Disney)
Credit decision-making in banks typically follows a hierarchical structure. At the lower level, bankers process applications, carry out interviews and visit sites, collect and compile hard and soft information, and provide recommendations; at the higher level, loan officers or committees issue final decisions based on this input. Although prior studies document personal biases throughout this hierarchy (e.g., Cortés, Duchin, and Sosyura (Reference Cortés, Duchin and Sosyura2016), Beck, Behr, and Madestam (Reference Beck, Behr and Madestam2018)), we know little about the role of personal values and, in particular, how higher-level bankers adjust for the biases of those below them. This article addresses these gaps by studying how bankers’ biospheric values interact with firms’ environmental credentials in the credit approval process.
The setting examined in this article provides an ideal environment for evaluating the role of personal values in the hierarchical decision process. We analyze comprehensive credit data from a mid-sized Chinese bank, which is linked with responses from a value survey involving its personnel. The bank’s loan evaluation process is uniquely suited to identifying how applicants’ environmental orientations interact with bankers’ personal preferences. Applications are randomly assigned to a frontline banker, called a customer manager, who gathers hard and soft information and prepares a report that concludes with an overall recommendation—a holistic, discretionary assessment. Each report is then randomly forwarded to a loan officer for review and final judgment.Footnote 1 All firm-related information, including the environmental and safety scores set by the customer manager, is thus exogenous to the loan officer. Neither banker knows the other’s identity.
We measure bankers’ biospheric values using a proprietary survey that records their agreement with four statements related to respect for the Earth, unity with nature, environmental protection, and pollution prevention. Although average values are relatively high, there is substantial variation across bankers. We classify bankers with high values as “green” and those with low values as “brown.” This heterogeneity is strongly associated with how customer managers assess firms’ environmental attributes when making loan recommendations. Controlling for a broad set of soft and hard firm information, the environmental score is the most economically important predictor of overall recommendations for customer managers in the greenest quintile, whereas it is statistically insignificant in the brownest quintile. In the full sample, a 1-standard-deviation increase in the environmental score corresponds to a 0.26-standard-deviation increase in the overall recommendation, placing it among the most economically significant inputs.
Our analysis shows that brown loan officers tend to downgrade green applications with strong environmental scores from customer managers, whereas green officers tend to discount brown applications that receive low environmental scores. Discretion is therefore used both to counter and to reinforce green bias, depending on the officer’s values. These discretionary adjustments are concentrated in borderline cases and arise in settings where customer managers appear overoptimistic, an outcome associated with weaker loan performance. When officers intervene, their actions align with personal values, consistent with targeted corrections to perceived misratings rather than indiscriminate activism.
The economic magnitude of these value‑driven interventions is nontrivial. For applicants with mid‑range baseline approval prospects, a 1-standard-deviation increase in the environmental score is associated with an approval probability roughly 15 percentage points higher when evaluated by a green loan officer, but about 5 percentage points lower when evaluated by a brown officer—a 21‑percentage‑point spread. Assuming denied applicants largely turn to higher‑cost alternative finance, this gap corresponds to an estimated 1.7% higher cost of debt for greener firms reviewed by brown officers. By contrast, gender‑based interactions between bankers and borrowers are economically and statistically weak, indicating that green bias is both more salient and more costly than gender bias in this context.
Taken together, our findings paint a nuanced picture. Hierarchies can balance bias: superiors have both the incentive and the discretion to offset subordinates’ value‑driven leniency. Yet because superiors bring their own values to the task, discretion simultaneously opens the door to new biases and potential misallocation. Recognizing which type of bias dominates, therefore, requires understanding the personal values of each layer in the hierarchy.
Is the interrelationship between biospheric values and the environmental score unique in its association with customer manager overall recommendations and loan outcomes? We show evidence it is. This applies to both components of the interrelationship, as the interaction variable loses much or all its explanatory power when one of them is varied. For example, substituting the variation in customer managers’ biospheric values with variation in their gender, level of education, age, or experience (which all have low correlations with biospheric values) renders the interaction variable unable to explain overall recommendations. Similarly, replacing the environmental score with another soft or hard variable eliminates or significantly weakens the explanatory power of the interaction. These findings suggest that there is something distinctive about the relationship between biospheric values and the environmental score—as the hypothesis of the significance of personal values would lead us to expect.
This research focuses on small firms, which are significant in their own right due to their important role in bank finance (e.g., Berger and Udell (Reference Berger and Udell1998)). Although bankers may approach loans to small and large firms differently—for example, because small firms produce less hard data (Cole, Goldberg, and White (Reference Cole, Goldberg and White2004))—the underlying economic and value-based factors guiding their decisions are likely similar. Thus, the findings from this study may also hold relevance for larger firms.
We also anticipate that our findings extend beyond the specific institutional setting examined. The biospheric values of the bankers in our sample align with those found in previous studies on U.S. and Chinese respondents (Bouman, Steg, and Zawadzki (Reference Bouman, Steg and Zawadzki2020), Wang, van der Werff, Bouman, Harder, and Steg (Reference Wang, Van der Werff, Bouman, Harder and Steg2021)). Both the overall recommendation and loan granting decision regressions include a comprehensive set of hard and soft variables widely recognized as important in credit decisions across various contexts. The high R 2 values of approximately 85% for both regressions suggest that these variables effectively capture the key determinants of loan recommendations and approvals, leaving little room for behaviors unique to our institutional setting. This supports the view that our findings extend to other banking environments, especially where lending to Small and Medium-sized Enterprises (SMEs) involves substantial discretion. The effects may be less pronounced in jurisdictions with strict green-lending mandates or mature Environmental, Social, and Governance (ESG) markets that limit value-driven discretion.
Our article aims to contribute to three different areas of literature. First, our article shows that loan officers often respond to positive recommendations on green applications with rejection decisions that “lean against the wind.” This pattern highlights the considerable discretionary authority of superiors in the banking hierarchy and their role in moderating subordinate decisions to contain bias. The interaction between different levels within the hierarchy ensures a comprehensive review process, essential for maintaining balanced credit assessments (Liberti and Mian (Reference Liberti and Mian2009)). This hierarchical dynamic can complement strategies such as rotating loan officers to avoid overly positive evaluations due to familiarity with clients (Hertzberg, Liberti, and Paravisini (Reference Hertzberg, Liberti and Paravisini2010)).
Second, besides other salient firm observables, such as comprehensive external or internal firm credit ratings (Cole (Reference Cole1998)), firm-bank distance (Agarwal and Hauswald (Reference Agarwal and Hauswald2010)), bank prospecting incentives (Cole, Kanz, and Klapper (Reference Cole, Kanz and Klapper2015)) or funding (Brown, Kirschenmann, and Ongena (Reference Brown, Kirschenmann and Ongena2014)), and/or monetary conditions (Jiménez, Ongena, Peydró, and Saurina (Reference Jiménez, Ongena, Peydró and Saurina2014)), our article documents that the green orientation of firms can be a potent novel factor determining their loan application success. This finding aligns with the increasing attention given to the green orientation of firm projects and their financing by banks (Dursun-de Neef, Ongena, and Tsankova (Reference Dursun-de Neef, Ongena and Tsankova2023)) and markets (Pástor, Stambaugh, and Taylor (Reference Pástor, Stambaugh and Taylor2021)).
Third and finally, we document that credit recommendations vary systematically with customer managers’ green preferences. This novel finding contributes to papers arguing that bankers’ gender (Beck et al. (Reference Beck, Behr and Madestam2018)), religion (Baele, Farooq, and Ongena (Reference Baele, Farooq and Ongena2014)), emotional state (Cortés et al. (Reference Cortés, Duchin and Sosyura2016)), attention and familiarity (Campbell, Loumioti, and Wittenberg-Moerman (Reference Campbell, Loumioti and Wittenberg-Moerman2019)), and/or commonality in ethnicity (Fisman, Paravisini, and Vig (Reference Fisman, Paravisini and Vig2017)), social connections (Haselmann, Schoenherr, and Vig (Reference Haselmann, Schoenherr and Vig2018)), and in general loan-specific preferences (Herpfer (Reference Herpfer2021)) may matter for credit outcomes. This finding also complements prior literature suggesting that personal values can shape the decision making of inspectors and other “street bureaucrats” (Lipsky (Reference Lipsky1980)), investors (Riedl and Smeets (Reference Riedl and Smeets2017)), corporate executives (di Giuli and Kostovetsky (Reference Di Giuli and Kostovetsky2014)), judges (Harris and Sen (Reference Harris and Sen2019)), and politicians (Washington (Reference Washington2008)).
The remainder of the article is organized as follows: Section II describes the institutional setting; Section III presents the data; Section IV reports the empirical results; and Section V concludes.
II. Institutional Setup
A. The Bank
The data for this study were obtained from a Chinese state-owned commercial bank that operates exclusively within its home province. The bank has over 30 branches, and it has a total asset value of over 100 billion RMB (equivalent to approximately 14 billion USD). Business loans account for roughly 90% of its loan portfolio. At the time of initiating the study, the bank employed around 1,600 individuals (with approximately 1,050 working full-time). Of these employees, there are roughly 1,100 frontline employees, with an additional 230 serving as customer managers and another 80 working as loan officers. Each branch provides retail services, with an average of five customer managers per branch.
B. The Granting of Loans
The loan approval process involves three parties: the applicant, the customer manager, and the loan officer. The interactions between them occur over five stages (as illustrated in Figure A1 in the Supplementary Material).
Firstly, a client—typically a small-medium business owner—expresses interest in obtaining a business loan. Front-desk personnel usually advise the client to submit an initial application that includes information such as the requested amount, collateral offered, purpose of the loan, and basic financial details about their business. If an initial evaluation suggests that the business meets the bank’s requirements for a loan, staff will provide customers with a formal loan application form and list of required supporting documents.
The bank collects various pieces of information from applicants including demographic data on owners or managers; type of business; primary industry sector; years of operation; desired loan amount; intended use of funds; and two most recent audited financial statements among other things. All loan applications are centralized in the bank’s head office system. The head office assigns a loan application to the nearest branch to the applicant’s (business’s) address, and then randomly assigns the application to a customer manager in that branch, subject to the current workload of the customer manager allowing her to accept more work. Reassignments of loan applications after the initial allocation are extremely rare, occurring only in exceptional cases such as long-term leave or staff turnover.
Secondly, once received by a customer manager at one of these branches, each application undergoes assessment to determine creditworthiness and probability of repayment. The collected data must be comprehensive, realistic, and specific to ensure that the application meets the bank’s loan-process requirements. To verify the authenticity of the submitted documents, customer managers conduct on-site visits to the businesses of over 70% of loan applicants. These visits, which include interviews to understand the loan’s purpose, associated risks, and business potential, often involve examining the business premises, inventory, assets, and overall financial status.
Thirdly, the customer manager prepares a due diligence report based on the investigation and submits it, together with the applicant’s supporting materials, to headquarters for approval. The report typically summarizes the applicant’s condition—covering its business, assets, debt, and expected repayment sources—and provides the customer manager’s assessment of the application. This assessment includes the following: i) the truthfulness of the stated loan purpose, ii) the applicant’s repayment ability, iii) the applicant’s willingness to repay, iv) creditworthiness, v) the environmental score, vi) the safety score, and vii) an overall recommendation. The overall recommendation reflects the manager’s judgment formed from both hard information in the file and soft information collected during interviews. It is not mechanically linked to the bank’s internal credit-risk score, which managers do not observe. Managers also cannot revise their soft-information scores after submission.
Fourthly, while the customer manager is responsible for preparing the application materials and writing a due diligence report, the final approval decisions are made by loan officers at the bank’s headquarters. After the due diligence report and supporting documents are submitted, the loan applications are assigned to loan officers at random by the bank’s central dispatcher algorithm. According to the bank, loan officers have no influence over the assignment process, and the assignment algorithm does not take loan officers’ characteristics into consideration. The only exception is the loan officer’s current workload, which may be taken into account to avoid overburdening her with too many applications at once.
Finally, the loan officer assigned to the application accesses the applicant’s information and reviews the due diligence report to determine whether to approve the application and to set the appropriate interest rate, loan amount, and loan term. In cases where the loan officer finds gaps or unclear information in the loan application documents or the due diligence report, they may request the customer manager to perform further investigations and obtain additional materials from the loan applicant. The loan application review process is carried out under full anonymity, with customer managers unaware of which loan officers will assess their applications and loan officers unaware of which customer managers prepared them. Loan officers exercise discretion over the evaluation, approval, and pricing of loans, with approximately 10% of the credit rating weight assigned to their discretionary judgment. The average duration of the loan process is about 10 business days.
C. Assessing the Greenness of Bankers and Loan Applicants
1. Green Credit Guidelines
China’s green credit guidelines, initiated in 2012 by the China Banking Regulatory Commission (CBRC), aim to encourage banks and financial institutions to lend to green projects and firms. These guidelines principally require banks to: i) issue more loans and provide better terms for green initiatives and ii) introduce incentives to enhance green lending. For example, meeting certain green credit targets may result in regulatory benefits like lower reserve requirements.
Our collaborating bank places importance on fulfilling its green credit obligations and has formed a dedicated team to supervise its green credit guidelines and processes. Green credit is characterized as loans awarded to projects and enterprises that contribute to green causes and comply with certain environmental metrics, such as pollution or emissions reduction and resource conservation. The procedure for green loans is distinct from that for standard loans. Customers initially specify if they want a green loan, and the bank assesses whether the loan’s purpose fits the green criteria. Loans falling outside these criteria go through the regular evaluation system.
Green loans in this bank are generally larger than SME loans and follow a distinct origination and approval process, often involving public programs and committee review. By contrast, all loan applications in our sample passed through the standard approval channel and were not explicitly labeled as green loans. This feature helps isolate the association between loan officers’ biospheric values and their decisions from any formal green lending mandate. The bank’s designated green loan portfolio exhibits lower nonperformance rates than conventional loans, suggesting that officers may be aware from internal reporting that green projects tend to perform well. This reduces the likelihood that brown officers’ reluctance toward green SMEs reflects information-based concerns and is instead more consistent with value-driven behavior. Although our sampling strategy limits the direct link to explicit green policies, it provides a cleaner setting for identifying how personal values relate to credit decisions.
2. Bankers’ Incentives
The pay for customer managers and loan officers consists of a fixed salary plus bonuses. Customer managers earn a bonus tied to the approved loan amount of each applicant they refer. Moreover, their evaluations consider other metrics, including customer satisfaction and loan performance. Although there are no immediate consequences for inaccurate assessments of soft metrics such as willingness to repay or environmental risk, such evaluations may affect customer managers’ long-term career prospects, particularly during promotions. Loan officers get bonuses based on both the default rate of the loans they approve and the total loan volume. There is no mandated approval rate for loan officers, and the bank sets no cap on the loan amounts they can approve.
Loan officers rank above customer managers in the organizational hierarchy. Performance-based promotions can elevate customer managers to loan officers, but the reverse is highly unlikely. The hiring of customer managers emphasizes sales capabilities over an understanding of credit risk or policy intricacies. In contrast, loan officers are primarily selected based on their depth of understanding of credit risk and bank lending policies.
The bank does not set direct, individual-level performance targets or bonuses for loan officers specifically tied to green lending. While environmental and social risk considerations are emphasized in training and compliance, they are framed as part of general credit risk assessment rather than green-loan–specific sales targets. Consequently, institutional incentives are unlikely to systematically bias loan officers either toward or against environmentally friendly borrowers.
III. Data
Our data set merges survey data on bankers with data related to the loan applications they oversee. Below, we detail the methods used to collect the survey data, examine its representativeness, and describe the criteria for selecting firms in the sample. We also discuss the bank’s approach to assessing the environmental risk of loan applicants. Definitions for all variables used in the study are provided in the Appendix.
A. Survey on Bankers
In October 2020, the bank conducted a survey on its full-time employees through its internal communication system, with a response rate of 85%. Most relevant for our purposes, the survey assessed the biospheric values of the bankers, which were measured using a set of statements and scales developed by de Groot and Steg (Reference de Groot and Steg2008) and Bouman, Steg, and Kiers (Reference Bouman, Steg and Kiers2018). The biospheric values were measured based on the customer manager’s or loan officer’s attitude toward four statements, namely, respecting the Earth, unity with nature, protecting the environment, and preventing pollution. Scores ranged from −1 to 7, where −1 represented strong disagreement and 7 indicated strong agreement. The average score of the attitudes toward the four statements is used to measure the greenness of the bankers.
B. Bankers’ Training
To ensure its employees are equipped with the necessary business skills and stay up to date with the latest industry legislation and regulations, the bank conducts training sessions several times a year. These sessions range from several hours to several days, and most of them are mandatory for the employees. Most relevant for our study purposes, between December 2020 and January 2021, a 3-h business-skills training was organized by the bank that included half an hour dedicated to discussing environmental friendliness in evaluating loan applications. During this session, the instructor presented the distribution of the biospheric values of the bank employees that were collected in the pretraining survey. This information allowed bankers to compare their biospheric values with those of their colleagues. The training was well-attended, with 91% of the bank’s customer managers and 94% of its loan officers present.
C. Prospective Loan Applicants’ Training
In April 2021, the bank made changes to its business loan due diligence report by adding environmental and safety scores. Bankers were instructed to assess these scores at the firm level, independent of any specific project the loan applicant might have sought funding for. The revised report was applied on an experimental basis to 2,994 potential loan applicants. These applicants were invited to a bank-sponsored training that lasted for 150 minutes between April and July 2021. (More details about these trainings are provided in Appendix A1 of the Supplementary Material.) Out of the 2,994 firms, 2,147 attended the training and 1,436 applied for a loan between April 2021 and February 2022. For this experiment, the bank revised its randomization algorithm to allocate loans randomly to customer managers and loan officers who responded to the survey. Our sample consists of these loan applications, with Appendix A2 of the Supplementary Material providing an empirical test whose results are consistent with the bank’s assertion of random assignment of the customer managers and loan officers to loan applications.
D. Assessing Environmental Risk
Given that the sample firms are small, they likely lack sufficient data for the bank to accurately assess their environmental risk using hard data alone. Therefore, the bank allows customer managers to assess this risk holistically, similar to how they evaluate other soft variables. Customer managers observe the industry of the loan applicants, which significantly influences environmental performance; for instance, firms in construction and industrial sectors often have direct environmental impacts from pollutants like dust, chemicals, and wastewater. Additionally, customer managers usually conduct site visits, giving them the opportunity to observe the firms’ environmental practices firsthand. Lastly, they interview the loan applicants, often owner-managers, who have a significant impact on the firm’s commitment to environmental measures (Roxas and Coetzer (Reference Roxas and Coetzer2012)). Even though banks might assess larger firms using different methods, this approach provides a solid foundation for making informed decisions about environmental risk.
IV. Results
A. How Green Is the Firm?
Table 1 presents descriptive statistics on loan applicants and loan outcomes, with Panel A providing information on the loan applicants themselves. The mean age of a loan applicant is 8 years, with mean total assets valued at 1.4 million RMB (about 200,000 USD). Applicant firms are notably profitable, with an average net profit ratio of 22%, although their small size limits absolute profits to an average of 92,000 RMB (about 13,000 USD). Additional information about the distribution of the loan applicants’ industry and company registration type can be found in Panels A and B of Table A1 in the Supplementary Material, respectively.
TABLE 1 Long description
Panel A. Descriptive Statistics on Hard Variables for Firms (N = 1,436):
* Firm age: Mean 7.68, Median 7, Std. Dev. 2.23, Min 4, Max 15.
* Total assets: Mean 1,405, Median 632, Std. Dev. 2,250, Min 200, Max 23,900.
* Current ratio: Mean 2.44, Median 1.53, Std. Dev. 2.39, Min 0.51, Max 10.35.
* Debt-assets ratio: Mean 0.53, Median 0.55, Std. Dev. 0.26, Min 0.08, Max 0.94.
* Net profit ratio: Mean 0.22, Median 0.18, Std. Dev. 0.17, Min 0.03, Max 0.67.
* Sales growth: Mean 0.17, Median 0.10, Std. Dev. 0.32, Min minus 0.31, Max 1.13.
Panel B. Descriptive Statistics on Soft Variables and Loan Outcomes (N = 1,436 for scores, N = 831 for loan details):
* Truthfulness: Mean 2.47, Median 3, Std. Dev. 0.75.
* Ability to repay: Mean 2.53, Median 3, Std. Dev. 1.02.
* Willingness to repay: Mean 2.38, Median 3, Std. Dev. 0.84.
* Environmental score: Mean 4.19, Median 5, Std. Dev. 0.99.
* Safety score: Mean 3.74, Median 4, Std. Dev. 1.00.
* Overall recommendation: Mean 3.86, Median 5, Std. Dev. 1.38.
* Loan granted: Mean 0.58, Median 1, Std. Dev. 0.49.
* Approved loan amount: Mean 357,581, Median 280,000, Std. Dev. 233,840.
* Collateral: Mean 0.20, Median 0, Std. Dev. 0.40.
* Annual effective interest rate: Mean 0.059, Median 0.057, Std. Dev. 0.005.
* Maturity: Mean 1.73, Median 2, Std. Dev. 0.44.
* Greater than or equal to 90 days overdue: Mean 0.036, Median 0, Std. Dev. 0.19.
Panel C. Distributions of Scores (Overall Recommendation, Environmental, and Safety):
* Score 1 (low): Recommendation 6.1 percent, Environmental 0.7 percent, Safety 0.1 percent.
* Score 2: Recommendation 17.8 percent, Environmental 7.8 percent, Safety 17.1 percent.
* Score 3: Recommendation 13.6 percent, Environmental 13.6 percent, Safety 14.3 percent.
* Score 4: Recommendation 9.7 percent, Environmental 27.5 percent, Safety 45.4 percent.
* Score 5 (high): Recommendation 52.9 percent, Environmental 50.4 percent, Safety 23.1 percent.
Panel B of Table 1 reports on loan outcomes and soft variables: approximately 58% of all applicants are granted loans for an average maturity of 1.7 years, while the average effective interest rate stands at 5.9% per annum. Short-term loans make appraising a firm’s environmental risk more straightforward, as climate-related risks are unlikely to materialize in such a limited timeframe. In all, 3.6% of the loans were at least 90 days overdue. The mean environment score is reported as being 4.19 out of 5, while safety scores come in slightly lower at an average rating of 3.74 out of 5; overall recommendation scores have a mean value of 3.86 out of 5.
The distribution across these scores can be seen in Panel C of Table 1 where more than half receive top marks for environmental friendliness and overall recommendations. Panel C of Table A1 in the Supplementary Material shows that industries prone to pollution, such as manufacturing and construction, have the lowest environmental scores, while less pollutive industries like wholesale and retail have the highest scores. Panel D of Table A1 in the Supplementary Material provides additional details regarding truthfulness and willingness to repay variables, whereas Panel E of Table A1 in the Supplementary Material focuses solely on the ability-to-repay variable.
As outlined in the Appendix, the scales used to measure each variable differ from one another but are generally easy to interpret. However, it should be noted that three specific variables require particular attention: truthfulness, environmental score, and safety score. The lowest category for each variable is defined as “Can’t judge.” To investigate whether bankers interpret “Can’t judge” statements as indicating lower scores (rather than being a manifestation of confusion that could be interpreted as either low or high scores), we cross-tabulate the three variables with the overall recommendation score and the loan approval indicator. The results are presented in Panels A–F of Table A2 in the Supplementary Material and show that all three variables are monotonically associated with the two outcome variables. This suggests that bankers likely interpret “Can’t judge” statements as corresponding to lower scores.
B. How Green Is Your Banker?
The first seven rows of Table 2, Panels A and B, outline the traits of 202 customer managers and 64 loan officers, respectively. At the study’s outset, these individuals were, on average, aged 34–35 and had about 5 years of experience working at the bank. More than two-thirds of each group are men, and 75% hold a bachelor’s degree.
TABLE 2 Long description
The table is divided into two sections: Panel A for Customer Managers (N = 202) and Panel B for Loan Officers (N = 64). The columns include Mean, Median, Std. Dev., Skew., Min, Max, and N.
Panel A. Customer Managers:
* Age: Mean 34.02, Median 33, Std. Dev. 5.69, Skew. 0.65, Min 20, Max 55.
* Experience: Mean 5.18, Median 5, Std. Dev. 2.22, Skew. 0.12, Min 1, Max 10.
* Biospheric values: Mean 4.91, Median 5.25, Std. Dev. 1.66, Skew. minus 1.30, Min minus 1, Max 7.
* Female dummy: Mean 0.27.
* High school or equiv.: Mean 0.18.
* Bachelor’s: Mean 0.75.
* Master’s or higher: Mean 0.07.
* Std. dev. of env. score: Mean 0.99, Median 0.96, Std. Dev. 0.61, Skew. 1.03, Min 0, Max 4.
Panel B. Loan Officers:
* Age: Mean 34.73, Median 35, Std. Dev. 7.16, Skew. 0.25, Min 21, Max 53.
* Experience: Mean 5.33, Median 5, Std. Dev. 2.71, Skew. minus 0.04, Min 1, Max 10.
* Biospheric values: Mean 5.46, Median 5.75, Std. Dev. 1.36, Skew. minus 1.40, Min 0, Max 7.
* Female dummy: Mean 0.31.
* High school or equiv.: Mean 0.13.
* Bachelor’s: Mean 0.83.
* Master’s or higher: Mean 0.05.
The primary metric for evaluating bankers’ values is their biospheric-values score. The mean biospheric values for customer managers and loan officers are 4.91 (with a maximum of 7) and 5.46, respectively. A score of 3 indicates that respondents consider the values to be “important,” while a score of 6 indicates that they view them as “very important.” These scores appear to reflect the biospheric values held by all bankers within the institution regardless of their department or position. Table A3 in the Supplementary Material reports that bankers who do not handle loan applications have a qualitatively similar mean biospheric-values score of 5.36, with little variation observed across departments or job positions. This figure is comparable to the mean scores of 5.13 and 4.79 reported in prior studies for U.S. and Chinese respondents, respectively (Bouman et al. (Reference Bouman, Steg and Zawadzki2020), Wang et al. (Reference Wang, Van der Werff, Bouman, Harder and Steg2021)). Both customer managers’ and loan officers’ biospheric values are negatively skewed (skewness between –1.3 and –1.4). In other words, a minority of people in both groups have environmental values that differ significantly from those of the majority (by being low). We will investigate later how the behavior of this minority group differs from that of the rest.
Row 8 in Table 2, Panel A, indicates the standard deviation of individual bankers’ environmental scores. This mean standard deviation of 0.99 is the same as the standard deviation computed from the aggregate sample presented in Table 1, Panel B.
C. Correlations
Panel A of Table 3 presents the correlations between the loan granting decision and firm soft information variables. Traditional soft variables exhibit high correlation with one another as well as with both the overall recommendation and loan granting decision, with correlations ranging from 0.77 to 0.88. On the other hand, environmental and safety risks display much smaller (though statistically highly significant) correlations with these two factors, ranging between 0.20 and 0.42.
TABLE 3 Long description
The table is divided into four panels.
Panel A. Correlations between Loan Grant Decision and Firm-Level Soft Information Variables (N = 1,436).
* Overall recommendation and Loan Granted: 0.88.
* Truthfulness and Loan Granted: 0.81; and Overall Recommendation: 0.80.
* Ability to repay and Loan Granted: 0.77; Overall Recommendation: 0.79; Truthfulness: 0.63.
* Willingness to repay and Loan Granted: 0.85; Overall Recommendation: 0.85; Truthfulness: 0.84; Ability to repay: 0.76.
* Environmental score and Loan Granted: 0.27; Overall Recommendation: 0.42; Truthfulness: 0.18; Ability to repay: 0.16; Willingness to repay: 0.16.
* Safety score and Loan Granted: 0.12; Overall Recommendation: 0.24; Truthfulness: 0.15; Ability to repay: 0.15; Willingness to repay: 0.15; Environmental score: 0.21.
Panel B. Correlations between Customer Managers’ Biospheric Values and Their Overall Recommendation Decisions (N = 202).
* Overall recommendation and Biospheric Values: minus 0.06.
* Environmental score and Biospheric Values: 0.04; Overall Recommendation: 0.41.
* Standard deviation of environmental score and Biospheric Values: 0.03; Overall Recommendation: minus 0.19; Environmental score: minus 0.53.
* Correlation of environmental score and overall recommendation and Biospheric Values: 0.25; Overall Recommendation: minus 0.06; Environmental score: 0.12; Standard deviation of environmental score: 0.10.
Panel C. Correlations between Loan Officers’ Biospheric Values and Their Loan Grant Decisions (N = 64).
* Loan granted and Biospheric Values: minus 0.13.
* Environmental score and Biospheric Values: 0.12; Loan Granted: minus 0.08.
* Correlation of environmental score and loan granted and Biospheric Values: 0.28; Loan Granted: 0.00; Environmental score: minus 0.03.
Panel D. Correlations between Customer Managers’ Biospheric Values and Their Other Personal Traits (N = 202).
* Age and Biospheric Values: minus 0.07.
* Female and Biospheric Values: minus 0.02; Age: minus 0.14.
* Education level and Biospheric Values: 0.06; Age: 0.05; Female: minus 0.13.
* Experience and Biospheric Values: 0.04; Age: 0.35; Female: minus 0.03; Education level: 0.01.
In Table 3, Panel B, we report on the correlations between customer managers’ biospheric values and their evaluations of the loan applicants. Biospheric values have a negligible correlation with both the average environmental score (r = 0.04; t = 0.54) and the standard deviation of the environmental score (r = 0.03; t = 0.37). These insignificant correlations are consistent with our underlying assumption that biospheric values do not influence how customer managers set environmental scores. The relationship between biospheric values and overall loan recommendations (r = −0.06) also lacks statistical significance.
The last row of Table 3, Panel B, reports on a layered correlation: specifically, how customer managers’ biospheric values correlate with the correlation between environmental scores and other soft variables. The 0.25 correlation between biospheric values and the relationship between environmental scores and overall recommendations is statistically significant at the 0.1% level. This finding suggests that for green managers, the relationship between environmental scores and overall recommendations is more pronounced than for brown managers. Our subsequent tests will build on this interaction.
Table A4 in the Supplementary Material demonstrates that the relationship between biospheric values and overall recommendations stands out among correlations with soft variables set by customer managers, which either lack statistical significance or are negative. This observation undermines the idea that green managers would also amplify other soft variables for environmentally friendly firms.
The environmental and other soft scores are defined by customer managers, which means that they are beyond the loan officers’ control. This setting provides a neat framework for investigating the relationship between environmental scores and loan officers’ credit approval decisions, as documented in Table 3, Panel C. We find no significant correlation between loan officers’ biospheric values and proportion of approved loans (r = −0.13; t = −1.07), nor between biospheric values and the (exogenously determined) average environmental scores (r = 0.12; t = 0.96). Importantly, the correlation between loan officers’ biospheric values and the secondary correlation between environmental scores and loan approval decisions is statistically significant (r = 0.28) at the 5% level. This finding aligns with that of customer managers, suggesting that green loan officers demonstrate a more marked positive relationship between environmental scores and loan approvals compared to their brown counterparts.
Panel D of Table 3 provides correlations between customer managers’ biospheric values and their gender, education, age, and experience. None of the correlations with biospheric values are statistically significant at conventional levels. Table A5 in the Supplementary Material details the same correlations for loan officers, where only education shows a statistically significant correlation.
D. Explaining Loan Applications’ Approval
1. The Role of the Firm’s Green Orientation for Its Loan Application Success
To understand which factors contribute to customer managers’ evaluations of loan applications, Table 4 presents results from linear probability model regressions of the overall recommendation on several predictors. The analysis begins with Panel A, detailing the coefficients and their corresponding t-values.
TABLE 4 Long description
The table is divided into two main sections: Panel A for Coefficients and t-values, and Panel B for Standardized Coefficients. The dependent variable is the Customer Manager’s Overall Recommendation.
Panel A. Coefficients and t-Values:
* Firm age: Values range from 0.006 to -0.010 across specifications 1 to 5, with t-values between 0.52 and -1.27.
* ln (total assets): Values range from 0.045 to -0.008, with t-values between 1.51 and -0.43.
* Current ratio: Shows significant negative coefficients from -0.039 to -0.025 (t-values -2.65 to -2.58).
* Debt-assets ratio: Highly significant negative coefficients from -3.730 to -0.339 (t-values -27.29 to -2.76).
* Net profit ratio: Coefficients from 0.211 to -0.103.
* Sales growth: Significant at 0.227 in specification 1, decreasing to 0.019 in specification 5.
* Truthfulness: Significant positive coefficients around 0.46 to 0.53 in specifications 2 to 5.
* Ability to repay: Significant positive coefficients from 0.332 to 0.367 in specifications 2 to 5.
* Will. to repay: Significant positive coefficients from 0.590 to 0.617 in specifications 2 to 5.
* Safety score: Significant positive coefficients from 0.090 to 0.115 in specifications 2 to 5.
* Env. score: Significant at 0.365 in specification 3.
* Env. score categorical (2 to 5): Significant positive coefficients increasing from approximately 0.98 to 1.97 in specifications 4 and 5.
* Cust. manager F E: ‘No’ for specifications 1 to 4, ‘Yes’ for specification 5.
* N: 1,436 for all specifications.
* R-squared: Increases from 0.432 in specification 1 to 0.876 in specification 5.
Panel B. Standardized Coefficients:
* This panel mirrors the variables in Panel A, showing standardized values. For example, the Debt-assets ratio standardized coefficient is -0.704 in specification 1, and the Env. score = 5 standardized coefficient reaches 0.713 in specification 5.
Specification 1 includes only hard firm variables as independent variables, with both debt-assets ratio and current ratio showing a significant negative association with the overall recommendation. The coefficient for the debt-assets ratio is as expected, while that for the current ratio is not.Footnote 2
Specification 2 adds traditional soft risk measures and safety score as independent variables. As expected, coefficients for traditional soft risk measures (truthfulness and reasonableness, ability to repay, and willingness to repay) are positive and have t-values greater than 8. Safety score also shows high significance with a t-value of 5.43. Soft variables appear to capture some of the explanatory power of hard variables, particularly the debt-assets ratio; at the same time, R 2 increases considerably, from 0.43 to 0.79.
Specification 3 includes the environmental score variable as a continuous variable in the regression equation, which commands the highest t-value at 14.84. With the exception of willingness to repay, the coefficients for all other variables decrease slightly from specification 2. However, the R 2 of the model continues to increase considerably, from 0.789 to 0.848.
Specification 4 models environmental score using dummy variables which show monotonic association with overall recommendation scores without any substantial effects on regression coefficients or R 2 values when customer manager fixed effects are added into regression equation. Since customer managers are branch-based and remained in the same roles throughout the observation period, these fixed effects effectively capture any branch-level variation. Collectively, these results suggest that customer managers’ values or traits have relatively little predictive power on the overall level of the recommendations.
Panel B of Table 4 displays standardized coefficients to show the economic significance and the relative importance of variables from Panel A. Assuming the model is correctly specified, standardized coefficient sizes indicate which variables are relatively more important (Darlington (Reference Darlington1968)). Although the exact ingredients and weights of the bank’s credit model have not been revealed to us, the inclusion of all ex ante well-justified variables that the bank collects data on and the high R 2 of the model suggest that we may not be too far off from the bank’s own credit model.
In specification 3, which considers all variables and treats the environmental score as a continuous measure, an increase of 1-standard-deviation in the environmental score corresponds to a 0.26-standard-deviation increase in the overall recommendation. This ranks the environmental score as the second most important variable, only surpassed by the willingness to repay, which holds a standardized coefficient of 0.37. In specifications 4 and 5, where the environmental score is represented by categorical variables, the category representing the highest score (5) is associated with about a 0.7-standard-deviation increase in the overall recommendation compared to the base score (1).
How does the coefficient for the environmental score depend on other variables in the model? Table A6 in the Supplementary Material shows an alternative regression starting with the environmental score variable alone (specification 1). The coefficient, 0.579, remains nearly unchanged when controlling for industry in specification 2. It decreases by about 20% in specification 3 when other hard variables are included and by about 37% in specification 4 (corresponding to specification 3 in Table 4) when all other soft variables are added. The t-value for the environmental score exceeds 14 in all specifications. Thus, while the coefficient decreases with additional controls, it remains quantitatively important and highly significant.
2. The Role of the Customer Managers’ Green Preferences
Table 5 examines how the interaction between customer managers’ biospheric values and the environmental score is associated with their overall recommendation. Panel A replicates the results of Table 4 specification 3 for each quintile of customer managers’ biospheric values. Our focus here is on any differences in coefficients for a given variable between high- and low-biospheric-values customer managers. The p-value of a χ 2 test that compares these coefficients appears in the rightmost column.
TABLE 5 Long description
Panel A presents coefficients and t-values by biospheric-values quintile. The dependent variable is the customer manager's overall recommendation. Key variables include Firm age, ln total assets, Current ratio, Debt assets ratio, Net profit ratio, Sales growth, Truthfulness, Ability to repay, Will. to repay, Safety score, and Env. score. Notably, the Env. score coefficient increases from 0.049 in the Low quintile to 0.539 in the High quintile, with a p-value of 0.000 for the difference between Low and High. Industry and Firm type F E are included. Sample sizes N range from 214 to 353 across quintiles, with R-squared values between 0.845 and 0.904.
Panel B lists standardized coefficients for the same variables across the five quintiles. The standardized Env. score coefficient shows a clear upward trend from 0.036 in the Low quintile to 0.386 in the High quintile.
Panel C details a pooled regression interacting biospheric-values quintile dummies with the environmental score across two specifications. Specification 1 includes quintile dummies and their interactions with Env. score, while Specification 2 includes customer manager F E. The interaction terms show a progressive increase in the impact of the Env. score on recommendations as biospheric values increase: from 0.065 for quintile 1 to 0.528 for quintile 5 in specification 1. Both specifications use a sample size N of 1,436 with R-squared values of 0.862 and 0.884 respectively.
Across all specifications, traditional soft risk measures are highly significant positive predictors, but this cannot be said for environmental scores. Specifically, we observe that the environmental score increases monotonically with increasing biospheric-values quintile: it has a coefficient of 0.049 (t = 1.14) in the bottom quintile and a coefficient of 0.539 (t = 12.17) in the top quintile; this difference is statistically significant at the 0.1% level. These findings, mirroring the layered correlation between biospheric values and the relationship between environmental scores and overall recommendations in Table 3, Panel B, suggest that environmental risk means different things to low- versus high-biospheric-values customer managers. Furthermore, the coefficients for sales growth and the safety score are significantly larger for the top quintile at the 5% level.
Panel B presents standardized coefficients by biospheric-values quintiles. We find that the environmental score is particularly significant in the top-three quintiles, with a standardized coefficient of 0.386 in the top-biospheric-values quintile, which is higher than any other variable. In contrast, it has little importance in the bottom quintile and falls fourth largest (after willingness to repay, truthfulness and reasonableness, and ability to repay) in the second-lowest quintile.
Panel C of Table 5 presents results from a pooled regression, as opposed to splitting the sample into quintiles. Specification 1 replicates Table 4 specification 3 with the following adjustments: the demeaned environmental score is interacted with customer manager’s biospheric-values quintile, and indicators for the biospheric-values quintiles are added to the regression.
The interaction variable in specification 1 of Panel C increases monotonically in biospheric-values quintile, from 0.065 (t = 1.75) to 0.528 (t = 12.37). This finding is consistent with the results in Panels A and B, indicating that environmental risk matters more for customer managers who hold higher biospheric values. The interaction effects remain qualitatively similar in specification 2, which adds customer manager fixed effects to the regression. In specification 1, the main effect for the lowest-biospheric-values quintile is significantly positive (t = 3.44).Footnote 3 This means that brown customer managers tend to be more lenient in their evaluations of firms on average. This finding is consistent with the slightly negative unconditional correlation between biospheric values and overall recommendations reported in Table 3, Panel B.
Table A7 in the Supplementary Material relaxes the assumption of equal distances between consecutive levels of overall recommendations by performing the analysis of Table 5, Panel C, using ordered logit. The results are qualitatively similar to those in Table 5, Panel C.
3. Loan Officers’ Loan Granting Decisions
What explains the loan granting decisions of loan officers? Table 6 evaluates this by presenting results of linear probability model regressions of the loan granted indicator on several predictors. The table follows a similar design to that used in studying overall recommendations made by customer managers (Table 4), with one key difference: we replace the overall recommendation variable with the loan granted indicator while also including overall recommendation as an additional explanatory variable. An important aspect of this setting is that the information on which loan officers base their decisions is provided by customer managers, making it exogenous to the decision-maker.
TABLE 6 Long description
A table titled Modeling Loan Granting Decisions with the dependent variable Loan Granted Dummy. It contains four specifications labeled 1 through 4.
Independent variables and their coefficients (with t-statistics in parentheses) include:
* Firm age: Values near 0.000 across all models, statistically insignificant.
* ln (total assets): Values between -0.002 and 0.001, statistically insignificant.
* Current ratio: Consistently negative at -0.008 or -0.007, significant at the 5 percent level.
* Debt-assets ratio: Strongly negative around -0.27, significant at the 1 percent level.
* Net profit ratio: Negative values between -0.032 and -0.045, insignificant.
* Sales growth: Positive coefficients between 0.047 and 0.056, significant at the 5 percent or 1 percent level.
* Overall recomm.: Positive coefficients around 0.14, highly significant at the 1 percent level.
* Truthfulness: Positive coefficients around 0.12, highly significant at the 1 percent level.
* Ability to repay: Positive coefficients around 0.058, highly significant at the 1 percent level.
* Will. to repay: Positive coefficients around 0.11, highly significant at the 1 percent level.
* Safety score: Positive coefficients around 0.015, significant at the 5 percent level.
* Env. score: Included in specification 2 (0.006) and as categorical dummies in 3 and 4, all statistically insignificant.
Bottom rows indicate:
* Loan officer F E: No for specifications 1, 2, and 3; Yes for specification 4.
* N: 1,436 for all models.
* R-squared: Increases from 0.841 in specification 1 to 0.850 in specification 4.
We find that the overall recommendation is overwhelmingly significant in predicting accepted loan applications, with t-values exceeding 15 across all specifications. The bank reports that a predetermined model, which remained constant during the sample period, accounts for approximately 90% of the loan application assessment process, while the remaining 10% is left to the discretion of loan officers. Since we lack specific information on the bank’s model, we must speculate on the variables and estimate the parameters based on the data. Our in-sample R 2 stands at around 84%, which is not far from the 90% included in the bank’s model. Incorporating loan officer fixed effects into the model has little impact on regression coefficients or R 2.
We are particularly interested in the environmental score as an explanatory variable, which we omit in specification 1, add as a continuous variable in specification 2, and split into indicators in specifications 3 and 4. However, none of the environmental score variables are found to be significant in any specification. This suggests that on average, there is no association between environmental scores and loan granting decisions.
There are two possible explanations for this outcome. First, it could be that customer managers’ overall recommendations already take into account the environmental score; therefore, loan officers do not need to consider this factor further. However, this explanation seems unlikely given that traditional soft variables, safety score, debt-assets ratio, and current ratio remain statistically significant across all specifications with consistent signs (as observed in Table 4). These findings suggest that these variables continue to predict loan acceptance even after accounting for their relationship with overall recommendation. The second potential explanation is that some loan officers may behave in ways that neutralize any association between environmental risk and the dependent variable. We will explore this possibility more closely in our subsequent analysis.
4. The Role of the Loan Officers’ Biospheric Values
Table 7 investigates how the interaction between loan officers’ biospheric values and the environmental score is associated with their loan granting decision. Our analysis follows the structure of Table 5, which explores the determinants of overall recommendations as a function of customer managers’ biospheric values.
TABLE 7 Long description
Panel A presents coefficients and t-values by biospheric-values quintile from Low to High. Key variables include Firm age, ln total assets, Current ratio, Debt assets ratio, Net profit ratio, Sales growth, Overall recommendation, Truthfulness, Ability to repay, Willingness to repay, Safety score, and Environmental score. Notably, the Environmental score coefficient is negative -0.040 for the Low quintile and positive 0.083 for the High quintile, with a p-value for Low equals High of 0.000. Sample sizes N range from 218 to 354 across quintiles, with R-squared values between 0.822 and 0.901.
Panel B models decisions with biospheric-values interactions across two specifications. It includes interaction terms for each biospheric-values quintile multiplied by the Environmental score. The interaction for Quintile 1 times Environmental score is -0.035 in specification 1 and -0.034 in specification 2. The interaction for Quintile 5 times Environmental score is 0.064 in specification 1 and 0.063 in specification 2. Both specifications include firm controls and other soft variables, while specification 2 adds loan officer F E. Total N is 1,436 with R-squared values of 0.847 and 0.852.
Panel A of Table 7 presents the results of regression specification 2 from Table 6, separately for each biospheric-values quintile of loan officers. The coefficient for the environmental score is monotonically associated with these values. In particular, among top-biospheric-values quintile loan officers, the environmental score takes a highly significant positive coefficient (t = 4.52), indicating that green loan officers are more likely to approve loans to companies with high environmental scores. Conversely, among bottom-two biospheric-values quantiles, the environmental score takes a statistically significantly negative coefficient (t-values –2.71 and –1.97 for the bottom and second-to-bottom quintiles, respectively), and the difference between the top and bottom quintile is significant at the 0.1% level.
These findings, which echo the positive layered correlation between biospheric values and the relation between environmental scores and loan approvals in Table 3, Panel C, suggest that for low-biospheric-values loan officers, a higher environmental score is associated with a decrease in the likelihood of loan approval, holding other factors constant. This is consistent with these loan officers expecting the customer manager (who likely has higher biospheric values than the brown loan officer) to have allowed the environmental risk to influence the overall recommendation, which is a significant predictor of loan approval. By allowing the environmental score to have a negative partial correlation with the loan granting decision, the loan officers can potentially offset the effect of what they perceive as too environmentally friendly customer managers.
To summarize, the results suggest that the environmental score appears to have a different meaning for low- and high-biospheric-values loan officers. Some loan officers associate a positive weight to the environmental score, while others assign a negative weight. These opposite forces offset each other on average, which explains the nonsignificant coefficient for the environmental score in Table 6.
Table A8 in the Supplementary Material presents standardized coefficients, which suggest that for loan officers belonging to the high-biospheric-values group, the environmental score is the third-most important variable after overall recommendation and willingness to repay. However, for loan officers in the bottom quintile, the environmental score is the fifth-most important variable and has an opposite sign compared with that in the top quintile.
Panel B of Table 7 reports results from a pooled regression instead of splitting the sample into quintiles. Specification 1 replicates Table 6 specification 2 but interacts demeaned environmental scores with loan officer’s biospheric-values quintile while adding indicators for biospheric-values quintiles in regression.
We find that the main effects for the biospheric indicators for the three lowest quintiles are positive and significant, at least at the 10% level. This indicates that brown loan officers tend to be more lenient judges of firms than green loan officers. The coefficient for the interaction variable increases monotonically in biospheric-values quintile. The coefficient is highly significantly negative (t = −3.26) for the bottom quintile and highly significantly positive for the top quintile (t = 3.90). The negative coefficient for the bottom quintile is consistent with the result in Table 7, Panel A, suggesting that brown loan officers take a more critical view of loan applications with higher environmental scores, all other things being equal. These findings remain qualitatively unchanged in specification 2, which introduces loan officer fixed effects to the model, as well as in Table A9 in the Supplementary Material, which estimates the regression using a logit framework.
5. Optimality of Bias Balancing
Forms of Loan Officer Discretion
Green bias, exercised in the form of loan officer discretion, can manifest in four different ways: i) brown loan officers may deny credit to worthy green applicants; ii) brown loan officers may grant credit to unworthy brown applicants; iii) green loan officers may grant credit to unworthy green applicants; and iv) green loan officers may deny credit to worthy brown applicants. To analyze the creditworthiness of loan applicants, we estimate predicted approval using specification 2 in Table 6. We then examine actual loan approval decisions as a function of predicted approval deciles and loan officer’s biospheric values.
Table A10 in the Supplementary Material indicates that loan applications fall into four categories: none in deciles 1 through 3 were approved; three in decile 4 were approved; all but 33 in deciles 5 through 9 were approved; and all in decile 10 were approved. In effect, applications with very low predicted approval rates are always rejected, those with very high predicted rates are always approved, and those in the two middle categories may be either approved or rejected.
To better understand the role and consequences of loan officer discretion, we further examine the middle categories. They differ in a key respect: we can observe the performance of the three approved applications in decile 4, but we lack outcome data for the 33 rejected applications in deciles 5 through 9.
Customer Manager Overoptimism and Loan Officer Discretion
We next analyze whether the discretion exercised by loan officers depends on the perceived overoptimism of customer managers. Customer managers’ overoptimism is measured using the residual overall recommendation from specification 3 in Table 4, Panel A. By construction, this yields an approximately even split between overoptimistic and underoptimistic loan applications.
Panel C of Table A10 in the Supplementary Material shows that 29 of the 33 rejected applications in deciles 5 through 9 exhibit overoptimism, while only 4 exhibit underoptimism. A binomial test decisively rejects the null hypothesis of equal probability between the two types (p < 0.01%). This pattern suggests that loan officers, regardless of their biospheric values, are much more cautious toward applications they perceive as overly endorsed by customer managers while rarely rejecting other high-quality applications. A mirror result holds for the three approved applications in decile 4: all exhibit underoptimism. In these cases, loan officers appear to have exercised discretion to approve applications that were harshly assessed by customer managers.
To assess whether loan officers act in the bank’s best interest when exercising discretion, we examine whether overoptimism predicts loan underperformance. The evidence supports this view. Panel D of Table A10 in the Supplementary Material shows that overoptimism is positively correlated with loans overdue by at least 90 days (r = 0.115, significant at the 0.1% level) and with loans overdue by less than 90 days (r = 0.071, significant at the 5% level). These relationships suggest that correcting overoptimism serves the bank’s best interest and can improve credit allocation. In our sample, nearly 90% of discretionary rejections involve applications displaying overoptimism, indicating that such interventions are generally efficiency-enhancing. However, the small number of discretionary cases limits the precision of this estimate.
Loan Officer Discretion and Green Bias
Loan officers’ biospheric values relate to two dimensions: the extent to which they exercise discretion and the direction in which they apply it. Along the first dimension, officers at both extremes of the biospheric values scale exercise more discretion than those in the middle. For example, the greenest quintile rejected 8% of applications in deciles 5 through 9, while the brownest quintile and the second-greenest quintile each rejected 5%. The remaining two quintiles rejected only 1%–2% of applications. This pattern is mirrored in the R2 values reported in Table 7, Panel A, which follow a similar distribution across officer quintiles.
The second dimension deals with the loans in which loan officers choose to use their discretion. Brown loan officers rejected seven loans in deciles 5 or higher. Of these, six had an environmental score of 5, and one had a score of 3. All six loans with score 5 exhibited customer manager overoptimism, indicated by positive residual recommendations. In contrast, the loan with score 3 showed underoptimism. These findings suggest that brown officers exercise discretion not only toward green firms in general but particularly toward those that appear overly favored by customer managers.
The greenest quintile of officers presents a similar pattern, but in the opposite direction. These officers rejected 15 loans in deciles 5 or higher. Among these, eight had an environmental score of 2, four had a score of 3, and one had a score of 1. The environmental scores of loans rejected by brown and green officers differ significantly (t = 5.13), suggesting that discretionary rejections are guided by opposing environmental considerations.
Unlike other loan officers, those in the two greenest quintiles also exercised discretion by accepting loans in predicted acceptance decile 4. These loans had high environmental scores—two scored 5 and one scored 4. In retrospect, this discretion did not yield favorable outcomes, as all three loans became overdue, with one delayed by 90 days and the others by shorter durations.
Interpretation and Relation to Literature
Lipsky (Reference Lipsky1980) offers a valuable framework for understanding loan officer behavior, especially the use of discretion in environments characterized by limited resources and policy ambiguity. While he does not discuss the work of loan officers per se, his insights on frontline bureaucrats—such as inspectors and police—are directly applicable. Lipsky emphasizes that discretion is an inherent feature of street-level work. Inspectors, constrained by time and information, often resort to simplifications and heuristics that reflect their own values, prior experiences, and expectations. These cognitive shortcuts may embed subjective judgment into what appear to be routine procedural decisions.
Lipsky also describes how bureaucrats develop coping strategies to manage high workloads and uncertainty. One such strategy is routinization, where officials prioritize easily observable signals or investigate more deeply only when clear “red flags” arise. While efficient, this approach may introduce systematic bias. Furthermore, Lipsky notes that personal beliefs frequently influence enforcement decisions, with frontline workers justifying leniency or severity based on their own views of fairness or intent. This framework suggests the potential for both statistical and taste-based discrimination, as judgments may be shaped by assumptions about the characteristics of applicants. Limited oversight further reinforces the role of individual discretion, often leaving value-driven decisions unchecked.
Our results indicate that loan officers typically follow routine procedures when granting loans, spending an average of 1.5 h to review a 10-page small business application. Given this workload, it is reasonable for them to rely on straightforward decision rules unless a specific factor prompts closer examination of an application. Overoptimism by the customer manager—which is on average associated with loan underperformance—serves as one such red flag. When loan officers respond, their discretionary efforts reflect their personal values. While such efforts may improve overall credit allocation, the outcome depends on the individual characteristics of the loan officer exercising discretion.
6. Economic Significance
Cost of Capital
We examine the economic significance of green bias by linking environmental scores and loan officers’ orientations to loan pricing and, especially, approval decisions. We find that interest rates display limited association with environmental scores, and this pattern is similar for green and brown officers. Approval probabilities, however, differ across officer types. Table A12 in the Supplementary Material shows that among applicants with midrange predicted approval rates, a 1-standard-deviation higher environmental score is associated with approval probabilities that are roughly 15 percentage points higher under green officers and about 5 percentage points lower under brown officers, resulting in a 21-percentage-point difference.
To illustrate the potential economic implications of this difference, we use a simple refinancing framework in which rejected firms are assumed to rely on higher-cost external funding. Within this setup, the 21-percentage-point difference in approval probabilities aligns with an expected difference of approximately 1.7% in debt financing costs for greener firms when evaluated by brown rather than green officers. This figure is indicative and relies on auxiliary assumptions. Further details are reported in Appendix A3 of the Supplementary Material.
Gender Bias Versus Green Bias
How important is green bias relative to other distortions in bank lending? Our data include information on both the gender of bankers and the gender of firms’ main owners, allowing a direct comparison between green bias and gender bias (e.g., Beck et al. (Reference Beck, Behr and Madestam2018)) within our sample. Female-owned firms account for 25% of the sample. Table A13 in the Supplementary Material shows that mean overall recommendations, loan acceptance rates, interest rates, and default rates are similar for male- and female-owned firms.
Specification 3 in Table A13 in the Supplementary Material replicates specification 1 from Table 7, Panel B, but adds results for gender. The findings for biospheric values closely mirror those in specification 1 of Table 7, Panel B, which are repeated for reference in specification 2 of Table A13 in the Supplementary Material. The interaction term for female owners has the expected sign but is not statistically significant at conventional levels (t = 0.80). An analogous regression for customer managers, reported in Table A14 in the Supplementary Material, yields qualitatively similar results. These findings suggest that gender bias plays a less important role in our sample.
7. Taste-Based Versus Statistical Discrimination
Our analysis up to this point has offered an explanation that suggests green managers and loan officers provide higher ratings and credit recommendations to green firms based on their personal biospheric values. However, as suggested by Bénabou and Tirole (Reference Bénabou and Tirole2011) and our correlation evidence on bankers’ green traits, it is also possible that these bankers believe that green firms are better customers to the bank and therefore deserve higher ratings or credit recommendations.
Does our evidence on the values-based channel align more closely with taste-based or statistical discrimination? As Levitt ((Reference Levitt2004), p. 433) observes, “in general, empirical tests have a difficult time distinguishing between taste-based and information-based models of discrimination.” In our setting, however, identification is aided by detailed measures of bankers’ values and beliefs. These measures, described in Appendix A4 of the Supplementary Material and summarized in Panel A of Table A15 in the Supplementary Material, originate from a follow-up survey administered in September 2023. The survey targeted staff who had served as customer managers, loan officers, or both since 2015, and response rates were high: 94% of customer managers and 89% of loan officers from the 2020 survey participated. Its primary purpose was to complement the earlier survey by gathering information on environmental beliefs and self-reported environmental knowledge while also introducing new questions that capture preferences aligned with biospheric values (referred to here as “green values”). Panel B of Table A15 in the Supplementary Material reports a correlation of 0.70 between biospheric and green values, indicating that the constructs are closely related and that respondent values appear stable over time. This high correlation also supports the credibility of the responses, given the challenge of maintaining consistent answers across two surveys conducted nearly 3 years apart.
Although bankers may not have enough data to fully calibrate “internal models” of green loan performance, many hold well-defined beliefs that could guide such assessments. We test whether these beliefs contain predictive information beyond what is captured by biospheric values, which would be consistent with statistical discrimination. Conversely, if biospheric values predict outcomes beyond beliefs, this would point to taste-based discrimination.
Table A16 in the Supplementary Material addresses this question. Panel A shows that brown loan officers are more likely to approve brown loans, even after controlling for green beliefs and subjective green information. This finding aligns with results from Table A17 in the Supplementary Material, which analyzes biospheric values within a subsample of loan officers with below-median green beliefs, and is consistent with taste-based discrimination. In contrast, Panel B of Table A16 in the Supplementary Material demonstrates that green beliefs have limited predictive power for loan officer behavior. Without controls, green beliefs are positively associated with green loan approvals (t = 1.85), and brown beliefs are negatively associated (t = –1.22). However, once biospheric values are included, these weak associations become statistically insignificant. This suggests that statistical discrimination, as proxied by green beliefs, does not independently explain loan officer decisions.
While these findings are consistent with taste-based discrimination, they should be interpreted with caution. Our green beliefs measure, used as a proxy for statistical reasoning, was developed specifically for this study and has not been validated in prior research. As such, it may be a noisy measure. Measurement error of this kind would tend to attenuate estimated effects, potentially biasing our test against finding support for statistical discrimination and making the separation between the two mechanisms less sharp than the results suggest.
8. Placebo Test
Thus far, we have demonstrated that bankers’ biospheric values interact with the environmental score to generate significant differences in outcomes between those with high and low biospheric values. To ensure that these differences are indeed related to biospheric values rather than other traits, we conduct a placebo test by dividing customer managers into three other sets of personal traits: gender, education, and age. As Panel A of Table A4 in the Supplementary Material shows that these traits have low correlations with biospheric values.
Panel A of Table A18 in the Supplementary Material reports the results of the placebo test for customer managers’ gender. None of the coefficients differ significantly between genders at conventional levels. Panels B and D show that the same is true for education level and experience at the bank, respectively. In Panel C, two variables differ at the 5% level between age quintiles, but neither of these variables is the environmental score, which is the variable of interest. Overall, the placebo test results are consistent with the idea that the environmental score does not play a special role in the interaction when there is no reason for it to do so.
9. Loan Performance
Do biospheric values predict loan performance? On one hand, green bankers may possess better knowledge about environmental matters, enabling them to evaluate firms’ environmental credentials more effectively and make informed credit decisions. On the other hand, personal values might override objective judgment, leading bankers to approve loan applications that would not meet standard business criteria. It is ex ante not obvious which of these forces is more influential. We examine this issue in Table 8.
TABLE 8 Long description
Panel A, titled Loan Performance by Loan Officers’ Biospheric-Values Quintile, contains six rows. The Lowest quintile has 4 loans overdue 90 days or more out of 157 total (3%). Quintile 2 has 2 overdue out of 128 (2%). Quintile 3 has 1 overdue out of 147 (1%). Quintile 4 has 4 overdue out of 199 (2%). The Highest quintile has 19 overdue out of 200 (10%). The total across all quintiles is 30 overdue loans out of 831 (4%).
Panel B, titled Modeling Loan Performance with Loan Officers’ Biospheric Values and Interactions, presents four regression specifications where the dependent variable is a dummy for loans overdue 90 days or more. Key variables include:
* Debt-assets ratio: Positive and significant across all models (0.109 to 0.116).
* Ability to repay: Negative and significant across all models (-0.039 to -0.035).
* Willingness to repay: Negative and significant across all models (-0.260 to -0.278).
* Biospheric-values quintile 5: Positive and highly significant in specifications 2 and 3 (0.064 and 0.065).
* Interaction terms between biospheric quintiles and environmental scores are generally negative but not statistically significant.
* Loan officer F E (Fixed Effects) is only applied in specification 4.
* The sample size N is 831 for all models, with R-squared values increasing from 0.089 in specification 1 to 0.177 in specification 4.
Panel A of Table 8 presents descriptive statistics on the biospheric values-loan performance link, focusing on loans overdue by 90 days or more (henceforth, “bad” loans) and their association with loan officers’ biospheric-values quintile. We find that the percentage of bad loans varies between 1% and 3% for the four lowest biospheric-values quintiles but rises to 10% for the top quintile. Notably, all six loan officers with at least two bad loans are part of this top-biospheric-values group.
Panel B of Table 8 investigates the relationship between biospheric values and loan performance using regression analysis. While similar to Table 6 in other respects, the regression table now includes additional control variables for loan maturity and interest rate. Specifications 2 and 3 reaffirm the finding from Panel A that the top quintile of biospheric values is positively linked to bad loans, significant at the 1% level. When it comes to firm variables, our findings resemble those in the customer manager’s recommendation regression (Table 4) and loan officers’ loan granting regressions (Table 6): in general, the debt-assets ratio and current ratio exhibit significant positive associations with bad loans, whereas the ability and willingness to repay are significantly negatively associated with bad loans. Specifications 3 and 4 also explore interactions between loan officers’ biospheric values and the environmental score. However, none of these interactions are statistically significant. Table A19 in the Supplementary Material confirms that removing loan maturity and interest rate from our regression equation does not qualitatively alter our findings.
In addition to the 30 loans overdue by 90 days, our sample includes 60 loans that are less than 90 days overdue. Table A20 in the Supplementary Material re-estimates the analysis from Table 8 using this expanded sample. Panel A shows that, as in Table 8, late repayment is most frequent among the greenest quintile of loan officers, with a rate of 9% compared to 6%–7% in the other quintiles. However, this difference is not statistically significant at conventional levels. More importantly, Panel B shows that the interaction between the greenest quintile of loan officers and environmental scores is positive and statistically significant at the 10% level across both specifications. In other words, late repayment is more likely for loans with higher environmental scores when approved by the greenest officers. This finding is consistent with green officers being overly lenient toward loans with strong environmental characteristics.
Table 8 indicates that strong personal values might lead to poor loan performance. The potential for biased judgment stemming from these values, as suggested by the patterns in Table A20 in the Supplementary Material, could also explain why brown loan officers tend to view recommendations with high environmental scores negatively. However, we proceed with caution in drawing definitive conclusions from the findings on loan performance. The small number of bad loans—30 overdue by at least 90 days and 60 by less than 90 days—combined with the uncertainty in our estimates—limits our ability to draw firm conclusions about how personal values interact with environmental scores in shaping loan performance. Further research is needed to explore this relationship.
10. Additional Robustness Checks and Extensions
This subsection presents additional analyses that both test the robustness of our findings and extend them to related questions. First, the four subcomponents underlying biospheric values are highly correlated, and re-estimating the interaction between each subcomponent and environmental scores yields patterns consistent with the baseline results (Table A21 in the Supplementary Material). Second, splitting industries into terciles based on average environmental scores confirms that both the downgrade behavior of brown officers and the upgrade behavior of green officers arise across all industry groups, with the strongest downgrade behavior in the greenest industries (Table A22 in the Supplementary Material). Third, redefining loan officers’ environmental orientation using quartiles rather than quintiles produces similar interaction patterns, indicating that the main results do not depend on the exact classification of biospheric values (Table A23 in the Supplementary Material). Finally, we extend the analysis by examining loan terms conditional on approval; firm characteristics strongly predict acceptance but have limited power in explaining collateral, interest rates, or maturity, with soft variables generally insignificant and low R 2 values across specifications (Table A24 in the Supplementary Material). Appendix A5 of the Supplementary Material provides a more nuanced discussion of these results.
V. Conclusion
The hierarchical decision-making structure in our study, where loan officers evaluate the recommendations of customer managers, reflects common practices in many professions. For example, referees provide recommendations to editors in academic publishing, civil servants advise ministers in government, and junior consultants offer input to senior consultants in consulting firms. However, these systems generally lack the randomized assignment and anonymity present in our study, which makes it difficult to isolate the influence of personal biases and values in real-world settings.
Our article documents that customer managers tend to favor green firms, especially when they hold strong biospheric values. Meanwhile, a minority of brown loan officers push back against this trend by downgrading green firms. Although they do not know which customer manager initiated each case, these loan officers use their discretion to counter what they view as green bias, highlighting the moderating influence of superiors in the bank’s hierarchy.
The anonymity between customer managers and loan officers highlights the ability of hierarchies to separate decisions from personal relationships or reputational concerns, allowing loan officers to exercise discretion independently. However, this discretion also carries risks, as loan officers’ own values may influence their decisions, creating room for new biases even as others are mitigated. Anonymity also limits information flow, leaving loan officers without full context for customer managers’ recommendations and preventing customer managers from learning through feedback. These trade-offs reflect the broader challenge of designing hierarchies that balance fairness, objectivity, and coordination.
Appendix. Definition of Variables
Hard Firm Information
- Industry:
-
Indicator for the 10 industries represented in the sample, listed in Table A1, Panel A, in the Supplementary Material.
- Company registration type:
-
Indicator for one of the seven company registration types represented in the sample, listed in Table A1, Panel B, in the Supplementary Material.
- Firm age:
-
Age of the loan applicant firm.
- Total assets:
-
Total assets of the loan applicant in the most recent financial statement in RMB.
- Current ratio:
-
Current ratio of the loan applicant in the most recent financial statement.
- Debt-assets ratio:
-
Debt-assets ratio of the loan applicant in the most recent financial statement.
- Net profit ratio:
-
Net profit ratio of the loan applicant in the most recent financial statement.
- Sales growth:
-
Sales growth of the loan applicant calculated from the two most recent financial statements.
Soft Firm Information
- Truthfulness:
-
Categorical variable representing a customer manager’s subjective assessment of the truthfulness and reasonableness of the loan application’s stated purpose. 3 = True and reasonable; 2 = Not true and not reasonable; 1 = Cannot judge.
- Ability to repay:
-
Categorical variable representing a customer manager’s subjective assessment of the loan applicant’s loan repayment ability. 4 = Very strong; 3 = Normal; 2 = Weak; 1 = No.
- Willingness to repay:
-
Categorical variable representing a customer manager’s subjective assessment of the loan applicant’s loan repayment willingness. 3 = Strong; 2 = Normal; 1 = Weak.
- Safety score:
-
Categorical variable representing a customer manager’s subjective assessment of the loan applicant’s safety risk. 5 = Negligible risk; 4 = Low risk; 3 = Moderate risk; 2 = High risk; 1 = Cannot judge.
- Environmental score:
-
Categorical variable representing a customer manager’s subjective assessment of the loan applicant’s environmental risk. 5 = Negligible risk; 4 = Low risk; 3 = Moderate risk; 2 = High risk; 1 = Cannot judge.
- Overall recommendation:
-
Categorical variable representing the likelihood that a customer manager would suggest that the loan application be approved. 5 = Strongly recommend; 4 = Recommend; 3 = Neither recommend nor not recommend; 2 = Not recommend; 1 = Strongly not recommend.
Outcome Variables
- Loan granted:
-
Dummy variable indicating that the loan application has been granted.
- Pledge:
-
Dummy variable indicating that the applicant firm has pledged collateral for the loan.
- Annual effective interest rate:
-
Annual effective interest rate of the granted loan.
- Maturity:
-
Loan maturity in years.
- ≥ 90 days overdue:
-
Loan has not been serviced for at least 90 days.
- Approved loan amount:
-
Approved loan amount in RMB.
Customer Manager and Loan Officer Variables
- Age:
-
Age of the customer manager or loan officer at the time of the bankers’ survey.
- Experience:
-
Number of years the customer manager or loan officer has worked for the bank at the time of the bankers’ survey.
- Female dummy:
-
Dummy variable indicating that the customer manager or loan officer is female.
- High school or equivalent:
-
Dummy variable indicating that the customer manager’s or loan officer’s highest education is a high school degree.
- Bachelor’s:
-
Dummy variable indicating that the customer manager’s or loan officer’s highest education is a bachelor’s degree.
- Master’s or higher:
-
Dummy variable indicating that the customer manager’s or loan officer’s highest education is a master’s degree or higher.
- Education level:
-
Continuous variable assigned a value of 1 for high school or equivalent, 2 for a Bachelor’s degree, and 3 for a Master’s degree or higher.
- Biospheric values:
-
Measured by the average score for the attitude of the customer manager or loan officer toward four statements: i) Respecting the Earth; ii) Unity with nature; iii) Protecting the environment; and iv) Preventing pollution. Scores range from −1 to 7, with −1 representing strong disagreement and 7 indicating strong agreement. The statements and the scale are from de Groot and Steg (Reference de Groot and Steg2008) and Bouman et al. (Reference Bouman, Steg and Kiers2018). The survey took place in October 2020.
Supplementary Material
To view supplementary material for this article, please visit http://doi.org/10.1017/S0022109025102561.
Funding Statement
Bu and Liao acknowledge research support from the Australian Research Council under project No. LP200301118. Keloharju acknowledges research support from the Academy of Finland (Grant Nos. 292363 and 319316) and OP Group Research Foundation. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.
Open access
