The effects of autocatalytic trade cycles on economic growth.
Bakker, Jurrien J. ; Afonso, Oscar ; Silva, Sandra T. 等
Introduction
Innovation is essential to technological-knowledge progress, which,
in turn, is an engine of economic growth (Gil et al. 2013.). Hence, most
countries should be interested in ways to improve their
technological-knowledge competence.
The causes for innovation are often best reviewed in a competitive
framework: entities, such as governments and firms, that innovate, are
expected to fare better than ones which do not innovate (Werker 2003).
However, this may not be the only reason to innovate: Cassell (2008)
found that their decision to innovate was dependent on more factors than
simply to save costs.
A traditional way of innovating is through research and development
(R&D) efforts. Empirical evidence of this growth mechanism has been
shown in, for example, Lichtenberg (1993) and Coe and Helpman (1995). At
the theoretical level, the first-generation of comprehensive, well
articulated general equilibrium growth models based on R&D that seek
to explore the role of technological knowledge change in the economic
growth process, are centred on two types of R&D
processes--horizontal and vertical.
In the first one, R&D is directed at developing new
horizontally differentiated goods, an approach followed in prominent
works by Romer (1986, 1990), Rivera-Batiz and Romer (1991), Grossman and
Helpman (1991) and Barro and Sala-i-Martin (2004). Since there are no
quality advances, no good ever becomes obsolete. Firms that become
producers remain leaders from then on without further R&D effort,
since they are granted a patent that lasts forever.
In the vertical process, R&D is instead directed at developing
new vertically differentiated qualities of each good, an approach that
was first developed by Segerstrom et al. (1990), Grossman and Helpman
(1991), and Aghion and Howitt (1992). The resulting models are called
Schumpeterian (inspired by the Schumpeterian concept of creative
destruction), or quality ladder models, since, assuming that the
leadership of the firms that use the state-of-the-art qualities is only
temporary--permanently subject to destruction by new qualities resulting
from successful R&D.
Hence, when integrated in endogenous growth models, R&D
activities--either horizontal or vertical--result in
technological-knowledge progress, which, in turn, is the primary
determinant of growth. Two major characteristics of technological
knowledge are essential for its role as an endogenous engine of
growth--non-rivalry and partial non-excludability. Technological
knowledge is non-rival in the sense that the marginal costs for its use
by an additional firm are negligible; and it is partially non-excludable
since the returns to private investment in its production are partly
private and partly public. As a result, the total return on innovation
for the society as a whole (the social return) is greater than the
private return.
As a result, in general, it is considered decreasing returns to
R&D (e.g. Ha, Howitt 2007). The phenomenon of decreasing returns to
R&D means that innovating becomes more and more costly as measured
in R&D costs per increase in quality of goods or decrease in factor
inputs. This is often caused by the exhaustion of technological
paradigms that the R&D effort has taken place in so far (e.g. Dosi
1982). However, Madsen (2007) has suggested that decreasing returns to
R&D cannot always be easily found.
To overcome this problem a new technological paradigm needs to be
found. In fact, one needs to innovate more radically. Here, radical
innovation is not limited to a particular good or service strictly
deviating from the path it has taken so far, being defined as deviating
enough to start improving a good at a reasonable R&D efficiency
again. This kind of innovation is distinct from the incremental type,
which does not involve new paradigms--e.g. Dosi (1982) and Freeman
(1991).
For countries that are not leaders (some developed and developing
countries), the first major way of overcoming the problem of decreasing
returns to R&D is by imitating innovations from leaders. When
successful, imitation allows for the technological-knowledge difusion
embodied in a good, as the imitator reverse-engineers that good.
Imitation is often a less costly process than innovating: Mansfield et
al. (1981) reports that the cost of imitation is, on the average, about
65% of the cost of innovation. Thus, imitation will be often the best
choice for follower countries to catch up (Afonso 2012, 2013; Motohashi
et al. 2012). However, the imitating process on its own does not bring
innovation in a global system; it only aids its diffusion (Afonso 2012,
2013).
Moreover, the existence of spillovers is not sufficient for
diffusion and catching up by the follower countries. In fact, diffusion
requires the followers to absorb the technological-knowledge spillovers;
i.e. to have the capacity to incorporate the innovative spillovers in
their own R&D. Imitation capacity is enhanced by domestic policies
promoting R&D (e.g. Aghion et al. 2001), by the degree of openness
and other trade policies (e.g. Coe et al. 1997), and decreases with the
human-capital gap in relation to the leaders (e.g. Nelson, Phelps 1966).
In the presence of imitation capacity, the advantage-of-backwardness
assumption is observed. This assumption considers that the rate of
technological-knowledge progress in the followers is an increasing
function of the gap between their own technological-knowledge level and
that of the leaders (e.g. Barro, Sala-i-Martin 1997; Afonso 2012, 2013).
The combination of old ideas and concepts into new ideas, or
recombination, is often a successful strategy for innovation (e.g.
Galunic, Rodan 1998). A good strategy is to collaborate with other
partners in order to promote exchanges between their distinct knowledge
stocks.
International trade can play a major role by aiding the
recombination of ideas and concepts. Trading also allows for a more
efficient knowledge flow in general (e.g. Coe et al. 1997; Afonso 2012).
Recombination often occurs when goods are imported. It can lead to a
variety of not only adapted goods, but also entirely new uses. Besides,
having recombination by the importer, problems can be discussed in
different contexts and techniques for production can be learned from
abroad to the advantage for exporting firms (e.g. Silva et al. 2013).
Recently most research in the area of international trade and
innovation is focused on the effects of export, which are important
indeed. The export-space (types of goods that are exported) of a country
is a good predictor for new export (e.g. Hidalgo et al. 2007). Export
variety is important for economic growth (e.g. Saviotti, Frenken 2008).
Firms that start to export are learning faster than firms that do not
(e.g. Silva et al. 2013).
Imports account for economic growth as well since they provide
essential factors for production and allow for the transfer of knowledge
that can be used to improve production and to foster innovation (e.g.
Bayoumi et al. 1999). The easiest way that import can contribute to
local innovation is by aiding imitation, especially for less-developed
countries (e.g. Afonso 2012).
Still, even a view that considers both imports and exports could be
too limited. First, because international trade does not occur only as
an independent bilateral experience since trade relations are mutually
dependent. Thus, a network approach could be valuable. In this line,
Shih et al. (2009), instead of purely bilateral indicators, as bilateral
import and export parameters, considered the centrality of countries in
the trade network as a measure for diffusing technological knowledge.
Moreover, technological knowledge that comes from trade import is
likely to decline over time if the relation is only one-sided. This is
because less and less new ideas, goods and processes can be transferred,
since the unused technological knowledge is decreasing. Over time the
technological knowledge import would be scaling with the technological
knowledge production in the exporting country. And, since its production
faces decreasing returns to R&D, this flow of new knowledge probably
dries up as well.
To overcome these problems and to keep on innovating, a
self-enforcing, or autocatalytic, process needs to be instated. An
autocatalytic system will create exponential returns due to its positive
feedback mechanism. Positive feedback mechanisms are possible only when
there is a cyclical nature to information within a system. After all
there needs to be a feedback, which implies that information of any
activity is related back to the source. The notion of cyclical
autocatalytic processes is not new and is often derived from chemistry
and ecology (e.g. Matutinovic 2005).
Combining the notion of autocatalytic cyclical processes with the
importance of trade to innovation provokes the question: Could
autocatalytic trade cycles be a positive feedback system for innovation?
The nature of innovation suggests they can: innovation will be more
persistent if it occurs in a cyclical system. This is because in a cycle
there will be a continuous recombination of knowledge stocks, which
should then lead to continuous innovation. This way of
technological-knowledge production does not face decreasing returns
since technological paradigms will be often shifted. As discussed above,
trade is conducive, not only to copying and diffusing technological
knowledge but also to producing innovation itself. Hence, it seems that
autocatalytic trade cycles produce continuous innovation. If that is
true then the partners of these autocatalytic trade cycles will
experience more economic growth than otherwise.
This discussion leads to the main research question of this paper:
is it beneficial, in terms of economic growth, for a country to be part
of an autocatalytic trade cycle? If this is true, then innovating and
creating economic growth while being part of an autocatalytic trade
cycle should be more efficient than innovating outside of a cycle. Thus,
this indicates the follow up question: is it possible to develop a
policy that takes advantage of autocatalytic trade cycles? If the
results from the first research question allow policy implications to be
derived, then the question can be answered positively. However if the
results from the first research question are more ambiguous this will
not be possible.
The remainder of this paper will provide an answer to these two
research questions. In the following section the research questions will
be more elaborated on by stating hypotheses that relate to these
questions. When testing these hypotheses, answers to the research
questions can be provided. In Section 2, the methodology for testing the
aforementioned hypotheses will be explained. Section 3 discusses the
results of these tests. General policy recommendations based on these
results are provided in Section 4. Finally, the conclusions of this work
on autocatalytic trade cycles will be shown, based on a comparison
between the answers to the first and the second research questions.
1. Hypotheses
To investigate the research questions a new approach is proposed to
capture different economic aspects. It was chosen because, at present,
there has been no research on autocatalytic trade cycles.
For studying different autocatalytic trade cycles it is important
to define the length of such a cycle, which will be defined by the
number of countries that are involved in it. The hypotheses relate to
the effects of autocatalytic trade cycles on the economic performance of
the involved countries. These hypotheses will start at the relation
between autocatalytic trade cycles and general economic performance, and
then move to the different cycles per good category, and will finally
consider the length of the autocatalytic trade cycles.
The first research question is related to the idea that countries
that are part of autocatalytic trade cycles can economically benefit
from it. Hence, the following hypothesis is considered:
H(i): Countries that are part of autocatalytic trade cycles
experience more economic growth than countries which are not. If it is
beneficial to be part of an autocatalytic trade cycle then over time
trade connections that constitute an autocatalytic trade cycle will be
more likely to grow than connections that do not constitute
autocatalytic trade cycles: the innovation that should occur will likely
increase the trade flow. Since this happens more in autocatalytic trade
cycles, trade flows that constitute these cycles will be growing faster.
Therefore, a second hypothesis is put forward:
H(ii): In a trade network, the trade flows that constitute
autocatalytic trade cycles are relatively higher than flows that are not
part of an autocatalytic trade cycle. Since the autocatalytic trade
cycle argument revolves around innovation, it is expected that more good
categories which are experiencing major changes due to innovation will
have more important autocatalytic trade cycles. Since these changes are
not spread equally through time and good categories, the importance of
autocatalytic trade cycles will vary across these two parameters. This
brings up two additional hypotheses:
H(iii.1): The importance of autocatalytic trade cycles will vary
for different categories of goods and times.
H(iii.2): The autocatalytic trade cycles from innovative goods will
be more important to the economic growth than the autocatalytic trade
cycles from less innovative goods.
However, not all autocatalytic trade cycles have to bring in an
equal amount of economic growth. One would expect that the length of a
cycle is an important factor. Since innovation often occurs due to the
recombination of ideas, one would expect longer autocatalytic trade
cycles to be more conducive to innovation. Nonetheless, for novelty to
occur it needs to fit in with the socio-economic system in which the
innovation is taking place. The more different the socio-economic system
is the less likely the innovation will fit and thereby the performance
of goods will deteriorates. This holds true even for high-tech goods
(e.g. Getler 1995). Thus, a better fit of an innovation is more likely
to happen with shorter cycles, as ideas and practices get less affected
by other systems. When combining these two aspects, a trade-off is
expected. But the results of this trade-off might be different for
distinct categories of goods, relying on the innovation that is
happening at that time in the category. Hence:
H(iv): There is an optimal length at which autocatalytic trade
cycles are most conducive to economic growth. This optimal cycle length
may vary per category of goods.
2. Methodology
2.1. The trade network
The methodology for testing the hypotheses is based on extracting
cycles from the international trade data in the National Bureau of
Economic Research (NBER) database (Feenstra et al. 2005). This database
lists all international trade flows, documented by the United Nations
(UN) from 1962 to 2000. The trade flows are listed by category of goods
on the basis of the Standard Industrial Classification in four digits
(SIC 4). After the collection of data, the autocatalytic trade cycles
are calculated for ten different classes of trade, indicated by the
first digit of the SIC 4 classification. This is done per year over the
period from 1962 to 2000.
The main decision here is to consider which trade flows are
relevant: if all trade flows are included, the trade network will be too
dense to have a relevant measure of the number of autocatalytic trade
cycles. Therefore, the less important trade flows need to be excluded
from the analysis. To assist this procedure a quality measure
[q.sub.i,j] for a trade flow [t.sub.i,j] from country [c.sub.i] to
[c.sub.j] with value [v.sub.i,j] will be introduced:
[q.sub.i,j] = [v.sub.i,j]/MINTRADE ([c.sub.i], [c.sub.j]). (1)
In (1), MINTRADE is an operator which picks the minimum sum of the
imports and exports of the countries involved. This sum of imports and
exports is also referred to as a measure of openness for a given economy
if it is divided by Gross Domestic Product (GDP). It was decided not to
divide it by GDP since the inclusion of GDP in the selection process
would mean that openness of the economy would be a major factor
influencing the number of links per country.
Since this paper deals mainly with the structure of the network it
is better to make the selection process based mostly on the trade volume
per country. This means, from a network perspective, that selection is
solely based on the property of the links, regardless of the properties
of the nodes. Hence, with a substantial reduction of links, most nodes
will be more equal in their number of links in the trade network. This
occurs because nodes with a large number of links will lose more links
than nodes with a small number of them, since the quality of the links
will be lower when a node has a larger number of links. Moreover, this
means that here the autocatalytic trade cycles refer to the structure of
the trade per node. It does not value more open economies more than more
closed economies. There is also the advantage that trade flows involving
countries with very open economies (e.g. because they are at a trade
nexus such as Singapore and the Netherlands) will be not be weighed
disproportionally.
To choose the minimum quality value qmin from which trade flows are
admitted, the following reasoning is used: to assert a relative
independence for countries who are part of autocatalytic trade cycles it
is important that their autocatalytic trade cycles do not overlap too
much. A key measure for this is the average path length of the trade
network related to the cycle length. If the average path length is
shorter than the cycle length most countries will share at least one
autocatalytic trade cycle. Thus, the cycle length cannot exceed the
average path length too much.
The network that is used sometimes includes countries that are not
part of any cycle. If they were included in the measure for the average
path length, this measure would be distorted. This is because they
contribute with their lower connections to a higher average path length.
Hence, the countries that are in autocatalytic trade cycles are
connected more than the average path length indicates. To omit this
problem an adjusted measure for average path length is used in which
these countries are excluded. To estimate the average path length l the
formula as derived in Fronczak et al. (2004), for a random network with
N nodes and a distribution of k links per node is used:
l = -2 < ln k >+ ln < N >+ ln < k (k -1) >
-[gamma]/ln N + ln < k (k -1) > - ln[beta] + 1/2. (2)
In (2), [gamma] [approximately equal to] 0.5772 is Euler's
constant while [beta] is a constant depending on the kind of
distribution of links we expect. It will be assumed that the
distribution corresponds roughly to that of an Erdos-Renyi random
network with a Poisson distribution. (1) The validation of this is that,
as mentioned earlier, most nodes are becoming more equal in the amount
of links. Within the trade networks in study, often a large portion of
links (75-90%) needed to be deleted in order to achieve the desired
average path length. Therefore, this procedure negated most of the
original distribution of links per node, resulting in a Poisson
distribution. This was validated by examining the distribution of links
in a number of different trade networks after a significant reduction of
low quality links. With this distribution [beta] would have the
following value: N < k >.
From Fronczak et al. (2004) it seems this formula underestimates l
slightly for low N(N~[10.sup.2]); this is not a problem since the
average path length sets a maximum for the cycle length, hence a slight
underestimation of this parameter will lead to results that are more
reliable than needed. The following procedure is applied: the minimum
quality value qmin is raised, thus allowing more flows to be deleted
until the required average path length is achieved. Occasionally this
procedure leads to an empty network, because the required average path
length could not be achieved. In this case the network with the highest
average path length, which has appeared in the previous procedure, will
be chosen. These path lengths are often very close to the required
average path length. Judging from the likely values for N and the
distribution of k, the highest average path length that can be obtained,
while still including most countries, is 2. This indicates that the
maximum cycle length that can be measured is a cycle including 4
countries.
2.2. Determining the effect of autocatalytic trade cycles on
economic growth
The process of determining the effect of autocatalytic trade cycles
will yield 6 indicators per category of goods, which combines magnitude
of autocatalytic trade cycles per each length of the autocatalytic trade
cycle per se. In this case there are cycles with 2, 3 and 4 countries.
The magnitude of an autocatalytic trade cycle is defined here as its
circulation value, or as the lowest value of the trade flows that it
consists of. The 6 indicators are probably highly correlated, since the
likelihood of a country being in a cycle with 4 countries grows as the
number of cycles with 3 countries, involving that country grows. As a
result, a multiple regression with all indicators as independent
variables will suffer from multicollinearity. To overcome this problem,
a factor analysis will be used to reduce these 6 indicators into 1 or 2
composite variables, depending on the outcome of the analysis.
Because of the nature of the selection process, the major
confounding variables here are imports and exports, as they will
probably influence the trade flows that were chosen in the trade network
and they will also have an effect on economic growth. The third
confounding variable to consider is GDP since it can be assumed that
countries with different GDP levels will have different trade
structures, e.g. a core periphery structure in the world economy. GDP
will also affect GDP growth because different sized economies will grow
at different rates. These variables will be used to predict economic
growth as measured in the relative increase of real GDP. (2) Because it
is expected that innovation processes operate on timescales larger than
a year and on different timescales for different goods, time series for
the 10 different goods will be constructed. In these time series the
composite variables will estimate both the annual rate of GDP growth and
the rate of aggregate GDP growth for consecutively more years ahead.
2.3. Determining the effect of autocatalytic trade cycles on trade
To test hypothesis (ii) the number of autocatalytic trade cycles
that a certain trade flow is in will be determined. Due to the selection
process of the trade network a comparison between flows that are in or
out of autocatalytic trade cycles cannot be made. This is because trade
flows that are not in the network are by definition smaller than flows
that are in the trade network, since small trade flows will have a
higher chance of being deleted than larger trade flows.
Therefore, in order to test hypothesis (ii), a comparison will be
made between the number of autocatalytic trade cycles a trade flow is in
and its value. Since there are 3 different sizes of cycles there will be
3 different indicators. With these indicators a multivariate linear
regression with the trade value as a dependent variable will be run.
Unfortunately, the problem of multicollinearity is harder to solve
because there are less indicators; however, the problem will be also
smaller due to the same reason.
2.4. Determining the effect of autocatalytic trade cycles in
different categories of goods
Hypotheses and (iii.2) suggest that the autocatalytic trade cycles
associated with different goods will have different effects on economic
growth. To test for this, the 6 indicators, for the combined value and
number of 2, 3 and 4 sized autocatalytic trade cycles, will be used to
compare the effects for each category of goods. These categories are
based on the first digit from the SIC 4 classification (Table 1).
A high correlation will be expected in these indicators for the
same reason as described in Section 2.2. Thus, the same process of using
factor analysis to extract 1 or 2 composites out of these 6 indicators
will be used. The main comparison will be made on the normalized effect
of the composites on growth. This effect will be determined using
multivariate linear regressions, with the same confounding variables as
mentioned in Section 2.2. A time series approach will be used because
there is no a priori assumption on the typical timescale of innovation.
This holds even more for different goods categories. It can be seen if
autocatalytic trade cycles from different categories of goods have
different effects on different timescales.
2.5. Determining the effect of autocatalytic trade cycles of
different lengths
Hypothesis (iv) deals with the effects of different lengths.
Indicators of autocatalytic trade cycles of different lengths are often
highly correlated. If this is the case it is not possible to compare the
effects of different lengths of autocatalytic trade cycles in that
particular category of goods. However, if the indicators are not highly
correlated, testing hypothesis (iv) becomes a possibility. If the
abovementioned indicators do not correlate very well, the factor
analysis will show it. Hence, the same set of composite indicators and
regressions can be used to test this hypothesis for a limited set of
goods.
3. Results
3.1. Hypothesis (i): the effects of autocatalytic trade cycles on
GDP growth
This hypothesis deals with the effect of autocatalytic trade cycles
on GDP of the countries involved. The country to country trade data was
used to create a network out of which the autocatalytic trade cycles
could be calculated. The factor analysis that followed indicated that
two separate variables were needed to represent the indicators from the
autocatalytic trade cycles: variable 1 to represent data mostly from the
2 and the 3 sized cycles and variable 2 to represent the 4 sized cycles.
Results from the linear regressions with GDP per year as a
dependent variable, and in years following, are depicted in Figure 1.
For variable 1 there is a strong and significant correlation between the
indicator and GDP increase per year. It also appears that there are two
timescales: one of about 2-3 years later and other of 9 years later.
Variable 2 seems to have an insignificant and small effect. Only at 2
and 6 years there is a significant effect which could point to a certain
timescale, but since that effect is quite small this is not a robust
finding.
Results from the regressions with aggregated GDP over a certain
amount of years (see Fig. 1) confirm the previous findings: variable 1
has a significant impact on GDP growth for all time periods, while
variable 2 is mostly small and insignificant. From the time scales for
variable 1 only the scale around 11 years is visible in the aggregated
GDP regressions. The 6 year time scale of variable 2 becomes more
insignificant, while the 2 year time scale is still significant. But
since there is relatively little difference between the graphs on small
time scales this could be expected. Results of the regressions indicate
a significant relation between the amount, both in value and number, of
autocatalytic trade cycles and GDP growth. Thus the hypothesis should be
accepted.
[FIGURE 1 OMITTED]
3.2. Hypothesis (ii): the effect of autocatalytic trade cycles on
trade flows
In this hypothesis the effects of being part of autocatalytic trade
cycles on the size of the trade flows is examined. If the hypothesis is
to be accepted there should be a significant positive effect of being in
an autocatalytic trade cycle for a trade flow. In total 1.241.627 trade
flows were included to test this hypothesis.
The covariance matrix shows that there is a reasonable, but small
correlation between the dependent variable and the 3 indicators (Table
2). Unfortunately the correlation between 3 and 4 sized cycles is very
large. This means that their respective coefficients are less reliable;
this is compensated for by the large number of cases that could be
examined.
Results of the linear regression in Table 3 indicate that there is
a significant relationship between the value of a trade flow and its
presence in one or more cycles. However, the relationship differs for
different sized cycles. Being in a 2 sized cycle has no significant
effect on the trade flow. Being part of a 3 sized cycle however, does
have a major and significant effect on the size of the flow. Being in a
4 sized cycle has a significant negative relationship to magnitude of
the trade flow.
Looking at the standardized coefficients, it is possible to see
that being in more autocatalytic trade cycles is overall positive for
the value of a trade flow. However, the size of the trade flow is a
significant variable to consider.
3.3. Hypothesis (iii): the importance of autocatalytic trade cycles
from different categories of goods
Now the independent variables are the autocatalytic trade cycles
per category of goods. The first part of the hypothesis states that
autocatalytic trade cycles in different categories have different
effects on GDP growth. The second part of the hypothesis states that the
effect on GDP growth should be more profound for more innovative goods.
To test this hypothesis the following method was used: first
different trade networks were made for different goods; second the
autocatalytic trade cycle indicators were calculated for each country;
then factor analyses were made to determine the composite variables for
each category of good; finally two series of regressions were made, one
with GDP growth per year in x years following, and another with
aggregated GDP over x years as dependent variables.
Since there were 10 categories of goods tested, only a few results
will be discussed in detail here. These results should provide a clear
picture of the overall trends. From the other categories a short summary
will be made.
3.3.1. Results from some categories of goods
Results that will be discussed in detail are from the following
categories of goods: SIC 2, SIC 4 and SIC 9. (3) These categories were
chosen because they represent either a group of categories, (SIC 2 and
SIC 9) or are unusual compared to the other categories (SIC 4).
Regarding the results from category SIC 2 (see Fig. 2), the factor
analysis indicated two variables. Variable 1, mainly dealing with
2&3 sized autocatalytic trade cycles and variable 2, mainly dealing
with 4 sized autocatalytic trade cycles. The results of the year on year
growth of GDP show that there is a significant relation between variable
1 and year on year GDP growth for the entire time span of 15 years,
although the strength is decreasing over time. The inverse is true for
variable 2, which is strongly and significantly negative for most values
and also decreasing over time. The results from year on year GDP
regressions are confirmed by the aggregate GDP growth. These series show
a stable and significant relation from both variables with aggregate GDP
growth.
The category of goods of SIC 4 had the following characteristics
(see Fig. 3): the factor analysis indicated two variables, as was the
case in SIC 2. Variable 1 included all indicators (variable 1), while
variable 2 only contained the numbers of cycles (variable 2). The year
on year regressions show that variable 1 has a small and mostly
insignificant relation to the dependent variable. There are two
significant timescales: a negative for the first 3-4 years and a
positive for a timescale of roughly 11 years. Variable 2 is mostly
positive and significant except for the time scale of 11 years, which
could be explained by the rise in the coefficient of variable 1. The
results from the regressions with the aggregate GDP growth show a more
clear result: variable 1 is negative, strong and significant related to
the aggregate GDP growth, while variable 2 is the inverse of this.
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
The category of goods of SIC 9 had the following results (see Fig.
4): the factor analysis was inconclusive over the amount of variables to
be included. Since variable 2 had no interpretable relation to the
autocatalytic trade cycle indicators, it was decided to include only one
variable. The standardized coefficient of the variable for regression
for year on year GDP growth is positive and significant with a maximum
at a 7 year timescale and a minimum around an 11 year time scale. The
results from the series of regressions with aggregate GDP growth show a
clear maximum around 7 years, indicating a 7 year time scale, while also
at other time scales showing continuously positive significant results.
[FIGURE 4 OMITTED]
3.3.2. Summary of the results for the other categories of goods
Results from the other categories of goods could be gathered into
three different groups: the first group has categories similar to SIC 2
with two variables representing respectively 2&3 sized cycles and 4
sized cycles. This group contains SIC 0 and SIC 8. Furthermore, the
second group is similar to SIC 9 and has one variable representing all
indicators. SIC 3 and SIC 7 are represented in here. Finally, the last
group had insignificant results for all variables that were tested and
consists of SIC 1 and SIC 5, and SIC 6.
The group similar to SIC 2 contains the following categories of
goods: SIC 0 and SIC 8. In both of these categories variable 1 is
positively related to GDP growth on a year on year basis, while variable
2 is insignificant but mostly negative for most of the time span.
Variable 1 for both SIC 0 and SIC 8 has significant results for the time
scales of 2 and 9 years. In the analysis of aggregate GDP growth
variable 2 is significantly negative for both categories. Variable 1 is
constant and almost significant for SIC 0 and only significant at a time
scale of 3 years for SIC 8.
In the second group, which is similar to SIC 9, there is only one
variable to represent all of the autocatalytic trade cycle indicators.
This group consists of SIC 3 and SIC 7: on a yearly basis the
coefficients from both categories are mostly insignificant. It is mostly
positive for SIC 3, while SIC 7 gives a more mixed view. On the
aggregate GDP growth the variable from category SIC 3 has a significant,
but not very strong, relationship with the dependent variable. This
relationship has its maximum at roughly 7 years. The variable from the
SIC 7 category has a continuously positive relationship with a
significant maximum at a time scale of around 3-4 years. After that
maximum the relation becomes insignificant for the remainder of the
timescales.
3.3.3. Conclusion
In the majority of the 10 examined categories, there was a
significant relationship between the composite variable and the GDP
growth. This relation differs per category, and per time scale (see
Table 4).
The categories with the strongest relation between their respective
autocatalytic trade cycles and GDP growth were manufacturing, transport
and governmental services. Therefore, it seems that industries, where
large capital investments need to be made, profit more from
autocatalytic trade cycles than industries without those requirements.
This does not fully correspond to hypothesis (iii.2). However, since the
goods categories were fairly broad, this warrants further investigation.
3.4. Hypothesis (iv): the optimal length of an autocatalytic trade
cycle
Hypothesis (iv) states that there is an optimal size of an
autocatalytic trade cycle and that this optimal length probably varies
for different categories of goods.
The optimal length of an autocatalytic trade cycle was hard to
verify since the lengths of autocatalytic trade cycles are highly
correlated (Table 3). However, for a number of distinct categories of
goods there was a comparison possible between 2&3 sized
autocatalytic trade cycles, on the one hand, and 4 sized autocatalytic
trade cycles, on the other hand.
These categories were SIC 0, SIC 2, SIC 8 and the overall trade
between countries. The variables representing 2&3 sized
autocatalytic trade cycles were consistently positively related to GDP
growth, while the variables representing 4 sized autocatalytic trade
cycles were consistently negatively related to GDP growth. Therefore,
for these categories the optimal length of autocatalytic trade cycles is
either 2 or 3. The results from hypothesis (ii) indicate that this
optimal autocatalytic trade cycle length is more likely to be 3 than 2.
For the other categories this optimal autocatalytic trade cycle
length is probably larger than 3. However, this is hard to test for
since it was impossible to test for larger autocatalytic trade cycles.
Thus, it is plausible that an optimal autocatalytic trade cycle length
exists and that this is different for different categories. Therefore,
hypothesis (iv) is validated.
4. General policy recommendations
This section will serve as a bridge between the hypotheses dealing
with the effects of autocatalytic trade cycles, and the implementation
of these results.
A number of different policies can be implemented to benefit from
the effects of autocatalytic trade cycles, as discussed in Section 3.
They can be either to change the composition of the economy into more
profitable sectors for autocatalytic trade cycles, or to promote the
strengthening of already existing autocatalytic trade cycles. The
different measures shall be better discussed.
4.1. Changing the composition of the economy
Results in Section 3 show that generally being a part of an
autocatalytic trade cycle is beneficial for growth. Indeed, Section 3.3
describes how different goods have different effects. The largest
effects were registered in manufacturing, both light and heavy,
infrastructure, personal services and (semi-) public goods. However, the
effects had often mixed components. This was the case in light
manufacturing, infrastructure and semi public goods (SIC 8). Thus, the
effects are reduced in these sectors. That leaves heavy manufacturing,
services and the governmental sector. Since the governmental sector is
hardly associated with export/import, and this relation does not seem to
be caused by a direct effect, heavy manufacturing and services are the
sectors that can be promoted.
This selection of sectors leaves two different policies with two
different effects. For a sustained long term effect it is better to
support heavy manufacturing, which will result in a growth that
maximizes about 7 years later but will make effects relatively early and
constant. For a more direct effect, it is better to support the services
industry, which has a maximum effect in 2 years, but this effect will
dissipate quite soon.
It has to be noted that heavy manufacturing requires strong
investments in capital and, since this has a lag in itself, results may
take a long time to be visible. However, the development of a services
industry should not be taken lightly since it will require a significant
investment in human capital, which could take a long time to accumulate
as well.
4.2. Changing regulations with respect to trade
Open economies will be more likely to be part of autocatalytic
trade cycles. However, this paper is related more to the structure of
trade flows rather than to the openness of economies (see Section 2.1).
Hence, just being more open does not mean that autocatalytic trade
cycles are automatically beneficial. However, it is important to be open
in sectors in which these autocatalytic trade cycles are present. The
sectors that are most likely to contribute to economic growth are heavy
manufacturing and services (see Section 4.1). This does not mean that
opening the economy for other sectors is wrong. The autocatalytic
process described in the introduction could work in any industry. This
process will probably function better when the goods are more
innovative. This was tested with hypothesis (iii.2), but since the
tested categories were too broad, this could not be fully verified. This
reasoning suggests that a country should focus to be open in industries
that are innovative and that can profit from international cooperation.
Making the economy more open can be done by reducing trade barriers
on both the export and the import sides of a sector. Foreign entities
often have the possibility to add more innovation to goods than domestic
industries. However, this openness should not destroy the domestic
industry, for without it the capacity to adapt to foreign innovations,
or adoptive capacity (Arrow 1969), will be reduced.
4.3. Forging trade connections
Results of Section 3 suggest that forming autocatalytic trade
cycles is beneficial for economic growth. This is mainly true for small
sized cycles. Thus, a possibility to involve foreign policy presents
itself.
Foreign policy can aid the forming of autocatalytic trade cycles by
promoting domestic goods and services in other countries. This can
create innovation and thereby economic growth back home as was indicated
by Saviotti and Frenken (2008) and Silva et al. (2013). But not only the
export side should be facilitated; promoting the internal market to
foreign companies could also be a worthwhile endeavour. The research by
Bayoumi et al. (1999); Coe and Helpman (1995) indicates that trade
imports support imitation and domestic innovation. This is especially
true when foreign companies start adapting their products and services
to the internal market, thereby contributing their own innovations to
domestic goods. When the trade connection is formed through both imports
and exports, it can lead to the creation of autocatalytic trade cycles.
Industrial policy can also aid forming new autocatalytic trade
cycles: industries should be more focused towards markets from which the
policy-maker knows he/she can learn and which will return the business
directly or indirectly. Also cooperation with importing and exporting
firms abroad can help to start new autocatalytic trade cycles. This
creation would occur in a similar but more direct way than the process
in this paragraph.
4.4. Making existing autocatalytic trade cycles more autocatalytic
Existing autocatalytic trade cycles can also be made more
autocatalytic by providing facilities for trading companies to discuss
changes and problems with their suppliers and customers in other
countries. The more problems and solutions can be diffused the more
ideas and more innovation can be created (Thompson 1965). Making
autocatalytic trade cycles more autocatalytic can be done by directly
bringing suppliers, firms and clients more in touch with each other.
Firms often have too little an idea of whom they are dealing with, and
can profit more by sharing common problems (Dyer 1996). The government
can facilitate these gatherings by promoting factory visits by
foreigners, and by promoting visits abroad. It also should provide a
service to analyse where problems and miscommunication between
international partners persist since cultural and language differences
can make these trade cycles less autocatalytic (Getler 1995).
This should also include future customers, suppliers and domestic
firms. However, this procedure may not be an easy task since most of the
information related to innovation is often proprietary, a trade secret
or patented. Thus, firms will be often reluctant to cooperate to a full
extent to an open sharing of information, unless there is a large degree
of trust between the sharing firms (Kale et al. 2000).
The solution can be the creation of a forum in which trade problems
can be discussed, presenting opportunities for companies to innovate
without giving away too much information. The more open this forum is,
the more players can inform themselves and either present solutions or
innovate to supply better goods. Furthermore, an open exchange can
produce trust so that firms are willing to share more information.
Hence, all players will benefit.
Conclusions
In this paper the effects and implications of autocatalytic trade
cycles have been discussed. Autocatalytic trade cycles are cycles that
are conducive to innovation for the countries involved. The literature
on innovation suggested that autocatalytic trade cycles could be formed
by countries in an trade cycle. Since innovation translates into
economic growth, autocatalytic trade cycles should be an indicator for
this growth.
For these research questions several hypotheses were formed. These
hypotheses were tested with UN trade data over the period 1962-2000,
obtained via the NBER. This data was filtered using an approach which
worked by selecting trade flows. This meant that the selection process
was blind to the total openness of a country, which allowed looking more
into the effects of the structure of the trade network.
The above methodology allowed the construction of indicators which
represented the value and number of three different sized autocatalytic
trade cycles. One or two variables were formed from these indicators,
dependent on the results of a factor analysis. These variables were then
be used to test the hypotheses. It was confirmed that the first variable
from the autocatalytic trade cycles in the general trade network
correlated significant and positively with GDP growth. The second
variable was mostly negative but insignificant. This confirmed the
hypothesis that autocatalytic trade cycles correlate positively with GDP
growth.
The second hypothesis was that trade flows that are being a part of
more autocatalytic trade cycles are bigger than trade flows that are
part of less autocatalytic trade cycles. The trade flows that were used
to test this were all part of at least one autocatalytic trade cycle.
The results of the linear regression showed that only being a part of
more 3 sized trade cycles was significantly positive related to the size
of a trade flow. Being part of a 4 sized cycle was found to be
significantly negative related to this size.
Results confirmed that the effects of autocatalytic trade cycles
differ significantly for different categories of goods. Three goods
categories showed a significant positive relation with GDP growth: heavy
manufacturing, personal services and public administration. Another
three categories had one variable out of the two that related positively
and significantly to GDP growth: light manufacturing, infrastructure and
another section of services. The testing of the second part of the
hypothesis was less conclusive: It was likely, but not certain that more
innovative categories of goods would profit more from autocatalytic
trade cycles. The categories of goods were too broad to confirm this
hypothesis.
The fourth hypothesis could also be validated with the help of the
results from testing hypotheses (ii) and (iii). It was confirmed that an
optimal autocatalytic trade cycle length probably existed and that, for
most categories of goods, this optimal length was likely to be 3. A
general policy recommendations could be made: among others, more direct
cooperation with foreign governments, as well as opening certain markets
for trade.
This paper has shown in general that autocatalytic trade cycles are
having a significant effect on economic growth, thereby answering the
main research question.
However, a central issue with variables related to the total trade
network is that parameters of single countries are always externally
affected by actions of other countries. Thus, it is impossible to
completely separate the measures on ego networks from the total network
configuration, even when strict requirements on average path length are
being fulfilled. Another consideration, which is related to the study of
countries over the years, is that these countries are interconnected on
several other parameters. This means that the success or failure of
economic policy needs to be always seen as being partially dependent on
the success and failures of others. This is true in times of global
recessions and global economic booms.
Next to these general considerations a more specific issue was
encountered when reviewing the indicators of autocatalytic trade cycles.
This review yielded often two variables: One significantly positive
associated with GDP growth; and one significantly negatively associated
with GDP growth. While it was expected to see a divergence in the effect
of differently sized autocatalytic trade cycles as hypothesis (iv)
indicated, the negative association with GDP growth was unexpected. This
indicates that autocatalytic trade cycles have negative effects on GDP
growth. An explanation for this is that longer autocatalytic trade
cycles bring more competition, while not contributing as much to local
innovation.
Future work could focus on better understanding the effects of
autocatalytic trade cycles and producing more tailored advice. As the
results from hypotheses (iii.2) and (iv) have shown, the categories of
goods are still very broad and it would be important to know if the
results still hold on a more detailed level. Furthermore, the effects of
autocatalytic trade cycles, and that of different goods, may differ on a
variety of parameters of the countries involved. For example: the
development level, the amount of human capital, the types of industrial
activities, and the geographical location of the country. Another aspect
that is worth looking into is what the effects are of different partners
of the autocatalytic trade cycles.
Also, it would be important to see how autocatalytic trade cycles
influence other variables besides GDP growth, such as the human
development index, employment or one of the several innovation
variables. Autocatalytic cycles can also occur within the borders of a
single nation, especially when this nation has a large differentiated
economy (e.g. the USA). A research on the effects of autocatalytic trade
cycles between different sectors within an economy could help to
understand better the effects of autocatalytic trade cycles, because it
reduces the effects of culture, language and geography in these cycles.
It would be interesting to see the relationship between the optimal
length of an autocatalytic trade cycle and these three parameters. From
a policy perspective it would also be interesting to see if
autocatalytic trade cycles could be used in the same way as exports are
used in Hidalgo et al. (2007). This means that the presence of a country
in an autocatalytic trade cycle could be an important indicator for the
industry of that particular country. Finally, it is relevant to
understand that this research has been undertaken on a macro scale. The
precise workings of autocatalytic trade cycles are still unknown. It
would be important to understand under what conditions these cycles form
and which actors are involved.
In conclusion, to the best of our knowledge, this paper is the
first one to discuss the effects and implications of autocatalytic trade
cycles. It is interesting to see that a variable, like autocatalytic
trade cycles, has such a relation with GDP growth. This is even more
interesting since there is no a priori reason for the effects observed,
other than the autocatalytic trade cycle argument provided in this
paper. The results also provide a further insight in the ways that
innovation is operating. This highlights the importance on studying and
using innovation to improve economic growth and to improve the general
welfare of societies. It would therefore be a worthwhile endeavour to
research this subject in depth, using the results of this subfield to
improve the welfare of our societies.
doi:10.3846/16111699.2012.720596
Caption: Fig. 1. Relation between the time scale and the
standardized coefficients for the composite variables
Caption: Fig. 2. Relation between the time scale and the
standardized coefficients for the composite variables
Caption: Fig. 3. Relation between the time scale and the
standardized coefficients for the composite variables
Caption: Fig. 4. Relation between the time scale and the
standardized coefficients for the composite variables
References
Aghion, P.; Howitt, P. 1992. A model of growth through creative
destruction, Econometrica 60(2): 323-352.
http://dx.doi.org/10.2307/2951599
Aghion, P.; Harris, C.; Howitt, P.; Vickers, J. 2001. Competition,
imitation and growth with step-by-step innovation, Review of Economic
Studies 68(3): 467-492. http://dx.doi.org/10.1111/1467-937X.00177
Afonso, O. 2012. Scale-independent North-South trade effects on the
technological-knowledge bias and on wage inequality, Review of World
Economics 148(1): 181-207. http://dx.doi.org/10.1007/s10290-011-0109-7
Afonso, O. 2013. Diffusion and directed technological knowledge,
human capital and wages, Economic Modelling 31: 370-382.
http://dx.doi.org/10.1016/j.econmod.2012.11.011
Alesina, A.; Spolaore, E. 1997. On the number and size of nations,
Quarterly Journal of Economics 112(4): 1027-1056.
http://dx.doi.org/10.1162/003355300555411
Arrow, K. 1969. Classificatory notes on the production and
transmission of technological knowledge, American Economic Review,
Papers and Proceedings 59(2): 29-35.
Barro, R. J.; Sala-i-Martin, X. 2004. Economic growth. 2nd ed.
Cambridge, MA: MIT Press.
Bayoumi, T.; Coe, D.; Helpman, E. 1999. R&D spillovers and
global growth, Journal of International Economics 47: 399-428.
http://dx.doi.org/10.1016/S0022-1996(98)00018-X
Cassell, M. 2008. Why governments innovate: adoption and
implementation of open source software by four european cities,
International Public Management Journal 11(2): 193-213.
http://dx.doi.org/10.1080/10967490802095680
Coe, D. T.; Helpman, E. 1995. International R&D spillovers,
European Economic Review 39(5): 859-897.
http://dx.doi.org/10.1016/0014-2921(94)00100-E
Coe, D. T.; Helpman, E.; Hoffmeister, A. 1997. North-south R&D
spillovers, Economic Journal 107(440): 134-149.
http://dx.doi.org/10.1111/1468-0297.00146
Dosi, G. 1982. Technical paradigms and technological trajectories:
a suggested interpretation of the determinants and directions of
technical change, Research Policy 11(3): 147162.
http://dx.doi.org/10.1016/0048-7333(82)90016-6
Dyer, J. 1996. Specialized supplier networks as a source of
competitive advantage: evidence from the auto industry, Strategic
Management Journal 17(4): 271-291. http://dx.doi.org/10.1002/
(SICI)1097-0266(199604)17:4<271::AID-SMJ807>3.0.CO;2-Y
Erdos, P.; Renyi, A. 1960. On the evolution of random graphs,
Hungarian academy of sciences 5: 17-61.
Feenstra, R. C.; Lipsey, R. E.; Ma, H.; Mo, H. 2005. World trade
flows: 1962-2000, National Bureau of Economic Research (NBER) Working
paper series 11040 [online], [cited January 2012]. Available from
Internet: http://www.nber.org/papers/w11040
Freeman, C. 1991. Innovation, changes of techno-economic paradigm
and biological analogies in economics, Revue Economique 42(2): 211-231.
Fronczak, A.; Fronczak, P.; Hofyst, J. A. 2004. Average path length
in random networks, Physical Review 70: 56-110.
Galunic, D. C.; Rodan, S. 1998. Resource recombinations in the
firm: knowledge structures and the potential for schumpeterian
innovation, Strategic Management Journal 19: 1193-1201.
http://dx.doi.org/10.1002/(SICI)1097-0266(1998120)19:12<1193::AID-SMJ5>3.0.CO;2-F
Getler, M. S. 1995. Being there: proximity, organization, and
culture in the development and adoption of advanced manufacturing
technologies, Economic Geography 71(1): 1-26.
http://dx.doi.org/10.2307/144433
Gil, P. M.; Brito, P.; Afonso, O. 2013. Growth and firm dynamics
with horizontal and vertical R&D, Macroeconomic Dynamics 17(7):
1438-1466. http://dx.doi.org/10.1017/S1365100512000181
Grossman, G. M.; Helpman, E. 1991. Innovation and growth in the
global economy. Cambridge, MA: MIT Press.
Ha, J.; Howitt, P. 2007. Accounting for trends in productivity and
R&D: a schumpeterian critique of semi-endogenous growth theory,
Journal of Money, Credit and Banking 39: 733774. http://
dx.doi.org/10.1111/j.1538-4616.2007.00045.x
Hidalgo, C. A.; Klinger, B.; Barabosi, A.-L.; Hausmann, R. 2007.
The product space conditions the development of nations, Science 324:
317-482.
Kale, P.; Singh, H.; Perlmutter, H. 2000. Learning and protection
of proprietary assets in strategic alliances: building relational
capital, Strategic Management Journal 21(3): 217237.
http://dx.doi.org/10.1002/(SICI)1097-0266(200003)21:3<217::AID-SMJ95>3.0.CO;2-Y
Lichtenberg, F. 1993. R&D investment and international
productivity differences, in H. Siebert (Ed.). Economic growth in the
world economy. Tubingen: J. C. B. Mohr. 89-110.
Madsen, J. 2007. Are there diminishing returns to R&D?,
Economics Letters 95(2): 161-166.
http://dx.doi.org/10.1016/j.econlet.2006.09.009
Mansfield, E. M.; Schwartz; Wagner, X. 1981. Imitation costs and
patents: an empirical study, Economic Journal 91(364): 907-918.
http://dx.doi.org/10.2307/2232499
Matutinovic, I. 2005. The microeconomic foundations of business
cycles: from institutions to autocatalytic networks, Journal of Economic
Issues 39(4): 867-898.
Motohashi, K.; Lee, D.; Sawng, Y.; Kim, S. 2012. Innovative
converged service and its adoption, use and diffusion: a holistic
approach to diffusion of innovations, combining adoption-diffusion and
use-diffusion paradigms, Journal of Business Economics and Management
13(2): 308-333. http://dx.doi.org/10.3846/16111699.2011.620147
Nelson, R.; Phelps, E. 1966. Investment in humans, technological
diffusion, and economic growth, American Economic Review 56: 69-75.
Rivera-Batiz, L.; Romer, P. 1991. Economic integration and
endogenous growth, Quarterly Journal of Economics 106(2): 531-555.
http://dx.doi.org/10.2307/2937946
Romer, P. 1986. Increasing returns and long-run growth, Journal of
Political Economy 94(2): 1002-1037. http://dx.doi.org/10.1086/261420
Romer, P. 1990. Capital, labor, and productivity, Brookings Papers
on Economic Activity (Microeconomics Special Issue): 337-367.
Saviotti, P.; Frenken, K. 2008. Export variety and the economic
performance of countries, Journal of Evolutionary Economics 18: 201-218.
http://dx.doi.org/10.1007/s00191-0070081-5
Segerstrom, P. S.; Anant, T. C. A.; Dinopoulos, E. 1990. A
schumpeterian model of the product life cycle, American Economic Review
80(5): 1077-1091.
Shih, H.; Tung-Lung, X.; Chang, S. 2009. International diffusion of
embodied and disembodied technology: a network analysis approach,
Technological Forecasting & Social Change 76: 821-834.
http://dx.doi.org/10.1016/j.techfore.2008.09.001
Silva, A.; Afonso, O.; Africano, A. P. 2013. Economic performance
and international trade engagement: the case of portuguese firms,
International Economics and Economic Policy 10(4): 521-547.
http://dx.doi.org/10.1007/s10368-012-0221-8
Thompson, V. 1965. Bureaucracy and innovation, Administrative
Science Quarterly 10: 1-20. http://dx.doi.org/10.2307/2391646
Werker, C. 2003. Innovation, market performance, and competition:
lessons from a product life cycle model, Technovation 23: 281-290.
http://dx.doi.org/10.1016/S01664972(01)00109-2
Received 09 March 2012; accepted 09 August 2012
Jurrien J. Bakker (1), Oscar Afonso (2), Sandra T. Silva (3)
(1) Managerial Economics, Strategy and Innovation (MSI), Faculty of
Economics and Applied Economics, K.U. Leuven, Naamsestraat 69 bus 3500,
B 3000 Leuven, Belgium
(2) CEFUP, OBEGEF, NIFIP, Faculty of Economics, University of
Porto, Rua Roberto Frias, s/n, 4200-464 Porto, Portugal
(3) CEFUP, Faculty of Economics, University of Porto, Rua Roberto
Frias, s/n, 4200-464 Porto, Portugal
E-mails: (1) j.j.bakker@student.tue.nl; (2) oafonso@fep.up.pt
(corresponding author); (3) sandras@fep.up.pt
(1) This is a network with the underlying assumption that the
chance of a link existing between two nodes is equal for all nodes. This
is unlike, e.g. a scale free network, where this is dependent on the
number of links a node already has. Erdos and Renyi (1960) have shown
that the distribution of links in a random network follows a Poisson
distribution.
(2) GDP data was obtained from the World Bank. 2011.[Online],
[cited January 2012]. Available from Internet: http://www.worldbank.org.
(3) Descriptions of the categories are given in Table 1.
Jurrien J. BAKKER (25) has obtained a Bsc in Applied Physics and a
Msc in Innovation Sciences at Eindhoven University of Technology. He has
completed an Erasmus semester in Business and Innovation at the Faculty
of Economics, University of Porto. He is currently a PhD student at the
faculty of applied economics Catholic University of Leuven.
Oscar AFONSO (44) has obtained MA and PhD degrees in Economics from
the University of Porto. He is Assistant Professor at Faculty of
Economics, University of Porto, and researcher at CEFUP (Center in
Economics and Finance), OBEGEF (Observatory in Economics and Management
of Fraud) and NIFIP. He has published a book, book chapters and articles
in Acta Oeconomica, Advances in Management and Applied Economics,
Applied Economics, Applied Economics Letters, Economic Modelling,
Economics Letters, Ekonomiaz, Economics Research International, Energy
Journal, European Research Studies Journal, Intereconomics,
International Economic Journal, International Trade Journal, Japanese
Economic Review, Journal of Business Economics and Management, Journal
of International Trade and Economic Development, Macroeconomic Dynamics,
Manchester School, Metroeconomica, Open Business Journal and Review of
World Economics. He has been teaching Computational Economics (Doctoral
Program in Economics), Economic growth (Master and Doctoral Program in
Economics), Macroeconomics (Undergraduate Economics) and International
trade (Undergraduate Economics).
Sandra T. SILVA (39) is Assistant Professor at Faculty of
Economics, Porto University (FEP) and researcher at CEFUP (Center in
Economics and Finance). She holds a MA and a PhD, both in Economics,
from FEP. She has published a pedagogic book on macroeconomics, some
book chapters and articles related with evolutionary economics,
innovation and economic growth, published in international journals such
as Journal of Evolutionary Economics, Journal of Economic Interaction
and Coordination, Evolutionary and Institutional Economics Review,
Organisational Transformation and Social Change. Current research
interests are on evolutionary economics, innovation, economic growth,
poverty and inequality. She has been teaching several courses at FEP,
for example, Macroeconomics (Undergraduate Economics, Master in
Economics and Doctoral Programme in Economics), Economic Growth
(Undergraduate Economics), and Theories and Systems of Innovation
(Master in Innovation Economics and Management).
Table 1. Broad classification of the different
categories of goods
Category Description
SIC 0 Agriculture, fishing and forestry
SIC 1 Mining and construction
SIC 2 Light manufacturing
SIC 3 Heavy manufacturing
SIC 4 Infrastructure and communication
SIC 5 Retail and wholesale trade
SIC 6 Services (financial)
SIC 7 Services (personal and business)
SIC 8 Services (health, legal,
educational, cultural, social
and consulting)
SIC 9 Public administration
Table 2. Covariance matrix depicting the correlation between the
independent and dependent variables in the linear regression
Correlation Trade value Nr. of 2 sized
cycles
Trade value 1 0.0716
Nr. of 2 sized cycles 0.0716 1
Nr. of 3 sized cycles 0.1578 0.5087
Nr. of 4 sized cycles 0.1129 0.5135
Correlation Nr. of 2 sized Nr. of 2 sized
cycles cycles
Trade value 0.1578 0.1129
Nr. of 2 sized cycles 0.5087 0.5135
Nr. of 3 sized cycles 1 0.9267
Nr. of 4 sized cycles 0.9267 1
Table 3. Results of the linear regression with the value of the
trade flow as a dependent variable
Unstandardized Standardized
coefficients coefficients
Value Standard Value Standard
deviation deviation
Constant -30165 896 0 0
Nr of 2 sized cycles 2296 1384 0.00157 0.001014
Nr of 3 sized cycles 50955 313 0.377 0.0023
Nr of 4 sized cycles -238 2.33 -0.237 0.0023
Significance
Constant 0
Nr of 2 sized cycles 0.097
Nr of 3 sized cycles 0
Nr of 4 sized cycles 0
Table 4. Overview of regression results per goods category
Category Description Contents composite
variable
SIC 0 Agriculture, fishing 2&3 sized cycles
and forestry 4 sized cycles
SIC 1 Mining and All indicators
construction
SIC 2 Light manufacturing 2&3 sized cycles
4 sized cycles
SIC 3 Heavy manufacturing All indicators
SIC 4 Infrastructure and Value of cycles
communication Number of cycles
SIC 5 Retail and wholesale All indicators
trade
SIC 6 Services (financial) All indicators
SIC 7 Services (personal All indicators
and business)
SIC 8 Services (other) 2&3 sized cycles
4 sized cycles
SIC 9 Public administration All indicators
Category Relation to aggregated Typical times
GDP growth cale (s)
SIC 0 Non-significant positive n/a
Significant negative n/a
SIC 1 Significant negative 3-
SIC 2 Significant positive 5+
Significant negative n/a
SIC 3 Significant positive 7+
SIC 4 Significant negative 9-
Significant positive 9+
SIC 5 Non-significant positive n/a
SIC 6 Non-significant 2(+), 13(-)
SIC 7 Significant positive 2+
SIC 8 Significant positive 3+ and 9+
Significant negative 9-
SIC 9 Significant positive 7+
Notes: significance at the 0.05 level for a number of years.
Timescale indicates extremes in the relation of the variable
with aggregate GDP growth. + is a positive maximum and - is a
negative minimum. When no clear extreme was observed, this is
denoted by n/a.
