Download Analyzing Export Diversification: Macroeconomic Factors Influencing and more Lecture notes Business in PDF only on Docsity!
Final Version, January 2006
CAF RESEARCH PROGRAM ON DEVELOPMENT ISSUES
“Explaining Export Diversification:
An Empirical Analysis”(*)
Authors: Ricardo N. Bebczuk and N. Daniel Berrettoni
Institution: Department of Economics, Universidad Nacional de La Plata,
Argentina
(*) We gratefully acknowledge the financial support from the CAF. We thank Carlos D’Elía for his efficient research assistanship. We appreciate the helpful comments of Manuel Agosín, Luis Miguel Castilla, Osmel Manzano, Ramón Espinasa and Andrés Rodríguez-Clare, as well as seminar participants at the 2005 LACEA Meeting. The usual disclaimer applies. Comments welcome at [email protected] and [email protected].
Introduction
The secular concentration of Latin American exports in primary products has been highlighted as a major drawback for the development prospects of the region. Adverse and volatile terms of trade, slow productivity growth, and relatively low value added are some of the issues that have been raised against this primary product dependence since the seminal work of Prebisch (1959). In addition to reducing the dependence on fluctuating commodity prices, diversification into other sectors, especially those more intensive in technology, is prone to trigger knowledge spillovers from the exposure to international markets, management and marketing practices, and production processes.
This fascinating debate has anything but faded away since then, and has been picked up by the recent growth literature (see for instance Warner and Sachs (1995), Maloney (2002), and Lederman and Maloney (2003)). These and other papers have thoroughly studied whether export and factor endowment structures influence growth, and have presented case studies to understand why diversification strategies put in place in the last half century in our region have failed.
Though the relationship between concentration, volatility and low growth could be well founded, it does not mean that increasing levels of export diversification can guarantee by themselves higher levels of growth. Between the mid-1960s and the late 1990s, most Latin American countries diversified their export structure (see Table 1 below and World Bank (2002)), but yet they were unable to achieve considerable levels of GDP expansion.
In particular, some of the recent studies have challenged the “resource curse” view, as became known the negative relationship between natural resource abundance and growth. Based on a cross-country analysis, Ng (2005) argues that natural resource abundance is beneficial for economic development showing a positive relationiship between resource abundance and output level. Lederman and Maloney (2003) find evidence in cross-section that what reduces growth prospects is concentration in export revenues and that there is no evidence that natural resource abundance is detrimental to growth.
The cross-country exercise revolves around a Herfindahl index using export data disaggregated at 2-digit SITC into 69 sectors. The Herfindahl index equals the sum of the squared proportional exposures to each sector and hence has a maximum of 1 when the country is completely focused on one sector, so lower values of the index indicate more diversification. Information is available on 56 countries with annual data for 1962-
2002.^2
To motivate our subsequent investigation, Figure 1 displays the simple average of the Herfindahl index for the complete sample, where it is apparent that diversification has consistently going up until the mid-1990s. Breaking down the sample by region, such downward trend has taken place in all cases but the European Union and North America. These two regions appear to have Herfindahl indices well below the world average in 1970 as well as in 2000, without much variation over the three decades.
Figure 1
Herfindahl Index Sample Average 1962-
0.1 0
0.1 2
0.1 4
0.1 6
0.1 8
(^2) The countries in the sample, grouped by region, are: Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Paraguay, Peru, Uruguay and Venezuela (South America); Canada, Mexico and the United States (North America); Austria, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, Spain, Sweden, United Kingdom and Belgium (European Union); Turkey, Cyprus, Estonia, Hungary, Latvia, Lithuania, Malta, Poland and Slovakia (other European countries); China, Hong Kong, Indonesia, Japan, South Korea, Malaysia, Singapur, Thailand and Taiwan (East Asia); Algeria, Egypt, Kenya, Morocco, South Africa and Tunisia (Africa); Australia, India, New Zealand, Brunei, Costa Rica and Vietnam (other countries).
Table 1 Region No. of countries
Herfindahl Index in 1970
Herfindahl Index in 2000
Total Sample 56 0.203 0. South America 10 0.354 0. North America 3 0.073 0. European Union 14 0.076 0. Other European countries 9 0.163 0. East Asia^9 0.151^ 0. Africa 6 0.268 0. Other countries 5 0.365 0.
Other remarkable patterns emerge from the inspection of time series by region, as shown in Figures 2 to 8. Most prominently, we observe a convergence process whereby countries having disparate diversification indexes at the start of the sample period gradually began, with particular intensity since the 1970s and early 1980s, to attain similar values within regions (but not always between regions, as suggested by the last table). Most countries lie in the 0.05-0.15 range by the early 2000s. Even though there are outliers in all regions (Mexico, Venezuela, Indonesia, Malta, Algeria) and the downward dynamics is uneven across regions (for instance, NAFTA, East Asia and the smaller South American countries have more stable indexes across time than the other regions), by and large some commonality is apparent. We will claim later on that this trend is to a great extent explained by common domestic macroeconomic developments (economic growth, increased trade openness), but the international context also has some bearing (as captured by annual dummy variables).
Figure 5 Herfindahl Index: Rest of Europe, 1962-
0
1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Tur Est Hun Lat Slov Cyp Lit Pol Malt Figure 6 Herfindahl Index: Africa, 1962-
0
1
1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Arge Ken Moro Sud Egy Tun Figure 7 Herfindahl Index: Latin America, 1962- Big countries
0
1
1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Arg Bra Chi Col Mex Ven
Figure 8 Herfindahl Index in Latin America: 1962- Small Countries
0
1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1 998 2000 2002 Bol Ecu Par Per Uru
The paper is structured as follows: In Section 1 a theoretical background is advanced and applied to the econometric cross-section analysis. Section 2 presents some robustness checks. Section 3 is devoted to the recent Argentine case. Some conclusions and recommendations close.
1. Theoretical considerations and cross-country results
Although the benefits of a diversified export base has been well-established in the literature, there exists no unified theoretical framework to rely on when it comes to uncover the macroeconomic drivers of export diversification. Furthermore, trade research treats export diversification from a social, aggregate standpoint, while the decision to diversify is made by individual firms from the private sector (provided the government has no decisive influence on export markets, as it is the case in most countries). One intuitive, but utterly misleading, approach would be to extrapolate the principles of financial diversification to export diversification. In the former case, any given firm or individual can purchase at low cost financial assets to construct a portfolio with a better expected risk and/or return performance. Conversely, in the latter case, even though the firm can still have a belief about future international prices, exchange rates and other relevant parameters, the return to the new export project will depend in the end on the ability of the very firm to profitably produce the good, which in turn responds to both micro and macroeconomic conditions. Likewise, risk is heightened by
country’s comparative advantage and the exploitation of increasing returns to scale via export specialization. Per capita GDP is another variable a priori linked to diversification. Richer economies tend to be economically and institutionally more stable, and such environment mitigates the business risks perceived by domestic producers, thus making diversification less imperative. Nevertheless, as these economies are characterized by higher total factor productivity and a better business climate, entrepreneurs may find it more appealing to broaden their productive mix. The access to credit, the quality of infrastructure, the gross investment ratio and the level of foreign direct investment are indicators of macroeconomic efficiency and strength that may likely enhance the growth prospects of firms, even though the implications for export diversification remain an open theoretical question.
The impact of the observed export composition is examined by including the shares of fuel, manufactures and agricultural exports in total exports. The expected sign on the fuel exports share is positive, as fuel exporters enjoy substantial rents that potentially downplay the long-run benefits of export diversification –a syndrome sometimes referred to as Dutch disease.^4 Since the industrial sector is the one offering a wider menu of alternative products vis-à-vis other sectors, a negative sign is expected for the manufactures exports share. A high reliance on agricultural exports may bring about either lower diversification –because of the rather limited possibilities of diversifying within this sector- or higher diversification –because agricultural exporters may be willing to diversify away the volatility and downward trend observed in the price of primary products-. Annual time effects are also included in the regressions to capture any time trend unrelated to economic fundamentals.
Table 2 displays the baseline regressions and our main findings. The usable sample comprises 1,180 observations covering the 1970-2002 period. Explanatory variables are one-year lag values under the sensible presumption of a delayed impact on diversification.^5 Fixed and random effects results are shown in the table.^6 Results do not
(^4) Besides this country-level factor, it is possible that trade surplus and the corresponding low real exchange rate reduce the productive incentives for non-oil exporters. 5 The use of lagged explanatory variables also helps to deal with their potential endogeneity provided that future values of the Herfindahl index have no influence whatsoever on the control set. While there are no strong reasons to believe on reverse or simultaneous causation, the use of proper instruments is always advisable to avoid any suspicion of endogeneity, but this approach is extremely difficult, if not impossible, to implement in most econometric applications.
change much from one method to the other, but a Hausman test suggests that the fixed effects estimator is preferred on consistency grounds. Goodness of fit, as reflected in the R Squared and the Wald and F Statistics of joint statistical significance, is quite satisfactory. An overall reading of the evidence supports the notion that richer, more efficient, more stable and more open countries tend to focus rather than to diversify exports. As revealed by the econometric outcome, variables typically associated with good macroeconomic performance (exports, GDP, investment rate, credit, infrastructure) have positive and highly significant loadings, meaning that they act in favor of more concentration rather than more diversification. This finding, yet quite robust, clashes against the common knowledge that export concentration should attenuate over the process of economic development. This belief comes from the normative observation that concentration in primary products has pervasive effects on trade and economic growth, while the results of this paper have more to do with private business incentives, which seem to encourage domestic firms to focus in order to take advantage of specialization-based economies of scale and, conversely, to diversify more intensely when macroeconomic risks are more evident. Along with this, manufactures exporters appear to be more diversified, in marked contrast to fuel exports. Time dummies (not reported) confirm that diversification has been on a systemic rise since the 1970s, regardless of idiosyncratic national factors. These dummy variables help reconciling the summary statistics in the Introduction –which showed a generalized move towards diversification- with the macroeconomic estimates. At first sight, the increase in diversification seems at odds with the focus-promoting effect of total exports and GDP, two variables that have gone up over the last decades. However, the advantage of multivariate regressions is that it isolates the independent effect of different variables. In the present case, the global bias toward diversified exports appears to be explained by global rather than by national factors.
(^6) The inclusion of the lagged Herfindahl index may be warranted on inertia considerations, in which case the presence of country fixed effects would render the employed methods inconsistent, forcing to resort to GMM-based or related estimators. However, autocorrelation is sometimes merely the consequence of omitting the underlying fundamentals. To check this, the inclusion of the lagged Herfindahl index in several (unreported) random and fixed effects, as well as GMM, regressions, yielded a significant estimate close to 1, but turned all the control variables not statistically significant. This suggests a serious multicolinearity drawback, explained by the likely empirical relationship of the lagged Herfindahl index with the lagged control set.
In the next table the sizable quantitative impact of the statistically significant variables under the fixed effects estimation is presented. A one-standard-deviation increase in the explanatory variables explain changes in the Herfindahl index that in no case fall below 16.1% from the average Herfindahl index of 0.158: Table 3 Explanatory Variable Average for the whole sample
Standard Deviation for the whole sample
Associated change in the Herfindahl index
Percentage change from the average Herfindahl index
Gross Fixed Capital to GDP
Per Capita GDP 8975 7314 0.019 30.5% Telephone lines (per 1,000 people)
Manufactures exports to total exports
Fuel exports to total exports
Exports to GDP 29.4 21.1 0.031 48.4%
2. Robustness Checks
A number of additional specifications were carried out to evaluate the robustness of the results just displayed. Such tests proved to be quite reassuring of the above findings, even though some surprising estimates came out along the way. Regional dummies were included in the first column of Table 4 in a random effects regression taking as an arbitrary benchmark the European Union. Except for North America, regional dummies have a positive and significant explanatory power over and above country fundamentals. In particular, these dummies account for nothing less than 19.7 and 18.5 percentage points in the Herfindahl index for South American and African countries. In any case, the fact that the control set estimates show no noticeable changes and that the R Squared goes up only moderately reinforces the confidence in the empirical model. Also, unreported regressions alternatively excluding the European Union and North America and including only these countries show that total exports and per capita GDP remain highly significant, although the latter lose significance when entered jointly with exports. In Column (2) the Manufactures export share is replaced by the Agricultural export share -they are not entered jointly due to multicolinearity. The observed positive
coefficient is in line with the hypothesis that this kind of exports is less prone to diversification.
The last regression drops the countries in the top quartile according to per capita GDP and adds the squared GDP to assess the presence of non-linearities. From the estimated coefficients, it seems that diversification increases at low income levels (up to $11, based on the regression) and that concentration is a prevalent feature in richer countries. The increasing complexity of the economic structure as the country goes from an agriculture-intensive production to a more balanced one is probably behind this temporal dynamics.^7 But this argument tacitly implies a monotonic relationship rather that the U-shaped pattern unveiled by the data. Consistent with the discussion of Section 1, a suitable rationale needs to take into account that there is a broad array of pro- and against-diversification factors, and that incentives in one or the other direction supposedly change over the development process. The present evidence points to the preference for diversifying away productive risks in low-income (and likely high- volatility) countries and for the exploitation of scale economies in rich and stable nations.
Quite interestingly, this finding coincides with independent work by Imbs and Wacziarg (2003), who characterizes sectoral diversification as per capita GDP grows. Using employment and value-added measures for 99 countries over 1969-1997, they find that productivity diversification rises at low GDP levels and declines later on.^8
Other unreported regressions were run. In order to pick up any possible exchange rate effect, we loosely calculated a proxy for the real exchange rate (the nominal exchange rate to the consumer price index), but it was not significant. Additionally, we rerun the baseline regressions using 5-year averages instead of yearly data. This time frequency could be helpful in case that fundamentals have a greater delayed impact on diversification than the one assumed under the original, annual database. Even though some coefficients do change, the total exports and GDP effects remain strong. Finally,
(^7) By the way, the below average Herfindahl index of European and North American countries at the beginning of the sample period can be understood from their relatively high GDP levels at the time. 8 Their interpretation goes along similar lines to that of Section 1, but they add another incentive for diversification: in the presence of high trading costs and consumer preference for product variety, countries will broaden their productive menu as long as income and demand grows.
Table 4 Regional Dummies
Agricultural Exports
Non linearity for GDP () Explanatory Variables Trade Variables: Exports to GDP 0. (6.22)**
Manufactures exports to total exports -0. (-2.96)***
Fuel exports to total exports 0. (12.0)***
Agricultural Exports to Total Exports 0. (4.28)*** Other Macroeconomic Variables: Per Capita GDP 0. (3.57)***
(Per Capita GDP)^2 0. (2.96)** Gross Fixed Capital to GDP 0. (4.68)***
Credit to the Private Sector to GDP 0. (1.19)
Telephone lines (per 1,000 people) 0. (4.0)***
Net Foreign Direct Investment to GDP
South America Dummy 0. (7.48)*** North America Dummy 0. (0.91) Non EU European countries Dummy 0. (2.06)** Oriental Asia Dummy 0. (2.2)** Africa Dummy 0. (6.22)*** Other Countries Dummy 0. (2.18)** Method Random Effects Fixed Effects Fixed Effects No. Observations 1180 1170 879 Year dummies Yes Yes Yes Wald (FE)/F (RE) Statistic (p-value) 953.17 (0.000) 23.57 (0.000) 18.25 (0.000) Adjusted R Squared 0.664 0.1602 0. ***Significant at 1%, **Significant at 5%, Significant at 10% () For GDP<$12,860 (fourth quartile excluded)
3. Some data and reflections on Argentina
The purpose of this section is to study export concentration approaching the recent Argentine experience. Given the sharp devaluation of early 2002, the crisis represents an interesting natural experiment to assess changes in export patterns following macroeconomic adjustment^10. More specifically, we try to find out if the change in the economic situation after the steep devaluation of early 2002 constitutes a new environment that favors export diversification among Argentine firms and, especially, if this diversification process occurs across or within firms. This is a relevant issue for economic policy because if export diversification occurs across firms, it is necessary new firms to enter the export business; conversely, if export diversification is a within- firm phenomenon, eventual supporting policies should be targeted towards existing exporting firms.
As mentioned in Section 1, from a theoretical point of view, it is not clear whether firms should diversify or concentrate their export supply. That is why this question must be addressed on the basis of empirical work. To undertake the analysis two main databases of Argentina are used: The figures on aggregated exports are taken from INDEC^11 , while the firm-level exports come from DGA.^12 The latter records the export value for all exporting firms. For the purpose of the present analysis, six-digit level was used for 1998, 2001 (pre-devaluation) and 2004. 13
The first part of this section describes changes in the real exchange rate and export concentration in Argentina after the crisis and the second part discusses export diversification at the firm level.
(^10) The descriptive analysis carried on here covers a short time period for a specific country case, so it cannot be strictly compared with the long run cross-country econometric analysis of the previous section. 11 12 Instituto Nacional de Estadísticas y Censos (National Institute of Statistics and Census). 13 Dirección General de Aduana (General Direction of Customs). The disaggregation level is six and eight-digit Harmonized System, respectively, for 2001 and 2004. So, in order to make possible the comparison, a six-digit (that is, subheading) level was used for the analysis. This includes 4,200 products exported by Argentina in 2004.
the goods sold abroad increased by 30%, from US$ 26,600 millions to US$ 34, millions15,16.
Digging into diversification trends, an interesting but still unexplored issue is to what extent the export expansion is explained by an increase in the volume of the pre-crisis exporting basket and/or by new goods within a more diversified export supply. Table 5 shows that, according to data at a six-digit Harmonized System level, export concentration slightly increased between 2001 (pre-devaluation) and 2004 (pos- devaluation), with the Herfindahl index going up by just 3%. However, the aggregate numbers hide sector-specific patterns, which actually show huge disparities. In fact, only Industrial Manufactures and especially Fuels and Energy tended to diversify their export supply.
Table 5 Herfindahl Index in Argentina, 1990- By Type of Good
Sector/Period 1991-2000 2001 2004 Primary Products 0.108 0.125 0. Agri-manufactures 0.088 0.133 0. Industrial Manufactures 0.017 0.014 0. Fuels and Energy 0.476 0.331 0. Total 0.025 0.029 0. Source: Own elaboration based on INDEC database
To try to understand the impact of the crisis on export diversification at the level of firms, two kinds of questions are addressed in what follows: (a) Has export diversification patterns change after the 2002 crisis among the top exporters in business both in 2001 and 2004? These firms represent about 82% of total Argentine exports in 2004; and (b) Has export diversification occur across firms or within firms?
(^15) During the same period, quantities exported increased by 13%. (^16) It is worth to mention that in 2002 the Argentine Government introduced changes in export taxes. The average level for these type of tax is 5,3%. The tax for Primary products (8,8%) is greater than for Agri- manufactures (5,8%) and Industrial Manufactures (5,1%).
The next step is to analyze within-firm and across-fim diversification for the top exporters who exported in each and every of the following years: 1998, 2001 and
- In order to measure within-firm diversification, a Herfindahl index for each of the firms of the sample is calculated for the three years. To get an aggregate idea of diversification, both a simple and a weighted (by export volume) average are computed. A similar procedure was employed to measure across-firm diversification.
Table 6 Diversification for top 500 exporters in Argentina Herfindahl Index
Simple average Weighted average Total Across firms Within firms Across firms Within firms 1998 0,032 0,802 0,601 0,394 0, 2001 0,035 0,795 0,606 0,409 0, 2004 0,048 0,783 0,576 0,355 0, Source: Own elaboration based on DGA database
As Table 6 shows, across and within-firm Herfindahls went down between 2001 and 2004, both for simple and weighted averages. This means that, after the 2002 devaluation, product supply became more diversified among the top 500 exporting firms.^18
Nevertheless, the first column in Table 6 indicates that, taking the top 500 exporters as a whole, the level of concentration increased between 2001 and 2004, thus confirming the previous finding on aggregate exports. In view of the claims from the previous paragraph, this finding might be puzzling. However, a plausible explanation is that devaluation makes profitable for a wide set of firms to sell abroad products that, before the change in relative prices, could only be exported by a few firms that were able to compete in international markets. In that way, firms diversify their export menu (within- (^17) The top 500 exporting firms in 2004 that also exported in 2001 account for 442 firms in 1998. (^18) The inter-year differences are statistically significant at 5% according to their respective mean difference tests.
