RDP 2005-01: Long-Term Patterns in Australia's Terms of Trade 3. Terms of Trade Volatility

The Australian terms of trade have had many episodes of exceptionally high volatility, as seen in Figure 1. Notably, as Table 5 summarises, dividing the sample into thirds, volatility was substantially higher in the middle portion of the sample.^[9] The standard deviation and proportion of volatile years in the most recent third of the sample have been remarkably similar to the first third. The middle part of the sample stands out for its higher volatility.

Abstracting from a sustained decline throughout the 1890s and sharp reversal around the turn of the century, the terms of trade were quite stable from 1870 until World War I. Following World War I the terms of trade were subject to many sharp swings, frequently doubling or halving within the space of just a few years. They declined rapidly after the war, only to more than double in three years with the 1920s boom. This was short lived, with the terms of trade halving leading into the Great Depression. Indeed, many authors, including Valentine (1987), have stressed the role of falling commodity prices, and so export earnings, in transmitting the Great Depression to Australia. Another large upward spike preceding World War II was followed by the terms of trade almost tripling over the period 1944–1951. This rapid growth was due to an increase in the price of wool exports resulting from the Korean War. The impact of shocks to individual commodities was all the greater because exports were highly concentrated in just a few goods. During the post-World War II episode, wool made up over 50 per cent of Australia's goods exports by value. While there have been more recent cyclical swings, notably the spike induced by the mineral price boom in the early 1970s and the OPEC oil price shock, the volatility of the goods terms of trade has declined sharply over recent decades, as seen in Table 5.

A likely reason for the reduction in terms of trade volatility in recent decades is the diversification of Australia's export base. If the prices of goods exported are not perfectly correlated, then a broader export base may lead to lower price volatility. Even if the range of goods exported does not change, so long as exports become less concentrated in just a few of those goods, volatility can decline. The volatility of the terms of trade could also have been reduced by diversification into manufactures and services trade, which will have reduced the compositional difference between Australia's import and export baskets. Note that export composition was broadly stable during the first half of the 20^th century, so changing composition cannot account for the increase in volatility during the inter-war years.

Changes in the volatility of Australia's terms of trade may also have been driven by volatility of global commodity prices. This in turn may have been influenced by changes in global exchange rate regimes. Cashin and McDermott (2002) and Cuddington and Liang (1998) find that real commodity prices have been more volatile during floating exchange rate periods. The fixed exchange rate regimes prior to World War I and post-World War II may have been associated with the less volatile terms of trade during these periods. The flexible exchange rates in the inter-war years may then be associated with a more variable terms of trade. But this observation does not accord with the slightly more stable terms of trade in the era of flexible exchange rates, whether measured from the float of the Australian dollar in 1983 or the end of Bretton Woods in 1972. For example, the standard deviation of the log terms of trade from 1960–1983 was 0.089 but thereafter has been only 0.074. Alternatively, high commodity price volatility in the inter-war years (and the breakdown of fixed exchange rate regimes) may have been due to geopolitical instability.

To test whether there have been statistically significant breaks in the volatility of Australia's terms of trade we use the Bai and Perron (1998) test, which endogenises the selection of multiple breakpoints. We measure volatility as the absolute value of annual changes in the log terms of trade, and look for breaks in the mean of this series (Figure 7). We use absolute changes, as do Ahmed, Levin and Wilson (2002), rather than squared changes because the latter magnifies the amplitude of large changes and so is presentationally more cumbersome. The test results using squared annual changes in the log terms of trade lead to the same conclusions. Table 6 reports the results of this test over the 1870–2004 sample.

The sequential test fails to find a break for both terms of trade series but this is most likely because the test cannot reject zero breaks against one break. The double maximum tests clearly reject zero breaks against an unspecified number of breaks for both series. After allowing for one break, the SupF tests reject one break in favour of two, but not two in favour of three at the 5 per cent significance level. In addition, the two information criteria tests indicate two breaks. Together these results strongly suggest that there are two breaks in the volatility of the terms of trade. Conditional on there being two breaks, the Bai and Perron test selects 1923 and 1953 as the most likely break dates.

The volatility of the terms of trade was significantly higher over the period 1923–1952 than before or after these years, as shown in Figure 7. This episode of increased terms of trade volatility is coincident with the inter-war period of flexible exchange regimes. The end of World War II and subsequent establishment of the Bretton Woods fixed exchange regime in 1946 may have been important factors in locating the subsequent fall in terms of trade volatility. The sharp increase in wool prices between 1946 and 1951, when it was 44 per cent of goods exports, may account for not finding a break in Australia's terms of trade at the time Bretton Woods commenced. As mentioned, diversification of Australia's export base may have contributed to the post-World War II fall in terms of trade volatility. While the broadening of the export base was relatively rapid, it did take a number of years and so this explanation may be less able to tie down a particular breakpoint.

3.1 Decomposing Volatility

To better understand the changes in its volatility, we decompose the variance of the terms of trade into the variance and covariance of its components. The log of the terms of trade, tot, is rearranged to express it as the difference between detrended log export and import prices, as shown in Equation (3),

where and , are the logged export and import price series at time t and d_t is a common HP filter used to detrend the prices. Detrending avoids overstating the variability due to a common trend in the two series, notably inflation.^[10]

The volatility of the terms of trade is decomposed as in Equation (4).

The sample is split in 1923 and 1953, the break dates selected by the Bai and Perron test. Table 7 reports these results in two parts, to highlight the increase and subsequent decrease in terms of trade volatility. For brevity, we only report results for the goods terms of trade as the results are similar to those for the goods and services series.

The volatility of the goods terms of trade was almost five times higher over the period 1923–1952 than 1870–1922, and more than three times higher than from 1953–2004. The rise and subsequent fall in export price volatility is almost entirely responsible for this development. In fact, import price variance was 36 per cent lower in the middle period and almost unchanged in the post 1953 sample. The covariance of detrended export and import prices was lower from 1923–1952 than 1870–1922, contributing to an increase in terms of trade volatility. From 1923–1952 and 1953–2004 the results indicate a negative correlation between import and export prices, but these correlations are sensitive to the detrending series used.

3.2 Changes in Export Price Volatility

Since export prices made the largest contribution to the rise and subsequent fall in terms of trade volatility, we consider their role in greater detail. As documented in Section 2.1.1, there was little change in Australia's export composition until the mid 1950s but there was substantial diversification thereafter. So while an increase in price shocks caused the increase in export price volatility in the inter-war years, both diversification of the export base and a decline in price shocks may have contributed to lower export price volatility from the mid 1950s. To determine the relative importance of these effects in reducing export price volatility we compare the volatility of the detrended export price series for different subsets of goods exports. Table 8 reports these results for various episodes, starting in 1904 because of the availability of component price data.^[11]

The effect of diversification on export price volatility can be seen by looking across the columns for each sample. Australia's exports remained primarily rural goods from 1904–1952, and as expected the rural goods series had similar volatility to the broader export price series. After 1953, diversification from rural goods into metals, coal and gold reduced the volatility of export prices, as seen in the last row of Table 8. However, diversification to include a broader range of commodities and manufactures did not significantly reduce export price volatility. This may be because rural goods, metals, coal and gold still constituted on average two-thirds of goods exports over the period 1953–2004.

Comparing the rows for each subset of goods exports, it can be seen that their price volatility was substantially higher over the period 1923–1952 than before or after this period. This suggests that an increase in price shocks was primarily responsible for the increase in export price volatility from 1923–1952. However, the results in Table 8 are not conclusive because within each grouping of goods considered, the weights vary over time. So changes in the composition of exports at a more disaggregated level may be causing an overestimation of the impact of price shocks.

To investigate whether this is the case we have made use of century-long price data for 19 commodities and produced two fixed-weight commodity price indices. By definition these series do not permit any diversification across commodities. Appendix A contains a list of the commodities included in the index. In 1905 these commodities comprised 82 per cent of total goods exports. While their share of total exports declined steadily they still represented 44 per cent of goods exports in the year 2000. The standard deviations of the constructed commodity price series are reported in Table 9.

Using 1905–1955 average value weights, over which time exports were primarily rural, the commodity price index volatility increased substantially over the period 1923–1952. This confirms our finding from Table 8 that increased price shocks caused the increase in export price volatility in the inter-war years. From the 1950s, the share of rural goods fell steadily, as mineral exports became more prominent. If Australia's exports from 1904–1952 had instead been as it was on average in the latter half of the 20^th century, the period 1923–1952 would not have been one of relatively high export price volatility. This suggests that higher price shocks during the period 1923–1952 were particularly pronounced for Australia's traditional major commodity exports. But the fall in volatility over the period 1953–2004 for the index using average 1905–1955 export value weights indicates that even if Australia's export base had not diversified beyond traditional commodity exports, volatility would still have fallen.

These results can also be seen in individual commodity price series. Table 10 reports the standard deviations of detrended logged prices for 14 important commodities for which a long time series of data are available. It can be seen that commodities prominent in Australia's exports in inter-war years, notably wool and wheat, experienced much larger price shocks over these years.

Footnotes

The pattern of goods terms of trade volatility in Table 5 is virtually identical for the broader goods and services terms of trade. [9]

λ = 100 was used as the smoothing parameter as is standard for annual data, though results are generally robust to the use of a range of lambda values greater than 100. Using the Australian GDP deflator instead of an HP filter trend gives qualitatively similar results. [10]

Because the move to a floating exchange rate in 1983 may mechanically impart volatility in the Australian export price series, we repeated the calculations in Table 8 using the Australian import price series to detrend. The results using this series are qualitatively similar to those shown in Table 8. [11]