Spatial Aspect of Global Value Chain in East Asia: How Ports and Airports Shape Industrial Clusters in East Asia

Abstract

This paper examines how geography matters for the location of industries in East Asia, employing regression analyses on a novel and comprehensive regional GDP dataset. This study examines how geography affects industrial location patterns, particularly the role of infrastructure, such as ports and airports. This paper analyzes the current economic geography of East Asia using the novel dataset. The regression analyses utilize location quotients as the dependent variable and incorporate explanatory variables, such as domestic/foreign market access, per capita income, population density, and distance-based dummies for ports and airports. The findings reveal that the determinants of industrial location differ significantly across industries. The relative importance of domestic versus foreign market access and proximity to ports and airports varies across sectors. The results imply that countries/regions cannot easily host industries of their choice, as different industries require distinct locational characteristics.

1. Introduction

There are two seemingly contradictory perceptions of geography in the globalized world. One is represented by the phrases “the death of distance” (Cairncross (2001)) and “the world is flat” (Friedman (2006)). Information and communication technology and ever-lowering trade costs make economic activities scattered worldwide, yet connected by Global Value Chains (GVCs). On the other hand, the other perception of geography is represented by the phrase “geography matters”. Even in our highly globalized world, physical proximity is still vital for economic activities, making industrial agglomerations possible.

Indeed, economic interactions are still greatly affected by distance. Crafts and Venables (2003) showed that the trade flows between countries 8000 km apart from each other are reduced to only 7% for countries located at a distance of 1000 km. Foreign direct investment (FDI) keeps 42% of the volume of trade within an 8000 km distance. At the same time, technology flows, which intuitively would seem indifferent to distance, are mostly affected by distance and reduced to only 5% if two places are located at a distance of 8000 km.

The world seems to be in the middle of being flat. Krugman and Venables (1995) show a “history of the world” in which economic activities are distributed unevenly when the level of trade costs is in the middle. Suppose the world is composed of two regions, North and South, and the economy comprises an agricultural sector with a constant return to scale (CRS), and a manufacturing sector with an increasing return to scale (IRS). When the trade costs are very high, both regions have manufacturing sectors to serve their domestic markets and have the same real wages. However, once the trade costs are lowered below a certain threshold, the North begins to host more manufacturing sectors to serve both regions because of the work of the IRS. In this phase, the industrialized North earns higher wages than the agricultural-based South. Then, the trade costs are further lowered, and it becomes profitable for a firm to locate in the South, which provides cheaper labor and has less severe market competition, causing the relocation of the manufacturing sector. The income gap between North and South starts to narrow as trade costs decline, and finally, the real wages of both regions converge at a trade cost of zero. Today, they seem to be in the “core-periphery” phase toward convergence.

In this paper, we examine how geography matters for the location of industries in East Asia by a novel geo-economic dataset. Specifically, we focus on the distance from infrastructure, such as airports and ports, and conduct a regression analysis to examine the determinants of industrial location. While it is plausible that proximity to airports and seaports gives locational advantages to industrial agglomerations, two key aspects remain unclear: (1) the specific industries that benefit more from such proximity to transport infrastructure, and (2) the distance thresholds within which the advantages of being near airports or seaports can be effectively exploited. An empirical investigation is necessary to clarify these hypotheses and quantify the relationships between industrial clusters and their access to major transportation hubs.

This paper makes two contributions to the previous literature. First, we focus on the role of infrastructure, such as ports and airports, in addition to traditional location determinants, such as market access and labor costs, proxied by per capita income. Second, we investigate industrial locations in East Asia. The previous literature has investigated mainly the US and Europe, or industrial locations within a specific country, such as China (Cheng and Kwan 2000) and India (Saikia 2011), while not even investigating East Asia. These contributions are made possible by our novel dataset of GDP by industry, covering 16 East Asian countries/regions at the sub-national level and port and airport location data, with the realistic route selection mechanism implemented in the CGE model, as discussed below. Hitherto, the lack of a reliable and comprehensive dataset of industry-specific GDP at a sub-national level for the East Asia region has impeded such research. Moreover, while many studies have used Euclidean distance as the distance measure, in reality, the transportation modes utilized differ by industry, rendering Euclidean distance a poor approximation of actual market access.

This paper is structured as follows: First, we briefly examine what economic theories say about the location of industries. Second, we present the current economic geography of East Asia through a novel and comprehensive regional GDP dataset. Third, we examine how geography affects the location of industries by regression analyses. Fourth, building upon the results from the previous section, we discuss and contrast our findings with those of the existing literature. The last part concludes this paper with some policy implications.

2. Economic Theories on the Location of Industries

A large amount of the literature has investigated the determinants of the location of industries, since Marshall (1890) identified three main sources of industrial agglomeration, namely, labor market pooling, input sharing, and knowledge spillovers. However, until spatial economics developed in the 1990s, mainstream economics had analyzed the location of industries as a part of international economics. The core concept of industrial location in global economics is a well-known comparative advantage. The theory of comparative advantage argues that a country/region hosts the industry that is comparatively advantageous to that country/region. The source of the comparative advantage may be an abundance of a specific production factor, like labor or capital (Heckscher–Ohlin theorem), or a higher production efficiency (Ricardian theorem).

Traditional international economics is based on CRS and cannot treat trade costs well. Spatial economics, or new economic geography, is a new theory that explains complex production and trade patterns in a globalized, but not flat, world. Spatial economics explains the location of industries and trade patterns as the interactions between concentration and dispersion forces (Fujita et al. (1999)). One typical example of an agglomeration force is the IRS in production. The IRS makes producing goods in one big plant more economical than multiple smaller plants. On the other hand, a typical example of a dispersion force is the high cost of transport, including various border costs (Anderson and Van Wincoop (2004)). If it is costly to export goods from one country to another, firms may choose to locate their factories in each country and serve the market where they produce the goods to save on transport costs.

Market demand is another crucial factor in spatial economics. When firms choose where to locate, they prefer to locate in larger markets to minimize transport costs. But it is precisely this movement, “located in the larger market”, which makes larger markets even larger. This mechanism is the primary circular causation that makes industrial agglomerations.

Krugman (1993) showed the relationship between the IRS, transport costs, and the number of agglomerations that appeared in a hypothetical world, the so-called racetrack economy, in which every city is located on a circle at an equal distance. Krugman showed that higher economies of scale and lower trade costs result in fewer agglomerations. Industries with a larger-scale economy tend to be located in fewer places. In comparison, industries with a smaller-scale economy tend to be located in more places. The number of agglomerations declines as trade costs are lowered.

Among empirical works, Midelfart et al. (2004) investigated the location of industries in Europe, and found that the industrial structure of European countries converged during the 1970s, but has diverged since then. They concluded that factor costs and geographical factors, such as the existence of an educated labor force, access to intermediate goods, and the degree of scale economy, can largely explain the cross-country differences in industrial structure. Brülhart and Traeger (2005) found that during the period 1975–2000 in Europe, the manufacturing sector exhibited a statistically significant increase in its concentration relative to the spatial distribution of overall employment, while simultaneously experiencing a statistically significant decrease in its concentration relative to geographic space. Cheng and Kwan (2000), by estimating the effects of determinants on foreign direct investment (FDI) in 29 Chinese regions from 1985 to 1995, found that large regional markets, good infrastructure, preferential policies, and FDI’s self-reinforcing effect had positive impacts, while wage costs had a negative impact and education’s effect was insignificant. Head and Mayer (2004) found that market potential explains some part of the location choice of Japanese foreign direct investment in Europe. Arauzo-Carod et al. (2010) surveyed the various literature on industrial location and concluded that the factors considered in location choice, such as tax policies, wages, and agglomeration forces, have not altered since the 1980s, but newer datasets and theoretical models have confirmed that these factors work as predicted by theoretical models.

3. Economic Geography of East Asia

East Asia has a considerable production base of manufacturing goods with GVCs among its member countries/regions. East Asia seems to be an example of a “flat world”, in which one product is been sent from one country to another to be a final product, through various production stages. However, this across-country value chain is economically viable because of considerable differences in nominal wages and factor endowments among the regions. Each production stage seems to be located in a suitable place to conduct that specific production stage utilizing these differences. For instance, the production stage, which is very labor intensive, was typically in China, a formerly labor-abundant country, and is now shifting to Vietnam, Cambodia, and Myanmar.

In this section, we depict the basic economic geography of East Asia using a comprehensive geographical dataset, the Geo-Economic Dataset for Asia (GEDA) 2005, compiled by IDE-JETRO. The GEDA dataset provides sub-national Gross Domestic Product (GDP) data for 25 sectors across the following industries: agriculture, fishing and forestry, mining, manufacturing (16 sectors), and services (7 sectors). The GDP figures are nominal and have been converted to US dollars using the average exchange rate between the local currency and US dollars in 2005, based on data from the International Financial Statistics (IFS) by the International Monetary Fund (IMF). The GEDA dataset encompasses regional demographic statistics including population and land area, coupled with the GDP by industrial sector.

The regional divisions in the GEDA follow the administrative boundaries within each country. Except for Hong Kong, Macao, Singapore, and Brunei, GEDA adopts a primary-level administrative division, although secondary-level administrative divisions are adopted for China, India, Indonesia, Bangladesh, and Myanmar. For more details on the data sources and methods used in compiling the GEDA, please refer to the “Data source and notes” section of the GEDA website (https://www.ide.go.jp/English/Data/Geda.html (accessed on 15 May 2024)).

Table 1. Number of administrative divisions by country/regions.

Bangladesh	64	Laos	17
Brunei	1	Macao	1
Cambodia	24	Malaysia	15
China	342	Myanmar	64
Hong Kong	1	Philippines	17
India	579	Singapore	1
Indonesia	435	Taiwan	25
Japan	47	Thailand	76
Korea	16	Vietnam	61
Total			1786

Source: author.

shows the number of administrative divisions by country/regions in the GEDA.

3.1. Basic Information

Figure 1 shows the population density by region in 2005. The regions with a high population density of over 500 people/square km are observed in the Ganges Basin, the coastal areas of China, and Java Island, in addition to most of the capital regions of each country. The distribution of population is very much uneven in East Asia.

Figure 2 shows the GDP per capita by region in 2005. The income distribution among East Asian regions is highly uneven, both internationally and intranationally. The regions with isolated high incomes are capital or mineral-resource-rich regions.

Figure 3 shows the GDP per square km, or “GDP Density”. This index seems to be a good measure of the agglomeration of economic activities (Panayotou 1997; Apergis and Ozturk 2015), because it is a product of the population density and GDP per capita, as follows:

East Asia’s economic geography is very much uneven (

Table 2. Summary of statistics of geo-economic data for East Asia, 2005.

	Obs.	Min	Median	Mean	Std. Dev.	Max	Min/Max
GDP per capita	1786	69.8	713.2	2365.5	6119.9	69,730	998
Population density	1786	0.26	261.5	704.3	2554.2	25,630	97,452
GDP per sq. km (‘000USD)	1786	0.4	190.0	3077.7	20,851.9	408,800	1,064,583

Source: author.

). For per capita income, the lowest to highest ratio is almost 1:1000; it is 1:100,000 for population density; and it is 1:1,000,000 for GDP density.

3.2. Industrial Agglomeration

One of the indices used to measure the degree of industrial agglomeration is the location quotient (LQ). The LQ for industry i in region r is defined as follows:

$${LQ}_{ir}=\raisebox{1ex}{$q_{ir}/q_r$}\!\left/ \!\raisebox{-1ex}{$Q_i/Q$}\right.=\raisebox{1ex}{$q_{ir}/Q_i$}\!\left/ \!\raisebox{-1ex}{$q_r/Q$}\right.$$

(1)

where q_ir is industry i’s employment in region r, q_r is the total employment in region r, Q_i is industry i’s employment in the reference area, and Q is the total employment in the reference area. The LQ is a theoretically substantiated agglomeration index that exhibits a relatively unbiased nature, rendering it a suitable index for evaluating economic activity concentration or spatial clustering within a specific geographic region relative to a broader reference area (Billings and Johnson 2012).

Here, we set 16 countries/regions in East Asia as the reference areas. For example, if the automotive industry accounts for 3% of the total employment in a sub-national region, while it accounts for 2% of the total employment in 16 East Asian countries/regions, then the LQ for the automotive industry for the region is 3%/2% = 1.5. So, an LQ > 1 means an industry concentrates in a region more than the average for the reference area.

Unfortunately, the GEDA 2005 has no information on the numbers employed by region and industry. Alternatively, we can calculate the proxy of employment by industry and region as follows:

$$q_{ir}={GDP}_{ir}/{GDP}_r·{Pop}_r$$

(2)

In this case, the LQ is calculated as follows:

$${LQ}_{ir}=\raisebox{1ex}{$q_{ir}/{Pop}_r$}\!\left/ \!\raisebox{-1ex}{$\sum _r{q}_{ir}/\sum _r{Pop}_r$}\right.$$

(3)

Figure 4 and Figure 5 show the LQ of each region’s automotive and food processing industries in 2005. The automotive industry agglomerates in relatively small areas, while the food processing industry is located in broader regions.

Next, we compare the degree of agglomeration between industries. The Theil index is used for this purpose and is calculated as follows:

$$T_i={\sum }_r{q}_{ir}/Q_i·\mathrm{l}\mathrm{n}\left({LQ}_{ir}\right)$$

(4)

The index takes 0 if the distribution of a specific industry perfectly follows the distribution of a reference value, in this case, total employment. A bigger index value means more concentration compared with the reference value. One of the merits of the Theil index is that inequality/dissimilarity can be decomposed into multiple layers, such as between-country and within-country inequality (Conceição and Ferreira (2000)).

Figure 6 shows the Theil index for ten industries. The Theil indices for the agricultural and services sectors are very low, showing that these industries exist in every region and country. The automotive sector has the highest index value among the manufacturing industries, and the electric and electronic (E&E) sector follows. These industries tend to agglomerate in a smaller number of regions. The food industry has the lowest Theil index, showing that the industry tends to be located in many regions. If you look at the between-country Theil index, the textile and garment industry has the highest value, and the wood industry follows, meaning that these industries tend to agglomerate in selected countries/regions. On the other hand, the petroleum and chemical, food processing, and iron and steel industries have the lowest values, meaning that these industries tend to be located in every country.

A theoretical interpretation is as follows. As discussed in Section 1, spatial economics predicts that an industry with a higher scale economy and lower trade costs is located in fewer places. The automotive industry has a higher IRS, while the E&E industry has lower transport costs than their goods’ value. So, the higher concentration of both industries is understandable. The food industry has a lower IRS, higher transport costs, and less concentration.

For the between-country Theil index, the textile and wood and paper industries have relatively higher values, i.e., higher international concentration. This seems to be explained not by spatial economics but by the comparative advantage of traditional international economics. The textile industry is highly labor intensive, and the wood and paper industry is resource-based. Both industries seem to be internationally concentrated in countries/regions with abundant labor and wood resources. As for the automotive industry, the between-country Theil index is low, meaning there is less international concentration. It can be interpreted that the automotive industry’s trade costs are very high because of various protectionist policies in each country. So, they tend to be located in each country while showing a high intranational concentration.

4. What Are the Determinants of the Spatial Distribution of Economic Activities?

In this section, we conduct an econometric analysis to identify the determinants of the spatial distribution of economic activities. Specifically, we conduct a regression analysis to identify the locational factors for industries in 2005 for each of the 1786 sub-national regions across the 16 countries/regions covered in the GEDA dataset. The dependent variables used are the location quotients (LQs) for each industry, while various explanatory variables are included in the regression model.

Spatial economics predicts that economic agglomerations tend to be located near large markets. One of the measures for the goodness of a location is market access. Market access is a concept that measures the proximity to markets, classically defined as the distant weighted sum of GDP for each market as follows:

$${MA}_r={\sum }_s{GDP}_s/T_{rs}$$

(5)

where ${GDP}_s$ denotes the Gross Domestic Product for region s, and $T_{rs}$ represents the trade costs incurred between region r and region s.

Redding and Venables (2004) showed that market access in a country is highly correlated with its per capita income. They used the theory-backed market access variable, but here, we calculate it using Equation (5) for simplicity. The GDP data used in this analysis include not only the 1786 sub-national regions across the 16 countries/regions covered in the GEDA dataset, but also the GDP data for 68 additional countries, represented by their respective capitals. This allows us to consider market access to regions outside of East Asia as well.

For the trade cost measures, we utilize the road network data used in a CGE model called IDE-GSM (see Kumagai et al. 2013). Instead of using Euclidean distances, we constructed trade cost data that consider actual road, rail, sea, and air transportation routes, rather than just direct distances. We use the functionality implemented in IDE-GSM to select the optimal multimodal transportation network for each industry–origin–destination combination. We also consider the waiting time and costs at border crossing points to construct the trade cost data when calculating the trade costs for foreign markets.

Market access can be divided into domestic market access (DMA) and foreign market access (FMA). DMA shows whether a region has good access to the market within the same country, while FMA shows whether an area has good access to foreign markets. The importance of market access may differ by industry, and the relative importance of FMA and DMA may vary, too.

Here, we use the LQ as the dependent variable. For the independent variables, we include DMA and FMA. So, we may estimate using the following equation:

$${LQ}_{ir}={\alpha }_i+{\beta }_{1i}\log \left({DMA}_r\right)+{\beta }_{2i}\log \left({FMA}_r\right)+{\varphi }_i{X}_r+{\epsilon }_{ij}$$

(6)

where X_r is the vector of other control variables, such as the GDP per capita and the population density of a region, and ${\epsilon }_{ij}$ is the error term.

In addition to the above equation, we try to capture the impacts of transport infrastructure, such as ports and airports, on the location of industries. Limao and Venables (2001) showed that infrastructure is an essential determinant of trade costs, and an improvement in the quality of infrastructure from the median to the 25th percentile makes trade costs 17% points lower, equivalent to 2358 km closer to all trade partners.

To see the effects of transport infrastructure on the location of industries, we add the port/airport dummy into the estimation of Equation (6). If the nearest port/airport for a region is within specific distance brackets from the capital city of that region, the corresponding dummy variable takes 1. In constructing these dummy variables, we considered 191 ports and 65 airports in 16 countries/regions covered by the GEDA. The distance from the ports and airports to each region is based on the Euclidean distance between the ports/airports and the capital cities of the respective regions. By estimating the coefficients of each dummy variable, we can detect if access to the port/airport matters for the industrial agglomeration. The rationale for employing Euclidean distance here is rooted in its simplicity, which facilitates the creation of distance-based dummy variables and enhances the interpretability of the results. The trade costs, accounting for multimodal route choice behavior when constructing the market access variable, encompass various elements, rendering it difficult to reduce the discussion to a mere distance measure. Furthermore, since the distances from each capital city to the nearest port or airport are predominantly domestic and relatively proximate, these can be well approximated with Euclidean distances.

Now, the equation used to estimate is as follows:

$${LQ}_{ir}={\alpha }_i+{\beta }_{1i}\log \left({DMA}_r\right)+{\beta }_{2i}\log \left({FMA}_r\right)+{\delta }_i^d{PD}_r^d+{\gamma }_i^d{AD}_r^d+{\varphi }_i{X}_r+{\epsilon }_{ij}$$

(7)

where ${PD}_r^d$ denotes a port dummy variable that takes a value of 1 if the distance from region r to the nearest port falls within the distance bracket d, and 0 otherwise. The distance brackets considered are less than 50 km, 51–100 km, 101–150 km, 151–200 km, 201–300 km, and 301–500 km. Similarly, let ${AD}_r^d$ be an airport dummy variable defined analogously, taking a value of 1 if the distance from region r to the nearest airport lies within the distance bracket d, and 0 otherwise. The distance brackets for airports are specified identically to those for ports.

While there are endogeneity concerns regarding the relationship between the locations of ports and airports and the locations of industries, the major ports and airports in many East Asian countries/regions were established much earlier, often during the colonial period, long before industrialization began. Consequently, endogeneity concerns are considered to be low, and it is difficult to find the proper instrumental variables. So, we use the common OLS for the estimation measure. Given these constraints, this study’s estimated relationship between ports/airports and industrial agglomerations should be interpreted as a correlation rather than as a causal effect. However, we posit that regardless of whether ports/airports induce industrial agglomerations or industrial agglomerations drive the construction of ports/airports, the underlying fact remains that the existence of such industrial agglomerations necessitates the presence of ports or airports.

4.1. Automotive Industry

Table 3. Results of regression.

	AUTO		E&E		TEXT		FOOD		WOOD		PTCM		IRON
Intercept	−4.439		−59.434	***	−32.463	***	9.182	**	2.493		−14.809	**	−24.359	***
	(7.627)		(6.721)		(6.309)		(4.640)		(8.450)		(5.800)		(6.463)
DMA	0.305	***	−0.048		0.206	***	0.043		−0.139		0.006		0.280	***
	(0.077)		(0.068)		(0.064)		(0.047)		(0.086)		(0.059)		(0.065)
FMA	−0.127		2.534	***	1.343	***	−0.449	**	−0.006		0.563	**	0.831	***
	(0.348)		(0.307)		(0.288)		(0.212)		(0.386)		(0.265)		(0.295)
Port50	−0.381	**	0.414	**	0.435	***	0.278	**	1.140	***	0.761	***	−0.192
	(0.186)		(0.164)		(0.154)		(0.113)		(0.206)		(0.141)		(0.157)
Port100	−0.143		0.845	***	0.460	***	0.033		0.682	***	0.173		−0.154
	(0.185)		(0.163)		(0.153)		(0.113)		(0.205)		(0.141)		(0.157)
Port150	−0.158		0.514	***	0.794	***	0.138		0.708	***	0.221		0.115
	(0.188)		(0.166)		(0.156)		(0.114)		(0.208)		(0.143)		(0.159)
Port200	−0.320		0.417	**	1.317	***	0.147		0.587	**	0.271	*	0.047
	(0.206)		(0.182)		(0.171)		(0.126)		(0.229)		(0.157)		(0.175)
Port300	0.077		0.287	*	0.484	***	0.250	**	0.546	***	0.072		0.011
	(0.185)		(0.163)		(0.153)		(0.113)		(0.205)		(0.141)		(0.157)
Port500	−0.090		0.050		0.103		−0.163		0.314	*	−0.014		−0.010
	(0.170)		(0.150)		(0.141)		(0.104)		(0.189)		(0.129)		(0.144)
Airport50	0.849	***	1.165	***	1.016	***	−0.350	**	−0.648	**	−0.066		−0.162
	(0.235)		(0.207)		(0.195)		(0.143)		(0.261)		(0.179)		(0.199)
Airport100	0.501	**	0.380	*	0.731	***	−0.139		−0.882	***	−0.091		0.047
	(0.223)		(0.197)		(0.185)		(0.136)		(0.247)		(0.170)		(0.189)
Airport150	0.088		0.137		0.346	**	0.120		−1.062	***	−0.261		−0.160
	(0.212)		(0.187)		(0.176)		(0.129)		(0.235)		(0.162)		(0.180)
Airport200	−0.244		0.084		0.142		0.096		−0.861	***	−0.163		−0.250
	(0.219)		(0.193)		(0.182)		(0.134)		(0.243)		(0.167)		(0.186)
Airport300	−0.132		0.077		0.047		−0.183	*	−0.847	***	−0.152		0.060
	(0.163)		(0.144)		(0.135)		(0.099)		(0.181)		(0.124)		(0.138)
Airport500	−0.080		−0.013		−0.103		−0.013		−0.700	***	−0.087		−0.076
	(0.146)		(0.129)		(0.121)		(0.089)		(0.162)		(0.111)		(0.124)
GDPPC	0.193	**	0.423	***	−0.324	***	0.047		0.158	*	0.266	***	0.110
	(0.084)		(0.074)		(0.069)		(0.051)		(0.093)		(0.064)		(0.071)
POPD	0.018		0.096	**	0.016		0.078	***	0.091	*	0.113	***	−0.044
	(0.045)		(0.040)		(0.037)		(0.027)		(0.050)		(0.034)		(0.038)
obs=	1722		1722		1722		1722		1722		1722		1722
Adj. R2	0.09		0.19		0.09		0.03		0.05		0.10		0.07

Note: standard errors in parentheses. ‘***’, ‘**’, and ‘*’ shows significance at 1%, 5%, and 10% level, respectively. Source: author.

shows the estimation results. The coefficients of each explanatory variable on the LQ differ significantly by industry. DMA has positive and statistically significant effects on the automotive industry, while the coefficient of FMA is negative and statistically insignificant. This means that size and domestic market access are essential for the automotive industry. Considering that the coefficients of the port dummies are generally negative and statistically insignificant, the general locations of the automotive industry in East Asia in 2005 seem to not be export-oriented. The positive and statistically significant impact of airports within 100 km are interpreted as indicating that the ease of international travel is vital for the automotive business, because the automotive industry in developing Asia was dominated by MNCs in 2005.

4.2. E&E Industry

For the E&E industry, FMA has positive and statistically significant impacts, while DMA does not. This means that access to foreign markets is very important for the industry, while the size and access to local markets do not matter much. Considering that both the port dummies within 300 km and the airport dummies within 100 km have positive and statistically significant effects, the E&E industry in East Asia in 2005 seems to have been highly export-oriented and integrated into GVCs.

4.3. Textile and Garment Industry

DMA and FMA positively and statistically significantly impact the textile and garment industry. Exports are crucial for this industry, since both the port dummies within 300 km and the airport dummies within 150 km have positive and statistically significant effects. Still, the domestic market is also essential, indicated by the coefficient of DMA. One notable result is that the coefficient of the GDP per capita is negative and statistically significant. This means that the textile and garment industry tends to avoid locations in regions with higher incomes. This result is consistent with the intuition that the industry needs abundant, low-cost labor.

4.4. Food Processing Industry

For the food processing industry, the coefficient of DMA is positive but statistically insignificant, and the FMA is negative and statistically significant. The results are hard to interpret. The coefficients of the port/airport dummies are also hard to explain. Considering the positive and highly significant coefficient of population density, and the lowest adjusted R² value of 0.03 among all industries, the food processing industry is located in highly populated regions to serve their needs, and other factors are hard to detect in this analysis.

4.5. Wood and Paper Industry

For the wood and paper industry, both the coefficients of DMA and FMA are negative, but statistically insignificant. The coefficients of the port dummies within 500 km are positive and statistically significant, while those for the airport dummies within 500 km are negative and statistically significant. The port dummies are easy to interpret if you consider that wood products need ports for export, but interpreting the airport dummies is difficult. The wood and paper industry is resource-based and less affected by market access, but ports seem essential.

4.6. Petroleum and Chemical Industry

DMA is not statistically significant for the petroleum and chemical industry, while FMA is positive and significant. The port dummy within 50 km is positive and highly significant, while the airport dummies are all statistically insignificant. The petroleum and chemical industry seems export-oriented, and ports are essential.

4.7. Iron and Steel Industry

DMA and FMA are positive and statistically significant for the iron and steel industry. The other explanatory variables are all statistically insignificant. From this analysis, the industry seems to locate nearer to larger markets within a country and internationally.

Figure 7a–c graphically show the port/airport dummies coefficients listed in Table 2 for selected industries. It indicates that the economic impacts of ports/airports differ significantly by industry, and how far their impacts last also differ by industry. For the E&E industry (Figure 7a), the impacts of airports are highest within 50 km, but then fade out fairly quickly. On the other hand, the impacts of ports last as far as 200–300 km from the ports. Airports have no statistically significant impacts on the petroleum and chemical industry (Figure 7b), while ports have substantial and strong impacts only within 50 km. For the textile and garment industry (Figure 7c), the economic impacts are most considerable in the 151–200 km bracket. In contrast, the economic impacts of airports are most substantial within 50 km, and statistically significant up to within 150 km.

5. Discussion

The regression analysis conducted in the previous section reveals several interesting facts. First, the factors influencing industrial agglomeration differ significantly across industries. For some industries, like the automotive industry, domestic market access is crucially important, while for the E&E industry, access to foreign markets is extremely important. This can be understood by considering the history of the automotive industry in East Asia, which developed as an import-substitution industry in most countries (Busser 2013; Natsuda et al. 2013; Okamoto and Sjöholm 1999), except in Japan and South Korea. Conversely, the E&E industry in East Asia has been an export-oriented industry (Hobday 2001; Rasiah 2008) from its inception, making access to foreign markets important, and is consistent with the results. On the other hand, access to domestic and foreign markets is found to be unimportant for resource-based industries like the wood and paper industry. This is unsurprising, considering that the location of industries utilizing natural resources depends on their geographic endowment.

Second, the importance of ports and airports also varies significantly across industries. For the E&E and textile and garment industries, proximity to both airports and ports is important, while for the wood and paper and petrochemical industries, only proximity to ports matters. These results indicate that the research by Song and van Geenhuizen (2014), which claimed that ports promote regional output in general, is inadequate, as it fails to consider industry differences. In contrast, Huang (2011) estimated the impact of ports on industrial agglomeration by industry, showing a positive effect for industries like tobacco, sericulture, and oil refining, but not conducive for the textile industry or clerical machinery manufacturing. While their results are broadly consistent with our study, the finding that ports do not positively influence textile agglomeration differs, possibly due to differences in industrial classification or because Shunquan’s study focused solely on China, whereas our analysis encompasses a broader geographic scope.

Third, the distance to ports and airports is found to have varying impacts on industrial agglomeration across industries. To our knowledge, such findings are novel. For instance, while being within 50 km of an airport is important for the electronics industry, the textile and garment industry benefits most when a port is located 151–200 km away, and the petrochemical industry benefits from being within 50 km of a port. Considering the time sensitivity of the electronics industry, proximity to airports is crucial. The observation that petrochemical plants worldwide are often located adjacent to ports aligns with our finding that being within 50 km of a port is advantageous for this industry. Proximity to ports facilitates the cost-effective transportation of bulk raw materials, such as crude oil and natural gas, as well as the shipment of finished petrochemical products. For the textile and garment industry, as road transportation is feasible, proximity to ports is important but does not necessarily require extreme closeness, making the result persuasive.

The one thing we can determine from the above results is that time costs are the key to understanding the impacts of ports/airports. Air cargo is mainly used to save time costs; thus, it must be very near to firms. In the case of the E&E industry, it should be located within 50 km to take advantage of the airport, while it is acceptable for the textile and garment industry to be within 100 km. Maritime transport is not very time conscious, and the impacts last as far away as 300 km for some industries.

6. Summary and Policy Implications

The distribution of industry is naturally uneven. However, the degree of unevenness differs by industry. From the viewpoint of spatial economics, industries with a higher IRS and lower transport costs tend to agglomerate in a small number of regions.

What affects the location of an industry may differ significantly by industry. The relative importance of DMA and FMA differs. A DMA-sensitive industry is more challenging to host in countries/regions with smaller markets. Developing infrastructure and reducing tariffs and non-tariff barriers can increase FMA.

The importance of specific transport infrastructure also differs by industry. For the E&E industry, the proximity to airports is essential, while the proximity to ports is essential for the petroleum and chemical industries. The impacts of ports on industrial locations generally last within 300 km, while the impacts of airports last within 50–100 km. This paper implies that a country/region cannot host a specific industry of choice as one likes. Different industries are located in places with different characteristics. This paper makes a significant contribution to the formulation of regional industrial promotion policies, especially by demonstrating that the distance from crucial infrastructure, such as airports and seaports, is an extremely important determinant of industrial location.

This detailed analysis of the relationship between industrial location and infrastructure within East Asia was made possible by combining a new geo-economic dataset, called the GEDA, and the realistic route selection mechanism implemented in the CGE model, called IDE-GSM. This represents a significant advantage of our paper over the existing literature.

While the data used in this study are from 2005, the GEDA dataset is the only comprehensive source covering industry-specific GDP at the sub-national level across multiple East Asian countries; moreover, industrial agglomerations form gradually over time and tend to persist once established, so the insights gleaned from the 2005 data remain relevant for understanding the determinants of industrial location patterns in the region today, despite the desirability of more recent data.

Here, we analyzed the economic geography at an industry level. Still, it might not be fine-enough criteria to analyze the spatial aspect of GVCs, because the further fragmentation of the production process makes a specific production process separately located from other processes in the same industry (Kimura (2006); Baldwin (2013)). For instance, in the automotive industry, the production of engines has a more significant scale economy. It tends to be located in a limited number of places, while wire harness assembly can possibly be located in a larger number of places. Considering these points, future research should employ more disaggregated industry classifications and newer data. Moreover, it is desirable to utilize panel data sources, which would necessitate the construction of detailed regional- and industry-level GDP datasets for East Asia.