Master Thesis
The influence of transport infrastructure on
international trade
MSc. International Economics & Business
August 2008
Author: Research Supervisor:
Desislava P. Cvetkova Dr. Ger J. Lanjouw
Stud. No. S1737309 Faculty of Economics University of Groningen Co-assessor:
The Netherlands Dr. E.H. Van Leeuwen
d.p.cvetkova@student.rug.nl
ABSTRACT
TABLE OF CO TE TS 1. Introduction……….…...4 2. Literature review……….………..………...6 2.1. Transport costs……….……....6 2.2. Gravity model………..9 3. Methodology……….…. …...…….11 3.1. Sample description………...11 3.2. Data sources………...14 3.3. Model specification………14 4. Statistical analysis………..……….…....19 4.1. Descriptive statistics………...19 4.2. Diagnostic checks………..20
4.3. Testing the model………..22
5. Model estimation…………....………....23
6. Results……….………...25
7. Conclusion and discussion……….26
1. Introduction
The role of transport infrastructure in economic life draws the attention of many researchers. A pioneer in that area is Aschauer (1989) who tests the relationship between the so-called “core infrastructure” (streets, highways, airports, electrical and gas facilities, water systems and sewers) and economic growth in the USA for the period 1949-1985. But the real importance of transport infrastructure is difficult to be determined because besides its direct effects on a country’s economic development there are many indirect effects which cannot be measured. For example, better infrastructure leads to lower transport costs and better economic performance. According to the tax policy of a country, this better economic performance can lead to higher tax revenues and as a consequence, to greater amount of money in the government budget. As a result, this amount of money can be used for improving health care or reducing poverty. That is the reason why the World Bank put such great efforts to improve the existing transport infrastructure in the developing countries. Furthermore, the increasing world population increases the necessity of adequate transport infrastructure in order to satisfy the growing needs of people for trade and traveling.
Transport infrastructure is also a key element of the Lisbon strategy for the development of the European Union. The aim of this strategy is to make the European Union the most competitive knowledge-based economy in the world (European Commission, 2005). For this purpose, a Trans-European Transport Network should be developed which has to reduce transport costs in the region and as a result to improve countries’ competitiveness.
time. On the other hand, all countries from the region are less developed than Western European countries. So for these countries improving their transport infrastructure and in that way improving their economic indicators has an essential role to catch up with the other countries from the European Union and to reach real convergence in the whole region. So the research question of this paper will be as follows:
“What is the influence of transport infrastructure on trade flows for the countries from Eastern and South-Eastern Europe?”
On the basis of the research question, several subquestions can be derived. They are:
1. What is the influence of road transport infrastructure on Eastern European trade flows? 2. What is the influence of railway transport infrastructure on Eastern European trade
flows?
3. What is the influence of air transport infrastructure on Eastern European trade flows?
In order to answer these questions I will use panel data for 35 countries from the European region. These countries are the 27 members of the European Union plus Albania, Bosnia and Herzegovina, Croatia, Macedonia, Serbia, Norway, Switzerland and Turkey. Including all countries from the region allows us to make a comparison between countries with good and countries with poor infrastructure and their international trade performance. Moreover, for international trade the available infrastructure in the whole region is more important than the picture for a certain country. The estimated period includes eight years, from 1998 till 2005. Data before 1998 are not available for all Eastern European countries. Furthermore, the economic situation in those countries was very uncertain. For example, some Eastern European countries suffered hyperinflation, Yugoslavia in 1993, Poland between 1990-1993, Bulgaria in 1997 (www.imf.org ). Moreover, the wars which started at the beginning of 1991 lead to the formation of several new countries from the territory of former Yugoslavia which are also included in the sample. They are Bosnia and Herzegovina, Croatia, Macedonia and Serbia. As a result of this research I would like to see how transport infrastructure affects foreign trade in Europe. I expect to find a positive relationship between these two variables. Furthermore, I expect that transport infrastructure will have a bigger effect on Eastern European exports because I assume that its influence is stronger at the beginning and after some sufficient level its importance starts to decrease.
countries from the region. Furthermore, aggregate data about export flows are used. But transport infrastructure has a different influence on different industries.
The rest of this paper is organized as follows. The first part represents an overview of the relevant literature. The following sections are devoted on the methodology of the paper and they include sample description, data sources, as well as model specification and estimation. Next, results will be represented and discussed. The last section consists of conclusion and discussion.
2. Literature review 2.1. Transport costs
exports to the foreign country but on the way back they are empty. As a result of this limitation, the Herberg model is relevant only for a small part of transport services such as pipe lines. The major part of transport services which includes transportation of goods in both directions such as transportation by ships, trucks or trains cannot be explained in this way. The next model described by Casas is the so-called “Falvey model”. This model is similar to the previous one but it improves it as assuming that market conditions are leading for choosing which country to provide the transportation of goods. The other important assumption is that transport technology is the same for both countries. At the end, Casas develops a model in which market conditions and the level of technology in both countries determine their participation in the transportation services.
Nowadays, Hummels (2007) gives an explicit explanation of the importance of transport costs for the development of international trade.In his paper “Transportation Costs and International Trade in the Second Era of Globalization”, he explains the increase in international trade in recent years with the decrease in transport costs due to improved transport infrastructure and technological innovations. He gives as an example the introduction of containerization which lead to the ability of using several kinds of transport modes without the necessity of packing and unpacking several times and in this way not only reduced transport costs but also the time for transportation. Hummels compares transport costs with tariff barriers but emphasizes that in contrast to tariffs which importance is reduced nowadays as a result of the creation of regional trade agreements, the importance of transport costs increases. In his paper, Hummels focuses more on sea and air transport costs and as a result his conclusions are based on these kinds of transport modes. One of the limitations of his paper is that he uses data only for New Zealand and the United States. But as he explains, usually transport costs are measured as the cost of shipping relative to the price of the good but most countries do not give sufficient data about shipping costs. He also emphasizes that transport costs play a vital role in relative prices between exporters and as a result determine the trading partners. That is why countries trade primarily with neighbours. But the improvement of air and sea transport infrastructure reduces this tendency. He concludes that transport costs depend on the distance on which the goods have to be transported, the quality of transport infrastructure and the quantity of transported goods.
Sankaran and Wood (2007) estimate the effect of traffic congestion on the supply chains. In their analysis, based on New Zealand’s experience, they conclude that traffic congestion leads to an increase in distribution costs. They point out two possible reasons for this outcome. The first one is that traffic congestion leads to a delay in supply. At the same time, customers always want to receive their goods on time. As a result, logistic companies have to increase the number of vehicles which they use for transportation as well as the length of the working day. The other explanation for the increase in distribution costs is retailers’ policy. Nowadays, most retailers prefer small lots of goods but more frequent orders because in that way they are able to reduce their expenditures for storage. On the other hand, fewer transported goods but on shorter intervals accelerate congestion.
There is no doubt that the leading factor for traffic congestion is the number of passenger cars. This assertion is confirmed by Goodwin (2004). He also states that traffic congestion leads to a change in the chosen transport mode for the consignors. He gives as an example the United Kingdom where congestion makes railway freight transport more desirable for some consignors. He illustrates the so-called “speed-flow” curve, according to which the increase in traffic leads to the decrease in speed. As a result, the governments of the countries try to stimulate the use of alternative ways for traveling such as using public transport or cycling. An interesting solution to the problem is suggested by the Dutch experience. The Dutch transport network is characterized by very good quality and as a result, it is one of the densest networks in Europe. As Bovy (1998) states, it is not profitable to build a good transport infrastructure if it is not used very often. So traffic congestion until a certain level can be a consequence of a good government policy which is the case with the Netherlands. The major negative effect of the congestion for the economy concerns road freight. As a result, the Dutch government concentrates its efforts mainly on reducing traffic congestion for the economically efficient road users. This group of users is called “target group” and it includes freight traffic and business travelers. For this purpose, the government offers preferential services to that group of users, such as the ability to use exclusive lanes. For example, there is a separate truck lane on the motorway around Rotterdam. The government also has planned to exempt the “target group” from paying user charges. In this way the Dutch government wants to keep the Netherlands’ position in European goods distribution network (Bovy, 1998).
countries in the world, most researchers use distance as a proxy for transport costs in their analyses.
2.2. Gravity model
Most of the studies which analyze the effect of transport infrastructure on international trade are based on the gravity model. This model is similar to Newton’s gravitation law and according to it bilateral trade flows depend on countries’ market size which is measured by countries’ GDP and the distance between them where distance is used as a proxy for transport costs. This model also assumes that, the more similar two countries are, the more they will trade with each other. The basic gravity equation, illustrated in the same way but with different initials by Alan Deardorff (1995), can be presented as follows:
Fij = А*(GDPi * GDPj / Dij) where
Fij represents trade flows from country i to country j
А is a constant
GDPi and GDPj are the gross domestic product of country i and country j
D is the distance between country i and country j
The gravity model of international trade was first developed by Tinbergen in his book “Shaping the World Economy-Suggestions for International Economic Policy” in 1962. After Tinbergen, most researchers augment the standard model as including more and different variables and in that way, improving the results. Despite its empirical success, the gravity model suffers of one major limitation which is the lack of strong theoretical foundations. The first attempt to underpin the model theoretically is made by Anderson in 1979. After him there are several other attempts for underpinning the model but although its theoretical foundations are still not very strong, it is widely used in the literature.
Taking also into account the distance between countries and countries’ GDP they find that transport infrastructure has a positive effect on trade flows.
In contrast to Bougheas, Demetriades and Morgenroth’s research, Kristjánsdóttir (2005) find that participation in regional trade organisations matters for international trade. In her paper “A Gravity Model for Exports from Iceland” she uses the gravity model in order to estimate Iceland’s trade flows. Kristjánsdóttir also tests the significance of the regional trade agreements and different industries for Iceland’s exports. She concludes that participation in EFTA, EU, and NAFTA influence Iceland’s trade.
In another study “Infrastructure, geographical disadvantage, transport costs and trade” Limao and Venables (2000) investigate the effect of transport infrastructure on trade flows for Sub-Saharan African countries. According to them distance alone is not enough to explain transport costs. That is why in their model they include several variables which affect transport costs. These variables consist of characteristics of the journey between the two countries and also characteristics of the countries themselves. There are two variables which characterize the journey between the two countries - distance between capitals and if countries share common border and for the characteristics of the two countries they take into account if the country is an island or a landlocked country. After that they include these transport costs in the gravity equation and estimate trade flows. They conclude that “Poor infrastructure accounts for 40% of predicted transport costs for coastal countries and up to 60% for landlocked countries”( 2000, Limao&Venables, 2). They also find that the dummy for a common border is significant for African countries.
Nag and Nandi (2006) however, find that the common border is not significant for trade between India and Pakistan. They conclude that sometimes political factors have a bigger influence than shorter distance.
border and if the countries are landlocked or islands. In their study Piermartini and Nordas (2004) argue that transport costs are highly correlated with the quality of the infrastructure and bad infrastructure is reflected in higher transport costs. To measure the quality of infrastructure they construct an index which includes the number of paved airports per 1000 square km, paved roads as a percentage of total roads, telephone lines per 1000 people, port efficiency index and median clearance time. They make several important conclusions. The first one is that port efficiency has a major role for bilateral trade. The second one is that distance as a proxy for trade costs does not change its weight with the improvement of the infrastructure. And finally, they emphasize that trade costs and respectively quality of infrastructure have different importance for different sectors.
3. Methodology
3.1 Sample description
Although the gravity model is so popular in the literature devoted to international trade flows, it has one major disadvantage, which is described by Piermartini and Theh (2005). In their discussion paper for the WTO, they state that, when we include countries’ specific variables in the gravity model, it estimates their role only for bilateral trade. Moreover, the gravity model can explain only bilateral trade but not total trade. So instead of using bilateral trade flows as a dependent variable in my model I will use countries’ total exports. Exports is chosen as a dependent variable mainly because as an indicator for a country’s output it gives more information about a country’s world competitiveness than imports. Although in the relevant literature some researchers use import flows or total trade flows, export flows have a predominant role and in this way my choice is confirmed. Some of the researchers who use export flows in their studies are Bougheas, Demetriades and Morgenroth (1999), Kalirajan (2007), Kristjánsdóttir (2005), Nag and Nandi (2006), etc. In order to control for countries’ size, I will adjust total exports as dividing it by countries’ GDP. This adjustment is needed because I include countries of different size in my sample. As in the gravity model I assume that total exports depend on country characteristics and transport costs.
Explanatory variables
congestion can essentially deteriorate the quality of the existing transport infrastructure and as a result, increase transport costs. Other variables measuring the quality of transport infrastructure are the available motorway and railway network. In order to present all kinds of transport modes and also because I estimate total countries’ exports, I will use registered carrier departures as a proxy for air transport infrastructure. Although, according to me, this transport mode is not so essential for European trade because it is more useful for long distances. But as WTO statistics indicates, intra-regional trade has a major share of total trade for the European countries. And the last variable in this study measures the number of people who have access to internet. Limao and Venables (2000) and also Piermartini and Nordas (2004) use the number of telephone lines per person as a proxy for the quality of infrastructure but according to me, nowadays, internet has an increasing role not only for the world economy but also for everyday people’s life. Today, people can buy almost everything which they need via internet. Moreover, internet banking increases its popularity every year which additionally accelerates e-trade. Internet also makes the receiving of business information easier and in that way, can stimulate international trade. So I assume that countries in which most people have free access to the global network have better trade performance. Although tariff barriers are an essential part of trade costs in international trade I decided not to include them in my model. My choice is based on the fact, that tariff barriers have different importance for different sectors. But as I estimate total exports, their role will be unclear.So the economic model of this paper is as follows: t_exp/t_gdp = f (t_gdp, t_area, cpi*oer, fdi, l_locked, distance, eu_m, tr_ac, pas_cars, t_motor, t_rail, reg_car, int_users)
Where:
t_exp/t_gdp = total exports of goods and services divided on total GDP of country i at time t (Euros)
gdp = total GDP of country i at time t (Euros) t_area = total area of country i
cpi*oer = consumer price index (CPI) of country i at time t (2000=100) multiply with the official exchange rate index (OER) of country i at time t (2000=100)
fdi = net inflows of foreign direct investments (FDI) of country i at time t (Euros) l_locked = a dummy equals to 1 if the country is landlocked and 0 otherwise distance = the distance between country’s i capital and Rotterdam
t_motor = total motorway network of country i at time t t_rail = total railway network of country i at time t
reg_car = registered carrier departures worldwide of country i at time t int_users = internet users of country i at time t (per 1000 people)
There are three variables related to transport which concern road infrastructure. They are total motorway network, people killed in road traffic accidents and the number of passenger cars. But as World Bank statistics indicate, for some countries and especially for most East European countries this kind of transport is predominant and it carries around 50% of freight ton-km in a country. So, I assume that road transport infrastructure should be presented with more variables in my model.
3.2. Data sources
Data needed for this research paper are collected from several international resources. Data for total exports measured in US dollars are available at the site of the World Trade Organization. The values of countries’ GDP in US dollars are taken from International Monetary Found statistics. Whether a country has access to sea is checked by the site: www.worldatlas.com. The distance between Rotterdam and the capitals of all countries is measured with the aid of
www.viamichelin.com site. The United Nations Economic Commission for Europe (UNECE) presents data about the number of people killed in road traffic accidents, the number of passenger cars and the available motorway network. Finally, most of the data are collected from the World Development Indicators (WDI). These data represent countries’ total area, consumer price index (CPI), official exchange rate, net inflows of FDI in US dollars, total railway lines, registered carrier departures worldwide and the number of internet users. Total exports, total GDP and net inflows of FDI are converted from US dollars to Euros using the exchange rate given by the Eurostat for the relevant year.
3.3. Model specification
will increase the number of observations. Furthermore, previous studies demonstrate that this kind of data has many advantages and leads to better results. In his paper “Benefits and limitations of panel data”, Chen Hsiao (1985) describes in detail the advantages of panel data comparing them with cross-section and time-series data. Here I will give only a brief description of the most common advantages.
Advantages of panel data
- Combining cross-section with time series data increases the number of observations and in this way improve the estimation results.
- Reduces the problem with multicollinearity.
- They allow us to control for country unobserved heterogeneity.
The main disadvantage of panel data is that using more complicated methods for estimation may create new problems and also may require many and more specific assumptions. Other problems with panel data described by Baltagi and Raj (1992) can include a bias in sample selection, errors in measurement, attrition bias, etc.
There are several models for analyzing panel data based on different assumptions. These models can be divided into linear and nonlinear, static and dynamic, models with fixed or random effects, with fixed individual, fixed time or fixed individual and time effects, etc. Again, I will describe only the most frequently used models in the relevant literature.
Fixed effects models The basic equation is: yit = αit + βxit + uit
In this equation the intercept αi represents individual specific effects which in my case are
European countries’ specific effects. These effects include factors such as geographical location, cultural differences and historical determinants which influence export orientation, for example, former participation of the Eastern European countries in COMECON or trade relations between some of Western European countries and their ex colonies. When the intercept is the same for all countries (αi = α) and it is also constant over time we have the
so-called constant coefficient model. But this model is not used very often in the literature. When the intercepts vary across countries but not over time we have the so-called fixed effects model called also least square dummy variables model. This is the most frequently used model in the literature. In almost all papers which I read, the researchers use this model in some part of their analysis.
In this model European countries and time - specific effects are captured by the error terms. In this model OLS estimator is not sufficient because it neglects the information of the covariance matrix and as a result GLS estimator is a better solution.
yit = β0 + β1xit + uit
where uit = µi + υit
In the second equation µ is a random variable which represents country effects. It can represent time effects. The second variable υ is the error term which is specific for each country. An important assumption is that both variables have a normal distribution.
Random coefficient models yit = βitxit + uit
where uit is the random error term
This kind of models assumes that βit coefficients are random variables. A common feature of
these models is that they try do reduce the number of the estimated parameters with the aid of different assumptions. According to the assumptions which are taken the models can be divided also into stationary and nonstationary random coefficients models. The common estimators for these models are full maximum likelihood or a GLS estimate. Egger (2001) argues that random effect models are not appropriate for estimation of the gravity model. He states that these models should be applied only when we assume that country - specific effects are not correlated with the independent variables. But in my model, I expect that country - specific omitted factors have influence on some of the independent variables such as GDP or transport infrastructure variables. Moreover, the countries in my sample are not randomly selected but they are concentrated only on the countries from the European region. So, these models do not look very appropriate for my research.
Dynamic models
yit = δyi,t-1 + β1xit + µi + uit
These models include lagged dependent variable, in my research this is countries’ total exports, but this variable is correlated with the error terms and as a result makes the estimation more complicated. In this case, fixed effects and random effect estimators are biased so an IV estimator is the appropriate one. But this estimator requires more assumptions and not all of them can be tested. Untested assumptions are the main disadvantage of this kind of models and as a result I will avoid them in my paper.
trade”, Cheng and Wall (1999) prove that this model is able to dissolve the problem with heterogeneity. In that paper they estimate trade flows a using standard gravity equations as well as an augmented gravity equations. In their analysis they compare a fixed effects model with a cross-section and pooled cross-section model and conclude that the fixed effects model eliminates the correlation between the residuals and the dependent variable. Moreover, in a fixed-effects model independent variables are assumed to be correlated with the country specific factors which is an important contribution in my case because some country-specific variables are very difficult to be measured. Such variables can include cultural differences or historical factors which are not measured in my model. Another benefit of this model is that as a simple model it does not take any complicated assumptions and as a result, is easy to be estimated. Choosing the right model is important in order to obtain sufficient results. So, I will check if my choice is correct. For this purpose, the Redundant fixed effects test and the Hausman specification test will be applied in the following section. The first one checks if fixed effects model is more appropriate than pooled regression model. The null hypothesis states that pooled regression model is the right choice. The Hausman specification test checks the choice between fixed effects and random effects models. The null hypothesis supports the use of random effects model.
Choosing an appropriate functional form
I assume that the improvement of the quality of transport infrastructure leads to an increase in total exports. However, when transport infrastructure reaches some sufficient level it continues to increase the volume of total exports but with a decreasing rate. This assumption is due to the fact that good transport infrastructure alone, cannot determine international trade flows. There are other factors, such as countries’ endowments, which have a predominant role. As a result, transport infrastructure can serve only as a stimulus or an obstacle for the increase in trade flows. But when it is available, the other determinants of international trade are more important. If we look at the graphs presented for different functional forms by Hill, Griffiths and Judge (2001), all “log-inverse”, “log-linear” and “log-log” forms look appropriate in this case. Most of the empirical studies based on the gravity model use “log-log” functional form. So I also decided to use “log-log” form in my model. Moreover, “log” dependent variable can reduce the opportunity for detection of heteroskedasticity.
The econometric model
countries’ capitals and Rotterdam ( in the literature for bilateral trade flows it is used the distance between countries’ capitals), the number of people killed in road traffic accidents and the number of passenger cars (in the literature as a proxy for the quality of road infrastructure it is used number of paved roads as a percentage of total roads), registered carriers departures (as a proxy for the quality of air transport infrastructure Piermartini and Nordas (2004) use number of airports per 1000 square km), and internet users per 1000 people ( Limao and Venables (2001) and Piermartini and Nordas (2004) use the number of telephone lines per person). So the econometric model of this research looks as follows:
Log t_exp/t_gdpit = β0it + β1 log t_gdpit + β2 log t_areait + β3 log cpi*oerit + β4 log fdiit + β5
l_lockedit + β6 log distanceit + β7 eu_mit + β8 log tr_acit + β9 log pas_carsit + β10 log t_motorit
+ β11 log t_railit + β12 log reg_carit + β13 log int_usersit + εit
The meaning of the variables is the same as it was described in the previous section.
On the basis of the independent variables in the model, thirteen hypotheses are derived. They are:
H1: A country’s GDP has a positive effect on total exports (β1 is positive). H2: A country’s total area has a positive effect on total exports (β2 is positive).
H3: Higher values of the product CPI*OER have a negative effect on total exports (β3 is negative).
H4: Foreign direct investments have a positive effect on total exports (β4 is positive). H5: Landlocked countries trade less (β5 is negative).
H6: Distance has a negative effect on total exports (β6 is negative).
H7: Membership in the European Union has a positive effect on total exports (β7 is positive). H8: A higher number of people killed in road traffic accidents has a negative effect on total trade (β8 is negative).
H9: A higher number of passenger cars has a negative effect on total trade (β9 is negative). H10: The motorway network has a positive effect on total exports (β10 is positive).
H11: The railway network has a positive effect on total exports (β11 is positive).
H12: Registered carrier departures have a positive effect on total exports (β12 is positive). H13: A higher level of internet users has a positive effect on total exports (β13 is positive).
to the fact that different kinds of transport modes are highly correlated between themselves. In order to estimate their effect on international trade most researchers construct an index and in this way they test their common significance. An example for such an empirical work is the research provided by Piermartini and Nordas (2004). Another option is to test their significance individually. For example, Bougheas, Demetriades and Morgenroth (1999) emphasize only on the effect of road transport infrastructure. In this research I will also test individually the influence of every kind of transport mode on total exports. Furthermore, the sample will be divided into two subsamples. The first one will include Western European countries and the second one will consist of Eastern European countries. In this way, I will be able to see if different quality of transport infrastructure leads to different influence on international trade. At the end, the appropriate model according to the applied tests will be estimated.
4. Statistical analysis
First, before starting to estimate the model, I have to make some corrections to my sample. As there are many missing data for Bosnia and Herzegovina and Serbia, I decided to exclude these countries from my model. Furthermore, as Cyprus and Malta are small islands’ countries for which road and railway transport infrastructure is not as important as water transport infrastructure is, these countries will be also excluded. Total exports for Belgium and Luxembourg for 1998 is given as one sum. So their volume of exports for 1998 will be derived using the data for 1998 and 1999 and assuming that the percentage of change is the same for these years. As there are many missing data, especially for transport variables, I will use unbalanced panel in my analysis. The unbalanced panel restricts the possibilities for estimation in Eviews, the program which I will use in my analysis. At the same time, it can be used for the fixed effects model and as a result, it is appropriate for my research.
The following subsections represent descriptive statistics, diagnostic checks and testing the right model.
4.1 Descriptive statistics
minimum is quite significant. Consequently, the standard deviations have as well very high values. The country with the highest volume of exports is Belgium which value is -0.117692 and the smallest volume is for Greece with a value of -2.710976. All values for the variable total exports are negative. This is due to the fact, that after adjustment of this variable dividing it by total GDP, all values for total exports are very small because total exports and total GDP have similar values for the countries. The maximum value for the variable total GDP is 28.44088 for Germany while the country with the minimum value is Albania with 21.60219. The biggest country is Turkey with maximum value of 13.57161 and the smallest one is Luxembourg with 7.857868. As concerning the variables which I use as a proxy for the transport infrastructure, the highest indicators has Germany with maximum value for the total motorways of 9.422463, for the total railway lines of 10.54868, and the registered carrier departures of 13.83909 while the worst values for the same variables has Estonia with a value of 4.304065 for the total motorways, Luxembourg with 5.616771 for the railway lines and Albania with 7.003065 for the variable registered carrier departures. When we look at the data for the total motorways however, we should take into account that Albania and Latvia have zero motorways and as a result the program exclude their values from the analysis as taking them for not available. Furthermore, the number of observations for this variable is the smallest one, there are only 206 observations. The highest value for internet users has Sweden, it is 6.637934, while the smallest one belongs to Albania, and it is -0.428434.
4.2. Diagnostic checks
Multicollinearity
between them is more than 0.80. As all three variables have the same purpose, that is to serve as a proxy for the quality of road transport infrastructure, I will exclude two of them. The variable for the total motorways is the only one used in the literature. So I also decided to keep this variable in my model. Moreover, there are many factors which can lead to accidents on the road and more passenger cars, for example the culture of driving after drinking or the lack of alternative transport mode. The expected high correlation between the different kinds of transport modes is not confirmed only for the total motorways and the total railway network. But as all other transport variables are highly correlated, I will estimate the effect of every kind of transport mode separately. Because of the high correlation between the variables for foreign direct investments and registered carrier departures, the variable FDI will be excluded from the regression for air transport infrastructure. With the same purpose, the total area will be excluded from the regression for railway transport infrastructure. So the three regressions for the different kinds of transport modes look as follows:
1. Log t_exp/t_gdpit = β0it + β1 log t_areait + β2 log cpi*oerit + β3 log fdiit + β4 l_lockedit + β5
log distanceit + β6 eu_mit + β7 log t_motorit + β8 log int_usersit + εit
2. Log t_exp/t_gdpit = β0it + β1 log cpi*oerit + β2 log fdiit + β3 l_lockedit + β4 log distanceit +
β5 eu_mit + β6 log t_railit + β7 log int_usersit + εit
3. Log t_exp/t_gdpit = β0it + β1 log t_areait + β2 log cpi*oerit + β3 l_lockedit + β4 log distanceit
+ β5 eu_mit + β6 log reg_carit + β7 log int_usersit + εit
The next step will be to check the model for heteroskedasticity and autocorrelation. Heteroskedasticity
Heteroskedasticity is often detected in cross-section data. In my case it can be due to the fact that I include countries of different size. That means that explanatory variables can have different influence on different countries. To put this in other way, that means that the variance (σ2) differs for different observations which is a violation of one of the major assumptions for linear regression model, the one for constant variance. The main consequences of the existence of heteroskedasticity in the regression model are (Hill; Griffiths; and Judge, 2001):
1. The least square estimator is not the best linear unbiased estimator anymore.
In order to test for heteroskedasticity, first, I run pooled OLS regression for the three kinds of transport modes independently and after that I apply the White Heteroskedasticity Test. The results are presented in tables 5, 7, and 9 in the appendix. In all three regressions the null hypothesis for absence of heteroskedasticity is rejected. As a result I will use White’s estimator which controls for existence of heteroskedasticity. In their paper “Improved Heteroskedasticity-Consistent Covariance Matrix Estimators”, Cribari-Neto, Ferrari and Cordeiro (2000) also state that this estimator is consistent when heteroskedasticity of unknown form is detected.
Autocorrelation
The next task is to test for autocorrelation. It is often detected in time-series data and as a result I also expect its presence in my model. Autocorrelation means that the error terms (et) are
correlated over time. The result is that the information obtained by the explanatory variables can be due to the data from the previous years. The consequences of the presence of autocorrelation are the same as those from the existence of heteroskedasticity. Mainly that (Hill; Griffiths; and Judge, 2001):
1. The least squares estimator is no longer the best linear unbiased estimator.
2. Standard errors are incorrect, which lead to misleading confidence intervals and hypothesis tests.
To test for the presence of autocorrelation I use Durbin-Watson test. In all three regressions, presented in tables 4, 6, and 8, the Durbin-Watson statistic is very low. In the regression which includes total motorway network it is 0.234, for railway regression is 0.291 and finally when I include registered carrier departures, it is 0.422. It should be around two in order to have no autocorrelation. To solve this problem I will include an autoregressive error of first order AR(1).
At the end, I run again pooled OLS but this time using the White’s estimator and the autoregressive error of first order AR(1). As we can see from tables 10, 11, and 12, now the Durbin-Watson statistic is close to two for all three regressions, which indicates that there is no more autocorrelation. Furthermore, the adjusted R-squared for all regressions are higher. Before controlling for heteroskedasticity and autocorrelation it is 0.603 for the road transport infrastructure, 0.514 for the railway infrastructure and 0.618 for the air transport. After the correction it is 0.947,0.934,and 0.891 respectively.
Before starting to estimate the model, I will test as well which model is the most appropriate for my case. For this purpose, the Redundant fixed effects test and the Hausman specification test will be applied. First, in order to apply the Redundant fixed effects test, I have to estimate the fixed effects model. For this purpose, all variables which are constant over time have to be excluded from the model because, as I explained earlier, their effects are captured by the intercept. The regression outputs for Redundant fixed effects test are presented in tables 13, 14 and 15 in the appendix. For all regressions, the null hypothesis which supports the choice of the pooled regression model is rejected in favor of the fixed effects model. Next, I will test if the fixed effects are as well better solution than the random effects for my model. For this purpose, the Hausman specification test is applied. The outputs of this test are presented in tables 16, 17 and 18. For this test, again, the null hypothesis which supports the choice of the random effects model is rejected in favor of the fixed effects model for all regressions. So, the fixed effects model looks the most appropriate for my research.
5. Model estimation
infrastructure, the coefficients of the dummy for EU membership and the intercept are not statistically significant at 5% level of significance. Although the coefficient for the registered carrier departures is significant and very high, its value is 0.201250, it has again negative sign. Next, I will structure the data as paneled and apply the fixed effects model in order to see if the results are the same after controlling for country specific omitted variables. The results are presented in tables 19, 20 and 21. This time in the regression for road transport infrastructures the variable total motorways and internet users are not statistically significant. In this regression only the hypothesis for the positive effect of foreign direct investments on total exports is confirmed. This hypothesis is also confirmed in the regression for the railway transport infrastructure. Unfortunately, the coefficient for the total railway lines is again not statistically significant at 5% level of significance. For air transport infrastructure, however, there is significant improvement. This time, the coefficient for the variable registered carrier departures is positive and statistically significant at 5% level of significance. The coefficient of internet users is also positive and statistically significant. On the contrary of my expectations, it seems that air transport infrastructure has the strongest positive effect on European exports. Now I will check if this assertion is also true for my two subsamples.
Western European subsample
The pooled OLS regressions for Western European countries are presented in tables 22, 23, and 24 in the appendix. There are three significant variables in the regression for the road transport infrastructure. They are FDI, the dummy for landlocked countries, and the total motorways. But only foreign direct investments and the dummy variable are with the expected sign. In the second regression, only the coefficients of the dummy for EU membership and the total railway lines are statistically significant but both are not with the expected signs. Finally, in the regression for the air transport infrastructure there are three variables which are not statistically significant at 5% level of significance. They are the total countries’ area, the variable cpi*oer, and the dummy variable for landlocked countries. This time two of the expected hypotheses are confirmed. The first one concerns the negative influence of the distance on total exports and the second one relates to the positive effect of the access to internet on the export flows. Although the coefficient of the variable for registered carrier departures is statistically significant, it has a negative value. When I use fixed effects model, the results are similar. There is only one confirmed hypothesis in the third regression and it concerns the internet users. The results from the fixed effects model are presented in tables 25, 26, and 27.
Eastern European subsample
All results for Eastern European countries are presented in tables 28, 29, 30, 31, 32, and 33. When I use the pooled regression model for the road transport regression, the variables for the product cpi*oer, the dummy for landlocked countries, the total motorways, and the internet users are statistically significant. But only internet users are with the expected sign. Although total railway lines are positive in the second regression, they are not statistically significant. In the last regression, the coefficients of the variables for the distance and internet users are statistically significant and with the expected signs. The coefficient of the variable registered carrier departures is also statistically significant but negative. The results of the fixed effects model are again better. The hypothesis for the positive effect of foreign direct investments on export flows is confirmed in the second regression and the hypotheses for the positive influence of registered carrier departures and internet users are confirmed in the third regression.
6. Results
In this section I will sum up and discuss the results for all variables derived in the previous section after controlling for country specific effects.
Because of the high correlation, I had to exclude three variables of all my regressions, the total GDP, the number of people killed in road traffic accidents, and the number of passenger cars in a country. Furthermore, the total area is excluded from the regression for railway transport infrastructure and foreign direct investments are excluded from the regression for air transport infrastructure. As a whole, for all regressions the adjusted R-squared is quite high. It is always more than 0.80.
Although the coefficient of the variable cpi*oer is statistically significant in most of the regressions, it is not with the expected sign. A possible reason for this result could be that when the consumer price index in a country is high, that means that the standard of living in this country is high. At the same time, this fact does not have a negative influence on total exports. The expected positive effects of foreign direct investments on total exports are confirmed for the whole sample and for the railway regression for Eastern European subsample.
The coefficient for the number of internet users is positive and statistically significant in the air transport regression for the whole sample, as well as for the two subsamples.
subsample. Although positive, it is not statistically significant at 5% level of significance for the Western European countries. The coefficient of the total railway lines is also statistically significant and positive for the Eastern European subsample when I use pooled regression model. But when I control for country specific omitted variables, it loses its significance.
Because of the great variance in the regressions’ outputs, it is difficult to make any interpretations of the results. As a whole, only the significance of the air transport infrastructure on total exports is confirmed. On the contrary of my expectations, road and railway transport infrastructure do not have any influence on the export flows. At the same time, when we interpret the total road and railway lines’ coefficients, we should take into account that these variables do not say a lot about the quality of these transport modes. Moreover, the variable registered carrier departures, is also not very good indicator about the quality of air transport infrastructure. It indicates more the availability of this kind of transport mode. As a consequence, the obtained results can be misleading.
7. Conclusion and discussion
From the existing literature based on the determinants of international trade flows it is clear that better transport infrastructure and consequently, lower transport costs play a key role in determining countries’ volume of trade. Moreover, it can represent an important advantage for the country which leads to an improvement of countries’ competitiveness in the world market. In this paper I focused on the influence of transport infrastructure on European exports. Three kinds of transport modes were examined: the road, railway, and air transport modes. In order to control for the advantages of the sea transport infrastructure, a dummy variable which indicates if a country is landlocked was introduced.
This research was based on the gravity model which is the most common model for analyzing international trade flows. In my analysis I used panel data for 31 European countries for a period of eight years. Furthermore, the sample was divided into two subsamples in order to see if there is essential difference in the values of the variables for Western and Eastern European countries. The results were obtained as using the pooled regression model as well as the fixed effects model.
subsamples. So my hypothesis that transport infrastructure has a bigger influence on Eastern European exports was not confirmed.
In return to the research question posed in the introduction: “What is the influence of transport
infrastructure on trade flows for the countries from Eastern and South-Eastern Europe?” The
answer is not the same for the examined transport modes. It seems that the air transport infrastructure is of greatest importance for this region. On the contrary of my expectations, it seems that the motorway and the railway transport infrastructure have no influence on the European exports.
So, it seems that as a whole existing transport infrastructure has a minor role in explaining international trade flows and other factors are of greater importance.
When we interpret the obtained results, however, we have to take into account some of the limitations of this paper. The most important one is that a possible explanation for some of the confusing results could be the lack of data about the quality of the chosen transport modes. There is no doubt that the quality of transport infrastructure is of great importance but unfortunately, it is very difficult to be measured. Moreover, exactly these countries which have the worst transport infrastructure, do not give sufficient data about its quality. As a result, I was not able to include such kind of information in my research. Furthermore, as I stated earlier, transport infrastructure has different influence on different industries. So the method of using aggregate data can be also a reason for the great variance in the obtained results.
References
Articles
Anderson, J. E. 1979. A Theoretical Foundation for the Gravity Equation. American Economic Review, 69, 1, 106-116
Aschauer, D. A. 1989. Is Public Expenditure Productive? Journal of Monetary Economics 23, 177-200
Bougheas, S., Demetriades, P. O. & Morgenroth, E. L. W. 1999. Infrastructure, Transport Costs and Trade. Journal of International Economics 47, 169-189
Bovy, P. H. L. 1998. Traffic Flooding the Low Countries: How the Dutch Cope with Motorway Congestion. Transport Reviews, 2001, Vol. 21, 'o. 1, 89-116
Brüderl, J. 2005. Panel Data Analysis. University of Mannheim, Available at:
http://www.sowi.uni-mannheim.de/lehrstuehle/lessm/veranst/Panelanalyse.pdf
Casas, F. R. 1983. International Trade with Produced Transport Services. Oxford Economic Papers 35, 89-109
Cheng, I.& Wall, H. 1999. Controlling for Heterogeneity in Gravity Models of Trade. Federal
Reserve Bank of St. Louis, Working Paper 99-010A, Available at:
htpp://www.stls.frb.org/research/wp/99-010.html
Christie, E. 2002. Potential Trade in East Europe: A Gravity Model Approach. South-East Europe Review, 4/2002, 81-102
Cribari-Neto, F.; Ferrari, S. L. P.; Cordeiro, G.M. 2000. Improved Heteroskedasticity-Consistent Covariance Matrix Estimators. Biometrika (2000), 87, 4, 907-918
Deardorff, A. V. 1995. Determinants of Bilateral Trade: Does Gravity Work in a Neoclassical World? 'BER Working Paper, 'o. 5377
Egger, P. 2000. A Note on the Proper Econometric Specification of the Gravity Equation. Economics Letters, 66, 25-31
Ellis, L. 2001. Measuring the Real Exchange Rate: Pitfalls and Practicalities. Research Discussion Paper 2001-04, Economic Research Department, Reserve Bank of Australia Fujimura, M. & Edmonds, C. 2006. Impact of Cross-border Transport Infrastructure on Trade and Investment in the GMS. ADB Institute Discussion Paper 'o. 48
Goodwin, P. 2004. The Economic Costs of Road Traffic Congestion. ESRC Transport Studies Unit, University College London
Head, K. 2003. Gravity for Beginners. Faculty of Commerce, University of British Columbia, Canada
Herranz–Loncan, A. 2005. The Spatial Distribution of Spanish Transport Infrastructure between 1860 and 1930. Springer-Verlag 2006, DOI 10.1007/s00168-006-0096-0
Hsiao, C. 1985. Benefits and Limitations of Panel Data. Econometric Review, 4(1), 121-174 Hummels, D. 2007. Transportation Costs and International Trade in the Second Era of Globalization. Journal of Economic Perspectives, Volume 21, 'o. 3, 131-154
Judson, R. A. & Owen, A. L. 1998. Estimating Dynamic Panel Data Model: A Guide for Macroeconomists. Economics Letters 65 (1999) 9 - 15
Kalirajan, K. 2007. A Comparative Analysis of Recent Export Performances of China and India. Asian Economic Panel, Brookings Institution
Khadaroo, J & Seetanah, B. 2007. Transport Infrastructure and Tourism Development. Annals of Tourism Research, Vol. 34, 'o. 4, 1021-1032
Kristjánsdóttir, H. 2005. A Gravity Model for Exports from Iceland. Department of Economics, University of Copenhagen
Limao, N. & Venables, A. J. 2000. Infrastructure, Geographic Disadvantage, Transport Costs and Trade. World Bank Economics Review 15, 451-479
Moneta, C. 1959. The Estimation of Transportation Costs in International Trade. The Journal of Political Economy, Vol. 67, 'o. 1, 41-58
Moreno, R & Lopez-Bazo, E. 2007. Returns to Local and Transport Infrastructure under Regional Spillovers. International Regional Science Review 30, 1: 47-71
Mundel, R. A. 1957. A Geometry of Transport Costs in International Trade Theory. The Canadian Journal of Economics and Political Science, August 1957, 331-348
Nag, B. & Nandi, A. 2006. Analysing India’s Trade Dynamics via-E-vis SAARC Members Using the Gravity Model. South Asia Economic Journal 2006, 7, 83
Piermartini, R. & Nordas, H. K. 2004. Infrastructure and Trade. Working Paper ERSD- 2004-04, World Trade Organization
Piermartini, R. & Teh, R. 2005. Demystifying Modelling Methods for Trade Policy. WTO Discussion Paper 'o. 10
Samuelson, P. A. 1954. The Transfer Problem and Transport Costs, II: Analysis of Effects of Trade Impediments. The Economic Journal, Vol. 64, 'o. 254, 264-289
Sanchez-Robles, B. 1998. Infrastructure Investment and Growth: Some Empirical Evidence. Contemporary Economic Policy, Vol. XVI, January 1998, 98-108
Sankaran, J.K.;Wood, L. 2007. The relative impact of consignee behaviour and road traffic congestion on distribution costs. Transportation Research Part B41 (2007), 1033-1049
Seetanah, B. 2006. Analysing Transport Capital as a Determinant of Tourist Arrival in a Co-integration and Error Correction Framework. International Review of Business Research Papers, Vol. 2, 'o. 2, August 2006, 15-29
Shepherd, B. & Wilson, J. 2006. Road Infrastructure in Europe and Central Asia: Does Network Quality Affect Trade? World Bank Policy Research Working Paper 4102, December 2006
Tinbergen, J. 1962. Shaping the World Economy-Suggestions for an International Economic Policy. American Economic Review, Vol. 53, 'o. 3, 1963, 514-516
Yafee, R. 2003. A Primer for Panel Data Analysis. University of 'ew York, Information Technology Services, 2003
Books
Hill, R. Carter; Griffiths, William E.; and Judge, George G. 2001. Undergraduate Econometrics. 2nd Edition, John Wiley & Sons, Inc.
Linnemann, H. 1966. An Econometric Study of International Trade Flows. Amsterdam, 'orth-Holland
Mátyás, L. & Sevestre, P. 1992. The Econometrics of Panel Data: Handbook of Theory and Applications. Advanced Studies in Theoretical and Applied Econometrics, Vol. 28
Raj, B.& Baltagi, B. 1992. Panel Data Analysis. Empirical Economics, Vol. 17, 'o. 1 Trans-European Transport Network. TEN-T Priority Axes and Projects 2005. European Commission, 2005
Internet sources
International Trade Statistics 2007, World Trade Organization World Bank (2007). World Development Indicators, CD Rom
www.ecb.int www.unece.org www.viamichelin.com www.worldatlas.com
Lecture otes:
Duncan, A. 2000. An Introduction to Panel Data Analysis. Available at:
