arXiv subjects

Matthew Kovach

Publications and source records attributed to Matthew Kovach.

12 recordsLinked to original sources

arXiv2026

Do People Follow AI Advice? Evidence from a Pension Portfolio Choice Experiment

We study how differences in AI-generated financial recommendations are transmitted into individual portfolio choices. In an experiment with 400 employed adults enrolled in workplace defined contribution pension plans in South Korea, participants allocate a hypothetical pension balance across eleven products and may revise it after receiving one of two fixed AI-generated recommendations. A $2 \times 2$ design randomizes recommendation content and whether the recommendation includes a short rationale. Approximately 37$\%$ of the experimentally induced difference between the aggressive and conservative recommendations passes through to final portfolios. This causal contrast changes expected portfolio return, volatility, allocations across risk grades, and the number of products held, but produces no detectable difference in computed Sharpe ratios. 81$\%$ of participants revise. Among revisers, 95$\%$ move toward the assigned recommendation and implement about half of the suggested adjustment. Rationales do not detectably alter pass-through. These results show that users partially and selectively transmit recommendation content into economically meaningful differences in risk exposure while retaining substantial weight on their initial choices.

econ.GN↗

arXiv2026

Learning from an Unknown DGP: Experimental Evidence on Belief Updating with AI Recommendations

We use a controlled experiment to study how beliefs are updated after receiving qualitative information (AI recommendations) from an unknown data-generating process (DGP). Across 60,252 pairs of prior and posterior beliefs, we document three behavioral patterns: updates close to zero when recommendations confirm extreme priors, larger updates when recommendations contradict extreme priors, and smaller updates for intermediate priors. These three behavioral patterns suggest four testable properties of belief updating, which we assess at the aggregate and individual levels. Finally, we examine how well updates are captured by three models of belief updating.

econ.GN↗

arXiv2026

The Focal Quantal Response Equilibrium

We propose a generalization of Quantal Response Equilibrium (QRE) built on a simple premise: some actions are more focal than others. In our model, which we call the Focal Quantal Response Equilibrium (Focal QRE), each player plays a stochastic version of Nash equilibrium as in the QRE, but some strategies are focal and thus are chosen relatively more frequently than other strategies after accounting for expected utilities. The Focal QRE is able to systematically account for various forms of bounded rationality of players, especially regret-aversion, salience, or limited consideration. The Focal QRE is also useful for explaining the observed heterogeneity of bounded rationality of players across different games. We show that regret-based focal sets perform relatively well at predicting strategies that are chosen more frequently relative to their expected utilities.

econ.TH↗

arXiv2026

Misspecified Model Estimation and Its Impact on Predictions

We study a linear statistical model where outcomes depend on regressors with fixed population coefficients and observation-specific latent coefficients, along with measurement errors. A decision-maker estimates population coefficients and uses the estimates to predict the latent coefficients for a given observation. We analyze how misspecification of some population coefficients distorts predictions, investigating comparative statics with respect to: (1) residual information in regressors associated with misspecified coefficients after projecting out those associated with free coefficients, (2) alignment between misspecification vector and latent-to-coefficient mapping. Applications include employee rating with unconscious bias and LLM-mediated consumer research.

econ.TH↗

arXiv2026

Can an LLM Learn Preferences from Choice Data?

Can large language models (LLMs) learn a decision maker's preferences from observed choices and generate preference-consistent recommendations in new situations? We propose a portable Simulate-Recommend-Evaluate framework that tests preference learning from revealed-choice data by comparing LLM recommendations with optimal choices implied by known preference primitives. We apply the framework to choice under uncertainty using the disappointment aversion model. Recommendation accuracy improves as models observe more choices, but learning is heterogeneous across preference types and LLMs: GPT learns risk aversion better than disappointment aversion, Gemini performs best in high disappointment-aversion regions, and Claude shows the broadest effective learning across parameter regions.

econ.GN↗

arXiv2025

Inertial Updating with General Information

We study belief revision when information is represented by a set of probability distributions, or general information. General information extends the standard event notion while including qualitative information (A is more likely than B), interval information (A has a ten-to-twenty percent chance), and more. We behaviorally characterize Inertial Updating: the decision maker's posterior is of minimal subjective distance from her prior, given the information constraint. Further, we introduce and characterize a notion of Bayesian updating for general information and show that Bayesian agents may disagree. We also behaviorally characterize f-divergences, the class of distances consistent with Bayesian updating.

econ.TH↗

arXiv2023

Ambiguity and Partial Bayesian Updating

Models of updating a set of priors either do not allow a decision maker to make inference about her priors (full bayesian updating or FB) or require an extreme degree of selection (maximum likelihood updating or ML). I characterize a general method for updating a set of priors, partial bayesian updating (PB), in which the decision maker (i) utilizes an event-dependent threshold to determine whether a prior is likely enough, conditional on observed information, and then (ii) applies Bayes' rule to the sufficiently likely priors. I show that PB nests FB and ML and explore its behavioral properties.

econ.TH↗

arXiv2023

Reference Dependence and Random Attention

We explore the ways that a reference point may direct attention. Utilizing a stochastic choice framework, we provide behavioral foundations for the Reference-Dependent Random Attention Model (RD-RAM). Our characterization result shows that preferences may be uniquely identified even when the attention process depends arbitrarily on both the menu and the reference point. The RD-RAM is able to capture rich behavioral patterns, including frequency reversals among non-status quo alternatives and choice overload. We also analyze specific attention processes, characterizing reference-dependent versions of several prominent models of stochastic consideration.

econ.TH↗

arXiv2023

Inertial Updating

We introduce and characterize inertial updating of beliefs. Under inertial updating, a decision maker (DM) chooses a belief that minimizes the subjective distance between their prior belief and the set of beliefs consistent with the observed event. Importantly, by varying the subjective notion of distance, inertial updating provides a unifying framework that nests three different types of belief updating: (i) Bayesian updating, (ii) non-Bayesian updating rules, and (iii) updating rules for events with zero probability, including the conditional probability system (CPS) of Myerson (1986a,b). We demonstrate that our model is behaviorally equivalent to the Hypothesis Testing model (HT) of Ortoleva (2012), clarifying the connection between HT and CPS. We apply our model to a persuasion game.

econ.TH↗

arXiv2022

Ordered Surprises and Conditional Probability Systems

We study conditioning on null events, or surprises, and behaviorally characterize the Ordered Surprises (OS) representation of beliefs. For feasible events, our Decision Maker (DM) is Bayesian. For null events, our DM considers a hierarchy of beliefs until one is consistent with the surprise. The DM adopts this prior and applies Bayes' rule. Unlike Bayesian updating, OS is a complete updating rule: conditional beliefs are well-defined for any event. OS is (behaviorally) equivalent to the Conditional Probability System (Myerson, 1986b) and is a special case of Hypothesis Testing (Ortoleva, 2012), clarifying the relationships between the various approaches to null events.

econ.TH↗

arXiv2022

Behavioral Foundations of Nested Stochastic Choice and Nested Logit

We provide the first behavioral characterization of nested logit, a foundational and widely applied discrete choice model, through the introduction of a non-parametric version of nested logit that we call Nested Stochastic Choice (NSC). NSC is characterized by a single axiom that weakens Independence of Irrelevant Alternatives based on revealed similarity to allow for the similarity effect. Nested logit is characterized by an additional menu-independence axiom. Our axiomatic characterization leads to a practical, data-driven algorithm that identifies the true nest structure from choice data. We also discuss limitations of generalizing nested logit by studying the testable implications of cross-nested logit.

econ.TH↗

arXiv2021

Conservative Updating

This paper provides a behavioral analysis of conservatism in beliefs. I introduce a new axiom, Dynamic Conservatism, that relaxes Dynamic Consistency when information and prior beliefs "conflict." When the agent is a subjective expected utility maximizer, Dynamic Conservatism implies that conditional beliefs are a convex combination of the prior and the Bayesian posterior. Conservatism may result in belief dynamics consistent with confirmation bias, representativeness, and the good news-bad news effect, suggesting a deeper behavioral connection between these biases. An index of conservatism and a notion of comparative conservatism are characterized. Finally, I extend conservatism to the case of an agent with incomplete preferences that admit a multiple priors representation.

econ.TH↗