TitleEssays on Bayesian analysis of financial economics
NameLi, Liuling (author), Tsurumi, Hiroki (chair), Swanson, Norman (internal member), Mizrach, Bruce (internal member), Goldman, Elena (outside member), Rutgers University, Graduate School - New Brunswick,
Bayesian statistical decision theory,
DescriptionThis dissertation consists of three essays with each essay forming a chapter. The regression models in these three chapters are different but share the same feature: the error terms of the models all follow ARMA-GARCH error processes generated either from normal or exponential power distributions.
In the first chapter I present a spot asset pricing model that is known as the CKLS model. Two CKLS models are compared. In one model the ARMA-GARCH error process is generated by the exponential power distribution while in the other model the error process is generated by the normal distribution. Using monthly U.S. federal funds rate I estimate the parameters of the CKLS models. From the predictive densities I obtain the distributions of the mean squared errors of forecast (MSEF) and the predictive deviance information criterion (PDIC). In addition I use the Bayes factor and the deviance information criterion (DIC). Markov Chain Monte Carlo (MCMC) algorithms, which are stochastic numerical integration methods, are used. I find that in general the CKLS model with the error term generated by the exponential power distribution is chosen over the model with the normal error term.
In the second chapter I first compare two MCMC algorithms: random walk draw and non-random walk draw for a Markov switching regression model. Two Markov switching models are compared: one with the variance of the normal distribution generated by the state space variable and the other with the constant variance. The realized volatilities of MMM Company are used to estimate and compare the models. The mean squared errors (MSE) and mean squared errors of forecast (MSEF) are used as the model selection criteria. I find that the model with the constant variance is chosen over the model with the state space variance by the MSE but the latter is chosen over the former by the MSEF.
In the third chapter I estimate a bivariate copula model. Each of the two regressions is generated by the exponential power distribution. I use monthly data on SP500 and FTSE100. Results show that the correlation parameter for SP500 and FTSE100 is .6893.
NoteIncludes bibliographical references (p. 86-96)
Noteby Liuling Li
CollectionGraduate School - New Brunswick Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.