Essays On Methods Of Demand And Production Function Estimation

Ruizhi Ma, University of Pennsylvania


Estimating consumer demand is fundamental to analyzing pricing decisions, welfare gains from new products, and changes in market structure. The first two chapters of this dissertation analyze econometric methods that allow researchers to estimate richer distributions of heterogeneity, and therefore more flexible demand models.The first chapter compares several newly developed methods for estimating individual heterogeneity by Monte Carlo simulations. I use them to estimate a simplified mixed logit model without price endogeneity. I find the method from Malone et al. (2019) achieves good performances with a significantly lower computational cost, and the method from Cheng et al. (2019) is well-suited for scenarios with sparse type interactions. I provide some recommendations on how an empirical researcher could use these methods in practice. The second chapter extends the first chapter by adapting the fixed-grid likelihood method (from Malone et al. 2019, henceforth FG) and a clustering method (from Cheng et al. 2019, henceforth CSS) to the full mixed logit model with price endogeneity and unobserved consumer demographics. I compare FG, CSS, and a conventional parametric MLE procedure using real panel data. I compare their predictions on welfare estimates in three hypothetical scenarios: a new product, a merger, and a divestiture. I find that the parametric approach distorts the welfare predictions. The third chapter estimates how weather affects Chinese manufacturers' productivity during 1998-2007 and predicts how climate change would affect their productivity by 2040-2042. We use Abito (2020)’s production function estimation method, allowing firm-specific fixed effects in productivity. We find most industries in our sample exhibit persistent differences in firm-level productivity, and that weather significantly affects a firm’s productivity. We use the estimated effects to predict the mean productivity level in 2040-2042 for each firm. Comparing to the firms’ historical mean productivity levels, we find the productivity will be lowered by about -4% by 2040-2042 on average across firms. We also find several industries where Ackerberg et al. (2015)’s productivity estimates induce significantly biased climate predictions. These industries amount to a large portion of the real value-added output, and are where the productivity heterogeneity is the largest and most persistent.