RDP 2005-01: Long-Term Patterns in Australia's Terms of Trade 2. Trends in the Terms of Trade

Prebisch (1950) and Singer (1950) suggested that countries that primarily export commodities, and import manufactures, had experienced declining terms of trade. Further, they could expect ongoing falls. Prebisch and Singer, and subsequent work, have proposed a range of theories to account for this phenomenon.

The most common theory is that the demand for raw commodities declines, at least proportionately, with ongoing economic development. This lower income elasticity of demand for commodities compared to other goods results in relative price falls as income rises. Another explanation notes that manufactured goods are differentiable, unlike homogenous commodities. As a result, producers of manufactured goods may have greater market power. Productivity increases will then lead to smaller relative price falls for manufactures than for commodities.[3] A less likely justification is that increases in the supply of commodities are met with larger price falls because the demand for commodities is less price elastic. Such an effect would have to be large enough to overcome presumably higher productivity growth in manufactures that is likely to lead to faster growth in production of manufactures than for commodities.

Prebisch and Singer's hypothesis reversed the view previously held by 19th century economists. The conventional wisdom had been that decreasing returns to scale in primary commodity production and constant or increasing returns to scale in the manufacturing sector, combined with population growth, would see relative commodity prices, that is the ratio of commodity prices to manufactures prices, increase over time.

Prebisch and Singer's early empirical work precipitated many studies questioning the validity and robustness of their findings. A frequently cited study is Grilli and Yang (1988). Using a commodity price index they constructed from 24 primary commodities they found a statistically significant fall of 0.6 per cent annually in non-oil commodity prices relative to manufactured goods prices over the period 1900–1986 (hereafter this series is referred to as relative commodity prices). The vast literature testing the Prebisch-Singer hypothesis has produced many conflicting results due to differences in methodology, even though much of it relies on the Grilli-Yang data. For example, Cuddington, Ludema and Jayasuriya (2002) suggest that relative commodity prices experienced a one-time downward jump in 1921, rather than having an ongoing negative trend.

Despite the methodological differences, work to date broadly supports the Prebisch-Singer hypothesis. For example, Lutz (1999) argues that the univariate trend model estimated by Grilli and Yang (1988) is inappropriate because commodity and manufactures prices are cointegrated. Nonetheless, Lutz finds a statistically significant long-run decline in real commodity prices of 0.9 per cent per annum. More recently, Cashin and McDermott (2002) found a larger 1.3 per cent annual decline in the relative price of commodities using a different data source, the Economist commodity price index. In this longer sample, 1862–1999, they found no evidence of a break in the trend.

There is some evidence that declining relative commodity prices have resulted in a negative trend in Australia's terms of trade. A common, and reasonable, assumption is that a small economy such as Australia takes world prices, and so its terms of trade, as given. Sapsford (1990) finds that there was a significant downward trend from 1951 to 1987, even though there was no trend in the first half of the century. Gruen and Kortian (1996) also find a negative trend in the Australian terms of trade.

In this study we use two measures of Australia's terms of trade on an annual basis. The goods and services terms of trade incorporates the prices of all exports and imports. We also consider the goods terms of trade, for which the Prebisch-Singer hypothesis will be more relevant and is not subject to the difficulties in the measurement of service prices. This series is more consistent over time since terms of trade data before 1949 do not include the, admittedly small, trade in services. Throughout we use the terms of trade in logs, as the change in the logged series between two periods measures the proportionate change in the terms of trade. Descriptions of all data and their sources are given in Appendix A.

Before assessing whether there is a trend in Australia's terms of trade we briefly examine the stationarity of these series. Knowing the degree of integration is important for choosing appropriate econometric techniques to test for a trend. But a finding of a unit root in the terms of trade would be of interest in itself, indicating that shocks to the terms of trade are permanent, and so that export and import prices are not cointegrated.

Table 1 reports results from two common unit root tests. The sample is split in 1955 because the substantial diversification of Australia's exports from this date, as documented in Section 2.1.1, may have changed the behaviour of export prices and so the terms of trade.

Table 1: Unit Root Tests
  Intercept Intercept and trend
Log goods terms of trade
1870–2004 *** *** ***
1870–1954 *** **
1955–2004 *** *
Log goods and services terms of trade
1870–2004 *** * ***
1870–1954 *** **
1955–2004 ***
Notes: ERS denotes the Elliot, Rottenberg and Stock (1996) unit root test, for which the null hypothesis is that the series contains a unit root. KPSS denotes the Kwiatkowski et al (1992) test for which the null hypothesis is stationarity. The Newey-West lag selection criteria was used for the KPSS test and the Bayes information criteria was used to select the number of lags for the ERS test. ***,** and * denote rejection of the null hypothesis at the 1, 5 and 10 per cent levels of significance respectively.

These results suggest that both the goods, and goods and services, terms of trade are stationary series, at least around a trend. The tests cannot reject that the terms of trade contain a unit root after 1955, but this is possibly due to the small sample size.

As a first test it is illustrative to fit a simple linear time trend, t, to the natural logarithm of the terms of trade, tott, as in Equation (1):

These results, reported in Table 2, indicate that there has been a statistically significant negative trend in Australia's terms of trade. Over the full sample there has been a −0.3 per cent trend in the goods terms of trade, and a smaller −0.1 per cent trend in the goods and services terms of trade.

Table 2: Trend in the Terms of Trade
tott = a + βt + εt
  1870–2004 1870–1954 1955–2004
α 4.936***
β −0.003***
Q(1) 78.71*** 49.04*** 17.79***
Q(5) 130.51*** 79.87*** 20.45***
Arch-LM 21.73*** 13.01*** 6.47**
  Goods and services
  1870–2004 1870–1954 1955–2004
α 4.734***
β −0.001***
Q(1) 76.38*** 47.46*** 15.99***
Q(5) 125.47*** 77.36*** 17.98***
Arch-LM 28.56*** 11.33*** 6.07***
Notes: ***,** and * denote rejection of the null hypothesis at the 1, 5 and 10 per cent levels of significance respectively. Figures in parentheses are Newey-West robust standard errors. Q(1) and Q(5) are Ljung-Box statistics for autocorrelation at 1 and 5 lags respectively. The null hypothesis is that the series contain no autocorrelation. Arch-LM is the Lagrange multiplier test for autoregressive conditional heteroskedasticity at 1-lag, with the null hypothesis of no heteroskedasticity.

However, the full sample results mask a story of two parts. We split the sample in 1955 based on graphical observation, previous work and evidence that the structure of trade changed after this date, as detailed in Section 2.1.1. After 1955 the trend is larger than over the full sample, −0.6 per cent for the goods terms of trade and −0.4 per cent for the goods and services terms of trade. Before 1955 the trend is statistically insignificant for both series. While these results indicate the negative trend has been lessened by the inclusion of services trade, the difference in the size of the trend is heavily influenced by data in just two years, 1958 and 1959. The significant negative coefficient lends support to the theory that declining relative commodity prices impart a negative trend in Australia's terms of trade, though the magnitude is less than that found for relative commodity prices.

However, the Ljung-Box statistics reported for the regressions in Table 2 indicate significant autocorrelation in the residuals of Equation (1), resulting from some persistence of the terms of trade. We can account for the residual autocorrelation by adding a lagged dependent variable to our trend regressions, as shown in Equation (2).

Lutz (1999) argued that Equation (2) does not properly account for cointegration between import and export prices because it assumes that they are cointegrated with a long-run elasticity of unity. However, our finding of stationarity for the terms of trade suggests it is an appropriate specification. Table 3 reports the results of this regression.

Table 3: Trend in the Terms of Trade: Allowing for Persistence
tott = α + βt + ρ(tott−1) + εt
  1870–2004 1870–1954 1955–2004
α 1.197***
β −0.001***
ρ 0.758***
Q(1) 2.43 1.44 1.19
Q(5) 7.55 3.07 5.47
Arch-LM 26.90*** 16.77*** 1.39
ρ-MU 0.796
Half-life 3.03
[1.88, 7.21]
[1.88, ∞]
[1.14, ∞]
  Goods and services
  1870–2004 1870–1954 1955–2004
α 1.202***
β −0.000*
ρ 0.746***
Q(1) 1.71 1.04 1.45
Q(5) 6.25 2.59 4.78
Arch-LM 28.24*** 15.31*** 3.41*
ρ-MU 0.784
Half-life 2.84
[1.78, 6.40]
[1.79, ∞]
[0.96, ∞]
Notes: ***,** and * denote rejection of the null hypothesis at the 1, 5 and 10 per cent levels of significance respectively. Figures in parentheses are Newey-West robust standard errors. Q(1) and Q(5) are Ljung-Box statistics for autocorrelation at 1 and 5 lags respectively. Arch-LM is the Lagrange multiplier test for autoregressive conditional heteroskedasticity at 1-lag. ρ-MU is the Andrews (1993) median unbiased estimate of ρ, figures in brackets below represent a 95 per cent significance upper bound on ρ. The numbers in brackets beneath the half-life represent a 90 per cent confidence interval for the half-life of a shock to the terms of trade.

The lagged dependent variable is highly significant in all samples. In each regression, its inclusion reduces the coefficient on the trend by around two-thirds. Over the full sample, there is still stronger evidence of a trend in the goods terms of trade, though even in this series it is just −0.1 per cent per annum. This coefficient roughly accords with the observed 12 per cent decline over the past 135 years. If we extend our sample to include two years of the projections for the goods and services terms of trade the trend is insignificantly different from zero. The trend is economically and statistically insignificant over the first part of the sample for both series. After 1955 it appears slightly stronger. However, we cannot reject that the trend has the same coefficient in the two sub-periods.

As noted earlier, we have some priors that the behaviour of the terms of trade may have changed after 1955, but this may not be the appropriate timing. To account for this we use a test that endogenises the selection of the breakpoint by searching over all possible breakpoints. The Sup(t) test, described in Cashin and McDermott (2002), uses standard t statistics for the null hypothesis that there has been no change in the growth rates. But the critical values for this test are increased to account for the greater chance of erroneously finding a break when searching over multiple possible breakpoints. While there are episodes when the trend in the terms of trade appears to have changed, over the full sample the Sup(t) test is not able to find evidence of a statistically significant break.[4] However, the Sup(t) test only allows one break. To allow for the possibility that a single break was not found because multiple breaks exist, the Bai and Perron (1998, 2003) test was also applied. The Bai and Perron test gave inconsistent results, but overall was not supportive of a break in the trend, especially so after allowing for persistence. The results of these tests are available from the authors.

While a casual observation suggests the trend decline in the terms of trade may have accelerated in the second half of the century, at least before projected rises are included, statistical tests do not support this conclusion. Overall, these results indicate that there is at most a weak negative deterministic trend.

The statistically significant coefficients on the lagged terms of trade indicate that the terms of trade is relatively persistent. But these coefficients are downward biased because they are on the lagged dependent variable. To get a more accurate estimate of the persistence we also report the median unbiased estimates of the lag term, ρ-MU, based on Andrews (1993), which corrects for this bias. These results are shown in Table 3. Note that the downward bias in ρ is greater in the smaller samples. The half-lives of a shock to the terms of trade as determined by ρ-MU are shown in the last row of Table 3, together with 90 per cent confidence intervals. Shocks to the terms of trade are found to be transitory, consistent with the finding of stationarity. Over the full sample, half of a shock is found to dissipate within around three years, though in the second half of the century point estimates suggest that shocks were less persistent with half-lives of around two years. Interestingly, these results indicate less persistence than typically found for individual commodities (see for example Cashin, Liang and McDermott 2000).

2.1 Decomposing the Level of the Terms of Trade

After accounting for the persistence of shocks, the negative trend in the terms of trade is found to be very small. This is perhaps surprising given the composition of Australia's trade and the stylised fact of falling relative commodity prices. In this section we investigate why the trend in the terms of trade has not been greater. We focus on goods imports and exports since these prices relate more directly to the Prebisch-Singer hypothesis and the trend including services is only marginally different.

Figure 2 plots Australia's goods terms of trade together with relative commodity prices, the ratio of world commodity prices to world manufactures prices. This series is an extension of the Grilli-Yang data that are available from 1900 and have been used to highlight the Prebisch-Singer effect.

Figure 2: Relative Commodity Prices and Australia's Terms of Trade

The Australian terms of trade clearly experienced larger swings in the middle part of the 20th century than did the relative commodity prices. This relates to large price movements for specific commodities that represented a large portion of Australia's exports, such as the spikes in wool and metals prices in 1951 due to the Korean war. Despite these differing cycles, over the first three-quarters of the century the total change in the two series were remarkably similar. However, since the mid 1970s real commodity prices have fallen at a much faster rate than Australia's terms of trade.

To better understand the factors influencing the level of the terms of trade we consider import and export prices individually. Figure 3 shows the ratio of export prices to the Grilli-Yang commodity price index and the ratio of import prices to the world manufactures series (the ratios of the two numerators and of the two denominators from the series in Figure 2). These comparisons are meaningful because the majority of Australia's exports are commodities, while imports are mostly manufactures.

Figure 3: Relative Import and Export Prices

The ratio of Australian import prices to world manufactures prices has been remarkably constant for most of the past century. This is not so surprising given the high proportion of manufactures in Australia's import basket. But after a pickup in this ratio in the 1970s, in part due to higher oil prices, Australia's import prices have been falling relative to world manufactures prices.

The large swings in the ratio of Australian export prices to world commodity prices clouds an interpretation of trends. Nevertheless, it appears to have increased over the course of the century. This has supported the level of the terms of trade and accounts for the smaller trend in Australia's terms of trade than in the Grilli-Yang relative commodity price series.

2.1.1 Export price developments

In this section we explore reasons why Australia's export price series appears to have risen relative to world commodity prices.

Protopapadakis and Stoll (1986, p 350) suggest that the law of one price ‘is a usable approximation of the behavior of commodity prices for macroeconomic purposes’ in the long run. Given that the majority of Australian exports have been commodities, this implies that differences between the Australian export prices and world commodity prices must be due to compositional differences rather than different prices for identical commodities.[5] Related to this, Australia's exports have become significantly more diversified, both within commodity classes and into manufactures, over the past four decades. Figure 4 shows the value shares of Australia's goods exports over the past century for some major export classes.

Figure 4: Australia's Goods Export Composition

Clearly there have been some striking changes in Australia's export composition. Since the 1950s, wool's share of exports has been in sharp decline. Around its peak, wool averaged 39 per cent of goods exports in the period 1941–1951 but by 1994–2004 it was only 3 per cent of goods exports. The falling share of wool exports is largely explained by the slow growth in the volume of wool exports relative to other exports. Over this 50 year period, total goods export volumes increased over 15-fold, but wool export volumes less than doubled. The decline is also explained by the collapse in the price of wool following its peak in 1951 during the Korean War. Only during the late 1980s boom did wool regain its 1951 nominal price. In contrast over this period the nominal world commodity price index rose almost three-fold.

The sharp decline in the share of wool in exports through the 1960s marked the beginning of a dramatic change in the composition of Australian exports. Other primary rural commodities that had been the mainstay of Australian exports – meat, dairy, cereals and other food – also declined in share. Their place was taken by the rapid expansion of mineral commodity exports, notably coal & coke, and metal ores & scrap. In recent decades, manufactures have also become an increasingly important component of Australia's exports.

Smith (1987) suggests that the rapid increase in Australia's mineral exports beginning in the 1960s was a result of demand from Japan rather than increases in world mineral prices. Despite the low extraction costs of minerals in Australia, transport costs were sufficiently high to prevent the development of a viable export market for some commodities before the economic development of the Japanese economy. As evidence of this, Smith (1987) notes that Australia's traditional mineral exports had been high value-to-bulk commodities such as copper, lead and zinc. Japan's prominence in Australia's commodity exports at the time is illustrated by the fact that by 1969/70 Japan imported 65 per cent of Australia's metal ores, coal, gas and petroleum exports.[6]

The diversification of Australia's export base may have changed the growth rate of Australia's export price series and explain export prices outperforming world commodity prices. However, this does not appear to have been the case, at least from 1904–1975. Over this period the rural subcomponent of exports recorded only slightly slower price growth than the all goods price index, 3.38 per cent per annum versus 3.45 per cent.[7] And even the broader sub-index of commodities, rural, metals, coal and gold, experienced faster price growth than world commodity prices: 3.45 per cent versus 2.94 per cent. (These, and all prices hereafter, are in Australian dollars.)

The availability of disaggregated price data allow a more detailed examination of relative price performance for the period after 1975. Diversification into mineral exports began in the 1960s but it was not until the mid 1970s that these commodities materially contributed to faster growth in the prices of Australian exports. The first two columns in the top panel in Table 4 show that this diversification added about 0.4 per cent per annum to the rate of growth of export prices over the period 1975–2004. Diversification beyond this narrow grouping of rural goods, metals, coal and gold to other commodities resulted in slightly faster growth, as seen by comparing the second and third columns. While they are only a small share of exports, petroleum products and natural & manufactured gasses have made a material contribution to the faster growth from diversification (columns three versus four).

Table 4: Component Export Price Series
Average yearly percentage growth
Commodities Memo item: world commodity prices
Rural Rural, metals, coal & gold All commodities excluding petroleum & gas All commodities
1975–2004 3.0 3.4 3.5 3.7 3.2
1975–1990 6.4 7.1 7.3 7.2 5.6
1990–2004 −0.6 −0.5 −0.4 0.2 0.8
Manufactures All goods
  Chemicals Machinery & transport equipment Other manufactures All manufactures  
1975–2004 3.4 3.0 3.0 3.0 3.6
1975–1990 7.4 7.6 6.3 6.8 7.2
1990–2004 −0.7 −1.7 −0.3 −1.0 −0.2
Notes: Rural refers to cereals, dairy, dried & canned fruit, hides & tallow, meat, sugar and wool. The all goods series shown is a reconstructed series rather than the actual series to maintain consistency with the derived series for the subsets of goods. The correlation with the actual series is 99 per cent. Due to a series break, some goods from Statistical International Trade Classification (SITC) category 63 are included in category 24, and some goods from category 51 are included in category 28 from 1974/75 to 1977/78.

Diversification into manufactured exports actually reduced aggregate export price growth over this period (column four in the top panel versus column five in the bottom panel). Indeed, over the full 28 years none of the sub-components of manufactures exports have outperformed total goods exports. More recently, over the period 1990–2004, commodity export prices have risen somewhat less rapidly than world commodity prices, but the growth in total export prices has been dragged further down by manufactures exports.

In summary, the diversification of Australia's export base into goods with faster price growth than traditional rural commodities has substantially boosted the growth of export prices. This began with diversification into mineral exports in the 1960s and has continued with diversification into a broader set of commodities. At least since 1975, on average the broadening of exports into manufactures exports has not increased export price growth.

2.1.2 Import price developments

In this section we examine the fall in Australia's import prices relative to world manufactures prices since the mid 1980s.[8] The difference in the growth of these series is presumably attributable to compositional differences between the goods Australia imports and the world manufactures price index, again because Australia is likely to be a price taker for these goods on world markets.

Since 1985/86, around the time the downward trend in import prices relative to world manufactures prices became apparent, Australia's import prices of elaborately transformed manufactures (ETMs) have grown at a rate close to that of total import prices, −0.7 per cent versus −0.3 per cent. Also, ETMs accounted for 84 per cent of Australia's imports on average over this period. Hence, differential price growth for non-manufactures cannot explain the difference between the growth in prices of goods imported by Australia and world manufactures prices.

Over the period 1985/86 to 2003/04 there has been little change at the 2-digit SITC level of disaggregation in the type of manufactures imported by Australia. In contrast, the source of imports now differs substantially from those used to construct the world manufactures price index, which is based only on industrialised country manufactures. An increasing proportion of Australia's imports come from non-industrialised countries, notably China and the ASEAN countries (Figure 5). The share of imports from non-Japan Asia has increased for all major classes of manufactured goods (Figure 6). This suggests that Australia's import prices have risen at a slower rate than world manufactures prices because of the substitution to cheaper imports sourced from Asia.

Figure 5: Source of Australian Imports
Figure 6: Manufactures Imports from Non-Japan Asia


The increasing commodification of some manufactured goods may act to counter this effect. [3]

The test is sensitive at endpoints, which were excluded. The period examined for breaks was 1891–1984. [4]

Our own examination of comparable Australian and international commodity prices suggested there were important deviations in price growth over periods of up to 10 years, but that over the long run, price growth was equivalent. [5]

There were restrictions on the export of iron ore and magnesium from 1939–1960, however, Smith (1987) suggests their removal was not a dominant factor in the mineral boom. [6]

The rural subcomponent includes cereals, dairy, dried & canned fruit, hides & tallow, meat, sugar and wool. [7]

Differences in index construction methodology may account for some of the difference. The Australian goods import price series is a Paasche price index whereas the world manufactures price index is a periodically re-based fixed-weight index. This means that the Australian import price series is downward biased while the world manufactures series is upward biased. It is unlikely that these methodological differences could account for more than a small part of the difference between the series. [8]