In Essay 1, we quantitatively investigate the determinants of capital structure at the industry level using a structural market equilibrium model. We analytically derive the unique equilibrium in which firms’ financing and production decisions as well as the market structure are endogenously determined. We then estimate the model industry by industry. Median industry leverage decreases with the product substitutability, increases with the fixed cost of production, decreases with the capital and labor intensities, and decreases with the idiosyncratic and persistent components of productivity risk. Quantitatively, the elasticity of industry leverage with respect to the fixed cost of production is greater than that of the product substitutability; the impact of the labor intensity is larger than that of the capital intensity; and the effects of idiosyncratic productivity risk dominate those of the persistent productivity risk. The proportional benefits and costs of debt, which play key roles in determining leverage, have relatively modest quantitative impacts. We also lay indicate a direction for future work where we intend to study how managerial characteristics in tandem with industry characteristics affect industry and firm leverage.

In Essay 2, we aim to predict urban household living standards (HLS) at a granular spatial and temporal resolution in years where traditional survey or census data is unavailable. Our approach leverages over two decades of confidential data on Household Living Standards in Mexico with data sources traditionally underutilized in economic or risk research, such as satellite imagery from the Landsat program. Using this unique dataset, we train neural networks composed of convolutional neural networks and long and short-term memory networks as well as convolutional neural networks and transformers that allow us to capture the spatial and temporal changes from the images and generate HLS predictions with an accuracy of over 52 percent. The resulting predicted living standard dataset at the granular urban primary sampling unit level and yearly frequency allows us to study several questions in Mexico, including the evolution of the concentration of assets in areas at high risk from natural disasters and the socioeconomic impact of extreme weather events. Our method is easily replicable and holds the potential to inform the targeting of social programs and the study of questions related to the evolution of HLS in developing economies where survey data is unavailable or infrequent.