RDP 2001-01: The Decline in Australian Output Volatility 5. Data and Results

Before presenting the main results I consider the appropriate way to include the variables in the VAR. This decision is driven by the results of ADF tests presented in Table A1 in the appendix. Output is unquestionably non-stationary while its growth rate is stationary so it is included in the VAR in growth rates. The evidence for unemployment and the terms of trade is more mixed. Over some samples they appear stationary and over others it is not possible to reject the null of integration. As there is not much to be gained one way or the other I include these variables in levels. Ultimately, the choice does not have a strong effect on the results.

The table below shows the variances of the structural shocks across the four samples considered in this paper. Clearly the shocks were much smaller in the 1990s than in any other period.

It is worth reminding the reader at this point that the ‘1970s’ in this table applies to the expansion from 1974:Q3 to 1983:Q2. Thus, the first OPEC oil shock actually falls right at the end of the ‘1960s’ sample. This is part of the explanation for the relatively low TOT structural variance in the ‘1970s’ and the high variance in the ‘1960s’.^[9] With regard to the interpretation of the structural variance of the productivity shocks this is not as much of an issue. Output was highly volatile in the 1960s even in the absence of any effect from the large OPEC oil shock. Table 4 below presents the sensitivity of each system to the different structural shocks at a horizon of one year. The results are all normalised so that the sensitivity of the 1990s equals one and so cannot be directly compared across columns.

We see that the 1990s were more sensitive to TOT shocks than in previous periods while there has been a reduction in the sensitivity to demand shocks. Nonetheless, the changes in sensitivity to demand and TOT shocks are significantly smaller than the changes in the structural variances.

The finding that output growth is more sensitive to TOT shocks in the 1990s is, at first glance, counterintuitive. Standard international economics theory suggests that a floating exchange rate should insulate the economy from TOT shocks more than a fixed-rate regime. We might then expect that output would be less responsive to TOT shocks in the 1990s. Unfortunately it is difficult to make a direct connection between the theory and the empirical results here. Part of the difference stems from the fact that appropriate dynamic responses are not well understood in these circumstances. Furthermore, the results reported above are for output growth rather than the level of output which is what is commonly included in the theory. Looking at the results for the level of output suggests that there has been little change in the sensitivity of output to TOT shocks. This may suggest that the moving peg that prevailed before the float of the dollar may have provided the same level of insulation as the float. This is, however, moving a little far from the focus of this paper and I will leave further investigation to other work.

The decline in the sensitivity of the economy to demand shocks provides some evidence that structural changes have helped reduce the volatility of output. Specifically, the economy is now somewhat more resistant to temporary shocks. The particular structural changes may include the microeconomic reform of the 1980s as well as the introduction of the medium-term inflation target in the 1990s.

Overall, however, Tables 3 and 4 suggest that the primary reason for the decline in output volatility in the 1990s was a decline in the volatility of the underlying shocks. The final piece of information required is to look at a variance decomposition to see which shocks are the most important in explaining overall output variability. Table 5 shows this decomposition at a one year horizon. Results are similar for longer horizons.

The dominance of productivity shocks in all periods leads to the strong conclusion that the decline in the volatility of productivity shocks was the most important single factor in the decline in output volatility in Australia in the 1990s. More precisely, given the identification strategy, the decline in the volatility of permanent shocks is the most important single factor explaining the decline in output volatility in Australia – and these permanent shocks are best thought of as productivity shocks. Demand and TOT shocks are relatively insignificant. Nonetheless, they both have parts to play in the decline.

The increased sensitivity of the 1990s to TOT shocks shows up in a larger contribution to the forecast error variance (FEV) for TOT shocks despite the lower variance of TOT shocks during the 1990s. This suggests that an important area to watch in terms of output volatility is the terms of trade. Relatively minor fluctuations in the TOT (from a historical perspective) could have a significant effect on output volatility. For example, volatility in the TOT similar to that of the 1980s could raise the volatility of output by almost 50 per cent.^[10] The reduction in demand shocks' weight in the variance decomposition is a reflection that demand shocks and the structural sensitivity of the economy to demand shocks have reduced by significantly more than the average.

These results are broadly in line with previous work on Australia by Glenn Otto (1995, 1999). He found that TOT shocks were relatively insignificant and that productivity shocks explained much of the forecast error variance in his models. Nonetheless, it is interesting to see that the predominant influence on Australian output volatility comes from the shocks hitting the economy rather than the structural responses to those shocks.

Thinking about the various shocks identified raises some substantial questions. While it is fairly easy to identify causes of fluctuations in Australia's terms of trade, it is more difficult to think about causes of fluctuations in productivity. The next section speculates on possible interpretations of the results.

Footnotes

The other part of the explanation is that quarter-to-quarter volatility in the 1960s was very high whereas the quarter-to-quarter volatility was less in the 1970s despite the overall range for the TOT being larger in the 1970s. These variances are residuals from an estimation and so do not necessarily match standard deviations of the series over the various samples. [9]

This is calculated by looking at the FEV of output one year out if the economy was subjected to productivity and demand shocks of the same size as in the 1990s but TOT shocks from the 1980s. [10]