RDP 2013-07: An Empirical BVAR-DSGE Model of the Australian Economy 1. Introduction

An important aspect of the Australian economy recently has been the divergence in growth between the commodity and non-commodity sectors, reflecting strong demand for commodities from abroad, particularly China (see Kearns and Lowe (2011) and Plumb, Kent and Bishop (2013)). In this paper we develop a multi-sector DSGE model with an explicit commodity sector that allows the price of commodities to differ from that of domestically produced consumer goods. This is important given the sizeable increase in the terms of trade for Australia since the mid 2000s. The domestic economy in our model is treated as ‘small’ in that it has no pricing power in world markets and external demand is treated as exogenous.

DSGE models place a large weight on economic theory; for example, they are derived from agents' utility and profit maximising behaviour. This enables economic interpretations to be given to the shocks, and hence DSGE models can be useful for scenario analysis. However, this comes at a cost: the model may be misspecified, and the high degree of structure in the model can place tight restrictions on its parameters. These tight restrictions reduce statistical uncertainty but may impede the forecasting performance.

At the other end of the spectrum, unrestricted VARs have many free parameters, and therefore can provide a better in-sample description of the data than a DSGE model. However, these parameters may be imprecisely estimated, particularly in small samples, which can reduce the VAR's forecasting performance. One possibility is to use a long run of data, but this is undesirable if the economy has experienced structural change in the period under investigation. Another possibility is to use Bayesian methods, which can limit over-fitting by introducing prior information or shrinking the parameter estimates towards some value.

In this paper we study the forecasting performance of a small open economy VAR model when information from the DSGE model developed earlier in the paper is used as a prior. We refer to this as an empirical BVAR-DSGE model.^[1] The aim of this approach is to reach a compromise between theory and data that may be useful for forecasting.^[2]

Many papers have estimated BVAR-DSGE models and evaluated their forecasting performance (see Del Negro and Schorfheide (2004) and Del Negro et al (2007) for the United States, Hodge et al (2008) for Australia and Lees et al (2011) for New Zealand). However, the methodology used in these papers, following Del Negro and Schorfheide (2004), does not impose the restrictions necessary to ensure that the small economy does not affect the large economy. To achieve this, we follow the estimation approach outlined in Robinson (2013).

We compare the point forecasting performance of our BVAR-DSGE model, which has exogenous foreign variables and error-correction terms arising from non-stationary technology, to those from the DSGE model alone, a small open economy variant of the Minnesota prior, and univariate autoregressive models. The main finding of this paper is that the BVAR-DSGE model generally improves the forecasts from the estimated DSGE model (as measured by their root mean squared error (RMSE)). However, we also find that this model does not perform better than the univariate benchmarks.

The rest of the paper is structured as follows. Section 2 discusses the DSGE model of the Australian economy and possible options for modelling the foreign sector. Section 3 describes the forecasting ‘competition’. Section 4 briefly reviews the estimation results from each of the models. We then examine the results in Section 5 and conclude.

Footnotes

We refer to it as an ‘empirical’ BVAR-DSGE model as the data are used in the construction of the prior, as well as in the BVAR estimation. [1]

BVAR-DSGE models are only one method of using information from DSGE and VAR models together (see Gerard and Nimark (2008) and Bache et al (2011) for alternative approaches in the context of central bank forecasting that have yielded promising results). [2]