Journal of Economic Geography

Journal of Economic Geography Advance Access published online on April 27, 2009
Journal of Economic Geography, doi:10.1093/jeg/lbp015

This Article Abstract FREE Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Request Permissions Google Scholar Articles by Boulhol, H. Articles by de Serres, A. Search for Related Content Social Bookmarking          What's this?

© The Author (2009). Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org

Have developed countries escaped the curse of distance?

Hervé Boulhol and Alain de Serres

OECD, Economics Department, 2, rue André Pascal, 75775 Paris Cedex 16, France. email <herve.boulhol@oecd.org>

Date submitted: 25 April 2008    Date accepted: 18 March 2009

JEL classifications: F12, F15, R11, R12

	Abstract

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 
This article applies for the first time the framework developed by Redding and Venables (2004, Journal of International Economics, 62: 53–82) on a panel dataset restricted to advanced countries over 1970–2004, and shows that the cost of remoteness remains significant. Second, the article highlights that the elasticity of aggregate income to distance to markets in the Redding–Venables model is severely biased upwards in cross-section samples that mix both developing and developed countries, most likely due to the inability to adequately control for heterogeneity in technology levels across countries. Also, the effect of distance is robust to whether the trade equation is specified as linear in logarithm or nonlinear in level.

Keywords: economic geography, market access, distance

	1. Introduction

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 
Everyone who has seen satellite pictures of the earth at night has undoubtedly been struck by the clustering of economic activities as depicted by the concentration of city lighting across the globe. Such representations powerfully display agglomeration in a few locations and prompt the observer to wonder whether being surrounded by neighbours matter for a country's wealth. To what extent are countries remote from centres of economic activity hindered in their development process?

The impact of distance on income levels can materialize through various channels including trade, foreign investment, knowledge spillovers and technology diffusion, all of which are hampered by remoteness. Distance directly raises transport costs and thereby reduces trade in much the same way as a tax on exports or a tariff on imports, although without the benefits of tax receipts. By segmenting markets, distance also limits the extent to which domestic firms can operate on an efficient scale and, more generally, exploit increasing returns to scale. Also, by providing a natural shelter from foreign competition, the pressure on domestic companies to be efficient and innovate is weakened. Furthermore, the clustering of innovative activities suggests that there are extra benefits associated with such concentration.

There is widespread evidence that a better access to markets contributes to raising income levels. The model developed by Redding and Venables (2004) has led to a workhorse methodology to assess the impact of proximity to markets on income levels. It has been tested in different contexts and all of the studies find a strong relationship. Redding and Venables apply their framework to a cross-country sample of 101 countries, while Breinlich (2006), highlighting that regional income levels in the European Union display a strong core-periphery gradient, tests the impact of market access using a panel of European regions over 1975–1997. Head and Mayer (2006) conduct a similar exercise based on European sectoral data over a shorter period.¹

The current article brings three contributions. First, it investigates for the first time whether proximity to markets is a significant determinant of GDP per capita in a panel covering only developed countries. Access to markets varies widely between Australia and New Zealand, Japan, North America and Europe. Even within the European Union, centrality to markets is very different between the core and periphery countries. Yet, in a broad sample covering both least and most developed countries, Australia and New Zealand come out as outliers in the relationship between distance to markets and GDP per capita, suggesting that they have overcome the tyranny of distance (Dolman et al., 2007). In contrast, according to the results presented in this article, remoteness from markets relative to the average developed country might contribute negatively to GDP per capita by as much as 12% in Australia and New Zealand. Conversely, the benefit from a favourable location might be as high as 6% of GDP in the case of Belgium and the Netherlands.

Second, the article elucidates why the cross-section sample used by Redding and Venables leads to the false impression that developed countries have escaped the curse of distance. The main reason is straightforward: a cross-section that mixes both low- and high-income countries cannot control satisfactorily for the wide differences in the level of technical efficiency, which drastically biases upwards the sensitivity of GDP per capita to proximity to markets. In addition, the empirical analysis based on a larger panel of countries suggests that the impact of access to markets might be twice as large for developing as for developed countries. The bottom line is that focusing on a more homogenous group leads to more reliable estimates and, as the sample used in this study covers a large period between 1970 and 2004, the panel dimension can be used to control for national idiosyncrasies.

Third, the article focuses on the order of magnitude of the sensitivity of GDP per capita to distance to markets, which depends both on the elasticity of GDP to market and supplier access and on that of market and supplier access to distance. This overall sensitivity is estimated to have been roughly stable since 1970 at around –0.10, rather than –0.70 implied in Redding and Venables: increasing distances to all markets by 10% would reduce GDP by 1%. This 1–7 difference in estimates is explained as follows. Taking advantage of panel versus cross-country data reduces the estimated elasticity by a factor of 2.5; it is further divided by 2 when the sample is restricted to developed countries, as the elasticity seems higher for developing countries; the inclusion of additional control variables such as physical and human capital accumulation reduces it by a factor of 1.4. Moreover, the estimated elasticity of GDP to distance is robust to whether the trade equation is specified as linear in logarithm or nonlinear in level, even though, consistent with a recent literature, the estimated distance elasticity of trade is sensitive to the specification of the gravity equation that is needed to compute access to markets. In short, the answer to the question raised in the title is: no, but the curse of distance is much lower than previously estimated.

The remainder of the article is organized as follows. The next section presents the theoretical background linking access to markets and income, and provides the intuition for why the choice of the sample might matter. Section 3 focuses on the construction of the access to markets indicators, based on a trade equation specified in log-linear or nonlinear level form, while Section 4 assesses their impact on GDP per capita. Finally, Section 5 concludes.

	2. Market and supplier access

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 
2.1. Theoretical background
Measuring access to markets is not a straightforward exercise. Market potential is a common measure of proximity to markets, which dates back to Harris (1954), and is defined as the sum of all countries’ GDP weighted by the inverse of the bilateral distance. Although it is an intuitive indicator of centrality, market potential is an ad hoc way of capturing the influence of distance to markets. In particular, the weighting of markets in the market potential computation is based solely on distances, regardless of the true accessibility of these markets. Being closer to a large market is more beneficial more the country is open. In that respect, market potential is a very crude measure of market access. A better approach consists in looking not only at the potential, but also rather at the actual accessibility to countries’ markets.

The new economic geography literature has revived the concept of proximity to markets and formalized the role of economic geography in determining income levels. Starting from a framework developed by Fujita et al. (1999), Redding and Venables (2004) have introduced two dimensions of proximity: market access refers to the access to final customers, while supplier access refers to the access of producers to intermediate goods. They show how better market and supplier access could raise the prices of the internationally immobile factors. The intuition for this result is the following and Appendix A provides the details of this theoretical framework.

Based on a model with fixed costs of production and free entry under monopolistic competition, a firm which faces a large demand for its products is able to charge a higher price. Consequently, in equilibrium, this tends to boost labour demand, inducing firms to pay higher wages. This is the market (domestic and foreign) access effect. In addition, due to increased competition, prices of intermediates are lower in large markets. Firms having a favourable access to suppliers benefit from lower costs of inputs, which also leads to higher wages in equilibrium. This is the supplier (domestic and foreign) access effect. As summarized by Redding and Venables, ‘transport costs or other barriers to trade mean that more distant countries suffer a market access penalty on their sales and also face additional costs on imported inputs. As a consequence, firms in these countries can only afford to pay relatively low wages – even if, for example, their technologies are the same as those elsewhere’.

The additional important insight of the Redding and Venables model consists of using trade data to reveal both observed and unobserved determinants of market and supplier access, MA and SA, respectively. As discussed below, MA and SA are constructed using the estimates of a bilateral trade equation based on a gravity-like specification which relates bilateral trade flows to trade partners’ characteristics and trade costs.

The effects of market and supplier access on wages, w, are captured in the following equation:

(2.1)

where and β are the intermediate input and labour shares in gross output, respectively, > 1, is the elasticity of substitution between the different varieties and a_it is the level of technical efficiency. Note that the impact of supplier access is directly related to the share of intermediates. Also, wages are more sensitive to access conditions when the labour share is low because, in that case, wages react more to changes in prices.

2.2. Importance of the sample
The level of technical efficiency, a_it, is a critical component in the relationship linking the access variables to income levels, i.e. equation (2.1). Since it is not directly observable, failure to adequately control for technical efficiency might bias the key parameters of interests. This issue is similar to the one highlighted by Hall and Jones (1999), among others, in the growth literature, since differences in technology are generally treated as residuals giving rise to omitted variable bias. The resulting bias is likely to be severe in cross-section analysis, while panel data enables to introduce the whole battery of country and year fixed effects, as well as country-specific time trends.

In order to illustrate this issue, Figure 1 reproduces Figure 3 from Redding and Venables (2004) and adds three trend lines.² The log of market access appears on the x-axis, while the log of GDP per capita is represented on the y-axis, both in 1994. The steepest trend line captures the relation identified by Redding and Venables. The slope is 0.51 and, with the best efforts to add control variables in order to account for technology and other fundamental determinants of income levels (primary resources, tropical area, institutions, etc.), Redding and Venables reach a reduced slope of around 0.30. The middle trend line is obtained, using the same data, for the cross-section limited to OECD countries. The slope is reduced from 0.51 to 0.09. When the cross-section is further restricted to the high-income OECD countries—excluding Eastern Europe members, Korea, Mexico and Turkey—in order to focus on a more homogenous group of countries, the slope falls to 0.05, but remains significantly different from 0.

View larger version (31K): [in this window] [in a new window] [Download PowerPoint slide]   Figure 1. GDP per capita and market access, Redding and Venables (2004). Data are from Figure 3 in Redding and Venables (2004). Only fitted trend lines are added.

 
Clearly, an estimated slope of 0.30 would imply an unrealistic impact of remoteness for countries, such as Australia and New Zealand, as well as for Belgium and the Netherlands. It is argued in Section 4, where the impact is quantified, that the flattest curve provides more realistic estimates, which has two implications. On the one hand, there is strong qualitative support for the relationships identified by Redding and Venables, even though the importance of paying a closer attention to heterogeneity is underscored. On the other hand, there is collateral damage: the flatter and more reliable line cannot be easily reconciled with the structural parameters of the model, suggesting the need to amend it somehow so as to better match the data. This is beyond the scope of this article.

	3. Construction of the access indicators

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 
Market and supplier access measures are derived from the estimation of a gravity-like relationship, and two methods have been used to construct the access indicators. Section 3.1 is based on a log-linear specification of trade flows, while Section 3.2 discusses the estimation of the trade equation in level using the pseudo-maximum-likelihood estimator proposed by Santos Silva and Tenreyro (2006).³ Section 3.3 provides the results from both methods.

3.1. Log-linear gravity equation
Gravity-like trade equations have traditionally been estimated from log-linear specifications. As is common in the literature, trade costs, , are assumed to depend on three variables: bilateral distance, common border and common language. Noting X_ij,t as the export from country i to country j in year t and d_ij the bilateral distance between the two countries, the model leads to the following trade equation, which is estimated for each year between 1970 and 2005 using the OECD International Trade by Commodity Statistics database covering 98.5% of world goods trade flows:

(3.1a)

where trade costs enter into the gravity equation via the freeness of trade, the so-called phi-ness, , which is inversely related to trade costs:⁴

(3.1b)

The market capacity, m_jt, and the supplier capacity, s_j, are unobservable characteristics, even though they can be given a theoretical interpretation, as explained in Appendix A. In practice, they are proxied, for each year, by importer and exporter country fixed effects. Although the specification (3.1a) focuses on international trade flows, domestic ‘phi-ness’, _iit, is needed to compute the domestic component of market and supplier access (see below). As in Redding and Venables, ‘intra-country’ phi-ness is computed based on (3.1b) applied to internal distance, common border and common language.

The main distance measure in this study, distcap, commonly combines geodesic capital-to-capital distances between countries and internal distances based on surface areas, i.e. . Robustness of the results was checked using the four distance indicators in the economic geography database of CEPII (Centre d’études prospectives et d'informations internationales). In particular, results are shown for the alternative measure based on city-level data, distcities. More precisely with distcities, both internal distances and distances between countries are based on bilateral distance between the largest cities of each country, weighted by their share in the overall country's population.⁵

Heteroskedasticity in trade-level equations leads to inconsistent estimates in the log specification, as shown by Santos Silva and Tenreyro (2006). Two solutions are proposed to address this issue. First, in order to reduce heterogeneity in trade levels while sticking to the log specification (equation 3.1), the world was split into the 32 following areas, thereby grouping countries that bear a low weight in trade flows: Africa, Australia, Austria, Belgium, Brazil, Canada, China, CIS countries, Denmark, Eastern Europe, Finland, France, Germany, Greece, Ireland, Italy, Japan, Korea, Latin America (other than Brazil and Mexico), Mexico, the Middle East, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, Turkey, UK, USA and Asia (other than the countries already included).⁶ This solution is consistent with the focus placed in this study on developed countries, since having many low-weight countries in the sample might distort the relevant estimates. Second, in case heteroskedasticity remains problematic, the nonlinear estimator proposed by Santos Silva and Tenreyro is applied to a large sample made of about 180 trade partners, as discussed in the following section. Market and supplier access are then constructed from the estimated parameters of the bilateral equation according to:

(3.2)

For all the countries, market access (supplier access, respectively) is computed as a weighted sum of unobserved importer characteristics m_kt (exporter characteristics s_kt, respectively) of all countries. Only the weights put on each partner change across countries, with these weights being a function of the estimated trade costs. If a given country k has a large market capacity m_k, countries having low trade costs with country k, i.e. a high freeness of trade, are associated with a high weight on m_k and tend to have a high market access. A similar argument applies to supplier access with respect to supplier capacity. Note that the same principle applies to market potential, which weights all countries’ GDP by the inverse of the bilateral distances.

In addition, it should be emphasized that market and supplier access, like market potential, are measures of market density as much as market size. The weighting of the domestic market is related to internal trade costs which are greater for a large and sparsely populated country than for a small and densely populated one. In the case of market potential, for example, the domestic market (GDP) is divided by the internal distance, itself an indicator of surface area. Furthermore, if a large autarkic country is split into two (or more) symmetric regions opened to each other, then each region would have a smaller domestic market access, but the same total market access as the whole country, provided that the overall phi-ness is measured correctly. Indeed, in the case of two regions, 1 and 2:

(3.3a)

(3.3b)

3.2. Poisson pseudo-maximum likelihood estimator
Santos Silva and Tenreyro (2006) show that heteroskedasticity in the trade-level equation might generate biases in the log-linearized specification (equation 3.1a). When the equation is specified in levels, it is very likely that the variance of the residuals depend on the explanatory variables such as importer and exporter characteristics (that cover observed ones like GDP) and distance to markets, especially when the sample is very heterogeneous in terms of development levels and trade flows. In that case, this heteroskedasticity generates biases in the log-linear specification because, consistent with Jensen's inequality [E (ln y) ln E (y)], the expected value of the logarithm of the residuals depends on the higher moments of the residuals, including the variance. Formally:

(3.4)

(3.5)

where the Zexplanatory variables include, for each year, importer and exporter fixed effects, (log of) bilateral distances and other control variables influencing the phi-ness of trade.

A natural way to overcome this problem consists in estimating the level equation in (3.4) using a nonlinear estimator. Santos Silva and Tenreyro propose the Poisson Pseudo-Maximum Likelihood estimator (PPML), assuming that the variance of X_ij is proportional to its expectancy, which is likely to make this estimator more efficient than the simple nonlinear least squares. Indeed, it is unrealistic to assume that the variance of estimated trade flows is the same for small/remote and large/central countries, as implicit in the nonlinear least squares specification. In addition, a level specification allows the inclusion of zero trade flows, even though Santos Silva and Tenreyro show ex post that this does not make a material difference. Data for the estimation of the level specification in (3.4) by PPML is the IMF Direction of Trade Statistics database covering about 180 trade partners.⁷ Market and supplier access are then computed from the estimates of the fixed-effect parameters and other β-parameters according to equation (3.2).

3.3. Results
The most important and significant parameter in estimating (3.1a and b) or (3.4) is that related to distance, which is shown in Figure 2. It varies from –0.81 in 1970 to –0.99 in 2005 in the log-linear specification (Table A1, upper panel shows the estimates of the gravity equations), implying that an increase of 10% in the distance triggers a decrease of 9% in trade flows on average, consistent with the order of magnitude found in the literature.⁸ Indeed, the meta-analysis of 103 papers carried out by Disdier and Head (2008) indicates that the elasticity of trade to distance is around –0.9 and that trade decreases with distance by roughly the same amount today than 30 years ago, with an increase in the impact of distance since the late 1980s that is consistent with the findings reported in Figure 2.

View larger version (29K): [in this window] [in a new window] [Download PowerPoint slide]   Figure 2. Elasticity of trade flows to distance (absolute value). Elasticity of trade flows to distance is estimated through the specification in equations (3.1a and b) and (3.4).

 
The elasticity of trade to distance is an order of magnitude lower in absolute terms when estimated in the nonlinear specification using the PPML estimator (Table A1, lower panel). The average estimated elasticity is around –0.70 (the average standard error equals 0.03) and fluctuates much less between 1970 and 1990 than its log-linear counterpart (Figure 2B). From the end of the 1980s, the elasticity is estimated to have steadily increased in absolute term from –0.63 in 1988 to –0.75 in 2005 (see Bosquet and Boulhol, 2009, for a detailed analysis of the ‘distance puzzle’ using various estimators). By comparison, Santos Silva and Tenreyro report an elasticity of –0.75 in 1990 based on the same methodology.

Table 1 reports the computed values for the 2005 access variables from the log-linear specification, as well as market potential for comparison. For presentational purposes, all indicators are scaled such that the average across countries in the table is 100 in each year.⁹ The range extends from around 25 for Australia, New Zealand and Brazil to above 200 for the very open core European economies of Belgium and the Netherlands. Unsurprisingly, the share of the domestic market in the total measures is the highest for Japan, which is both very densely populated and a large market, and for the USA, at around three-quarters and two-thirds, respectively. That share is about one-half for Germany, and one-tenth for Northern European countries.

View this table: [in this window] [in a new window]   Table 1. Domestic and foreign components of market access, supplier access and market potential, 2005a

 
An alternative measure consists in using the distance indicator, distcities. This measure entails some differences depending on the size of the countries (Table 2). In particular, the fact that Canada appears to have a larger access than the USA is entirely due to the capital-to-capital measure, as the capital-to-capital distance is relatively low between the two countries, leading to an overestimated access of Canada to the US market.¹⁰

View this table: [in this window] [in a new window]   Table 2. Market and supplier access, alternative distance measure, 2005a

 
Differences in access to markets across countries are very persistent as shown in Figure 3A, which presents market access estimates in 1970 and 2005. The main changes relate to the market access gain of China and Korea. In addition, Spain and Portugal record significant improvements in market access due to their better integration in the European Union, while Canada and Mexico have benefited from both NAFTA and the dynamic US market. In contrast, Switzerland, Australia and most of Northern European countries have seen their relative position deteriorate by 15–20%, which can be measured as the vertical distance to the diagonal.¹¹

View larger version (19K): [in this window] [in a new window] [Download PowerPoint slide]   Figure 3. Market access, logarithm. (A) Average across countries = 100 for each year, 2005 versus 1970, log-linear specification [The relative change between 1970 and 2005 compared to the average country is read from the graph as the vertical distance to the diagonal (as a first-order approximation). For example, for Australia, the log of market access was 3.40 in 1970 and 3.22 in 2005. The vertical distance is therefore –18%. The exact change is exp (–0.18) – 1 = –16.5%.]. (B) Log-linear versus PPML specification, 2005.

 
Figure 3B compares (log of) market access measures in 2005 obtained from the log-linear and PPML estimators. The two measures are strongly correlated with a linear correlation coefficient of 0.97 across countries. However, there is one important result to keep in mind for the remaining of this study: regressing the log of ‘PPML’ market access on the log of ‘log-linear least-squares’ market access gives a coefficient of 0.71, which of course reflects primarily the ratio of the estimated elasticity of trade to distance between the two estimators. For example, based on the PPML indicator, Australia has a market access in 2005 that is equal to 37% of the country average, whereas it is only 25% with the log-linear indicator, in which distance has a greater impact.

	4. Impact on GDP per capita

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 
This section analyses the sensitivity of GDP per capita to market and supplier access in OECD countries, starting in Section 4.1 by the access indicators based on the log-linear trade equation. The following section analyses whether the PPML specification of trade flows, the inclusion of developing and least-developed countries and of other control variables affect the results. Finally, Section 4.3 quantifies for each country the effect of distance to markets on GDP per capita relative to the average OECD country.

4.1. Impact on OECD countries based on the log-linear trade equation
The impact of access variables on GDP per capita is first estimated using a sample of 21 OECD countries over 1970–2004.¹² The empirical specification takes advantage of the panel dimension and include various combinations of country and year fixed effects and country-specific time trends in order to control for the level of technical efficiency (log a_it in equation 2.1). In addition, estimates based on instrumental variables are reported in order to address the potential endogeneity of the access variables. As in Redding and Venables (2004), GDP per capita is taken as a proxy for the price of the immobile factor, w:

(4.1)

Table 3 reports a first series of results when equation (4.1) is estimated in levels (first three columns) or in first-differences which wipe out the country fixed effects (last three columns). Comparing the estimates between these two panels suggest that auto-correlation in residuals might be a serious issue, while comparing them within panels reveals some potential collinearity issues between the two access measures. Faced with the latter problem, Redding and Venables (2004) use the structural parameters of the model to constrain the parameters related to the two access variables. In the expression in brackets in equation (2.1), the weight of supplier access is /( – 1). Assuming a value for the share of intermediates, , of between 0.5 and 0.6, and a value for the elasticity of substitution, , of between 6 and 10, as is generally estimated, that weight varies between 0.55 and 0.72. The results hereafter were robust to any specific choice within that range and the reported results use the set that is preferred by Redding and Venables, i.e. = 0.5 and =

View this table: [in this window] [in a new window]   Table 3. GDP per capita and economic geography: market and supplier access separately

 
The model gives some order of magnitude of the expected impacts. According to (2.1), the parameter for the weighted sum of the market and supplier access is 1/(β ). Taking the share of labour in output β between a third and a half, the 6–10 range for the elasticity of substitution implies an expected parameter between 0.2 and 0.5. For example, an estimate of 0.2 means that if distances to markets were reduced by half for a given country, market and supplier access would both increase by roughly 0.9 * ln(2) = 62%, based on a distance elasticity of trade of –0.9, and GDP per capita would increase by 0.2 * (1 + 0.6) * 0.62 20%. Redding and Venables (2004) find an estimate between 0.30 and 0.50 (but based on a distance elasticity of trade of –1.7, more on this below), while Breinlich (2006) reports an estimate for market access of 0.25. After controlling for human capital stock across regions, Head and Mayer (2006) find a parameter of 0.10 (0.11 without controlling), and from a different theoretical model, Hanson (2005) finds 0.25 (0.35 without controlling).

The estimates for the constrained model are presented in Table 4. The fixed-effect estimator produces an estimate of 0.23 for the access variable.¹³ Since the explanatory variables are meant to capture market sizes, there is a well-acknowledged endogeneity issue, which is most obvious with Harris’ market potential. To take this into account, Redding and Venables lag the right-hand side term. When the access variable is lagged three times, the parameter, reported in the second column, is still very significant, albeit roughly halved. A better treatment of this issue is presented below.

View this table: [in this window] [in a new window]   Table 4. GDP per capita and economic geography: weighted sum of market and supplier accessa

 
Another serious concern is related to serial correlation. Indeed, taking into account the first-order auto-correlation, the estimate (columns 3–4) falls to around 0.09. Excluding country fixed effects (column 5) gives very close results, suggesting that the access variables capture variations through time as well as across countries.¹⁴ Given the estimated level of the auto-correlation parameter, a reasonable alternative is indeed the first-difference specification. As indicated in columns (6–7), it produces an estimate of around 0.07.

As an attempt to overcome the potential endogeneity bias, previous studies have used the sum of the distances of each country to Tokyo, Brussels and New York as an instrument. However, the choice of these three locations is in itself endogenous. One could expect that a researcher doing the same exercise in 30 years would include Beijing, Sao Paolo and Moscow. An appealing instrument is to sum the distances to all the other countries (Head and Mayer, 2006). In order to take advantage of the panel dimension of the data, the effect of this time-invariant instrument is allowed to vary through time. In other words, the proposed instruments are where the h_t are time dummies. The IV-estimates of 0.10 is still very significant.

Also, a few countries might well drive these results. As a robustness check, because Australia and New Zealand on the one hand, Belgium and the Netherlands on the other hand, hold a specific geographical position, the first difference specification was re-estimated excluding either of these two pairs and both of them. Instead of the 0.080 parameter obtained for the full sample, that exercise leads to significant estimates between 0.062 and 0.087, respectively.

Finally, the effects on GDP per capita are robust to the choice of the distance definition, i.e. distcities instead of distcap, essentially because differences between distance measures reflect mostly a level effect that are controlled for by country fixed effects.

4.2. PPML trade equation, wider sample and other control variables
The estimated parameter of market and supplier access based on the PPML trade specification is somewhat higher than its log-linear counterpart, reflecting the different sensitivity of trade to distance across estimators. The first three columns of Table 5 report the estimates using the PPML trade specification to construct the access variables limiting the GDP sample to the same 21 countries to distinguish the impact of the trade specification from that of the country sample. These columns correspond to the estimates in columns 3, 4 and 6 of Table 4 for the log-linear version, respectively. The market and supplier access parameter is estimated around 0.13 compared with 0.09 using the log-linear trade specification. This difference is consistent with the relation highlighted in Figure 3B as indeed 0.09/0.710.13, and Section 4.3 shows that the implied effect of the geographical position on GDP per capita across countries is not affected by the choice of the trade specification.

View this table: [in this window] [in a new window]   Table 5. GDP per capita and economic geography: weighted sum of market and supplier accessa

 
Broadening the sample to take into account developing and least-developed countries leads to a higher parameter (middle columns of Table 5).¹⁵ It is estimated at 0.35 when controlling for country and year fixed effects and at around 0.25 either with the same specification in first differences or when adding country-specific time trends, which might be important additional controls in such a heterogeneous panel of countries. Although the 0.25 mark is of a similar order of magnitude to that found by Redding and Venables from a cross-section analysis, this simple comparison is misleading.¹⁶ Indeed, the access variables constructed in Redding and Venables are based on an elasticity of trade to distance of –1.74 which is 2.5 times what is used here. That is, a similar parameter estimate means an implied effect on GDP per capita that is 2.5 times larger in cross-section than in panel regression.¹⁷

There are at least two explanations for the twice as large estimates between the world sample and the 21 developed countries one estimated in country-year panels (middle versus left part of Table 5). First, the whole battery of fixed effects and country-specific time trends might still be insufficient to control for differences in technology levels. And, to the extent that this uncontrolled component is correlated with the access variables, the main parameter of interest would be biased. This possibility seems especially relevant when the sample is very heterogeneous in terms of country's development levels. Second, access to markets might, however, be more influential for less developed countries. To test whether data support this conjecture, the access variable is interacted with a dummy taking the value of 0 for our 21 OECD countries and 1 otherwise. According to the estimates reported in the right part of Table 5, the parameter of access to markets is around 0.13 for the advanced countries, consistent with what found on the limited sample, and about twice as much for the developing countries. Disentangling between these two explanations is beyond the scope of this study.

One might be concerned that the above estimates might not be robust to the inclusion of the usual determinants of GDP per capita or to a better accounting of the dynamic adjustment towards the steady state. Therefore, an important robustness check consists in investigating whether the market and supplier access variables are still significant in an augmented-Solow framework. In a related research, Boulhol et al. (2008) show that the access variables parameter remains highly significant even though it is reduced by one-third when physical, human capital and population growth are included in the regression, as well as when accounting for the heterogeneity across countries of short-term coefficients. The first column of Table 6 reproduces their central estimate, obtained from the log-linear trade specification, where the access to markets parameter is of 0.056 compared with 0.094 without the augmented-Solow explanatory variables (third column of Table 4). Respectively, with PPML-based indicators, the parameter is 0.082 as reported in the second column, instead of 0.116 (first column of Table 5). This result suggests that the impact of centrality to markets acts for the most part on top of these usual determinants.

View this table: [in this window] [in a new window]   Table 6. GDP per capita and economic geography with Solow-augmented regressors

 
Even though the elasticity of trade to distance does not seem to have decreased over time in absolute terms, it is still conceivable that distance plays a lesser role today if the impact of access to markets has become less relevant. However, this hypothesis is not supported by the data. In the last two columns of Table 6, the access variable is interacted with a post-1987 dummy, and the reported interaction parameter is not significant.

4.3. Size of the effect on GDP per capita
To sum up, market and supplier access seem to have a very significant impact on GDP per capita; the estimates herein are below most of those obtained in previous studies, but in line with those of Head and Mayer (2006). Of course, the samples are not the same. However, the failure to take into account country's idiosyncrasies in cross-section analysis might be the primary source of the differences in key parameter estimates. Nevertheless, this lower order of magnitude implies a sizeable impact over the wide spectrum of developed countries. Based on the level of the access variables reported in Table 1 and a parameter of 0.056 with the full set of control variables (Table 6, column 1), distance to markets is estimated to penalize Australia and New Zealand by 12% of GDP compared to the average country in the sample, whereas, at the other extreme, Belgium and the Netherlands would benefit by 6–7%.

Even so, the effect of the geographical position is tremendously lower than that obtained based on cross-section analysis, such as in Redding and Venables and which has been replicated herein. Indeed, the elasticity of GDP per capita to distance to all markets, , is the product of the elasticity of GDP per capita to market and supplier access and of that of access measures to distance. Formally, combining equations (3.1b), (3.2) and (4.1):

(4.2)

where _M is the parameter reported for the weighted sum of market and supplier access (in the constrained specification _s = 0.6 _M). Table 7 reports what would be the impact for each country in terms of GDP of having the average OECD market and supplier access. Taking Australia as a striking example, GDP would be increased by 20% based on the estimation ignoring the usual Solow-augmented control variables (columns 1–2) and by 13% when taking them into account (columns 3–4, our baseline estimates). When pooling both developed and developing countries, an assumption that seems to be rejected by the data, the effect increases to 44% (column 5, without the Solow-augmented variables). Based on the cross-section analysis, using either our sample and specification (column 6) or those of Redding and Venables (column 7), Australia and New Zealand would triple their GDP by moving to the average country location! This is because the elasticity of GDP to distance to all markets is around –0.09 in our base case versus –0.68 in Redding and Venables.¹⁸ That is, reducing distances to all markets, including the domestic one, by 10% is estimated to increase GDP by 0.9% in our baseline, instead of a boost of 6.8%, or >7 times greater, based on cross-section analysis that mixed both developed and developing countries.¹⁹

View this table: [in this window] [in a new window]   Table 7. GDP estimated impact of moving each country to the average country in terms of geographical location, 2004 (in percentages)a

 

	5. Conclusion

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 
Distance continues to shape trade across countries to the same extent it did 30 years ago, notwithstanding the overall evolution of transport costs. As a result, remote countries are still penalized relative to more centrally located ones. The impression derived from Redding and Venables (2004) that distant developed countries, such as Australia and New Zealand, had largely escaped the ‘curse of distance’ is due to the inability to adequately control for heterogeneity in technology levels in cross-section samples that mix both developing and developed countries. In contrast, by focusing on a more homogenous panel sample, this article shows that the negative impact of distance to markets contributes significantly to GDP per capita even for the most developed countries. It is also argued that this focus provides a more precise estimate of the effect of distance, which is found to be some order of magnitude lower than previously estimated.

	Appendix A: The Redding and Venables model

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 
The theoretical foundations are standard in the New Economic Geography literature. The world consists of i = 1, ..., R countries and the focus is on the manufacturing industry that produces differentiated varieties under increasing returns to scale. On the demand side, each firm's product is used both in consumption and as an intermediate good, based on a constant elasticity of substitution, > 1, in both cases. Demand for goods in location j results from the maximization of the representative consumer's constant elasticity of substitution (CES) utility function:

(A.1)

where z denotes manufacturing varieties, n_i the number of firm/varieties in country i, and x_ij (z) the country j demand for variety z produced in i, or simply x_ij under the standard symmetry assumption. The resulting demand facing a firm in i from country j is:

(A.2)

where p_ij is the price of varieties produced in i and sold in j, E_j is country j 's total expenditure on manufactures and is the price index for manufactures in country j. As seen in (A.2), the own price elasticity of demand is and Redding and Venables highlight that the position of the demand curve facing each firm i in market j is given by the term E_jG_j^–1 m_j, which they refer to as the ‘market capacity’ of country j.

On the supply side, technology is Cobb–Douglas in three types of inputs. One is an internationally immobile factor that is interpreted as labour, with price w_i and input share β. The second is an internationally mobile factor with price v and input share . The third is a composite intermediate good with price G_i and input share such that + β + = 1. _ij being iceberg trade costs (i.e. _ij – 1 measures the proportion of output that is lost in shipping from i to j and _ij = 1 corresponds to costless trade), profits of the representative firm in country i is:

(A.3)

where a_i and F denote marginal input requirement (i.e. technical-level efficiency) and fixed costs, respectively. Given (A.2), whatever the destination country, profit-maximizing firms set a single FOB price p_i, which is a mark-up over marginal costs:

(A.4)

Free-entry condition leads to x_i = ( – 1)F and equation (A.2) entails:

(A.5)

The price p_i in a country i is a sum of the market capacity of all countries weighted by a function of trade costs. Bearing fixed costs of production under monopolistic competition, a firm which is located close to markets faces a large demand for its differentiated products and is able to charge a higher price. Consequently, in equilibrium, these firms pay higher wages. This is the market (domestic and foreign) access effect, which is formally obtained by combining (A.5) with (A.4):

(A.6)

The price of intermediates is the price index for manufactures, which is now given by:

(A.7)

Being close to a large supplier (high n_i, low p_i) generates lower prices of intermediates. The term n_ip_i^1– s_i measures the ‘supply capacity’ of country i, and the price of intermediates is a function of the sum of supply capacities weighted by a function of trade costs. As seen in (A.6), lower input costs of intermediate goods enable firms to pay higher wages in equilibrium ceteris paribus. This is the supplier (domestic and foreign) access effect. Therefore, better access to markets raises the prices of the internationally immobile factors through both the market and supplier access effects.

Finally, bilateral trade flows are directly derived from (A.2), as the exports from i to j, X_ij are equal to:

(A.8)

Exports from i to j depend on three terms: the supply capacity of the exporter country, the market capacity of the importer country and the trade costs between the two countries. Market and supplier access, MA and SA, respectively, are defined as the trade-costs-weighted sum of market and supply capacity, respectively, of all partner countries:

(A.9)

Equation (A.7) directly relates the price of intermediates to supplier access according to:

(A.10)

Finally, A being a constant, equation (A.6) becomes:

(A.11)

Incorporating the time dimension leads to the wage equation:

(A.12)

View this table: [in this window] [in a new window]   Table A1. Bilateral trade equation, 1970–2004a

 

	Acknowledgements

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 
The authors would like to thank numerous OECD colleagues, in particular, Sveinbjörn Blöndal, Jørgen Elmeskov, Christian Gianella, David Haugh, Peter Hoeller, Vincent Koen, Jean-Luc Schneider and Andreas Wörgötter, for their valuable comments as well as Philippe Briard and Martine Levasseur for technical assistance and Caroline Abettan for editorial support. The article has also benefited from comments by members of the Working party No. 1 of the OECD Economic Policy Committee, as well as the participants to ‘The Gravity Model’ Conference, Groningen, October 2007.

	Notes

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 
¹ Concurrently, Hanson (2005) develops a different strategy with respect to market access and tests it using data covering US counties. Combes and Overman (2004) present a survey of studies for various European countries replicating Hanson's approach developed in an earlier version.

² The authors wish to thank Stephen Redding for providing the data.

³ We are grateful to two anonymous referees for having suggested to implement the PPML estimator.

⁴ When trade costs are prohibitive = + and = 0; when they are negligible = 1 and = 1.

⁵ For more details, see the notes of the database at http://www.cepii.fr/distance/noticedist_en.pdf.

⁶ For each year, the sample is therefore composed of a maximum of 32 * 31 = 992 observations. For areas covering more than one country, the bilateral explanatory variables in (3.1b) have been weighted by the average population over the period of the countries composing each area.

⁷ We are thankful to Clément Bosquet who implemented the PPML estimator using the DOTS database in the course of his Master Thesis about the so-called ‘distance puzzle’.

⁸ As an example based on this order of magnitude, the sole effect of distance implies that the value of France's imports from Australia or New Zealand should be only 3% of that from the UK. In reality, this ratio was 15% and 5% in the 1970s for Australia and New Zealand, respectively; 8% and 2% in the 1980s; 4% and 1% in the period between 1990 and 2005.

⁹ For each date t, the constant _t in equations (3.1a and b) cannot be identified due to country fixed effects, and both log s_it and log m_jt are known up to a constant. This means that, for each date, the level of phi-ness, the market and supplier capacity are known up to a multiple constant. It follows that the log of market and supplier access variables are known up to a time dummy and, therefore, that only the relative evolution between countries through time can be depicted. In the econometric estimation of an equation such as (2.1), this is unproblematic insofar as the specification includes year fixed effects.

¹⁰ The other main difference between the two measures relates to the internal distance for Japan. Using the distance-to-cities measure, Japan is roughly at the average level, while the area-based measure places Japan in a better position.

¹¹ Despite the rising economic importance of Asia, the impact for other OECD countries’ access to markets is muted for two reasons. First, the share of Asia minus Japan in world GDP has gained <4 points since 1970. Second, even Australia and New Zealand, often seen as significant beneficiaries of strong growth in Asia, are far from the centres of growth in this area. For example, the geodesic capital-to-capital distance between Australia, on the one hand, and China and Korea, on the other hand, is 9000 and 8400 km, respectively. For Germany, these distances are 7400 and 8100 km, respectively. Moreover, Australia and New Zealand have not benefited from the large regional integration, such as driven by NAFTA or the European Single Market Programme.

¹² Relative to the sample of 24 OECD countries, Korea, Mexico and Turkey are excluded, mainly due to data availability over the 1970s, but also to restrict the dataset to the high-income OECD countries.

¹³ Because the right-hand side variables are generated regressors, the standard errors might be underestimated (Pagan, 1984). To check the extent of the standard error bias, bootstrap techniques are used to obtain appropriate standard errors for the specification in levels without first-order correlation. Each bootstrap re-samples the 36,000 bilateral trade observations to re-calculate market and supplier access, the resampling being done within country-pair bins. The standard errors are based on 200 replications of equation (3.4) estimation. This exercise suggests that the reported standard errors are underestimated by 10–20%, which is not problematic given the high-level of significance of the results.

¹⁴ The fact that including or excluding fixed effects leads to similar orders of magnitude when controlling for autocorrelation does not mean that cross-country estimates, which cannot control for country fixed effects, are acceptable. The optimal panel estimator depends on the level of autocorrelation (Wooldridge, 2002, Ch. 10). When the autocorrelation parameter, , equals 0, the within estimator is more efficient, while when equals 1, the first-difference estimator is more efficient. Therefore, as is estimated at around 0.80, the efficient estimator reported in columns (3–4) is close to first-differencing: the fixed effects that must be included in the specification (equation 4.1) are wiped out by the first-difference transformation. However, it remains that the cross-section estimator will/might be severely biased because the dependent variables are correlated with the fixed effects, and therefore with the cross-section residuals for year t, e_i + u_it using notations of equation (4.1). Table 7 (column 6) in Section 4.3 below provides the cross-section estimate on the large sample, showing the extent of that bias.

¹⁵ The GDP per capita data comes from Alan Heston, Robert Summers and Bettina Atten, Penn World Table Version 6.2, Center for Intenrational Comparisons of Production, Income and Prices at the University of Pennsylvania, September 2006.

¹⁶ Here is how the estimate herein compares with those reported by Redding and Venables. First, it was checked that using the sample limited to their 101 countries does not alter our estimates. Second, the main distance measure used here is their second distance measure. They obtained a parameter of 0.51 for market access (Table 2), which is equivalent to 0.51/1.6 = 0.32 in terms of the weighted sum of market and supplier access (weighted sum = market access + 0.6 supplier access; besides, this ratio of 1.6 is consistent with what is found for the parameter based on their third distance measure as can be inferred for comparing their Tables 2 and 3: 0.40/1.6 = 0.25). Adding some control variables leads to a reduction of 20% in the parameter (e.g. from 0.25 to 0.20 in their Table 3).

¹⁷ Another way to see this consists in estimating the equation in cross-section using our PPML indicators for consistency. The estimated parameter for the weighted sum of market and supplier access is 0.92 (SE 0.11) in 1970, 1.14 (SE 0.13) in 1987 and 0.97 (SE 0.14) in 2004, i.e. more than 3 times what is obtained based on panel estimates (see Table 7 below, comparison of columns 5 and 6). On top of the cross-country/panel distinction, the estimates by Redding and Venables are produced from both a smaller panel of 101 countries and a trade equation specified in logarithm.

¹⁸ Given the estimated elasticity of trade to distance, the market access of Australia and New Zealand is 3% of the average OECD country in Redding and Venables, versus 30% in our study. See the note in Table 7 for computation details.

¹⁹ In a parallel research, Bosker and Garretsen (2008) assess the impact of market access on sub-Saharan Africa based on panel data. Their results are broadly consistent with those herein. Controlling for fixed effects leads to a market access parameter that is divided by about 3. In their baseline, the elasticity of trade to distance is estimated at –1.5, and that of GDP to market access at 0.076, leading to an elasticity of GDP to distance of –1.5 * 0.076 = –0.11, as they only consider market access (ignoring supplier access).

	References

TOP Abstract 1. Introduction 2. Market and supplier... 3. Construction of the... 4. Impact on GDP... 5. Conclusion Appendix A: The Redding... Notes Acknowledgements References

 

Bosker EM, Garretsen H. Economic geography and economic development in Sub-Saharan Africa. (2008) Mimeo. Available at: http://maartenbosker.googlepages.com/.

Bosquet C, Boulhol H. Gravity, log of gravity and the ‘distance puzzle’? (2009) Working Paper 09-14, GREQAM, University Aix-Marseille.

Boulhol H, de Serres A, Molnar M. The contribution of economic geography to GDP per capita. OECD Economic Studies (2008) 2:287–323.

Breinlich H. The spatial income structure in the European Union – What role for economic geography? Journal of Economic Geography (2006) 6:593–617.[Abstract/Free Full Text]

Combes P-P, Overman HG. The spatial distribution of economic activities in the European Union. In: Handbook of Urban and Regional Economics—Henderson V, Thisse J-F, eds. (2004) 4. Amsterdam: Elsevier-North Holland.

Disdier A-C, Head K. The puzzling persistence of the distance effect on bilateral trade. Review of Economics and Statistics (2008) 90:37–48.[CrossRef][Web of Science]

Dolman B, Parham D, Zheng S. Can Australia match US productivity performance? (2007) Australia Government Productivity Commission, Staff Working Paper.

Fujita M, Krugman P, Venables AJ. The Spatial Economy: Cities, Regions, and International Trade (1999) Cambridge, MA: MIT Press.

Hall RE, Jones CI. Why do some countries produce so much more output per worker than others? Quarterly Journal of Economics (1999) 114:83–116.[CrossRef][Web of Science]

Hanson GH. Market potential, increasing returns and geographic concentration. Journal of Internaional Economics (2005) 67:1–24.[CrossRef]

Harris C. The market as a factor in the localization of industry in the United States. Annals of the Association of National Geographers (1954) 44:315–348.

Head K, Mayer T. Regional wage and employment responses to market potential in the EU. Regional Science and Urban Economics (2006) 36:573–594.[CrossRef][Web of Science]

Pagan AR. Econometric issues in the analysis of regressions with generated regressors. International Economic Review (1984) 25:221–247.[CrossRef][Web of Science]

Redding S, Venables AJ. Economic geography and international inequality. Journal of International Economics (2004) 62:53–82.[CrossRef][Web of Science]

Santos Silva JMC, Tenreyro S. The log of gravity. Review of Economics and Statistics (2006) 88:641–658.[CrossRef][Web of Science]

Wooldridge J. Econometric Analysis of Cross-section and Panel Data (2002) Cambridge, MA: MIT Press.

CiteULike    Connotea    Del.icio.us    What's this?

This Article Abstract FREE Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Request Permissions Google Scholar Articles by Boulhol, H. Articles by de Serres, A. Search for Related Content Social Bookmarking          What's this?

Online ISSN 1468-2710 - Print ISSN 1468-2702

Copyright © 2009 Oxford University Press

Oxford Journals Oxford University Press

• Site Map
• Privacy Policy
• Frequently Asked Questions

Other Oxford University Press sites: Oxford University Press Oxford Journals China Oxford Journals Japan American National Biography Booksellers' Information Service Children's Fiction and Poetry Children's Reference Corporate & Special Sales Dictionaries Dictionary of National Biography Digital Reference English Language Teaching Higher Education Textbooks Humanities International Education Unit Law Medicine Music Online Products Oxford English Dictionary Reference Rights and Permissions Science School Books Social Sciences Very Short Introductions World's Classics