This paper develops a flexible approach to predict the price effects of horizontal mergers using ML/AI methods. While standard merger simulation techniques rely on restrictive assumptions about firm conduct, we propose a data-driven framework that relaxes these constraints when rich market data are available. We develop and identify a flexible nonparametric model of supply that nests a broad range of conduct models and cost functions. To overcome the curse of dimensionality, we adapt the Variational Method of Moments (VMM) [Bennett and Kallus, 2023] to estimate the model, allowing for various forms of strategic interaction. Monte Carlo simulations show that our method significantly outperforms an array of misspecified models and rivals the performance of the true model, both in test sample prediction and in counterfactual merger simulations. As a way to interpret the economics of the estimated supply function, we simulate pass-through and find that the model learns markup and cost functions that imply approximately correct pass-through. Applied to the American Airlines-US Airways merger, our method produces more accurate post-merger price predictions than traditional approaches. The results demonstrate the potential for ML/AI techniques to enhance merger analysis while maintaining (nonparametric) structure from economics.

We develop a method to recover primitives from data generated by artificial intelligence (AI) agents in strategic environments such as online marketplaces and auctions. Building on the design of leading online learning AIs, we assume that agents minimize their regret. Under asymptotic no regret, we show that the time average of play converges to the set of Bayes coarse correlated equilibrium (BCCE) predictions. Our econometric procedure is based on BCCE restrictions and convergence rates of regret-minimizing AIs. We apply the method to pricing data in a digital marketplace for used smartphones. We estimate sellers' cost distributions and find lower markups than in centralized platforms.

We propose a simple method to estimate demand in markets for differentiated products. The method augments price and quantity data with triplets data (of the form "product A is closer to B than it is to C'') obtained from an online survey. Using a machine learning algorithm, the triplets data are used to estimate an embedding---i.e., a low-dimensional representation of the latent product space. Distances between pairs of products, computed from the embedding, discipline substitution patterns in a simple log-linear demand model. This approach solves the dimensionality problem of product-space demand models (too many cross-price elasticity parameters to estimate). We illustrate the performance of the method by estimating demand for ready-to-eat cereals and comparing our estimates to those obtained from the standard method of (BLP). We find that our elasticity estimates imply credible substitution patterns and compare favorably to the BLP estimates. Beyond our current implementation of the method, the embedding data can be incorporated in either characteristic-space demand approaches, or in more complex product-space models.