Session 2: Empirical Implementation of Theoretical Models of Strategic Interaction and Dynamic Behavior

We develop a general model of discrete choice that incorporates peer effects in preferences and consideration sets. We characterize the equilibrium behavior and establish conditions under which all parts of the model can be recovered from a sequence of choices. We allow peers to affect preferences, consideration, or both. We show that these peer-effect mechanisms have different behavioral implications in the data. This allows us to recover the set and the type of connections between the agents in the network. We then use this information to recover each agent’s preferences and consideration mechanisms. These nonparametric identification results allow for general forms of heterogeneity across agents and do not rely on the variation of either exogenous covariates or the set of available options (menus). We apply our results to model expansion decisions by tea chains and find evidence of limited consideration. We simulate counterfactual predictions and show how limited consideration slows market penetration and competition.

This paper introduces a new method to derive moment restrictions in dynamic discrete choice models with strictly exogenous regressors, fixed effects and logistic errors. We show how the structure of logit probabilities and basic properties of rational fractions can be used to construct moment functions free of the fixed effects in a way that scales naturally with the lag order and the number of observed periods. We demonstrate the approach in binary response models of arbitrary lag order, first-order panel vector autoregressions and dynamic multinomial logit models. The semiparametric efficiency bound is characterized for the leading binary case with one lag. Finally, we illustrate our results in an application investigating the dynamics of drug consumption among young people.

This paper examines how individual identity influences group behavior through social interactions. I study a discrete choice model in which people are affected differently by different members of their network, conforming to the actions of some peers while deviating from the actions of others. Under this generalized framework, I explore what aggregate outcomes arise from noncooperative decisionmaking. I analyze uniqueness and stability of equilibria, and I characterize how negative spillovers impact social welfare. I then show how to take the model to data, introducing a novel identification strategy that leverages within-network variation in individual characteristics to account for unobserved network effects. I also show how to construct internal instruments to overcome the issue of measurement error, which is a primary source of endogeneity in models with incomplete information. Lastly, I apply my method to data from the large-scale education experiment Project STAR, where I find strong evidence that classroom peer effects differ by gender.

Comparative statistics in high dimensional models are empirically demanding and analytically complicated. Complete identification of the correspondingly many models parameters is required, which is often infeasible due to data availability. I derive novel, sharp bounds on high dimensional comparative statics that depend only on low dimensional sufficient statistics, the knowledge of which is often more feasible, and have a simpler functional form relative to the exact relationship. I demonstrate application in canonical models across economics, and address existing methodological limitations in the research on peer effects, the gains from trade, and price cost passthrough.

We develop a new approach to estimating flexible demand models with exoge-nous supply-side shocks. Our approach avoids conventional assumptions of exogenous product characteristics, putting no restrictions on product entry, despite using instrumental variables that incorporate characteristic variation. The proposed instruments are model-predicted responses of endogenous vari-ables to the exogenous shocks, recentered to avoid bias from endogenous char-acteristics. We illustrate the approach in a series of Monte Carlo simulations.

Linear models are widely used in applied work. In this paper we explore which quan-tities can be learned if some coefficients are heterogeneous and unrestricted. In a paneldata setting, this corresponds to having a coefficient varying unrestrictedly across unitsand time periods. Our main focus is on linear combinations of the coefficients, such asweighted averages. We provide necessary and sufficient conditions for such quantities tobe point-identified, while also studying the identification of quadratic forms such as vari-ance components. We study three regimes, when the covariates are strictly exogenous,sequentially exogenous, and endogenous, and characterize conditions for identification ineach case. We illustrate the analysis by revisiting the impact of smoking during pregnancy on the weight of children at birth.

Data-driven decision-making involves estimating the value associated with each possible decision and select-ing the optimal estimated choice. This type of decision making is at the heart of a huge number of modern marketing applications, including ad creative choice, algorithm optimization, personalized targeting, A/B testing, pricing, and assortment optimization. In practice, it is crucial not only to estimate the optimal policy but also to accurately measure the incremental value or lift of that policy. In this paper, we first demonstrate theoretically that selecting the optimal policy based on estimated effects from data leads, on average, to overly optimistic evaluations of the policy value, a phenomenon known as the winner’s curse. This is true no matter what best methodology is used to estimate the policy’s effectiveness. We then empirically illustrate that the winner’s curse arises in a wide range of key marketing applications, including A/B testing, personalized targeting, and counterfactual estimation using structural models, and that its magnitude can be substantial even within realistic parameter ranges commonly seen in the literature. Given the generality of this problem across diverse settings, we propose a correction method based on a non-continuous bootstrap approach designed to effectively mitigate the winner’s curse in nearly all scenarios. Finally, we benchmark our proposed method against several existing context-specific solutions, demonstrating that our bootstrap-based correction consistently performs well and frequently outperforms alternative methods across important marketing contexts.

Decision-making processes increasingly rely on the use of algorithms. Yet, al- gorithms’ predictive ability frequently exhibits systematic variation across sub- groups of the population. While both fairness and accuracy are desirable proper- ties of an algorithm, they often come at the cost of one another, calling policy- makers to assess this trade-off based on finite data. We provide a debiased ma-chine learning estimator for a theoretical fairness-accuracy frontier put forward by Liang, Lu, Mu, and Okumura (2024), derive its asymptotic distribution, and propose inference methods to test hypotheses that have received much attention in the fairness literature, such as (i) whether fully excluding group identity from use in training the algorithm is optimal and (ii) whether there are less discriminatory alternatives to an existing algorithm. We also provide an estimator for the distance between a given algorithm and the fairest point on the frontier, and characterize its asymptotic distribution.

Multiple hypothesis testing practices vary widely, without consensus on which are appropriate when. This paper provides an economic foundation for these practices designed to capture leading examples, such as regulatory approval on the basis of clinical trials. In studies of multiple treatments or sub-populations, adjustments may be appropriate depending on scale economies in the research production function, with control of classical notions of compound errors emerging in some but not all cases. In studies with multiple outcomes, indexing is appropriate and adjustments to test levels may be appropriate if the intended audience is heterogeneous. Data on actual costs in the drug approval process suggest both that some adjustment is warranted in that setting and that standard procedures may be overly conservative.

This paper studies the sharp testable implications of an additive random utility model with a discrete multi-valued treatment and a discrete multi-valued instrument, in which each value of the instrument only weakly increases the utility of one choice. Borrowing the terminology used in randomized experiments, we call such a setting an encouragement design. We derive inequalities in terms of the conditional choice probabilities that characterize when the distribution of the observed data is consistent with such a model. Through a novel constructive argument, we further show these inequalities are sharp in the sense that any distribution of the observed data that satisfies these inequalities is generated by this additive random utility model.

Designing individualized allocation of treatments so as to maximize the equilibrium welfare of interacting agents has many policy-relevant applications. Focusing on sequential decision games of interacting agents, this paper develops a method to obtain optimal treatment assignment rules that maximize a social welfare criterion by evaluating stationary distributions of outcomes. Stationary distributions in sequential decision games are given by Gibbs distributions, which are difficult to optimize with respect to a treatment allocation due to analytical and computational complexity. We apply a variational approximation to the stationary distribution and optimize the approximated equilibrium welfare with respect to treatment allocation using a greedy optimization algorithm. We characterize the performance of the variational approximation, deriving a performance guarantee for the greedy optimization algorithm via a welfare regret bound. We implement our proposed method in simulation exercises and an empirical application using the Indian microfinance data (Banerjee et al., 2013), and show it delivers significant welfare gains.

This paper connects the literature on demand estimation to the literature on causal inference by interpreting nonparametric structural assumptions as restrictions on counterfactual outcomes. It offers nontrivial and equivalent restatements of key demand estimation assumptions in the Neyman–Rubin potential outcomes model, for both settings with market-level data (Berry and Haile, 2014) and settings with demographic-specific market shares (Berry and Haile, 2024). This exercise helps bridge the literatures on structural estimation and on causal inference by separating notational and linguistic differences from substantive ones.