VAR Analysis of Heterogeneous Panel Data
Cumulative thesis
Date of Examination:2025-01-29
Date of issue:2025-10-02
Advisor:Prof. Dr. Helmut Herwartz
Referee:Prof. Dr. Helmut Herwartz
Referee:Prof. Dr. Thomas Kneib
Referee:Prof. Dr. Simone Maxand
Files in this item
Name:Empting_Thesis.pdf
Size:1.23Mb
Format:PDF
Abstract
English
Vector-autoregressive (VAR) analysis is a central instrument of empirical macroeconomics and the starting point for numerous extensions of multivariate time series analysis. For example, structural VAR models allow to identify causal relationships in dynamic systems and thus to quantify the effects of political interventions. Panel VAR models combine time series of many survey units such as countries to increase the dataset and thus achieve more precise estimates. The three papers in this the dissertation draw on this toolkit. The first paper introduces the R-package pvars. The aim of the package is to provide a seamless implementation of VAR methods for heterogeneous panel data, thus filling a gap in the available software packages. In particular, pvars provides panel cointegration rank tests that take into account cross-sectional dependencies in the error terms and structural breaks in the deterministic terms. Based on these specifications, panel SVAR models with pooled cointegration vectors can be estimated and identified by different panel methods. The paper’s contribution lies in presenting the methodology, its modular implementation in R, and the practical application: Two empirical examples from the literature are reproduced step by step and help users to get started and perform their own analyses. The second paper revisits the long-lasting debate about whether public investment – for example in the context of the Green New Deal – promotes economic growth. The aim is to systematically revise the extensive empirical literature using a new panel dataset covering 23 OECD countries from 1960 to 2019. To this end, a heterogeneous panel VAR model is employed, which combines the dynamic panel model and the vector error correction model as two previous approaches. The empirical results show that two homogeneous long-run relationships exist within the panel and that the declining productivity over time is a key factor in explaining the puzzling and often insignificant impulse response functions at the country level. The third paper analyzes the armament dynamics in the Arab-Israeli conflict between 1962 and 2010 using SVAR models. In contrast to earlier work that relied on Granger causality in reduced-form VAR models, this study takes a data-driven identification approach. Using independent component analysis, two types of shock are identified: country-specific and regional geopolitical tensions. The latter are shown to be more relevant for explaining the armament dynamics in Israel, Egypt, Jordan and Syria. Particularly for the rivalry between Israel and Egypt, reciprocal reactions and mutual reinforcements are found, which reflect typical patterns of an arms race.
Keywords: panel vector autoregressive models; structural vector autoregressive models; pooled cointegrating vectors; independent components; panel cointegration rank tests; public capital; arms race
