Three papers in regional dynamics and panel econometrics
This dissertation includes three chapters that cover broad topics in economics. The first chapter explores how the US Government's Capital Purchase Program, a large capital injection to local and regional banks through a stock purchase agreement, impacted local establishment dynamics such as entry, exit, employment expansion, and employment contraction following the 2008 Financial Crisis. The Capital Purchase Program dispersed over \$200 billion dollars to banks hoping to prevent failure and ease tightened lending conditions. I estimate the direct effects of a county having a bank receive Capital Purchase Program funds on local business dynamics in the seven years following treatment, as well as spillover effects as entrepreneurs and business in neighboring regions travel to gain access to credit. Estimates show the CPP had no effect on establishment entry and exit, nor employment expansion and contraction. This paper establishes that the business-lending aims of the CPP were not realized in the communities and regions that received funds, and casts further doubt on meaningful pass through of CPP funds to desirable local economic activity.
The second chapter develops a joint hypothesis centered Wald test over fixed effects in large N small T panel data models with symmetric serial correlation within cross sectional observations. The enables joint hypothesis tests over inconsistently estimated fixed effects, such as the traditional varying intercept model as well as models with individual specific slope coefficients. I establish two different set of assumptions where feasible tests exist. The first assumption requires that individual errors follow a stationary $\ARp$ process. Under this assumption all second and fourth cross product moments can be consistently estimated while allowing for individual specific hypothesis and covariates to vary across individuals and time with individual specific slopes. The second feasible test requires individuals to have coefficient slopes that are shared among all individuals in a known grouping structure under the null. This set of assumptions enables estimation of a completely unconstrained variance-covariance matrix and higher cross product moments for individuals. Examples of these tests arise in wanting to establish latent panel structure, such as unobserved grouping of individuals, wanting to compare different models of teacher or firm value added against each other, or testing whether or not fixed effects can be approximated by Mundlak-Chamberlain devices.
Finally, the third paper estimates how messages displaced on Dynamic Message Boards, large signs either adjacent to or displayed above roads, impact near to sign accidents. In this research, I look at the traffic-related messages such as ``drive sober,'' ``x deaths on roads this year,'' and ``click it or ticket,'' displayed on major highways, on reported near-to-sign traffic accidents. This provides estimates of the impact of different types of nudges on road safety behavior. To estimate the causal effect of these nudges, we build a new high-frequency panel data set using the information on the time and location of messages, crashes, overall traffic levels, and weather conditions using the data of the state of Vermont over a three year time period. I estimate models that control for endogeneity of displayed messages, or allow for spillover effects from neighboring messages.