RDP 2008-08: The Role of International Shocks in Australia's Business Cycle 1. Introduction

There is little consensus on the role played by the rest of the world in a small open economy's business cycle. In the case of Australia, Dungey (2002) estimates a structural vector autoregression (SVAR) model, which implies that international factors account for 32 per cent of output forecast errors over a one-year horizon, while domestic GDP shocks remain the dominant contributor. A SVAR model for Australia by Brischetto and Voss (1999) reveals that only around 5 per cent of output forecast errors stem from foreign factors. On the other hand, using an estimated new Keynesian dynamic stochastic general equilibrium (DSGE) model, Nimark (2007) concludes foreign shocks explain over 50 per cent of the variance in Australian output around its trend while domestic output shocks account for only 8 per cent. Using a different criteria, Dungey and Pagan (2000) simulate data from a SVAR model and find that recessions would have been less severe in the absence of foreign disturbances, while cumulated movements during the expansion phase would also have been smaller.

This paper argues that the different findings of the studies cited above can be understood as resulting from the difficulty of deciding how to appropriately identify the structural disturbances relevant to a small open economy. Traditional SVAR models employ zero-type restrictions, which may introduce substantial misspecifications that could lead to invalid inference. At the same time, identification of structural disturbances by means of cross-equation restrictions from a small DSGE model may be too stringent a method to capture the complex dynamics of the data-generating process. This paper contributes to this debate by developing a SVAR model of the Australian economy using robust sign restrictions derived from an estimated DSGE model. One key element of this approach is that it allows for a theoretically consistent view of the relationships between the set of macro variables without imposing the full DSGE structure or potentially invalid zero-type restrictions used in SVAR models.

Earlier sign restriction VAR studies focus mainly on identifying a subset of structural disturbances; examples include Faust (1998) and Uhlig (2005) who identify only monetary policy shocks. More recent studies by Canova and De Nicolo (2002) and Peersman (2005) apply the sign restriction methodology to identify all shocks in the VAR model. All these studies, however, are based on large economies with little discussion of the role of exchange rates. One exception is Farrant and Peersman (2006), who investigate the role of exchange rates in an open economy setting. However, the role of international factors is not explicitly discussed in that study.

The use of restrictions derived from a theoretical model to aid VAR estimation is not new. McKibbin, Pagan and Robertson (1998) use the McKibbin-Sachs Global (MSG2) model to restrict the long-run behaviour of a VAR, while the short-run features are left unrestricted. Dungey and Pagan (forthcoming) try to reconcile their earlier SVAR model with restrictions implied by a simple open economy DSGE model. Peersman and Straub (2004) use a calibrated real business cycle model to derive sign restrictions in order to identify technology shocks.

The starting point of this paper is to use the Beveridge-Nelson decomposition to extract the cyclical component of GDP, which will be used as a measure of Australia's business cycle. A slightly modified version of the small open economy model proposed in Monacelli (2005) and Galí and Monacelli (2005) is then estimated using maximum likelihood. The estimated model is used to determine a set of robust sign restrictions for the VAR analysis. The small open economy assumption is imposed on the VAR model by restricting the impact of domestic variables on foreign variables. The ultimate aim of the analysis is to map the set of statistical relationships estimated from the reduced-form VAR back into a set of structural disturbances for economic interpretation. To do this, an algorithm similar to that proposed by Canova and De Nicolo (2002) is used to trace out all possible orthogonal vector moving average (VMA) representations of the VAR that are consistent with the sign restrictions derived from the estimated DSGE model. Since there is not enough information to uniquely identify a set of structural disturbances, the median impulse approach suggested in Fry and Pagan (2005) is used to summarise the results.

The analysis reveals several interesting results. First, the Beveridge-Nelson decomposition produces a plausible measure of Australia's output fluctuations.

The characteristics of the cyclical behaviour match previous business cycle studies using factor models such as Gillitzer, Kearns and Richards (2005). Second, in contrast to previous zero-type restriction SVAR studies, foreign factors account for over half of the output forecast errors whereas innovations from output itself have only a modest effect. The result is robust across different foreign specifications using data for the United States and the G7 countries.

The rest of the paper is organised as follows. Section 2 describes the Beveridge-Nelson decomposition used to extract the cyclical component of GDP. Section 3 outlines the estimated small open economy DSGE model together with the data used in the analysis. A set of robust sign restrictions are derived from the estimated DSGE model for the open economy SVAR. Section 4 describes the estimation and identification of the open economy SVAR model. Section 5 summarises the estimation results. Finally, Section 6 reviews the main findings.