Project 16: Evaluating a Policy to Combat Loan Delinquency

A bank implemented a policy focused on reducing delinquency through three main mechanisms:

• Renegotiation and restructuring of existing debts: offering longer terms, partial debt forgiveness, and reduced interest rates to delinquent clients;
• Conditional credit offers to clients with regular status;
• Active monitoring and support for clients at risk of delinquency.

Side effect: such policies may lead to excessive relaxation of credit criteria, granting credit to riskier clients and creating the opposite effect of what was intended.

The central question is: was the policy effective in reducing loan delinquency?

The variables appear to be well balanced for both groups, which is essential to ensure they do not determine the results.

To solve the problem, we used:

• Difference-in-Differences (DiD): to estimate the causal impact by capturing the pre- and post-policy periods;
• Propensity Score Matching (PSM): to balance covariates that differ between treatment and control groups.

Dependent Variable: Delinquency

Covariates included in the models:

• Age
• Income
• Credit History
• Number of Dependents
• Time as Client

This combination allows us to estimate the causal effect of the policy while ensuring that results are not driven by biases.

Note: for the diff-in-diff model to be valid, the parallel trends assumption must hold—meaning that, in the absence of treatment, the outcome variable would evolve similarly for both treatment and control groups. This assumption will be tested using a placebo effect.

• The interaction coefficient treatment_dummy:post_dummy from the model with covariates is 0.061 (significant, p < 0.001).
• This indicates an increase of 6.1 percentage points in delinquency in the treatment group due to the policy.

Considering:

• Number of clients: 3,932
• Average loss per defaulting client: R$ 1,500

The estimated effect of the policy is:

• Additional defaulting clients: 0.061 × 3,932 ≈ 240
• Estimated financial loss: 240 × 1,500 ≈ R$ 360,000

In other words, the policy produced the opposite effect of what was intended, increasing defaults and causing direct financial losses for the bank.

Robustness checks confirm the result:

• PSM with caliper: policy effect still positive (higher delinquency)
• PSM with alternative variables: similar effect
• Placebo test: no effect detected in the pre-policy period, supporting the model’s validity.

The policy did not reduce loan delinquency and led to additional financial losses. The main reason is that renegotiation measures and credit concessions ended up relaxing risk criteria, allowing higher-risk clients to continue or resume borrowing, thereby increasing delinquency rates.

• Models estimated with R using packages: dplyr, tidyr, MatchIt, fixest, ggplot2, broom, kableExtra, ggplot2, cowplot