Causality in structural vector autoregressions: Science or sorcery?

This paper presents the structural vector autoregression (SVAR) as a method for estimating dynamic causal effects in agricultural and resource economics. We have a pedagogical purpose; we aim the presentation at economists trained primarily in microeconometrics. The SVAR is a model of a system, whereas a reduced-form microeconometric study aims to estimate the causal effect of one variable on another. The system approach produces estimates of a complete set of causal relationships among the variables, but it requires strong assumptions to do so. We explain these assumptions and describe similarities and differences with the classical instrumental variables (IV) model. We demonstrate that the population analogue of the Wald IV estimator for a particular causal effect is identical to the ratio of two impulse responses from an SVAR. We further demonstrate that incorrect identification assumptions about some components of the SVAR do not necessarily invalidate the estimated causal effects of other components. We present an SVAR analysis of global supply and demand for agricultural commodities, which was previously examined using IV. We illustrate the additional economic insights that the SVAR reveals, and we articulate the additional assumptions upon which those insights rest.