TL;DR
This paper models a bank's lending decisions under capital constraints (caps and floors) using dynamic programming, revealing how these regulations lead to inefficiencies and unfairness. It shows that optimal lending policies become more complex and can significantly reduce profitability compared to an unconstrained scenario.
What Problem Does This Paper Address?
Financial regulators often impose caps or floors on bank lending to influence economic activity. These regulations, however, can create complex decision-making problems for banks, potentially leading to inefficient allocation of capital and unfair treatment of borrowers. The paper addresses the question of how these constraints affect a bank's optimal lending strategy, its profitability, and the fairness of its decisions. Specifically, it investigates the trade-offs between maximizing profit and adhering to regulatory limits, and how these trade-offs manifest in accept/reject decisions for loan applicants.
Methods
The core of the paper's methodology is the application of Markov Decision Processes (MDPs) to model the lending problem. This approach allows for the optimization of sequential decisions under uncertainty. The authors develop dynamic programming equations to represent the expected future reward for a bank considering a loan applicant, taking into account the available capital, time remaining, and the applicant's characteristics (loan amount and probability of repayment). They analyze different scenarios, including fully funded, partially funded, and not funded loans when a cap is reached. The paper also explores a discrete-time approximation of the continuous-time model for numerical analysis and compares the optimal capped/floored policies with suboptimal heuristics.
Datasets
Not discussed in the paper. The paper uses simulated parameters for its numerical examples rather than real-world datasets.
Key Insights and Ideas
- Regulatory Constraints Create Inefficiency and Unfairness: The central finding is that caps and floors, while intended to manage the economy, introduce significant inefficiencies and can lead to unfairness in lending decisions.
- "Polygamous Marriage" Analogy: The problem is framed as a variant of the "polygamous marriage" or "secretary problem," where a decision-maker must choose from a sequence of applicants with limited information and a finite horizon.
- Dynamic Cut-off Scores: Unlike a simple credit score threshold, the optimal lending decision (represented by a cut-off probability of repayment) is dynamic. It depends not only on the applicant's creditworthiness but also on the remaining capital, time, and the loan amount requested.
- Overshooting Effect: In the "fully funded" cap scenario, the model allows for "overshooting" the capital limit, which can sometimes lead to higher profits by allowing the bank to wait for a very good borrower even if it means temporarily exceeding the cap.
- Loan Amount Matters: The paper demonstrates that the distribution of loan amounts requested by applicants significantly impacts the optimal policy, contrary to what might be intuitively expected (i.e., that more capital always leads to less restrictive policies).
- Floors are More Damaging than Caps: The analysis suggests that imposing lending floors can have a more severe impact on profitability than imposing caps, forcing banks to lend even when it's not profitable to meet the minimum requirement.
Key Results
- Profit Reduction: Capping lending can lead to profit drops of up to 60% compared to an uncapped scenario. Imposing floors can result in profit reductions of up to 82% in extreme cases.
- Inconsistent Accept/Reject Decisions: Capping can make the accept/reject decision inconsistent, with the required "good" credit score varying significantly (from 0.857 to 1.00) depending on available capital and time.
- Loan Amount Preference: The model often favors borrowers requesting smaller loan amounts, as this allows for more flexibility and potential overshooting.
- Floor Policy Behavior: The optimal policy under a floor constraint often resembles a "bang-bang" strategy, switching between accepting all borrowers and only profitable ones, becoming more lenient as the floor is approached.
Limitations and Future Work
- Assumptions: The model relies on several assumptions, including a Poisson arrival process for applicants, known distributions of loan amounts and repayment probabilities, and specific cost/reward structures.
- Discrete vs. Continuous Time: While the paper uses a continuous-time model, it also employs a discrete-time approximation for numerical examples, which introduces its own set of approximations.
- Future Work Suggestions:
- Deferred Decisions: Modeling situations where lenders can defer decisions to a later point in the lending horizon to gain more information.
- Credit Union Extension: Adapting the model to credit unions where capital availability (based on deposits) is variable and unknown.
- More Complex Cost Functions: Incorporating more nuanced cost structures beyond simple profit and loss.
Summary updated