Research Highlights Podcast
May 16, 2023
Gender bias in bank lending
Michelle Brock discusses discriminatory lending practices toward women in Turkey.
Source: Natee Meepian
Around the world, female entrepreneurs borrow less than their male counterparts. Many people suggest that the reason for this gap comes down to the fact that women select into less capital-intensive industries.
But in a paper in the American Economic Journal: Applied Economics, authors J. Michelle Brock and Ralph De Haas show that implicit bias against women leads to more onerous guarantor requirements on loans. The findings come from a lab-in-the-field experiment conducted with over three hundred Turkish loan officers and real-life loan applications.
Brock says that the additional collateral requirements placed on female entrepreneurs could be a significant barrier to women running businesses. But there may be steps that banks can take to mitigate the problem.
Brock recently spoke with Tyler Smith about her and De Haas’s experiment in Turkey and what lessons policymakers should take away from the results.
The edited highlights of that conversation are below, and the full interview can be heard using the podcast player.
Tyler Smith: What inspired you to investigate gender discrimination in bank lending?
Michelle Brock: At the time that we started this study, there was so much that we collectively, researchers and practitioners, did not know about lending biases, and so we were driven by the need to measure bias precisely. Banks don't necessarily understand when they have bias in their lending.
We also were driven by some really intriguing research that was available at the time, suggesting that lending biases likely went beyond the fact that women are selecting into smaller and less capital-intensive firms. If you're a bank, the fact that a woman is in a smaller firm or a less capital-intensive firm usually means you will allocate less credit to them. And that has nothing to do with gender because you would do the same thing for a man who is applying with a smaller firm or a less capital-intensive firm.
In the loan portfolio—looking at who they're lending to and how much credit everybody's getting—it's really hard to piece out from that what exactly is going on with bias. The literature has had what we call omitted variables bias. So that is what then inspired us to study the gender biases in the laboratory setting.
Smith: Can you explain the lab in the field experiment that you used and how it helps you answer your research questions?
Brock: The outcomes we're interested in are whether or not men or women get offered credit at different rates, and whether or not they have different collateral requirements—specifically guarantor requirements.
We wanted to understand the effect of gender, so we needed a way to randomize gender. And this is of course where the lab becomes very useful because even in an RCT it's really hard to randomize gender. We brought loan officers in, and we worked to gather data from a portfolio of past applications that had been reviewed, either accepted or denied. We never shared the outcome with the loan officers, but we did ask them to review these files. We said look at this file, and tell us what you think. Should they get credit? We said we will give you points if you pick somebody to give credit who has a performing loan because that's how they're paid in real life. To randomize gender, we removed the person's name and all identifying information from each of these applications. But when we handed it to a loan officer doing the experiment with us, we randomly assigned a name at that time to that file. We gave that same file to a variety of different people with a variety of different names, male and female.
This is really exciting because we can do that thing you always hear about in econometrics class where we can hold all else equal and actually compare a firm when it's a male versus female application.
Smith: What do we learn from this experiment then? What are the main findings?
Brock: The main one that's really striking is that we don't see direct discrimination. We do not see that people are differentially assigning credit to men versus women. If the firm looks good on paper, they're equally likely to say a man can have the credit as a woman, but they are more likely, about 26 percent more likely, to require a woman to provide a guarantor. This is particularly pronounced in the sectors that are male dominated, and it is driven by scores on an implicit association test, which measures implicit bias. We also find a result of secondary importance. The people who are younger and have less experience in their careers are more likely to levy this extra guarantor requirement than people who are older and have more experience. While older people do show stronger implicit biases, it's not affecting their choices on the credit allocation as much.
We don't see direct discrimination. If the firm looks good on paper, they're equally likely to say a man can have the credit as a woman, but they are more likely, about 26 percent more likely, to require a woman to provide a guarantor.
Michelle Brock
Smith: How do you think your findings could be applied in practice to reduce gender discrimination in the financial sector?
Brock: When it comes to policy recommendations, we really lean into the result that it's driven by implicit bias. How can you reduce the extent to which that bias is reflected in behavior? Implicit biases are the result of just being part of society. They're really hard to change. So we're not going to suggest that we change implicit biases in all of society. Rather we ask what banks or firms can do to reduce the chances that that comes through on how their staff are making choices. For one, you can limit ambiguity. When you're making really fast decisions, that bias is more likely to come through in your behavior. Obviously, the private sector is very fast paced, but maybe we need to take a step back and offer a little more time for reviewing these things. Also, it would help to limit distractions when applications are being reviewed. So how can we limit ambiguity and time pressure?
We can make bank-wide targets for lending to women without guarantors. This is what's called a comply or explain policy. We can have mixed teams where more senior people have a little bit more mentoring in those early stages of the individual's career. On the more subliminal side, we can integrate successful female entrepreneurs into training programs. This simply increases the visibility for women as business people, the pictures you see in your training programs or in your daily work, positive aspects of women in business, etcetera.
One question that we have is whether or not automated credit scoring can help. Are there any ways in which AI can help us make less biased decisions now? We're not the first to ask that question. And there's a bunch of research suggesting that actually AI is just going to reinforce the biases that already exist in the data. But at one point we thought there was promise in this space. Is there a way to help it deliver on that promise? We're not sure about that, but it seems like it could still be an interesting space.
♦
“Discriminatory Lending: Evidence from Bankers in the Lab” appears in the April 2023 issue of the American Economic Journal: Applied Economics. Music in the audio is by Podington Bear.