Nonparametric Methods in Economics and Finance: Dependence, Causality and Prediction

PhD Thesis# 386
Dr. Valentyn (V.) Panchenko
Prof. dr. C.H. Hommes, Dr. C.G.H. Diks


In this thesis we investigate a number of issues related to nonparametric and semiparametric methods in financial econometrics. Compared to parametric methods, nonparametric and semiparametric methods require fewer assumptions on distributions and/or functional forms and allow for more flexibility in financial modelling and prediction. The thesis is divided into three major themes: dependence, causality and prediction. First, we develop nonparametric test for serial independence using quadratic forms and multiple bandwidths. Further, we use the concept of copulas to characterise cross-sectional dependence between asset returns and suggest a test for parametric copulas. In the subsequent chapters we discuss nonparametric testing for Granger non-causality. Next, we develop a semiparametric procedure which uses a parametric conditional copula and a nonparametric marginal. This procedure may be used for predicting joint distributions of future asset returns.

Publisher of the TI-theses is: Rozenberg Publishing Services