Three Empirical Essays in Public Economics
Loading...
Authors
Chicola, Randall M.
Issue Date
2024
Type
Dissertation
Language
Keywords
Almost Ideal Demand , Keras , Machine Learning , Neural Networks , Sequential Deep Learning , Web Scrapping
Alternative Title
Abstract
This dissertation uses applied microeconometrics methods to investigate research questions in public economics (first two chapters) and to explore the use of deep learning to construct data useful in public economics and other fields (third chapter). The second and third chapters focus on Nevada. The first essay examines the elasticity response of consumers to the imposition of online sales taxes. The second essay examines the effect of the Nevadan labor force on Nevada county budgets with a particular focus on employees who commute out of their place of residence to work. The third essay uses Clark County Nevada home and property attribute data to infer precinct-level income. The first paper: Implications of Sales Tax Enforcement on E-commerce: Evidence from Nielsen Consumer Panel Data is a solo paper that investigates how sensitive are ecommerce purchases to state sales taxes. It accomplishes this by aggregating Nielsen Consumer Panel data to determine participating panelists' expenditure shares in each panel year and applying a structural regression model underpinned by microeconomic foundations. Sales taxes were not always applied to e-commerce until court cases settled the legal notion of physical nexus . As this legal issue was resolved, states started adopting online sales tax policies for e-commerce at different times. This paper exploits timing variation in the imposition of state sales tax onto online purchases to estimate demand elasticities for online and traditional brick-and-mortar retail shopping. The large elasticities imply that collecting sales taxes from online retailers partially shifts consumption back to brick-and-mortar retailers. The policy effects for sales taxes in Nevada are then compared to other Western states. The second paper: Joint Prediction and Simulation of Labor Force and Fiscal Conditions of Nevada Counties uses Nevada county labor and budget data to examine the dynamic relationships between workers who live in their place of work employment and those who commute outside their county of residence for work to determine the fiscal impact on Nevada county budgets. After the model is specified through a system of equations with interlinked variables for the estimation process, a simulation is performed to assess the impact of changes in exogenous employment on the labor and fiscal status of each Nevada county. The third chapter: Constructing Precinct Level Income Variables Using Deep Learning uses property and home feature data scrapped from the Clark County Assessor's office as a basis for a neural network model that predicts income shares at the precinct level. The contribution of this work is a method for training a neural network model at a given level of geographic granularity, Census blocks, to extrapolate income share estimates at a different granularity at the precinct level. This method is then compared to standard Ordinary Least Squares (OLS) regression by comparing the mean squared error (MSE) of the respective prediction methods.
Description
Citation
Publisher
License
Creative Commons Attribution 4.0 International