Individual welfare analysis: Random quasilinear utility, independence, and confidence bounds

Junlong Feng, Sokbae Lee

TL;DR

This paper introduces a robust framework for calculating how much a price hike hurts a specific individual, rather than just the "average consumer." By inverting a cutting-edge statistical test for independence (Chatterjee’s $\xi$), the authors provide a way to find reliable "welfare loss" bounds that are computationally efficient and resistant to the common data pitfalls that plague traditional economic models.

What Problem Does This Paper Address?

In economics, "welfare analysis" measures how changes in the market—like a new tax or a supply chain disruption—affect people's well-being. Traditionally, this is done at the aggregate level: we calculate how the average household is affected.

However, averages lie. A gasoline price hike might be a minor annoyance for a wealthy urbanite but a catastrophe for a low-income rural commuter. Capturing this "individual heterogeneity" is notoriously difficult because we cannot observe every person's internal preferences or the random "shocks" that influence their choices. Furthermore, standard statistical methods often fail when the data is "weak" (i.e., when the variables we use to identify the model don't provide strong signals) or when the relationship between variables is highly nonlinear.

Methods

The authors build their framework on a Random Quasilinear Utility model. This model allows for multiple goods and accounts for unobserved individual-level preference shocks.

The technical "secret sauce" of the paper is the use of Chatterjee’s Correlation Coefficient ($\xi$).

  1. Inverting the Test: Instead of just trying to find a single "best" estimate for consumer preferences, the authors "invert" a test of independence. They look for all possible preference parameters that are consistent with the assumption that unobserved shocks are independent of exogenous variables (like distance to a supply source).
  2. Joint Limiting Distribution: The authors prove a new mathematical result: a vector of Chatterjee’s statistics for different goods follows a joint normal distribution. This allows them to build a "box" of plausible parameters for many goods simultaneously.
  3. The Intersection Approach: To make the bounds even tighter, they intersect their new confidence set with traditional ones (like those from 2SLS or OLS).
  4. Scalable Optimization: They demonstrate that the welfare loss function is monotonic and concave. This means that finding the "worst-case" and "best-case" welfare loss for an individual is a simple convex optimization problem that can be solved in milliseconds, even for many goods.

Datasets

The authors validate their method using both simulations and real-world data:

  • 2001 National Household Travel Survey (NHTS): Used to analyze gasoline demand among 3,640 households.
  • Stanford Basket Dataset: A household-level scanner panel dataset used to analyze food demand (specifically ice cream vs. other foods) across hundreds of households over two years.
  • Monte Carlo Simulations: Synthetic data used to test the speed and accuracy of the bounds under various conditions (weak instruments, correlated shocks, etc.).

Key Insights and Ideas

  • Robustness to "Weakness": Traditional methods (like 2SLS) can produce wildly inaccurate results if the "instruments" used are weak. Because this method is based on inverting a test of independence, it remains valid even under partial identification or weak instruments.
  • Computational Efficiency: Inverting Chatterjee’s test is roughly 1,000 times faster than inverting other popular independence tests (like Hoeffding’s D). This makes individual-level analysis feasible for large datasets.
  • Independence as a Diagnostic: If the "intersection" of the new confidence set and a traditional one is empty, it serves as a "warning light" that the underlying economic model might be misspecified.
  • Beyond Averages: The method allows researchers to plot a "welfare loss curve" for a specific population, showing exactly who is hit hardest by a policy change.

Key Results

  • Gasoline Application: The study found that higher-income individuals actually suffer a larger absolute welfare loss from gasoline price increases. This is because they tend to have a higher marginal utility for driving (they value it more) compared to lower-income groups who may rely more on public transit.
  • Food Application: In the ice cream study, the authors were able to calculate tight welfare bounds for 39 specific households. They showed that even within a small sample, there is significant "visible" heterogeneity in how families react to price changes.
  • Statistical Power: Simulations showed that while Chatterjee’s test is sometimes criticized for low power in small samples, the "Intersection Approach" successfully tightens the bounds without sacrificing the desired confidence level (e.g., 95%).

Limitations and Future Work

  • Corner Solutions: The current model assumes "interior solutions"—meaning it assumes consumers buy at least a little bit of every good. It does not yet handle "zero consumption" (e.g., someone who never buys gasoline) gracefully.
  • Income Effects: The quasilinear utility model assumes there is no "income effect" (where a price change makes you feel poorer and thus changes your demand for all goods). The authors mitigate this by letting parameters vary by income group, but a more general utility function is a goal for future research.
  • I.I.D. Assumption: The theory currently relies on the assumption that data points are independent and identically distributed, which might be a stretch for complex panel data. The authors suggest this as a "convenient approximation" for now.

Summary updated

Abstract

We introduce a novel framework for individual-level welfare analysis. It builds on a parametric model for continuous demand with a quasilinear utility function, allowing for heterogeneous coefficients and unobserved individual-good-level preference shocks. We obtain bounds on the individual-level consumer welfare loss at any confidence level due to a hypothetical price increase, solving a scalable optimization problem constrained by a novel confidence set under an independence restriction. This confidence set is computationally simple and robust to weak instruments, nonlinearity, and partial identification. The validity of the confidence set is guaranteed by our new results on the joint limiting distribution of the independence test by Chatterjee (2021). These results together with the confidence set may have applications beyond welfare analysis. Monte Carlo simulations and two empirical applications on gasoline and food demand demonstrate the effectiveness of our method.
BibTeX
@article{feng2024individual,
  author = {Junlong Feng and Sokbae Lee},
  title = {{Individual welfare analysis: Random quasilinear utility, independence, and confidence bounds}},
  journal = {Journal of Econometrics},
  year = {2024},
  doi = {10.1016/j.jeconom.2024.105927},
}

Loading related content...

Figure