This article discusses dynamic analyses using the vector autoregressive (VAR) model for
mixed-frequency data. The model estimation is achieved by representing the original model
just with current and lagged observable variables. Such representation is accomplished
through recursive substitution of unobservable variables with lagged observable variables.
The consistent estimation of model parameters is facilitated by the classical minimum
distance estimation that uses lagged variables as instruments. Conventional dynamic
analyses, which include forecasting with the VAR model, are possible after model
estimation. The proposed method differs from other approaches in three aspects. First, unlike
a Bayesian approach, the proposed classical method does not require any specific prior
distribution of coefficients. Second, an “explicit” identification condition is suggested for the
model. Finally, the proposed method can estimate the error variance consistently, which is
critical for dynamic analyses.