Abstract
Vector autoregressions are popular data generating models in applied econometrics. They are frequently used to examine forecasts, to study the sources and the characteristics of economic fluctuations, and to test hypotheses which involve rational expectations. This dissertation studies vector auoregressions subject to cointegration or steady state restrictions. A theory is developed to analyze, e.g., how important growth innovations are at the business cycle horizon. The question whether possible nonstationarity (cointegration) needs to be accounted for prior to conducting inference from tests of economic theory constraints is also addressed. A simple likelihood ratio test is proposed to examine cointegration and cross equation restrictions jointly. The major theoretical results are examplified by data analysis of simple growth and term structure models.