ROAD SECTOR DEVELOPMENT AND ECONOMIC

GROWTH IN ETHIOPIA1

Ibrahim Worku2

Abstract

The study attempts to see the trends, stock of achievements, and impact of road
network on economic growth in Ethiopia. To do so, descriptive and econometric
analyses are utilized. From the descriptive analysis, the findings indicate that the
stock of road network is by now growing at an encouraging pace. The government’s
spending has reached tenfold relative to what it was a decade ago. It also reveals
that donors are not following the footsteps of the government in financing road
projects. The issue of rural accessibility still remains far from the desired target level
that the country needs to have. Regarding community roads, both the management
and accountancy is weak, even to analyze its impact. Thus, the country needs to do
a lot to graduate to middle income country status in terms of road network
expansion, community road management and administration, and improved
accessibility. The econometric analysis is based on time series data extending from
1971‐2009. Augmented Cobb‐Douglas production function is used to investigate the
impact of roads on economic growth. The model is estimated using a two‐step
efficient GMM estimator. The findings reveal that the total road network has
significant growth‐spurring impact. When the network is disaggregated, asphalt
road also has a positive sectoral impact, but gravel roads fail to significantly affect
both overall and sectoral GDP growth, including agricultural GDP. By way of
recommendation, donors need to strengthen their support on road financing, the
government needs to expand the road network with the aim of increasing the
current rural accessibility, and more attention has to be given for community road
management and accountancy. Lastly, gravel road expansion has to be made to
meet the target level of the road network and simultaneously ascertain rural
accessibility, thereby improving agricultural productivity and market access of the
poor rural population.

Author Keywords: Road sector development, Stock of achievements, Impact on Overall and
Sectoral Economic Growth, Two‐step GMM Estimation, Ethiopia

1
The final version of this article was submitted in October 2011.
2
Ethiopia Support Strategy Program II, International Food Policy Research Institute, Addis Ababa,
Ethiopia

101
Ibrahim Worku: Road sector development and economic growth in Ethiopia

1. Introduction

Since 1993/94, the Ethiopian government has been implementing various reforms
that have involved the processes of structural adjustment programs along with
commercialization of agriculture, private sector development, and a number of
related poverty alleviation programs. Successful implementation of the programs
requires an efficient infrastructural system. In particular, road transport is supposed
to create a network over a wide array of infrastructural facilities. In addition, the
road transport sector is essential for developing countries for the reason that
provision of other advanced means of transportation is expensive. For instance, Fan
and Rao (2003), citing numerous studies, indicated that public spending in rural
infrastructure is one of the most powerful instruments that governments can use to
promote economic growth and poverty reduction and among these services road
transport sector is considered as the crucial one.

A well‐developed road transport sector in developing countries is assumed to fuel

up the growth process through a variety of activities of the development endeavors
of a nation. Among these, creation of market access opportunities for agricultural
products is the major one. The issue of market access is more relevant for a country
like Ethiopia where rural population accounts for about 85% of the national
population who are engaged in production for both the domestic and international
market. Moreover, road transport facilities play a role in both the production and
consumption decisions of every household in their day‐to‐day activities. Besides,
road transport facilities are essential for expanding education, health service
provision, trade facilitation – both within the country and the export market, and
better public as well as private service provisions, including banking and insurance
services, to the destitute and marginalized rural dweller. Likewise, roads serve as key
infrastructural units, which provide linkages to other modes of transportation like
railways, shipping, and airways.

In Ethiopia road transport is the dominant mode and accounts for 90 to 95 percent
of motorized inter‐urban freight and passenger movements. However, because of its
limited road network, provision of infrastructure has remained one of the
formidable challenges for Ethiopia in its endeavor towards socio‐economic

102
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

development and poverty reduction (ERA, 2008a).The Ethiopian Road Authority

(ERA) investigated the link between the country’s development plan and the road
sector policy (ERA 2008a). The study generally indicated that there is a well‐
established nexus between the development plan of the nation “A Plan for
Accelerated and Sustained Development to End Poverty, Ethiopia’s version of
Poverty Reduction Strategy” (PASDEP) and the road sector policy.

Because of the imperative nature of the sector, it is relevant to undertake critical

investigation of the road transport sector of Ethiopia. Government expenditure
patterns in Ethiopia have changed dramatically over the last few years. Thus, it is
also important to study the trends in the levels and composition of government
expenditures in the road transport sector, and assess the impact of these changes
over time. It is even more important to analyze the contribution of such expenditure
in the overall growth aspiration of the nation. The study will also provide important
information for more efficient targeting and use of limited financial resources of the
country in designing an efficient and effective road transport system and developing
the road transport policy of the nation. Added to this, making such an investigation
contributes to the stock of knowledge regarding the road transport sector.

The aim of the study is to analyze the performance of the road transport sector in
Ethiopia. The study will have the following specific objectives:
• The study will analyze the stock of achievements and the performance of the
road network in Ethiopia;
• The study will review the available reports and policy strategy documents;
• The study will identify and characterize the links that exist between road
network development and overall and sectoral growth; it will also attempt to
capture the impacts of different types of road on the overall and sectoral
economic growth;

Based on the findings of the aforementioned analysis, the study will come up with
recommendations for improved performance of the road sector development on the
overall and sector specific economic growth.

To analyze the stock of achievements in the road transport sector of Ethiopia, time
series data are used for the econometric analysis. The major data source regarding

103
Ibrahim Worku: Road sector development and economic growth in Ethiopia

most of road related variables is obtained from the Ethiopian Road Authority (ERA).
Other relevant variables are collected from government authorities such as the
Ministry of Finance and Economic Development (MOFED), and the Central Statistical
Authority (CSA 2008; CSA 1970‐2010). In addition, whenever necessary, African
Development Indicators CD‐ROMs and other relevant publications are also used in
the data compilation process.

In order to achieve the stated objectives, various empirical methodologies are

utilized. In analyzing the trends in the road transport sector over the last few
decades, both descriptive statistics and econometric methods are employed. Time
series analysis is used to investigate the trends in the road transport sector over the
previous years and its impact on overall and sectoral economic growth. Descriptive
analysis is deployed to capture the link between the road transport sector and
performance indicators, like rural accessibility, road density, road network, road
financing, and community road management. The study also attempts to investigate
the stock of road transport in Ethiopia and its impact on economic growth. To this
end, the study reviewed the scattered knowledge in the area and uses a number of
tools in analyzing its impact on overall and sectoral economic growth.

The limitation of the study is that it is difficult to deal with the socio‐economic
impact of the sector. Analyzing the socio‐economic impact requires a baseline
survey on the status of the society before and after provision of road infrastructure.
This task would require a significant stock of data, time, and financial resource. It
also requires a long‐term plan, which is designed in parallel with a new road project
intervention. Upcoming studies might deal with this issue to see its impact on socio‐
economic development.

The study is organized as follows. Section two provides an overview of the road
sector in Ethiopia. Section three presents descriptive analysis of the road sector:
road network, road density, rural and urban road accessibility, community road
network, and financing of the sector. The conceptual framework and econometric
model specification and discussion on data related issues are discussed in section
four, as well as the econometric analysis along with estimation issues and discussion
on the findings. Section five finalizes the study with a brief conclusion and stating
plausible policy recommendations.

104
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

2. Overview of road sector policies in Ethiopia

The starting point is the federal government’s vision, which is to transform Ethiopia
from a least developed country into a middle‐income country by 2028, by sustaining
the two digit economic growth registered in the recent years (2003–2010/11).
Achieving this Government vision requires sustainable growth of the Ethiopian
economy, which in turn depends on the development of infrastructure in general
and expansion and improvement of the road network of the country in particular
(MOFED 2006).

The Ethiopian economy is highly dependent on agriculture, which accounts for

around 50 percent of the gross domestic product (GDP). An estimated 85 percent of
the population is directly or indirectly depending on the agricultural sector. More
than 90 percent of export earning is generated from the agricultural sector. Second
to the agricultural sector, services account for more than one‐third of economic
activity. The composition of service earnings has shifted only slowly in response to
economic liberalization, with recent slight growth in the construction, transport, and
tourism sectors. Contribution of the construction industry to GDP at constant factor
cost is about 6 percent for 2006/07 (Central Statistical Authority 2008). On the other
hand, industry accounts for almost 12 percent of economic activity where most of
the manufacturing firms are concentrated in Addis Ababa. However, these days it is
also common to see manufacturing industries being established in some other cities
and towns. Industrialization of towns outside of Addis Ababa obviously requires
more road infrastructure and efficient transportation operation.

Since its commencement the Ethiopian Roads Authority (ERA) has administered the
road sector. ERA was established in 1967 by proclamation No 256/67 to provide for
the control and regulation of travel and transport on the road. The ERA is
responsible for the use of all roads within Ethiopia, vehicles using these roads, and
to all matters relating to road transport activities of the country. After the downfall
of the military government, ERA restructured its obligations with a vision to ensure
the provision of a modern, integrated, and safe road transport service to meet the
needs of all the communities of a strong and unitary economic and political system
in Ethiopia.

105
Ibrahim Worku: Road sector development and economic growth in Ethiopia

When we look at the road network of the country over the past five decades,
compared to the year 1951 the total road network has increased with factor seven
to reach the level in 2009. In 1951 the total stock of road network was only 6400 km;
in 2009 that is 46812 km (ERA 2009). The rise in the length of road is due to the
emphasis given to the sector. In particular, the current government, the Federal
Democratic Republic of Ethiopia, has placed increased emphasis on improving the
quality and size of the road infrastructure. To address the constraints in the road
sector related to restricted road network coverage and low standards, the
Government originally formulated a 10‐year Road Sector Development Program in
1997 (RSDP 1997‐2007).

The first phase of the RSDP (1997‐2002) focused on the restoration of the road
network to an acceptable condition. Specifically, the program focused on (1)
rehabilitation of main roads; (2) upgrading of main roads; (3) construction of new
roads; and (4) regular maintenance on the network. Side by side, the program also
considered major policy and institutional reforms.

The program was launched with a very significant donor support to create adequate
capacity in the road sector and to facilitate the economic recovery process through
the restoration of essential road network. The first five year of the program (RSDP I),
1997‐2002, was officially launched in September 1997, and has been completed in
June 2002. Accomplishment under RSDP II is rather encouraging. The total
disbursement rate of investment on federal and regional roads for the 5 years of
RSDP II is about 125% and 73%, respectively, whilst the corresponding physical
accomplishment is 134% and 145% of the planned. Within the ten years period of
the program, the total disbursement of projects planned for the execution amount
25.4 billion Birr (US$ 2.9 billion). This would enhance the integration of domestic
markets and the potential growth of exports in terms of volume and international
competitiveness (ERA 2008b). Having looked at policies road sector policies,
hereafter achievements and constraints of the sector are described.

106
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

3. Descriptive analysis of relevant factors of the road sector

In this section, attention is given to the road network, road density and accessibility,
financing, construction, and maintenance and betterment costs, as factors are
indicators for the sector’s contribution to a wide range of growth inducing factors.

3.1. Road network

In 1951 the total stock of road network was only 6400 km of which 3400 km was
asphalt and the remaining 3000 km was gravel road. This entire network was found
only in urban areas. When the Imperial regime lost power, the network has reached
to 9160 km in 1973. On average, the network has been growing at a rate of 2.05
percent per annum over the period 1951‐1973. During the Derg regime, 1974‐1991,
the stock road increased to 19017 km with a growth rate of 6.2 percent per annum.
With the current EPRDF regime, the road network has reached 46812 km in 2009
with an average annual growth rate of 9.35 percent. Over the period 1991 to 2009,
28731 km of new road network was constructed.

As shown in Figure 1, the red lines depict asphalt roads while the grey ones show the
graveled roads. As it can clearly be seen from the figure, the development of road
network is yet to go far. A large space in the country is networked with only a few
roads. Though the development is good, more construction is important for
connecting the remote areas. Especially, the rural part of Ethiopia is less networked
with roads.

According to World Bank (2010), only 10 percent of the rural population lives within
two kilometers of all weather roads. Thus, the remaining 90 percent of rural people
live at a distance of more than two km from all weather roads. The
underdevelopment of the road network has its implication for the development of
the agricultural sector which is the mainstay of the rural people and the country in
general. Visual inspection indicates that a lot has to be done to put a sufficient
network in the country.

107
Ibrahim Worku: Road sector development and economic growth in Ethiopia

Figure 1: Ethiopian road network pictorial representation

Source: World Bank (2010)

Similarly, Table 1 shows the road network length in Ethiopia by type. Though there
was an increase in the length of roads between 1974 and 1989, it was somewhat
constant in the years 1989 to 1991. After, the takeover of EPDRF the government
has invested much in construction of asphalt roads. Especially after 2001 there is a
significant growth in asphalt road length. However, there is a negative growth in
gravel road length. This happened in the recent years like 2003, 2005 and the last
two years.

One possible cause for the negative growth in gravel roads would be the fact that
community roads, which could be considered as part of gravel road, are being
constructed with Productive Safety Net Projects (PSNP). This type of road is not
counted or included as gravel road for the very reason that it fails to meet the
standard set by the Ethiopian Roads Authority (ERA). In addition, either the federal

108
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

or regional road authorities do not administer this type of road. Another cause may
be the fall in expenditure for maintenance and reconstruction, mainly over the
period 2003 and 2005.

Table 1: Classification of road network (length in km)

Growth Growth
Year Asphalt Gravel Rural Urban Total Growth
Asphalt Gravel
1974 3360 5900 9260 9260
1975 3280 ‐2.38 6080 3.05 9360 9360 1.08
1976 3200 ‐2.44 6200 1.97 120 9400 9520 1.709
1977 3126 ‐2.31 6290 1.45 652 9416 10068 5.756
1978 3051 ‐2.4 6801 8.12 790 9852 10642 5.701
1979 3115 2.1 7328 7.75 1091 10443 11534 8.382
1980 3285 5.46 7328 0 1595 10613 12208 5.844
1981 3515 7 7430 1.39 1830 10945 12775 4.644
1982 3769 7.23 8532 14.83 2630 12301 14931 16.877
1983 3916 3.9 8532 0 3053 12448 15501 3.818
1984 4000 2.15 8738 2.41 3420 12738 16158 4.238
1985 4042 1.05 8788 0.57 3808 12830 16638 2.971
1986 4050 0.2 8989 2.29 4198 13039 17237 3.6
1987 4062 0.3 8994 0.06 5158 13056 18214 5.668
1988 4109 1.16 9270 3.07 5232 13379 18611 2.18
1989 4109 0 9270 0 5232 13379 18611 0
1990 4109 0 9287 0.18 5550 13396 18946 1.8
1991 4109 0 9298 0.12 5610 13407 19017 0.375
1992 3542 ‐13.8 8966 ‐3.57 5573 12508 18081 ‐4.922
1993 3555 0.37 9011 0.5 5800 12566 18366 1.576
1994 3622 1.88 10100 12.09 7812 13722 21534 17.249
1995 3630 0.22 12000 18.81 8043 15630 23673 9.933
1996 3656 0.72 12133 1.11 9100 15789 24889 5.137
1997 3708 1.42 12162 0.24 10680 15870 26550 6.674
1998 3760 1.4 12240 0.64 11737 16000 27737 4.471
1999 3812 1.38 12250 0.08 12600 16062 28662 3.335
2000 3824 0.31 12250 0 15480 16074 31554 10.09
2001 3924 2.62 12467 1.77 16480 16391 32871 4.174
2002 4053 3.29 12564 0.78 16680 16617 33297 1.296
2003 4362 7.62 12340 ‐1.78 17154 16702 33856 1.679
2004 4635 6.26 13905 12.68 17956 18540 36496 7.798
2005 4972 7.27 13640 ‐1.91 18406 18612 37018 1.43
2006 5002 0.6 14311 4.92 20164 19313 39477 6.643
2007 5452 9 14628 2.22 22349 20080 42429 7.478
2008 6066 11.26 14363 ‐1.81 23930 20429 44359 4.549
2009 6938 14.38 14234 ‐0.9 25640 21172 46812 5.53
Source: ERA 2009, *the road network in Eritrea is deducted

109
Ibrahim Worku: Road sector development and economic growth in Ethiopia

3.2. Road density

The proper level of road network is assessed by road density, which is measured by
road length per 1000 persons or by road length per 1000 km2. In the three RSDP
periods, there was a plan to increase the road density form 0.43 to 1.5 km per 1000
persons and from 21 to 116 km per 1000 km2, starting 1997 through 2009.

At the end of the first phase the road density has increased achieving the target of
the government. In 2002 the road density was exactly at the aimed level, which is
0.49 km per 1000 persons whereas the road length per 1000 km2 is more than the
target level by 30.27 km per 1000 km2.

When the second phase of RSDP continued, the government has also targeted for
higher levels, i.e. targeted road density of 0.5 km per 1000 persons and 30 km per
1000 square km. At the end of RSDP II, road density has reached 0.55 km per 1000
persons and 38.6 km per 1000 km2 in the year 2007. The accomplishment of the
second phase was thus a success.

Table 2: Road densities per 1000 persons and per 1000 km2
Road Density /1000 Road density /1000sq. Total Road
Year
person km Length
1997 0.46 24.14 26550
1998 0.46 25.22 27737
1999 0.47 26.06 28662
2000 0.5 28.69 31554
2001 0.5 29.88 32871
2002 0.49 30.27 33297
2003 0.49 30.78 33856
2004 0.51 33.18 36496
2005 0.51 33.6 37018
2006 0.53 35.89 39477
2007 0.55 38.6 42429
2008 0.56 40.3 44359
2009 0.57 42.6 46812
Source: ERA (2009)

However, the targeted figures were a bit high for the third phase, which could not
be attained at the end of the period. This phase is the shortest period, which lasted

110
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

only for two years. At the expiration of RSDP III, road density of Ethiopia has reached
0.57 km per 1000 persons and 42.6 km per 1000 km2, where 1.5 km per 1000
persons and 116 km per 1000 km2 were the targeted ones. Although road density
has increased, it has not improved that much as planned. At the year 2009, the road
density is still much below the average road density of Africa, that is, 60 km per 1000
km2 (ERA 2008b).

Table 3: Performance of road infrastructure of African regions

Average road length per capita, km/ 1000 persons
Regions 1960‐1975 1976‐1985 1986‐1997
Central 6.6 4.6 4.2
Eastern 3.3 2.8 2.2
Northern 3.8 2.9 2.4
Southern 7.1 6.3 5.6
Western 2.8 2.7 2.2
Source: Richaud, C. et al. (1999)

In spite of the successes achieved in improving the road density of Ethiopia, the
current level is much below the different African regions in 1997 (Table 3). Ethiopia’s
road density is even less than the average road density of Eastern Africa. This
indicates that even if there is much investment on construction and maintenance of
roads, the need for further development in the sector remains.

In the calculation of road density, the total road network is used which is a sum of
the good and bad condition ones. For a better indication of the network level, it is
good to investigate also the condition of the roads. Table 4 shows that the total road
in good condition has increased after the implementation of the RSDP programs,
while total road in bad condition has decreased. At completion of the last RSDP
program, the targeted proportion of good condition roads is not yet attained. In the
year 2009, the total good condition road has reached 54% while the planned was
82%.

111
Ibrahim Worku: Road sector development and economic growth in Ethiopia

Table 4: Proportion of different condition of roads in RSDP program

Year Good Fair Poor
1997 22 26 52
1998 23 26 51
1999 25 30 45
2000 28 32 41
2001 28 29 43
2002 30 30 40
2003 32 30 38
2004 37 28 35
2005 39 26 35
2006 47 22 31
2007 49 22 29
2008 53 20 27
2009 54 24 22
Source: ERA (2009)

International Road Federation Statistical Report (2006), as can be observed form

Table 5, indicated that both lower and middle‐income countries have a developed
road network system as reflected by road density per square kilometer. The current
average road density per square kilometer of lower middle‐income countries is 0.30
km/km2. Ethiopia’s current road density is only 0.0468 km/km2 of the total land
area. If we consider 80% of the land area to be populated, the country by now has
0.058 km/km2 of road density. Clearly, one can easily observe that the Ethiopian
Road Network compares poorly even with other low income countries – including
some Sub‐Saharan Africa. This indicates that a lot has to be done in expanding the
road network, which requires about four fold of the existing road network.

112
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

Table 5: Road densities of lower middle income countries

Lower Middle Income Road Density Lower Middle Income Road Density
Countries (Km/Sq. Km) Countries (Km/Sq. Km)
Algeria 0.05 Indonesia 0.19
Djibouti 0.12 Kiribati 0.92
Egypt 0.09 Micronesia 0.34
Iran 0.11 Philippines 0.67
Iraq 0.10 Samoa 0.82
Jordan 0.08 Thailand 0.11
Morocco 0.13 Tonga 0.91
Tunisia 0.12 Vanuatu 0.09
Bolivia 0.06 Albania 0.63
Brazil 0.20 Armenia 0.27
Colombia 0.10 Azerbaijan 0.68
Cuba 0.55 Belarus 0.45
Dominica Republic 0.26 Bulgaria 0.39
Ecuador 0.15 Georgia 0.29
El Salvador 0.48 Kazakhstan 0.03
Guatemala 0.13 Macedonia 0.34
Guyana 0.04 Moldova 0.39
Honduras 0.12 Serbia 0.16
Jamaica 1.94 Turkmenistan 0.05
Nicaragua 0.14 Ukraine 0.29
Paraguay 0.07 Angola 0.04
Peru 0.06 Cameron 0.11
Suriname 0.03 Cape Verde 0.25
China 0.20 Congo 0.05
Fiji 0.19 Lesotho 0.20
Namibia 0.05 Sri Lanka 1.48
Swaziland 0.21
Average 0.30 Km/Sq. Km
Source: International Road Federation Statistical Report (2006)

3.3. Road accessibility

Access refers to the opportunity to use or the right to or the ability to reach some
destiny. Accessibility is measured as the percentage of population having access to
all weather roads. The benefits of having access to a road network is measured in
terms of reductions in monetary costs or time needed by beneficiaries to access
output markets or key public social services like health and education.

113
Ibrahim Worku: Road sector development and economic growth in Ethiopia

The accepted theory, according to ERA’s (2008b) study, is that accessibility has three
elements: 1) the location of the individual; 2) the location of the supply, service, or
facility to which the individual needs access; 3) the link to bring the two together.
The study used three approaches, namely, the random model approach, the graph
theory approach, and the squire grid approach to cover the country’s network
demand. This demand was estimated as such that all rural population could have
access to all weather roads within a 5 km distance.

According to the ERA study the country is required to construct 200,000 km of

optimum national road network, which is considered as a target road network on
the assumption that it will give reasonably good accessibility. Whereas, for the
country to be competitive enough and enter into middle income category, the
targeted road density which secures the rural population to have access to all
weather road is estimated to be 0.3 km/km2, the average road density of the lower
middle countries. In this case the road transport network has to reach 330,000 km.
The same study defined the concept measured in terms of average distance from
the road network and proportion of area farther than 5 km from all weather roads
as lack of access, which deprives people from the opportunities to improve their
lives. Access is composed of two elements: mobility, reflecting the ease or difficulty
in traveling to a service or facility; and proximity of those services and facilities. The
study considered access to be one key element in providing the opportunity for both
social and economic development, and a key determinant of both poverty itself and
opportunities to escape from the poverty trap.

When we look at the recent trend regarding society’s access to the all weather road
network, we find a slight improvement over the past seven years. However, in 2008
only about 33% of the rural population had access to an all weather road within a
distance of 5 km. Given the fact that around 80 million people are living in rural
area, such a low rate exacerbates the problem of poverty. Improving the current
access rate should be a major concern of the country’s road sector expansion
program.

Similarly, African Development Indicators (ADI) (2008/09) data indicates that the
country has made an effort to provide access to all weather roads, though it is not
satisfactory. The graph below depicts that the problem of accessibility is resolved

114
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

only very sluggishly. Within a period of seven years (2002 to 2008), an additional
seven percent of the rural population is provided with access to all weather roads
(from 26% to 33% of the rural population). ERA (2008b) study also indicates that
with the recent construction of new roads, the average distance from a road has
been reduced from 21km in 1997 to 11.7 km in 2009.

Figure 2. Trends over access to all‐season road per percentage of rural population

Access to an all-season road (% rural population)

35%
30%
25%
20%
15%
10%
5%
0%
2002 2003 2004 2005 2006 2007 2008

Source: ADI (2009)

On the other hand, the proportion of area farther than 5 km from all weather roads,
which was 79% in 1997, has been reduced to 65.3% in 2009. Therefore, the issue of
accessibility calls for a kind of ‘big‐push’ approach in expanding all weather roads for
the destitute rural poor. The problem of accessibility could also be addressed
through a well‐designed planning process coinciding with the parallel trends towards
the decentralization of decision making and the concern to involve the local
communities in the decision making process. The effort made so far towards the
improvement of main roads and rural roads is a necessary but not sufficient
measure to enhance rural accessibility.

However, the future is not unwelcoming. It has been observed that a continuous
support from the government and a serious commitment of the sector offices and
other stakeholders would enable to achieve enhanced access of the rural
population.

115
Ibrahim Worku: Road sector development and economic growth in Ethiopia

3.4. Road sector financing

Construction and maintenance of roads is not an easy investment. It requires a huge

amount of expenditure with foreign currency. This huge amount of money is not
affordable only by the government of Ethiopia from internal sources. Hence, the
government of Ethiopia finances road construction and maintenance through loans
and aid from other countries and organizations in addition to domestic finance. In
recent periods, the lion share of finance is the government of Ethiopia followed by
International Development Association. The other following financers are European
Union and African Development Bank.

Looking at the trends in the road sector financing (Figure 3), the pattern shows that
the internal financing in 2008/09 has increased to be more than ten times what it
has been a decade before. The trend also indicates that the total amount of external
financing is also growing but it has only multiplied by a factor five over the same
comparable period. The domestic financial expenditure share in 1997 has been only
twice that of the external source. Whereas, the recent expenditure is about four
times that of the external source in the face of growing financial expenditure over
the sector.

Figure 3: Share of internal and external financing in the road sector

9,000.0
8,000.0
7,000.0
6,000.0
5,000.0
4,000.0
Internal
3,000.0
External
2,000.0
1,000.0
0.0

Source: ERA, 2008b and author’s compilation

116
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

Additionally, the government of Ethiopia’s financial expenditure in the RSDP

programs had also been consistently rising (see also Table 6).

All in all, the above discussion reveals that the government is making a relentless
effort towards the development of the road network. In this regard, one key
constraint is the availability of sufficient funding to achieve an adequate road
network comparable with at least the African average of 60km per 1000 km2 (ERA
2008b). The ERA (2008b) study also indicated that inadequate funding and resource
mobilization are major problems in the road sector. Insufficient public resources
lead to under‐funding of road transport infrastructure needs (road network
maintenance needs as well as road network expansion). The problem is found to be
worse in expanding the network in rural areas. The available fund for rural roads
development and maintenance is limited. The possible remedial measure would be
to look for an alternate and sustainable finance.

Table 6: Source of finance by the pursued RSDP programs

Financier Disbursement % of
RSDP I RSDP II RSDP III (2yrs) (12 yrs) contribution
IDA 1,432.9 3,135.3 1,848.8 6,416.9 14.2
EU 678.1 1,049.7 2,334.7 4,061.9 9.0
ADB 506.4 517.8 275.7 1,299.8 2.9
NDF 14.8 63.9 56.9 135.6 0.3
Japan 164.9 380 274.6 819.5 1.8
Germany 27.7 302.6 54.8 385.0 0.8
Sweden 0.0 5.0 0.0 5.0 0.0
Ireland Aid 2.6 20.9 0.0 24.1 0.1
UK 23.2 135.1 28.2 186.4 0.4
OFID 0.3 293.3 167.9 461.5 1.0
BADEA 0.0 59.9 133.9 193.7 0.4
SDF 0.0 39.3 92.9 132.2 0.3
GOE 3,455.5 8,669.5 11,139.1 23,264. 51.3
Road Fund 978.2 2,555.8 3,053.1 6,587.1 14.5
Community 0.0 884.8 469.6 1,354.4 3.0
TOTAL 7,284.6 18,112. 19,930.0 45,327. 100.0
Source: ERA (2009)

117
Ibrahim Worku: Road sector development and economic growth in Ethiopia

3.5. Few points on community roads

According to the ERA (2008b) study, community roads are being developed in
different woredas. The roads are being constructed as part of the Productive Safety
Net Project (PSNP). Table 7 depicts that 57,000 km of community roads were
constructed in the period 2004‐2008 under the “food for work” scheme. Within this
food for work, substantial amounts of community roads are being maintained and
constructed in all the regions.

Table 7: Existing stock of community roads (2007)

Existing in 2006 Constructed in 2007 Total
Region
(Km) (Km) Community
Tigray 5,803 578 6,381
Afar ‐ ‐ ‐
Amhara 11,980 4,000 15,980
Oromia 13,839 3,489 17,328
Somali 1,273 ‐ 1,273
SNNP 10,754 2,027 12,781
Benishangul‐Gumz 2,230 1,001 3,231
Gambella 464 ‐ 464
Total 46,640 11,124 57,438
Source: ERA, 2008b

As per the same study, it shows that community roads are not well accounted and
managed by either the federal or regional road administration. In this regard the
study clearly cites the problem as follows:

However, it’s worth noting that, for these community roads their design standard,
location, condition, ownership and for that matter their existence is not known in
the form of a national database. Further, the issue of how sustainable (in terms of
both financially and administratively) these community roads are with respect to
development and maintenance is unclear – as there is no institution that finances
the maintenance costs of these rural/community roads and nobody is accountable
to them institutionally both at the federal and regional levels. The mileage of
community roads will continue to grow rapidly – and the needed maintenance costs
for rural/community roads alone would become substantial. (ERA, 2008a)

118
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

Community road development and its sustained maintenance is a key point in

ensuring the accessibility/mobility of the rural poor and to connect them to the
improved federal and regional road network. Hence, an in‐depth assessment of the
status of the community road development program and the future development
and maintenance strategy is recognized as an urgent task.

4. Econometric modeling and the data

In order to analyze the economic impact of the road sector on the growth process of
a country, studies adopted ‘augmented Solow growth model’. Recently, there is a
wide array of literature written towards identifying the relationship between
economic growth and road sector development. However, the focus of previous
studies has been to investigate the link between infrastructure, which is an
aggregated measure, and economic growth.

Previous studies, for instance, Fan et al. (2002), Fan and Chan‐Kang (2005), Canning
and Bennathan (2000), Canning and Pedroni (2004) used the standard Cobb‐Douglas
type production function to analyze the impact of infrastructure on the overall GDP
growth, which per se is assumed to be a measure of overall economic growth.
Dercon et al. (2008) also used an analogous type of specification to see the impact of
road and agricultural extension on growth and poverty reduction in a panel data set
of selected fifteen Ethiopian villages. Fan and Chan‐Kang (2005) indicated that there
is strong link between road development, economic growth, and poverty reduction.
In specifying the production function, we need to take into account different
specifications for paved and gravel roads on the overall productivity. For the reason
cited above, augmented Solow trans‐log production function is used to see the
impact of road on GDP growth.

Few of the previous studies consider the issue of diminishing return, which could be
captured by looking to the coefficient of the square of road network. If the
coefficient happens to be negative, the negative sign indicates that road network,
used in the regression, is abided by “neoclassical” paradigm. In case the coefficient
happens to be positive, the sign indicates that road network is a self sustaining

119
Ibrahim Worku: Road sector development and economic growth in Ethiopia

input, governed by an “endogenous growth” paradigm, which is mostly believed to

be true for developed countries.

The possible existence of reverse causality, i.e. the fact that infrastructure
accumulation could be driven by productivity growth, need to be accounted in
analyzing the relationship between infrastructure variables and economic growth. In
this regard, Canning (1999), Canning and Bennthan (2000), Canning and Pedroni
(2004) and Calderon (2009 explicitly stated the need to account for the possible
existence of reverse causality in estimating the growth function which is augmented
with an infrastructure variable, like road network.

The study also looks at the impact of road network on sectoral GDP. For instance,
Dorosh et al. (2009) analyzed the importance of road connectivity to agricultural
productivity in Africa. The findings indicate that lower return from having high
density is exhibited to be low for West Africa. Whereas, longer travel time decreases
total crop production, and reducing travel time significantly increases adoption of
high‐input/high‐yield technology in East Africa. The findings showed the importance
of increased road connectivity in East Africa.

Following the aforementioned studies, this study utilizes similar specifications to see
the impact of roads on the overall economic growth. The general specification of the
model is an augmented Solow Growth model. Such a model is basically a log
transformation of the Cobb‐Douglas production function. The logarithmic form of
this production function of Solow growth model allows including any relevant
variable which affects the growth of GDP. It also allows us to include a dummy
variable which captures the impact of any policy intervention in the analysis period.
In addition, Fan and Chan‐Kang (2005) indicated that different types of roads can
create different economic return. Therefore, it is important for us to see the
implication of road quality on the productivity.

4.1. Conceptual framework of the model

Having the theoretical inter‐linkage on the relevance of road network to economic

growth from the above discussion, the following section provides the conceptual

120
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

framework used in modeling road network. To do so, the starting point is the
simplest production function that has the following specification.

GDP = f (L, K) (1)

Where, GDP is the gross domestic product of the country, L is labor and K denotes
capital. The above general Cobb‐Douglas type functional specification will be
augmented with road so as to identify its impact on economic growth. Accordingly,
the above functional specification will be reformulated as

GDP = f (L, K, R) (2)

Where, R is road length and the others are as stated above.

Here again, Fan and Chan‐Kang (2005) criticize previous studies for not recognizing
the fact that the return from different types of road on economic growth might be
different. Thus, in this study the road variables will be classified into paved and
gravel road in order to see the impact of the difference on the overall growth. The
function, which takes care of the difference on road type, will be specified as;

GDP = f (L, K, RT) (3)

GDP = f (L, K, Rp, Rg) (4)

Where Rp and Rg represent paved and gravel roads respectively.

We can also specify the production function for agricultural GDP, with the inclusion
of gravel roads as an input. Such specification makes it interesting for the reason
that the agricultural populations rarely have the opportunity to use paved roads for
transportation purposes and thus the impact might be negligible for agricultural GDP
growth. Therefore, we might anticipate at this stage to state the functional
specification on agricultural productivity. The model could now be stated as;

AGDP = f (L, K, RT) (5)

AGDP = f (L, K, Rp, Rg) (6)

121
Ibrahim Worku: Road sector development and economic growth in Ethiopia

Where AGDP denotes the agricultural GDP and Rg is gravel road length and Rp paved
road length, and the rest are as already defined.

In addition to the AGDP, the impact of roads could be seen on non‐agricultural

sector GDP growth rates by disaggregating it into industrial GDP (IGDP) and service
sector GDP (SGDP). For this purpose, similar type of production functions will be
estimated with their respective variables. The regression is run on the total road
network.

IGDP = f (L, K, RT) .(7)

IGDP = f (L, K, Rp, Rg) .(8)

And the service sector GDP specification will be

SGDP = f (L, K, RT) .(9)

SGDP = f (L, K, Rp, Rg) .(10)

In order to account for other factors, which are missing in the above functional
specification, once again, the model is augmented with policy dummy to see the
impact of any relevant policy intervention in the regression function. At this stage,
the policy dummy is entered in the functional specification on the aggregate
production function. Thus, the specification will be stated as follows:

GDP = f (K, H, Rp, Rg, Dpi) .(11)

Dpi denotes dummy for policy interventions, which are introduced to account for any
policy intervention over the analysis period.

4.2. Specification of the model

Once the variables within the model are clearly defined, the next step is to derive
out the estimable production function. Accordingly, the aggregate Cobb‐Douglas
production function along with the road component, which could be estimable, can
be reached through the following procedure,

122
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

GDPt = γ t H tα K tβ (R pt .Rgt )
ρ
(12)

Where: Ht is human capital at time t

Kt is physical capital at time t
Rpt and Rgt is road network for paved and gravel roads, respectively at time t and
whereby the variables are written in per worker terms, and α , β and ρ are
parameters of interest.

The model is then transformed to the logarithmic form whereby the resulting
equation is set as follows.

ln GDP t = γ + α ln h t + β ln k t + ρ 1 ln γ + ρ 2 ln γ gt + ε t
pt
.(13)

Given the above specification, policy and other structural shift indicator dummies
are introduced in the following specification.

ln GDPt = γ + β 1 ln ht + β 2 ln k t + β 3 ln γ pt + β 4 ln γ gt + β 5 D p + ε t
(14)

Where Dp is policy dummy reflecting the introduction of a new policy which is

related to road provision.

Finally, once the estimable functional form specification is reached in the estimation
procedure, GDP will further be disaggregated to identify the impact of road on
agricultural, manufacturing and service sector productivity. In addition, the model
will also be estimated for the agricultural GDP for a more disaggregated rural road
network to see whether there exists a significant relationship among these variables.
At the end, the marginal return per unit length of different types of road on each of
the three sectors productivity will be computed.

4.3. The data

The objective of the study is to estimate the impact of road infrastructure on the
overall economic growth. A time series data spanning over the period 1971‐2009 is

123
Ibrahim Worku: Road sector development and economic growth in Ethiopia

used for the analysis. The aggregate output function is setup with a set of
explanatory variables constituting physical capital, human capital, and road network
and policy intervention dummy variable. The detail on the source and type of each
of the variables is presented below.

Real GDP per worker is obtained from ministry of finance and economic
development (MOFED). The data set from the period 1991 to 2009 is not compatible
with prior year’s data. To deal with this problem, prior year growth rate of GDP is
used to extrapolate and have consistent GDP values with their respective GDP
growth rates.

Given the above functional specification, most of the variables could be obtained
from different macroeconomic databases. What is not clearly explained in the above
general specification is as to how the physical capital and human capital variables
could be handled in the model. While the capital variables could be derived from
Kohler’s (1988) capital accumulation function, which is referred as perpetual
inventory method. The procedure for deriving the capital stock is set as follows.

Kt = It + (1 − δ )kt −1 (15)

Where Kt is capital stock in period t which could be computed as the sum of It, which
is gross capital formation in year t. δ is the rate of depreciation of capital. Fan and
Chan‐Kang (2005) took the rate of depreciation to be 10%. This study will also adopt
similar assumption to arrive at the stock of capital.

Following Fan and Chan‐Kang (2005) and similar other studies, the initial capital
stock is computed using the following mathematical representations.
It
kt = .(16)
(δ + r )
Where r is the real interest rate and δ is rate of depreciation of capital. Using the
above formula, the initial stock of capital is computed for the year 1970. Previous
studies tend to make assumption on the real interest rate and the depreciation rate.
For instance, Kohler (1988) computed the initial capital stock to be 7.4 times gross
capital formation of the beginning year, which is 1978, taking 8 percent depreciation

124
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

rate of the stock of capital. In this study too, I assumed 8 percent depreciation rate.
The computed initial capital stock is found to be 5.88 times the gross capital
formation of the year 1970. Then after, the remaining capital stock data are
computed using the above stated capital accumulation function.

As already explained above, the data on physical capital was constructed using the
perpetual inventory method. To implement it, the initial level of the capital stock
was estimated using data on the capital stock and real output from MOFED. The
year 1970 is taken as the initial period for which gross capital formation is used. The
physical capital is set up as net of gross capital formation for each period. Since the
stock of road network is used as one variable, it is deducted from the gross capital
formation to avoid double counting.

For human capital, secondary school enrolment rate is used as a proxy. Secondary
school enrolment rate is obtained from Central Statistical Agency annual bulletin.
The best proxy for human capital as indicated by Fan and Chan‐Kang (2005) study is
average years of schooling. However, Ethiopian data regarding years of schooling are
not available. With regard to the human capital variable, Calderón (2009) choose
secondary school enrolment to account for human capital in dealing with the impact
of infrastructure and growth in Africa. Not only Calderon (2009), similar other
studies, for instance Cohen and Soto (2001) also used secondary school enrolment
as a proxy for human capital in the growth regression function. So, following the
above studies, taking secondary school enrolment as a proxy for human capital in
this study is justified.

Finally, the labor for converting all inputs in per worker terms is obtained from
African Development Indicators (ADI) (2009) CD‐Rom, which is part of World Bank’s
World Development Indicators. Electricity Power generation is also obtained from
ADI CD‐Rom.

Road network with a classification of paved urban road, gravel urban road, gravel
rural road and road total network (in kilometers) for the entire period under
consideration is obtained from Ethiopian Road Authority. In the econometric
estimate all the variables are expressed in per worker units.

125
Ibrahim Worku: Road sector development and economic growth in Ethiopia

In addition, a number of ERA publications, ADI, IMF web data set, MOFED and CSA
data are used for the descriptive analysis part.

4.4. Econometric analysis

4.4.1. Test for unit roots

The data set deployed for this study is a time series data. When dealing with time
series data it is important to test the stationary or non‐stationary nature of the data
set for the reason that non‐stationary variables might lead to spurious regression. In
this regard Harris (1995) stated that:

…models containing non‐stationary variables will often lead to a problem of

spurious regression, whereby the results obtained suggest that there is
statistically significant relationship between the variables in the regression
model when in fact all that obtained is evidence of contemporaneous
correlation rather than meaningful causal relation.

Thus, the first step is to test the stationary nature of individual variables that will be
included in the regression. To test the stationary nature of the variables, the
Augmented Dickey‐Fuller (ADF), the modified version of the Dickey‐Fuller, test is
used. According to the ADF test, null hypothesis is that the variable is assumed to
have/contain a unit root. The time series nature of the data will be tested against
the alternative, where a stationary process generates the variable. Other common
unit‐root tests that could alternatively be used to test stationarity include the DF‐
GLS test of Elliot, Rothenberg, and Stock (1996) and the Phillips‐Perron (1988) test.

Pperron test in stata command performs the Phillips‐Perron test that a variable has
a unit root. The null hypothesis is that the variable contains a unit root, and the
alternative is that the variable is generated by a stationary process. Pperron uses
Newey‐West standard errors to account for serial correlation, whereas the
augmented Dickey‐Fuller test implemented in dfuller uses additional lags of the first‐
difference variable. Stata automatically select the appropriate lag length when we
use pperron. So, this study uses both the pperron and ADF tests to check the
stationary nature of the variables.

126
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

In testing the unit root, most macroeconomic variables, in practice, are suspected of
showing a time trend. To control for the trending nature of the variables, the graphic
visual inspection test is also conducted. A summary table on test for stationarity of
the variables is reported following the graphic trend test (Annex Figure A.1).

With the exception of agricultural GDP per worker, urban gravel road and urban
road, the other variables exhibit trends. Therefore, unit root tests require a trend
term in both the ADF and pperron tests.

Table 8. Summary table on Phillips‐Perron test for unit root Newey‐West lags
stata routine lag length selection
Statistic Z(t) Statistic Z(t) at level Statistic Z(t) at
at level a with trend b * first difference c
lnrgdp_pw ‐0.926 ‐0.731 ‐5.701
lnargdp_pw ‐2.407 ‐2.167 ‐6.193
lnsrgdp_pw 1.209 ‐0.428 ‐5.729
lnirgdp_pw ‐1.189 ‐1.696 ‐4.887
lncapnet_pw 1.747 ‐1.335 ‐4.260
lnhcap_pw ‐1.813 ‐2.283 ‐4.313
lnelect_pw 0.389 ‐3.035 ‐7.596
lnroadtot_pw ‐0.977 ‐2.693 ‐6.457
lnroadtotsq_pw ‐0.425 ‐2.560 ‐5.461
lnaspurb_pw ‐1.630 ‐1.194 ‐5.707
lngrvroad_pw ‐1.599 ‐1.553 ‐5.903
lnrdcapnet_pw 1.429 ‐1.954 ‐4.016
Notes: a 1%, 5%, and 10% Critical Values at levels are ‐3.662, ‐2.964 and ‐2.614
b 1%, 5%, and 10% Critical Values at levels with trend are ‐4.260, ‐3.548 and ‐3.209
c
1%, 5%, and 10% Critical Values at difference are ‐3.668, ‐2.966 and ‐2.616
Lag length of three is chosen to be the optimal lag length
* Based on the graphics inspection, de‐trending is made for variables showing a trend
behavior

As the above table indicates, all of the variables are found to be non‐stationary in
levels whereas few of the variables are marginally stationary in the case of ADF test,
only at 10% significant value. The ADF test at times might lead to acceptance of
stationarity even though the variables are not stationary. For this reason, only the

127
Ibrahim Worku: Road sector development and economic growth in Ethiopia

pperron test is chosen. Both tests indicate that all of the variables seem to be
stationary at first difference. In other words, all of the variables are found to be
integrated of order 1, and hence are called I (1) variables. Thus, to be safe the order
of integration for all the variables is considered to be I (1)

Regarding the existence of trend component within a time series data set,
Harris 1995 underlined that trend in a data set can lead to spurious correlation
that may imply relationships between the variables in a regression equation,
when all that is present are correlated time trends. The time trend in a trend
stationary variable can either be removed by regressing the variable on time
trend (with the regression forming a new variable which is trend‐free and
stationary) or nullified by including a deterministic time trend as one of the
regressers in the model. Harris (1995)

4.4.2. Test for co‐integration

Having tested our time‐series for stationarity, the next step of time‐series analysis is
testing for co‐integration, which amounts to checking whether the linear
combination of the variables is also stationary or not. It requires that the variables of
interest have the same order of integration. It is only when the variables are
integrated of the same order that a linear relationship among them can be
expected. Variables are said to be co‐integrated if a long run equilibrium relationship
exists among them.

Engle and Granger (1987) argue that for such relationships to exist, the error terms
of the model should be stationary. We have applied the Engle‐Granger procedure to
test for co‐integration. When variable x is said to Granger‐cause variable y if, given
the past values of y, past values of x are useful for predicting y. The first stage of the
co‐integration test involves estimating model/equation and saving the error terms.
In time series analysis, estimating a relationship between non‐stationary variables
that are not co‐integrated gives rise to the problem of spurious regression; the error
term in the regression is non‐stationary, producing a high degree of "noise" in the
relationship, and inconsistent parameter estimates. In order to test the stationary
nature of the data set both the ADF and pperron tests are applied on the error
terms. If the error terms are found to be stationary, the variables are said to be co‐

128
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

integrated and this necessitates the estimation of an Error Correction Model (ECM)
involving long run relationships. If, on the other hand, the variables are not co‐
integrated, then the regression leads to spurious results. A number of approaches
are forwarded to deal with such kind of problem. Modeling should proceed with the
differenced time‐series. Annex Table A.1 reports the test statistics from the unit root
tests. As can be seen from the table, the first and second stage estimation results of
the ECM show that ECM could be applicable to estimate the model.

The Engle‐Granger two step estimates indicate that the variables are co‐integrating
at 10% significant level. The Durbin‐Watson test statistic also shows the existence of
serial correlation in both models. Therefore, it is possible for us to estimate the
model using ECM. However the problem with OLS is that it fails to account for the
problem of endogeneity within the model, which is created, in particular, due to the
existence of physical capital as one potential explanatory variable in the model that
is significantly correlated with GDP. In addition, it is also difficult to deal with a
model of variable with a suspect of heteroskedasticity of the error terms.

4.4.3. Method of estimation and pre‐estimation requirements

When the above problems are persistent in a time series data set, the best way to
deal with is to opt for generalized method of estimation (GMM) technique for a
number of significant advantages. The recently devised two‐step GMM estimator
would be an ideal tool to deal with multiple time series data with endogenous
variable and as well is suspected of having heteroskedasticity. According to Baum et
al. (2007), unlike instrumental variables or two‐stage least square estimators, GMM
estimator does not require additional assumption on the error terms. The former
estimators are specific cases of GMM estimation, which require homoskedastic
assumption and independent error terms. Without stating those assumptions one
can estimate a model efficiently and consistently using two‐step GMM estimation
technique. In addition, Baum et al. (2007) state that the two‐step GMM estimator
could also be applied when the errors are serially correlated, which is a typical
nature of time series data set that we are now dealing with. (See also Bond, Hoeffel,
and Temple 2001).

129
Ibrahim Worku: Road sector development and economic growth in Ethiopia

In general, the model that is used to estimate the time series data will have the
following econometric specification.

y t = β i x t + ε t .(17)

β i s’ are the coefficients that are going to be estimated.

Since some of the regressors are endogenous, E( xt ε t ) ≠ 0.

To estimate this model, it requires instruments for endogenous variables in such a way
that the instrumented variable need not be correlated with the error term so that

E(Zt,
ε t ) = 0.

Lanne and Saikkonen, (2009) highlight the problem in using the first‐differenced
GMM panel data estimator to estimate cross‐country growth regressions, which
essentially is similar to time series data set. When the time series are persistent, the
first‐differenced GMM estimator can poorly behave, since lagged levels of the series
provide only weak instruments for subsequent first‐differences. According to the work
of Caselli, Esquivel and Lefort (1996) this problem may be much more serious in practice.
The authors suggest using a more efficient GMM estimator that exploits stationarity
restrictions, and this approach is shown to give more reasonable results than first‐
differenced GMM in our estimation of an empirical growth model. Accordingly, this
study utilized two‐step efficient GMM estimator to handle this problem.

To do so, the model is estimated using two‐step GMM estimator with the robust
option for the reason that such an estimate is liable for inefficient estimate of
downward biased standard errors. So, the robust two‐step estimation procedure is
chosen to estimate the above specified model with an automatic bandwidth
selection with robust standard error.

Another key issue when estimating a model using two‐step GMM is that capital is
assumed to be endogenous. In addition, human capital is suspected to be potentially
endogenous. To account for these restrictions, Baum et al. (2007) estimation
technique allows us to instrument the model for the problem of endogeniety and

130
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

setup ortogonality condition on the suspect of instrument. Thus these facts are well
taken care of in the estimation procedure with an ivreg2, gmm2s orthog bw( ) stata
installed command.

After conducting Granger causality test, for instance, Rouvinen (2002) also used
lagged levels from the second one onwards as instruments. In this study, to avoid
over‐fitting the model, the second and third lags and lagged differences are used as
potential instruments.

Care has to be made over the selection of ideal instrument, when estimating time series
model with endogenous variable. After estimating the model, one has to cautiously look
at test of exogeniety of the instruments before interpreting the result.

4.4.4. Discussion on post‐estimation requirements and discussion on the results

Having paid enough attention for all the pre‐estimation factors regarding two‐step
efficient GMM estimation, the resulting summary table on eight models is presented
below (Table 9). These models are regressions of the growth function, with
dependent variables labeled as total real GDP worker, on agriculture, service and
industrial GDP.

lnrgdp_pw = f (lncapnet_pw, lnhcap_pw, lnelect_pw, lnroadtot_pw, lnroadto

tsq_pw, prersdp, rsdp1, rsdp2, rsdp3, year)
lnrgdp_pw = f (lncapnet_pw, lnhcap_pw, lnelect_pw, lnaspurb_pw, lngrvroad_pw,
lnroadtotsq_pw, prersdp, rsdp1, rsdp2, rsdp3, year)
lnargdp_pw = f (lncapnet_pw, lnhcap_pw, lnroadtot_pw, lnroadtotsq_pw,
prersdp, rsdp1, rsdp2, rsdp3, year)
lnargdp_pw = f (lncapnet_pw, lnhcap_pw, lnaspurb_pw, lngrvroad_pw,
lnelect_pw, lnroadtotsq_pw, prersdp, rsdp1, rsdp2, rsdp3, year)
lnirgdp_pw = f (lncapnet_pw, lnhcap_pw, lnroadtot_pw, lnelect_pw,
lnroadtotsq_pw, prersdp, rsdp1, rsdp2, rsdp3, year)
lnsrgdp_pw = f (lncapnet_pw, lnhcap_pw, lnelect_p, lnroadtot_pw,
lnroadtotsq_pw, prersdp, rsdp1, rsdp2, rsdp3, year)
lnsrgdp_pw = f (lncapnet_pw, lnhcap_pw, lnaspurb_pw, lnelect_pw,
lngrvroad_pw, lnroadtotsq_pw, prersdp, rsdp1, rsdp2, rsdp3, year)
Where lnrgdp_pw, lnargdp_pw, lnirgdp_pw and lnsrgdp_pw are logarithm of real
GDP of total, agricultural, industrial, and service GDP in per worker units respectively.

131
Ibrahim Worku: Road sector development and economic growth in Ethiopia

lncapnet_pw is capital stock net of road in per worker units

lnhacap_pw is logarithm of human capital in per worker units
lneclect_pw denotes logarithm of electricity power generation in per worker units
lnroadtot_pw is the logarithm of total road network in per worker
lnaspurban_pw is the logarithm of asphalt urban road network in per worker
lngrvroad_pw is the logarithm of gravel road network in per worker
prersdp is policy dummy for the period 1991‐1997 to account for change of regime
rsdp1 is policy dummy for the period 1997‐2001 to account for the first RSDP program
rsdp2 is policy dummy for the period 2002‐2006
rsdp3 policy dummy for the period 2006‐2009
year time dummy included to account for trending nature of the fundamental
components of the model

For all the above alternative models estimation is undertaken. But, before
interpreting the finding of the regression result, the following post‐estimation issues
are considered. Otherwise the parameters would not be considered as efficient and
consistent to make a valid statistical inference.

Stata estimate for two‐step GMM estimation is produced with a number of

statistical outputs that are indicators of the efficiency of the estimators. For the
parameters to be interpreted, the first task is to check the consistency of these
statistical results with the expected hypothesis that need to be satisfied (Baum et al.
2010). According to Hayashi (2000), two‐step efficient GMM estimator needs to
pass test of exogeneity of instruments, weak identification and overidentifying
restrictions. These three statistics are of our interest that the estimated models
need to satisfy.

First, we need to test that the instruments are exogenous with respect to the
instrumented variable. Theoretically, this condition is satisfied when lagged
differences and lag levels are taken as instruments with GMM estimates. In this
regard, Lanne and Saikkonen (2009), for instance, argue that when there is causality,
as causality is already tested using Engle‐Granger two‐step procedure, within a time
series framework it would be appropriate for one to use lagged variables and lagged
differences as best instruments in GMM estimation. Following this proposition,
capital is instrument by its second and third lagged level and lagged difference. In

132
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

this regard, Hayashi (2000) sets the requirement over the null statistics. The C
statistic (also known as a "GMM distance" or "difference‐in‐Sargan" statistic) allows
us to test exogeneity of one or more instruments. Under the null hypothesis that
both the smaller set of instruments and the additional, suspect instruments are
valid, the C statistic is distributed as chi‐squared in the number of instruments
tested. Note that failure to reject the null hypothesis requires that the full set of
orthogonality conditions be valid. Test of exogeneity is reported along with the
summary table for all the models under consideration. The models pass the test of
exogeneity.

Second, the instruments should be relatively highly correlated with the

instrumented variable, so that they act as an effective predictor. Weak identification
tests are a category of tests to check for this condition, and the specific version the
stata routine reports Kleibergen‐Paap rk Wald F statistics. When the independently
and identically distributed (i.i.d.) assumption is dropped and ivreg2 reports
heteroskedastic, AC, HAC or cluster‐robust statistics, the Anderson LM and Cragg‐
Donald Wald statistics are no longer valid. In these cases, the LM and Wald versions
of the Kleibergen‐Paap (2006) rk statistic, stata output reports also distributed as
chi‐squared with (L1‐K1+1) degrees of freedom. The rk statistic can be seen as a
generalization of these tests to the case of non‐i.i.d. errors. To pass the test, a value
of ten or greater is typically required. As shown in the regression summary table
(Table 9), for each model these instruments pass the test.

Third, the instruments should not influence the dependent variable except via the
instrumented variable. Overidentification tests are a category of tests used to test
for this condition, and the specific version which is presented with the Stata routine
output is called the Hansen J statistic. According to Hayashi (2000), under the
assumption of conditional homoskedasticity, Hansen's J statistic is consistent in the
presence of heteroskedasticity and (for HAC‐consistent estimation) autocorrelation,
where a p‐value result that is greater than 0.1 is typically considered passing. As
shown in Table 9, the three instruments pass this test. For all the models under
consideration, the Hansen J statistics is greater than 0.4.

Stata output also automatically reports tests of both underidentification and weak
identification. The underidentification test is an LM test of whether the equation is

133
Ibrahim Worku: Road sector development and economic growth in Ethiopia

identified, i.e., the excluded instruments are "relevant", meaning correlated with the
endogenous regressors. Under the null that the equation is underidentified, the
statistic is distributed as chi‐squared with degrees of freedom (L1‐K1+1). A rejection
of the null indicates that the model is identified. For the models under
consideration, the models pass the underidentification tests (Baum 2006).

In sum, while estimating the models, the estimator is forced to compute estimators
that are heteroskedasticity and autocorrelation consistent (HAC) standard errors. To
do so, while estimating the models automatic bandwidth selection and robust
option are imposed to produce an estimate which accounts for the existence of
heteroskedasticity and autocorrelation in time series data set (HAC). With the HAC
standard errors, various summary statistics are “robustified” as well, in the sense
that the post estimation criteria are satisfied.

Once all the post estimation issues are well handled, it would be safe for us to report
the estimated results. Accordingly, the following section discusses on the findings of
these results.

134
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

Table 9: Summary on two‐step GMM estimation results of the impact of road network on overall and sectoral GDP
Variables| model1 model2 model3 model4 model5 model6 model7 model8
Dependant| RGDP RGDP ARGDP ARGDP SRGDP SRGDP IRGDP IRGDP
Explanatory
Phy_capital | .356*** .184* ‐.0304 ‐.117* .423*** .43*** .428*** .436***
Hum_capital | .0539*** .0455** .038 .0572* ‐.0585* ‐.0644*** .0661*** .0632***
Electricity | .345*** .165 .352 .294 .286* ‐.00567 .251* ‐.0209
Road | .575* 1.31*** 1.18*** .568**
Road2 | ‐.401** ‐.186 ‐.759*** ‐.537*** ‐.126 ‐.453*** ‐.122
Pre_RSDP | ‐.0921*** .0222 .0209 .129* ‐.378*** ‐.22*** ‐.211*** ‐.0764
RSDP 1 | ‐.0253 .131* .162*** .256*** ‐.341*** ‐.18** ‐.152*** ‐.0284
RSDP 2 | .0316 .117 .162* .15 ‐.253*** ‐.121 ‐.0788 .0264
RSDP 3 | .194*** .222* .444*** .27*** ‐.172* ‐.112 .0808 .126
Year | ‐.0182*
Asphalt | .582*** .585*** .62*** .426**
Gravel | ‐.0699 ‐.374*** .214 ‐.102
Road*Ph_capit|
Constant | 45.3*** 10.9*** 17.3*** 9.1*** 11.7*** 10.2*** 8.05*** 7**
r2_a | .808 .786 .623 .609 .856 .873 .928 .933
rmse | .0416 .0438 .0598 .0609 .0583 .0535 .0481 .0454
N| 35 35 35 35 35 35 35 35
K| 20.406 87.309 106.206 42.852 106.206 131.911 106.206 99.961
K P| .5148 0.5166 0.4921 0.3173 0.5201 0.4201 0.6580 0.5863
J| 0.5368 0.4467 0.3280 0.1302 0.2671 0.2073 0.4782 0.4475
C| 0.8396 0.8894 0.4969 0.9401 0.7833 .7036 0.5633 0.4835
F| 2685 978 157 115 2520 696 2378 1828

Notes: * p<0.05; ** p<0.01; *** p<0.001, all the explanatory variables are in log and per‐worker units

135
Ibrahim Worku: Road sector development and economic growth in Ethiopia

As can be seen from Table 9, eight models are estimated. The general model, the
first model labeled as model1, is on the impact of roads on aggregate real GDP per
worker. The result indicates that road network per worker is positively related with
economic growth.

In the second model, the study examined the impact of different types of road
(classified as asphalt and gravel road) on economic growth, captured by logarithm of
overall and sectoral real GDP growth. The findings indicate that expansion of asphalt
road has positive influence on overall economic growth. The coefficient for asphalt
road is also statistically significant. Similarly, though it is statistically insignificant,
gravel road has positive impact on growth.

Model3 up to model8 are estimated with the aim of investigating the impact of road
network on sectoral output, i.e. agricultural GDP, industrial GDP and service GDP. All
of them are expressed in per worker terms. In order to see the impact of road
quality on total and sectoral GDP, each sector GDP is estimated on asphalt and
gravel road. Accordingly, model3 and model4 are estimation results on the impact of
total and disaggregated road on agricultural GDP. In both cases, road network has
positive influence on the growth of agricultural GDP. The impact of total road is
statistically highly significant. Nonetheless, the impact of the disaggregated road,
asphalt and gravel, fails to be as such significant. The possible explanations would be
that the road sector is not highly integrated with the agricultural sector in terms of
accessibility to the rural dweller. The other possible explanation would be that
disaggregated roads fail to reach the threshold level, which is required spurring
agricultural GDP growth. Moreover, another plausible explanation ascribed for such
unexpected finding might be the fact that the rural area is not as such networked
with asphalt and gravel road and access to these networks is somehow thin which
essentially proves lesser impact on agricultural GDP growth. Rather, in this area
community roads play a more significant role than these two types of road
networks. As already explained in the descriptive analysis section, community road
is not well accounted by the road authority for it to be in incorporated within the
regression equation.

Model5 and model6 are on the impact of road on industrial GDP. In here, again both
total road network and classified roads are found to have positive impact on this

136
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

sector GDP growth. Here again, the asphalt road has significant and positive impact
on industrial GDP. Intuitively, this is true for the reason that the asphalt road is
networked where industries are operating, i.e. city areas. This shows that road
network is concentrated around cities. In contrast, gravel network in urban areas are
not prevalent. Thus, their impact on this sector GDP is negligible.

When we look at model7 and model8, we find that the impact of road on service
sector GDP is positive for the total and asphalt road network. The result is
concurrent with the intuitive perception that road quality is essential for the well
performing service sector.

In sum, the findings of all the previous models are consistent with the intuitive and
theoretical explanation. Having this fact in mind, road network is found to have
much more pronounced impact on sectoral GDP than on total GDP. For the road
sector to have significant output on each sector, sector specific objectives regarding
road network expansion is relevant. That is, for the impact of road on agricultural
GDP to be significant, the road network fails to reach the threshold level for it to
have significant impact on this sector. In addition, gravel road expansion is not as
such important for increasing service sector GDP.

In the estimation process, I tried to include other interaction terms, but it resulted in
multicollinearity. When it happened with two‐step efficient GMM estimator, the stat
routine command automatically drops those variables. So it is not possible to see the
impact of interaction terms.

Regarding the other inputs used in the models, in all cases the coefficients are found
to be with the theoretical expectation that both human and physical capital have
positive impact on economic growth.

Finally, when we come back to the impact of policy interventions introduced over
the period under study, for all of the models RSDP III, which has been introduced
over the period 2006‐2009, has more significant growth spurring impact than the
other two. Here it requires careful interpretation in that, RSDP II also has positive
and significant impact for most of the models. We need to note that policy
intervention on infrastructure would have a pronounced impact in later years than

137
Ibrahim Worku: Road sector development and economic growth in Ethiopia

the moment the policy is introduced. That is, we need to be cognizant of the fact
that such kind of policies, policy on infrastructure, will have impact lag. It would take
time for it to spur growth than having a direct and shorter impact. Such investments
will have a very strong impact over a long time horizon. In all the models the policy
dummy for the period 1991‐1997 has a negative influence for the country and fails
to have any growth spurring impact. That period was more a period of political
stabilization than growth. The result is also consistent with this fact.

5. Conclusions

This study investigated achievements in the road sector in Ethiopia and its impact on
both the overall and sectoral economic growth. To do so, the study econometrically
analyzed the impact of roads on economic growth by using a time series data set. To
this end, the study reviewed theoretical and empirical researches related to the road
development over a wide array of perspectives. It also reviewed relevant literature
on the link between road transport and economic growth and certain aspects of
road transport in the context of Ethiopia and some other countries experiences.

Ethiopia has a respectable growth performance in the post‐1991 period. In

particular, over the period 2004‐2009, the economy averaged 10 percent growth
rate of GDP, which signals the country’s future economic prospects and lays down
promising momentum for further economic growth. Keeping up the growth
momentum and ascertaining its sustainability is a key to boost domestic investment
and also attract foreign investment. In this regard, to sustain the growth
performance of the economy, through creation of a favorable macroeconomic
environment, developing vital infrastructure, ensuring the quality of institutions as
well as improving the quality of human capital, road network development is
supposed to play a major catalytic role.

The findings from the descriptive analysis indicate that the government in the recent
decade is making a relentless effort towards expanding the road network of the
country. However, an important key indicator is the issue of accessibility. The
country’s overall accessibility is far below from what is needed to graduate to the
club of lower middle income countries. Accessibility is a good indicator to investigate

138
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

the impact of poverty on socioeconomic development. This indicator is linked with

rural population. The issue of access is not a challenge for the urban population.
Therefore, in order to reduce rural poverty or provision of public utilities, expanding
the road network in rural areas would be the best way to reach the rural population.
The recent trend shows that the government is showing strong commitment to
expand and improve the current performance. The analysis also shows that the
government in this decade has made sufficient attention in financing road projects.
By now the government’s expenditure has reached tenfold of what it has been a
decade before. The analysis also reveals that the donor’s share of money spent on
the road network is not increasing as such in the face of the rising road expenditure
by the government and the expansion of the road network.

Community roads should be given sufficient attention both in terms of expansion,

management, and accountancy by either regional or federal road authorities. At this
point, Ethiopian Road Authority should design an easy way to get detailed
information regarding community road networks from regional road authorities.
Future community road expansion need to be an integral part of the road networks
as these might be an easy way to ascertain access to the destitute rural poor.
Community roads are supposed to better reflect the community demand of which
roads should be constructed or upgraded.

When we come to findings of the econometric results according the link between
road length and economic growth, the results indicate that road network per worker
is positively related with economic growth and that expansion of asphalt road has a
positive influence on overall economic growth. Similarly, though statistically
insignificant, gravel road has a positive impact on economic growth.

Finally, at the heart of the above analysis one key question is, are the efforts made
so far sufficient to spur the overall economic growth of the nation and thereby to
have an impact on the livelihood of the rural poor? The impact is seen to be less
strong on the agricultural GDP growth and addressing the issue of accessibility to the
rural poor. The effort is relatively better and has pronounced impact on industrial
and service sector GDP. Nevertheless, for it to have a far‐reaching impact on
agricultural growth and poverty reduction, a lot more has to be done.

139
Ibrahim Worku: Road sector development and economic growth in Ethiopia

References

African Development Indicators (ADI). 2009. CD‐ROM.

Baum, C. F. 2006. An Introduction to Modern Econometrics Using Stata. College Station,
TX: Stata Press.
Baum, C. F., M. E. Schaffer, and S. Stillman. 2007. Enhanced Routines for Instrumental
Variables/GMM Estimation and Testing. Boston College Economics Working
Paper 667.
Baum, C. F., M. E. Schaffer, and S. Stillman. 2010. IVREG29: Stata module for extended
instrumental variables/2SLS and GMM estimation (v9). Boston College
Department of Economics, Statistical Software Components S4254010.
Bond, S., A. Hoeffler, and J. Temple. 2001. GMM Estimation of Empirical Growth Models.
CEPR discussion paper 3048.
Calderón, C. 2009. Infrastructure and Growth in Africa. Policy Research Working Paper
4914. African Sustainable Development Front Office, World Bank.
Caselli, F., G. Esquivel, and F. Lefort. 1996. Reopening the Convergence Debate: A New
Look at Cross‐Country Growth Empirics. Journal of Economic Growth 1 (3): 363‐
389.
Canning, D. 1999. Infrastructure’s Contribution to Aggregate Output. Development
Research Group Public Economics, Policy Research Working Paper 2246.
Washington, D.C.: World Bank.
Canning D. and Bennthan. 2000. The Social Rate of Return on Infrastructure Investments.
Development Research Group, Public Economics and Private Sector
Development and Infrastructure Group. World Bank.
Canning, D., and P. Pedroni. 2004. The Effect of Infrastructure on Long Run Economic
Growth. Harvard University, Williams College.
Central Statistical Authority (CSA). 2008. Summary and Statistical Report of the 2007
Population and Housing Census, Population Size by Age and Sex. Addis Ababa:
CSA.
______. 1970‐2010. Annual Report, Statistical Abstract. Addis Ababa: CSA.
______. National Abstracts. www.csa.gov.et.
Cohen, D., and M. Soto. 2001. Growth and Human Capital: Good Data, Good Results.
OECD Development Centre Technical Papers 179. Paris: OECD.
Dercon, S., D. O. Gilligan, J. Hoddinott, and T. Woldehanna. 2008. The Impact of
Agricultural Extension and Roads on Poverty and Consumption Growth in
Fifteen Ethiopian Villages, Food Consumption and Nutrition Division, IFPRI

140
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

Discussion Paper 00840. Washington, D.C.: International Food Policy Research

Institute.
Dorosh, P., H.‐G. Wang, L. You, and E. Schmidt. 2009. Crop Production and Road
Connectivity in Sub‐Saharan Africa: A Spatial Analysis. Africa Infrastructure
Country Diagnostic (AICD). Washington, D.C.: World Bank.
Engel, R., and C. Granger. 1987. Cointegration and Error Correction: Representation,
estimation and testing. Econometrica 55: 251‐276.
Elliott, G., T. J. Rothenberg, and J. H. Stock. 1996. Efficient Tests for an Autoregressive
Unit Root. Econometrica 64: 813‐836.
Ethiopian Roads Authority (ERA). 2008a. Rural Transport Infrastructure. Addis Ababa.
______. 2008b. Enhancing Rural Accessibility. Concept Paper. Addis Ababa: Federal
Democratic Republic of Ethiopia, Ministry of Works and Urban Development.
Ethiopian Roads Authority (ERA). 2009. RSDP Performance: Twelve Years Later. Addis
Ababa.
Fan, S., and C. Chan‐Kang. 2005. Road Development, Economic Growth, and Poverty
Reduction in Chain. Research Report 138. Washington, D.C.: International Food
Policy Research Institute.
Fan, S., and N. Rao. 2003. Public Spending in Developing Countries: Trend,
Determination and impact. EPTD Discussion Paper 99. Washington, D.C.:
International Food Policy Research Institute.
Fan, S., L. Zhang, and X. Zhang. 2002. Growth, Inequality and Poverty in Rural China, The
Role of Public Investment. Research Report 125. Washington, D.C.: International
Food Policy Research Institute.
Hayashi, F. 2000. Econometrics. Princeton: Princeton University Press.
Harris, R. I. D. 1995. Using Cointegration Analysis in Econometric Modelling. FT Prentice
Hall Publication.
International Monetary Fund. http://www.imf.org/external/data.htm
International Road Federation Statistical Report. 2006. World Road Statistics.
www.irfnet.org.
Kleibergen, F., and R. Paap. 2006. Generalized Reduced Rank Tests Using the Singular
Value Decomposition. Journal of Econometrics 127(1): 97–126.
Kohler, F. D. 1988. The Effect of Defense and Security on Capital Formation in Africa: An
Empirical Investigation. Rand Publication, N‐2635‐USDP, Santa Monica.
Lanne, P., and M. Saikkonen. 2009. Noncausal Autoregression, Instrumental Variables,
Test of Overidentifying Restrictions. Journal of Economic Literature.

141
Ibrahim Worku: Road sector development and economic growth in Ethiopia

Ministry of Finance and Economic Development (MoFED). 2006. Ethiopia: Building on

Progress, A Plan for Accelerated and Sustained Development to End Poverty
(PASDEP) (2005/06‐2009/10). Volume I. Addis Ababa.
______. Macroeconomic Unit, data base.
Phillips, P. C. B., and P. Perron. 1988. Testing for a Unit Root in Time Series Regression.
Biometrika 75: 335‐346.
Richaud, C., K. Sekkat, and A. Varoudakis. 1999. Infrastructure and Growth Spillovers: A
Case for Regional Infrastructural Policy in Africa. OECD mimeo
Rouvinen, P. 2002. R&D–Productivity Dynamics: Causality, Lags, and Dry Holes. Journal
of Applied Economics V (1): 123‐156.
World Bank. 2010. Ethiopia’s Infrastructure: A Continental Perspective. Africa
Infrastructure Country Diagnostic Country Report, The International Bank for
Reconstruction and Development. Washington, D.C. World Bank.

142
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

Annex Table A.1. Error Correction Model

reg lnsrgdp_pw lncapnet_pw lnelect_pw lnhcap_pw lnroadtot_pw rsdp2 rsdp3 year
Source| SS df MS Number of obs = 39
Model | 1.63507733 7 .233582476 F( 7, 31) = 73.76
Residual| .098164414 31 .003166594 Prob > F = 0.0000
Total| 1.73324175 38 .045611625 R‐squared = 0.9434
Adj R‐squared = 0.9306
Root MSE = .05627
lnsrgdp_pw Coef. Std. Err t P>|t| [95% Conf. Interval]
lncapnet_pw .7795128 .1345526 5.79 0.000 .505091 1.053935
lnelect_pw | ‐.0150187 .1194418 ‐0.13 0.901 ‐.2586219 .2285846
lnhcap_pw | .1508255 .0337444 4.47 0.000 .0820033 .2196477
lnroadtot_pw .0550367 .1519828 0.36 0.720 ‐.2549343 .3650076
|
rsdp2 | .1385427 .0381224 3.63 0.001 .0607916 .2162937
rsdp3 | .2481983 .0666627 3.72 0.001 .1122389 .3841577
year | ‐.0448544 .0079673 ‐5.63 0.000 ‐.0611038 ‐.0286051
_cons | 91.0019 14.73099 6.18 0.000 60.95784 121.046

. predict resid
. estat dwatson
Durbin‐Watson d‐statistic( 8, 39) = 1.22548
. pperron resid, noconstant regress

Phillips‐Perron test for unit root Number of obs = 38

Newey‐West lags = 3

‐‐‐‐‐‐‐‐‐‐ Interpolated Dickey‐Fuller ‐‐‐‐‐‐‐‐‐‐‐‐

Test 1% Critical 5% Critical 10% Critical
Statistic Value Value Value
‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
Z(rho) 0.148 ‐12.420 ‐7.508 ‐5.404
Z(t) 1.966 ‐2.639 ‐1.950 ‐1.605
‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
resid | Coef. Std. Err. t P>|t| [95% Conf. Interval]
‐‐‐‐‐‐‐‐‐‐‐‐‐+‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
resid |
L1. | 1.003896 .00181 554.63 0.000 1.000229 1.007564
‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐

. estat dwatson
Durbin‐Watson d‐statistic( 1, 38) = 1.682735

143
Ibrahim Worku: Road sector development and economic growth in Ethiopia

Annex Table A.2: Stata output on the Engle‐granger two‐step procedure test of co‐integration
. reg lnsrgdp_pw lncapnet_pw lnelect_pw lnhcap_pw lnroadtot_pw rsdp2 rsdp3 year

Source | SS df MS Number of obs = 39

‐‐‐‐‐‐‐‐‐‐‐‐‐+‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐ F( 7, 31) = 73.76
Model | 1.63507733 7 .233582476 Prob > F = 0.0000
Residual .098164414 31 .003166594 R‐squared = 0.9434
Adj R‐squared = 0.9306
Total | 1.73324175 38 .045611625 Root MSE = .05627

lnsrgdp_pw | Coef. Std. Err. t P>|t| [95% Conf. Interval]

lncapnet_pw| .7795128 .1345526 5.79 0.000 .505091 1.053935
lnelect_pw | .0150187 .1194418 0.13 0.901 .2586219 .2285846
lnhcap_pw | .1508255 .0337444 4.47 0.000 .0820033 .2196477
lnroadtot_pw | .0550367 .1519828 0.36 0.720 .2549343 .3650076
rsdp2 | .1385427 .0381224 3.63 0.001 .0607916 .2162937
rsdp3 | .2481983 .0666627 3.72 0.001 .1122389 .3841577
year | .0448544 .0079673 5.63 0.000 .0611038 .0286051
_cons | 91.0019 14.73099 6.18 0.000 60.95784 121.046
. predict resid
. estat dwatson
Durbin‐Watson d‐statistic ( 8, 39) = 1.22548
. pperron resid, noconstant regress
Phillips‐Perron test for unit root Number of obs = 38
Newey‐West lags = 3

Interpolated Dickey‐Fuller
Test 1% Critical 5% Critical 10% Critical
Statistic Value Value Value
‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
Z(t) 1.966 ‐2.639 ‐1.950 ‐1.605

‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
resid | Coef. Std. Err. t P>|t| [95% Conf. Interval]
‐‐‐‐‐‐‐‐‐‐‐‐‐+‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
resid | L1. | 1.003896 .00181 554.63 0.000 1.000229 1.007564

144
 Ethiopian Journal of Economics, Volume XIX, No. 2, October 2010

Annex Figure A.1: Graphic visual inspection test on the existence of trend
8 .2

7 .4
7 .2
8

ln a rg d p _ p w
ln rg d p _ pw
7.8

7
7 .6

6 .8
7 .4

6 .6
1970 1980 1990 2000 2010 1970 1980 1990 2000 2010
Year Year
6 .4
7 .2

6 .2
7
ln s rg d p_ pw

ln irg d p _ p w
6 .8

6
6 .6

5 .8
6.4

5 .6

1970 1980 1990 2000 2010 1970 1980 1990 2000 2010
Year Year
-8 .4
8 .5
8

-8 .6
ln a s p u rb _ p w
7 .5
ln ca p _ p w
7

-8 .8
6 .5

-9
6

1970 1980 1990 2000 2010 1970 1980 1990 2000 2010
Year Year

145
Ibrahim Worku: Road sector development and economic growth in Ethiopia

-7 .2
-7 .6

-7 .3
-7 .7

ln u rb ro a d _ p w
ln g rv u rb _ p w

-7 .4
-7 .8

-7 .5
-7 .9

-7 .6
-8

1970 1980 1990 2000 2010

1970 1980 1990 2000 2010 Year
Year

-6.8
-7

-7
-8

-7.2
ln grvro ad_pw
lnrurgroad_pw
-9

-7.4
-10

-7.6
-11

-7.8
-12

1970 1980 1990 2000 2010 1970 1980 1990 2000 2010
Year Year
4
-6.8

3.5
lnroadtotsq_ pw
-7

3
ln roa dtot_ pw

2.5
-7.2

2
-7.4

1.5

1970 1980 1990 2000 2010

1970 1980 1990 2000 2010 Year
Year
11
19

10
18

ln ro a d th c a p _ p w
ln roa dtcap _p w

9
17

8
16

7
6
15

1970 1980 1990 2000 2010 1970 1980 1990 2000 2010
Year Year

146