1995
es
Corden, W. Max
Corden, W. Max
Una zona de libre comercio en el Hemisferio Occidental: posibles implicancias para América Latina
En: La liberalización del comercio en el Hemisferio Occidental - Washington, DC : BID/CEPAL, 1995 - p. 13-40
2014-01-02T14:51:16Z
hdl:11362/36988
c e pa l
review
110
•
august
2013
109
Terms of trade and output fluctuations
in Colombia
Gonzalo Hernández
ABSTRACT
This paper explores the impact of the terms of trade on output fluctuations in
Colombia, a developing country where as much as 62% of export earnings come
from just four commodities: oil (42%), coal (14%), coffee (5%), and nickel (1%). This
research was prompted by: the particular role of short-run fluctuations in developing
economies, the fact that the Colombian terms of trade are procyclical, and the
discussion on economic policies towards sterilization of the effects of commodity
prices. Following time series analysis for the period 1994 -2011, robust evidence was
found indicating that around one third of Colombia’s quarterly growth is attributable
to changes in the terms of trade.
KEYWORDS
International trade, terms of trade, economic growth, productivity, time series analysis, Colombia
JEL CLASSIFICATION
F41, F44, O54
AUTHOR
Gonzalo Hernández, Professor, Department of Economics, Pontificia Universidad Javeriana, Bogota,
Colombia. Ph.D candidate, Department of Economics, University of Massachusetts, Amherst, United States,
gonzalo@econs.umass.edu
110
c e pa l
review
110
•
august
2013
I
Introduction
The role of short-run output fluctuations in developing
economies is particularly important. Developing countries
are usually more exposed than developed economies to
the effects of macroeconomic ups and downs. In addition,
welfare implications may be asymmetrical, depending on
the degree of development. Some examples of a possible
uneven effect are: first, that developing economies lack
the proper social safety nets to mitigate the impact of bad
phases on the poorest population; second, poverty and
unemployment in developing countries make people less
capable of adjusting their consumption when temporary
shocks appear;1 and third, the more variable tax base
may constrain both the ability of the public sector to
implement long-run projects necessary to remove the
obstacles that hinder the development of these economies
as well as the responsiveness of short-run fiscal policy.
This study focuses on the terms of trade to explain
these output fluctuations. This decision is motivated by
the literature on development macroeconomics based on
a small open economy framework.2 In particular, the
dependent economy model (with its three goods variant:
exportable, importable and non-tradable) assumes that
small economies face an infinitely elastic global demand
for their goods, and an infinitely elastic supply of imported
goods.3 This means that prices of exports and imports
are determined in the international markets where the
The author would like to acknowledge helpful comments from
Arslan Razmi, Gerald Epstein, Christian Rojas, Leila Davis, Martin
Rapetti, and an anonymous referee, on previous drafts. The views
expressed in this article are those of the authors and do not necessarily
reflect the views of eclac.
1 Economic theory usually assumes that more volatile consumption
decreases individuals’ utility in the presence of risk and incomplete
financial markets.
2 See, for example, Agénor and Montiel (2008).
3 For this study, the dependent economy with three goods seems
to be a more convenient framework than the Mundell-Fleming
model, where the terms of trade, when variable, are endogenous. The
endogeneity in the Mundell-Fleming model occurs because there is
some degree of producer market power in the exportable good. The
price of the exportable good may be altered by internal conditions (that
is, domestic demand) even if the economy is a price taker regarding
the importable goods. Likewise, the two goods dependent economy
model (traded and nontraded) has its own limitation: both exportable
and importable goods are aggregated in a composite good (the traded
good). Therefore, the variability of the terms of trade is not defined
and cannot be the origin of macroeconomic fluctuations. See, for
example, Greenwood (1984) and Buiter (1988), for other dependent
economy model specifications.
domestic economy has no market power. The framework
predicts that external shocks to terms of trade may be an
important source of output fluctuations in the domestic
economy. An improvement in the terms of trade, say,
because of a boom in commodity prices, works as an
incentive to expand output in the sectors that benefit from
a higher price. However, the shock may result also in an
appreciation of the real exchange rate that increases real
wages in the sectors that compete with importable goods.
Therefore, the initial aggregate output increase might be
offset by the loss of competitiveness in the sectors that
compete with importable goods (Dutch Disease). The same
mechanism may be easily extended to other exportable
goods. The net result, however, depends theoretically on
critical assumptions regarding the labour markets and the
degree of price flexibility (market-clearing conditions).
The most common assumption is that the non-tradable
sector clears through price variations rather than through
an adjustment in output. Furthermore, whether these
effects are displayed in the short run will depend not
only on the type of market adjustment but also on its
speed, the reaction of the economic policy authorities
to changes in this relative price, the degree of openness,
the degree of specialization in exportable goods, and the
exchange-rate regime, among other elements.
Some facts justify the selection of Colombia as a
case study. First, recent Colombian exporting structure
seems to support the role of exogenous terms of trade
as in the three goods model. Annual data for 2010 show
that Colombian exports (62%) are concentrated mainly in
four commodities in respect of which Colombian market
power is negligible: petroleum and derivatives (42%), coal
(14%), coffee (5%), and nickel (1%). Second, quarterly
information for the period 1994-2011 reports a positive
correlation between quarterly variations in terms of trade
and quarterly growth of gross domestic product (gdp)
equal to 0.35. This magnitude is important relative to other
studies on developing economies (see section II). Lastly,
the period 1994-2011 shows high variability in both the
terms of trade and gdp. This variability is useful for testing
the validity of the results from the time series analysis.
Although this study is limited to aggregate results,
there are other channels that can illustrate the relevance of
terms of trade in Colombia’s economic performance in the
short run. Let’s take again a commodity price boom as an
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
review
example. Once the commodity price rises, extra profits will
be generated for the firms involved directly or indirectly in
the production of that commodity. Thus, this shock fosters
the expansion of consumption and output in other sectors.
In addition, a higher level of wealth allows investors to
access financial credit more easily. This credit is available
due to the greater availability of foreign currency, which
relaxes the monetary constraints. Therefore, the process
boosts credit, investment and consumption, and also
profits for the financial system, which currently accounts
for around 18% of total Colombian value added. This
mechanism is clearly plausible in Colombia where there
has been a large accumulation of international reserves,
and where, despite the central bank’s inflation targeting
policy, some interventions have been made to curb the
appreciation of the exchange rate.
Another reason for the procyclical terms of trade
in Colombia may be found in the public sector. Around
60% of the total volume of exported oil is exported by
Empresa Colombiana de Petróleo (ecopetrol). Some of
the revenues obtained by this institution make up part of
the revenues of the non-financial public sector. In addition,
it is reasonable to expect that by increasing profits of the
firms, and stimulating the economy, a commodity price
boom will also expand tax revenues. The result is not
necessarily a fiscal surplus. For instance, Kaminsky (2010)
finds evidence of a procyclical fiscal policy in middle
income countries when terms-of-trade shocks occur.4
Regarding international trade, after the United States
and the European Union, the Bolivarian Republic of
Venezuela and Ecuador are the most important markets
for Colombian exports. These two countries are oil
exporters and net buyers of Colombian manufactures.
This means that a commodity price boom that increases
the income of these trading partners may also increase the
demand for Colombian products. However, preliminary
evidence shows that the Colombian current account is
not positively correlated with the terms of trade.5
This paper sets out to resolve the theoretically
ambiguous relationship between terms of trade and output.
Specifically, this paper presents a time series analysis
that examines the relationship between quarterly gdp
growth and quarterly variations in the terms of trade.
110
•
august
2013
For that purpose, a price index has been constructed for
the four main exportable commodities, and a simple
econometric methodology (Box-Jenkins methodology)
is used, which is consistent with: the exogeneity of
the Colombian terms of trade, the non-co-integration
between gdp and terms of trade, and the stationarity of
the key variables. The study offers different robustness
tests, starting with the inclusion of relevant control
variables whose absence may cause a potential bias in
the estimate for terms of trade. For example, real and
nominal exchanges rates are two such control variables
because a negative effect of the variations in the terms of
trade on short-run output fluctuations could be associated
with a Dutch disease mechanism. Nevertheless, it is not
clear a priori either that positive terms of trade shocks
result in an appreciation or that an appreciation is going
to decrease aggregate output unambiguously. First, nontraded goods production could increase with the shock
while the expansion of the real income would be adjusted
by a change in output rather than by a change in prices.
Second, an eventual appreciation of the nominal exchange
rate, given a larger supply of foreign currency, could
have expansionary effects on output just as a nominal
devaluation may have contractionary effects. In a seminal
theoretical model, Krugman and Taylor (1978) describe
this possibility.6 Among different mechanisms presented
by the authors, one of them, following the Kaleckian
tradition, states that an appreciation may redistribute
income from profits and rent to wages. The reduction in
the price of imported inputs is automatically translated
into a reduction in the price of home goods, which
increases real wages. Because the marginal propensity
to consume is higher for workers than for capitalists, the
redistribution from wages to profits increases aggregate
demand and domestic output.
Thus, following a review of the related literature
in section II, the empirical strategy in section III is
essential to evaluate the direct effect of changes in the
terms of trade after taking into account eventual indirect
effects through other variables. Concluding remarks are
presented in section IV.
6
4 See, for example, Tornell and Lane (1999) and Frankel (2010)
for institutional aspects explaining procyclical public expenditure
in developing economies. For Latin America, see, for example,
Medina (2010).
5 See Obstfeld (1982), Svensson and Razin (1983) and Kent and
Cashin (2003) for discussions about the effects of the terms of trade
on the balance-of-payments current account.
111
See Lizondo and Montiel (1989) for a detailed overview of the
theory on contractionary effects of devaluation applied to developing
countries. Razmi (2007) extends the theoretical framework of Krugman
and Taylor (1978). This extension, including the role of transnational
corporations and the type of trading partners for exports (developing or
industrialized economy), suggests that the likelihood of contractionary
short-run effects of devaluation may be greater for developing
economies. As an opposite example, Reinhart and Reinhart (1991)
find that a devaluation is expansionary in the short run in Colombia
in a simulation-based model with a neo-Keynesian structure.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
112
c e pa l
review
110
•
august
2013
II
Related literature
Empirical evidence on the effects of the terms of trade
on output fluctuations in developing economies may
be classified in three groups: (i) studies that describe a
correlation between business cycles and cycles in terms
of trade as a stylized fact; (ii) simulation-based models;
and (iii) vector autoregression models (var).
Agénor, McDermott and Prasad (2000) find, for
instance, a strong positive correlation for Colombia,
Mexico and the Republic of Korea between the cyclical
components of industrial output and the terms of trade (with
both the Hodrick Prescott and the band-pass methodologies
using quarterly data). Also in this group of papers, Parra
(2008), with quarterly data from 1994 to 2007, reports a
correlation equal to 0.24 for Colombia, and Mahadeva and
Gómez (2009), a positive correlation between the terms
of trade and real gdp per capita for Colombia equal to
0.32 (using annual data for 1970-2007).7 However, this
type of stylized fact becomes more persuasive when it is
used either for the calibration of simulated-based models
or for the specification of an econometric model.
For instance, Mendoza (1995), in a seminal work in
the second category, not only reports a positive correlation
between terms of trade and gdp but also claims that his
intertemporal model predicts that terms of trade shocks
may account for between 37% and 56% of the actual
variability of gdp in developing countries. This outcome
depends of course on the particular setup of his three
goods model (exportable, importable, and non-traded
goods). In that framework, the dominant effect that
explains the short-run effect of the terms of trade on
output is basically that terms-of-trade gains induce an
increase in the marginal profitability of the exportable
sector, which fosters an investment boom in that sector.
Investment corresponds to an international and domestic
reallocation of capital where the importable goods sector
is the only source of domestic capital (not the non-traded
sector). On the other hand, labour supply is inelastic in
traded-sector industries, and the labour supply response
in the non-traded sector is assumed to be negligible. After
the short-run impact, adjustment mechanisms start to
work to drive the economy to a long-run equilibrium,
which is by definition equal to the initial equilibrium.
7
See Rand and Tarp (2002) for a description of stylized facts of the
business cycles in developing countries.
The adjustment of the real exchange rate towards its
long-run equilibrium reduces the short-run interest rate
differential, and thus, the foreign capital that entered the
domestic economy during the investment boom flows
back out. As expected, the initial gdp boom weakens.
Although Mendoza’s framework (1995) presents a
plausible scenario for the positive correlation between
the terms of trade and gdp in the short run, different
theoretical assumptions could obviously tell a different
story. Indeed, empirically, in his own sample, some
countries displayed a negative correlation: Algeria
(-0.234), Democratic Republic of the Congo (-0.107),
Egypt (-0.455), Philippines (-0.285) and Tunisia (-0.309).
These cases are not, however, covered by the general
equilibrium model in his paper.
Kose and Riezman (1999) and Kose (2002) offer other
examples of how empirical evidence may be conditioned
by the particular theoretical setup. Kose and Riezman
(1999), developing a general equilibrium model for a
small open African economy with two sectors (exportable
primary goods and non-traded goods), find that world price
shocks can explain around 45% of output fluctuations,
basically because both the primary good and the nontraded sectors use imported capital goods as factors of
production. Therefore, a decline in the international prices
of imports leads to an expansion of aggregate output. On
the other hand, Kose (2002) finds that disturbances in
the prices of capital goods and intermediate goods may
account for 87.6% of the output variability. The greater
effect in this case occurs because the author focuses on
main export and import prices (which are more sensitive
than terms of trade to productivity shocks), and the role
of intermediate inputs in the non-traded sector, which,
according to his small open economy model, does not
face any limit on the supply of capital.
The third group of studies have used var techniques to
examine the effects of terms of trade on output fluctuations
in developing countries They include Hoffmaister, Roldós
and Wickham (1998); Hoffmaister and Roldós (2001);
Ahmed (2003); Broda (2004); Izquierdo, Romero and
Talvi (2007), and Raddatz (2007). These analyses are
usually based on long-run theoretical models whose
reduced forms become specific structural vars. For
instance, Hoffmaister, Roldós and Wickham (1998) point
out that terms-of-trade shocks act through the price of
intermediate inputs, assuming that a positive change in
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
review
this price behaves as negative technological progress. This
way, positive terms-of-trade shocks are positive supply
shocks that relax the intermediate inputs constraint.
Regardless of the specifics of the technique, most
of the literature suggests a positive effect of terms of
trade on output fluctuations in developing countries.
However, some of the documented literature undermines
the role of international prices. For example, Broda
(2004) affirms that his evidence contradicts that from
Mendoza (1995). Broda, working with a sample of
75 developing countries with annual data from 1973 to
1996, finds that the contribution of terms-of-trade shocks
accounts for less than 10% of actual real gdp volatility
in countries with flexible exchange regimes. Similarly,
Ahmed (2003), who studied the economic fluctuations
of six Latin American economies (Argentina, Bolivarian
Republic of Venezuela, Brazil, Chile, Colombia and
Mexico), concludes that the terms-of-trade shocks may
account for (although significant in statistical terms) less
than 8% of domestic output fluctuations.
The terms of trade have also been used as a control
variable in explaining the relationship between the
110
•
august
113
2013
short-run fluctuations in gdp and other variables in Latin
American countries, for example, in Barro (1979) and
Edwards (1983 and 1986). Consistent with the studies
described before, the effect is usually positive. However,
Edwards (1983) finds that the estimate of the effect of
terms of trade on output is only significant for Chile and
Mexico, and not significant for Brazil, Colombia and
Peru. Furthermore, Edwards (1986), who checks whether
a devaluation of the nominal exchange rate may display
contractionary effects in the short run, concludes that
the terms-of-trade effect on real output in developing
countries is negligible.
To the author’s knowledge, no previous study
determines the extent to which output fluctuations in the
recent Colombian context are attributable to the terms of
trade. Colombia’s own specific features and status as a
developing economy may allow the use of a simple but
powerful econometric tool to pursue that quantification
and test its robustness. Given that the background literature
contains several cases of positive, negative and null
effects of the terms of trade, the question examined in
the Colombian case is fundamentally empirical.
III
Empirical strategy
This section aims to offer a parsimonious model for
Colombia for the period 1994-2011 to describe its output
fluctuations, to estimate the partial effect of the terms
of trade on gdp variations, and to test the significance
of that estimate, using quarterly data.
This period of analysis was selected for several
reasons. First, the data are available without substantial
methodological changes in the national accounts and the
balance of payments.8 Second, the analysis excludes
one of the most important structural break points in the
Colombian economic policy: trade liberalization in the
early 1990s. Third, the period includes: the commodity
price boom that started in 2003, the subsequent downturn
at the end of 2008 (for the Colombian terms of trade),
and a recovery starting in 2009. In the same way, this
period also includes the sharpest recession known in
The information was obtained directly from the National
Administrative Department of Statistics (dane) of Colombia; the
dataset of International Financial Statistics (ifs) does not report
quarterly gdp data for Colombia until 1994.
8
Colombian economic history (year 1999) and a period
of high growth (2003-2007) (see figure 1).
Regarding the statistical procedure, this paper
follows the Box-Jenkins technique for a univariate
model. The type of model that is estimated is usually
known in the literature as the autoregressive moving
average model with exogenous variables (armax), a
model for stationary series with three components:
(i) the autoregressive part (ar), (ii) the moving average
part (ma), and (iii) the set of other explanatory variables
(x). The general model is thus:
n
yt = a +
/m y
p t-p +
p=1
n
/i n
q
n
t-q +
q=1
/c
i, m Xi, t - m + nt (1)
m=0
Where y represents the dependent variable (a stationary
series of gdp in this paper), t indexes time, μ is the error
term, X is the set of explanatory variables (stationary,
and that includes the terms of trade), and α, λ, θ, and γ,
the parameters to estimate.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
114
Figure 1
c e pa l
review
110
•
august
2013
Real gdp and terms of trade
(Logarithmic scaling)
80
75
3.5
3.0
70
2.5
65
2.0
60
1.5
55
1.0
50
45
40
1994 1996 1998 2000 2002 2004 2006 2008 2010
gdp (trillions of pesos, at constant prices, 2000)
0.5
1994 1996 1998 2000 2002 2004 2006 2008 2010
totcl
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia and the
central bank of Colombia (Banco de la República de Colombia).
totcl: terms of trade for four commodities (oil, coal, coffee and nickel).
There are several reasons that justify the specification
in equation (1) given that gdp and terms of trade are
not cointegrated (see table A.1).9 First, stationary
series reduce the possibility of spurious correlations
due to similar trends between the dependent and an
explanatory variable. Second, the Wold decomposition
shows that any stationary process can be approached
through the combination of both the autoregressive and
the moving average models. Third, the combination
of both components contributes to the parsimony of
the model, once the autocorrelation of the errors that
would affect the significance tests is taken into account.
Lastly, the use of an arma model allows one to control
for any possible persistence of output fluctuations.10 In
addition to estimating the contemporaneous effect, this
specification allows an estimate of the total effect of the
terms of trade on gdp over time.
Besides the arma specification and the terms of
trade, control variables must be considered in the set
of explanatory variables. The main reason is that their
omission may result in a biased estimate for the effect
9 See, for example, Montenegro (2002).
10 See, for example, Nelson and Plosser (1982), Campbell and Mankiw
(1987), and Blanchard and Quah (1989), for more information on the
persistence of output fluctuations.
of the terms of trade. From the aggregate demand side,11
robustness tests include two groups of monetary variables:
lending interest rates, and exchange rates. Interest rates
may be important in the determination of the investment
component, which explains most of the variability in
gdp, and may also be responsive to changes in the terms
of trade through the relaxation of balance-of-payment
constraints. On the other hand, by including the nominal
and real exchange rates, it is possible not only to test the
robustness of the effect of the terms of trade but also to
examine if the short-run effect of a depreciation (or an
appreciation) of the exchange rate is contractionary (or
expansionary). Lastly, quarterly growth of United States
gdp (gdpus) and net financial flows (inflows minus
outflows) in the Colombian balance of payments (nff)
will also be treated as control variables. Both variables
might also be correlated simultaneously with the terms
of trade and Colombian gdp.
The specification leaves aside technological shocks,
which are an important element in the literature on real
business cycles.12 This decision is justifiable on three
11 See Shapiro and Watson (1989), in whose work the source of output
fluctuations is divided between demand and supply components.
12 See Mankiw (1989) for a criticism of the real business cycle theory
and Holland and Scott (1998) for an empirical defence of the technical
change explaining the business cycle in the United Kingdom.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
review
110
counts. First, given the volatility of the quarterly data,
technological shocks correlated with the terms of trade
and that can explain variations of gdp quarter to quarter
are unlikely (even if some amplifiers are considered).
Second, proxies of technical change, like total factor
productivity, are not usually reliable, especially in
developing countries. Third, despite the fact that a clear
identification is impossible, the arma specification is
already controlling for the new information (innovations)
through its moving average term, including non-observable
shocks that affect output.13
The study does not take into account the expectations
of economic agents or the management of these
expectations as an instrument of economic policy through,
august
115
2013
for example, monetary policy. This is a limitation of the
study, despite the difficulty of finding a good proxy for
that variable.
While further research can explore whether particular
variables may improve the fit of the model, the main
purpose of the study is not to obtain a forecasting model,
but to evaluate the role of the terms of trade.
1. Variables and data description
— Gross domestic product (gdp)
The quarterly data for real gdp, seasonally
adjusted, was obtained from the National Administrative
Department of Statistics (dane) of Colombia (see table
A.2). The dependent variable is the first difference of the
logarithm of gdp (dlgdp) for Colombia (approximately
quarterly gdp growth) (see figure 2). This transformation
is necessary for two reasons: it defines the variable in
terms of output fluctuations, and it fulfils the stationarity
requirement in the Box-Jenkins technique. According to
different tests, weak stationarity of dlgdp is verified by
rejecting the null hypothesis that this series has a unit
root (see table A.3).
13 The El Niño Phenomenon, another supply-side shock not correlated
with the terms of trade but potentially useful for understanding the
nature of the Colombian business cycle, was examined in a previous
analysis not reported in this paper. Using the multivariate El NiñoSouthern Oscillation index, I used different alternative definitions to
create a dummy variable, according to whether the quarter was in the
warm phase or not, whether the quarter was in a warm phase with an
index that was one standard deviation higher than the average or not,
or whether the absolute value of the index was relatively high to its
average. No clear relationship between El Niño and gdp was found
for the period under analysis.
Figure 2
•
First difference of the logarithm of gdp
0.04
0.03
0.02
0.01
0.00
-0.01
-0.02
-0.03
1994
1996
1998
2000
2002
2004
2006
2008
2010
dlgdp
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia.
dlgdp: first difference (quarter to quarter) of the logarithm of real gdp.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
review
As an alternative definition of output fluctuations,
the cyclical component of gdp (gdpcycle) was estimated
by means of the Hodrick-Prescott filter. This series is
also stationary.
•
2013
Correlation dlgdp and dltotcl
(Correlation coefficient simple: 0.35)
0.04
0.03
— Terms of trade and related prices
Two definitions for the terms of trade are used
in this paper. The first was constructed with statistical
information from the balance of payments and the
wholesale imports price index from the central bank of
Colombia. This definition, called totcl, corresponds to
the ratio (pxcl/pi), where the denominator is the wholesale
imports price index and the numerator is a Laspeyres
index14 for the basket of the most important Colombian
exportable commodities (oil, coal, coffee and nickel). I
use the variable dltotcl (unit value of all Colombian
exports). The second definition is called tott, available
also from the central bank of Colombia, and is the ratio
of the wholesale exports price index to the wholesale
imports price index (px/pi). The transformed variable is
called dltott (first difference of logarithm of tott).
As a preliminary graphical diagnosis of the key
relationship in this paper, figures 3 and 4 present the
correlation between the output fluctuations and the
variations in the terms of trade in Colombia. Figure 3
depicts the simple correlation (the correlation coefficient
is 0.35). Figure 4 shows the correlation between the
cyclical components of gdp and totcl (the correlation
coefficient is 0.48). Besides the positive correlation, both
scatter plots suggest that these correlation coefficients
are not augmented by potential outliers. Most of the
observations in the sample follow the same pattern
described by the simple ordinary least squares (ols)
regression between dlgdp and dltotcl in figure 3 and
between gdpcycle and totclcycle in figure 4.
This empirical analysis assumes, based on the
dependent economy framework, and the construction
of our series related to terms of trade, that the terms
of trade are exogenous and that they cause the output
fluctuations, not the other way around. Although this is a
very plausible assumption for Colombia as described in
the introduction, a Granger causality test was performed.
The test suggests the non-rejection of this assumption
(see table 1).
august
Figure 3
0.02
DLGDP
110
0.01
0.00
-0.01
-0.02
-0.03
-0.6
-0.4
-0.2
0.0
0.2
0.4
DLTOTCL
Source: prepared by the author on the basis of data from the National
Administrative Department of Statistics (dane) of Colombia and the
central bank of Colombia (Banco de la República de Colombia).
dlgdp: first difference (quarter to quarter) of the logarithm of real gdp.
dltotcl: first difference (quarter to quarter) of the logarithm of the
terms of trade for four commodities (oil, coal, coffee and nickel).
Figure 4
Correlation cyclical components of gdp and totcl
(Correlation coefficient simple: 0.48)
3 000 000
2 000 000
GDPCYCLE
116
1 000 000
0
-1 000 000
-2 000 000
-1.0
-0.5
0.0
0.5
1.0
1.5
TOTCLCYCLE
Source: prepared by the author on the basis of data from the National
Administrative Department of Statistics (dane) of Colombia and the
central bank of Colombia (Banco de la República de Colombia).
14 A
Paasche index was also calculated but it did not exhibit a
substantial difference from the Laspeyres one.
totclcycle: cyclical component of the terms of trade for four
commodities (oil, coal, coffee and nickel).
gdpcycle: cyclical component of real gdp.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
Table 1
review
110
•
august
117
2013
Granger causality test
Null hypothesis: dlgdp does not Granger cause dltotcl
P value
Observations
Lag length 1
Lag length 2
Lag length 3
Lag length 4
0.80
60
0.79
59
0.82
58
0.52
57
0.41
59
0.33
58
0.22
57
Null hypothesis: dltotcl does not Granger cause dlgdp
P-value
Observations
0.15
60
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia and the
central bank of Colombia (Banco de la República de Colombia).
dlgdp: first difference (quarter to quarter) of the logarithm of real gdp.
dltotcl: first difference (quarter to quarter) of the logarithm of the terms of trade for four commodities (oil, coal, coffee and nickel).
This diagnostic test also suggests that dltotcl
does not Granger cause dlgdp (although the p values are
relatively smaller than those in the other hypothesis in
table 1). More formal empirical results on the relationship
dltotcl and dlgdp will be presented in the following
section. In addition to the terms-of-trade definitions
above, four more related prices are used (as alternative
to totcl) in the right hand side of the regressions: a
Laspeyres index for the prices of oil, coal, coffee and
nickel (pxcl), the oil prices (oilpr), the wholesale imports
price index (pi), and the wholesale exports price index
(px). The transformed and stationary variables are called
dlpxcl, dloilpr, dlpiifs and dlpxifs, respectively.
— Lending interest rates, exchange rates, net financial
flows, and United States output fluctuations
Four stationary control variables are included in
the right hand side of the regression. The first is the
first difference of the nominal lending interest rate
(dnir) which was obtained from International Financial
Statistics (ifs) and corresponds to a weighted average of
effective rates for the whole banking system including
all types of credit. As an alternative, the first difference
of the real interest rate (drir) was calculated using the
inflation in the producer price index.
The second variable is the nominal depreciation of
the exchange rate (the first difference of the logarithm of
the nominal exchange rate (dlner). dlner corresponds
to quarterly depreciation of the exchange rate when the
value is positive. Likewise, the first difference of the
logarithm of the real exchange rate (dlrer) is also used.
The third variable is the first difference of the
net financial flows (inflows minus outflows in the
Colombian balance of payments). Although interest
rates and exchanges rates should capture the role of
financial flows to some extent, this variable is included
as a potential omitted variable.
Lastly, the econometric analysis controls for the
quarterly growth of United States gdp (dlgdpus).
This series is available in ifs. The United States is the
destination for approximately 40% of total Colombian
exports and 70% of Colombian oil exports.
Control variables may be correlated with each
other. For instance, changes in the structure of interest
rates along with some degree of capital mobility may
put pressure on the exchange rates and such correlation
could affect the respective significance tests. However,
the key issue in this paper is a possible bias that may
exist if these control variables are omitted given their
simultaneous correlation with terms-of-trade and output
fluctuations.
2. Econometric results
The specification of the arma component of the model
was based on the correlogram for the dependent variable
(see table A.4), and a set of regressions (see table A.5) that
evaluate the significance of the estimated coefficients for
the arma elements (dlgdp as the left hand side variable).
Both the autocorrelation and the partial correlation functions
suggest a specification around the arma (3, 3). However,
the set of regressions (even including a fourth lag with
an possible economic interpretation) reveals a robust and
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
118
c e pa l
review
parsimonious specification. Table A.5 shows the results
for regressions, including: (i) only the autoregressive
elements (column 1); (ii) only the moving average
elements (column 2); (iii) a baseline regression with all
the arma elements (column 3); (iv) a specific regression
obtained after removing one by one the elements whose
estimated coefficients were not significant at the 5% level
in the baseline regression (column 4); and (v) individual
regressions for the elements ar(2), ma(3) and ma(4)
(columns 5 to 7), which along with the regressions in
columns 1 and 2 show that the elements ar(2) and ma(3)
are the most robust. Therefore, a parsimonious version of
the arma (2, 3) was used without including the first lag
for the autoregressive component and without the first
and second lags for the moving average (column 8). The
estimates in the arma (2, 3) are robust to the inclusion
of the terms-of-trade and control variables. The number
of lags was reasonable for interpreting the effects on the
dependent variable in the short run. Given the quarterly
Table 2
110
•
august
2013
data, the second lag in the autoregressive component
refers to a half-year lag.
Table 2 (column 1) reports positive and significant
estimates for the arma coefficients. While the estimates
for the moving average coefficient can be associated
with the effect of the statistical innovations, the estimate
in the autoregressive part suggests the existence of an
important degree of persistence in the Colombian gdp
fluctuations. All the estimates for this arma model are
significant at least at the 5% level. The arma model
can explain 17% of the total variation in the dependent
variable. The Durbin h’s statistic, the p value of the ChiSquare test, the Breusch-Godfrey test for the residual, and
the correlogram of the residual (see table A.6) suggest
the absence of autocorrelation. Furthermore, given the
assumption of weak stationarity, dlgdp does not face
heteroskedasticity. This means that the t-statistics and
the p-values used to establish significance at the 1.5%
and 10% are reliable.
Terms of trade and output fluctuations I
Dependent variable: dlgdp (quarterly growth rate of real gdp)
(1)
(2)
Constant
ar(2)
ma(3)
dltotclt
Total effect
(including persistence)
0.0082***
(3.00)
0.3273***
(3.14)
0.3377**
(2.40)
0.0075***
(4.87)
0.0339***
(3.95)
R2
R2 adjusted
Durbin-Watson statistic
Durbin h
Prob. Chi-square (Breusch-Godfrey)
S.E. of regression
Akaike’s information criterion
Schwarz information criterion
F-statistic (p-value)
Observations
0.17
0.14
0.71
0.20
0.01
-6.06
-5.96
0.00
67
0.12
0.11
1.96
0.00
61
(3)
0.0082***
(2.72)
0.3227***
(2.96)
0.3659**
(2.56)
0.0215***
(4.08)
0.0318***
[10.24]
0.26
0.22
0.14
0.32
0.01
-6.04
-5.90
0.00
59
(4)
Standardized variable
0.3227***
(2.96)
0.3659**
(2.56)
0.2315***
(4.08)
0.3418***
[10.24]
0.26
0.22
0.14
59
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia and the
central bank of Colombia (Banco de la República de Colombia).
Note:
(t-statistic), *p0.10, **p0.05, ***p0.01, [Chi-square].
Least squares and ma derivatives that use accurate numeric methods.
Consistent standard errors.
dlgdp: first difference (quarter to quarter) of the logarithm of real gdp.
dltotcl: first difference (quarter to quarter) of the logarithm of the terms of trade for four commodities (oil, coal, coffee and nickel).
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
review
Column 3 of table A6 corresponds to the regression
including dltotcl:
DLGDPt
= a + c1 DLTOTCL + m2 DLGDPt - 2 (2)
+ i3 nt - 3 + nt
The estimate for the effect of dltotcl on dlgdp
is significantly positive at the 1% level. The magnitude
of the estimate for the contemporaneous effect means
that a 1% increase in the growth of the terms of trade
increases by 0.02% the quarterly growth of gdp (holding
other variables constant). This magnitude is important.
One standard deviation in dltotcl (equal to 13.22%)
will change the quarterly growth of gdp by 0.28%. This
change is around 23% of one standard deviation in the
quarterly growth of gdp (column 4, table A.6). Once
the persistence effect is calculated, the same standard
Table 3
110
•
august
119
2013
deviation of dtotlc is associated with a change around
34% of one standard deviation in the quarterly growth of
the gdp. Therefore, one third of the quarterly variability
in gdp is driven by the terms of trade for the four most
important Colombian commodities.
The terms of trade effect holds when the definition
of the terms of trade is extended to include the unit
value of all Colombian exports (dltott) (column
1, table 3). The estimate is higher but the standard
deviation of dltott is lower (5.79%). The independent
variable still accounts for around 27% of one standard
deviation of the gdp growth (column 2, table 3). Table
3 (columns 5, 6, 7 and 8) and table 4 (using cyclical
components) also offer evidence confirming that prices
of the four most important Colombian export goods,
in particular oil, are the ones that lead the short-run
effect on output.
Terms of trade and output fluctuations II
Dependent variable: dlgdp (Quarterly growth rate of real gdp)
dltottt
dlpxt
dlpit
dlpxclt
dloilprt
Total effect
(including persistence)
(1)
(2)
Standard
deviation
0.0364* 0.1715*
(1.69)
(1.69)
0.0566
0.2664
[2.37]
[2.37]
0.20
R2
0.16
R2 adjusted
Durbin h
0.15
Prob. Chi-Square (Breusch-Godfrey) 0.22
S.E. of regression
0.01
Akaike’s information criterion
-6.07
Schwarz information criterion
-5.94
F-statistic (p-value)
0.00
Observations
67
0.20
0.16
0.15
67
(5)
(6)
(7)
(8)
(9)
Standard
deviation
(3)
(4)
0.0253
(1.33)
0.0028
(0.09)
0.0365*
(1.69)
-0.0367
(-1.03)
0.0252***
(4.29)
0.0377***
[11.96]
0.0197
(0.61)
0.0260***
(3.96)
0.0147*** 0.2043***
(3.29)
(3.29)
0.0219*** 0.3052***
[7.24]
[7.24]
0.19
0.15
0.47
0.23
0.01
-6.06
-5.93
0.00
67
0.17
0.13
0.75
0.20
0.01
-6.03
-5.90
0.01
67
0.20
0.15
0.08
0.22
0.01
-6.04
-5.88
0.01
67
0.28
0.22
0.50
0.33
0.01
-6.03
-5.85
0.00
59
0.27
0.23
0.28
0.33
0.01
-6.06
-5.92
0.00
59
0.25
0.21
0.65
0.31
0.01
-6.03
-5.89
0.00
59
0.25
0.21
0.65
59
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia and the
central bank of Colombia (Banco de la República de Colombia).
Note:
(t-statistic), *p0.10, **p0.05, ***p0.01, [Chi-square]
Least squares and MA derivatives that use accurate numeric methods. Consistent standard errors.
The arma component is included in all the regressions but not reported.
dltott: first difference (quarter to quarter) of the logarithm of the terms of trade (wholesale export and import prices).
dlpx: first difference (quarter to quarter) of the logarithm of the total exports price index (wholesale).
dlpi: first difference (quarter to quarter) of the logarithm of the total imports price index (wholesale).
dlpxcl: first difference (quarter to quarter) of the logarithm of the Laspeyres price index for exports of four commodities (oil, coal, coffee and nickel).
dloilpr: first difference (quarter to quarter) of the logarithm of the Laspeyres price index for oil.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
120
c e pa l
Table 4
review
110
•
august
2013
Estimate
Adjusted R2
Durbin h.
Prob. Chi-square
(Breusch-Godfrey)
S.E. of regression
(x100.000)
Akaike’s information
criterion
Schwarz information
criterion
F-statistic (p-value)
Observations
Terms-of-trade and output fluctuations III
(Cyclical components)
totclt
542 217**
(2.42)
0.75
0.59
0.80
5.7
29.41
29.58
0.00
61
2 170 543**
(2.36)
0.74
0.73
0.85
5.3
29.29
29.45
0.00
69
pxt
14 030*
(1.95)
0.74
0.72
0.66
5.4
29.31
29.47
0.00
69
pit
-105.4
(-0.01)
0.72
0.89
0.72
5.6
29.36
29.52
0.00
69
pxclt
5 928***
(3.00)
0.74
0.53
0.75
5.6
29.39
29.56
0.00
61
oilprt
12 718**
(2.64)
0.74
0.59
0.75
5.7
29.41
29.6
0.00
61
tottt
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia and the
central bank of Colombia (Banco de la República de Colombia).
Note:
(t-statistic), *p0.10, **p0.05, ***p0.01.
Least squares and MA derivatives that use accurate numeric methods.
Consistent standard errors. AR(1) MA(2) and MA(3) are included but not reported.
totcl: terms of trade for four commodities (oil, coal, coffee and nickel).
tott: terms of trade (wholesale export and import prices).
pxt: total exports price index (wholesale).
pit: total imports price index (wholesale).
pxcl: Laspeyres price index for exports of four commodities (oil, coal, coffee and nickel).
oilpr: Laspeyres price index for oil.
3. Robustness tests
The results in section III.2 are robust to the inclusion of
the control variables: lending interest rates, exchange
rates, net financial flows, and quarterly growth of United
States gdp (see table 5). The estimate for dltotcl not only
remains significant at 1% in most of the regressions (at
5% in column 4, table 6) but also its magnitude is stable.
In order to expose the results to a stronger robustness
test, lagged control variables were included that were
independently significant when a regression for dlgdp
was run. These variables are: dnir and dlgdpus (both
lagged two quarters). Once these variables are included,
only dnir (-2) remains significant (columns 1 and 3,
table 6). The estimates for standardized dltotcl are
still robust and the total effect, including persistence,
accounts for 30% of gdp variability (column 4, table 6).
Results for standardized variables (column 4,
table 6) also report a theoretically consistent negative
effect of dnir (-2), which is significant and important
in magnitude. Although the estimate is not robust when
arma components are removed, the inclusion of dnir
(-2) increases the R2 from 0.17 to 0.39. Although it is true
that the purpose of this paper is not to evaluate either the
model’s forecasting properties or the robustness in the
estimate for the effect of dnir (-2), the negative estimate,
along with a higher R2, may reflect the fact that dnir
is acting through investment, which is the aggregate
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
Table 5
review
110
•
august
121
2013
Robustness to additional variables I
Dependent variable: dlgdp (quarterly growth rate of real gdp)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Constant
0.0082***
(2.72)
0.0082***
(2.72)
0.0081***
(2.70)
0.0079**
(2.60)
0.0081***
(2.71)
0.0086***
(3.07)
0.0078**
(2.66)
ar(2)
0.3227***
(2.96)
0.3204**
(2.29)
0.3231***
(2.90)
0.3356***
(3.12)
0.3241***
(3.06)
0.3105***
(3.00)
0.2855**
(2.44)
ma(3)
0.3659**
(2.56)
0.3624**
(2.19)
0.3626**
(2.50)
0.3657**
(2.52)
0.3581***
(2.53)
0.3627**
(2.42)
0.3843**
(2.58)
dltotclt
0.0215***
(4.08)
0.0221***
(3.86)
0.0217***
(4.11)
0.0279**
(2.57)
0.0226***
(4.02)
0.0237***
(2.69)
0.0201***
(3.63)
dnirt
drirt
dlnert
dlrert
dlgdpus
dnff
0.26
0.22
0.14
0.32
0.01
-6.04
-5.90
0.00
59
0.26
0.21
...
0.29
0.01
-6.00
-5.83
0.00
59
0.26
0.20
0.15
0.34
0.01
-6.01
-5.94
0.00
59
0.27
0.21
0.27
0.36
0.01
-6.02
-5.84
0.00
59
0.26
0.21
0.26
0.33
0.01
-6.01
-5.84
0.00
59
0.26
0.21
0.25
0.29
0.01
-6.01
-5.83
0.00
59
R2
R2 adjusted
Durbin h
Prob. Chi-square (Breusch-Godfrey)
S.E. of regression
Akaike’s information criterion
Schwarz information criterion
F-statistic (p-value)
Observations
0.0144
(0.23)
-0.0008
(-0.29)
0.0234
(0.77)
0.0127
(0.56)
-0.0878
(-0.32)
0.0000
(0.48)
0.26
0.20
0.41
0.35
0.01
-6.00
-5.93
0.00
58
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia, the central
bank of Colombia (Banco de la República de Colombia) and International Financial Statistics (ifs).
Note:
(t-statistic), *p0.10, **p0.05, ***p0.01.
Least squares and MA derivatives that use accurate numeric methods. Consistent standard errors.
dltotcl: first difference (quarter to quarter) of the logarithm of the terms of trade for four commodities (oil, coal, coffee and nickel).
dnir: first difference (quarter to quarter) of the nominal lending interest rate.
drir: first difference (quarter to quarter) of the real lending interest rate.
dlner: first difference (quarter to quarter) of the logarithm of the nominal exchange rate.
dlrer: first difference (quarter to quarter) of the logarithm of the real exchange rate.
dlgdpus: first difference (quarter to quarter) of the logarithm of the real gdp of the United States.
dnff: first difference (quarter to quarter) of net financial flows.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
122
Table 6
c e pa l
review
110
•
august
2013
Robustness to additional variables II
(Including lags)
Dependent variable: dlgdp (quarterly growth rate of real gdp)
(1)
dltotclt
0.0182***
(3.81)
dnirt-2
-0.1777**
(-2.61)
(2)
0.0189***
(3.08)
0.0146**
(2.33)
-0.1813***
(-2.72)
dlgdpust -2
Total effect dltotcl
(including persistence)
Total effect dnirt-2
(including persistence)
0.39
0.34
-2.33
0.43
0.01
-6.20
-6.02
0.00
59
0.27
0.21
0.34
0.24
0.01
-6.02
-5.84
0.00
59
0.41
0.35
-1.77
0.42
0.01
-6.19
-5.98
0.00
59
R2
R2 adjusted
Durbin-Watson statistic
Durbin h
Prob. Chi-square (Breusch-Godfrey)
S.E. of regression
Akaike’s information criterion
Schwarz information criterion
F-statistic (p-value)
Observations
-0.1762
(-1.19)
(3)
-0.2637
(-1.61)
(4)
Standard
variation
(5)
Non-arma
0.1961***
(3.81)
-0.3552**
(-2.61)
0.0319***
(4.05)
-0.1251
(-1.28)
(6)
Non-arma
standard
variation
0.3432***
(4.05)
-0.2500
(-1.28)
0.3076***
[8.46]
-0.5573**
[4.96]
0.18
0.15
2.09
0.01
0.00
61
0.18
0.15
2.09
0.00
61
0.39
0.34
-1.77
0.00
59
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia, the central
bank of Colombia (Banco de la República de Colombia) and International Financial Statistics (ifs).
Note:
(t-statistic), *p0.10, **p0.05, ***p0.01, [Chi-square].
Least squares and ma derivatives that use accurate numeric methods. Consistent standard errors.
The arma component is included but not reported in (1) to (4).
dltotcl: first difference (quarter to quarter) of the logarithm of the terms of trade for four commodities (oil, coal, coffee and nickel).
dnir: first difference (quarter to quarter) of the nominal lending interest rate.
dlgdpus: first difference (quarter to quarter) of the logarithm of the real gdp of the United States.
demand component whose variations explain most of
the short-run variation in gdp. Although investment is
one fourth of Colombian gdp, while consumption is
two thirds, investment is the most volatile component
of gdp (its standard deviation is 8 times greater than for
consumption). An initial exploration of the channels
in aggregate demand relevant to understanding more
deeply the significant and robust effect of terms of
trade on output fluctuations in Colombia (see table 7)
suggest that investment (dli) is the main channel.
One standard deviation in dltotcl seems to explain
one third of the variability in investment (only for the
contemporaneous effect). Future research will be oriented
towards examining what type of investment terms-oftrade shocks are fostering in the short run. This might
also require a better understanding of the mechanisms
through which terms of trade may affect credit markets
and interest rates. The next channel suggested in table 7,
but apparently less responsive, is public expenditure
(dlg). Lastly, the current account (dlx for exports and
dlm for imports) does not seem to be correlated with
terms of trade in the short run.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
Table 7
Dependent variable:
review
110
•
august
Terms of trade and aggregate demand components
(Ordinary least squares (ols) regressions)
(1)
(2)
(3)
(4)
(5)
(6)
dlc
dli
dlg
dlx
dlm
dli
Constant
0.0068***
(4.53)
0.0086
(0.82)
0.011***
(4.18)
dltotclt
0.0130
(1.36)
0.2096***
(3.30)
0.0330**
(2.22)
dnirt-2
R2
R2 adjusted
Durbin-Watson statistic
S.E. of regression
F-statistic (p-value)
Observations
123
2013
0.0099**
(2.36)
-0.0296
(-0.74)
0.0129**
(2.04)
0.0037
(0.36)
0.0972
(1.65)
0.1939***
(3.22)
0.02
0.004
1.33
0.01
0.26
61
0.10
0.09
2.05
0.09
0.01
61
0.04
0.03
1.93
0.03
0.11
61
0.01
-0.003
1.98
0.03
0.37
61
0.06
0.05
1.42
0.05
0.05
61
-1.0042
(-1.65)
0.19
0.16
2.25
0.08
0.00
61
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia, the central
bank of Colombia (Banco de la República de Colombia) and International Financial Statistics (ifs).
Note:
(t-statistic), *p0.10, **p0.05, ***p0.01.
dli: first difference (quarter to quarter) of the logarithm of real aggregate investment.
dlc: first difference (quarter to quarter) of the logarithm of real aggregate consumption.
dlg: first difference (quarter to quarter) of the logarithm of real aggregate public spending.
dlx: first difference (quarter to quarter) of the logarithm of real exports.
dlm: first difference (quarter to quarter) of the logarithm of real imports.
dltotcl: first difference (quarter to quarter) of the logarithm of the terms of trade for four commodities (oil, coal, coffee and nickel).
dnir: first difference (quarter to quarter) of the nominal lending interest rate.
IV
Concluding remarks
Although it is sometimes claimed that a positive
correlation between the terms of trade and aggregate
output can be established a priori, a vast literature
describes the complexity of the relationship of these
two variables. First, a positive as well as a negative
correlation have been found in some developing
countries. Second, the usual theoretical framework used
to describe small open economies permits outcomes
in which the relationship may be negative or null. A
lot depends on the plausibility of the assumptions
made for a particular economy and the way in which
domestic markets adjust after external shocks. The
idea of an ambiguous effect has lately been part of a
debate in Colombia about the perverse effects of the
terms of trade and the well-known Dutch Disease. This
outcome, which is commonly associated with the long
run, might also act in the short run depending on how
fast possible contractionary effects of a commodity
price boom can be transmitted.
The estimate of the impact of the terms of trade on
gdp is found not only to be significantly positive but also
to be very great in magnitude. One standard deviation
in the growth of the terms of trade results in around
one third of one standard deviation of quarterly gdp
growth. The results are robust to different specifications
that include: price components of the terms of trade,
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
124
c e pa l
review
alternative definitions of business cycles, and control
variables whose omission might lead to a biased estimate.
In addition, depreciation of the nominal exchange
rate does not seem to have a significant short-run effect
as stated by the contractionary devaluation hypothesis.
This might be important when analysing potential
new policies, costly or distortionary, oriented towards
controlling the appreciation of the nominal exchange
rate that Colombia and other developing countries
have been experiencing lately. Likewise, this short-run
dynamic might complement analysis that suggests that
devaluation is a useful tool for growth. On the other hand,
this paper finds preliminary evidence that supports the
belief that the lending interest rate can have a negative
effect on output fluctuations.
In summary, robust evidence was found in support
of the hypothesis that the terms of trade played an
important role in determining the short-run variations
in gdp in Colombia over the period 1994-2011. Results
from simple specifications for stationary series, justified
by time series tests (cointegration and Granger), along
110
•
august
2013
with the particular features of the Colombian economy,
suggest that the terms of trade are exogenous and a source
of the output fluctuations as described in the three goods
model for a dependent economy. At least in the short
run, evidence does not indicate that eventual negative
effects of the terms of trade (Dutch Disease), if they
exist, can offset the positive effects on aggregate output.
Preliminary evidence also indicates that investment
may be the most important demand component driving
the aggregate outcome. One limitation of this study is
the use of aggregate data. Therefore, a future extension
would be to study the relationship between terms of trade
and investment demand disaggregated by components
and by industrial sectors to determine the foundation of
the observed fast adjustment of the external shocks in the
short run. Future research related to this finding will also
explore short-run effects of terms of trade fluctuations on
credit markets, interest rates, and investment in Colombia.
These studies would allow a more detailed evaluation of
the mechanisms behind the quick investment and output
responses to external shocks.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
review
110
•
august
125
2013
Annex
Table A.1
Summary of cointegration tests (gdp and totcl)
Sample: 1993Q4 2011Q2
Included observations: 59
Series: totcl gdp
Lags interval: 1 to 2
Selected (0.05 level*) Number of cointegrating relations by model
Data trend:
Test Type
Trace
Max-Eig
None
None
Linear
Linear
Quadratic
No intercept
No trend
0
0
Intercept
No trend
0
0
Intercept
No trend
0
0
Intercept
Trend
0
0
Intercept
Trend
0
0
* Critical values based on MacKinnon, Haug and Michelis (1999).
Information criteria by rank and model
Data trend:
None
None
Linear
Linear
Quadratic
Rank or
No. of CEs
No intercept
No trend
Intercept
No trend
Intercept
No trend
Intercept
Trend
Intercept
Trend
0
1
2
0
1
2
0
1
2
Log likelihood by rank (rows) and model (columns)
-864.2456
-860.1549
-859.5365
-864.2456
-859.1981
-855.9931
-860.4798
-856.0961
-855.9931
-860.4798
-854.2298
-851.0928
-858.4805
-852.4016
-851.0928
Akaike’s information criterion by rank (rows) and model (columns)
29.56765
29.56457
29.6792
29.56765
29.56604
29.62689
29.50779
29.49478
29.62689
29.50779
29.46542
29.52857
29.50781
29.43734*
29.52857
29.85991
29.99361
30.23282
29.93036
30.00074
30.23282
Schwarz information criterion by rank (rows) and model (columns)
29.84935*
29.98712
30.2426
29.84935*
30.0238
30.26071
29.85991
29.98776
30.26071
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia and the
central bank of Colombia (Banco de la República de Colombia).
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
126
c e pa l
Table A.2
review
110
•
august
2013
Data and sample definition
Code
Definition
Source
Coverage
d
First difference (quarter to quarter) of...
dl
First difference (quarter to quarter) of the logarithm of...
gdp
Real gdp
National Administrative Department of
Statistics (dane)
1994 I - 2011 II
totcl
Terms of trade for four commodities (oil, coal, coffee
and nickel)
Central bank, author’s calculations
1996 I - 2011 II
tott
Terms of trade (wholesale export and import prices)
Central bank
1994 I - 2011 II
px
Total exports price index (wholesale)
Central bank
1994 I - 2011 II
pi
Total imports price index (wholesale)
Central bank
1994 I - 2011 II
pxcl
Laspeyres price index for exports of four commodities
(oil, coal, coffee and nickel)
Central bank, author’s calculations
1996 I - 2011 II
oilpr
Laspeyres price index for oil
Central bank, author’s calculations
1996 I - 2011 II
nir
Nominal lending interest rate
International Financial Statistics
1994 I - 2011 II
rir
Real lending interest rate (using inflation of the producer International Financial Statistics, author’s
price index)
calculations
1994 I - 2011 II
ner
Nominal exchange rate (pesos per United States dollar)
Central bank
1994 I - 2011 II
rer
Real exchange rate (using producer price index)
Central bank
1994 I - 2011 II
gdpus
Real gdp of the United States
International Financial Statistics
1994 I - 2011 II
nff
Net financial flows (inflows-outflows)
Central bank
1996 I - 2011 II
c
Real aggregate consumption
National Administrative Department of
Statistics (dane)
1994 I - 2011 II
i
Real aggregate investment
National Administrative Department of
Statistics (dane)
1994 I - 2011 II
g
Real aggregate public spending
National Administrative Department of
Statistics (dane)
1994 I - 2011 II
x
Real exports
National Administrative Department of
Statistics (dane)
1994 I - 2011 II
m
Real imports
National Administrative Department of
Statistics (dane)
1994 I - 2011 II
Source: prepared by the author on the basis of information from National Administrative Department of Statistics (dane) of Colombia, the
central bank of Colombia (Banco de la República de Colombia) and International Financial Statistics (ifs).
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
Table A.3
Variable
dlgdp
dltotcl
dltott
dlpx
dlpi
dlpxcl
dloilpr
gdp (cycle)
totcl (cycle)
tott (cycle)
px (cycle)
pi (cycle)
pxcl (cycle)
oilpr (cycle)
dnir
drir
dlner
dlrer
dlgdpus
dnff
dlc
dli
dlg
dlx
dlm
review
110
•
august
2013
127
Augmented Dickey-Fuller unit root tests
MacKinnon one-sided p-values
Null hypothesis: variable has a unit root
0.002
0.000
0.000
0.000
0.000
0.000
0.000
0.001
0.000
0.000
0.000
0.008
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia, the central
bank of Colombia (Banco de la República de Colombia) and International Financial Statistics (ifs).
Note: Schwarz information criterion.
dlgdp: first difference (quarter to quarter) of the logarithm of real gdp.
dltotcl: first difference (quarter to quarter) of the logarithm of the terms of trade for four commodities (oil, coal, coffee and nickel).
dltott: first difference of the logarithm of tott.
dlpx: first difference (quarter to quarter) of the logarithm of the total exports price index (wholesale).
dlpi: first difference (quarter to quarter) of the logarithm of the total imports price index (wholesale).
dlpxcl: first difference (quarter to quarter) of the logarithm of the Laspeyres price index for exports of four commodities (oil, coal, coffee and
nickel).
dloilpr: first difference (quarter to quarter) of the logarithm of the Laspeyres price index for oil.
gdpcycle: cyclical component of real gdp.
totclcycle: cyclical component of the terms of trade for four commodities (oil, coal, coffee and nickel).
tott: terms of trade (wholesale export and import prices).
pi: total imports price index (wholesale).
px: total exports price index (wholesale).
pxcl: Laspeyres price index for exports of four commodities (oil, coal, coffee, and nickel).
oilpr: Laspeyres price index for oil.
dnir: first difference (quarter to quarter) of the nominal lending interest rate.
drir: first difference (quarter to quarter) of the real lending interest rate.
dlner: first difference (quarter to quarter) of the logarithm of the nominal exchange rate.
dlrer: first difference (quarter to quarter) of the logarithm of the real exchange rate.
dlgdpus: first difference (quarter to quarter) of the logarithm of the real gdp of the United States.
dlg: first difference (quarter to quarter) of the logarithm of real public spending.
dlx: first difference (quarter to quarter) of the logarithm of real exports.
dlm: first difference (quarter to quarter) of the logarithm of real imports.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
128
c e pa l
Table A.4
review
110
•
august
2013
Specific arma
(Correlogram for dlgdp)
Lag
Autocorrelation
Partial correlation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
0.09
0.30
0.24
-0.07
-0.04
0.00
0.04
-0.10
0.19
0.01
-0.03
0.11
0.00
-0.08
0.02
-0.10
-0.01
-0.20
0.03
-0.07
-0.11
0.04
-0.14
-0.02
-0.08
-0.01
-0.12
-0.05
0.09
0.30
0.22
-0.20
-0.20
0.04
0.23
-0.09
0.09
0.00
-0.07
0.03
0.07
-0.09
-0.05
-0.10
0.14
-0.23
0.01
0.08
-0.05
-0.08
-0.07
0.03
0.02
-0.03
-0.04
-0.07
Q-statistic
0.52
7.17
11.48
11.82
11.93
11.93
12.06
12.88
15.78
15.78
15.83
16.93
16.93
17.50
17.52
18.48
18.48
22.24
22.30
22.78
23.96
24.09
26.06
26.11
26.86
26.87
28.57
28.81
Prob. Q statistic
0.47
0.03
0.01
0.02
0.04
0.06
0.10
0.12
0.07
0.11
0.15
0.15
0.20
0.23
0.29
0.30
0.36
0.22
0.27
0.30
0.30
0.34
0.30
0.35
0.36
0.42
0.38
0.42
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
Table A.5
review
110
•
august
129
2013
Specific arma
(Regressions for arma components)
Dependent variable: dlgdp (quarterly growth rate of real gdp)
(1)
(2)
(3)
(4)
Specific I
(5)
(6)
(7)
(8)
Specific II
Constant
0.0081*** 0.0085*** 0.0082*** 0.0082*** 0.0079*** 0.0085*** 0.0082*** 0.0082***
(3.33)
(3.82)
(3.27)
(3.48)
(3.80)
(4.43)
(6.08)
(3.00)
ar(1)
0.0498
(0.33)
-0.5984*
(-1.89)
ar(2)
0.3426***
(3.69)
-0.5704**
(-2.36)
ar(3)
0.2285*
(1.98)
0.2419
(1.02)
ar(4)
-0.2163*
(-2.00)
-0.1293
(-0.65)
0.3724*** 0.3055***
(2.79)
(3.12)
0.3273***
(3.14)
ma(1)
0.0096
(0.07)
0.6289***
(2.81)
ma(2)
0.4205*** 1.0886***
(3.97)
(5.75)
ma(3)
0.3443**
(2.33)
0.2901
(1.51)
0.5052***
(6.24)
ma(4)
-0.0671
(-0.48)
0.6383*** -0.4013***
(3.91)
(-4.36)
0.18
R2
0.12
R2 adjusted
Prob. Chi-square
0.23
(Breusch-Godfrey)
S.E. of regression
0.01
Akaike’s information
-5.99
criterion
Schwarz information criterion -5.82
F-statistic (p-value)
0.02
Observations
65
0.3593**
(2.55)
0.3377**
(2.40)
-0.0981
(-0.94)
0.24
0.19
0.86
0.28
0.17
0.72
0.27
0.23
0.69
0.09
0.08
0.46
0.08
0.07
0.03
0.01
-0.01
0.01
0.17
0.14
0.20
0.01
-6.10
0.01
-6.00
0.01
-6.16
0.01
-6.00
0.01
-6.00
0.01
-5.92
0.01
-6.06
-5.94
0.00
69
-5.70
0.01
65
-6.03
0.00
67
-5.98
0.01
67
-5.93
0.02
69
-5.86
0.50
69
-5.96
0.00
67
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia.
Note:
(t-statistic), *p0.10, **p0.05, ***p0.01.
Least squares and ma derivatives that use accurate numeric methods. Consistent standard errors.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
130
c e pa l
Table A.6
review
110
•
august
2013
Correlogram of residuals for the selected regression a
Lag
Autocorrelation
Partial correlation
Q-statistic
Prob. Q statistic
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
0.03
0.07
-0.03
-0.22
-0.16
0.01
0.09
-0.04
0.22
-0.01
-0.09
0.04
0.00
-0.10
0.07
0.00
0.09
-0.13
0.08
-0.02
-0.05
0.07
-0.10
0.01
-0.02
0.06
-0.06
0.01
0.03
0.07
-0.03
-0.23
-0.15
0.05
0.12
-0.11
0.15
0.00
-0.09
0.04
0.08
-0.08
0.02
-0.03
0.16
-0.22
0.06
0.07
-0.05
-0.03
-0.06
0.01
0.01
-0.02
0.00
-0.05
0.08
0.47
0.53
4.13
5.98
5.99
6.62
6.75
10.59
10.60
11.33
11.43
11.44
12.21
12.58
12.59
13.25
14.88
15.46
15.49
15.78
16.23
17.28
17.30
17.33
17.70
18.16
18.16
0.47
0.13
0.11
0.20
0.25
0.34
0.16
0.23
0.25
0.33
0.41
0.43
0.48
0.56
0.58
0.53
0.56
0.63
0.67
0.70
0.69
0.75
0.79
0.82
0.84
0.87
Source: prepared by the author on the basis of data from the National Administrative Department of Statistics (dane) of Colombia.
a
See column (1) of table 2.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández
c e pa l
review
110
•
august
2013
131
Bibliography
Agénor, P.R and P.J. Montiel (2008), Development Macroeconomics,
Princeton, Princeton University Press.
Agénor, P.R., C.J. McDermott and E.S. Prasad (2000), “Macroeconomic
fluctuations in developing countries: some stylized facts”,
World Bank Economic Review, vol. 14, No. 2, Washington,
D.C., World Bank.
Ahmed, S. (2003), “Sources of economic fluctuations in Latin America
and implications for choice of exchange rate regimes”, Journal
of Development Economics, vol. 72, No. 1, Amsterdam, Elsevier.
Barro, R.J. (1979), “Money and output in Mexico, Colombia, and
Brazil”, nber Chapters, Cambridge, Massachusetts, National
Bureau of Economic Research.
Blanchard, O.J. and D. Quah (1989), “The dynamic effects of aggregate
demand and supply disturbances”, American Economic Review,
vol. 79, No. 4, Nashville, Tennessee, American Economic
Association.
Broda, C. (2004), “Terms of trade and exchange rate regimes in
developing countries”, Journal of International Economics,
vol. 63, No. 1, Amsterdam, Elsevier.
Buiter, W.H. (1988), “Structural and stabilization aspects of fiscal
and financial policy in the dependent economy”, Oxford
Economic Papers, New Series, vol. 40, No. 2, Oxford, Oxford
University Press.
Campbell, J.Y. and G. Mankiw (1987), “Are output fluctuations
transitory?”, The Quarterly Journal of Economics, vol. 102,
No. 4, Oxford, Oxford University Press.
Edwards, S. (1986), “Are devaluations contractionary?”, The Review
of Economics and Statistics, vol. 68, No. 3, Massachusetts,
The mit Press.
_____ (1983), “The short-run relation between growth and inflation in
Latin America: comment”, American Economic Review, vol. 73,
No. 3, Nashville, Tennessee, American Economic Association.
Frankel, J.A. (2010), “The natural resource curse: a survey”, nber
Working Paper, No. 15836, Cambridge, Massachusetts, National
Bureau of Economic Research.
Greenwood, J. (1984), “Non-traded goods, the trade balance, and the
balance of payments”, The Canadian Journal of Economics,
vol. 17, No. 4, Wiley.
Hoffmaister, A.W. and J.E. Roldós (2001), “The sources of
macroeconomic fluctuations in developing countries: Brazil
and Korea”, Journal of Macroeconomics, vol. 23, No. 2,
Amsterdam, Elsevier.
Hoffmaister, A.W., J.E. Roldós and P. Wickham (1998), “Macroeconomic
fluctuations in sub-Saharan Africa”, IMF Staff Papers, vol. 45,
No. 1, Houndmills, Palgrave Macmillan.
Holland, A. and A. Scott (1998), “The determinants of UK business
cycles”, Economic Journal, vol. 108, No. 449, Wiley.
Izquierdo, A., R. Romero and E. Talvi (2007), “Business cycles in
Latin America: the role of external factors”, Working Paper,
Washington, D.C., Inter-American Development Bank.
Kaminsky, G.L. (2010), “Terms of trade shocks and fiscal cycles”, nber
Working Paper Series, No. 15780, Cambridge, Massachusetts,
National Bureau of Economic Research.
Kent, C.J. and P. Cashin (2003), “The response of the current account to
terms of trade shocks: persistence matters”, imf Working Papers,
No. 03/143, Washington, D.C., International Monetary Fund.
Kose, M.A. (2002), “Explaining business cycles in small open
economies: how much do world prices matter?”, Journal of
International Economics, vol. 56, No. 2, Amsterdam, Elsevier.
Kose, M.A. and R.G. Riezman (1999), “Trade shocks and macroeconomic
fluctuations in Africa”, cesifo Working Paper Series, No. 203,
Munich, Center for Economic Studies and Ifo Institute for
Economic Research.
Krugman, P. and L. Taylor (1978), “Contractionary effects of
devaluation”, Journal of International Economics, vol. 8, No. 3,
Amsterdam, Elsevier.
Lizondo, J.S. and P.J. Montiel (1989), “Contractionary devaluation in
developing countries: an analytical overview”, imf Staff Papers,
vol. 36, No. 1, Houndmills, Palgrave Macmillan.
MacKinnon, J.G., A.A. Haug and L. Michelis (1999),
“Numerical distribution functions of likelihood ratio tests for
cointegration”, Journal of Applied Econometrics, vol. 14, No. 5,
John Wiley Sons.
Mahadeva, L. and J. Gómez (2009), “The international cycle and
Colombian monetary policy”, Borradores de Economía, No. 557,
Bogota, Bank of the Republic.
Mankiw, N.G. (1989), “Real business cycles: a new Keynesian
perspective”, The Journal of Economic Perspectives, vol. 3,
No. 3, Nashville, Tennessee, American Economic Association.
Medina, L. (2010), “The dynamic effects of commodity prices on fiscal
performance in Latin America”, imf Working Papers, vol. 192,
No. 201, Washington, D.C., International Monetary Fund.
Mendoza, E.G. (1995), “The terms of trade, the real exchange rate,
and economic fluctuations”, International Economic Review,
vol. 36, No. 1, Pennsylvania, University of Pennsylvania/Osaka
University Institute of Social and Economic Research Association.
Montenegro, A. (2002), Series de tiempo, Bogota, Pontificia
Universidad Javeriana.
Nelson, C.R. and C.L. Plosser (1982), “Trends and random walks in
macroeconomic time series: some evidence and implications”,
Journal of Monetary Economics, vol. 10, No. 2, Amsterdam,
Elsevier.
Obstfeld, M. (1982), “Aggregate spending and the terms of trade:
is there a Laursen-Metzler effect?”, The Quarterly Journal of
Economics, vol. 97, No. 2, Oxford, Oxford University Press.
Parra, J.C. (2008), “Hechos estilizados de la economía colombiana:
Fundamentos empíricos para la construcción y evaluación de
un modelo dsge”, Borradores de Economía, No. 4602, Bogota,
Bank of the Republic.
Raddatz, C. (2007), “Are external shocks responsible for the instability
of output in low-income countries?”, Journal of Development
Economics, vol. 84, No. 1, Amsterdam, Elsevier.
Rand, J. and F. Tarp (2002), “Business cycles in developing countries:
are they different?”, World Development, vol. 30, No. 12,
Amsterdam, Elsevier.
Razmi, A. (2007), “The contractionary short-run effects of nominal
devaluation in developing countries: some neglected nuances”,
International Review of Applied Economics, vol. 21, No. 5,
Abingdon, Routledge.
Reinhart, C.M. and V.R. Reinhart (1991), “Output fluctuations and
monetary shocks: evidence from Colombia”, imf Staff Papers,
vol. 38, No. 4, Houndmills, Palgrave Macmillan.
Shapiro, M.D. and M.W. Watson (1989), “Sources of business cycle
fluctuations”, nber Working Papers, No. 2589, Cambridge,
Massachusetts, National Bureau of Economic Research.
Svensson, L.E.O. and A. Razin (1983), “The terms of trade and
the current account: the Harberger-Laursen-Metzler effect”,
The Journal of Political Economy, vol. 91, No. 1, Chicago,
The University of Chicago Press.
Tornell, A. and P.R. Lane (1999), “The voracity effect”, American
Economic Review, vol. 89, No. 1, Nashville, Tennessee, American
Economic Association.
Terms of trade and output fluctuations in Colombia • Gonzalo Hernández