Using subspace algorithm cointegration analysis for structural estimation

Structural vector autoregression (SVAR) is often used as empirical instrument in order to examine business cycle fluctuations. Its popularity caused a comprehensive controversy in the literature. In this paper, we want to contribute to this discussion and introduce an alternative structural estimation method. Since we focus on the context of cointegration, our approach should be viewed as alternative to standard methods in this area, e.g. structural vector error correction models (SVECMs). We pursue the subspace cointegration analysis by Bauer and Wagner (2002, 2003, 2009) in using their subspace algorithm for structural estimation. Thereby, we follow a recent study by Kascha and Mertens (2009) in comparing our structural estimation approach to its standard counterpart in a Monte Carlo simulation. We analyze the estimated impulse responses and structural shocks and relate them to their true complements in terms of correlation and mean squared error. The results illustrate that our procedure is a serious option to structural estimation. However, our findings depend on the underlying model parametrization but stress the fact that it can dramatically outperform SVECM because of better small sample properties. (JEL E32, C15, C52)