Nonstationary Time-Series Modeling versus Structural Equation Modeling: With an Application to Japanese Money Demand

The issues of identification, estimation, and statistical inferences of nonstationary time series and simultaneous equation models are reviewed. It is shown that prior information matters and the advantage of dichotomization of the traditional autoregressive distributed lag model into the long-run equilibrium relation and the short-run dynamic adjustment process as an empirical modeling device may be exaggerated. A Japanese money demand study is used to illustrate that a direct approach yields a more stable long-run and short-run relationship and has better predictive power than the approach of letting the data determine the long-run relationship and modeling the short-run dynamics as an adjustment of the deviation from its equilibrium position.