Job Market Paper
Distance affects the cost of moving goods and people across space. Traditional spatial economics models explore this relationship, but recent evidence suggests that geography also affects the flow of information. To investigate this hypothesis, I study the causes of brand sales growth over time and space. I analyze data from a large set of branded retail products sold in different regions in the United States and document a series of stylized facts about their life-cycle. I find that brands typically sell to a small number of locations that tend to be geographically close, and growth usually happens around previously successful markets. Furthermore, I decompose sales into three components: customer base, prices, and quantities per customer. Almost all of the variation in brand sales, both across locations and over time, comes from the first term. The evidence suggests that geography plays a vital role in customer acquisition, but not due to differences in prices. Motivated by these findings, I propose a model in which information about brands' existence spreads geographically, similarly to how contagious diseases spread. Consumers aware of a brand might 'infect' others with that knowledge, and the probability of contagion depends on their location. Additionally, brands have different costs to deliver their goods to different markets. I use the predictions for the correlation of brand sales and customer base across regions to estimate the model using Simulated Methods of Moments and find that information frictions are more severe between distant locations. Furthermore, eliminating the role of distance in contagion increases consumer welfare by 32.5%. These results highlight the importance of geography for the spread of information about brands. This relationship allows for the description of brand dynamics across space and has significant welfare implications.
Internal Gravity and Customer Base
How does distance affect the sales of brands in the United States? To answer that question, I use brand data across 44 different regions in the US. At the brand level, most of the effect of distance on sales is associated with a reduced number of customers in distant locations rather than sales per customer. At the aggregate level, most sales reduction comes from having fewer brands serving other areas. To understand what drives these patterns, I introduce a trade model between regions with shipping costs and search frictions between brands and final consumers. The preliminary estimates suggest that the shipping costs are a log-linear function of distance. However, information frictions are lower at the origin than at other destinations but are not affected by distance otherwise.
Geographic Spillovers and Exporter’s Growth
Does selling to a country increase the demand for a firm's product in neighboring locations? To answer this question, I study the post-entry sales of Brazilian exporters in different markets. The border-effect is associated with higher initial sales, higher growth, and lower probability of exit, controlling for firm-year fixed effects. Even accounting for the spell-length of exporting events, which serves as a control for unobserved demand heterogeneity, I find that selling to a bordering country in the previous period increases sales by more than 10% at any given age. These results suggest that the persistent geographic spillovers on exporters' sales are more likely due to increasing demand than to lower fixed costs.
Work in Progress
A Method for Estimating Bounds on Switching Costs Using Intertemporal Moment Inequalities (with Vitoria R. de Castro)
We present a method to estimate fixed and sunk costs using moment inequalities in random utility models. This approach avoids error specification assumptions in dynamic discrete choice models where the number of alternatives is small, but the set of static choices and error dimensionality are large. Moments are constructed using revealed preference conditions on intertemporal decisions. These conditions impose bounds on continuation values. Crucially, they also impose restrictions on the set of utility shocks consistent with choices in each state. We combine these restrictions with conditional choice probabilities from frequencies in the data to generate intertemporal moment conditions for the estimation of switching and fixed costs.