Canadian-American Macroeconomic Interactions: A Sensitivity Analysis Based on Alternative Structural Interpretations of VAR Models

Abstract

<p>A number of recent studies have taken a comprehensive view of Canadian-American macroeconomic interactions and have employed vector autoregressive (VAR) modeling techniques to explore the impact of American events on the Canadian economy. The main reason behind the we of the VAR framework is that it requires a minimal set of restrictions, as opposed to traditional econometric practice in which models are generally overidentified. The VAR modeling is highly helpful in exposing more fully the dynamic relationships among model variables. Although the use of VAR modeling may be considered preferable from some points of view, it has been criticized for its atheoretical nature. The critics have argued that any multi-equation model must be provided with set of economically meaningful identifying restriction if purposeful interpretation is required. In the standard VAR framework, identification is achieved simply by imposing the Choleski decomposition on the error covariance matrix. This implies that the underlying economic structure is strictly recursive. Such an assumption may be considered inappropriate since there is no reason that structuraI recursivity should be a prerequisite for the identification of all VAR moles. Much of the evidence from Canadian VAR studies is subject to that limitation. Concern about the possible inappropriateness of the recursivity assumption has led recently to a number of structural VAR studies. These studies remove the neccessity of recursivity and permit the (ezact) identification and estimation of contemporaneous structural relationships based on economic theory. The present study relies on structural VAR modeling and re-examines primarily the contribution of American economic fluctuations to the variability of Canadian economic variables. A variety of representative structures (including neoclassical and Keynesian-type structures) are imposed alternatively on a seven-equation Canadian-American VAR system and their implications for model interpretation are considered. The VAR system includes Canadian and American output, price levels, and money stocks, us well as the exchange rate. In order to capture the effects of the energy shocks of the 1970s the world oil price is also included, as an exogenous variable. The data set is quarterly and seasonally unadjusted. It covers the period 1970:2 to 1987:4 (1970 was the first year in recent times in which the Canadian exchange rate was allowed to float against other currencies.) It turns out that the size and significance of structural estimates are highly sensitive to the ways in which one identifies a model. The sensitivity of the estimates is more prominent in non-recursive, contemporaneous structures. Impulse response functions and variance decompositions are also affected by the choice of identifying priors. Again, the sensitivity of impulse response functions and variance decompositions is more obvious when their calculation is based on alternative non-recursive, contemporaneous structures.</p>