Demand models’ substitution predictions inform merger review and the value of product variety. Using the 2015 withdrawal of the regional ice cream brand Blue Bell, I test how well models estimated on data through March 2015 predict the amount of demand retained by rivals and its allocation among them. Relative to markets Blue Bell never served, rivals recaptured about three quarters of its volume, and private label took half of that gain. Logit and random-coefficient models without nesting retain too little demand. A nested logit that pools private label with Blue Bell’s mainstream tubs comes close both on how much demand stays with rivals and on which rivals receive it, and random-coefficient nested logit (RCNL), which allows consumer heterogeneity on the same nests, performs similarly. In this application, where private label sits in the nests matters more for these predictions than how much taste heterogeneity a model allows.

We develop a method to estimate discrete games where firms make a large number of interdependent discrete decisions, based on the optimal-transport inference framework of Li and Henry (2025). We apply it to retail distribution choices in carbonated beverages, tea, and yogurt, where firms decide whether to enter and if so, which products to sell in each local market. Using firms’ product-choice decisions, we derive product-level bounds on incremental fixed costs under weak behavioral assumptions. These bounds allow us to estimate a fixed-cost function that accommodates economies or diseconomies of scope, so that the cost of adding a product may depend on a firm’s product portfolio. We find evidence for economies of scope, where the incremental fixed cost of adding a product is lower when the firm sells more products in the same local market. Counterfactual simulations show that these scope economies have important implications. First, policies that increase a market’s size raise consumer surplus primarily through the increase of incumbent firms’ product variety rather than through entry by new firms. Second, counterfactual mergers among small owners lead to fixed-cost savings, which raise product variety and consumer surplus.

Public support for electric vehicles has long focused on purchase subsidies, and federal funding is now shifting toward the charging network, yet the welfare return on charging subsidies is rarely measured. I evaluate the first round of the National Electric Vehicle Infrastructure (NEVI) program in Texas, which commits $31.8 million to 50 new charging stations along interstate corridors. Using ZIP-code-level registrations of each EV model from 2017 to 2023, I estimate a random-coefficients demand model, disciplined by micro moments on EV buyers’ income and homeownership, in which local charging availability enters the value of each model. Most models record no sales in most ZIP-code-quarters; rather than dropping these zeros or aggregating to the state level, I bound demand with the moment inequalities of Gandhi et al. (2023) and construct confidence sets following Chernozhukov et al. (2007). Simulating a year with and without the 50 stations, I find expected annual welfare gains, summing consumer surplus, vehicle producer profits, and emission reductions, of about $15.2 million against $31.8 million in fixed costs, implying a payback period of roughly two years.

I develop a nested two-stage demand framework for viewer engagement and model streamer learning, effort, and exit, to quantify the welfare contribution of marginal streamers and evaluate counterfactual monetization, retention, and merger scenarios.