Date of Award

December 2014

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Economics

Advisor(s)

William C. Horrace

Keywords

Econometrics, Energy Economics, Extreme Value Theory, Nonparametric Estimation, Spatial Econometrics

Subject Categories

Social and Behavioral Sciences

Abstract

Abstract

Chapter 1

This essay examines the effect of state renewable energy policies in inducing innovation and the spillover effect of these policies on innovation in neighboring states. The analysis is conducted with patent data related to renewable technology using wind power for the United States over the period 1983-2010. We run a panel data regression of a log transformation of states' yearly patent counts on state renewable energy policies and spatially weighted average of renewable energy policies in neighboring states using the Tobit model with individual effects. The results show that renewable energy rules, regulation and mandates such as interconnection standards, net metering and renewable portfolio standard enacted in neighboring states have shown a statistically significant positive spillover effect in increasing the number of patent applications in that state. However, financial policies such as tax incentives and subsidy policies implemented by neighboring states have shown statistically significant negative effects on technological innovation within that state.

Chapter 2

In this essay, we have conducted a Monte Carlo Study of the prediction performance of various nonparametric estimation methods for spatially dependent data, such as the nonparametric local linear kernel estimator, the Nadraya-Watson estimator, and the k-Nearest Neighbors method developed by Hallin et al. (2004b), Lu and Chen (2002), P.M. Robinson (2011) and Li and Tran (2009). With data sampled on a rectangular grid in a nonlinear random field, the results show that nonparametric local linear kernel method has the best performance in terms of mean squared prediction error. The Nadaraya-Watson estimation method also performs well. In general, these two nonparametric methods consistently outperform the k-Nearest Neighbors method and the maximum likelihood method regardless of the data generating process and sample size. However, the maximum likelihood method does not perform well because the spatial weight matrix can only be used to estimate linear structures while the true data generating process is nonlinear. This also gives some support to the idea of using nonparametric methods when various misspecification may exist either in the functional form or spatial weight matrix for spatially dependent data.

We use these methods to predict county-level crop yields with spatially weighted precipitation. The results are generally consistent with the simulation results. The nonparametric local linear kernel estimator has the best prediction performance. The Nadaraya-Watson estimator also performs better than the k-Nearest Neighbors method and the maximum likelihood estimator. However, with an inverse distance weighting matrix, the maximum likelihood estimator outperforms the k-Nearest Neighbors method in predicting crop yield.

Chapter 3

This essay uses the "exceedances over high threshold model" of Davidson and Smith (1990) to investigate the univariate tail distribution of the returns on various energy products such as Crude Oil, Gasoline, Heating Oil, Propane and Diesel. The bivariate threshold exceedance model of Ledford and Tawn (1996) is also used to study the tail dependence between returns on various pairs of selected energy products. Tail index estimates for univariate threshold exceedance models show that these returns generally have fat tails similar to those of a Student's t-Distribution with 2 to 5 degrees of freedom except that for Crude Oil where the tail index estimates are closer to that of a normal distribution. We also estimate the tail dependence index for four pairs of energy products, crude oil/gasoline, crude oil/heating oil, crude oil/propane, crude oil/diese. The correlation coefficients implied by the dependence index estimates show that correlations conditional on threshold exceedance are generally higher than the unconditional correlation between crude oil/heating oil and crude oil/gasoline. However, there is some variation in the implied correlation between crude oil/propane and crude oil/diesel. Whether the extreme correlation will be higher or lower than the unconditional correlation depends on the threshold chosen.

Access

Open Access

Download

Share

COinS