Journal of Development Economics 78 (2005) 49 – 59

www.elsevier.com/locate/econbase

Testing for trends in the terms of trade between

primary commodities and manufactured goods
George P. Zanias*
Athens University of Economics and Business, Department of International and European Economic Studies,
76 Patission Street, Athens 104 34, Greece
Accepted 1 August 2004

Abstract

This paper joins the statistical debate on the terms of trade between primary commodities and
manufactured goods by contributing to the methodological discussion and presenting new evidence
using data series covering almost the whole of the 20th century. Using statistical tests that take into
account breaks in the series, it is found that over the 20th century the relative prices of primary
commodities dropped to nearly one-third of their level at the beginning of the century in two
binstallmentsQ, when random shocks led to structural breaks, and not in a gradual way as implied by
either a deterministic or stochastic trend. Possible reasons for the structural breaks and their policy
implications are discussed.
D 2005 Elsevier B.V. All rights reserved.

JEL classification: F14; O19

Keywords: Primary commodities; Terms of trade; Structural breaks

1. Introduction

Since Prebisch (1950) and Singer (1950) challenged the maintained view by classical
economists of improving terms of trade of primary commodities relative to manufactured
goods, and suggested instead a secular deterioration, research on this subject has been
thriving. A variety of empirical results have been obtained so far, which are capable of

* Tel.: +30 1 8203171; fax: +30 1 8214122.

E-mail address: zanias@aueb.gr.

0304-3878/$ - see front matter D 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.jdeveco.2004.08.005
50 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59

supporting conflicting views. According to the list of the most important studies on the
terms of trade issue, which was originally compiled by Nguyen (1981) and extended by
Diakosavvas and Scandizzo (1991), neither side of the controversy can claim victory. The
results obtained on secular trends differ according to the period explored, the definitions
used and the estimation techniques employed. From the studies included in this list, about
one-third confirm the Prebisch–Singer hypothesis and one fourth disprove it.
The basic task in these studies was to prove, or disprove, the existence of a time trend in
the data. In the cases that this is confirmed, economic forces exist which dpushT the terms
of trade to follow a continuously improving or worsening deterministic course, with
deviations from it being temporary. However, the predominant movement of a series over
time in one direction may also be due to certain shocks that have permanent effects on the
series, either in the form of a stochastic trend or of structural break(s).
The separation of trends from structural breaks in the series requires carefully designed
tests. This task is pursued in this paper for the case of commodity terms of trade using a
data set covering almost the whole 20th century (1900–1998). This data set is probably the
best that exists on this issue, and was developed at the World Bank by Grilli and Yang
(1988) for the period 1900–1986 and extended to 1998 at the IMF. The original GY data
set (1900–1986) has so far been used by researchers, including the compilers themselves,
in order to provide more accurate evidence on the evolution of the primary commodity
terms of trade. The compilers of the original data set, using the dtraditionalT approach
confirm the existence of a negative time trend in the relative primary commodity prices,
which is implicit in the works of Prebisch and Singer.
Among the other researchers who used the GY data set, Cuddington and Urzua (1989)
[CU] introduced the distinction between deterministic and stochastic trends as well as the
confounding effects of possible structural breaks when conducting unit root tests. CU
suggested the existence of a structural break in 1920. Powell (1991), also, recognized the
possibility of structural shifts in the terms of trade series and suggested three breaks, one in
1920 and two more at later dates. In both papers the timing of the shifts is selected in an ad
hoc manner, while Powel, in addition, uses test statistics that do not take the shifts into
account and the shift variables are arbitrarily constructed not allowing the measurement of
the impact of the different structural breaks.
In comparison to previous researchers, this paper: first, uses an extended GY data set
covering almost the whole 20th century. Second, endogenizes the search for structural
breaks. Third, uses more appropriate and more powerful tests for breaks. More
specifically, the Lumsdaine–Papell unit root test is used which tests for the presence of
two structural breaks at the same time. In this way, testing for trends in the commodity
terms of trade leads to some very interesting results about the way the commodity terms of
trade evolve over time and which has totally different policy implications than those of a
deterministic or stochastic trend.

2. The data set and the testing procedure

The GY series of the commodities terms of trade (COMTT) are the ratio of an index of
non-fuel commodity prices (COM) and a price index of manufactures (MUV). COM is an
 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59 51

5.4

5.2

5.0 LCOMTT
Logarithms

4.8

4.6

4.4

4.2

4.0
00 10 20 30 40 50 60 70 80 90
Year

Fig. 1. Grilli and Yang terms of trade series (1900–1998).

index of prices of 24 internationally traded non-fuel commodities weighted by the 1977–

1979 values of world exports of each commodity. MUV is the United Nations index of
export unit values of manufactures, in which GY filled the two gaps that this index had
during 1914–1920 and 1939–1947. The original GY data set covered the period 1900–1986.
Cashin and McDermott (2002) extended1 this data set to 1998, using the GY weights and
the IMF commodity price indices, which are very close to those of the World Bank.
The logarithm of the terms of trade index LCOMTT [= LOG (COM/MUV)] appears in
Fig. 1. The beyeball testQ clearly suggests downward movement in the commodity terms of
trade over time but raises the question about whether this is due to: (1) a deterministic
trend and/or (2) a stochastic trend and/or (3) structural breaks in the level or trend. The
nature of the downward movement in the commodity terms of trade has important policy
implications.
The testing procedure should encompass the above three possibilities of non-stationary
behaviour (e.g. a downward movement) of the terms of trade. Thus, if a structural break
occurs at time 1 b T b b t, the relevant set of testing hypotheses is the following:
Null: LCOMTTt ¼ l þ LCOMTTt1 þ et ð1aÞ

Alternative: LCOMTTt ¼ l þ bt þ ðl2  l1 ÞDUt þ et ð1bÞ

OR: LCOMTTt ¼ l þ b1 t þ ðl2  l1 ÞDUt þ ðb2  b1 ÞDTt þ et ð1cÞ

where: DUt = 1 if t N T b and 0 otherwise; DTt = t  T b if t N T b and 0 otherwise; l 2  l 1
denotes the magnitude of the change in the intercept and b 2  b 1 the change in the slope of
the trend function occurring at time T b.

1
Cashin and McDermott used the extended data set in this study, but the actual data were provided through
personal communication with Paul Cashin.
52 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59

No breaks are assumed under the null in the above set of hypotheses. The sampling
distribution used in the testing procedure has to reflect this choice. Zivot and Andrews
(1992) and Perron (1997) developed such distributions for the case of one structural break,
and Lumsdaine and Papell (1997) for the case of two breaks.
The beyeball testQ suggests that two structural breaks may have occurred in the terms of
trade series. Furthermore, the timing of the breaks cannot be determined a priori. The
testing procedure developed by Lumsdaine and Papell takes into account both these
features and it is used here. Thus, it tests for two structural breaks in the series over the
time period considered, and it uses a search procedure for breaks because they are treated
as unknown occurrences.
The testing procedure requires the estimation of the following augmented regression
equations to test null against the first and the second alternative hypothesis respectively:



D ðLCOMTÞt ¼ l̂þ bˆ t þ #̂1 DU1t kˆ 1 þ #̂
#2 DU2t kˆ 2 þ â LCOMTTt1
X
k
þ ĉj DðLCOMTTÞtj þ ê1t ð2aÞ
j¼1


    
DðLCOMTÞt ¼ l̂ þ bˆ t þ #̂1 DU1 kˆ 1 þ #̂2 DU2 kˆ 2 þ þ fˆ 1 TD1 kˆ 1 þ fˆ2 TD2 kˆ 2
X
k
þ âLCOMTTt1 þ ĉj DðLCOMTTÞtj þ ê2t ð2bÞ
j¼1

where k 1 = T b1/T and k 2 = T b2/T are the points in time that the two breaks take place.
Selection of the breakpoints requires the estimation of the above equations for all possible
breakpoints. This means estimation of Eqs. (2a) and (2b) by ordinary least squares for all
distinct pairs of values of (j 1,j 2) for j 1,j 2 = j 2 / T,. . .,j (T  1) / T and j 1 p j 2. The
breakpoints selected are the ones that give the least favorable result for the null
hypothesis. This means selecting from all regressions the minimum t-statistic for testing
a = 1. The so-estimated t-statistic is then compared to the critical values calculated by
Lumsdaine and Papell.
In order to minimize possible problems caused by an incorrect choice of the lag-length
(see Cuddington and Liang, 2000), the general-to-specific method is used here, and
throughout the paper, to select the lag length (k). This means, setting k = k max and working
backward until the t-statistic on the coefficients of the lag is greater than pre-specified
value, set here in the neighbourhood of 1.60, which corresponds to the 10% level of
significance. This sequential selection procedure has the advantage over information-based
rules (Akaike and Schwartz) of showing less size distortions while having comparable
power (Ng and Perron, 1995).
The search procedure produces the smallest t-value for a when k 1 = 1920 and k 2 = 1984.
Thus, the first break occurred in 1920, which was also found by Cuddington and Urzua,
and the second in 1984, which could not be found by previous researchers who worked on
the GY data set because their sample ended in 1986 and the permanent nature of the 1984
break could not be confirmed. The estimations of the testing equation for breaks in 1920
 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59 53

and 1984 are the following (t-values appear in the brackets below the coefficients and the
Lumsdaine and Papell critical values for two structural breaks below each equation):

DðLCOMTTÞt ¼ 3:19  0:0007t  0:21DU1t  0:23DU2t

ð7:23Þ ð1:11Þ ð4:92Þ ð5:17Þ

 0:63LCOMTTt1 þ 0:28DðLCOMTTÞt1 ð3aÞ

ð7:24Þ ð3:05Þ

Box-Ljung Q 12 = 6.13 ( p = 0.91); critical values: 6.24 (5%), 6.94 (1%)

DðLCOMTTÞt ¼ 3:37 þ 0:005t  0:28DU1t  0:23DU2t  0:006TD1t

ð7:31Þ ð1:26Þ ð4:56Þ ð3:80Þ ð1:45Þ

 0:002TD2t  0:68LCOMTTt1 þ 0:31DðLCOMTTÞt1 ð3bÞ

ð0:36Þ ð7:26Þ ð3:29Þ

Box-Ljung Q 12 = 9.23 ( p = 0.68); critical values: 6.82 (5%), 7.34 (1%).

According to these results, the unit root hypothesis is rejected in both cases, at the 1%
level of significance in the first case and close to this level in the second case. Therefore,
the terms of trade series do not follow a stochastic trend. The only shocks that had a
permanent effect on the series are the ones that led to the two structural breaks in 1920 and
1984. Therefore, the series either follow a deterministic trend or they are stationary. The
statistical test on the time trend coefficient is conditional on the presence of a unit root.
Because the null hypothesis of a unit root is rejected, one is tempted to use the student-t
distribution to test the statistical significance of the time trend coefficients, which indicates
that the coefficients of all the trend variables are statistically insignificant. This conclusion
becomes stronger if we use the, even higher, critical values of the Augmented Dickey
Fuller test (= 2.79 at the 5% level of significance), which are conditional on the presence of
a unit root.
Additionally, and since the unit root hypothesis has been rejected, a trend stationary
(TS) model with the appropriate structural breaks and correction for serial correlation is
estimated:

LCOMTTt ¼ 5:07  0:0013t  0:36DU1t  0:32DU2t þ et ð4Þ

ð124:3Þ ð1:30Þ ð6:30Þ ð5:37Þ

where the disturbance term follows an AR(2) process:

et ¼ 0:73 et1  0:27 et2 þ ut ð5Þ
7:22Þ ð2:66Þ

Q 12 = 3.48 ( p = 0.97); R 2 = 0.88.

The estimated TS model clearly produces a statistically insignificant time trend
coefficient and highly significant coefficients on the two intercept-shift dummy
variables.2

2
Similar are the results when allowing for changes in the trend slope at the time of the breaks.
54 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59

3. Terms of trade and the structural breaks

The above testing procedure rejects the presence of both a stochastic and a
deterministic trend in the terms of trade series and suggests that the bdownward
movementQ, which the beyeball testQ indicated, is due to two structural breaks that took
place in 1920 and 1984. This result differs from those which disregard the possibility of
breaks, and which are obtained either by estimating a TS model or applying a more
conventional unit root test, like the Augmented Dickey-Fuller (ADF) test3. Also, the
results confirm, through endogenizing the search procedure, the 1920 structural break,
which had been indicated in ad hoc manner by previous researchers, and finds a second
break in 1984.
In the absence of both a stochastic and a deterministic trend, the evolution of the terms
of trade series is described by the following estimated equation, which is appropriately
corrected for autocorrelation:

LCOMTTt ¼ 5:05  0:41DU1t  0:36DU2t þ et ð6Þ

ð131:7Þ ð9:55Þ ð7:65Þ

where:

et ¼ 0:74 et1  0:26 et2 þ ut ð7Þ

ð7:30Þ ð2:61Þ

R 2 = 0.88; F = 173.2; Q (12) = 3.76 ( p = 0.96); Jarque-Bera = 0.40 ( p = 0.82).

According to these results, the 1920 structural break is responsible for an
unfavorable 41% reduction in the primary commodity terms of trade and the 1984
break for a 36% reduction compared to the level established after the first break. The
cumulative effect is a reduction in the terms of trade at the end of the 20th century to
a level equal to nearly one-third of that existing at the beginning of the century (62%
drop). The terms of trade series together with these two structural breaks are shown in
Fig. 2.
What has caused these structural breaks which led to such deterioration in the
primary commodity terms of trade last century? Speculation is unavoidable in trying to

3
The ADF can only be used to test the null of a unit root against stationarity or trend-stationarity.
Application to the extended GY data set produces the following result (t-values in the brackets below the
coefficients):

DðLCOMTTÞt ¼ 1:50  0:0022t  0:30 LCOMTTt1 þ 0:16DðLCOMTTÞt1 ð8Þ

ð4:05Þ ð3:54Þ ð4:05Þ ð1:52Þ

Box-Ljung Q 12=4.8 ( p = 0.96); A3=14.5.

The existence of a unit root is rejected at the 1% level of statistical significance (critical value = 4.04).
A deterministic trend is statistically significant at the 1% level of significance (critical value = 3.53). This
result is also confirmed by the A3 Dickey-Fuller statistic (1% critical value to test for unit root and no
trend = 8.73). This makes the commodity terms of trade to be a stationary variable around a negative
deterministic trend, a finding that supports the Prebisch–Singer thesis.
 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59 55

5.4

5.2
Logarithms
5.0 LCOMTT

4.8

4.6

4.4

4.2

4.0
00 10 20 30 40 50 60 70 80 90
Year

Fig. 2. Commodity terms of trade and structural breaks.

give a plausible answer to this question. One possible explanation, suggested also by
Powell, which however needs further exploration by future researchers, is associated
with the observation that both breaks took place after a commodity price boom. With
regard to the first structural break, it is possible that, stimulated by the boost in
commodity prices with which the First World War became associated, productivity
increases were introduced that led to the built up of an excess production capacity not
commensurate to the changes in demand at that time. The disequilibrium was rectified
by a move to a new equilibrium position associated with lower primary commodity
prices. The drop in the prices of manufactured goods was much smaller and the 1920
jump in the terms of trade obtains.
Expansion in the production capacity well beyond that dictated by changes in
demand was also built up in the 1970s following the 1972–1973 oil crisis and the
1972–1974 world food crisis that led to sharp but temporary rises in the prices of
agricultural products.4 Production capacity expanded in order to meet the sharp increase in
import demand. In fact, imports of agricultural products increased more than three-fold in
a very short period from the early 1970s to 1980. The surge in demand was halted as
developing countries were facing a slow down in income growth and had to service an
enlarged external debt. Large countries like China and India managed to raise their
domestic production and the European Community reduced its import requirements
(Alexandratos, 1988). The resulting collapse in the international prices of agricultural
commodities provided the impetus to bring for the first time this sector (with a minor
exception in 1962) into the GATT talks for trade barrier reduction (Uruguay Round). Since

4
The prices, for example, of cereals doubled in 1973–1974.
56 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59

then, international primary commodity prices have been stabilized at a lower equilibrium
level.

4. Concluding remarks

In sum, after accounting two structural breaks in 1920 and 1984, the terms of trade
series is stationary; it does not follow a negative deterministic or stochastic trend. This,
however, does not disprove the Prebisch–Singer thesis with regard to the direction of the
barter terms of trade long run movement. Although the relative prices of primary
commodities (compared to the manufactured goods) have not been falling gradually in
the way implied by a time trend, two major negative structural breaks last century led to
a decline in the relative prices of primary commodities. These downward shifts brought
the terms of trade down to nearly one-third of their level at the beginning of the century.
This is a very large deterioration in the terms of trade. In the intervening period of more
than six decades between the two breaks, and despite the fluctuations, the mean level of
the terms of trade was maintained.
This finding has very important implications especially for the developing countries,
which rely heavily on earnings from primary commodity exports to finance their
development process. Specialization in primary commodity production seems to be
detrimental in the long run. The finding of this study that the decline in the primary
commodity terms of trade takes place in binstallmentsQ rather than in a gradual way, has
further serious policy implications since, unlike gradual trends, the random shocks that
are capable of having a permanent effect (structural break) cannot be easily predicted.
Failure to predict the shocks may have serious consequences, as was the case with the
more recent structural break. Thus, false reading about the future evolution of export
earnings compounded by relatively low real interest rates led to a surge in external
borrowing by developing countries and the well known debt problems of the 1980s
following the collapse in international commodity prices.
Such behavior of primary commodity terms of trade suggests greater export
diversification by the developing countries. This is a risk reducing and export revenue
increasing policy, and it should be a prime goal. Many developing countries have done
considerable progress in this direction, especially after the last debt crisis. Of course, not
all developing countries are affected in the same manner, since there may be specific
commodities that exhibit even a positive trend. To the extent that declining terms of
trade exist for the agricultural products produced by developed countries, this has
serious implications for the funds devoted to farm support to sustain certain income
levels as well as for the design of farm adjustment policies.

Acknowledgements

This paper is based on research supported by the FAO. Useful comments from
Fernando Zegarra and Jacques Vercueil of the FAO and, especially, from an anonymous
referee are gratefully recognized.
 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59 57

Appendix A. Price indexes of primary commodities (COM) and manufactured goods

(MUV), and barter terms of trade (COMTT) 1900–1998
Year COM MUV COMTT
1900 19.3090 14.6070 132.1900
1901 18.2360 13.8580 131.5919
1902 18.1450 13.4830 134.5769
1903 19.0060 13.4830 140.9627
1904 20.5860 13.8580 148.5496
1905 21.6210 13.8580 156.0182
1906 21.6100 14.6070 147.9428
1907 22.7570 15.3560 148.1961
1908 20.4270 14.2320 143.5287
1909 21.5540 14.2320 151.4474
1910 22.6300 14.2320 159.0079
1911 21.9090 14.2320 153.9418
1912 22.6400 14.6070 154.9942
1913 20.4610 14.6070 140.0767
1914 20.2100 13.8580 145.8363
1915 24.4680 14.2320 171.9224
1916 31.9330 17.6030 181.4066
1917 39.3960 20.9740 187.8326
1918 42.0280 25.4680 165.0228
1919 39.2080 26.9660 145.3979
1920 41.9510 28.8390 145.4662
1921 21.3560 24.3450 87.7223
1922 21.9100 21.7230 100.8608
1923 26.4070 21.7230 121.5624
1924 26.5210 21.7230 122.0872
1925 29.3810 22.0970 132.9638
1926 25.7580 20.9740 122.8092
1927 25.1430 19.8500 126.6650
1928 24.4230 19.8500 123.0378
1929 23.2660 19.1010 121.8051
1930 18.2770 18.7270 97.5970
1931 13.6100 15.3560 88.6298
1932 10.7970 12.7340 84.7887
1933 12.5910 14.2320 88.4696
1934 15.7630 16.8540 93.5267
1935 17.2940 16.4790 104.9457
1936 18.4180 16.4790 111.7665
1937 21.3610 16.8540 126.7414
1938 16.5520 17.6030 94.0294
1939 16.0190 16.1050 99.4660
1940 17.2370 17.6030 97.9208
1941 20.0930 18.7270 107.2943
1942 23.0730 21.7230 106.2146
1943 24.2830 24.3450 99.7453
1944 25.2430 27.7150 91.0806
1945 25.8320 28.4640 90.7532
1946 31.2320 28.8390 108.2978
1947 40.3890 34.8310 115.9570
1948 38.7220 35.5810 108.8277
(continued on next page)
58 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59

Appendix A (continued)
Year COM MUV COMTT
1949 35.8450 33.3330 107.5361
1950 39.2630 30.3370 129.4228
1951 48.0930 35.9550 133.7589
1952 40.5080 36.7040 110.3640
1953 37.8970 35.2060 107.6436
1954 38.5650 34.4570 111.9221
1955 38.2330 34.8310 109.7672
1956 39.8950 36.3300 109.8128
1957 40.1080 36.7040 109.2742
1958 36.2310 36.3300 99.7275
1959 37.1130 36.3300 102.1552
1960 37.3270 37.0790 100.6688
1961 36.4660 37.4530 97.3647
1962 36.4660 37.4530 97.3647
1963 41.4190 37.4530 110.5893
1964 41.0460 38.2020 107.4446
1965 38.1190 38.9510 97.8639
1966 37.9350 39.7000 95.5541
1967 36.8460 39.7000 92.8110
1968 37.4310 39.3260 95.1813
1969 39.7610 40.4490 98.2990
1970 42.2010 42.6970 98.8383
1971 42.3240 45.3180 93.3933
1972 46.6250 48.6890 95.7608
1973 69.4720 58.8010 118.1477
1974 102.4100 71.1610 143.9131
1975 85.1560 79.0260 107.7569
1976 83.1100 78.6520 105.6680
1977 93.1250 86.5170 107.6378
1978 93.6270 98.8760 94.6913
1979 113.2500 114.610 98.8133
1980 138.8300 125.470 110.6480
1981 117.9400 119.1000 99.0260
1982 96.7840 115.7300 83.6291
1983 102.7800 110.4900 93.0219
1984 103.5400 108.6100 95.3319
1985 91.2680 109.5900 83.2813
1986 88.3580 130.3000 67.8112
1987 94.1814 146.7000 64.2000
1988 117.0870 156.8700 74.6400
1989 121.7570 155.6000 78.2500
1990 118.5769 172.0500 68.9200
1991 107.8065 172.0500 62.6600
1992 108.3645 177.2400 61.1400
1993 100.4330 166.8600 60.1900
1994 121.2223 170.2800 71.1900
1995 143.2787 185.4500 77.2600
1996 132.5100 178.7100 74.1500
1997 121.5606 166.9100 72.8300
1998 106.1538 163.5400 64.9100
Source: 1900–1986: Grilli and Yang (1988).
1987–1998: Cashin and McDermott (2002).
 G.P. Zanias / Journal of Development Economics 78 (2005) 49–59 59

References

Alexandratos, N. (Ed.), 1988. Agriculture Towards 2000. An FAO Study. Belhaven Press, London.
Cashin, P., McDermott, C.J., 2002. The long-run behaviour of commodity prices: small trends and big variability.
IMF Staff Papers 49, 175 – 199.
Cuddington, J.T., Liang, H., 2000. Purchasing power parity over two centuries? Journal of International Money
and Finance 19, 753 – 757.
Cuddington, J.T., Urzua, C.M., 1989. Trends and cycles in the net barter terms of trade: a new approach.
Economic Journal 99, 426 – 442.
Diakosavvas, D., Scandizzo, P.L., 1991. Trends in the terms of trade of primary commodities, 1900–1982.
Economic Development and Cultural Change 39, 231 – 264.
Grilli, E.R., Yang, M.C., 1988. Primary commodity prices, manufactured goods prices, and terms of trade of
developing countries: what does the long-run show. World Bank Review 2, 1 – 48.
Lumsdaine, R.L., Papell, D.H., 1997. Multiple trend breaks and the unit root Hypothesis. The Review of
Economics and Statistics 79, 212 – 217.
Ng, S., Perron, P., 1995. Unit root tests in ARMA model with data dependent methods for the selection of the
truncation lag. Journal of the American Statistical Association 90, 268 – 281.
Nguyen, T., 1981. Trends in the terms of trade of LDCs. Journal of Economic Studies 8, 46 – 56.
Perron, P., 1997. Further evidence on breaking trend functions in macroeconomic variables. Journal of
Econometrics 80, 355 – 385.
Powell, A., 1991. Commodity and developing countries terms of trade: what does the long-run show? The
Economic Journal 101, 1485 – 1496.
Prebisch, R., 1950. The Economic Development of Latin America and its Principal Problems. United Nations,
New York.
Singer, H., 1950. The distribution of gains between investing and borrowing countries. American Economic
Review, Papers and Proceedings 40, 473 – 485.
Zivot, E., Andrews, D.W.K., 1992. Further evidence on the great crash, the oil price shock and the unit root
hypothesis. Journal of Business and Economic Statistics 10, 251 – 270.