Distribution-valued Causal Machine Learning: Implications of Credit on Spending Patterns

2025-09-03

AI Summary

Prior studies show credit limit increases raise average consumption, but scalar effects obscure heterogeneous consumer responses across the expenditure distribution. Method: This paper introduces the first causal framework for estimating the distributional effect of continuous treatment—credit limit adjustments—on consumption expenditure distributions within the Wasserstein space. We propose a distributional double machine learning estimator that jointly employs neural function regression and conditional normalizing flow networks to nonparametrically estimate generalized propensity scores and capture complex nonlinear treatment–outcome relationships. Contribution/Results: Our method accurately identifies distributional shifts across multiple simulation scenarios. Empirically, we find that credit limit increases primarily elevate consumption at upper quantiles rather than uniformly raising overall expenditure. This work establishes a novel paradigm and scalable toolkit for distributional causal inference with continuous treatments in fintech applications.

Technology Category

• Machine Learning: Causal Learning
• Search and Optimization: Distributed Search
• Multiagent Systems: Mechanism Design

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience

Abstract

Fintech lending has become a central mechanism through which digital platforms stimulate consumption, offering dynamic, personalized credit limits that directly shape the purchasing power of consumers. Although prior research shows that higher limits increase average spending, scalar-based outcomes obscure the heterogeneous distributional nature of consumer responses. This paper addresses this gap by proposing a new causal inference framework that estimates how continuous changes in the credit limit affect the entire distribution of consumer spending. We formalize distributional causal effects within the Wasserstein space and introduce a robust Distributional Double Machine Learning estimator, supported by asymptotic theory to ensure consistency and validity. To implement this estimator, we design a deep learning architecture comprising two components: a Neural Functional Regression Net to capture complex, nonlinear relationships between treatments, covariates, and distributional outcomes, and a Conditional Normalizing Flow Net to estimate generalized propensity scores under continuous treatment. Numerical experiments demonstrate that the proposed estimator accurately recovers distributional effects in a range of data-generating scenarios. Applying our framework to transaction-level data from a major BigTech platform, we find that increased credit limits primarily shift consumers towards higher-value purchases rather than uniformly increasing spending, offering new insights for personalized marketing strategies and digital consumer finance.

Problem

• Estimating credit limit effects on spending distribution
• Proposing causal inference for continuous treatment impacts
• Addressing heterogeneous consumer responses to credit

Innovation

• Distributional Double Machine Learning estimator
• Neural Functional Regression Net architecture
• Conditional Normalizing Flow Net scoring