RDP 2001-01: The Decline in Australian Output Volatility 1. Introduction

The 1990s have been a period of generally strong growth and extended expansion throughout much of the world. In the US the expansion beginning in March 1991 has set a record as the longest post-war expansion. The expansion in Australia from the June quarter of 1991 has also been long by historical standards. When combined with low inflation, this robust growth has led to economic conditions unlike any seen since the 1960s. This paper aims to shed more light on the sources of the recent output growth experience and thereby, indirectly, on the prospects for the future.

The most obvious feature of the recent growth experience is that output growth has been much smoother than in previous periods. Table 1 shows the facts on output variability over the past forty years. It shows the variance of output growth over the four cycles that generally cover the 1960s, 1970s, 1980s and 1990s.[1]

Table 1: Variance of GDP(E)[2]
1960s 14.60
1970s 3.34
1980s 1.75
1990s 0.63

Note: Calculated as the variance of quarterly GDP(E) growth, gross national expenditure plus net exports, over the periods identified in footnote 1.

It is clear that there has been a significant decline in GDP(E) volatility. Much of this change in volatility can be traced to a change in the volatility of the inventories cycle. It is not clear whether this is a statistical artefact, given that it is so large compared to other output measures, or a real effect. To remove the influence of the inventories cycle and focus on more fundamental forces, I will concentrate on the growth of output minus inventories. A detailed examination of the inventories cycle in Australia will be left for another paper.

Removing the influence of inventories from the data there is still a significant decline in output volatility. Table 2 shows the variance of domestic final demand plus net exports.[3]

Table 2: Variance of Domestic Final Demand Plus Net Exports
1960s 1.23
1970s 1.57
1980s 1.73
1990s 0.81

Note: The variances for the 1970s and 1980s are affected by one outlier in each case. Deletion of the outlier leads to a variance of 1.31 for the 1970s and 1.16 for the 1980s. The sample periods are those identified in footnote 1.

We can see that the volatility of output in the 1990s (abstracting from the inventories cycle) is still the lowest it has been since the start of the quarterly national accounts data – the volatility of the 1990s is less than half that of the 1970s or 1980s. The pattern is also slightly different, suggesting that the 1980s were a time of relative volatility and that the 1960s, despite the picture given from the GDP(E) data, were a time of relatively smooth expenditure growth.

The volatility of output growth is important in assessing the prospects for economic growth. Quite apart from the confidence that smooth, consistent growth engenders, there is an effect on the potential length of expansions. With high volatility it is much easier for random shocks to cause the economy to contract and start a recession. The work of Hamilton (1989) seems to suggest that recessions, once begun, have different dynamics from expansions. Thus, to the extent that recessions are not merely a continuation of previous dynamics, their initiation is worth avoiding.

There is also a literature that looks at the correlation between output growth and its volatility (see Kormandi and Meguire (1985) for example). The initial finding was that higher volatility is correlated with higher growth. This was rationalised as an application of asset-market ideas about risk and return trade-offs; higher growth (return) should be accompanied by higher volatility (risk). However, this has been questioned by some more recent papers (Ramey and Ramey (1995), Martin and Rogers (2000)). It would be interesting to see how the recent run of data, including the substantially improved output performance in the US over the past few years, affects this story.

These factors should provide adequate reason to look at the sources of output volatility. Hopefully, a better understanding of their causes will allow for better understanding of any shocks that occur and provide a better foundation for forecasts of output growth. The rest of this paper looks at the nature of volatility in Australian output growth and how it has changed over the past decades. It identifies whether the reduced output volatility is primarily due to shocks or the improved stability of the economy. The paper also considers some potential explanations for the reduced variability of shocks. Section 2 conducts a brief literature review to place this work in the context of previous research. Section 3 then discusses the distinction between shocks and structure in more detail. Section 4 presents the technical details of the structural VAR model that will be used to separate the influences of shocks and structure and Section 5 presents the results. Section 6 speculates on causes of the decline in the supply (or productivity) shocks identified in Section 5. Section 7 concludes. Details of the data used are contained in Appendix A.


The cycles are defined from trough to trough. The cycle of the ‘1960s’ is defined as 1961:Q4 to 1974:Q2, the cycle of the ‘1970s’ as 1974:Q3 to 1983:Q2, the cycle of the ‘1980s’ as 1983:Q3 to 1991:Q2 and the expansion of the ‘1990s’ by 1991:Q3 to the present, 2000:Q4. Including the early 1990s recession in the ‘1990s’ sample, and similarly adjusting earlier samples, does not unduly affect the results. [1]

The high volatility in the 1960s and 1970s is not present in the series for GDP(A). The divergence comes from (a) the statistical discrepancy and (b) the process of chain linking the real GDP series. These seem to lead to much greater volatility in earlier observations. Nonetheless, the same pattern is present in the GDP(A) series. These results are presented to provide a consistent basis for comparison with the results of Table 2. For comparison, the variances of GDP(A) in the 1960s, 1970s, 1980s and 1990s are 2.41, 1.15, 0.91 and 0.35. [2]

This measure is essentially GDP(E) less inventories but the use of chain-linked series creates some minor differences in earlier periods. [3]