We discuss the Markov-switching vector autoregressive (MS-VAR) class of nonlinear time series models that can be used to analyze recurring discrete structural changes in time series. Hamilton's (1989) seminal Markovswitching (MS) model of the U.S. business cycle triggered considerable interest in the MS approach in economics. Most empirical applications to date have focused on the business cycle and financial markets, but we see potential for MS-VAR models in agricultural economics, for example, in price transmission analysis. In the following, we first provide an overview of the MS-VAR framework. We then present an illustrative application to maize price transmission between Tanzania and Kenya. The article closes with a discussion of strengths, weaknesses, and potential uses of the MS-VAR approach in price transmission analysis.
A Brief Overview of MS-VAR ModelsFollowing Krolzig (1997), the basic idea behind the MS-VAR class of models is that the parameters of a VAR process are allowed to depend on an unobserved regime variable s t ∈ {1, . . . , M}, representing M possible states of the world. In its most general form, the MS-VAR is given bywhere y t = (y 1t , . . . , y Kt ) , t = 1, . . . , T is a Kdimensional time series vector, (s t ) and A j (s t ) are matrices of intercepts and autoregressive (AR) parameters of appropriate dimension, and (s t ) is the variance-covariance matrix of a Gaussian zero-mean error process u t . In (1) the terms (s t ), A j (s t ), and (s t ) describe the dependence of the respective parameters on the unobserved regime s t . The intercept in (1), for example, will be regime-dependent as follows:Hence, the nonlinear data generating process in (1) can be described as piecewise linear (i.e., as linear conditional on the regimes).Depending on which parameters in (1) are allowed to be regime-dependent, different subclasses of the MS-VAR result. Krolzig (1997) proposes the terminology MSx(M)-VAR(p) to distinguish between them, where M is the number of regimes, p the order of the VAR, and x indicates which parameters are regime-dependent. Thus, a MSI(M)-VAR(p) refers to a model in which the intercept is regime-dependent, while MSA(M)-VAR(p) and MSH(M)-VAR(p) refer to regime-dependence in the AR terms A j (s t ) and the error covariance (s t ), respectively. These elements can be combined, so that the "full-blown" model in (1) can be considered a MSIAH(M)-VAR(p).A further sub-class of MS-VAR models that is not immediately apparent from (1) is the MSM(M)-VAR(p) in equation (3) y t − (s t ) = p j=1 A j (y t− j − (s t− j )) + u t ,