Context-Dependent Heterogeneous Preferences: A Comment on Barseghyan and Molinari (2023)

Matias D. Cattaneo, Xinwei Ma, Yusufcan Masatlıoĝlu

Abstract

Barseghyan and Molinari (2023) established sufficient conditions for semi-nonparametric point identification in mixture models of decision-making under risk, assuming unobserved but context-independent risk preferences. This comment extends their framework to allow for context-dependent risk preferences. We demonstrate that the underlying distribution of heterogeneous preferences remains identifiable even when individuals' risk attitudes shift across different decision environments, leveraging discontinuities in the derivatives of choice probabilities.

Introduction

A central goal in empirical economics is to recover unobserved risk preferences from observed choices, such as insurance deductible selections. While recent advances have enabled identification in models with unobserved heterogeneity and limited consideration, they often rely on the assumption of context-independence—that a decision-maker's utility function remains constant across different choice environments. However, behavioral evidence suggests that preferences can be domain-specific. This paper addresses whether mathematical identification is still possible when risk attitudes are malleable and context-dependent.

Theoretical Framework: A Mixture Model of Preferences

The proposed framework models a population composed of three distinct types of decision-makers:

  1. Type-ν: Individuals who consistently use utility function $U_\nu$ across all contexts.
  2. Type-ω: Individuals who consistently use utility function $U_\omega$.
  3. Context-Dependent Type: Individuals who switch between utility functions (e.g., using $U_\nu$ in Context I and $U_\omega$ in Context II).
The three types of decision-makers in the mixture model: (a) Type-ν, (b) Type-ω, and (c) Context-dependent preferences modeled via a joint distribution.

As illustrated in Figure 2, the model accounts for both the marginal distributions of preferences ($F$ and $G$) and the joint distribution for context-dependent types, often represented via a copula function $C(F(\cdot), G(\cdot))$.

Identification Strategy

To uniquely determine the model's parameters from choice data, the authors employ a "discontinuity-in-derivative" approach. This method analyzes how choice probabilities respond to marginal changes in prices (or contract attributes). By identifying "indifference points"—where an infinitesimal price change causes a specific type to switch choices—and examining the limits of these changes from different directions, the researchers can isolate the mixing probabilities (the population share of each type) and the specific shapes of the preference distributions.

A key insight is that context-dependent types exhibit unique "binding" behaviors. While a context-independent type is rarely at the edge of indifference in multiple scenarios simultaneously, a context-dependent type can be. This creates detectable discontinuities in the aggregate data that allow for the recovery of the joint preference distribution.

Key Results

The paper presents two primary identification theorems:

  • Theorem 1 (Full Consideration): Under the assumption that consumers evaluate all available options, the mixing probabilities ($\alpha, \beta$) and the preference distributions ($F, G$) are point-identified.
  • Theorem 2 (Limited Consideration): The identification result holds even when consumers have random consideration sets (i.e., they only observe a subset of options), provided there is sufficient variation in prices across contexts.

These results generalize the findings of Barseghyan and Molinari (2023), proving that context-independence is a sufficient but not necessary condition for semi-nonparametric identification in these mixture models.

Conclusion and Limitations

This research demonstrates that context-dependent risk preferences can be rigorously identified using econometric techniques. However, the practical application of these results requires significant price variation in the data to locate the necessary indifference points. Furthermore, the model assumes "narrow bracketing," where decisions in one context are not directly influenced by the specific prices of another. Future empirical work is needed to implement these identification results using real-world datasets, such as household insurance choices.

Summary updated

Abstract

Barseghyan and Molinari give sufficient conditions for semi-nonparametric point identification of parameters of interest in a mixture model of decision-making under risk, allowing for unobserved heterogeneity in utility functions and limited consideration. A key assumption in the model is that the heterogeneity of risk preferences is unobservable but context-independent. In this comment, we build on their insights and present identification results in a setting where the risk preferences are allowed to be context-dependent.
BibTeX
@article{cattaneo2023context,
  author = {Matias D. Cattaneo and Xinwei Ma and Yusufcan Masatlıoĝlu},
  title = {{Context-Dependent Heterogeneous Preferences: A Comment on Barseghyan and Molinari (2023)}},
  journal = {Journal of Business and Economic Statistics},
  year = {2023},
  doi = {10.1080/07350015.2023.2216740},
}
Figure