Body Mass and Creditworthiness: Evidence from Loan Officers in Uganda

2019-11-01

2020-09-01

Access to credit:
- choice of granting a meeting
- choice of qualifying the application
- interest rate charged

Creditworthiness

Perceived Financial Ability:
- perception as able to put loan money to productive use

Access to credit:
(1) Meeting: A Yes/No choice of meeting the applicant to discuss and verify information;
(2) Qualifying: a 1-5 Likert scale assessment of the probability of qualifying the requested application.

Creditworthiness: a 1-5 Likert scale assessment of the creditworthiness of the applicant

Perceived Financial Ability: a 1-5 Likert scale assessment of the probability that the applicant puts the money to productive use.

Interest Rate.
Trust in the information provided (only when more information is provided).
Loan's officers attention:
- choice of receiving more information
- time spent on each application

Explicit Bias

Interest Rate: a 1-3 scale choice of the interest rate charged to the applicant (lower than standard, standard, higher than standard, not applicable). NA is selected for those loan officers who have no discretionality to change interest rate.

Trust in the information provided: A 1-5 Likert scale assessment of the probability that the applicant information is believed to be trustworthy.

Loan's officers attention: Yes/No choice of receiving more information on an applicant.

Explicit Bias: Pointing on a silhouette, whether an individual moving from low to high BM silhouette would more/less/equally likely to receive a loan.

Loan officers are asked to evaluate 30 hypothetical loan applications, in three sets. Each set is a treatment arm, the design exploits within subject variation and the order of the treatment arms is fixed. Each treatment arm varies according to how much and which information is presented. In the first treatment arm, respondents only see the picture, age and loan profile requested. In the second treatment arm, respondents only see the picture, age and loan profile requested but can choose to learn occupation and income of the applicant. In the third treatment arm, participants see all the information (including the picture) at once.

Within each treatment arm, the applications vary by characteristics; in each application, the applicants' age, nationality and place of residence are defined statically, while the applicants' loan profile, collateral, reason for loan, occupation, income (profits and revenues), gender, portrait, body mass are cross-randomised. Body mass is randomized by assigning the high body mass or low body mass photomorphed version of the randomly selected portrait. To avoid suspicions, portraits are randomized without replacement across the full set of applications each respondents receives so that no picture is seen twice by the same respondent.

Using the procedure detailed in the appendix, I first build 30 loan applications with all the cross randomized information, except body mass. Then I create the final set of 60 applications by creating two versions (a high body mass or low body mass) of each application. The 10 applications are randomly assigned to each arm stratified by characteristics. The order of the applications within each arm is random.

For each application, the loan officer answers the following questions:

1) Would you like us to refer to you this applicant to meet and discuss this loan application? Yes No

2) Based on your first impression, how likely would you be to approve this loan application? Not at all likely 1 2 3 4 5 Very likely

3) “Creditworthiness describes how likely a person is to repay a financial obligation according to the terms of the agreement.” Based on your first impression, how would you rate the person’s creditworthiness? Not at all creditworthy 1 2 3 4 5 Very creditworthy

4) Based on your first impression, how likely do you think this person would be to put the loan money to productive use? Not at all likely 1 2 3 4 5 Very likely

5) If you had to grant this loan to this person, what is the interest rate would you charge to this applicant?
Lower than standard - Standard - Higher than Standard
Interest rate question is only displayed to loan officers which have discretionality to change interest rates.

When answering the questions, the loan officers are informed that the applicants are hypothetical. Honest answers are incentivised by informing the loan officers that their answers will be used to provide referrals to prospective clients whose characteristics match with their preferences. The matching is performed using a machine learning algorithm, following Kessler et al. (2019).

The design allows to investigate whether
(1) body mass affects access to credit and creditworthiness,
(2) whether reducing asymmetric information on income changes the relevance of body mass in determining the decision,
(3) at which point of the decision making process does discrimination bites.

Relevant heterogeneity analysis include: 1) gender of applicant and loan officer; 2) body mass of the loan officer; 3) age of the loan officer; 4) perception of high body mass. If heterogeneity in the share of applicant invited is detected (either at the loan officers' level or at the institution level), I will compare "lemon dropping" types to "cherry picking" ones.
The incentives structure of the experiment is based on the idea that it is worthwhile for loan officers to receive good applicants. This incentives are the strongest when the loan officers' pay is tied to performance. While incentives might still be aligned also in the absence of performance pay (good applicants' referrals reduce screening and verification costs), nonetheless they are likely to be less relevant. Hence, if heterogeneity by performance pay will emerge, I will compare answers of high incentives vs low incentives loan officers.

The main statistics monitored are the difference in primary and secondary outcomes when applications are randomly assigned a high body mass vs a low body mass picture. I look at the difference in the full sample, and by treatment arm. When applicable, I assess the variation in this difference when the applicants 's wealth is high, and when applicants' wealth is low.

In addition to computing basic statistics, I will also identify parameters estimating the comparable regressions specifications. I will estimate the regressions by treatment arm, and on the full sample. I will also investigate how the other characteristics in the applications affect the main outcomes and how do these interact with the body mass variable, with a special interest on wealth and gender of the applicant.

To the basic regressions, I will add sequentially 1) loan profiles fixed effect; 2) loan officers fixed effect; 3) photo morphed pictures might differ on other dimensions, on top of body mass and wealth perception. That is why, at the end of the experiment the very same pictures are evaluated in terms of health, attractiveness, longevity, and self control. The information on wealth will be exploited as sanity check. The ratings will be pooled with the ratings from a similar rating exercise which involved 600 laypeople from Kampala and the average ratings will be included as controls.

computer

Loan officer level; Application level.

Yes

minimum/expected 250 loan officers.

The sample size is obtained as follows. The set of institutions is obtained by randomly selecting 250 branches in Kampala, Mukono and Wakiso, stratifying by bank tier and location. Notably, commercial banks are excluded from the randomisation, as they are unlikely to lend as little as low as Ush. 1 million. Each selected branch is visited to ask whether they will be willing to participate to the study. If they agree, first it is verified that the institution offers both personal and business loans; second, it is verified that the institution offers as little as Ush. 1 million loans. If these conditions are met, 1 to 2 loan officers are interviewed depending on availability. The procedure continues until exhaustion of the list. The target loan officers number is 250 loan officers. If the 250 branches interview fail to reach the target number, we will perform a second randomisation among the remaining branches.

Outliers:
We consider as poor data and hence exclude from the sample those observations which show variation in the outcome variable of interest "qualify" for a given arm, and specifically bounce on 3. Furthermore, we will consider as outliers those observations which show a consistent negative correlation between "meet" and "qualify".

30 applications per loan officer, for a total of expected 250 loan officers.

This is not an RCT. minimum 250 loan officers per treatment arm, 10 applications per treatment arm.

In Kessler et al. (2019), the effect of gender on hiring interest for the STEM majors subsample ranged from 0.399 to 0.516 sd., with a single treatment arm. Because the effect the effect of body mass is presumably mediated by wealth, I target a smaller MDE. Using data from power calculations, 2500 observations allow me to pick up for choice of meeting an effect slightly larger than 0.05 in each arm.