Transportation Research Part A 74 (2015) 73–90

Contents lists available at ScienceDirect

Transportation Research Part A

journal homepage: www.elsevier.com/locate/tra

Transport infrastructure and long-run economic growth

in OECD countries
Minoo Farhadi ⇑
Department of Econometrics and Business Statistics, Monash University, Victoria 3800, Australia

a r t i c l e i n f o a b s t r a c t

Article history: This paper provides an empirical evaluation of the growth impact of public infrastructure
Received 3 March 2014 in a panel of 18 OECD countries during 1870–2009. This study goes beyond the traditional
Received in revised form 31 January 2015 analysis of growth accounting models by exploring the indirect effect of stock of core
Accepted 9 February 2015
infrastructure on output growth through its impact on productivity. Constructing a long-
Available online 6 March 2015
run historical dataset on infrastructural capital formation spanning from 1870, estimated
results show that growth in both labour productivity and total factor productivity are
Keywords:
positively, but not substantially, influenced by growth in the stock of infrastructure. Fur-
Economic growth
Public expenditure
thermore, applying the system GMM technique (Generalised Method of Moments) revels
Infrastructure that although rate of returns to investment in infrastructure exceed the private rate in
Productivity OECD countries, it is not as high as positive externalities associated with investment in
Returns rate equipment and structure investment.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction

An important part of government expenditure is the finance and maintenance of infrastructure capital as one of the
productive sectors in any economy. Public infrastructure forms a means by which governments can effectively promote
economic growth, as it performs a vital role in stimulating countries’ economies, so much so that the World Bank refers
to public capital, especially infrastructure, as the ‘‘wheels’’ of any economic activity (World Bank, 1994). Amongst different
types of infrastructure, transport as a productive public expenditure is one of the critical sectors in any economy, since an
economy can benefit from transport facilities by accelerating access to the services, increasing the market mobility, saving
time and reducing business costs.
The purpose of this paper is to provide empirical evidence for investigating the influence of government expenditures on
infrastructure in stimulating the economic performance of OECD countries. To address this, it is crucial to know whether an
increase in public investment in infrastructure have an indirect and significant effect on the long-run economic growth of
OECD countries by increasing their productivity. In order to answer this question, this study empirically tests a growth
model in which the connection between productivity and public spending in infrastructure is examined. Furthermore, the
rate of returns to infrastructure investments is also investigated, so as to determine the extent to which rate of returns to
investments in the transport sector exceed its private returns amongst the OECD members.
This paper contributes to the literature in three ways. The first one is the construction of a long-run historical dataset on
infrastructural capital formation, over the last 140 years for a panel of 18 OECD countries. To the best of my knowledge, the
existing literature has only covered post-war periods for cross-section studies, from 1950 onward, and this is the first time

⇑ Tel.: +61 4 4997 7360.

E-mail address: [email protected]

http://dx.doi.org/10.1016/j.tra.2015.02.006
0965-8564/Ó 2015 Elsevier Ltd. All rights reserved.
74 M. Farhadi / Transportation Research Part A 74 (2015) 73–90

that a longer perspective is taken into account in analysing the links between investment in infrastructure and economic
growth.
Secondly, this paper presents a model in which the role of stock of infrastructure investment as an influential factor of
economic growth is examined through productivity gains. While a few studies have addressed the indirect impact of infras-
tructure on economic growth through raising total factor productivity (TFP) or labour productivity, they only considered
regional or industrial levels, and did not take into account cross-country levels (see Hulten and Schwab, 1991; La Ferrara
and Marcellino, 2000). It is worth mentioning that two different measurements of productivity are used in this study, namely
labour productivity and total factor productivity. This is done to enable us to analyse the differences in the impact of public
infrastructure on each measure over the time period considered, and across selected countries.
Finally, in contrast to the existing literature, which has used ‘‘total public expenditure’’ as a proxy for infrastructure invest-
ment, this paper focuses on a more precise classification of infrastructure called ‘‘core infrastructure’’ which includes roads,
highways, airports, railways, inland waterways and public transport, on productivity growth. Public investment expenditure,
which has been used in many previous studies, cannot properly represent infrastructure investment, since, rather than only
taking into account infrastructure, it contains more dimensions, and therefore is not a reliable proxy (Gramlich, 1994;
Fernández and Montuenga-Gómez, 2003). In line with what was mentioned earlier, this paper demonstrates that the final
impact of infrastructure on economic growth is due, not only to the input role that infrastructure has in the production func-
tion as a public good, but also to the indirect effect of capital infrastructure in shifting productivity in an economy in various
ways, such as improvements in learning by doing, increasing the efficiency of labour, and saving working time.
The paper is organized as follows. Next section reviews the existing literature on the links between investment in public
capital and economic growth. In Section 3 the anatomy of public infrastructure growth over the last 140 years is presented.
Section 4 describes the model specification, variables, and the methodology, while Section 5 devoted to the empirical results
and the robustness check of investigating the link between infrastructure investment and productivity. Section 6 is devoted
to exploring the social rate of returns to investment in transport infrastructure. Last section concludes the paper.

2. Infrastructures and economic growth: a review

Following the publication of the seminal work of Barro (1991), who introduced government spending into the production
function as a public good, many scholars focused on the impacts of fiscal policy, through public expenditure and taxation, on
long-run economic growth. Prior to Barro, the study of infrastructure as a productive sector dates back to the pioneering con-
tribution of Aschauer (1989), in which he highlighted the importance of productive government services for the US economy
in the 1970s, although his results were challenged by scholars such as Gramlich (1994) and Sturm (1998), who proved that
the actual growth impact of real government capital is much smaller than what Aschauer reported.
According to Romp and De Haan (2007), the existing empirical literature on public capital can be grouped into four
methodological strands, including production function approach, cost function approach, growth models, and Vector
Auto-Regressive Models. Moreover, each approach can be categorized into three different subsections: (1) national and
cross-country, (2) regional, and (3) industry studies.
First, the production function approach, which models the amount of output that can be produced for each factor of pro-
duction. Assuming technological constraints, public infrastructure enters these models as an input supported by government
(see, for instance, for cross-country studies: Esfahani and Ramı´rez, 2003; Boopen, 2006; for regional studies see, Picci, 1999;
La Ferrara and Marcellino, 2000; and for industry studies see, Fernald, 1999; Yeoh and Stansel, 2013). Second, the cost func-
tion approach, which assumes infrastructure as a free input provided by the government with the cost savings impact. The
main aim of studies which used this approach is to determine whether or not increases in infrastructure endowment
decrease the cost of output (see for cross-country studies: Demetriades and Mamuneas, 2000; Loizides and Tsionas, 2002;
for regional studies, La Ferrara and Marcellino, 2000; Vijverberg et al., 2011; and for industry studies, Moreno et al.,
2003). Third, the endogenous growth models, which try to investigate the growth impact of infrastructure, based on the idea
that economic growth is not driven merely by exogenous factors, but is in fact an endogenous phenomenon (see for cross-
country studies: Easterly and Rebelo, 1993; Égert et al., 2009; for regional studies, Mas et al., 1996; Jayme et al., 2009; and for
Industry Studies, Cellini and Torrisi, 2009). Finally, data-oriented models which do not rely heavily on economic theory but
analyse the relationships between several data series, including infrastructures and GDP (for cross-country studies see,
Kamps, 2005; Ghani and Din, 2006; Guo et al., 2011 and for regional studies, Herranz-Loncán, 2007; Márquez et al., 2011).
To conclude, although many empirical studies have examined the growth impact of public infrastructure, but an overall
review of the literature highlighted the existing gaps in this area of research, suggesting the need for a new dimension of
time, necessity of using a proper proxy for infrastructure investment, and applying a new growth model in which infrastruc-
ture investment is examined through productivity gains. The following sections attempt to address the above mentioned
issues.

3. Anatomy of public infrastructure growth over the last 140 years

Before examining the productivity impact of infrastructure investment, it is worth looking at the long run trend of public
spending on infrastructure and its decomposition results for the 18 OECD countries, namely, Australia, Belgium, Canada,
 M. Farhadi / Transportation Research Part A 74 (2015) 73–90 75

0.06

% of GDP
0.04

0.02

0
1870 1890 1910 1930 1950 1970 1990 2010
Year

Fig. 1. Trend of government expenditure in infrastructure as a percentage share of GDP (1870–2009).

Table 1
Average annual growth rates in infrastructure investment in OECD countries.

Average annual growth rate (1870–1913) Second Industrial Revolution 1.253

Average annual growth rate (1913–1950) World War period 0.864
Average annual growth rate (1950–1973) Golden period 2.242
Average annual growth rate (1973–2009) Period of new opportunities 0.247
Average annual growth rate (1870–2009) 0.924

Note: The average geometric annual growth rate (as a share of GDP) is based on infrastructure investment
data from 1870 to 2009 for 18 OECD countries.

Denmark, Finland, France, Germany, Ireland, Italy, Japan, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, the UK
and the USA, over the period 1870–2009 based on the dataset constructed for this study.1
Fig. 1 provides some information on the yearly trend of government expenditure on core infrastructure as a share of GDP
over the last 140 years. Except for a small reduction in trend for the first three years, the rest of the nineteenth century expe-
rienced a moderate increase in public investment. This is due to the great investment efforts that led to the spread of sub-
stantial railway networks across the OECD countries, especially in the Western Europe (Carreras and Tafunell, 2006).
Improvements in infrastructure investment took a step forward at the end of the 19th century, with the Second Industrial
Revolution leading to a period of stability and development. However, the start of World War I in 1914 was the end of such
enhancements, and instead, uncertainty affected all kinds of investments. Although a recovery in infrastructure expenditure
started soon after WWI, it was short-lived, as the great depression and the occurrence of WWII again pushed the economy
backward and provided an unstable and volatile environment for investment (see, Carreras and Tafunell, 2006).
The post-war years are commonly considered a golden age period for the world’s economies, especially for the OECD
countries (Madsen, 2008). Therefore, the significant rise in public infrastructure spending is due to the extraordinarily suc-
cessful recovery efforts soon after WWII. From 1973 to 1990, infrastructure investments remained fairly stable, followed by
some small fluctuations, after which we can see a reduction in the total trend of public investment, up to 2009. To gain
insights into the long-term trend of infrastructure spending, the average geometric growth rate for the four major periods,
1870–1913, 1913–1950, 1950–1973, and 1973–2009 is calculated (see Table 1).
As can be seen, the government investment in core infrastructure was high and considerable (around 1.25%) in the late
nineteenth century and the first decade of twentieth. As has been mentioned, the period 1870–1913 is identified as a time of
stability following the Second Industrial Revolution, accompanied by industrialization and urban development. In contrast,
the average annual growth rate dropped to 0.86% for the period 1913–1950, which is not surprising, due to the previously-
mentioned harmful wars, extreme inflations and deflations and great depression which pushed the economies backward
during this period, thus reducing the amount of investment in various sectors.
As is shown in the third row, the average annual growth rate over the post-war period increased considerably, to 2.24%. In
fact, the golden age years helped economies to recover from the damage caused by the wars by extraordinary efforts in
increasing the amount of investments on machinery, building and infrastructure. In contrast, from 1973 onward, govern-
ment spending on infrastructure dropped significantly, by almost 2%, reaching the lowest average annual growth rate
amongst all of the periods considered. To sum up, comparing the results over these significant periods of time shows that
this average annual growth rates of government spending on infrastructure for the period (1913–1950) and from 1973
onward are lower than the total average of this rate over the entire period (1870–2009). However, it is interesting to see that
the highest rate of government expenditure on transportation, far above the total average annual growth rate over the period
1870–2009, occurred during the golden age.
In the next step, in order to see how this trend worked out within each specific country, the average annual growth rates
for some of the individual countries are decomposed. As can be seen in Table 2, following the same pattern as before, the

1
The construction of data is detailed in the data section and Data Appendix.
76 M. Farhadi / Transportation Research Part A 74 (2015) 73–90

Table 2
The average annual growth rate for individual countries.

Average annual growth rate AUS DEN FRN GER ITA NET POR SPA UK USA
1870–1913 1.811 2.737 0.366 1.464 0.494 0.775 1.005 0.023 1.412 1.024
1913–1950 0.346 0.019 4.531 1.657 1.058 0.794 2.127 0.0311 0.661 0.313
1950–1973 2.412 5.155 2.529 3.311 1.950 0.357 2.258 0.386 6.581 3.252
1973–2009 1.736 0.882 0.079 2.2 0.961 1.694 1.182 1.796 0.085 0.502

Note: The average geometric annual growth rates are conducted for Australia, Denmark, France, Germany, Italy, the Netherlands, Portugal, Spain, the UK and
the USA.

investment ratio was high and considerable during the industrialisation period, while this rate decreased significantly in the
war period for the majority of countries. With the exception of Spain and Portugal, a huge reduction in the growth rate of
public spending on infrastructure took place over the period 1973–2009. This downward trend increases our awareness
of the negative productivity effect of such an expenditure decline on countries’ economic growth rates (Harchaoui et al.,
2004; Vuchelen and Caekelbergh, 2010).
In order to have a closer look at the infrastructure expenditure trend over the past 140 years, countries, which have a few
exceptions in their average annual growth rates, are discussed in more detail.
In Spain, starting in 1875, the main investment in infrastructure was allocated to the provision of railroads, with other
types of infrastructure taking small fractions. Since 1895, the importance of investments in ports, communications, urban
transport and electricity distribution has increased. During the period of 1923–1936, government expenditure was devoted
to roads, ports and hydraulic transport, while the private sector was responsible for the huge investments in telecommuni-
cations, electricity, and urban transport. This huge effort in the infrastructure sector in the late nineteenth and early twen-
tieth centuries had a prominent impact on the productivity of the Spanish economy (see, Herranz-Loncán, 2007). It is worth
mentioning that Spain and Portugal are amongst those less-developed economies which have received substantial subsidies
following the Cohesion Policy. This policy was established to encourage investment, improve long-run economic growth and
promote equality amongst countries within the European Union (See amongst others, Sosvilla-Rivero and Herce, 2008). As a
result, while in majority of OECD countries investment in infrastructure has declined over the last two decades, Spain’s pub-
lic capital as a share of GDP has increased. The significant influence of such this investment on the productivity growth of
Spanish economy has been shown in various different studies (amongst others, see Fernández and Montuenga-Gómez,
2003; Finley, 2005).
In the USA, the federal government established and developed various infrastructure projects, such as transportation,
telegraph and railroad networks, in the late nineteenth century and early twentieth century in order to enhance the econ-
omy. However, in contrast to most European and South Pacific countries, remarkable efforts in infrastructure investment also
took place in the war periods, in order to increase America’s national security and military power. In fact, various types of
infrastructure, such as railroads, airfields, military facilities and highways, were a target of government investment during
WWI and WWII, and especially the fundamental efforts which were undertaken in 1916, 1921 and 1956, which significantly
boosted the economic growth of the USA (Gibson, 2011). Over the post-war period (1950–1979), the average annual growth
rate of public infrastructure investment rose at the same rate as the economic growth of the USA, while these rates both fell
together afterwards (Heintz et al., 2009).
Focusing on the Netherlands, one of the main public expenditures in the infrastructure sector started in 1860 with the
construction of the national railway network, followed by the expansion of this project to thousands of kilometres of railway
across the country by 1885. After this, the government’s initiative in the provision of waterways and shipping channels was
considerable, and the investment in this sector peaked in 1870, 1890 and 1907. In fact, its infrastructure endowment in the
second half of the nineteenth century made the Netherlands an economic super-power in the pre-industrial period through
the introduction of a new transport system and the construction of modern infrastructure. This endowment continued for
the following decades, and increased the role of infrastructure in the growth of productivity in this country over time
(see, Sturm et al., (1999) and Groote et al.,(1999) for comprehensive information about infrastructure development in
Netherlands back to 1853).
In France for many years investment level of infrastructure was about 1% of the GDP (Short and Kopp, 2005). French rail-
way played an important role during the war period and became one of the most efficient ways to get the troops across the
border, providing mobilisation of weapons and equipment and in general meeting all the military requirements. Further than
investing in its own rail network, France made a strategic relationship with Russia as a military alliance called the Franco-
Russian Alliance (finalised in 1894) which provided Russia with substantial loans, industrial enterprise, and construction of
strategic railways (Clark, 2012). After the World War II, all damaged railways were repaired or rebuilt due to their crucial
role to the mobility of French people (see Meunier, 2002).
Thus far, we have tried to decompose the long-run trend of public infrastructure growth into different periods of time, in
order to get a general idea of its movements and tendencies over time and across selected OECD countries. In the next sec-
tion, the labour productivity growth model, which was explained comprehensively in the previous section, will be estimated
in order to examine the indirect impact of public infrastructure on economic growth.
 M. Farhadi / Transportation Research Part A 74 (2015) 73–90 77

4. Empirical model, data and methodology

4.1. Empirical framework

To examine the proposed effect of public infrastructure expenditure on growth productivity, this study extended the
endogenous growth model introduced by Madsen (2010), in which productivity is driven by various different economic vari-
ables. The model underlying the empirical estimation can be described as follows:
 d  f
X X
D ln ðY=LÞit ¼ u1 þ u2 DTFit þ u3 DEDUit þ u4 D ln Sdit þ u5 ln þ u6 D ln Sitf þ u7 ln þ u8 POP
Q it Q it
þ u9 D ln INFRit þ Z it þ D þ eit ð1Þ

where D is a five-year difference operator, and the superscripts d and f stand for domestic and foreign, respectively. We use i
to index counties and t to index time. In Eq. (1), the dependent variable is economy-wide output per hour worked (Y/L); EDU
stands for educational attainment, S is the stock of knowledge, X is innovative activity, measured by patent applications as a
reasonably proper measure of research activity (Griliches, 1990); Q is the product variety, and (X/Q) is research intensity. DTF
is the distance to the frontier (the USA and/or the UK in our case). POP is population growth drag, Z is a set of control vari-
ables, D is a dummy variable, and e is a stochastic error term. INFR, which is the capital stock of transportation provided by
the government as a share of GDP, is added to the above model in order to examine its impact on the growth of labour pro-
ductivity. Construction of infrastructure and the other key variables is explained in data section and the Data Appendix.
Since Eq. (1) outlines the growth model which integrates various hypotheses on the determinants of productivity growth,
clarifying the reasons for choosing these variables are desirable and needed to be comprehensively explained.

4.1.1. Domestic and foreign knowledge (Sd,f, (X/Q)d,f)

Consider the following production function:

Y ¼ AK a L1a ;
where Y is real output, and A, K, and L are knowledge, capital, and labour, respectively. Based on the endogenous growth
theories, the growth in knowledge (gA) can be given by the following function (Ha and Howitt, 2007; Madsen, 2008):
 r
A_ t X
gA ¼ ¼c A/1 ; 0 < r 6 1; /61
At Q

Q / Lb in steady state;
where Q is product varieties, X is innovative activities, r is a duplication parameter which is equal to 1 if there are no
replications for new products and 0 if all innovations are duplications, c is a R&D productivity parameter, X/Q is known
as the research intensity, b is the coefficient of product proliferation, and finally, / represents knowledge production’s
returns to scale. To distinguish between different theories, it should be mentioned that, according to semi-endogenous
growth models, X is R&D input, / < 1 and b = 0, while based on Schumpeterian growth theories, X is the productivity adjusted
R&D, / = 1 and b = 1.
According to the recent endogenous growth models (Grossman and Helpman, 1991; Romer, 1990), international transi-
tion of knowledge can be measured by both patent and research intensity since increasing the variety and the quality of
intermediate inputs during the production process increases the efficiency of total factor productivity. Moreover, following
the seminal work of Coe and Helpman (1995), many empirical studies show that the productivity level of an economy
depends not only on its cumulative domestic capital, but also on foreign R&D capital (see for instance, Park, 1995;
Engelbrecht, 1997; Eaton and Kortum, 2001; Lumenga-Neso et al., 2005; Madsen, 2007, 2008; Acharya and Keller, 2009).

4.1.2. Distance to the frontier (DTF)

DTF is the distance to the frontier, calculated as DTF ¼ ½At1  At1 =At1 , where A is TFP at the world technological frontier.
In our case the highest of the US or the UK at a given point in time is taken into account. The UK was the frontier country
from 1870 up to 1906, the UK or the USA between 1906 and 1931, and the USA afterwards. According to Howitt (2000),
Aghion and Howitt (2006), Griffith et al. (2003), and Madsen (2008), the further away a country is from the world techno-
logical frontier, the better its chances of adopting new technologies, which are developed at the frontier, at lower costs, and
converging towards the technological frontier. Since the effective costs of innovations for these countries are lower, so the
rate of growth is faster than in countries which are located closer to the technological frontier (Aghion et al., 2005; Acemoglu
et al., 2006).

4.1.3. Human capital and its interaction with DTF (EDU.DTF)

Beside the importance of human capital as one of the driving factors of economic growth (Romer, 1990; Aghion and
Howitt, 1992, 1998), Nelson and Phelps (1966) believe that the interaction between human capital and the distance to
78 M. Farhadi / Transportation Research Part A 74 (2015) 73–90

Table 3
Descriptive statistics of the entire sample (1870–2009).

Y/L DTF EDU POP Sdit (X/Q)dit SImp

it (X/Q)Imp
it INFR
Mean 0.108 1.007 0.693 0.246 0.190 0.560 0.215 0.304 0.103
Median 0.103 0.688 0.625 0.158 0.111 0.601 0.191 0.385 0.065
Maximum 0.722 5.782 1.735 3.460 2.123 0.799 6.236 0.719 6.493
Minimum 0.712 0.280 0.020 1.741 0.651 0.345 1.132 0.519 0.076
Std. dev. 0.120 1.005 0.380 0.413 0.337 0.176 0.561 0.290 0.339

Note: Y/L is labour productivity, DTF is the distance to the frontier, EDU is educational attainments, POP is population growth drag, Sd and Simp are stocks of
domestic and imported knowledge, and X=Q d and X=Q imp are the domestic and imported research intensity, respectively. INFR is the capital stock of
infrastructure as a share of the GDP. The data period is from 1875 to 2009 for 18 OECD countries. The results are generated over time and countries and time
is considered annually.

the frontier also affects TFP growth. Therefore, the growth of total factor productivity (gA) is a function of the interaction
between human capital and the gap between the technology level (Tt) and the level of theoretical knowledge (At):
 
A_ t T t  At 
gA ¼ ¼ @ðEDUÞ @ð0Þ ¼ 0; @ðEDUÞ > 0:
At At

Benhabib and Spiegel (1994), by using this model but allowing a ‘‘catch-up’’ to the technology of leading countries, specify
the TFP growth as follows:
 max 
A_ t A
gA ¼ ¼ @ðEDUÞ t1 ;
At At1

where Amax
t1 is the total factor productivity of leading countries. This equation shows that countries can benefit from the level
of their human capital in two ways: first, through developing their own inventions, and second, by increasing their ability to
observe and adopt more advanced technologies which are produced elsewhere. Therefore, the more educated the labour
force a country has, the higher the TFP growth it can gain, and the faster it will catch up to the advanced technology
(Aghion et al., 2009; Madsen, 2010).

4.1.4. Interaction of research intensity with DTF ((X/Q)d.DTF)

The interaction between the distance to the frontier and research intensity has also attracted some attention. According
to Howitt (2000) and Griffith et al. (2003), besides the direct impact of domestic research intensity on economic growth, R&D
can influence TFP through its interaction with countries’ distance from the technology frontier by the following function:
   max     max 
A_ t X A X At1
gA ¼ ¼a þ b ln t1 þ c ;
At Q t1 At1 Q t1 At1

It should be mentioned that DTF as an explanatory variable is added alone into the baseline model (Eq. (1)), while these two
absorptive capacity terms (DTF.EDU and DTF.(X/Q)) are added for robustness check in order to test the sensitivity of results.

4.1.5. Population growth drag (POP)

Population growth is a drag on economic growth since land, as a fixed factor of production, has diminishing returns.
Unlike reproducibility of other types of capital, land, as a perfectly inelastic factor, cannot be expanded along with increasing
demand driven by population growth (Madsen, 2010).

4.2. Data

Eq. (1) is estimated over the period 1870–2009 for a panel of 18 OECD countries listed in the first paragraph in Section 3.
The summary statistics for the entire sample (Table 3) show that the estimated coefficients are not affected by any extreme
observations, since all of the variables are in the same range.2
In order to reduce the influence of business cycle, the data are reconstructed in 5-year differences; and the level variables
are measured as averages of the five years. The data are constructed as follows (see also the Data Appendix for more details):
Productivity: we have used two commonly-applied measures of productivity, namely labour productivity (Y/L) and total
factor productivity (TFP). The former is used as the dependent variable in our basic model and the latter is used as an alter-
native measurement in order to check the robustness of the results. Labour productivity is measured as the output per hour
worked, and the total factor productivity is constructed based on the homogeneous Cobb-Douglas technology assumption,
a ð1a Þ
ðY it =Liti K it i ), where Y is the real GDP, K is non-residential capital stock, L is labour input, measured as the economy-wide
employment multiplied by annual hours worked, and ai is the labour’s share of income.

2
The summary statistics of the key variables is presented in the Data Appendix.
 M. Farhadi / Transportation Research Part A 74 (2015) 73–90 79

Capital (K): capital stock is measured as the sum of the total investment in buildings and machinery. Stock of capital is
constructed by applying the perpetual inventory method with a depreciation rate of 3% for buildings and structures and
17% for machinery and equipment, following Madsen (2008).
Domestic stock of knowledge (S)d: patents are measured as patents granted to residents, and the perpetual inventory method,
with a depreciation rate of 15%, is applied for computing the knowledge stock. The initial stock of knowledge is constructed as
the number of patents in 1870 divided by the average geometric annual growth in patents over the period 1870–2009.
Domestic research intensity (X/Q)d: research intensity is measured as patents divided by product varieties, to test for the
Schumpeterian growth theories and allow for the diminishing returns to R&D discussed by Ha and Howitt (2007). For
product variety and under the Schumpeterian growth theory, we follow Howitt and Aghion (1998) and Madsen (2008)
and use employment.
International knowledge (Sf, (X/Q)f): imported knowledge from country i to country j is calculated as follow:
 imp X 18   X d
X
¼ mijt þ ð1  qÞmSij;t1
Q it j¼1
Q jt
18 
X 
Simp
it ¼ mijt þ ð1  qÞmSij;t1 Sdjt ;
j¼1

where the superscripts d stand for domestic, i and j refer to the recipient and the source country, respectively; Mij is accu-
mulated imports of goods of high technological advancement from country j to country i; mSijt is a moving average of mijt
with geometrically declining weights; Yj is the nominal income of country j in USD, Sj is the stock of country j’s domestic
knowledge; and q = 0.2 is declining geometric weights related to previous import propensities. The bilateral weighted
import is used based on high technological products, since they usually represent intermediate products through which
technology can transmit internationally (Coe and Helpman, 1995).
Human capital (EDU): educational attainments are measured based on the gross enrolment rate (GER) for each level, pri-
mary, secondary and tertiary. GER is constructed from the fraction of the population enroled in each level of education based
on their age group.3
Population growth drag (POP): Population growth drag, calculated as (xit/(1  a))  n, in which xit is estimated as the
share of agriculture in the total GDP, following Denison (1967), n is the employment growth rate, and (1  a) is labour’s
share of income.
h i
DTF: which is constructed as At1  At1 =At1 , where A is the TFP of the frontier country. According to Howitt (2000), the
further away a country is from the world technological frontier, the better its chances of adopting new technologies with
lower costs.
Infrastructure (INFR): Based on the literature, the level of infrastructure endowment is measured in real terms, and it may
be considered as either a flow or stock variable. In this paper, the stock of infrastructure is used, since using stock rather than
flow gives us more robust results (Irmen and Kuehnel, 2009; Égert et al., 2009). Furthermore, applying the stock of infras-
tructure may reduce the reverse causality in the empirical models, as the responses (accumulated investment over years) go
from output to infrastructure, and so would be smaller in models with a stock variable rather than a flow variable (Arnold
et al., 2007).
In order to use the real value of infrastructure, a new historical data set is constructed which covers 18 OECD countries
over the period 1870–2009. Data is obtained from national sources which are different across countries and the measure-
ment errors are likely to be higher as we go back in time. However, these deficiencies have to be weighted against the large
benefits derived from using long-run data as discussed in the introduction. Following Aschauer (1989), this paper considers
infrastructure as a core physical structure in terms of quantity, which is includes roads, highways, airports, railways, and
inland waterways, are all provided by the government and calculated as a share of the GDP.
When calculating the stock measure of infrastructure from the financial flow, we use the perpetual inventory method
(PIM), which can be written as: kit = Iit + (1  d) ki,t1, where Iit is infrastructure investment and d is the rate of depreciation
(assumed to be 2%). The initial capital stock is obtained by dividing the initial investment by the sum of the depreciation
rates and the average annual growth rate in infrastructure investment over the entire data period (i0/(d + g)).

4.3. Econometric methodology

Eq. (1) is estimated by using pooled cross-sections and time series analyses to find out whether an investment in capital
infrastructure can increase the economic growth of OECD members through changes in labour productivity. The Hausman
specification test suggests a preference of fixed effects over random effects under the null hypothesis that the individual
effects are uncorrelated with the other regressors in the model (Hausman, 1978). To check for the existence of
multicollinearity, the correlation matrix has been checked, showing that the final estimates are less likely to be affected
by the multicollinearity problem. However, using White’s test reveals evidence of heteroskedasticity in the model and the

3
Tertiary is only defined as universities in Madsen (2014) and colleges and open universities are not included in this dataset. For further information on GER
and its construction please see Madsen (2014).
80 M. Farhadi / Transportation Research Part A 74 (2015) 73–90

Breusch-Godfrey Lagrange Multiplier (LM) test shows existence of a significant autocorrelation problem in our panel-data
model. Hence, in order to increase the efficiency of estimators and to correct for both heteroskedasticity and autocorrelation,
the model is estimated using feasible generalised (weighted) least squares (FGLS). Since we are working with a panel data set
in which time (T) is greater than number of individuals (N), (T > N), applying FGLS provides more efficient results than other
methods. The estimated results are discussed later.

5. Estimation results

The results from estimating Eq. (1) using two different dependent variables are presented in Table 4. The first column
represents the results when labour productivity is taken as the dependent variable, and the second column relates to the
model with the total factor productivity as a response variable.
Consider first the model in which the labour productivity growth is the dependent variable. As can be seen from Column
1, the positive and strongly highly significant DTF coefficient indicates that the relative backwardness of the countries of
interest plays a considerable role in the growth of productivity. This result supports the conditional convergence hypothesis,
following Howitt (2000), and Ha and Howitt (2007), amongst others, which states that the further away a country is from the
world technological frontier, the faster its technology can grow, due to the lower effective costs. The estimated coefficient of
change in educational attainments is statistically and economically significant, indicating the positive effect of changes in
this variable on the growth of productivity. Highly-significant estimated coefficient of population growth indicates that
population growth is a drag on productivity growth due to the semi-fixed nature of land as a production factor. The result
is consistent with the theory, and shows that the growth in OECD countries is negatively affected by the increases in popula-
tion over the last 140 years.
The estimated coefficients of the domestic stock of knowledge and domestic research intensity are both statistically sig-
nificant at conventional significance levels. Taking the magnitudes into account, these results provide more support for
Schumpeterian theories, in which domestic R&D has a long-term positive impact on economic growth. Focusing on spillover
variables, the magnitude of the estimated coefficient of knowledge spillover through the channel of imports is significantly
greater for research intensity (0.162) than for the elasticity of the growth in the stock of foreign knowledge (0.043), implying
that research intensity is a highly significant determinant of productivity growth in OECD countries. This result again pro-
vides strong evidence in favour of the Schumpeterian growth theory, which assumes that a one-off change in research inten-
sity will have a long-run effect on the growth, and is also consistent with the predictions of Ha and Howitt (2007) and
Madsen (2008), who found research intensity a growth-enhancing factor in OECD countries since 1870.
Finally, and most importantly, is the influential role of infrastructure for productivity growth. Focusing on the estimated
coefficient of infrastructure shows that the labour productivity is positively affected by the growth in the stock of infrastruc-
ture. The significance of the estimated coefficient at conventional significance levels indicates that government spending in
productive sectors such as infrastructure can enhance countries’ economic growth indirectly, through its impact on the
labour productivity. In fact, a 10% increase in the share of the transportation infrastructure expenditure increases the labour
productivity of the OECD countries by 0.14 percentage points. Although in terms of the magnitude this is not large enough,
but still can provide evidence in support of the indirect positive effect of infrastructure investment on growth through its
impact on the labour productivity of OECD countries since 1870.4
The result of regressing Eq. (1) using a different dependent variable is presented in Column 2. As can be seen, considering
the total factor productivity as a dependent variable, all of the parameter estimates have approximately the same coeffi-
cients, and they are all highly significant, and therefore influential for growth productivity. However, compared to the pre-
vious model, in which we had labour productivity as a dependent variable, the magnitude of estimated variables slightly
decreased. Focusing on infrastructure, when it comes to total factor productivity, the estimated elasticity of the growth in
stock of infrastructure does not change considerably, and it is still highly significant. This result represents the fact that
changes in the stock of government spending on infrastructure can enhance the economic growth of countries of interest
indirectly through raising the labour productivity and directly by affecting the total factor productivity. Although, consider-
ing the huge stock of high quality infrastructure in OECD countries, the magnitude of this impact is relatively minor com-
pared to the other explanatory variables.

5.1. Robustness check

Labour productivity growth is expected to be influenced by other factors as well as those we included in our basic regres-
sion, and therefore, it is necessary to verify the sensitivity of the estimated parameters in order to ensure that our main
results are robust to the inclusion of control variables and/or the exclusion of different countries.

4
Regarding the potential spillover effects of transport investments on trade and by that on knowledge transfer and labour productivity, it should be
highlighted that the productivity effects of investment in infrastructure are likely to be underestimated in the regressions in this paper since indirect
productivity effects may not be captured by the estimates. One potential indirect productivity effect is through trade. Increasing infrastructure investment is
likely to reduce transport costs between trade partners substantially through improved port facilities, rail lines, airports and road infrastructure. This will in
turn increase trade and, consequently, benefit growth through the potential positive growth effects associated with trade. Furthermore, increasing trade is
likely to increase imports of knowledge through the channel of imports, which in turn will enhance productivity as shown by Madsen (2007).
 M. Farhadi / Transportation Research Part A 74 (2015) 73–90 81

Table 4
Productivity growth equation (Eq. (1)).

Dependent variables Y1 Y2
(1) (2)
DTF 0.048⁄⁄⁄ 0.024⁄⁄⁄
(12.44) (8.64)
EDU 0.090⁄⁄⁄ 0.059⁄⁄⁄
(9.90) (12.66)
POP 0.044⁄⁄⁄ 0.025⁄⁄⁄
(15.42) (6.90)
Sitd 0.016⁄⁄⁄ 0.011⁄
(3.52) (1.96)
(X/Q)dit 0.186⁄⁄⁄ 0.113⁄⁄⁄
(7.83) (5.58)
SImp
it 0.043⁄⁄⁄ 0.023⁄⁄⁄
(20.92) (20.12)
(X/Q)Imp
it 0.162⁄⁄⁄ 0.057⁄⁄⁄
(7.17) (3.88)
INFR 0.014⁄⁄⁄ 0.010⁄⁄⁄
(5.60) (3.18)
R2 0.58 0.59
D.W. 1.96 1.94

Note: The following years are included in the 5-year difference estimates: 1875, 1880, 1885, 1890, 1895,
1900, 1905, 1910, 1915, 1920, 1925, 1930, 1935, 1940, 1945, 1950, 1955, 1960, 1965, 1970, 1975, 1980,
1985, 1990, 1995, 2000, 2005 and 2009. The number of observations is 504. The dependent variables are
Y1 = labour productivity and Y2 = total factor productivity (TFP). The independent variables, stocks of
knowledge (S), educational attainments (EDU), and stocks of capital infrastructure (INFR) are measured in
five-year differences, while the research intensity (X/Q), population growth drag (POP), and distance to the
technology frontier (DTF) are measured as averages over five-year intervals. Feasible generalised least
squares (FGLS) regressions were run, corrected for first-order autocorrelation and heteroscedasticity.
D.W. = Durbin–Watson test for first-order serial correlation. The numbers in parentheses are t-statistics. ⁄, ⁄⁄
and ⁄⁄⁄ denote significance at the 10%, 5% and 1% levels, respectively. See also the note to Table 3 for the
definition of the variables.

5.1.1. Adding control variables

Several control variables are added to Eq. (1), and estimated results are reported in Table 5. The first control variable is
trade openness. The positive and highly statistically significant estimated coefficient suggests that countries with a greater
degree of openness catch up to the world technological frontier more quickly and therefore experience a higher productivity
growth. The second added control variable is the level of educational attainments. As it is shown in the second column of
Table 5, the estimated coefficient of the level of education is not significant at the conventional 5% level, while the change
in educational attainments remain positive and highly significant, implying that only one-off changes in education have long
run and permanent effects on productivity growth.
Considering the absorptive capacity, the third additional control variable is the interaction between the distance to the
frontier and educational attainments, and also that between the distance to the frontier and research intensity. As can be
seen from Column 3, the estimated coefficients of (DTF.EDU) and (DTF.(X/Q)) are 0.020 and 0.094, respectively, both statis-
tically and economically significant, indicating the indirect role of education and patent applications in enhancing produc-
tivity growth for countries behind the frontier through the observation of new technologies developed elsewhere. The next
control variable is the capital-output ratio. Since ignoring the capital variable from the growth model produced biased
results, capital stock investments in building and machinery are added to the model in order to check the consistency of
the estimated results. The estimated results prove the positive and highly significant impact of capital investments in
enhancing labour productivity growth.
To take into account the structure of both taxation and expenditure, we include indirect tax as an additional control vari-
able for our basic model. The inclusion of indirect taxes in Eq. (1) demonstrates the negative impact of the other side of gov-
ernment activity on productivity growth. Inflation is the last control variable which is added to the model and the growth
impact of public infrastructure is tested in the presence of price stability.
The results given in Table 5 show that the magnitude and significance of the estimated coefficient of infrastructure invest-
ment remain at about the same conventional significant level when different control variables are added into the model. The
estimated elasticity of the changes in the stock of public infrastructure after including openness, the level of educational
attainments, inflation and tax are around 0.012, 0.014, 0.013 and 0.010, respectively, which are in the neighbourhood of
the baseline model (0.014). However, this magnitude decreases significantly when capital investment is added to the model
(0.009). This decrease may be due to the fact that the social rate of returns to investments in both machinery and equipment
are higher than the rate of returns in infrastructure investment in our sample of OECD countries, as will be discussed in more
detail later. In general, the explanatory variables are not sensitive to the inclusion of control variables, suggesting that our
baseline model is consistent.
82 M. Farhadi / Transportation Research Part A 74 (2015) 73–90

Table 5
Robustness checks for the productivity growth (Eq. (1)).

Openness Education Capital Tax Inflation DTF. (X) Horse race 9 countries
(1) (2) (3) (4) (5) (6) (7) (8)
EDU 0.088⁄⁄⁄ 0.090⁄⁄⁄ 0.077⁄⁄⁄ 0.090⁄⁄⁄ 0.111⁄⁄⁄ 0.090⁄⁄⁄ 0.093⁄⁄⁄ 0.056⁄⁄⁄
(10.71) (9.76) (12.93) (9.23) (12.23) (10.47) (15.38) (2.83)
POP 0.051⁄⁄⁄ 0.044⁄⁄⁄ 0.052⁄⁄⁄ 0.045⁄⁄⁄ 0.053⁄⁄⁄ 0.044⁄⁄⁄ 0.062⁄⁄⁄ 0.012
(19.70) (14.22) (12.96) (13.98) (15.26) (14.19) (16.54) (0.90)
Sdit 0.014⁄⁄ 0.017⁄⁄⁄ 0.0001 0.019⁄⁄⁄ 0.007 0.021⁄⁄⁄ 0.005 0.028
(2.37) (3.48) (0.03) (4.29) (1.47) (4.73) (0.93) (0.98)
(X/Q)itd 0.157⁄⁄⁄ 0.166⁄⁄⁄ 0.141⁄⁄⁄ 0.185⁄⁄⁄ 0.173⁄⁄⁄ 0.036 0.133⁄⁄⁄ 0.127⁄⁄
(5.42) (6.37) (6.05) (8.28) (6.43) (1.32) (4.34) (2.26)
SImp
it 0.034⁄⁄⁄ 0.043⁄⁄⁄ 0.034⁄⁄⁄ 0.044⁄⁄⁄ 0.048⁄⁄⁄ 0.042⁄⁄⁄ 0.036⁄⁄⁄ 0.016⁄⁄⁄
(17.61) (18.51) (13.93) (14.32) (22.59) (17.81) (11.75) (5.06)
(X/Q)Imp
it 0.175⁄⁄⁄ 0.146⁄⁄⁄ 0.113⁄⁄⁄ 0.169⁄⁄⁄ 0.170⁄⁄⁄ 0.151⁄⁄⁄ 0.151⁄⁄⁄ 0.074
(10.35) (6.47) (6.00) (7.50) (6.78) (6.77) (8.51) (1.58)
DTF 0.045⁄⁄⁄ 0.048⁄⁄⁄ 0.042⁄⁄⁄ 0.049⁄⁄⁄ 0.0453⁄⁄⁄ 0.058⁄⁄⁄ 0.037⁄⁄⁄
(12.46) (12.24) (12.31) (12.84) (13.27) (13.10) (6.01)
INFR 0.012⁄⁄⁄ 0.014⁄⁄⁄ 0.009⁄⁄ 0.013⁄⁄⁄ 0.010⁄⁄⁄ 0.022⁄⁄⁄ 0.008⁄⁄ 0.117⁄
(5.90) (5.60) (2.13) (5.04) (3.43) (8.53) (2.06) (1.68)
OPEN 0.054⁄⁄⁄ 0.044⁄⁄⁄ 0.079⁄⁄⁄
(14.16) (9.05) (7.47)
EDU-lev 0.0009 0.001⁄⁄⁄ 0.001⁄
(1.43) (3.73) (1.96)
CAP 0.138⁄⁄⁄ 0.131⁄⁄⁄ 0.004⁄⁄⁄
(18.34) (19.14) (2.84)
TAX 0.009⁄ 0.001⁄⁄ 0.370⁄⁄⁄
(1.88) (2.28) (3.09)
INFLA 0.003⁄⁄⁄ 0.002⁄⁄⁄ 0.098⁄⁄⁄
(23.32) (14.18) (6.35)
DTF.(X/Q)dit 0.094⁄⁄⁄
(13.49)
DTF.(EDU) 0.020⁄⁄
(3.02)
R2 0.61 0.58 0.75 0.65 0.66 0.66 0.78 0.56
D.W. 1.94 1.96 1.95 1.96 1.98 1.96 1.94 1.99

Note: The control variables are: OPEN = degree of openness, EDU_lev = level of educational attainment, CAP = capital-output ratio (capital here means
capital investment in both structure and machinery), TAX = indirect tax imposed by the government, INFLA = inflation rate, and DTF.(X/Q)dit and DTF.(EDU)
are the interaction between the distance to the world technological frontier (DTF) and research intensity and education, respectively. See also the note to
Tables 3 and 4.

In the final step, and to see whether the estimation results remain consistent when the model contains all of the control
variables, a horse race regression is estimated and results presented in Column 7 of Table 5. As is shown, the magnitude of
infrastructure, as well as of all of the other explanatory variables, remain fairly stable after including all of the control vari-
ables in the baseline model at the same time, though the significance level of INFR drops slightly.

5.1.2. Changing the number of countries in the sample

To control for a potential country sample bias, nine countries are chosen and the model is re-estimated with the horse
race regression, to make sure that the infrastructure and all explanatory and control variables remain consistent in this
new sample. The results are given in the last column of Table 5, and show that, except for a few variables, the exclusion
of nine countries from the sample does not affect the estimation results, and the variables are still statistically significant,
indicating that our baseline model is not sensitive to the different sample.

5.1.3. Alternative TFP measure

As has been mentioned and was shown in Table 4, two different measurements for TFP are used in order to check the
consistency of the baseline model. Furthermore, all of the above robustness checks have also been tested once more with
the total factor productivity as a dependent variable; the results are not reported here, but were consistent with the basic
model.
To sum up, the above table shows that no matter what control variables are taken into account or how many countries are
in the sample, the results for all estimated parameters remain consistent and significant, and indicate that the labour pro-
ductivity for the selected OECD countries over the past 140 years has been driven by public expenditure on infrastructure, as
well as other factors such as changes in educational attainments, distance to the frontier, domestic and foreign knowledge,
and capital investment.
 M. Farhadi / Transportation Research Part A 74 (2015) 73–90 83

6. Social returns to investment in transport infrastructure

In this section, we investigate the extent to which social returns to investments in the transport sector exceed its private
returns amongst OECD members. Exploring this issue requires us to find the best approach for examining the nexus between
‘‘per capita economic growth’’ and ‘‘capital stock of investment’’. To do so, we follow Madsen’s (2005) work in which he
pointed out the potential bias associated with traditional models regarding the estimation of the rate of returns to invest-
ments (see for example, De Long and Summers, 1991, and De Long, 1992) and so he tried to re-examine the link between
GDP growth and capital investment in equipment and structure sectors. Based on what is mentioned above, the purpose
of this section is to re-examine Madsen’s production function approach in two ways: first, extending the data period to
2009 and also number of countries to 18 OECD economies, and second, adding transport infrastructure into the model in
order to compare its social returns to investment with other two sectors, including machinery and equipment, and building
and structure. Thus, the extended equation can be written as follows:

D ln Yit ¼ a0 þ a1 D ln K eq st inf
it þ a2 D ln K it þ a3 D ln H it þ a4 D ln K it þ e1;it ð2Þ

where Y is the real gross domestic product, K eq

it is the capital stock of machinery and equipment, K st
it is the capital stock of
structure and non-residential building, H is hours worked, and K inf
it is the capital stock of infrastructure.

6.1. Data and methodology

Eq. (2) is estimated by using the system Generalised Method of Moments (GMM) estimator method, which has been used
recently for dealing with endogeneity biases and unobserved heterogeneity in the estimation. The two-step system-GMM leads
to more efficient estimation results in the presence of heteroscedasticity and serial correlation (Baum et al., 2003). Growth in
the real gross domestic product is used as a dependent variable and capital stock of both machinery and structure are construct-
ed by applying the perpetual inventory method with depreciation rate of 17.6% and 3% for each section, respectively.
To measure the capital stock of transportation, a depreciation rate of 2% is used in the perpetual inventory method to
convert the flow infrastructure to the stock one. Finally, to measure labour input, labour force which is measured as hours
worked is added into the model. Measuring labour inputs in this manner is theoretically much more satisfactory than what
has been done in previous studies (Madsen, 2005).
It is important to clarify why investment in education and domestic research intensity are not added into the Eq. (2). There
are two reasons: first, the focus of this part of paper is to present and compare the social returns to investment of capital stock
of machinery, equipment and infrastructure only. Second, I extended Madsen’s (2005) production function model which he
introduced to overcomes the estimation bias associated with the traditional models (investment–income ratio approach), and
all these models have used the same investment variables, as Eq. (2), in order to examine the rate of returns.
Table 6 presents descriptive statistics for the variables used in this empirical study for the entire period 1870–2009. As
can be seen, all of the variables are in the same range, and the estimated coefficients are not affected by any extreme
observations.

6.2. Results and robustness check

Eq. (2) is estimated based on 5-year non-overlapping differences. To deal with endogeneity and potential omitted variable
bias, the two-step system-GMM are applied. The basic results are illustrated in Table 7. Column 1 of the Table 7 shows esti-
mated results of the basic model (Eq. (2)) while the endogenous explanatory variables are instrumented with their appro-
priate lags. However, since lagged level of the endogenous variables may not be strong instruments, external instruments
can be used to get rid of possible endogeneity. In general, instrumental variables are used when explanatory variables are
correlated with the error term in the equation. In this situation, explanatory variables called endogenous regressor and
the ordinary linear regression produces biased results. To have unbiased and consistent results, the model needs external
variables that are not correlated with the error term but are correlated with the endogenous regressor to utilise as a proxy
or instrument in the model.
For this section, four external instruments are used, including: capital stock weighted by geographic proximity, popula-
tion growth, urbanisation, and gold reserves. Columns 2 and 3 represent the results after instrumental variables are applied.
The second column of the table shows the results after capital stock weighted by geographic proximity is included to the
model to represent neighbourhood effects of investments in infrastructure and in both machinery and structures sectors.5
For this purpose, these variables have been weighted by the inverse geographic distance between each pair of countries
(i and j) using the following formula:
X18
1
Xit ¼ Zit ;
j¼1
Dij

5
They are chosen as the instruments since it fixed investments in neighbouring countries are driven by the same forces.
84 M. Farhadi / Transportation Research Part A 74 (2015) 73–90

Table 6
Descriptive statistics (Eq. (2)).

Yit K eq
it
K st
it K inf
it
Hit
Mean 0.13 0.212 0.163 0.384 0.022
Std. dev. 0.133 0.199 0.122 0.257 0.043
Min 0.714 0.754 0.317 0.017 0.371
Max 0.886 0.937 0.991 2.098 0.399

Note: Yit (dependent variable) is the real gross domestic product, K eq

it
is the capital stock of machinery and equipment, K st
it is the capital stock of structure and
non-residential building, Hit is hours worked, and K inf
it is the capital stock of infrastructure. The data period is from 1875 to 2009 for the 18 OECD countries.

Table 7
Parameter estimates of Eq. (2).

(1) (2) (3)

D ln K eq
it
0.337⁄⁄⁄ 0.332⁄⁄⁄ 0.286⁄⁄⁄
(5.69) (7.05) (4.22)
D ln K st
it 0.257⁄⁄ 0.243⁄⁄ 0.387⁄⁄⁄
(2.60) (2.59) (3.40)
D ln K inf
it 0.211⁄⁄⁄ 0.220⁄⁄⁄ 0.188⁄⁄⁄
(4.85) (5.76) (4.08)
D ln Hit 0.120 0.187 0.114
(0.65) (1.40) (0.46)
Observations 504 504 504
AR(2)-test (p-value) 0.484 0.524 0.530
Hansen-test (p-value) 0.181 0.374 0.113

Notes: The following years are included in the 5-year difference estimates: 1875, 1880, 1885, 1890, 1895,
1900, 1905, 1910, 1915, 1920, 1925, 1930, 1935, 1940, 1945, 1950, 1955, 1960, 1965, 1970, 1975, 1980,
1985, 1990, 1995, 2000, 2005 and 2009. All of the explanatory variables except hours are treated as
endogenous. Roodman’s (2009) Xtbond2 specification is used to estimate the model in STATA. The final
number of instruments is 14 for the basic model and 16 and 17 for the second and third columns,
respectively, which are derived from ‘‘collapsed’’. Orthogonal deviations and Windmeijer corrections are
used to estimate the coefficient in a two-step system GMM. To check the validity of the instrument, the
Hansen test is used and the null hypothesis is reported. Also, the AR(2) test results with a null hypothesis
that ‘‘the error terms in the 1st-differenced regression exhibit no 2nd order serial correlation’’ are reported in
the last row of table. The joint significance of the coefficients of the F-test and the AR(1) test for first order
serial correlation is not reported here, but the results were satisfactory for regressions. Time dummies for
the war periods are included in the model for the estimation, and also log level of initial GDP for the
dynamic relationship, but the results are not reported here, again to conserve space. The second lags of the
explanatory variables are taken as instruments for the differenced equation, whereas first differences of
the explanatory variables are taken as instruments for the level equation. The numbers in parentheses are
t-statistics and are based on robust standard errors. ⁄, ⁄⁄ and ⁄⁄⁄ denote significance at the 10%, 5% and 1%
levels, respectively.

where D is the geographic distance between countries i and j, and Z is the investments in infrastructure and/or the sum of
investments in capital stocks of both machinery and structure. In Column 3, the population growth, urbanisation ratio and
annual gold reserves of selected countries (as a share of their nominal GDP) are included to the basic model as instruments.
The results are qualitatively similar to the estimated outcomes in Column 1, indicating the consistency of the baseline
model. According to Madsen (2005), since the social returns to investment cannot be read from the coefficient estimates
of the production function in Eq. (2) directly, so have to be transformed. In fact, the coefficients of capital stocks in that equa-
tion measure the gross rate of returns to investment, in each individual capital, multiplied by the ratio of output-capital.
Therefore, the marginal productivities of capital investments (ax.Y/Kx), under the prefect competition assumption, equal
the gross rates of returns to investment:
aeq :Y=K eq ¼ r eq þ deq

astr :Y=K str ¼ r str þ dstr

ainf :Y=K inf ¼ rinf þ dinf

where the gross returns to investment are equal to the net return plus the depreciation rate. Superscripts of eq, st and infr
stand for investments in equipment, structure and infrastructure, respectively. ri represents the net rate of returns to invest-
ment, and d is the depreciation rate. Based on what is mentioned above, the rate of real social returns to the stock of capital
investment is as follows:

ri ¼ ai :Y=K i  di
 M. Farhadi / Transportation Research Part A 74 (2015) 73–90 85

In order to see whether or not the social returns to investment in the categories mentioned exceed their private returns, the
following parameters are used:
1. The rate of returns to private investment for capital stocks is set to 6%.6
2. The depreciation rates for capital stocks are as follows: equipment 17.6%; structures 3%; and infrastructure 2%.

The output-capital ratio (Y/Ki), which is calculated based on the unweighted average of the data series, is set to 4 for
machinery and equipment, 7 for building and structure, and 1.5 for transportation over the period 1870–2009. Therefore,
we have:

Ereq ¼ a
^eq :4  0:236; ð3Þ
Erst ¼ a
^st :7  0:09; ð4Þ
inf
Er ^inf
¼ a :1:5  0:08; ð5Þ
eq st inf
where Er , Er , and Er represent the excess of social returns to private investments in equipment, structure and infrastruc-
ture, respectively.
Following the above Eqs. (3)–(5), and based on the estimated coefficients presented in Columns 1 to 3 of Table 7, the
excess returns to investment in machinery and equipment are estimated to be around 111% in model 1 decreasing slightly
to 109% and 90% in models 2 and 3, respectively. The calculated excess returns to investments for building and structure are
significantly higher, around 170%, 161%, and 262% in models 1, 2 and 3, respectively. This indicates that a one dollar invest-
ment in the building and structure sector leads to more positive externalise than equipment investment. Compared to struc-
ture and machinery investments, any investment in the infrastructure sectors provides a considerably lower productivity,
since the rate of returns in investment for the transport sectors is estimated to be around 23%, 25%, and 20% for models
1–3. These results show that although the rate of returns to investments in infrastructure is still high and positive, this rate
is not as high as the positive externalities associated with the other two sectors over the last 140 years.

6.3. Post-war estimation

In order to control the potential country sample bias and to check whether the estimated coefficients change for the
shorter period of time, Eq. (2) is estimated for the post-WWII period by applying two-step GMM estimator. Furthermore,
using post-war data also enables us to compare our estimation results with the findings of the available literature, which
mainly take into account the evidence post-1950.
The estimated coefficients of Eq. (2) for the post-war period are presented in Table 8. Column 1 represents the estimated
coefficients for the basic model when only lags of explanatory variables are used as internal instruments. The estimated
results after including external instruments (the same ones as before) are shown in Columns 2 and 3.
As can be seen, the magnitudes of the estimated coefficients of machinery and infrastructure investments have increased
significantly, while the corresponding coefficients for structural investments have lost its significance, compare to the pre-
vious section. As has been mentioned, the social returns to investments cannot be found directly from the coefficient esti-
mates of Eq. (2), so we have:

Ereq ¼ a
^eq :6  0:236 ð6Þ
st
^st :7:5  0:09
Er ¼ a ð7Þ
Erinf ¼ a
^inf :1:3  0:08 ð8Þ

All of the parameters are the same as before, except for the Y/Ki ratio, which is calculated based on the unweighted average of
the data series over the period 1950–2009. Using the above Eqs. (6)–(8), calculated excess returns for equipment investment
reveal that rate of returns to investments in machinery and equipment considerably exceeds the private rate after the war
period, where the excess returns to investment in machinery are 228%, 211%, and 235% for models 1–3, respectively.
However, the rate of returns for building and structure investment has decreased considerably, to around 67% (model 1),
79% (model 2), and 75% (model 3).
The positive externalities associated with investment in equipment show that this type of investment has well exceeded
the private rate since 1950. The finding of high social returns to investment in equipment is consistent with the work of De
Long (1992), Oulton and Young (1996), and Madsen (2005), who find that the rate of returns to investments in equipment
and machinery is positive and well exceeds the private returns to investment in structural investment; on the other hand, it
disagrees with those works which have found the social rate of returns to be low or zero for OECD countries (Auerbach et al.,
1993, amongst others). Focusing on the rate of returns for infrastructure, the results for the post-war period show that the
rate of returns to infrastructure investment have increased slightly, remaining at around 28%, for models 1, 2, and 3. The
finding that investments in infrastructure have been socially productive is consistent with the study by Demetriades and
Mamuneas (2000), who showed that an extra dollar of investment in public capital is more productive than private invest-
ment in the long-run. However, this is contrary to Kopp’s (2007) results, who found that the rate of returns to infrastructure

6
This rate is chosen following Madsen (2005).
86 M. Farhadi / Transportation Research Part A 74 (2015) 73–90

Table 8
Parameter estimates of Eq. (2) – post-war period.

(1) (2) (3)

D ln K eq
it
0.420⁄⁄⁄ 0.392⁄⁄⁄ 0.431⁄⁄⁄
(4.56) (4.37) (4.27)
D ln K st
it 0.102 0.118 0.113
(1.58) (1.49) (0.96)
D ln K inf
it 0.283⁄⁄⁄ 0.281⁄⁄⁄ 0.279⁄⁄⁄
(8.26) (7.31) (5.82)
D ln Hit 0.235 0.222 0.178
(1.00) (1.03) (0.37)
Observations 234 234 234
Hansen-test (p-value) 0.307 0.368 0.355
AR(2)-test (p-value) 0.602 0.562 0.461

Notes: The final number of instruments is 11 for the basic model and 13 and 14 for the models in Columns 2
and 3, which are derived from ‘‘collapsed’’. See also the notes to Table 7.

for Western European countries is not significantly high and its around 5% for majority of countries, although, his work is
substantially different to ours as we use different technique over an extended period of time and across a large number
of countries.
To sum up, our results shows that even though rate of returns to investment in infrastructure is relatively high and exceed
the private rate, but it is not as high as positive externalities associated with investment in equipment and structure invest-
ment across OECD countries.

7. Discussion and conclusion

The current focus of the policy and public debate is on the role of infrastructure development as a vehicle for growth.
Although much work has been done to represent the exact relationship between infrastructure and economic growth, the
final impact of productive public spending on countries’ economic growth is still being debated. Following Aschauer
(1989), this paper considers infrastructure as a core infrastructure which consists of various sub-sectors, including roads,
highways, airports, railways, and inland waterways, which are all provided by the government and are measured as a share
of the GDP. Since the growth-enhancing impact of infrastructure can be examined better over a longer period of time, and so
as to have a comprehensive interpretation of the annual growth rate of government expenditure on the transport sectors
over the last century, this paper used a brand new historical dataset on infrastructure. This exploitation of a new long-
run dataset for OECD countries makes a significant contribution to the literature, given that almost all of the available studies
have only really explored the post-WWII period.
Using an inclusive model in which the economic growth of OECD countries is determined indirectly by changes in the
stock of transport investment and other explanatory variables, it is shown that a 10% increase in the share of infrastructure
expenditure is likely to increases the labour productivity of the selected OECD members by 0.14 percentage points. Robust-
ness check shows that our main results are robust to the inclusion of control variables and/or the exclusion of different coun-
tries. All the coefficients of infrastructure are downward biased due to measurement errors. It should be mentioned that
although the productivity effect of investment in stock of transport infrastructure is concluded, but the magnitude of this
impact is relatively minor compared to other variables, taking into account the substantial amount of investment in infras-
tructure by OECD members. Also, the significance of this relationship is relatively lower compared to other variables tested.
Applying the two-step GMM method, the estimated social rate of returns to capital investments shows that although the
rate of returns to investments in infrastructure is relatively high and positive and exceed its private rate, but this rate is not
as high as the positive externalities associated with the investments in structure and machinery, where their rate of returns
considerably exceeds the private rate, especially after the war period.
The policy interpretation of results should take into account the point made by studies such as DeLong et al. (2012) where
they believe that government should invest in projects with positive real rate of return, with any magnitude. They argue that
even though additional spending on productive stocks of public infrastructure capital and private human capital had no
impact on current GDP, but these types of investments would enhance the present value of future real GDP.
Future research can further focus on comparing new capacity provision vs. good maintenance of existing infrastructures
and as well as differentiating across different modes of transport, based on availability of data.

Acknowledgments

I would like to express my sincere gratitude to Professor Jakob Madsen from Monash University for his valuable com-
ments, his helpful advice and for providing me with the necessary data to undertake this study. I am grateful to the editor
 M. Farhadi / Transportation Research Part A 74 (2015) 73–90 87

and anonymous reviewers for their insightful comments. Helpful comments and suggestions from participants at the 41st
Australasian Conference of Economists (ACE 2012) are also gratefully acknowledged.

Appendix A. Variable list

See Table A1.

Table A1
Descriptions of variables used in the Eqs. (1) and (2).

Variables name Descriptions In Model

Capital stock of infrastructure Infrastructure consists of sub-sectors include roads, highways, airports, rail Log growth
ways, inland water ways, all provided by government and calculated as a share
of GDP. Perpetual inventory method is used to calculate the stock of
infrastructure from financial flow (with 2% depreciation rate)
Capital stock of machinery Constructed by applying perpetual inventory method with depreciation rate of Log growth
17.6% for investments in machinery and equipment
Capital stock of structures Constructed by applying perpetual inventory method with depreciation rate of Log growth
3% for investments in non-residential
h structures
i and construction
Distance to the frontier Calculated based on: DTF ¼ At1  At1 =At1 , in which A is TFP at the world Level
technological frontier (the USA and the UK)
Domestic patent stock Patents applied for by residents. Perpetual inventory method is used to calculate Log growth
the stock measure using a 15% depreciation rate
Domestic research intensity Measured as patent applications of residents divided by employment Log level
Educational attainment The data are based on school enrolments going back to 1812 for primary Growth
education since the oldest age cohort in 1870 started their education in 1812.
School enrolment is transformed into enrolment rates and the educational
attainment of the population of working age is then computed as the sum of
average enrolment rates at each level of schooling multiplied by the fraction of
the working age population in each age cohort
Gold reserves Annual gold reserves/nominal GDP Log growth
Import of foreign research intensity Foreign patent applications divided by employment weighted by imports Log level
Import weighted foreign knowledge stock Foreign stock of domestic patents weighted by imports Log growth
Inflation Positive growth in consumer price index (CPI) Level
Labour productivity Measured as an output per hour worked, where hours worked is measured as Log growth
annual hours worked multiplied by economy-wide employment
Openness Import/real GDP Level
Population Log growth
Urbanisation (Urban population (cities over 100,000 inhabitants)/total population) ⁄ 100 Growth

Appendix B. Data Sources

Transport infrastructure:
Before 1975: P. Flora, 1983, State, Economy, and Society in Western Europe 1815–1975, Frankfurt: Campus Verlag. From
1975 onwards data are obtained from two different sources: 1 – OECD, National Accounts; 2 – Government finance statistics
yearbooks, except for the following countries which other sources are used: Australia, Vamplew, W. (1987). Australians, His-
torical Statistics. Fairfax, Syme and Weldon Associates: New South Wales; and Australian bureau of statistic at http://www.
abs.gov.au/AUSSTATS/[email protected]/ViewContent?readform&view=ProductsbyCatalogue&Action=Expand&Num=6.4. Canada,
M.C. Urquhart, 1965, historical statistics of Canada, Cambridge: Cambridge University Press. Statistics Canada, www.stat-
can.gc.ca. Portuguese; France, INSEE, Institut national de la statistique et des études économiques, http://www.insee.fr/fr/the-
mes/comptes-nationaux/souschapitre.asp?id=62.Japan, Budget Bureau, Ministry of Finance, http://www.stat.go.jp/
english/data/index.htm; Netherlands, 1800–1913: J.-P. Smits, E. Horlings, and J. L. van Zanden, 2000, Dutch GNP and its Com-
ponents, 1800–1913, Groningen http://www.eco.rug.nl/ggdc/PUB/dutchgnp.pdf. Portuguese historical statistics, http://fes-
rvsd.fe.unl.pt/unlfe/Monografias/2005/2005-0741/Ingl/Ingles.pdf. United States of America, Susan B. Carter and et al.,
2006, Historical statistics of the United States, earliest times to the present, millennial edition, Vol. V. Part E, Cambridge:
Cambridge University Press. White house historical budget, http://www.whitehouse.gov/omb/budget/Historicals. Congres-
sional Budget Office paper, 2007, Trends in public spending on transportation and water infrastructure, The Congress of
the United States; Also, from OECD, 2000, conference of ministers of transport (ECMT), investment in transport infrastruc-
ture, OECD publication some data are taken for the following countries during 1985–95: Australia, Austria, Denmark, Fin-
land, France, Germany Ireland, Netherlands, New Zealand, Norway, Spain, Swiss, and Sweden.
Educational attainment, See Madsen (2014).
Inflation, See Madsen (2003).
Openness, See Madsen (2009).
Genetic distance, Spolaore and Wacziarg (2009).
88 M. Farhadi / Transportation Research Part A 74 (2015) 73–90

Geographic distance: Geographic distance between two countries based on bilateral distances between the biggest cities in
the two countries, where inter-city distances are weighted by the share of the city in the overall country’s population. The
data are downloaded from Centre d’Etudes Prospectives et d’Informations Internationales http://www.cepii.fr/anglaisgraph/
bdd/distances.htm.
Gold reserve: From 1870–1945: World Gold Council. 1950–1998: World Gold Council. 1998–2011: IMF, Exchange Rate,
Fund Position, & International Liquidity, obtained from DataStream. Note: Some data are backdated to 1870 using the
sum of values from UK, USA, Austria, France, Italy and Sweden. Irish data are backdated using UK values only.
Urbanisation: Fraction of the population living in cities with more than 100,000 inhabitants. http://www.databanksinter-
national.com/53.html (Databanks International).
Capital stock of machinery and equipment and non-residential buildings and structures, See Madsen (2005).
Total factor productivity, labour productivity, economy-wide real GDP, total employment, labour’s share, land, population, bilat-
eral trade weights, average annual hours worked per employee, patents, international knowledge spillovers through imports, and
distance to the frontier, See Madsen (2007, 2008, 2009).

Appendix C. Descriptive statistics for main variables

See Tables C1–C3.

Table C1
Country-based descriptive statistics for the labour productivity (1870–2009).

Country Mean Median Minimum Maximum Std. dev.

AUS 0.079097 0.096183 0.10762 0.180395 0.069705
BEL 0.105183 0.102829 0.14967 0.330557 0.104319
CAN 0.099335 0.097099 0.11343 0.406756 0.103445
DEN 0.107436 0.116344 0.10979 0.336656 0.093744
FIN 0.121864 0.116216 0.04372 0.290131 0.084847
FRA 0.11013 0.112099 0.51844 0.472509 0.164811
GER 0.103086 0.113244 0.2846 0.367229 0.149235
IRL 0.111247 0.094796 0.15786 0.28362 0.093838
ITA 0.113764 0.106115 0.58824 0.722933 0.222249
JAP 0.158096 0.138521 0.22848 0.471558 0.129565
NET 0.097467 0.082989 0.71257 0.695261 0.20829
NOR 0.126647 0.141826 0.06158 0.280529 0.089196
POR 0.115408 0.095658 0.01726 0.357208 0.093865
SPA 0.114397 0.096011 0.14382 0.428807 0.110681
SWE 0.117222 0.125723 0.15884 0.262826 0.083322
SWI 0.102448 0.092116 0.05467 0.270701 0.088249
UK 0.081986 0.066736 0.01125 0.195822 0.064533
US 0.095958 0.095718 0.15423 0.402412 0.084566

Table C2
Country-based descriptive statistics for the TFP (1870–2009).

Country Mean Median Minimum Maximum Std. dev.

AUS 0.033827 0.044773 0.13863 0.141719 0.066714
BEL 0.091027 0.082261 0.01493 0.286573 0.073787
CAN 0.043647 0.031933 0.10034 0.282144 0.072987
DEN 0.057444 0.058701 0.09107 0.233309 0.060668
FIN 0.061808 0.064067 0.21846 0.200277 0.074519
FRA 0.060981 0.04811 0.06562 0.197753 0.066703
GER 0.05844 0.059532 0.15009 0.255471 0.077494
IRL 0.075638 0.047212 0.11864 0.363869 0.088847
ITA 0.057895 0.060154 0.51075 0.523079 0.164681
JAP 0.047669 0.036814 0.12334 0.202065 0.078967
NET 0.047143 0.047122 0.43565 0.528286 0.139534
NOR 0.064363 0.069356 0.08342 0.180262 0.067538
POR 0.046337 0.047316 0.09481 0.17708 0.061954
SPA 0.030368 0.019805 0.38274 0.555656 0.144942
SWE 0.065608 0.064275 0.06826 0.191161 0.054271
SWI 0.046823 0.041055 0.09707 0.250962 0.070279
UK 0.022717 0.030174 0.11734 0.109252 0.050165
US 0.053565 0.036244 0.13965 0.341107 0.074851
 M. Farhadi / Transportation Research Part A 74 (2015) 73–90 89

Table C3
Country-based descriptive statistics for the infrastrcture (1870–2009).

Country Mean Median Minimum Maximum Std. dev.

AUS 0.076508 0.055473 0.03343 0.669527 0.128892
BEL 0.097547 0.107188 0 0.159793 0.045365
CAN 0.038753 0.039124 0.12177 0.165428 0.065231
DEN 0.09661 0.116447 0.02604 0.206948 0.067598
FIN 0.082073 0.059524 0.04818 0.266397 0.092281
FRA 0.063242 0.044283 0.01332 0.221431 0.06166
GER 0.142439 0.050432 0.0764 1.22707 0.262698
IRL 0.002926 0.02761 0.07071 0.141622 0.058102
ITA 0.052684 0.054919 0.02985 0.211954 0.064501
JAP 0.110927 0.060586 0.0711 0.797634 0.183054
NET 0.059064 0.041837 0.03548 0.186172 0.058046
NOR 0.110051 0.108841 0.02768 0.404552 0.088883
POR 0.475183 0.101697 0.00146 6.49368 1.335194
SPA 0.015911 0.014831 0.02807 0.098776 0.027427
SWE 0.115092 0.106644 0.00104 0.259197 0.068754
SWI 0.156367 0.140549 0.020079 0.566477 0.095765
UK 0.100798 0.09437 0.00951 0.250961 0.064198
US 0.013487 0.005265 0.04375 0.097989 0.037733

References

Acharya, R.C., Keller, W., 2009. Technology transfer through imports. Can. J. Econ. 42, 1411–1448.
Acemoglu, D., Aghion, P., Zilibotti, F., 2006. Distance to frontier, selection, and economic growth. J. Eur. Econ. Assoc. 4 (1), 37–74.
Aghion, P., Howitt, P., 1992. A model of growth through creative destruction. Econometrica, 323–351.
Aghion, P., Howitt, P., 1998. Endogenous Growth. MIT Press, Cambridge.
Aghion, P., Howitt, P., 2006. Joseph Schumpeter lecture appropriate growth policy: a unifying framework. J. Eur. Econ. Assoc. 4 (2–3), 269–314.
Aghion, P., Howitt, P., Mayer-Foulkes, D., 2005. The effect of financial development on convergence: theory and evidence. Q. J. Econ., 173–222
Aghion, P., Boustan, L., Hoxby, C., Vandenbussche, J., 2009. The Causal Impact of Education on Economic Growth: Evidence from the United States. Working
Paper, Harvard University.
Arnold, J., Bassanini, A., Scarpetta, S., 2007. Solow or Lucas? Testing Growth Models Using Panel Data from OECD Countries. OECD Working Paper, No. 592.
Aschauer, D.A., 1989. Is public expenditure productive? J. Monet. Econ. 23 (2), 177–200.
Auerbach, A.J., Hassett, K.A., Oliner, S.D., 1993. Reassessing the Social Returns to Equipment Investment. NBER Working Paper, No. 4405.
Barro, R.J., 1991. Government Spending in a Simple Model of Endogenous Growth. NBER Working Paper, No.2588.
Baum, C.F., Schaffer, M.E., Stillman, S., 2003. Instrumental variables and GMM: estimation and testing. Stata J. 3 (1), 1–31.
Benhabib, J., Spiegel, M.M., 1994. The role of human capital in economic development evidence from aggregate cross-country data. J. Monet. Econ. 34 (2),
143–173.
Boopen, S., 2006. Transport infrastructure and economic growth: evidence from Africa using dynamic panel estimates. Empirical Econ. Lett. 5 (1), 38–52.
Carreras, A., Tafunell, X., 2006. Long term growth of the Western European countries and the United States, 1830–2000. Cuadernos Económicos de ICE 70
(2), 25–50.
Cellini, R., Torrisi, G., 2009. The Regional Public Spending for Tourism in Italy: An Empirical Analysis. MPRA Working Paper, No. 16131.
Clark, C., 2012. The Sleepwalkers: How Europe Went to War in 1914. Penguin, UK.
Coe, D.T., Helpman, E., 1995. International R&D spillovers. Eur. Econ. Rev. 39 (5), 859–887.
De Long, J.B., 1992. Productivity Growth and Machinery Investment: A Long-Run Look, 1870–1980. Cambridge University Press.
De Long, J.B., Summers, L.H., 1991. Equipment investment and economic growth. Q. J. Econ. 106 (2), 445–502.
DeLong, J.B., Summers, L.H., Feldstein, M., Ramey, V.A., 2012. Fiscal policy in a depressed economy [with comments and discussion]. Brookings Pap. Econ.
Act., 233–297
Demetriades, P.O., Mamuneas, T.P., 2000. Intertemporal output and employment effects of public infrastructure capital: evidence from 12 OECD economies.
Econ. J. 110 (465), 687–712.
Denison, E.F., 1967. Why growth rates differ? The Brookings Institution, Washington.
Easterly, W., Rebelo, S., 1993. Fiscal policy and economic growth. J. Monet. Econ. 32 (3), 417–458.
Eaton, J., Kortum, S., 2001. Trade in capital goods. Eur. Econ. Rev. 45 (7), 1195–1235.
Égert, B., Koźluk, T., Sutherland, D., 2009. Infrastructure Investment: Links to Growth and the Role of Public Policies. OECD publishing, No.686.
Engelbrecht, H.J., 1997. International R&D spillovers, human capital and productivity in OECD economies: an empirical investigation. Eur. Econ. Rev. 41 (8),
1479–1488.
Esfahani, H.S., Ramı´rez, M.a.T., 2003. Institutions, infrastructure, and economic growth. J. Dev. Econ. 70 (2), 443–477.
Fernald, J.G., 1999. Roads to prosperity? Assessing the link between public capital and productivity. Am. Econ. Rev., 619–638
Fernández, M., Montuenga-Gómez, V.M., 2003. The effects of public capital on the growth in Spanish productivity. Contemp. Econ. Pol. 21 (3), 383–393.
Finley, L.A. (Ed.), 2005. Focus on Economic Growth and Productivity. Chapter 6: Public Capital, Human Capital and Economic Growth: Portugal, 1977–2001,
Nova Publishers.
Ghani, E., Din, M.U., 2006. The impact of public investment on economic growth in Pakistan. Pak. Dev. Rev., 87–98
Gibson, J., 2011. Federal Investment in American Infrastructure: The Past, Present, and Future <http://www.examiner.com/article/federal-investment-
american-infrastructure-the-past-present-and-future>.
Gramlich, E.M., 1994. Infrastructure investment: a review essay. J. Econ. Lit. 32 (3), 1176–1196.
Griffith, R., Redding, S., Van Reenen, J., 2003. R&D and absorptive capacity: theory and empirical evidence. Scand. J. Econ. 105 (1), 99–118.
Griliches, Z., 1990. Patent Statistics as Economic Indicators: A Survey. NBER Working Paper, No. w3301.
Groote, P., Jacobs, J., Sturm, J.E., 1999. Infrastructure and economic development in the Netherlands, 1853–1913. Eur. Rev. Econ. Hist. 3 (2), 233–251.
Grossman, G.M., Helpman, E., 1991. Comparative Advantage and Long-Run Growth, NBER Working Paper, No. 2809.
Guo, J., Guo, J., Xia, J., 2011. Econometrical investigation on infrastructure investment and economic development in China: a case study using vector
autoregression approach. KSCE J. Civil Eng. 15 (3), 561–567.
Ha, J., Howitt, P., 2007. Accounting for trends in productivity and R&D: a Schumpeterian critique of semi-endogenous growth theory. J. Money Credit Bank.
39, 733–774.
90 M. Farhadi / Transportation Research Part A 74 (2015) 73–90

Harchaoui, T.M., Tarkhani, F., Warren, P., 2004. Public infrastructure in Canada, 1961–2002. Can. Pub. Pol., 303–318
Hausman, J.A., 1978. Specification tests in econometrics. Econometrica. J. Econ. Soc., 1251–1271
Heintz, J., Pollin, R., Garrett-Peltier, H., 2009. How Infrastructure Investments Support the US Economy: Employment, Productivity and Growth. Political
Economy Research Institute (PERI), University of Massachussetts Amberst.
Herranz-Loncán, A., 2007. Infrastructure investment and Spanish economic growth, 1850–1935. Explor. Econ. Hist. 44 (3), 452–468.
Howitt, P., 2000. Endogenous growth and cross-country income differences. Am. Econ. Rev., 829–846
Howitt, P., Aghion, P., 1998. Capital accumulation and innovation as complementary factors in long-run growth. J. Econ. Growth 3 (2), 111–130.
Hulten, C.R., Schwab, R.M., 1991. Is there too Little Public Capital? Infrastructure and Economic Growth. American Enterprise Institute for Public Policy
Research.
Irmen, A., Kuehnel, J., 2009. Productive government expenditure and economic growth. J. Econ. Surv. 23 (4), 692–733.
Jayme Jr., F.G., da Silva, G.J.C., Martins, R.S., 2009. Public Expenditure on Infrastructure and Economic Growth Across Brazilian States, Cedeplar, Universidade
Federal de Minas Gerais No.td346.
Kamps, C., 2005. The dynamic effects of public capital: VAR evidence for 22 OECD countries. Int. Tax Pub. Financ. 12 (4), 533–558.
Kopp, A., 2007. Aggregate Productivity Effects of Road Investment: A Reassessment for Western Europe1. The Management and Measurement of
Infrastructure: Performance, Efficiency and Innovation, 47.
La Ferrara, E., Marcellino, M., 2000. TFP, Costs, and Public Infrastructure: An Equivocal Relationship. IGIER Working Paper No. 176.
Loizides, J., Tsionas, E.G., 2002. Productivity growth in European railways: a new approach. Trans. Res. A: Pol. Pract. 36 (7), 633–644.
Lumenga-Neso, O., Olarreaga, M., Schiff, M., 2005. On indirect trade-related R&D spillovers. Eur. Econ. Rev. 49 (7), 1785–1798.
Madsen, J.B., 2003. Inflation and investment. Scot. J. Polit. Econ. 50 (4), 375–397.
Madsen, J.B., 2005. A century of economic growth: the social returns to investment in equipment and structures. Manch. Sch. 73 (1), 101–122.
Madsen, J.B., 2007. Technology spillover through trade and TFP convergence: 135 years of evidence for the OECD countries. J. Int. Econ. 72 (2), 464–480.
Madsen, J.B., 2008. Semi-endogenous versus Schumpeterian growth models: testing the knowledge production function using international data. J. Econ.
Growth 13, 1–26.
Madsen, J.B., 2009. Trade barriers, openness, and economic growth. South. Econ. J. 76 (2), 397–418.
Madsen, J.B., 2010. The anatomy of growth in the OECD since 1870. J. Monet. Econ. 57 (6), 753–767.
Madsen, J.B., 2014. Growth, human capital, and the world technology frontier. Rev. Econ. Stat. 96, 676–692.
Márquez, M.A., Ramajo, J., Hewings, J.G., 2011. Public capital and regional economic growth: a SVAR approach for the Spanish regions. Investigaciones
Regionales 21, 199–223.
Mas, M., Maudos, J., Pérez, F., Uriel, E., 1996. Infrastructures and productivity in the Spanish regions. Region. Stud. 30 (7), 641–649.
Meunier, J., 2002. On the Fast Track: French Railway Modernization and the Origins of the TGV, 1944–1983. Greenwood Publishing Group.
Moreno, R., Lopez-Bazo, E., Artis, M., 2003. On the effectiveness of private and public capital. Appl. Econ. 35 (6), 727–740.
Nelson, R.R., Phelps, E.S., 1966. Investment in humans, technological diffusion, and economic growth. Am. Econ. Rev. 56 (1/2), 69–75.
Oulton, N., Young, G., 1996. How high is the social rate of return to investment. Oxford Rev. Econ. Pol. 12 (2), 48–69.
Park, W.G., 1995. International R&D spillovers and OECD economic growth. Econ. Inq. 33 (4), 571–591.
Picci, L., 1999. Productivity and infrastructure in the Italian regions. Giornale degli economisti e Annali di Economia, 329–353.
Romer, P.M., 1990. Endogenous technological change. J. Pol. Econ. 98 (5), S71–S102.
Romp, W., De Haan, J., 2007. Public capital and economic growth: a critical survey. Perspektiven der Wirtschaftspolitik 8 (S1), 6–52.
Short, J., Kopp, A., 2005. Transport infrastructure: investment and planning: policy and research aspects. Transp. Policy 12 (4), 360–367.
Sosvilla-Rivero, S., Herce, J.A., 2008. European cohesion policy and the Spanish economy: a policy discussion case. J. Pol. Model. 30 (3), 559–570.
Spolaore, E., Wacziarg, R., 2009. The diffusion of development. Q. J. Econ. 124 (2).
Sturm, J.E., 1998. Public Capital Expenditure in OECD Countries: The Causes and Impact of the Decline in Public Capital Spending, Edward Elgar Pub.
Sturm, J.E., Jacobs, J., Groote, P., 1999. Output effects of infrastructure investment in the Netherlands, 1853–1913. J. Macroecon. 21 (2), 355–380.
Vijverberg, W.P., Fu, F.C., Vijverberg, C.C., 2011. Public infrastructure as a determinant of productive performance in China. J. Prod. Anal. 36 (1), 91–111.
Vuchelen, J., Caekelbergh, S., 2010. Explaining public investment in Western Europe. Appl. Econ. 42 (14), 1783–1796.
World Bank, 1994. Annual Report. Washington, World Bank.
Yeoh, M., Stansel, D., 2013. Is public expenditure productive? Evidence from the manufacturing sector in US Cities, 1880–1920. Cato J. 33 (1).