Investigating the impact of four dimensions of distance on bilateral interactions over time: Who trades with whom and why?

(1)

Investigating the impact of four dimensions of

distance on bilateral interactions over time:

Who trades with whom and why?

Abstract: Many studies on the topic of bilateral interactions between economies have proven that distance and barriers to international trade seem to remain large. However, the relative importance of the different dimensions of distance have not yet been clearly compared. The aim of this paper is to estimate, interpret and compare relative bilateral trade costs. This study makes use of a standard specification of the gravity model of international trade with a large panel data set that incorporates 192 countries over a period of 65 years. With the ongoing trade liberalization, technical innovations and interconnectedness forces that seem to drive trade might have shifted over the years. My empirical work on bilateral interactions delivers results that are consistent with the mainstream literature on the topic. Furthermore, it presents evidence on the fact that physical distance has become a relative more important component of trade costs over the years instead of a less important factor.

Keywords: International Trade, Gravity Equation, Trade costs

Name student: Fleur Ruiterman Student ID number: S1885731

Student email: fleurruiterman@live.nl Date Thesis: 7-7-2014

Name 1st supervisor: Dr. S. Brakman Name 2nd supervisor: Dr. R. Ortega Argiles

University of Groningen

Faculty of Economics and Business

(2)

2

LIST OF GRAPHS

Graph 1. Comparison of world trade and world GDP growth... Graph 2. Annual world GDP growth... Graph 3. World GDP in trillions of current US dollars... Graph 4. Annual motor gasoline prices in dollar per gallon... Graph 5. World international air cargo values... Graph 6. Aggregate world import tariff rates... Graph 7. Annual number of RTAs into force... Graph 8. Export value and GDP of importer for four country-pairs...

LIST OF TABLES

(5)

5

ABBREVIATIONS LIST

ANZCERTA The Australia New Zealand Closer Economic Agreement ASEAN Association of Southeast Asian Nations

CACM Central American Common Market

CAGE Cultural, Administrative, Geographic, and Economic distance framework CARICOM The Caribbean Community

CEPII Centre d'Etudes Prospectives et d'Informations Internationales COW Correlates Of War project

DOTS Direction of Trade Statistics

EIA Energy Information Administration

EU European Union

FDI Foreign Direct Investment

GATT General Agreement on Tariffs and Trade GDP Gross Domestic Product

HO Heckscher-Ohlin

ICAO International Civil Aviation Organization IMF International Monetary Fund

ISO International Organization for Standardization LSDV Least Square Dummy Variable

MRTs Multilateral Resistance Terms

NAFTA North American Free Trade Agreement

OAPEC Organization of Arab Petroleum Exporting Countries

OECD The Organisation for Economic Co-operation and Development OLS Ordinary Least Squares

PATCRA Papua New Guinea-Australia Trade and Commercial Relations Agreement RTA Regional Trade Agreement

SPARTECA The South Pacific Regional Trade and Economic Co-operation Agreement USA United States of America

(6)

6

1. INTRODUCTION

Remarkably, German exports to France are 14 percent higher than German exports to the United States, the economic giant of the world which has a GDP that is almost eight times as large as that of France (Brakman, Garretsen, & Marrewijk van, 2009). It would appear that there is a strong local flavour to the German top export markets; however seven of the eight German neighbours are included in the list with the top fifteen trading partners, the majority of which are rather small countries (Brakman, Garretsen, & Marrewijk van, 2009). This is a great example of how physical distance proves to be of importance. But what about the argument that Germany is trading more with France than the USA due to the fact that both Germany and France have joined the EU?

From the period 1950 onwards, international trade has grown almost twice as fast as the average world income. While the world´s Gross Domestic Product (GDP) grew by 3.8 percent annually, international trade has grown by 8 percent on average annually (Behar & Venables, 2010); (Rose, 2004). Graph 1 below shows the yearly growth rates of world trade and GDP using data from the World Bank1.

Source: World Bank This spur in growth might for one stem from a sharp decrease in trade costs as a result of global trade liberalization and technical innovations in communication and transportation (Novy, 2013). However, the world still is not as flat as Friedman (2005) wants us to believe it is. The impact of distance still remains a hurdle to trade; an empirical result that has been

1

Due to exogenous shocks, the year 1975 and 2009 can be assumed biased observations and have not been included in the periods average calculation. The average world growth in exports over the period 1970 until 2012 is 6.23 percent and the average world GDP growth over that same period is 3.66 percent according to the data that has been abstracted from the World Bank database.

0 2 4 6 8 10 12 14 19 60 19 62 19 64 19 66 19 68 19 70 19 72 19 74 19 77 19 79 19 81 19 83 19 85 19 87 19 89 19 91 19 93 19 95 19 97 19 99 20 01 20 03 20 05 20 07 20 10 20 12

Graph 1. World Trade and GDP growth, 1960-2012

(7)

7

proven by many studies on the topic (Disdier & Head, 2008). The absences of a common border, waterway access or adequate infrastructure are examples of factors that could explain the nonexistence of bilateral interactions (Ghemawhat, 2001). However, geographical features alone cannot always be hold accountable for the absence of trade. India and Pakistan for example, even though they share a border, a common language and a colonial history, do not conduct in business with each other as one might expect due to political hostility. Strong legal and financial institutions and trade facilitating borders are examples of factors which could remove political distance between countries, which in turn could boost bilateral trade flows between a country pair (Ghemawhat, 2001). Distance remains important, whether this is due to the cultural, political, geographical or economical (CAGE) dimension of distance (Ghemawhat, 2001).

(8)

8

Neoclassical trade theory proposes. Ricardo’s (1817) wine for cloth example is no longer feasible and international competition of trade can be characterized by its unpredictability, its suddenness and the shift of focus away from sectors/industries towards individuals (Baldwin, 2006). But what are the consequences of these developments on the direction of trade flows? And how can we explain the booming growth of world trade?

The existing trade flows between countries can be explained by multiple characteristics of both the country of origin and the country of destination. In order to be able to realize profits from international trade, buyers and sellers must incur a variety of costs other than production costs, which are known as the international trade costs or barriers to trade. These trade costs have, among other parts, distance- and border-related components and include transportation costs, travel costs, communication costs, customs costs (trade policy barriers), currency conversion costs, and transaction costs (Head, Elements of Multinational Strategy, 2007).

Krugman (1995) noted that there tend to be two mainstream arguments when it comes to the explanation of the booming growth of world trade. One argument that is often used is the decline in transportation and communication costs due to technology improvements that causes ever increasing world trade. The second theory is based on the removal of protectionist measures due to trade liberalization and agreements that in turn might have spurred trade. Transportation costs and communication costs are influenced by the physical dimension of distance and are shaped by factors such as water access, quality of infrastructure, fuel costs and transportation technology (Behar & Venables, 2010). These costs are especially important for products that are heavy or require strong coordination (Ghemawhat, 2001). However, trade costs do not arise from the physical dimension of distance only. A study by Wilson (2003) reveals that the average waiting time at a border could be translated into a travel of 1600 km in-land. Many have argued that transport costs have been dropping post World War II (WWII). These costs stem from border-related costs such as administrative and legal policies and could be labelled as costs from the political dimension of distance.

While the physical aspect of distance has become relatively less important due to transport innovation that has been boosted by both technical and institutional innovations such as containerisation2 and transport deregulation (Behar & Venables, 2010); (Hummels, 2007); (Buch, Kleinert, & Toubal, 2004), the increase of cooperation and negotiation between countries has removed some of the political distance; especially when it comes to tariffs (Hummels, 2007). The most prominent example of removing political distance between

2_{Containerisation has been used by transportation companies as a way of reducing their operational costs by}

(9)

9

trading partners is the foundation of the European Union (Ghemawhat, 2001). By, theoretically speaking, removing the borders between its members and allowing for free movement of capital and labour across the union, it promotes internal trade. Both developments; innovations and technical developments in transporting modes on the one hand and the drop in tariffs as a result of trade negotiations after the 1950s on the other hand, have made it unclear to what extent political distance is now more important in explaining bilateral trade flows than physical distance or the other way around (Hummels, 2007).

By exploiting a panel data set of 226 countries over a time period of 65 years, this paper has two main objectives. First, it investigates the importance of distance on international trade, looking both at the influence of remoteness and political factors. Second, it compares the different dimensions of distance and their relative effects on trade. It raises questions such as: to what extent has the proclaimed drop in transportation costs laid emphasis on the political distance between countries? In the end this study will compare all factors described above. To sum up: with the ongoing international effects of globalization, trade liberalization and decreasing transport, communication and coordination costs driven by technological changes, the following research question can be raised:

Which dimension of distance is relatively more important in explaining bilateral trade flows between countries over time?

The remainder of the paper proceeds as follows. Section 2 will introduce the literature available on the topic, the strengths and weaknesses of mainstream arguments and the existing literature gap. Section 3 will provide the hypotheses and the model that underpins this paper. Section 4 will introduce the data sources, the appropriate econometric methods and possible caveats that come with large panel sets and the gravity equation in particular. Section 5 will discuss the estimation results. Section 6 will provide a discussion and some limitations on this study. Also, it presents suggestions for future research. Finally, Section 7 will conclude the paper.

2. LITERATURE REVIEW

(10)

10

two or more dimensions of distance will be summarized and reviewed, as these are the main areas of interest to this study and in the end a proper literature gap will be revealed.

2.1 Gravity models of international trade

It is important to clarify what is meant by an international transaction, as the term encompasses many concepts. It is about cross-border transactions (not necessarily different ownership). Firms doing business abroad face extra challenges and hurdles that domestically active firms do not face. The price of cross-border business lies in the opportunity of reaping much higher profits. However, with these international transactions there also exist possible costs due to different forms of separation that extricate international trade from national transactions (Head, Elements of Multinational Strategy, 2007).

Trade theories justify the allocation and direction of economic activity between countries (Brakman, Garretsen, & Marrewijk van, 2009). One of the first and most famous economic theory on trade is the Neoclassical trade theory which proclaims that trade flows between nations is based on comparative advantages, resulting from technological differences or from factor abundance. However, now that we are dealing with rapid technological and institutional changes, this theory might no longer apply. All kinds of new ways of thinking have been introduced into the economic literature. The main point is that trade theorists are mostly interested in how market structure, production techniques, and consumer behaviour interact (Neary, 2004). The resulting factor and commodity prices determine the pattern of international trade flows. In their opinion, location is an exogenous factor and usually does not play a role of significance. Also, according to the HO model, country size has little to do with the direction and size of trade flows (WTO, 2013). However, gravity models of international trade are the exception to the rule3. These models are easily extended to include all kinds of transaction costs and take into account the location of one economy relative to another (Brakman, Garretsen, & Marrewijk van, 2009).

According to the gravity model of international trade, bilateral trade flows and physical distance between economies are negatively correlated. The model also proves that bilateral trade is proportional to the relative size of the markets that are trading (Buch, Kleinert, & Toubal, 2004). These empirical findings are some of the most robust empirical discoveries in the economic geography literature and have been proven by a latitude of different studies

3

(11)

11

which use the standard gravity model of international trade (Leamer & Levinsohn, 1995), (Disdier & Head, 2008).

Tinbergen (1962) was the first to use ‘gravity’ as the explanatory factor for the difference in trade flows between countries by introducing the traditional gravity equation. However it took some time to for the model to be accepted by mainstream economists due to the fact that is was considered to be more of a physics analogy than a proper economic model (Head & Mayer, 2013). Anderson (1979) was next to set forth the legacy of the model by actually taking price differences across countries into account. However, it did not have much impact4. Many years later, in times when ‘the world was proclaimed to be flat5’, McCallum (1995) and Leamer and Levisohn (1995) spurred the discussion by introducing the ‘border effects6’. Anderson and van Wincoop (2003) were first to convince the general public with their gravity methodology. They suggest that a gravity equation takes into account multilateral resistance terms (MRTs) or fixed effects, as they argue that the traditional gravity model is wrongly specified7. The latter introduced importer and exporter fixed effects, creating a unit-income-elasticity model (Silva & Tenreyro, 2006). The MRTs are assumed low if a country is remote from world markets, due to surroundings of oceans or mountains (WTO, 2013). According to Anderson & van Wincoop (2003), border effects have an asymmetric effect on countries that are different in size, having a larger effect on small countries. Feenstra (2004) reviewed three methods to account for price effects in the gravity equation; price Indexes, the computational method and country fixed effects. Silva and Tenreyro (2006) compare the traditional gravity equation with the Anderson-van Wincoop gravity equation. They argue that in both models biases are present for which they propose a simple Poisson pseudo-maximum-likelihood method8.

The gravity equation of international trade model has been used to serve as a means for many different studies all wanting to explain the existence or absence of bilateral trade flows. Some argue that transportation costs are the most important cause of the effects of distance

4_{His theoretical foundation was ought to be too complicated.} 5

Reference to the bestseller: ‘The world is flat’.

6

The concept of border effects and McCallum’s theory will be explained in chapter 2.4 in this study.

7_{Their approach makes the parameters in a standard linear equation non-linear and therefore they are not}

able to use the standard OLS regression (Disdier & Head, 2008).

8

(12)

12

and use the model to prove this point (Limao & Venables, 2001)9. Others insert the communication and information costs into the model in order to account for capital or intangible goods (Portes & Rey, 2001).

The amount of studies and papers which incorporate some form of the gravity model into their paper is humongous. The meta-analysis by Disdier and Head (2008) summarizes and examines 1467 distance effects estimated in 103 papers. They find that on average a 10 percent increase in distance drops bilateral trade by about 9 percent10.

2.2 Globalization

The international economic playground has changed due to all kinds of forces such as interconnectedness that together can be summarized into a singular term labelled ‘globalisation’. Globalisation refers to all processes that eventually will lead to international integration. The opening up of trade and the fading of borders through international integration also implies that the production per variety can increase, as a larger market makes it profitable to expand the scale of production (Krugman, 1979). The globalization of the world economy in turn seems so have decreased the importance of physical11 distance (Buch, Kleinert, & Toubal, 2004). The publication of the book ‘The world is Flat’ by Thomas Friedman (2005) reflects on the increasing globalized world with greater similarity and homogeneity between people across nations, based on ten political, economic and technological phenomena. Friedman states that technology has reduced the costs of engaging in activities across geographical space, thereby ‘flattening’ the world and increasing the world´s income almost exponentially.

However, Ghemawat (2001) stresses the importance of distance as a barrier to

international trade. This can be either physical distance, which may be hampered or alleviated by geographical phenomena such as mountain ranges or easy access to good waterways, or political, cultural, or social distance, which influences trade through extra time and effort in order to successfully engage in international business (Brakman, Garretsen, & Marrewijk van, 2009); (Ghemawat, 2001). These four dimensions of distance will in turn be used and

9

In this study, I will make use of both the terms ‘transportation costs (or transport costs)’ and ‘trade costs’. Whereas the former just implies the costs that come with the physical transport of a good from point A to B, the latter entails all possible costs that are fuelled by all dimensions of distance specified in this study.

10

According to the meta-analysis, this negative impact on trade rose around the 1950s and has since than remained high (Disdier & Head, 2008).

11_{Throughout this entire paper, the terms ‘physical’ and ‘geographical’ distance are used to define the same}

(13)

13

investigated in this literature review in order to find evidence for a shift in the relative importance of one of these factors.

2.3 Physical Distance

Physical distance deals with both the natural barriers to trade (oceans, mountains, and distance in km for instance) and the cost of trade (transport, travel, and communication for example). This subdivision of distance mostly affects industries or products with low value-to-weight ratio, products that are fragile or industries and products in which communications are vital (Ghemawat, 2001). In general, it will be harder to conduct business with countries that are far way than with countries which whom you share a common border.

That is why transportation costs have to be part of the analysis of trade between countries; they are crucial in determining the balance between agglomeration and spreading forces (Brakman, Garretsen, & Marrewijk van, 2009). There would be no geography without transport costs, and the whole exercise of transforming economic models into geographical economics models would become pointless (Brakman, Garretsen, & Marrewijk van, 2009). Obstfeld and Rogoff (2001) argue that transport costs cause the distance effect entirely. The export of goods and services from one country to another involves time and effort, and hence costs. Goods have to be physically loaded and unloaded and transported by trucks for example, before even reaching their destination. Also, a distribution and maintenance network has to be established, and the exporter will have to familiarize him- or herself with the (legal) rules and procedures in another country, usually in another language and embedded in a different culture (Brakman, Garretsen, & Marrewijk van, 2009). All this involves costs, which tend to increase with distance’s geographical dimension, whether transported across land, sea or through air (Ghemawat, 2001).

(14)

14

Behar & Venables (2010) argue that the fall in trade costs can be considered to be modest due to three reasons: the importance of fuel costs, the fact that it is not clear whether technical advance in transport has been greater then technical progress in other areas and the fact that the innovation in transport has been mostly focused on quality rather than on cost improvements.

Hummels (2007) finds that countries that share a border are likely (90 percent) to use transportation modes that go over land than vehicles that use the air or the sea. Graph 4 below portrays the motor gasoline prices from 1949 to 2009 and shows that until 2004 the dollar price per gallon has fluctuated around 1.5 dollar per gallon12. However, from 2003 onwards a steep rise in the price of gasoline is depicted, which still remains the case today.

Source: US. Energy Information Administration (EIA)13 Countries that do not share a common border usually make use of transportation across the sea or through the air. In fact, Hummels (2007) states that there has been a sharp increase in the air transport over the years mostly owing to the plunge in the cost of air shipping and the time saving costs that come from a much quicker delivery than by shipping. These sharp increases are shown in graph 5 on the next page14.

12

Except for the period around the mid 1970s. This can be elucidated by the oil crisis of 1973 in which an oil embargo by the Organization of Arab Petroleum Exporting Countries (OAPEC) (U.S. Departement of State, 2013). The peak in the data can therefore by explained by this exogenous shock.

13

The data used to create the graph considers the motor gasoline prices. The precise amounts can be found in the appendix in table 1. The motor gasoline prices are portrayed, since calculations on different modes of transportation by U.S. and Latin American data suggest that roughly 90 per cent of trade with land neighbours is executed over land by trucks and trains for instance (Hummels, 2007).

14

The data used to create this graph is obtained from the data set made available on the website

http://www.krannert.purdue.edu/faculty/hummelsd/research/jep/data.html by D. Hummels. The original data stem from the International Civil Aviation Organization (ICAO).

0 500 1,000 1,500 2,000 2,500 3,000 3,500 Do llar p e r gall o n year

(15)

15

Source: (Hummels, 2007)

2.4 Political Distance

Political distance is another one of the ramifications of the cultural, administrative (political), geographical, and economical (CAGE) distance framework by Ghemawat (2001). According to this framework, the political dimension of distance between two countries should increases with: the absence of a shared monetary or political association, the existence of political hostilities, the presence of weak legal and financial institutions and certain government policies. Political distance or separation deals with the movement of goods (customs), people (immigration), Money (currency exchange), capital (regulation, taxation) and ideas (censorship, firewalls). In a well-known article, McCallum (1995) presents a border puzzle in which Canadian provinces trade 22 times as much with each other than U.S. states and Canadian provinces. Anderson and Wincoop (2003), even though not convinced by McCallum’s puzzle, still find that borders effects are responsible for a reduction of bilateral trade levels. However, they estimated that border effects have a larger effect on countries that are smaller, because of an asymmetric effect (Feenstra, 2002).

Governments have invented all sorts of barriers to protect their domestic markets from foreign competition. This protectionism comes in the form of tariffs, quotas, restrictions on Foreign Direct Investment (FDI), and subsidizing and favouring domestic competitors (Ghemawhat, 2001). Due to the global opening up of economies and the tendency towards trade liberalization, some of the initial barriers such as quota’s or tariffs have completely faded. This specifically happened within regions of countries that signed a trade agreement that promotes free trade such as in the North American Free Trade Agreement (NAFTA) between the United States of America (USA), Canada and Mexico (Venables & Limao, 2002). International trade negotiations have slowly reduced tariff rates over the years, with import tariffs of the USA dropping to 25 percent of their original value of 6 percent in 1950 for example (Hummels, 2007). Clemens and Williamson (2002) claim that import tariffs

0 50000 100000 150000 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2004 To n /km (i n m ill io n s) year

(16)

16

worldwide have plunged from 8.6 to 3.2 percent in the period 1960 until 1995. Graph 6 below shows the proclaimed drop in tariffs using aggregate world data on import tariff rates15. These aggregate values show a larger variation in applied tariff rates, however the message remains the same; tariff rates have been dropping over the years16.

Source: World Bank The WTO (2014) received 585 notifications of Regional Trade Agreements (RTAs) of which 379 were in force. These RTAs can be considered to be reciprocal trade agreements between two or more partners. Preferential Trade arrangements (PTAs) are unilateral trade preferences. There tend to be a sharp increase in the number of RTAs that go into force in the period from 1958 until present (see graph 717).

Source: WTO

15_{The word aggregate import tariff rates dataset used for this graph represents the average of effectively}

applied rates weighted by the product import shares corresponding to each partner country.

16

The sharp decline in import tariffs shown in table 6 might be the aftermath of the foundation of the WTO in 1995 and its trade liberalizing policies.

17

Graph 7 shows the number of RTAs that actually went into force at each particular year. 3 5 7 9 11 13 15 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 A gg re gate wo rl d im p o rt tar iff rate s Year

Graph 6. Aggregate world import tariff rates, 1988-2012

0 5 10 15 20 25 1958 1961 1971 1976 1981 1985 1988 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013

(17)

17

Next to bilateral and unilateral measures that directly influence trust, there are political factors that might alter bilateral interactions unnoticed. Yu (2009) finds that the a higher democratization of a exporter’s economy ensures stronger institutions regarding product regulations for instance which in turn foster trust from the international community. Bénassy-Quéré et al. (2007) focus on the influence of domestic institutions as a key explanation of higher propensity of countries due to for instance the avoidance of extra costs that can arise from corruption or uncertainty. Ghemawat (2001) finds that political hostilities between countries may lead to a complete absence of trade all together even though the countries maybe sharing a border.

2.5 Economic distance

Physical and geographical dimensions of distance are not the only factors affecting trade that one should focus on according to Ghemawat (2001). Economic factors do play an important role as well. The world’s income has been booming since the last century and except for some major dips due to crises, the world shows GDP growth rates that are unseen throughout history. Graph 2 shows the annual world GDP growth rates over the period after the mid 1950s until present18.

Source: World Bank

18_{The extreme values such as in 1974, 1982 and 2009 stem from exogenous shocks such as the OPEC oil price}

shock, debt shocks and the credit crunch and should therefore not be considered in this study. -3 -2 -1 0 1 2 3 4 5 6 7 8 19 60 19 62 19 64 19 66 19 68 19 70 19 72 19 74 19 76 19 78 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06 20 08 20 10 20 12

(18)

18

Friedman states that technology has reduced the costs of engaging in activities across geographical space, thereby ‘flattening’ the world and increasing the world´s income almost exponentially as portrayed in graph 3 below.

Source: World Bank However, this huge increase in world income sadly comes with both winners and losers. Whereas world GDP growth seems too good to be true, GDP per capita in many countries is often lacking behind. The income of consumers of particular economies remains the most import aspect of the economical dimension of distance between countries as richer countries are ought to trade more with other rich countries (Ghemawat, 2001).

2.6 Cultural distance

That differences in culture can create a distance between economies is no rocket science. Deviations in social norms, values and beliefs could easily drive a wedge between countries (Ghemawat, 2001). Cultural distance also shows in the way that consumers chose specific products and have different preferences towards specific features.

Grossman (1996) finds that not only transport costs but unfamiliarity or cultural differences can explain the negative relationship between geographic distances and trade as well. Huang (2007) finds that ´high uncertainty-aversion countries, compared to low uncertainty-countries, trade disproportionately more with close neighbours and less with distant partners than what standard gravity models predict’.

19 60 19 62 19 64 19 66 19 68 19 70 19 72 19 74 19 76 19 78 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06 20 08 20 10 20 12 GDP (c u rr e n t tr ill io n s o f US $) year

(19)

19

2.7 Comparing different components

Many have agreed on the impeccable effect of distance on trade. However, every research on the topic of the explanation of trade flows between country-pairs includes different variables into its analysis to eventually conclude a slightly or complete different thing. What is most interesting to this study is to what extent they compare all components of distance to trade over time. Anderson and van Wincoop (2004) compare the relative importance of a whole bunch of elements that arise when engaging in international trade. They compare freight costs with transportation costs (not only freight costs, but also time costs), policy barriers (tariffs and non-tariff barriers), information costs, contract enforcement costs, costs associated with the use of different currencies, legal and regulatory costs, and local distribution costs. They find that trade costs break down in approximately 21 percent transportation costs and 44 percent border-related trade barriers. This means that they find evidence for the importance of border related costs compared to physical distance related costs. However, this finding is an aggregate finding and the study lacks comparison of the relative impact of the elements on trade over time.

Venables and Limao (2002) find that transport costs are very important in the ability of countries to participate in the global economy. The tendency of trade liberalization makes transportation costs as such somewhat more important that the official barriers, such as tariffs. Baier (2001) investigate the bilateral trade flow between 16 OECD countries in order to find the main reason behind world trade growth. They find that the relative contribution of trade liberalization was three times that of transport costs, meaning that a decline in political costs assured the growth in trade and therefore find evidence for the importance of border related costs compared to physical distance related costs.

In a seminal paper that looks at price volatility, Engel and Rogers (1996) showed that the dispersion of prices of similar goods increases with the distance between city pairs, a pattern that holds even within a country due to transport costs. Crucini, Telmer, and Zachariadis (1999) provide an interesting recent twist based on a large cross section of goods prices in European capital cities in 1985. In other words, the equally weighted average of goods prices in local currencies between two European cities, say, Paris and Bonn, is a good predictor of the nominal exchange rate in that year. This suggests that markets (in Europe at least) may, in fact, be more integrated, and borders may matter less than studies examining the variability of price differences would suggest. (Parsley & Wei, 2000).

(20)

20

dimensions of distance that make up ´trade costs´ has been performed over time. Studies include part of the distance dimensions, but mostly leave components out of their estimation. Also, the existing comparisons in the literature that come close to my research topic have all been lacking to include a comparison over time. This study is an attempt to fill that gap. The study involves as many country-pairs and as many years for which data is available.

3. THEORY AND MODEL

3.1 hypotheses

If one were to theoretically unfold aggregate trade costs, one would come across a huge range of costs items such as informational costs or language barriers that all might have an individual impact on trade. This chapter tries to combine literature and empirics in order to construe a set of hypotheses that underlie the research question. Hypotheses 1 until 5 represent the core assumptions of this study, whereas hypotheses 6 until 9 are included to give rise to the many control variables the standard gravity model possesses.

3.1.1 Physical Distance

Geographical distance is not only about how far away a country is in exit measurements. It has also to do with the topography, the size of the country and its transportation and communication infrastructures (Ghemawhat, 2001).

However, since good direct measures of distance costs are unavailable, geographic distance between countries is used in many studies in the gravity equations (Buch, Kleinert, & Toubal, 2004). Often, the distance is measured between two points which are usually the capital or largest cities of each country. However, sometimes aggregated measures of several cities are used (Behar & Venables, 2010). One way of measuring transport costs is by taking the cif/fob ratio, which is the ratio of data of free on board (fob) provided by the exporting country and the costs of insurance and freights (cif), provided by the importing country (Limao & Venables, 2001). Ghemawat (2001) found that the amount of trade between economies that are 1000 miles apart from each other is five times as much as economies five times as far apart from each other. Also, the quality of transportation services has improved sharply, which yields greater speed and reliability (Hummels, 2007).

(21)

21

not the only aspect to be considered19, they are the most practical and general measure to calculate trade costs that arise from moving products or services across space. However, there is still no consensus on the exact proxy for geographical distance (Huang, 2007)20. I therefore hypothesize:

Hypothesis 1:

Due to technical changes and innovations physical distance has become less important in explaining bilateral interactions through trade costs over time.

3.1.2 Political Distance

Where distance is a bilateral variable, political distance could be both a national specific variable (import tariffs) and a bilateral variable (trade regions). Behar and Venables (2010) extents the gravity model by including a variable called trade facilitation. Wilson et all. (2005) split up trade facilitation into four separate features: port facilities, customs handling, the regulatory environment and the availability of service sector infrastructure. In this study, the average import tariffs of all products per year, the membership of trade organizations/regions and the legal origin of countries will be analyzed to further investigate the importance of political distance for trade.

3.1.2.1 Tariffs

According to Hummels (2007), tariffs pose at least as large a barrier to trade as transportation costs do. However, tariff rates have been decreasing since 1950 with an

average world drop of 5.4 per cent between 1960 and 1995. Therefore, as tariffs are becoming less of a burden, transportation costs are becoming relatively more importance according to Hummels (2007). As stated before, Clemens and Williamson (2002) claim that import tariffs worldwide have plunged from 8.6 to 3.2 percent worldwide in the period 1960 until 1995. Graph 6 (see literature review) shows this drop. I therefore hypothesize:

Hypothesis 2a:

The lowering of the average world tariff rate has decreased the relative importance of political distance over time.

19

Other factors such as water access, the within-distance in the economy itself and landscape must be

considered as well (Ghemawat, 2001). These influencers will be explained and used in the empirical part of this paper.

20

(22)

22

3.1.2.2 Trade agreements & regions

Regional Trade Agreements (RTAs) remove barriers to trade and favour the country in the agreement over one that is not. Some RTAs that are commonly known are the ASEAN, the NAFTA and the EU. Since the RTAs are intended to boost trade and RTAs are currently everywhere, one should expect that:

Hypothesis 3:

The existence of regional trade agreements between two countries has become more important over time.

Ghemawat (2001) finds that economies that share a common membership in a regional trading bloc have an on average increase in bilateral trade flows of 330%. Tomz et al. (2007) find that if all GATT participants are rightfully accounted for, participation in the trade organization substantially increases trade. The WTO promotes cooperation, peace and prosperity among its members through transparency, quick dispute settlement and the reliance on rules rather than power. I therefore hypothesize:

Hypothesis 4:

The existence of a membership of the GATT/WTO of both countries in the country-pair has become more important over time.

3.1.2.3 Legal Origin

Legal origins theory is a point of view in which people link the country’s current economic state of development to its legal system. Some economists have classified countries on whether they adhere to common law or whether their legal system is based on French civil law, German civil law or Scandinavian civil law. Sharing a common legal origin will therefore enhance trade. I therefore hypothesize:

Hypothesis 5:

Sharing a common legal origin will increase the bilateral trade flows within a country-pair and will become more important over time.

3.1.3 Economical distance 3.1.3.1 Common Currency

According to the CAGE distance framework (Ghemawat, 2001), sharing a common currency increases bilateral trade by 340 per cent. Rose (2004) finds a positive correlation between the two variables as well. I therefore hypothesize:

Hypothesis 6:

(23)

23

3.1.4 Cultural distance 3.1.4.1 Common language

In Ghemawat (2001) ´s CAGE distance framework, the cultural dimension of distance states that countries that share a common language trade three times as more with each other than countries that lack a shared language. I therefore hypothesize:

Hypothesis 7:

Sharing a common language increases bilateral trade flows within a country-pair.

3.1.4.2 Common Colonizer post WWII

History is assumed to play a role in bilateral interactions (Eichengreen & Irwin, 1995). Sharing a common colonizer, increases trade in a country-pair due to having to overcome less information costs (WTO, 2013). This has been empirically proven by Anderson and van Wincoop (2003) and Rose (2004)21. I therefore hypothesize:

Hypothesis 8:

Sharing a common colonizer increases bilateral trade flows within a country-pair.

3.1.5 Set of sub-hypotheses

The gravity model usually consists of a comprehensive set of control variables (Huang, 2007). Sharing a common language for instance can be used to capture information costs (WTO, 2013). In order to create a model that is as complete as possible, a dummy variable on sharing a common border has been included.

3.1.5.1 Common border

According to Hummels (2007), sharing a common land border explains 23 percent of world trade. This is not surprising, ´as countries that are close to each other tend to learn each other’s language, tend to have direct interaction between their citizens for tourism or business, or because of better media coverage´ (Portes, Rey, & Oh, 2001). I therefore hypothesize: Hypothesis 9:

Sharing a common border increases bilateral trade flows within a country-pair.

3.2 Gravity model of International trade

This study is based on the approach of ‘the gravity model of international trade. A `demographic gravitation’ model of interactions between two locations was first developed by James Q. Stewart. By analogy with the Newtonian gravity model, he finds strong correlations for traffic, migration, and communication between two places, based on the product of their

21

(24)

24

population size and inversely related to their distance squared (Anderson J. , 2011). Jan Tinbergen (1962) was first to introduce a specific version of this model to the theory on bilateral interactions. This became known as ‘the gravity model’ in international economics and is being used as the backbone of the model used in this study and many others that investigate bilateral interactions. The gravity model relates trade to income and distance, the way Newton´s law of gravity is used in physics. Or as to put it differently; export values increase proportionately with the size of the destinations’ economy, which can be compared to the way that planets are attracted to each other proportionally to their size and proximity and hence the ‘gravity’ in this particular model of trade (Head & Mayer, 2013). However, unlike Newton’s equation, it was chosen in this study because it can be regarded as the standard for measuring the variables that explain international trade flows (Behar & Venables, 2010). It is used in many empirical research on the topic (Huang, 2007); (Leamer & Levinsohn, 1995); (Egger & Pfaffermayr, 2003) and comes in a broad spectrum of different interpretations. However, a standard specification of the model has been agreed on. Below two different standards of the model are portrayed: an econometricians perspective and a economists one.

A typical Gravity specification by econometricians (Mátyás, 1997):

(1)

(25)

25

A typical specification of the model by economists:

(2)

Where is the volume of trade measured as exports from country i to country j at time t. is the GDP of country i at time t. The same accounts for , which is the GDP of country j at time t. is the distance between country i and j. The equation is typically estimated in log linear form, such that can be interpreted as the elasticity of trade with respect to distance. Distance is a bilateral variable but most other variables, for example the landlocked dummy, are country specific (Behar & Venables, 2010).

3.2.1 Model

In order to be able to estimate the effects of both political and physical distance on international trade, I will make use of a standard version of the gravity model of bilateral trade as explained above, with the extension of specific variables that represent the levels of distance. Extraneous factors that might affect trade are included, a set of control variables has been included that capture: culture (sharing a common language), history (sharing a common colonizer, geography (sharing a common border) and economics (sharing a common currency) (Rose, Do We Really Know that the WTO Increases Trade?, 2004). The following specification of the model has been construed:

(3)

Where i denotes the country of origin, j denotes the country of destination, t symbolizes time in years, and the variables:

denotes the export value of goods by country i from country j at time t,

stands for real GDP at time t,

(26)

26

denotes the weighted distance between countries i and j in kilometres and functions as a proxy for physical distance in this study,

represents the area of the economies in square kilometers,

symbolizes the simple mean applied tariff and is the unweighted average of effectively applied rates for all products subject to tariffs calculated for all traded goods by the exporting country at time t,

stand for two dummy variables which equal 1 if the country of origin and the country of destination are GATT/WTO members at time t,

represents a dummy variable which equals 1 if there is a regional trade agreement at force between t,

he country-pair i and j at time t,

is a dummy variable that equals 1 if the country-pair i and j share a common legal origin,

is a dummy variable that equals 1 if the country-pair i and j share a common border,

is a dummy variable that equals 1 if there exist a common primary language between the country-pair i and j at time t,

is a dummy variable that equals 1 if nine percent of the population of both country i and j speak the same language at time t,

is a dummy variable that equals 1 if there exist a common currency between the country-pair i and j at time t,

is a dummy variable that equals 1 if the countries within the country-pair share a common colonizer post WWII at time t,

is a dummy variable that equals 1 if the country of origin is at war at time t, represents the time difference between country i and country j,

denotes the omitted possible other influences on the trade flow between i and j.

4. DATA AND METHODS

4.1 Data Collection

(27)

27

GDP of the economies in the sample is collected from the Centre d'Etudes Prospectives et d'Informations Internationales (CEPII). This French research centre also provides data on the weighted distance between sets of country-pairs. The data on population is obtained from the Penn World Table (version 7.122) and has been used in order to create the GDP per capita values. The Correlates of War project (COW) dataset has been used to provide data on the area of economies in square kilometers. I used the World Bank database for statistics on the average import tariff rates. The World Trade Organisation (WTO) database has been used to accrue data on the existence of possible WTO membership, for which I created a binary variable. The gravity model usually consists of a comprehensive set of control variables (Huang, 2007). Control variables are added to the gravity equation because they are thought to affect trade not through transport costs but otherwise (Behar & Venables, 2010). There are a lot of geographical variables that have influence on the cross-country variation of trade flows. Having a common border, having a larger country area, being an island, these are all positive influencers on bilateral trade flows (Behar & Venables, 2010). I exploit the COW dataset to obtain a number of country-specific variables. These include cultural (language), historical (colonizers) and geographical (landlocked) variables23.

The data on transport modes stem from the database by David Hummels. He published the data on air or sea transport values on his website.

4.2 Methods

The empirical research in this paper will be done by making use of a panel dataset which consist of a cross-sectional units (country-pairs) who will be observed over time (years). The estimation of the gravity model in this paper covers unique NUMBER unique country pairs from the year 1948 until 2006. The number of cross-sectional units in this study can be considered large and the time dimension wide (Hill, Griffiths, & Lim, 2007). The panel data accounts for individual heterogeneity since in allows for variables to change over time but not across entities. Since the purpose of this study is to find possible changes in importance of factors on bilateral trade flows, year dummies have been created so that different periods can be examined. In this study I will make use of all the data available for all countries there is available. This means that, after deleting countries for which almost no data was available, the COW database still covered bilateral trade data for 224 unique

22

Version 7.1 of the Penn World Table has been used, because it contains more observations than the existing sequence.

23

(28)

28

countries/economies (see table 1 in the appendix for the exact countries). The statistical program STATA will be used as an executive for this research, because Windows Excel does simply lack space for the enormous size of the sample.

Since I work with different databases, files and statistics had to be merged in order to empirically perform research. In order to be able to add extra variables into the original file, the ISO-country code variables had to be encoded, because STATA is not able to program words. Also, the ISO-country Codes tend to differ slightly in almost every single database. They have been adjusted to the most commonly used codes (see table 1 in appendix). These standardized ISO-codes than had to be matched with the IMF numerical ISO codes in the DOT database.

After deleting all trade regions and continents and all export values of zero’s, DOT database had to be merged with my COW database in order for an OLS to take place. After the cleaning up of the DOT dataset, only 192 unique country codes remain (see table 7 in the appendix). This means that in order for the two databases to merge properly, I have to delete over 30 countries. After merging several datasets, the panel data set had to be declared a panel data set in STATA. However, some duplicates existed. So in order to start my estimation, duplicates had to be found and deleted. The variables in the dataset have been transformed to their natural logarithms so that a log-linear model has been created. The log-linear specification makes it possible to interpret the parameters as being elasticities. In order to estimate the gravity model I will make use of ordinary least squares (OLS) regression, which is possible due to the linearity of the parameters in equation 3 above (Disdier & Head, 2008) ; (Rose, 2004).

Since I work with a panel dataset, random effects and fixed effects estimators will be used in order to check for the robustness (Rose, 2004); (Adkins & Carter Hill, 2008). However, it would be make sense to make use of the fixed effects model because it allows individual intercepts to be included to ‘control’ for country specific differences (Hill, Griffiths, & Lim, 2007).

Table 1 below portrays the descriptive statistics on the final panel dataset for this study. After all the adjustments that had to be made 17672 unique country-pairs remain. This means that a total of 327037 observations (unique-country pairs x years available for each country-pair) remain. This is a good thing, because it might overrule the missing data bias that comes with panel data research. However, the fact that there are so many observations makes it very time consuming to change one single variable in my main dataset.

(29)

29

dyad: 1112, 1122, ..., 148963 n = 17672 year: 1948, 1949, ..., 2006 T = 59

Delta(year) = 1 unit Span(year) = 59 periods

(dyad*year uniquely identifies each observation)

4.2.1. Caveats

The literature on exploring bilateral trade flows by using the gravity equation has come across the problem of the so called zeros. This entails the countries that do not trade with each other at all and therefore have a dependent variable that equals zero. In other words, “many potential bilateral trade flows are not active” (Anderson J. , 2011). I choose to leave those country-pairs out of this paper, since former research has proven that including them does not significantly influence the results (Silva & Tenreyro, 2006)24. However, this does not imply that it is not of an important matter and that it might be an interesting topic for future research on the matter.

Heteroskedasticity is a looming issue that might arise when estimating log-linearized models by OLS (Silva & Tenreyro, 2006). I will perform several tests in STATA to make sure that heteroskedasticity might not bias the results in this study.

5. EMPIRICAL RESULTS

5.1 Descriptive statistics

Table 7 in the appendix contains the summary statistics for all variables used in the estimation. The table summarizes cross-section and time series data by providing the standard deviations for all standard variables on the overall, between and within level. Some variables show a within standard deviation of zero. This means that there exist no variation within the observations on the individual dyad ids and that they will be eliminated by the fixed effects transformation due to their time invariance25.

As discussed before, the size of an economy has a positive influence on the value of exports. This is a classic result and robust result that stems from every standard specification of the gravity model of international trade. Graph 8 below shows how this finding applies to my construed dataset as well. It portrays the relationship between the GDP of the importer and the export value for four country-pairs over time.

24_{There has been a variety of non-linear methods that been created for this zero problem. Helpman et al.}

(2006) derives an estimation via maximum likelihood.

25

(30)

30

Graph 8. Export value and GDP of importer for four country-pairs

5.2 Interpretation of estimation

The behaviour of the observations in my dataset is observed across time. There are commonly two techniques that are used in order to analyse panel data: fixed effects or random effects. However, first, I performed an OLS regression of which the estimation results are shown in table 8 in the appendix. This particular regression model only displays the basic variables that are part of the most standard gravity model. It seems that except for the GDP per capita of the country that exports all variables are significant at the one percent level. It confirms the standard results; the GDPs and the area of the economies have a positive impact on the export value whereas the weighted geographical distance between them shows a negative impact. The model, however, seem to negatively relate the size of the area to trade.

Table 9 in the appendix shows the results for a least square dummy variable model (LSDV). The overall fit of this model can be considered to be relatively high (R-squared of 0.7035) and F-test is zero, which confirms the overall strength of the gravity equation. However, this time more variables are not significant. This could be the case due to the influence of the variable conflict, which has been dropped for the second part of the analysis.

Table 10 in the appendix portrays the estimation results by making use of a random effects model. The differences across the units are uncorrelated with the regressors, however the

(31)

31

Wald Chi variable seems to be huge. This time, almost all variables that are not included in the very basic gravity equation seem to be insignificant at the one percent level. Table 2 below portrays the estimation results and the R-squared of all three models used in this estimation. The LSDV model shows the strongest fit and the most variables which are significant at a very small percentage. The only important factor that remains insignificant in this particular model is the average import tariffs rate.

Table 2. Comparing the OLS, the LSDV and the random effects model ---

However, I have to be careful with the interpretation of the coefficients, because random effects estimations include both the within-entity and the between- entity effects.

(32)

32

Table 3. Hausman test

---- Coefficients ----

| (b) (B) (b-B) sqrt(diag(V_b-V_B)) | fixed random Difference S.E.

---+--- lgdp_expor~r | -.9578874 .38908 -1.346967 .2689735 lgdp_impor~r | 1.941376 .4105702 1.530806 .2363536 lgdpcap_ex~r | 1.220353 .1964935 1.02386 .2748468 lgdpcap_im~r | -1.098382 .1915064 -1.289888 .2297846 ltariff | -.0560231 -.0475442 -.0084789 .0180156 1.rta | .3856213 .3844525 .0011688 . 1.gatt_exp~r | .1966394 .1515857 .0450537 .0253237 1.gatt_imp~r | -.0154366 .0022303 -.0176669 . 1.comcur | .2157682 .0899603 .1258079 .0433987 --- b = consistent under Ho and Ha; obtained from xtreg B = inconsistent under Ha, efficient under Ho; obtained from xtreg Test: Ho: difference in coefficients not systematic

chi2(9) = (b-B)'[(V_b-V_B)^(-1)](b-B) = 103.27

Prob>chi2 = 0.0000

(V_b-V_B is not positive definite)

For the second part of my research, I will use the LSDV model and drop the variable conflict as it has a tremendous biased effect on my results due to missing values.

So far, even though both models represent a completely different estimation, one thing that is clear is that trade is indeed proportional to size and that is inversely proportional to distance. Nothing new under the sun.

5.3 Comparison of coefficients over time

(33)

33

Table 4. Comparison of coefficients over three periods

--- Variable | period1 period2 period3 ---+--- lgdp_expor~r | .77058701*** .92367345*** 1.2646835*** lgdp_impor~r | .69680284*** .81928721*** .99166212*** lgdpcap_ex~r | .04112639*** .07380844*** -.07543555*** lgdpcap_im~r | -.08309499*** -.07192655*** -.14782295*** ldistw | -.58365412*** -.95988691*** -1.0932716*** larea_expo~r | -.0499958*** -.09034457*** -.13626455*** larea_impo~r | -.11947512*** -.17977043*** -.18519456*** ltariff | -.00053667 -.2280165*** .04446391*** contig | .39979693*** .63535303*** 1.1277135*** comlang_off | -.02304418 .08662*** .40818913*** comcol | .24016033*** .14840629*** .62677873*** colony | .84677405*** 1.624776*** 1.4191871*** rta | .81940082*** .85721096*** .91159619*** gatt_expor~r | .15361586*** -.16567356*** -.14456595*** gatt_impor~r | .17524318*** -.04845186** -.19090671*** comleg | .29565629*** .24150229*** .37378332*** comcur | .65607614*** 1.0295364*** .46123207*** _cons | 9.1296953*** 9.7494571*** 6.2333863*** ---+--- r2 | .530771 .52395633 .59233712 --- legend: * p<0.05; ** p<0.01; *** p<0.001

There are a couple of conclusions that can be drawn from the data in table 4. First, in period 1 the most important positive influences on the bilateral interactions are sharing a regional trade agreement, sharing a colonial history and the GDP of the exporter and the importer. The most important negative influence on the bilateral interactions is physical distance. In period 1, tariffs seem to be insignificant. This can be explained due to a lack of appropriate import data on that period.

In period 2 the most important positive influences on bilateral interactions are sharing a common currency, sharing a colonial history and sharing a regional trade agreement. The most important negative influences this time are tariffs and distance. A couple of important things happen in the transition from period 1 to period 2. The significance of tariffs and its huge negative impact on trade are tremendous. This can be correct, because period 2 is known for its protectionism of domestic markets. Also the sharp increase of the negative impact of both geographical distance and the positive impact of sharing a common border are tremendous compared to period 1.

(34)

34

whereas in period 2 tariffs had a huge negative impact on trade, in period 3 it seems that import tariffs now stimulate trade slightly. The importance of sharing a common currency now almost dropped in its significant impact.

As far as the hypotheses were predicting various options of distance impact on trade, it can be established that hypothesis 1 can be rejected. Whereas innovation indeed might have taken place it seems that spatial transactions costs have not fallen over recent years, but instead they increased. If we look at the different effects that impacted trade, we see a steady increase in the negative impact that physical distance has on trade. This finding is also backed up by the fact that sharing a common border becomes more and more stimulating to trade over the years. One explanation is that during the second wave of globalization (1960s onwards), the drop in communication and coordination costs made the transport of goods relatively more expensive however. Hummels (2007) also noticed that the proclaimed drop in shipping costs due to innovation and technological developments seems lacking in the documentation of recent years. Hypothesis 2 and 3 can be accepted, stating that the effect of tariffs has decreased and the impact of regional trade agreements has increased. The data show an interesting fact, namely that tariffs might actually have a slight impact on trade in period 3. Being member of the WTO actually has a negative impact on trade according to my dataset. Therefore I have to reject hypothesis 4. Probably due to measurement errors or wrongly specified cells, the variable has gotten a negative sign. Hypothesis 5 can be accepted. Sharing a legal origin indeed has a positive effect on trade that increases over time. This can be explained by the fact that sharing a legal origin makes it a lot easier to conduct business. Hypothesis 6 has to be rejected. It seems that sharing a currency has a positive influence on trade, however this effect decreased over time. This could be due to the fact that nowadays it is a lot easier to exchange domestic currency against foreign currency. Both the positive effect of sharing of a common language and the sharing of a common colonizer and a colonial history seem to sharply increase over time. Therefore hypotheses 8 and 9 can be sustained. Hypothesis 10 can be accepted. However, the remarkable increase of the positive impact on trade by sharing a common border is humongous. Overall, I can almost accept all hypotheses.

Investigating the impact of four dimensions of distance on bilateral interactions over time: Who trades with whom and why?