M.Sc. International Economics and Business - University of Groningen
M.A. International Economy and Business - Corvinus University of Budapest
Double Degree Program
Master Thesis July 2013
The Impact of Infrastructure Development on
the Economic Growth in Europe
Georgina Csendes
S2418622
csendes.georgina@gmail.com
2
Abstract
In the current empirical analysis the production function approach is applied, in order to investigate the contribution of the line infrastructure to the economic growth in Europe. The broadest approach is based on the period 1999-2009 for eleven European countries. The panel dataset covers the 2004 enlargement of the EU, which allows comparing the effects of the infrastructure in old and new member states. Furthermore, as infrastructure connecting these countries, spatial interaction effects are examined as well. The results show a positive effect of the line infrastructure on economic growth, which is higher when spatial issues are accounted for.
3
Table of content
I. Introduction ... 4
II. Hypothesis ... 7
III. Methodology ... 8
1. Data and Variables ... 8
2. Model specification and estimation ... 10
2. a. Standard panels ... 10
2. b. Spatial panel models ... 13
III. Results ... 15
1. Standard panel models ... 15
2. Spatial panel models ... 18
3. Comparison of the results of the standard and the spatial panel model ... 19
IV. Conclusion ... 21
V. References ... 23
4
I. Introduction
The growth performance of the European Union has markedly changed after the 2004 enlargement, due to the larger single market and intensified trade between the old and new member states. However, the new member states, from which seven were parts of the Soviet Union, represent slower productivity growth (Timmer et al., 2007) and thus, make the EU diversified. In order to achieve convergence and economic development, funds are invested in wide range of infrastructure, such as innovation, research, telecommunication, environment, energy and transport (ERDF regulation, 2006). Since infrastructure provision is an important factor of shaping growth, European policy makers aim to improve regional GDP per capita, facilitate mobility, increase accessibility and remove bottlenecks in the EU transport system through investing in infrastructure (ESPON, 2003). The above mentioned facts indicate that the new member states, which mainly consist of transition economies, can better access financial assets and thus can invest into rebuilding or maintaining the infrastructure (Hoskinsson et al., 2012). Firstly, this also suggests that after the accession the impact of the infrastructure on economic growth is higher in the later joined countries. Secondly, the effects may be different due to the fact that in infrastructure abundant countries new infrastructure may have minimal impact due to diminishing returns (Oosterhaven and Knaap, 2003).
5
production researches, which cannot really answer the relevant policy questions. He suggests that economists should pay more attention to disaggregated rate of return studies and researches that examine the impact of policy changes. Romp et al. (2007) also provide a survey on how public capital can contribute to economic growth. The authors argue that public capital enters the macroeconomic production function in two different ways; i.e. directly as a third input of production and indirectly influencing production through multifactor productivity. They highlight the importance of public infrastructure in raising the production, since it lowers fixed costs, attracts companies and serves as a factor of production. However, these not necessarily indicate higher growth at national level, due to the differences in regional development. In La Ferrara and Marcellino (2000) the production function approach is applied at regional level. The authors examined the impact of public infrastructure on economic performance. In this study services from public capital is considered as a direct factor input and proxied by the stock of infrastructure. They find mixed effects on Italian macro-regions, since the productive structure and the level of development vary substantially across the Northern and Southern regions.
6
growth across the OECD countries. However, their annual time series growth regression suggests that the contribution of infrastructure on long run output levels is heterogeneous across countries. Moreover, the authors find a positive and non linear link between infrastructure and economic growth.
Considering the findings and arguments of the related literatures, there is a broad consensus among economists that infrastructure stock as a factor of production is an important determinant of economic growth. Furthermore they highlight that there could be differences in the way and level of contribution of infrastructure to economic growth, when the examined regions or countries vary in the initial level of infrastructure development. However, these researches do not pay much attention on the spillover effects of the transport infrastructure that may occur across the borders through network externalities (Egert et al., 2009; Delgado et al., 2000; Mankiw et al., 1992; Gramlich, 1994). The lower transport and production costs due to the transport infrastructure can facilitate the market expansion that links the different regions and countries. Since infrastructure is connecting geographical units, better infrastructure in the home country can attract production factors from the neighbouring countries as well (Romp et al., 2007). Therefore exports and imports may increase that may lead to a higher levels of output. The upsurge of import may put competitive pressures on local firms; however, these pressures could lead to improved efficiency, which may lower the costs of production, raise productivity and thus enhance aggregate growth (Lakshmanan, 2008). Furthermore, as far as economic growth is concerned, it is not merely determined by within country development but it could be affected by the growth rates of neighbouring countries (Debarsy et al., 2006). These facts indicate that an observation associated with a given country may be dependent on observations at other locations as well, due to spatial interactions (Elhorst, 2010; LeSage and Pace, 2009). In order to control for this spatial dependence, Del Bo et al. (2009) estimate an infrastructure augmented production function model by using spatial econometrics techniques, to evaluate the role of infrastructure in shaping GDP growth in Europe. According to their results, the estimates indicate a stronger positive relationship between infrastructure and economic growth, when spatial issues are fully accounted for. Considering their findings, in order to get more accurate results, it is crucial to control for spatial interactions.
7
old member states, which allows to control for the effects of the infrastructure before and after the accession in case of the new member states; and to compare the different effects experienced by new and old member states. The Czech Republic, Hungary, Poland, Slovakia and Slovenia represent the new member states. These countries are somewhat similar in size, in historical background and joined the EU together in 2004 following the same accession process. The selection of the old member states consists of Austria, Germany, France, Italy, Belgium and the Netherlands, which are the main trading partners and parts of the neighbourhood set of the new member states.
In line with the recent literature, the infrastructure stock needs to be considered as an input of the production process to capture its effects on economic growth. Thus, the applied models in the current analysis are based on the production function approach. On purpose to test the argument that infrastructure development has a positive effect on the economic growth standard panel models are applied. Moreover, assuming that spatial interaction effects exist, spatial panel models are used as well, in order to control for the spatial issues. The results of the standard and spatial panels are compared to see whether the spatial panel models give more accurate estimates.
II. Hypothesis
The related literature highlights the evidence of a strong link between infrastructure and economic growth. In most cases the production function approach is employed as an appropriate framework to determine the effect of the infrastructure on economic growth, through the production process. It has to be mentioned that depending on the selected countries, time periods and infrastructure units (length or capital) are used; the estimated results can be diverse. Moreover, spatial interactions may also influence the outcomes of the research because geographical units are examined, which are linked to each other by road networks and rail lines. Thus, more precise results can be obtained by controlling for these spatial dependence issues.
Based on these arguments and results, the current empirical analysis tests the following hypothesis by using both standard and spatial models on the basis of the production function approach:
8
III. Methodology
1. Data and Variables
The variables used in the empirical analysis are from 1999 to 2009 (t) for the eleven European countries (i).
Dependent variable
The main variable of interest is GDP per capita, as it controls for the country size differences. The dataset of this variable stems from the Eurostat database and expressed in current prices deflated by current PPP for each year (Eurostat, 2011). On the web site of the European Commission’s statistical database, possibilities are given to construct a suitable table of GDP indicies for the selected countries and years. The values of the dependent variable are specified in euro.
Independent variables
As the analysis focuses on the contribution of the line infrastructure development to economic growth, the total length of roads (km) and railways (km) for the selected eleven European countries and eleven years period are considered as the main explanatory variables. In the current analysis physical stock of infrastructure is used since it can ensure better estimates. Furthermore, using stocks, rather than investment flows may reduce the problem of causality; however, it does not eliminate it entirely (Egert et al., 2009; Romp et al., 2007, Sturm et al, 1999).
In order to obtain all values for one of the main independent variables, the length of roads, for the examined countries and years, two steps had been taken due to lack of available data in the primary source, Eurostat,.
9
In order to capture the other main line infrastructure variable, the length of rail lines, data was drawn from the World DataBank, World Development Indicators. The reason for considering only one source in this case was that the dataset available in Eurostat is lacking information for some countries; while in contrast, the World Bank database contains all the values for the length of railways for all the selected European countries and years.
Both the total network of roads and length of rail lines are expressed in kilometres.
Control variables
The choice of control variables was driven by the fact that there are other proximate sources of growth such as capital and labour force. This assumption is stemming from the Cobb Douglas production function approach (Egert et al., 2009; Del Bo et al., 2009; Mankiw et al., 1992; Gramlich, 1994).
Considering the above mentioned assumption, in order to obtain accurate results, two control variables are applied; the stock of real fixed capital and the total labour force. The total values of real fixed capital stock (1995 prices) were collected from the Socio Economic Accounts of the World Input-Output Database; while labour force is measured by participation rate (total employment levels per population with 15 years and over), data are drawn from Eurostat.
Spatial weight matrix
10
2. Model specification and estimation
To shed light on the long run effect of transport infrastructure in Europe, standard and spatial panel models are used. The main advantage of the panel model compared to the cross sectional and time series models is that it accounts for unobserved spatial and temporal heterogeneity (Baltagi, 2005). Not surprisingly, new and old member states of the European Union, and more generally spatial units, differ in their origin, religion, geographical location etc. These characteristics are usually considered as space specific and time invariant variables that affect the dependent variable. It is often difficult to measure and to account for these factors, but omitting them could bias the estimation (Elhorst, 2011). Using panel data makes it possible to correct for the above mentioned bias and it controls for the country specific effects (time invariant variables). In the panel models, time period specific effects can control for all spatial invariant variables and account for temporal heterogeneity (Baltagi, 2005). Considering these effects is crucially important in the view of the current empirical analysis, as during the period 1999-2009 European countries have experienced economic booms and busts that affected differently the examined countries’ economic growth. Thus these time specific effects can significantly influence the observations and so the results of the estimation (Elhorst, 2011).
2. a. Standard panels
The panel models are based on the line infrastructure augmented production function approach. The model construction begins with the following equation for all countries (i) and years (t).
Eq. (1)
Where Y represents the output, K is the capital, L accounts for the labour force and I denotes the infrastructure. These variables entered the equation as ratios, thus observations became comparable across countries:
Eq. (2)
11
Considering the production function approach, the line infrastructure system is assumed to be a factor of production. The underlying argument suggests that an increase in the level of infrastructure will enhance the production. Furthermore, besides the direct effect, it may increases the aggregate growth through external economies (Delgado et al., 2000; Egert et al., 2004; Mankiw et al., 1992; Gramlich, 1994); thus road networks and rail lines are the main explanatory variables.
The stock of physical capital and the labour participation rate are treated as control variables. It is crucially important to include these productive inputs as well, since these are the main ingredients of economic growth, therefore, considering them in the estimation leads to more accurate results (Hill et al, 2011;Delgado et al., 2000; Egert et al., 2004; Mankiw et al., 1992; Gramlich, 1994). While excluding them could lead to bias results due to the omitted variable problem.
As far as estimation issues are concerned, there could be other factors that may bias the accuracy of the outcomes of the estimated standard errors of the model, namely heteroskedasticity. The derivation of the basic model, however, may solve the problem (Eq.
2). To detect the presence of heteroskedasticity a Breusch-Pagan test is performed. This test
of heteroskedasticity is based on the variance function, thus according to the null hypothesis the data are homoskedastic when the variances of all observations are the same. Meanwhile the alternative hypothesis is that the data are heteroskedastic i.e. the variances are different. According to the result of the test (χ2
=70.55; p=0.000) the null hypothesis is rejected and one can conclude that heteroskedasticity exists in the model, thus new heteroskedasticity robust standard errors are obtained in order to correct for the bias and to be able to draw valid conclusions (Hill et al., 2011).
12
Accordingly, the derived equation expressed in computer shorthand notation is the following:
Eq. (3)
This model is first estimated by pooled Ordinary Least Square (OLS). Regarding the pooled OLS estimation the coefficients are assumed to be constant for all European countries in all time periods. Moreover, it does not allow for heterogeneity, the results of the estimation might be biased because it ignores the panel nature of the data. One remedy is to apply special panel regression models, such as fixed effects and random effects models (Hill et. al, 2011).
The fixed effects estimator controls for unobserved spatial and temporal heterogeneity. The country and time-period specific effects, which capture all the different characteristics, are treated as fixed effects (Elhorst, 2011). This estimation technique eliminates the country specific, time invariant characteristics from the explanatory variables and generates an unbiased model. Clearly, country specific time-constant unobserved heterogeneity no longer biases the results (Brüderl, 2005). Furthermore, a dummy variable is introduced in the fixed effects model for each country and for each year with one exception in order to avoid perfect multicollinearity (Hill et al, 2011).
In case of the random effects model, the country and time-period specific effects are treated as random variables with the underlying assumption that these variables and the error term ( are independent of each other. Furthermore, it also assumes that the country and time specific effects are independently and identically distributed with zero mean and variance and (Elhorst, 2011). In many cases, if the main assumption i.e. Cov( , ) = 0 is violated, the
random effect estimator will be biased, thus cannot solve the problem of unobserved heterogeneity (Brüderl, 2005).
13
analysis geographical units are examined, the individual intercepts are clearly fixed (Hill et al, 2011). As the difference in the compared coefficients is not systematic, and the outcomes are failed to meet the asymptotic assumptions of the Hausman test, an alternative test is needed to perform, an augmented Hausman test. After the estimation of the augmented model with random effect, and testing the significance of the variables, the result (χ2=147.95 and p= 0.000) shows that fixed effects model is the most suitable to use. Another advantage is the possibility of using cluster robust standard errors that solve the problem of heteroskedasticity (Hill et al., 2011).
Finally in order to investigate the possible differences of the new and old member states, the same estimation strategy is followed in subsamples. One of the subsamples covers the transition economies, while the other consists of the old member states. Furthermore, to see how the effects of the infrastructure development on economic growth vary across the transition economies before and after the accession to the European Union, two other subsamples are estimated with fixed effects.
2. b. Spatial panel models
So far standard panel models are considered, however, one of the aims of the current analysis is to control for the spatial interaction effects among geographical units by using spatial panel setting; and compare the estimated outcomes to the coefficients of the standard panel model (Fischer and Getis, 2010). Considering the linkages between European countries, due to the networks of line infrastructure, the spatial panel models may have the advantage of modelling the effects of the infrastructure development on economic growth in Europe more comprehensively, compared to the standard panel models (Elhorst, 2011).
14
Spatial lag model (SAR)
In order to capture the endogenous interaction effects among the dependent variables spatial lag model is introduced:
Eq. (4)
The variable is the spatially weighted dependent variable, where W is a row normalized spatial inverse-distance weight matrix. is called the spatial autoregressive coefficient (Elhorst, 2010, 2011; Del Bo et al., 2009; Drukker et al., 2011; Hughes, 2011, LeSage and Pace, 2009).
Spatial Durbin model (SDM)
The spatial Durbin model controls for both endogenous interaction effects among the dependent variable and the exogenous interaction effects among the independent variables.
Eq. (5)
In this case is the exogenous interaction effect among the independent variables (roads,
rails, capital and labour), while indicates the coefficients of them (Elhorst 2010, 2011; Del
Bo et al., 2009; Drukker et al., 2011; Hughes, 2011; LeSage and Pace, 2009).
Spatial error model (SEM)
Finally, interaction effects among the error terms can also be incorporated by considering spatial dependence in the error structure.
Where Eq. (6)
15
These models account for the possibility that the variables may be spatially linked among countries, thus by applying these spatial econometrics models more accurate results can be obtained (Del Bo et al., 2009). In order to fully exploit the advantages of the panel data, the spatial models (Eq. 4, 5, 6) are estimated by using fixed effects model with spatial and time period specific effects that control for all variables whose omission could bias the estimates (Elhorst, 2003, 2009, 2011).
III. Results
The results of the non spatial linear regression models and the spatial panel models are discussed in this section.
1. Standard panel models
Table 1 presents the main results of the non spatial linear regression which confirms the positive role of infrastructure. The model is estimated by using OLS and fixed time and country effects.
Table 1: Standard panel models – estimated with OLS and fixed effects
(1) OLS
(2) FE VARIABLES GDP/capita GDP/capita
Roads t-1 0.0367 0.125* (0.109) (0.0617) Rails t-1 0.230 0.254* (0.150) (0.116) Capital 0.970*** 0.962*** (0.0475) (0.107) Labour 0.622** 0.768*** (0.281) (0.212) Constant 0.0169*** 0.0166* (0.00511) (0.00824) R-squared 0.874 0.848 Number of country 11
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
16
fixed effects. Therefore, in Column (2), the model is estimated with the fixed effects estimator. In this case the road and rail variables have positive and significant effects at 10% on the GDP per capita growth, with the estimated coefficients of 0.125 and 0.254.
Considering the control variables, the results are in line with the expectations and verify the positive contribution to the GDP per capita growth. It is well seen that in all specifications the coefficients of the control variables are positive and significant at 1% and 5%.
The standard model, presented above, is estimated for the whole sample. In order to see the differences between the effects of the infrastructure on the new and old member states’ GDP per capita growth, the model is estimated for two subsamples in Table 2.
Table 2: Standard model for Eastern and Western countries – estimated by FE
(1) FE - CEE
(2)
FE - Western EU VARIABLES GDP/capita GDP/capita
Roads t-1 0.200* -0.0157 (0.0913) (0.0475) Rails t-1 0.381 0.0846 (0.249) (0.0935) Capital 0.908*** 1.101*** (0.157) (0.0233) Labour 0.739** 1.155*** (0.256) (0.0980) Constant 0.0361* -0.00587*** (0.0136) (0.00117) R-squared 0.845 0.954 Number of country 5 6
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
17
The results highlight the different effects of the infrastructure variables on GDP per capita growth between the transition economies and the Western countries. The outcomes partly support the assumption that less developed countries tend to have higher rates of return to infrastructure development than developed countries (Mankiw et al., 1992). However, the total physical capital and the labour participation rate have slightly higher positive effects on the old member states’ GDP per capita growth.
In order to see how the EU membership affected the correlation of the infrastructure variables and the GDP per capita growth, subsamples are estimated for the transition economies.
Table 3: Standard model for the Central Eastern European countries, before and after joining the European Union – estimated by FE
(1) FE – pre EU
(2) FE – post EU VARIABLES GDP/capita GDP/capita
Roads t-1 -0.483*** 0.198* (0.0258) (0.0916) Rails t-1 -0.007 0.495 (0.0337) (0.253) Capital 0.985*** 0.868** (0.0195) (0.208) Labour 0.775** 1.170** (0.1976) (0.298) Constant 0.0099*** 0.0226 (0.0017) (0.0150) R-squared 0.988 0.860 Number of country 5 5
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
18
2. Spatial panel models
Three different types of interaction effects were introduced in the methodology section, in order to explain why an observation on a given country may be dependent on observations at other countries. According to these effects, namely endogenous interaction, exogenous interaction and correlated effects; three different models are constructed.
Table 4 presents the results of these spatial panel models. In all cases the road variable shows positive significant impact on the GDP per capita growth. The highest coefficient of this variable, at 1% level of significance, is obtained in the spatial Durbin model (SDM) in Column (2), with a value approximately 0.144.
In contrast the coefficients of the rail lines variable are not significant, although they are positively correlated with the GDP per capita growth in each and every model. So far the results are in line with the expectation of the hypothesis.
Table 4: Spatial panel models – estimated by FE
VARIABLES (1) (2) (3) GDP/capita SAR SDM SEM
Roads t-1 0.121** 0.144* 0.111** (0.0596) (0.0738) (0.0497) Rails t-1 0.0104 0.0163 0.002 (0.0277) (0.0322) (0.0207) Capital 0.564*** 0.552*** 0.552*** (0.189) (0.186) (0.192) Labour 0.230 0.264** 0.194 (0.142) (0.124) (0.173) Spatial coefficient 0.152** 0.187*** 0.164* (0.0588) (0.0455) (0.111) Country fixed effects Yes Yes Yes Time fixed effects Yes Yes Yes
R-squared 0.866 0.853 0.851 Number of country 11 11 11
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
19
As far as the spatial parameter is concerned, it has positive and significant coefficient in all specifications; it takes the highest value in the spatial Durbin model (approximately 0.187).
The highest R2 (0.866) is provided by the SAR model, while the SDM has almost the same R2 (0.853) as the SEM model (0.851).
To decide which one of the spatial models is the most appropriate to describe the data, the Wald test is performed. The Wald test investigates the hypothesis whether the SDM model can be simplified to the spatial lag model, thus H0: θ=0. The second hypothesis tests if the
spatial Durbin model can be simplified to the spatial error model, thus H0: θ+ρβ=0. As the
SAR and SEM model are nested in the spatial Durbin model, the Wald test only requires estimating the unrestricted model (SDM) (Bruin, 2006; Korn and Graubard, 1990). According to the test result (F=23.97, p=0.000), both hypotheses are rejected, therefore the SDM model describes the data best. In addition to the Wald test results, there are other arguments in favour of the spatial Durbin model. This model accounts for the spatial dependence in the dependent variable (GDP per capita) and in the regressors, and thus it leads to more accurate estimates. Moreover the SDM accounts for the spatially lagged regressors (WX), which could serve as a control for omitted variables (Elhorst, 2010; LeSage and Pace, 2009; Belotti et al., 2013).
3. Comparison of the results of the standard and the
spatial panel model
Table 5 shows the different coefficient estimates for both the standard and the spatial model. Beyond the standard fixed effects model (Column (1)), the results of the spatial Durbin model can be seen as well. In Column (2.a) both country and time fixed effects are applied in the SDM estimation process. In Column (2.b) the direct effects can be seen, which show the impact of the given country’s explanatory variables on the given country’s GDP per capita growth. Column (2.c) represents the spatially lagged dependent variables, which are local indirect effects. The so called local spillovers are the effects of a given country’s explanatory variables on the nearby countries’ dependent variable. While Column (2.d) represents the global effects, i.e. how the given country’s explanatory variables affect not only the nearby countries’ GDP per capita growth, but the whole sample’s (Elhorst, 2010).
20
the estimated coefficients of the road variable are slightly different; however, they are higher in the spatial Durbin model with the approximate value 0.144
The estimates of the other main explanatory variables became lower and insignificant, when the spatial Durbin model (SDM) is considered.
Looking at the control variables, on the one hand, the coefficients represent positive and strongly significant correlation between physical capital and GDP growth, although it has negative insignificant local and global spillover effect. On the other hand, the estimated coefficients of the labour variable show positive and significant effects, in case of the standard fixed effects model and in the SDM model with both country and time fixed effects; moreover it has a significant direct effect as well.
Regarding the spatial effects, despite of the negative and insignificant coefficients of the spatially lagged independent variables’ (WX), the spatial coefficient is positive and strongly significant (0.187), thus spatial dependence exists.
Table 5: Comparison of the standard fixed effects model and the spatial Durbin model
VARIABLES (1) (2.a) (2.b) (2.c) (2.d) GDP FE SDM SDM Direct SDM Local (WX) SDM Global Roads t-1 0.125* 0.144* 0.149** -0.0203 0.0171 (0.0617) (0.0738) (0.0722) (0.0343) (0.0434) Rails t-1 0.254* 0.0163 0.0195 0.0263 0.0367 (0.116) (0.0322) (0.0302) (0.0262) (0.0324) Capital 0.962*** 0.552*** 0.547*** -0.0996 -0.0148 (0.107) (0.186) (0.203) (0.0708) (0.0933) Labour 0.768*** 0.264** 0.303*** 0.294 0.438 (0.212) (0.124) (0.108) (0.234) (0.290) Spatial coefficient 0.187*** (0.0455) Constant 0.0226 (0.0150) Country fixed effects
Yes Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes Yes R-squared 0.848 0.853
Number of country
11 11
21
Considering the R2 statistics the spatial Durbin model has somewhat similar R2 with the value of approximately 0.86, while in case of the standard fixed effects model this value is slightly lower (0.85), thus the SDM provides a better fit of the data.
The presented results confirm the hypothesis that the effects of the road network and rail lines on long term growth are positive. According to the estimated coefficients, only the road network has a significant effect on the GDP per capita growth. In case of rail lines, significant impacts cannot be seen in each and every specification.
IV. Conclusion
The current empirical analysis examined the long term relationship between infrastructure development and economic growth by using standard and spatial panel models on the basis of the production function approach.
Considering data from 1999-2009 for eleven European countries, the results reveal evidence of a positive effect of the transportation infrastructure on the long term growth; however, it is only robust in case of the road networks. When only the subsamples are considered, the results of the standard panel models suggest that the old member states of the European Union tend to have lower rates of return to infrastructure stock, compared to the new member states. This may stem from the diminishing returns to infrastructure development across the developed countries. The time period considered allowed accounting for the impact of the infrastructure stock before and after the membership of the European Union, in case of the new member states. Appropriately, the results are shown that after the transition economies joined the EU, they experienced higher positive and significant impact of the transport infrastructure on the GDP per capita growth.
22
23
V. References
Aschauer D.A. (1989). Is Public Expenditure Productive? Journal of Monetary Economics, vol. 23, pp. 177–200.
Atella V., Belotti F., Depalo D. and Mortari A. P. (2013). Measuring spatial effects in presence of institutional constraints: the case of Italian Local Health Authority expenditure. CEIS Tor Vergata Research Paper, vol. 11, issue 8, no. 278.
Baltagi B.H. (2005). Econometric analysis of panel data. 3rd edition. Wiley, Chichester. Brüderl, J. (2005). Panel Data Analysis. University of Mannheim. http://www.sowi.uni-mannheim.de/lehrstuehle/lessm/veranst/Panelanalyse.pdf
Bruin J. (2006). Newtest: command to compute new test. UCLA: Statistical Consulting Group. http://www.ats.ucla.edu/stat/stata/ado/analysis/
Calderon C. and Serven L. (2004). The effects of infrastructure development on growth and income distribution. World Bank Policy Research Working Paper, no. 3400.
Debarsy N. and Ertur C. (2006). The European Enlargement Process and Regional Convergence Revisited: Spatial Effects Still Matter. European Regional Science Association ERSA conference papers, no. ersa06p198.
Del Bo C., Florio M. and Manzi G. (2009). Regional infrastructure and convergence: Growth implications in a spatial framework. Transition Studies Review, vol. 17, pp. 475-493.
Delgado M. J. and Alvarez I. (2000). The effect of public infrastructure on private activity: Evidence from the Spanish regions. Universidad Complutense de Madrid, Documentos del Institutio Complutense de Analisis Economico, no. 0103.
Dreher A. (2006). The IMF and Economic Growth: The Effects of Programs, Loans, and Compliance with Conditionality, World Development, vol. 34, no. 5, pp. 769-788.
Drukker D. M., Peng H., Prucha I. R. and Raciborski R. (2011). Creating and managing spatial-weighting matrices using the spmat command. Statistical Software Components, Boston College Department of Economics.
Égert B., Koźluk T. and Sutherland D. (2009). Infrastructure and Growth: Empirical Evidence. OECD - CESifo working paper no. 2700, category 6: Fiscal policy, macroeconomics and growth. CESifo Group Munich.
Elhorst J.P. (2003). Specification and estimation of spatial panel data models. International Regional Science Review, vol. 26, no. 3, pp. 244-268.
24
Elhorst J.P. (2010). Applied spatial econometrics: Raising the bar. Taylor & Francis Group. Spatial Economic Analysis, vol. 5, no. 1., pp. 9-28.
Elhorst J.P. (2011). Spatial Panel Data Models.
http://www.york.ac.uk/media/economics/documents/seminars/2011-12/Elhorst_November2011.pdf
ERDF regulation (2006). Regulation (EC) No 1080/2006 of the European Parliament and of the Council of 5 July 2006 on the European Regional Development Fund and Repealing Regulation (EC) No 1783/1999. Official Journal L 210 of 31.7.2006.
ESPON (2003). Territorial impact of EU transport and TEN Policies. Second interim report of Action 2.1.1 of the European Spatial Planning Observation Network (ESPON).
Eurostat (2011). Purchasing Power Parities. Reference Metadata in Euro SMX Metadata Structure (ESMS). Eurostat, the statistical office of the European Union.
http://epp.eurostat.ec.europa.eu/cache/ITY_SDDS/EN/prc_ppp_esms.htm.
Eurostat (2013). Table of employment rate (15-64 years)
http://epp.eurostat.ec.europa.eu/portal/page/portal/employment_unemployment_lfs/data/datab ase.
Eurostat (2013). Table of GDP per inhabitant (euro).
http://epp.eurostat.ec.europa.eu/portal/page/portal/national_accounts/data/database. Eurostat (2013). Table of length of motorways, other roads (km).
http://epp.eurostat.ec.europa.eu/portal/page/portal/transport/data/database.
Fischer M. M. and Getis A. (2010). Handbook of Applied Spatial Analysis. Springer Heidelberg Dordrecht London New York.
Gramlich E. M. (1994). Infrastructure investment: A review essay. Journal of Economic Literature, vol. 32, no. 3, pp. 1176-1196.
Hill R. C., Griffiths W. E. and Lim G. C. (2011). Principles of Econometrics. John Wiley & Sons. Fourth Edition. International Student Version.
Hoskisson E. R., Wright M., Filatotchev I. and Peng W. M. (2012). Emerging multinationals from mid-range economies: The influence of institutions and factor markets. Journal of Management Studies. (In Press) downloadable at:
http://onlinelibrary.wiley.com/doi/10.1111/j.1467-6486.2012.01085.x/full
Hughes G. (2011). Implementing procedures for spatial panel econometrics in Stata. Stata User Group Meeting, 15th September 2011.
http://www.stata.com/meeting/uk11/abstracts/UK11_hughes.pdf
25
La Ferrara E. L. and Marcellino M. (2000). TFP, Costs, and Public Infrastructure: An Equivocal Relationship. IGIER (Innocenzo Gasparini Institute for Economic Research), Bocconi University. Working Paper, no. 176.
Lakshamanan T. R. (2008). The Wider Economic Benefits of Transportation: An Overview. Discussion paper. The wider economic benefits of transport: Macro-, meso- and micro-economic transport planning and investment tools. OECD and International Transport Forum. Round table: 140, pp. 51-64.
LeSage J. P. and Pace R. K. (2009). Introduction to Spatial Econometrics. Boca Raton, FL: CRC Press Taylor & Francis Group.
LeSage, J. P. (2008). An Introduction to Spatial Econometrics. Revue d'Economie Industrielle, 2008, vol. 123, no. 3, pp. 19-44.
Mankiw N. G., Romer D. and Weil D. N. (1992). A contribution to the empirics of economic growth. The Quarterly Journal of Economics, vol. 107, no. 2, pp. 407-437.
Oosterhaven J. and Knaap T. (2003). Spatial Economic Impacts of Transport Infrastructure Investments. Appeared in: Pearman A., Mackie P. and Nellthorp J. (eds) Transport Projects, Programmes and Policies: Evaluation Needs and Capabilities, pp. 87-1.
Romp W., De Haan J. and Sturm J. E. (2007). Public Capital and Economic Growth: Key Issues for Europe. In Public Investment and Public-Private Partnerships: Addressing Infrastructure Challenges and Managing Fiscal Risks, (eds) Schwartz G., Corbacho A. and Funke K., Washington, D.C.: IMF / Palgrave MacMillan, Chapter 1, pp. 11-20.
Sturm J. E., Jacobs J. and Groote P. (1999). Output Effects of Infrastructure Investment in the Netherlands, 1853–1913. Journal of Macroeconomics, vol. 21, no. 2, pp. 355–380.
The World Bank Database (2013). Table of length of rail lines (km).
http://data.worldbank.org/indicator/IS.RRS.TOTL.KM.
The World Bank Database (2013). Table of length of roads (total network (km)).
http://data.worldbank.org/indicator/IS.ROD.TOTL.KM.
The World Input Output Database (WIOD): Contents, Sources and Methods (2012). Table of real fixed capital stock (1995 prices). Socio-Economic accounts
http://www.wiod.org/database/sea.htm.
Timmer M., O’Mahony and van Ark B. (2007). The EU KLEMS Growth and Productivity Accounts: Overview. University of Groningen & University of Birmingham. downloadable at www.euklems.net.
Tobler W. (1970). A computer movie simulating urban growth in the Detroit region. Economic Geography, vol. 46, no. 2, pp. 234-240.
26
Appendix
1. Distance matrix – between population centroids (metres)
2. Properties of the inverse-distance row normalized spatial weight matrix Inverse-distance row normalized
matrix Dimensions 11*11 Values min 0 min>0 0.0365 mean 0.0909 max 0.2976
Countries Austria Belgium
Czech
