Output volatility in Australia

Abstract

A number of papers have documented a significant decline in real GDP volatility in several major OECD economies. Some authors have presented evidence to suggest that this is the outcome of a one-off structural break from a high to low volatility state whilst others have estimated regime switching models that indicate low volatility regime states have dominated in recent years. This article provides further evidence on the general properties of output volatility for Australia, including evidence of a significant moderation in output volatility for the country that occurred in the early 1980s. Estimates of various GARCH models of real GDP growth are also provided to further examine shorter term volatility features of the Australian economy that are associated with its business-cycle. A regime shift dummy is maintained in all models of the conditional variance in order to account for the regime shift in volatility and evidence is found of significant business-cycle effects, including leverage effects and asymmetries that suggest recessions are times of higher output volatility than economic expansions. Overall, it is concluded that the so-called ‘Great Moderation’ in macroeconomic instability, as documented here for Australia, is a result of a myriad of economic, institutional and policymaking changes.

Acknowledgement

The author would like to thank Gareth Leeves, John Foster, Alicia Rambaldi, Meryn Scott, the editor an an anonymous refree for helpful comments and suggestions. Alas all remaining errors and omissions are attributable to the author.

Notes

¹See Summers (2005) and Bernanke (2004).

²Such evidence regarding both long-run and short-run changes in volatility are often ignored in discussion of so-called ‘key features’ or ‘stylized facts’ regarding business-cycles. A recent paper by Cashin and Ouliaris (2004) is an example of such for Australia.

³Note that the procedure proposed by Bai and Perron (2003) was used to check if there are more structural breaks than just the one break tested for using the Andrews–Quandt procedure. The results obtained were consistent with those presented in Table 2 and are omitted for purposes of brevity, although they are available from the author upon request.

⁴It should be noted that Hansen (1992) points out that if both the parameters of the mean function and the variance have shifted, then the L test has only low power to detect the shift in the parameters of the mean function. Thus, the test does not allow us to discount the possibility that the mean function has shifted with any degree of certainty, although there is a clear indication that the variance is nonconstant.

⁵These results stand in contrast to the recent research on German output volatility by Fritsche and Kuzin (2005) that suggests that diminished volatility in output growth in that country is largely due to a decline in the persistence of the growth process (a change in the propagation mechanism possibly due to a change in the conduct of monetary policy) rather than a break in the variance.

⁶Results of the basic preliminary models are omitted for purposes of brevity but all detailed results can be obtained from the author upon request.

⁷BDS test (see Brock et al. (1996) – a portmanteau test for time based dependence in a series. e represents the embedding dimension whilst l represents the distance between pairs of consecutive observations measured here as a multiple of the SDs of the series. Under the i.i.d. null, the BDS test statistic is asymptotically distributed as standard normal. Bootstrapped p-values were calculated using 5000 repetitions.

⁸Various specifications of ARCH in mean (ARCH-M) models were also estimated but in no case was it found that the volatility entered the mean equation with a significant coefficient. This evidence is supportive of that found by Speight (1999) for the UK.

⁹McMillan and Speight (1997) find evidence of asymmetries of EGARCH form for US industrial production and of TGARCH asymmetries for German, Japanese and US industrial production.

¹⁰Once again, various specifications of GARCH in mean (GARCH-M) models were also estimated but in no case was it found that the volatility entered the mean equation with a significant coefficient.

¹¹Again, since no evidence was found of significant GARCH terms in the models, an ARCH(1) specification was used in each case.

¹²Alternative methods used to generate classical business-cycle dates, such as the popular ‘at least two negative quarters of real GDP growth’ scheme and the dates estimated using a two phase Markov switching model, such as that presented in Bodman and Crosby (2002), were also tried and similar results obtained to those presented here.

¹³The importance of inventories in the busines cycle has been discussed by authors such as Flood and Lowe (1995)

¹³See Summers (2005) and McConnell and Perez-Quiroz (2000).

¹⁴See Ho and Tsui 2003.