Attend Spring Meetings on development topics from Apr 18-23. Comment and engage with experts. Calendar of Events

Syndicate content

Sex and Credit: Is There a Gender Bias in Lending?

Thorsten Beck's picture

Group identity in the form of family, ethnicity, or gender is a powerful predictor of social preferences, as shown by theory and empirical work. In particular, people generally favor in-group over out-group members. Such favoritism can have positive or negative repercussions. On the one hand, it can lead to inefficient transactions and lost opportunities. On the other hand, group identity may also entail trust, reciprocity, and efficiency due to shared norms and understandings. In recent research with Patrick Behr and Andreas Madestam, we gauge these opposing hypotheses, examining one important form of group identity, gender, and the consequences of own-gender preferences for outcomes in the credit market. We use microcredit transactions as they are an ideal ground to test these different hypotheses, relying heavily on transaction between loan officers and borrowers.

Using a large dataset of loan transactions from a commercial microlender in Albania, we investigate whether the officer-borrower gender match influences the likelihood that borrowers return to the lender for additional credit. Specifically, we have a large dataset on up to 7,300 first-time borrowers, with detailed information on loan conditions, arrears and borrower characteristics. Critically, we have information on gender of both borrower and loan officer as well as information on previous experience of loan officers.  In addition, we use variation in financial market competition and in the number of officers employed in a given branch across bank branches and over time to gauge the variation of a possible own-gender bias with loan officers’ discretion.

Estimating the effect of own-gender preferences presents two main challenges. First, if male or female borrowers with certain characteristics are more likely to be assigned to the same or opposite-sex loan officers, the true effect of loan officer gender would be biased. Second, if unobserved borrower traits are correlated with borrower gender, and if these can be observed by the loan officers but not by the researchers, it is not clear whether a significant coefficient on gender is due to a loan officer bias or the unobservable traits.

We address these issues by exploiting a quasi-random component of the institutional setting of the lender: the fact that first-time borrowers are arbitrarily assigned to their respective loan officer, with the sector of activity and year of application being the only factors driving the match with a specific officer. Conditional on sector and year, the random assignment of borrowers to officers ensures that unobservable borrower characteristics are the same across all officers, regardless of officer gender. In particular, we compare the difference in credit market outcomes for male and female borrowers obtaining loans from male loan officers to the difference between male and female borrowers obtaining loans from female loan officers.

Using this difference-in-difference technique, we find:

  • Borrowers matched with opposite-sex loan officers are 11 percent less likely to apply for a second loan with the same lender as compared to borrowers assigned to same-sex officers.
  • This effect is particularly strong in the case of loan officers with limited experience with borrowers from the other gender, suggesting that the bias fades away with gender-specific learning-on-the-job. We also find that the effect only exists when loan officers have a sufficient degree of discretion, as proxied by low competition from other financial institutions and branches with a small number of loan officers.
  • First-time borrowers assigned to officers of the other gender pay, on average, 35 basis points higher interest rates compared to borrowers assigned to same-gender officers. Again, these effects are more pronounced (around one percentage point) when officers have less opposite-sex experience and more discretion (weaker outside competition and smaller branches).
  • Borrowers matched with officers with less exposure to the other gender and a large degree of discretion also receive between 4 to 24 percent lower loan amounts.

If information asymmetries between officers and borrowers were important, the variation observed in interest rates or loan amounts should be reflected in different arrear outcomes. However, our results also suggest that loan arrears are independent of the officer-borrower gender assignment.  The lack of any differential arrear outcome supports the existence of an initial taste-based bias rather than the notion of an information hypothesis where loan officers are more efficient when transacting with own-gender as compared to opposite-gender borrowers.

Our findings have important repercussions for both financial institutions and policy makers. First, our results should affect firms’ human-resource practices as loan officers’ opposite-gender experience has repercussions for the size of the own-gender bias. Second, from a policy perspective, our findings support the conjecture that financial market competition can be a powerful tool in dampening the biases of loan officers, and, ultimately, banks, against borrowers of a certain gender.


Submitted by Vighneswara Swamy on

Professor Beck,

Would be glad to be elucidated more on the part of
DD method as to how it could capture the causal effect?

Thank you for the question. We use the difference-in-differences specification to control for unobservable borrower and especially loan-officer characteristics. Specifically, by comparing same-gender and opposite-gender borrowers of one specific loan officer makes sure that our findings are not driven by loan officer idiosyncratic characteristics. Similarly, we do not compare female vs. male borrowers, but borrowers assigned to a same-gender or opposite-gender loan officer.

With regard to employing DD method, let me inform you that i too employed the same dd method in a similar study and when the paper was submitted to WBER journal, the editor states as follows:
"I am afraid however that you face the same problems as many that it is very difficult to find credible counterfactuals. Using the double difference method does not solve the problem, as it very likely that SG participants not only differ in their initial level of welfare (which the DD method would correct for) but in their welfare dynamic. At the WBER, we strive to only publish papers that rigorously establish causal effects."

Given this opinion of the WBER Editor, what would be your opinion?

Add new comment