European Economic Overview

Get Complete Project Material File(s) Now! »

Empirical Section

Model of Choice and Variables

The gravity model was indeed intended to be used in this paper. However, there are many variants in how to define the variables as well as the equation itself. Andersson and van Wincoop (2003) choice of multilateral resistance term did sound appealing but as stated above; a grouped variable will be achieved and thus it will be hard to observe the significance of each and every variable. Thus the more original option of Linnemann will be used where the model later will be double logarithmic for estimation purposes. Linnemann did reject the choice of introducing GDP per Capita into the model due to multicollinearity; thus a correlation matrix, appendix 4, was computed as there were suspicions regarding the correlation between the variables GDP and population. Indeed the correlation matrix showed that the variables were highly correlated depending on the year; between 81%-85%. Hence the variable population was removed and the model was modified to include Tinbergen’s model [4.3] where the GDP for each country was used along with the distance variable.
Two regressions will be run to have sufficient information regarding the coefficients and their estimators. The first regression model [5.1] will include all 29 countries for each year in the time period 1995-2011 but divide the years into 3 sets; 1995-2000, 2001-2006 and 2007-2011.
The second regression model [5.2] will include all 29 countries in the time period 1995-2011 and will calculate the impact of the sovereign debt crisis on the trade in the European Union. The variable will start taking effect the year 2007. Additionally, dummies are introduced for each and every year to allow for a different intercept in each year. As a result, the dummy coefficient of each year will generate the mean traded value in each year. To avoid the dummy variable trap the intercept term is omitted.
LnExpij = β0 + β1LnYi + β2LnYj + β3LnDij + β4Lan + β5Bor + β6Cur + β7LockExp + β8LockImp + β9EuExp + β10EuImp + β11EuCrisis + β12Year (1996, 1997, … , 2011) + εij [5.2]
The variables which will be used in the regressions will now be presented, along with their sources and the expected signs. A table will then be presented summarizing the hypothesis.
Total Export Value (LnExpij)
This variable contains the export data from country i to country j (the other 28 countries) for each and every year. This is the dependent variable in both of the regression models and was collected from Comtrade. The values was measured in current US dollars and then made logarithmic.
The GDP of the Exporting Country (LnYi)
This explanatory variable consists of the reported GDP of country i for each and every year and is measured in current US dollars. It was collected from UNdata base and should have a large positive impact on trade according to Linnemann (1966), Tinbergen (1962) and the main concept of Newton’s gravity model. As this variable increases, the more products will be available for exports.
The GDP of the Importing country (LnYj)
This explanatory variable was collected, reported and measured in the same way as the previous. This variable should as well have a large impact on trade according to Linnemann (1966), Tinbergen (1962) and the main concept of Newton’s gravity model. Additionally, Linnemann (1966) describes this variable as the ‘attraction’ variable as with an increased size the domestic demand will increase along with domestic income. Thus increasing the total amount imported from country i.
Geographical Distance (LnDij)
This explanatory variable was collected from and contains the geographical distance between the capitals for each and every country in the study. As Newton’s gravity model as well as Linnemann (1966) states, this variable should have the largest negative impact on trade. The trade between two countries diminishes as the distance increases. This variable was measured in kilometers and then made logarithmic. The cities used for measuring the distance can be found in appendix 5.
Common Currency (Cur)
This dummy variable was collected from Nationalencyklopedin and takes on the value of 1 if, and only if, country i and country j shares the same currency, otherwise 0. Sharing the same currency would induce a decreased distance between the objects (countries). Thus, as decreased distance increases the trade between the countries this variable would have a positive impact on trade according to Linnemann (1966).
Exporting Country in the European Union (EuExp)
This dummy variable takes on the value of 1 if, and only if, the exporting country is a part of the European Union, otherwise 0. According to Krugman and Obstfeld (2009) the economic integration would have a positive impact on trade due to comparative advantage and economies of scale. Due to these reasons along with being a customs union (where tariffs are less or removed for trade between the countries) this variable should have a positive impact on trade.
Importing Country in the European Union (EuImp)
This dummy variable takes on the value of 1 if, and only if, the importing country is part of the European Union. This variable should as the variable above have a positive impact on trade. And due to the preferential trading agreements countries are more willing to trade with other countries that are part of the European Union as well. Thus the trade flows would decrease if destined to a country part of the European Union, if the exporting country is not part of the customs union as well, which in turn implies tariffs, higher costs and lower demand (Krugman and Obstfeld, 2009). Hence it is hard to predict a hypothesized effect of this variable.
European Sovereign Debt Crisis (EuCrisis)
This dummy variable takes on the value of 1 if, and only if, the economic crisis is in effect. This variable start having an effect from year 2007 until 2011 and will capture the effect of the sovereign debt crisis in Europe. This variable should have a negative impact on trade.
Common Language (Lan)
This dummy variable was collected from Nationalencyklopedin and takes on the value of 1 if country i and country j shares the same language, otherwise 0. According to Linnemann (1966) this is a natural barrier to trade (included in distance) and thus sharing the same language should have positive impact to trade.
Common Border (Bor)
This dummy variable was collected from WorldAtlas and takes on the value 1 if and only if country i and country j shares a land border with each other, otherwise 0. However, if two countries are connected by an artificial construction (such as the Sweden-Denmark Bridge or UK-France tunnel) the value will be given 1. This is due to the reason that this transportation way does not require a shift of vehicle (i.e. truck-boat) in the transportation. Thus this new connection does not induce a major increase of transportation time. This variable should, if given value 1, have a positive impact on trade as sharing a common border naturally decreases the distance between countries compared to if they did not share a border as Linnemann (1966) argues.
Landlocked Exporter (LockExp)
This dummy variable was collected from WorldAtlas and takes on the value of 1 if, and only if, country i has access to the coast. According to Brakman, Garretsen and van Marrewijk (2009) this variable induces less transportation costs and less delays. However, this variable importance has declined through time as new transportation systems have developed – allowing a less costly and a more efficient way for transportation. Countries which have had access to the coast have improved this transportation system as well. On the other hand, countries located centrally in Europe such as Switzerland has no access to coast but do have a central geographical location close to big economies such as Germany and France. Thus the geographical location importance may overcome the significance of access to a coast. This variable should have a slightly positive effect on trade.
Landlocked Importer (LockImp)
This dummy was collected and reported in the same way as the variable above. It takes on the value 1 if, and only if, the importing country has access to a coast, otherwise 0. Due to the same reasons above, this variable should have a slightly positive effect on trade.
Year Dummies (Year)
This dummy variable is introduced to generate an intercept for each and every year. This is due to the reason that GDP and the exports contains continuous data and to investigate whether there is a yearly effect in a given year. Thus the value of 1 is given if, and only if, the year is of relevant interest. Otherwise 0. As trade can return both a positive, insignificant and negative mean value (compared to the benchmark year), this variable is hard to put a trend on (Gujurati and Porter, 2009).


Additional Assumptions

During the information retrieval it became clear that much data was not available for certain years. Additionally, Belgium and Luxembourg separated the year 1999 and was before that one single country. It was possible to include Belgium-Luxembourg in the studies but this would have induced problems with the choice of capital and thus the distance. Hence Belgium and Luxembourg enters the sample year 1999 and is not accounted for until then. The missing values as well as the zero-trade flows, mostly found during the 20th century, can be dealt with by three different guides the World Trade Organization (2012) offers.
The first way is to drop the observation with a zero-trade flow. This can be done if the zero-trade flows are randomly distributed and thus are not informative. The second way to approach this problem is the use of adding a small constant instead of the zero-trade flow. This method is of importance when small economies, such as Malta, chooses not to commit any trade due to distance, land locked reasons or excessive transportation costs. This has also been observed to be the case with Malta and Cyprus. The third way to approach the problem is the use of estimating the model in levels where each and every variable is estimated. This approach would induce a non-linear relationship.
The second and third way is incorrect if the OLS estimation method is used. If the second or third method were to be used inconsistent estimators in the results would have been obtained. “Second, the use of OLS estimation in levels is not supported by theoretically founded gravity equations that present a multiplicative form” WTO (2012).
Thus, even though many of the observation with a zero-trade flow can be considered random, the use of the first method is forced to be conducted with the remaining observations where transportation cost and distance might have been the case. The observations with zero-trade flows are thus considered to be random and removed.


The first regression will deal with the data in three sets, cross-sectional data for the years 1995-2011, by the use of ordinary least squares (OLS). This way the trend which the estimators are changing can be more specifically observed. The second regression will pool the data and, by the use of OLS once again, construct one regression instead of 17. This is the regression which will estimate the effect of the sovereign debt crisis in Europe with the start of 2007 and also create an intercept for each and every year (year dummies).
To remove cross-sectional correlation between the variables in the time trend, referred to as ‘Gold medal error’ in the paper written by Baldwin and Taglioni (2006), time invariant pair dummies can be created to be used in the pooled data for the second regression. This type of procedure is referred to as ‘The Fixed Effect Least-Squares Dummy Variable Model (LSDV)’. Gujurati and Porter (2009) state that this type of procedure requires many dummy variables in the model, and as the model already consists of a significant amount of dummy variables, introducing additional dummies would result in a higher possibility of multicollinearity. Additionally, as there exists variables that does not change with time (such as coastal access and partly language), “In some situations the LSDV may not be able to identify the impact of time-invariant variables… Incidentally, the time-invariant variables are sometimes called nuisance or lurking variables”, Gujurati and Porter (2009) p.598. Hence, this approach is disregarded.
It is also given naturally that due to the different sizes in GDP between the countries the variance in the exports cannot be the same (Gujurati and Porter, 2009). Additionally, due to the reason of the amount of dummy variables the variances are not expected to be equal. This can be explained by that each and every dummy variable measures different subjects and where the effect of one dummy variable can equal in more zeros meanwhile the other equals in more ones (Aczel and Sounderpandian, 2009).
White’s General Heteroscedasticity Test was conducted on the data and the plots of residuals against the predicted values can be found in appendix 6. It became evident that there was heteroscedasticity present in the data and the presence of heteroscedasticity makes the estimators inefficient, underestimate/overestimate the variances and therefore may cause the regression results to be misleading (Gujurati and Porter, 2009). Thus to remedy this problem the use of White’s heteroscedasticity-consistent standard errors are generated and replaced the regular standard errors generated from the OLS regressions.

Descriptive Statistics

The descriptive statistics can be found in table 5.3.1 and table 5.3.2 for the first and the last year, 1995 and 2011. The dependent variable, LnExp, increased by 0.6734 in the mean and by 0.6368 in the median. Indicating that the trade flows has increased on average but in the same time the central point has increased less and also stating that on average the bigger economies have increased their exports more than the smaller. This can also be verified by the values of GDP for the exporting countries as well as the importing countries where the mean has increased more than the median. However, many observations 1995 were missing and if these values corresponds to smaller countries then these results could be explained by the fact that even though the trade flows have increased – the new data from the former missing countries may have had an effect on the median as well as the mean. The standard error has decreased in value from the observation in 1995 to 2011 indicating that the most values have clustered in the middle.
The dummy variable for currency has increased to count for 33% of the countries conducting bilateral trade flows in year 2011. This is to be expected as the euro was introduced year 1999 and thus its impact is not to be significant before that. The variable counting for if the exporting respectively the importing country were a part in the European Union differs year 1995, and then both increases to account for 93% of the bilateral trade flow. The difference 1995 can be explained by the missing observation that were removed meanwhile the increase from 1995 to 2011 is to be expected as only Norway and Switzerland were not a part of the European Union 2011.
The explanatory variable LnDist’s maximum and minimum remains unchanged meanwhile mean and median have decreased, implying that either more trade has been done to countries located closer or that the missing values from 1995 were countries closer located in Europe and thus effecting the result.
The GDP for the exporting and importing countries remains closely to each other for both year 1995 as well as year 2011. The standard deviation has, however, decreased from year 1995 to year 2011; implying that the distance from the mean value has decreased in year 2011 and that the values are more clustered in the middle than before.
The dummy variable for border countries remained the same, which was to be expected or even an increase as Belgium and Luxembourg (bordered countries) came into the sample first 1999. The dummy variables for if the countries where landlocked or not, where the value of 1 was given if they had access to a coast, differs a bit year 1995 meanwhile they are the same year 2011. The difference year 1995 would most likely be due to the missing observations that were removed. The final dummy variable, for language, increases from year 1995 to year 2011. This is to be expected as Belgium and Luxembourg came into the equation 1999 where both of the country’s official language are shared by others.

Table of Contents
1. Introduction 
1.1 Previous Research
1.2 Disposition
2. European Union 
2.1 European Economic Overview
3. Theory 
3.1 Tariffs
3.2 Comparative Advantages
3.3 Economies of Scale
3.4 Optimum Integration
4. Model 
4.1 Theoretical Foundation
4.2 Further Developments
5. Empirical Section 
5.1 Model of Choice and Variables
5.2 Method
5.3 Descriptive Statistics
5.4 Regression Results
6. Analysis 
7. Conclusion
The European Union and the Euro A Gravity Approach on Bilateral Trade

Related Posts