| dc.description.abstract | This dissertation consists of three essays. Chapter II examines the possible
asymmetric response of gasoline prices to crude oil price changes using an error
correction model with GARCH errors. Recent papers have looked at this issue. Some of
these papers estimate a form of error correction model, but none of them accounts for
autoregressive heteroskedasticity in estimation and testing for asymmetry and none of
them takes the response of crude oil price into consideration. We find that time-varying
volatility of gasoline price disturbances is an important feature of the data, and when we
allow for asymmetric GARCH errors and investigate the system wide impulse response
function, we find evidence of asymmetric adjustment to crude oil price changes in
weekly retail gasoline prices
Chapter III discusses the relationship between fiscal deficit and exchange rate.
Economic theory predicts that fiscal deficits can significantly affect real exchange rate
movements, but existing empirical evidence reports only a weak impact of fiscal deficits
on exchange rates. Based on US dollar-based real exchange rates in G5 countries and a
flexible varying coefficient model, we show that the previously documented weak relationship between fiscal deficits and exchange rates may be the result of additive
specifications, and that the relationship is stronger if we allow fiscal deficits to impact
real exchange rates non-additively as well as nonlinearly. We find that the speed of
exchange rate adjustment toward equilibrium depends on the state of the fiscal deficit; a
fiscal contraction in the US can lead to less persistence in the deviation of exchange rates
from fundamentals, and faster mean reversion to the equilibrium.
Chapter IV proposes a kernel method to deal with the nonparametric regression
model with only discrete covariates as regressors. This new approach is based on
recently developed least squares cross-validation kernel smoothing method. It can not
only automatically smooth the irrelevant variables out of the nonparametric regression
model, but also avoid the problem of loss of efficiency related to the traditional
nonparametric frequency-based method and the problem of misspecification based on
parametric model. | en |
