پیش بینی اقتصاد باز با یک DSGE-VAR : رقابت شانه به شانه با پیش بینی های منتشر شده RBNZ

We construct a DSGE-VAR model for competing head to head with the long history of published forecasts of the Reserve Bank of New Zealand. We also construct a Bayesian VAR model with a Minnesota prior for forecast comparison. The DSGE-VAR model combines a structural DSGE model with a statistical VAR model based on the in-sample fit over the majority of New Zealand’s inflation-targeting period. We evaluate the real-time out-of-sample forecasting performance of the DSGE-VAR model, and show that the forecasts from the DSGE-VAR are competitive with the Reserve Bank of New Zealand’s published, judgmentally-adjusted forecasts. The Bayesian VAR model with a Minnesota prior also provides a competitive forecasting performance, and generally, with a few exceptions, out-performs both the DSGE-VAR and the Reserve Bank’s own forecasts.

Combining models has been demonstrated to improve forecasts in a number of contexts (see for example Elliott and Timmermann, 2005 and Goodwin, 2000; and Hall & Mitchell, 2007). Often this combination has been restricted to purely statistical models, rather than models developed from either microeconomic or macroeconomic theory. At the same time, policymakers often want structural models to assess alternative policies in the light of the Lucas critique, which stresses the dependence of reduced form parameters on control parameters set by policymakers. Del Negro and Schorfheide (2004) show how a structural Dynamic Stochastic General Equilibrium (DSGE) model can be combined with a vector autoregression (VAR) to provide a hybrid, DSGE-VAR model that forecasts well and provides structure that policymakers can use to evaluate alternative policies. While Bayesian VARs utilise time series priors to help improve the forecasting performances of unrestricted VARs, the DSGE-VAR utilises macroeconomic theory to provide the priors. Equally, including the VAR component in the hybrid model helps to reduce the potential misspecification imposed on the data by the DSGE model. In this paper we apply the DSGE-VAR methodology to New Zealand–a small, open economy with an inflation-targeting central bank. We estimate the five-variable DSGE model developed by Lubik and Schorfheide (2007) over the majority of the inflation targeting history of the Reserve Bank of New Zealand, since New Zealand has the longest such history of any explicit inflation targeter. The Lubik and Schorfheide (2007) DSGE model represents a minimal set of DSGE theory to apply to the data, and our VAR, based on the set of observables implied by the DSGE model, acts to mitigate any potential misspecification. Since the Reserve Bank’s many published forecasts over time are predicated on endogenous policy, they provide a unique benchmark amongst explicit inflation targeters against which to compare our DSGE-VAR forecasts. Because these forecasts are free to condition on any information set deemed relevant by the Reserve Bank (such as high frequency financial data, survey data, anecdotal evidence, institutional knowledge, or simply policymaker beliefs), these forecasts should set a relatively high benchmark for the DSGE-VAR, compared to, say, a random-walk, or the simple single-equation forecasting models that are frequently used as points of comparison for macroeconomic forecasting. The rest of the paper is organised as follows. Section 2 discusses the DSGE-VAR technology and outlines the Del Negro-Schorfheide algorithm we adopt as our estimation procedure. Section 3 outlines the Lubik and Schorfheide model, our parameter estimates, and the impulse responses implied by the model. Section 4 compares the out-of-sample forecasts of the DSGE-VAR to the official forecasts of the Reserve Bank of New Zealand. Concluding comments are made in Section 5.

This paper shows the benefits of utilizing DSGE-VAR models for forecasting in the open economy context. A VAR informed by Lubik and Schorfheide’s (2007) small open economy model produces forecasts which are comparable with, or, in the case of output growth, superior to, the Reserve Bank of New Zealand’s published judgement-adjusted forecasts. In addition, the forecasting performance of the DSGE model with no VAR correction is competitive. At longer horizons, where monetary policy is conventionally thought to have its greatest influence, both the DSGE-VAR and the DSGE model outperform the forecasts given in the Reserve Bank of New Zealand’s Monetary Policy Statement. The DSGE and DSGE-VAR models approach, but do not attain, the performance of the Bayesian VAR with the Minnesota prior. However, the DSGE-VAR is informative about the structure of the economy. The DSGE structure should be robust to the Lucas critique; a policymaker who fears the Lucas critique can work with a model which places a higher weight on the DSGE than is suggested by the data. We believe that the DSGE-VAR is a useful modelling technology for central banks. In addition to the competitive forecasting performance reported by Del Negro and Schorfheide (2004) for the U.S., the DSGE-VAR produces a good forecasting performance for New Zealand, a small open economy with an explicit inflation-targeting central bank. The ability to forecast well and yet obtain the economic structure suggests that the DSGE-VAR may also be a useful forecasting and policy analysis tool for other central banks.