Econometric Modeling for Functional-Coefficient VAR Models: Theories and Applications

Abstract

This dissertation proposes theories and applications for three new types of functional-coefficient VAR models. The first part of dissertation develops a vector autoregressive model for conditional quantiles with functional coefficients to construct a novel class of nonparametric dynamic network systems, of which the interdependences among tail risks such as Value-at-Risk are allowed to vary smoothly with a variable of general economy. The contributions to literature are four-fold in this part. First, the model setting is general enough to nest many well-known dynamic quantile models in the literature. Second, by allowing coefficients to vary with a smoothing variable, the proposed model provides a new tool to estimate the relationship between the interdependence of risk and the state variable of economy or time. Third, a new and simple-to-implement estimation procedure is developed for estimating the proposed quantile model with highly nonlinear structure and latent covariates. Finally, a large sample theory for the proposed estimator is established to construct confidence intervals for functional coefficients in the empirical study. The second part proposes a new class of functional-coefficient factor-augmented predictive VAR (FC-FAVAR) models. Different from the existing literature, this model setting allows both factor loadings of corresponding factor model and coefficients of this predictive VAR model vary with a smoothing economic variable, which adds additional information of variation in the factor structure and economic interpretability to the predictive model. Moreover, both observed variables and unobserved factor regressors in this new model are jointly imposed in a vector autoregressive form. In this way, some important information of model dynamic may be included in these lagged factors, which is helpful to enhance the ability of prediction. Finally, the proposed model is applied in both simulation and empirical study of one-step ahead prediction, which demonstrate its reliability in forecasting. In the third part, effects of monetary policy shocks on large amounts of macroeconomic variables are identified by a class of FC-FAVAR models. In the empirical study, I analyze the generalized impulse response functions (GIRF) estimated by the newly proposed model and compare my results with those from classical FAVAR models. The major contributions are two parts. In the empirical study, I provide an alternative way from econometric perspective to reduce price puzzle by using the proposed FC-FAVAR model, without introducing new variables or structure in the conventional macroeconomic model and replacing policy instruments.