RDP 2009-05: Macroeconomic Volatility and Terms of Trade Shocks 6. Robustness Checks

To test the robustness of our results, we considered a number of alternative specifications of our terms of trade and control variables, none of which substantively affects our conclusions.[17] Our key findings are also robust to the exclusion of various groups of countries (to control for endogeneity bias), alternative sample periods, and the inclusion of additional variables to control for other types of shocks.

While assuming that terms of trade shocks are exogenous is probably reasonable for small economies, it is not strictly true for very large countries or for countries that exert substantial pricing power in the markets for certain commodities. To explore this issue, we re-estimated our model excluding two groups of countries – the G7 economies and non-commodity producers – whose terms of trade have some chance of being influenced by domestic economic developments.[18] The motivation for the latter test is that, because commodities tend to be homogeneous goods, terms of trade shocks for most of these countries are likely to be exogenous with respect to domestic economic conditions.[19]

The results of these exercises, shown in Tables C2 and C3, are similar to our baseline regressions. Indeed, the only noticeable difference is that the coefficient on terms of trade volatility is often slightly larger, suggesting that terms of trade shocks are more disruptive for smaller economies and for commodity exporters. While excluding OPEC oil producers from our sample also produced qualitatively similar results, the floating exchange rate interaction term became significant at the 13 per cent level.[20] However, to the extent that Australia's terms of trade is reasonably correlated with that of oil exporters (reflecting Australia's status as a major exporter of coal and iron ore), we are inclined to retain this OPEC variation in our central specification.

Although the inclusion of time-fixed effects controlled for the existence of common shocks, it is still interesting to examine the extent to which the large amount of macroeconomic and terms of trade volatility that many economies experienced in the 1970s affects our sample. To do this, we re-estimate our equations over the period 1980–2005, which was a time of comparative macroeconomic stability for many economies in our sample.[21] As Table C4 shows, the interaction term on strict monetary policy becomes highly significant in the output volatility regressions, while our finding that floating exchange rate regimes cushion the impact of terms of trade shocks remains intact. The results from the post-1980 period suggest that the results based on the longer sample understate the role of strict monetary policy in helping to moderate the effects of terms of trade shocks.

Finally, despite our focus on the terms of trade shocks, there are likely to be other external factors that influence macroeconomic volatility (see Lui 2008, for example). While the time-fixed effects in our regressions will control for shocks that are common across countries (such as to productivity or technology), it is possible that our terms of trade variable is at least partly proxying for other external influences that directly affect domestic output volatility through trade and/or confidence channels. To control for this possibility, we included a measure of trading partner output volatility in our regressions. We construct this variable by aggregating the output volatility of each country's ten largest trading partners, using bilateral export weights at five-year intervals (for instance, output volatility for the period 1981–1985 is weighted by the 1980 export share; see Appendix A for more details). While we weight this variable on the basis of the top ten trading partners to ease the computational burden of this exercise, it provides a reasonably comprehensive coverage.[22]

Table C5 shows the results from the post-1980 regression controlling for trading partner output volatility (tpvol).[23] Most of the key results from Table C4 remain intact. While the terms of trade coefficient is positive but less consistently significant, the floating exchange rate and strict monetary policy interaction terms remain negative and are significant (by themselves) at the 5 per cent level. Interestingly, tpvol enters the model with a negative sign, although this term is generally insignificant. Further analysis revealed that while tpvol was statistically significant and positively related to output volatility in a simple regression with country-fixed effects, it became insignificant (and in some cases negative) once time-fixed effects were included. We also re-estimated our models with tpvol instead of time-fixed effects to control for global shocks, but the explanatory power of these models were somewhat lower. The results suggest that, to the extent that tpvol captures external influences relevant to macroeconomic volatility, these tend to be common across countries.


These included: adding the average level of inflation over the previous five-year block to the output volatility regressions to control for monetary shocks, including exchange rate volatility (measured in SDR or nominal effective exchange rate terms) as an additional variable and multiplying terms of trade volatility by an economy's trade share to control for the possibility that terms of trade volatility has a larger effect on more open economies. Results of these robustness tests are available from the authors on request. [17]

The economic significance of some of the G7 economies has declined in recent years relative to some of the larger developing economies, including China, India and Brazil. However, as our sample includes the period 1971–2005, the G7 probably reflects the most economically significant economies over this sample. [18]

We classified commodity producers as those countries for whom manufactured goods account for less than 50 per cent of export revenue. [19]

However, this change in significance mainly reflects a larger standard error – associated with the smaller sample size – as opposed to a meaningful reduction in the point estimate. [20]

Of course, even this period was not free of macroeconomic volatility in particular regions, most notably the ‘Peso Crisis’ in Latin America in the early 1990s, and the ‘Asian Financial Crisis’ of the late 1990s. [21]

For instance, we are able to capture at least to two-thirds to three-quarters of the value of exports for most countries in our sample. [22]

Comprehensive bilateral export weights are only available from 1980. [23]