job flows-130410 adm - iza institute of labor economics · above average job reallocation rate...
TRANSCRIPT
The Dynamics of Job Creation and Job Destruction: Is Sub-Saharan Africa Different?
Admasu Shiferaw University of Göttingen, Germany
Arjun S. Bedi International Institute of Social Studies, Erasmus University Rotterdam, The
Netherlands
April 2010
1
Abstract This paper analyzes the creation, destruction and reallocation of jobs in order to understand the micro-dynamics of aggregate employment change in African manufacturing. The nature and magnitude of gross job flows are examined using a unique panel data of Ethiopian manufacturing establishments covering the period 1996-2007. The paper also assesses the relative importance of firm demographics, size, capital intensity and the relative importance of within and across industry effects for job flows. The analysis shows that the rates and patterns of job creation and destruction in Ethiopia are comparable to patterns observed in other developed and emerging economies suggesting that African firms adjust their labor force in a manner broadly similar to firms elsewhere and that African labor markets are not uniquely restrictive in terms of undermining job reallocation across firms. We also find, as in other countries, that job reallocation is predominantly an inter-industry phenomenon and is relatively higher in industries dominated by smaller and younger establishments. The analysis shows that while there is no shortage of new and small enterprises that create jobs at the moment of entry, such establishments contribute less to job creation through post-entry expansion. The similarity in the micro-dynamics of employment generation in an African setting as compared to other countries suggests that the poor employment generation observed in African manufacturing may not be attributed to any peculiar features of African labor markets.
JEL Classification: J20, J23, J49 Key Words: Job Creation, Job Destruction, Job Reallocation, Firm Dynamics, Sub- Saharan-Africa, Ethiopia
2
1. Introduction
A growing share of manufacturing in GDP and in employment is a common feature
observed in successful developing countries. Manufacturing, however, has not been a
major source of gainful employment for the African labor force. The sector accounts for
less than 10 percent of total employment in the region except for the island economy of
Mauritius where it accounts for about 25 percent of employment.1 From a macroeconomic
perspective, among others, the lack of growth in manufacturing has been attributed to the
low rates of investment (Collier and Gunning, 1999), low level of domestic demand for
manufactures in Sub-Saharan Africa and the lack of export orientation of its manufacturing
sector. Notwithstanding the relative merits of the various aggregate level explanations, the
micro-dynamics underlying the lackluster aggregate employment performance of African
manufacturing and whether or not the underlying firm level processes of job creation and
job destruction which shape aggregate employment outcomes are different from the rest of
the world is not yet known.
Recent micro level studies indicate that the behavior of African manufacturing firms is not
very different from their counterparts in other developing and advanced economies despite
the fact that African manufacturing is still at an incipient stage. For instance, small firms in
African manufacturing grow faster than large firms (Bigsten and Gebreeyesus, 2007;
Gunning and Mengistae, 2001; Van Biesebroeck, 2005) while relatively efficient firms
stand better chances of survival as in other parts of the world (Frazer, 2005; Shiferaw,
2007, 2009; Söderbom et al., 2006). While some of these studies (Bigsten and
Gebreeyesus, 2007; Gunning & Mengistae, 2001; Van Biesebroeck, 2005) have examined
age and size effects on firm level net employment growth they do not examine the
dynamics of gross job flows - job creation, job destruction and job reallocation - which
underlie aggregate net employment growth. The aim of this article, which is based on data
1 This is far less than the nearly 30 percent share of manufacturing in total employment in East Asia, and the approximately 20 percent share in developed countries (ILO, 2009). In Ethiopia, Denu et al. (2005) note that in 1999, manufacturing employment accounted for 4.45 per cent of total employment. This figure includes employment in small‐scale and cottage/handicraft manufacturing establishments. Employment in medium and small scale manufacturing (more than 10 workers) accounts for about 10 percent of total manufacturing employment, and as shown in Figure 1 rose from about 80,000 workers in 1996 to about 115,000 workers in 2007.
3
from a Sub-Saharan African country, is to identify the underlying firm level patterns in
gross job flows and to provide a comparative analysis of the patterns observed in an
African labor market with those observed in other settings.
As reviewed in some more detail in the next section, the growing body of firm-level
analyses of gross job flows in manufacturing yields some clear stylized facts. For example,
in the context of developed countries, Davis and Haltiwanger (1992) and Baldwin et al.
(1998), show that there are high rates of job creation and destruction, in excess of 10
percent a year even within narrowly defined industries, reflecting the simultaneous creation
and destruction of jobs and the substantial firm-level heterogeneity in employment. Firm
level employment adjustments are mostly persistent rather than transitory, and adjustment
rates tend to be higher among smaller and younger firms as compared to larger and older
firms. Especially in the case of developed countries, job destruction is more volatile than
job creation which implies that the reshuffling of jobs across firms is countercyclical.
Researchers have also associated gross job flows with firm demographics to show the
relative contributions of the birth, expansion, contraction and death of firms. The growing
number of firm level studies for different countries, and the cross-country variation in the
rate of job flows observed in these studies serves as an indicator of the degree of labor
market flexibility and the efficiency of resource allocation. While Haltiwanger et al. (2008)
provide the most recent cross-country analysis of job flows using harmonized firm level
data for 16 countries, no African country features in their sample. In fact there are no
comparable studies on the nature and magnitude of job creation and destruction in Sub
Saharan African using establishment level data.
The current paper contributes by providing a firm level analysis of job flows in the context
of Sub-Saharan Africa. Apart from offering a rare description of job flows in this region, the
analysis is motivated by the following inter-related and policy relevant questions: Is the
lackluster aggregate employment performance of African manufacturing a result of limited
job creation, relative to other countries, or is it a result of simultaneous processes of job
destruction and job creation offsetting each other? More broadly, are African labor
markets flexible enough to accommodate a smooth reallocation of labor to its best use?
How distinct are firm demographics in this region and their relative importance for job
4
flows? Do business cycles play an important role in driving observed patterns of job flows
as compared to industry and firm-specific characteristics? These questions cannot be
answered by examining net employment change at higher levels of aggregation as the
latter could be consistent with any underlying rates of job creation and destruction. In this
paper we address these questions by analyzing job creation and destruction rates in
Ethiopian manufacturing using a unique census based establishment level panel data
covering the period 1996-2007. The nature of the data also allows us to measure the
relative importance of establishment entry and exit as well as establishment expansion and
contraction for job flows so that we can draw a more complete picture of employment
dynamics. Moreover, the relatively long span of the data and the distinct business cycles
that it captures allows us to determine whether observed patterns of job creation and job
destruction are primarily driven by business cycles or by technological factors and
employer specific characteristics.
The remainder of the paper is organized as follows. The next section provides a review of
job flows, highlighting the key stylized facts. Section 3 provides a description of the data
and briefly discusses the business environment and Ethiopia’s manufacturing sector.
Section 4 introduces the measurement framework. Section 5 provides a temporal and
cross-industry analysis of variations in job flows and decomposes job flows along various
dimensions. An econometric analysis of job flows using industry level characteristics is
provided in section 6 while section 7 contains concluding remarks and policy reflections.
2. Job flows – A review
Firm heterogeneity in employment is a prominent feature of gross job flows in developed
and emerging economies. Firms producing similar products experience a simultaneous
process of job creation and destruction and exhibit large variations in the rates of job
creation and destruction. For instance, during the 1970s and 1980s, new jobs were
created at the rate of 10 percent per annum in the manufacturing sectors of the USA and
Canada while job destruction occurred simultaneously at a comparable rate. In a recent
paper, Haltiwanger et al. (2008) study job flows in 16 developed and emerging economies
using harmonized firm level data sets from the 1990s. Their work goes beyond the results
obtained from a number of country specific studies and provides interesting insights on the
5
distribution of job flows across countries. They find job creation rates of about 12.7, 14.8
and 17.4 percent for OECD, Latin American and transition economies, respectively, with
corresponding job destruction rates of 12.7, 14 and 12.8 percent. When economic reforms
began in transition economies in the early 1990s, job destruction rates were much higher
than job creation rates before coming closer to that of OECD countries in the late 1990s
(Faggio et al., 2003).
In terms of patterns across industries, Haltiwanger et al. (2008) find a positive rank
correlation of job reallocation rates (the sum of job destruction and job creation rates)
across industries in their sample of 16 countries suggesting that some industries have
above average job reallocation rate across all countries. More specifically, Baldwin et al.
(1998) show that industry job-flow patterns are very similar across the United States and
Canada and that sectors which have a high job reallocation rate in Canada also have a
high job reallocation rate in the United States (a correlation of 0.83). These patterns
suggest that there may be common industry level characteristics such as technology, cost
and demand factors that drive job reallocation patterns across different countries. Two
other points which emerge are that the overwhelming fraction of job reallocation occurs
within industries rather than across industries and that firm decisions to create and destroy
jobs tend to be persistent, reflecting adjustments toward desired firm size rather than
temporary layoffs and rehires (Davis and Haltiwanger, 1990, 1992).2
Beyond these basic patterns an interesting aspect of the literature on job flows is the
cyclical nature of job reallocation across producers. In developed countries, the job
reallocation rate is countercyclical, that is, it intensifies during recessions or periods of net
employment loss. Baldwin et al. (1998) show that net employment growth in the
manufacturing sectors of the US and Canada is accompanied by a reduction in job
destruction rate without significant improvement in job creation rate. Similarly, net
employment loss at the aggregate level is mainly associated with a rapid increase in job
destruction with only a slight decrease in job creation. In other words, the variance of job
2 For instance, Baldwin et al. (1998) find that only 2 to 4 percent of excess job reallocation in Canada and the United States may be attributed to shifts across industries. Davis and Haltiwanger (1992) report that the average one‐year persistence rates for annual job creation and destruction lie between 68 and 81 percent.
6
destruction is higher than that of job creation leading to the countercyclical movement of
job reallocation. Campbell and Fisher (2000) argue that this is partly because of
asymmetric adjustment costs. Job creation not only involves the actual adjustment cost of
hiring new workers but also the expected cost of future separation, making job creation
less responsive to business cycles than job destruction. It should be noted that the
countercyclical nature of job reallocation is not universal. Even for the US, the
countercyclical pattern is observed mainly among larger and older firms while firms in
transition economies, where average firm size is smaller than that of the US, exhibit pro-
cyclical movements in job reallocation (Haltiwanger et al. 2008).
Turning to correlates of firm heterogeneity in job flows, various authors have examined the
role of characteristics such as firm size and age, and technological characteristics such as
industrial affiliation and capital intensity. In general, job creation and destruction rates
decline with firm size and age although at the aggregate level the bulk of (or size weighted)
gross job flows is accounted for by larger and older firms (Davis and Haltiwanger, 1990,
1992; Haltiwanger et al., 2008). Similarly, explicit decomposition analyses of firm
demographics and job flows have shown that during the 1970s and 1980s, new
establishments accounted for 20 percent of job creation in US manufacturing while firm
closures accounted for 25 percent of job destruction (Davis and Haltiwanger, 1990). The
bulk of labor adjustment therefore takes place among continuing firms. Comparable
numbers are not available for transition economies as the existing firm level data for these
countries does not capture firm entry and exit.
With empirical evidence on job flows emerging for a growing number of countries, the
cross country variation in job reallocation across firms has become an important indicator
of labor market flexibility. In this regard the experience of developed countries, particularly
the US, serves as a benchmark to gauge the efficiency of labor allocation in emerging
economies. Haltiwanger et al. (2008) show that while a large share (close to 60 percent) of
the cross country variation in job flows can be explained by industry and firm size effects
(the firm size effect being dominant), a significant proportion of cross country variation
maybe linked to differences in labor market regulations and differences in business
environment across countries. Thus, countries with restrictive labor laws exhibit relatively
7
less reallocation of jobs across firms, the reductions being stronger in those industries with
inherently high job reallocation rates.
3. Data and background
This paper uses establishment level panel data from Ethiopian manufacturing covering the
period 1996-2007. The data come from the annual manufacturing census carried out by
the Central Statistical Authority (CSA) of Ethiopia. The census covers all establishments
with at least 10 workers each of which are identified by unique identification numbers. The
number of establishments increases from 623 in 1996 to 1339 in 2007 and contains a total
of 10,305 observations (establishment-years). About two-thirds of the manufacturing
establishments are located in and around the capital city, Addis Ababa, and about 70
percent are small producers with less than 50 employees.
In 1992, Ethiopia launched a comprehensive set of economic reforms marking the
country’s transition to a market based economy after 17 years of socialism and military
dictatorship. The first few years saw the opening up of the economy to international trade
and witnessed greater participation of the private sector (Shiferaw, 2009). Except for a rise
in inflationary pressure since 2005, macroeconomic conditions have been stable and the
government continues to spend aggressively on physical infrastructure.3 The economy
grew at a respectable average annual rate of 5.6 percent during the 1990s and has
continued to grow at even higher rates (average annual rate of about 8 percent) since the
turn of the century. The business environment has also witnessed substantial
improvements. According to the World Bank’s “Doing Business” report for 2009, it takes 7
procedures to start a new business in Ethiopia as compared to the African average of 10
procedures. The time taken to go through these procedures has declined from about 44
days in 2003 to 16 days in 2009 which is again far less than the 2009 regional average of
about 45 days. However, other aspects of the business climate are not as favorable. The
legal system remains unreliable and it takes an average of 690 days to enforce contracts
3 Between 2002 and 2007, the Ethiopian government’s capital expenditure increased from 31.8 percent to 51.8 percent of total government expenditure within which the share of economic development projects (such as roads, bridges and dams) increased from 20 percent in 2002 to 32 percent in 2007. The remainder is spent on social development projects and multi‐sector development projects (IMF, 2008).
8
and the country is ranked very low (below 100) in terms of protecting investors and
registering property. Although labor laws have been adjusted twice (in 1993 and 2003), the
country is ranked 94 in the world in terms of the ease of hiring and firing workers. Since
the independence of Eritrea in 1993, Ethiopia has become landlocked and has shifted its
trade gateway almost entirely to the smaller and more expensive port of Djibouti in the
aftermath of the 1998-2000 border conflict with Eritrea. Political tension and uncertainty
remain high, particularly since the disputed elections in 2005 that seriously damaged the
domestic and international standing of the current government.
Unsurprisingly, the Ethiopian manufacturing sector is dominated by light consumer goods
industries. About 60 percent of total manufacturing employment is located in the textile and
garments (36 percent) and food and beverage (24 percent) industries. These two
industries also account for about 50 percent of total manufacturing sales. In terms of
temporal patterns, Figure 1 plots manufacturing employment and sales in Ethiopia
between 1996 and 2007. The figure reveals very little change in manufacturing
employment during the first six years of the sample period culminating with an absolute
decline in 2001. Since 2002, this trend has reversed and the sector has experienced
strong employment growth. The figure suggests a natural way of dividing the data in order
to examine cyclical patterns in job flows, that is, a period 1996 to 2001 and a period since
2002 and one of the tasks we undertake in this paper is to breakdown this aggregate
picture and investigate the underlying micro-dynamics in line with the literature on gross
job flows.
4. Measuring job flows
This section introduces various concepts and outlines the framework used in this paper
and the existing literature to measure employment changes. We begin by defining a
measure of establishment level employment growth which is in turn linked to industry level
measures of job creation and job destruction. Employment growth, g at the establishment
level, between time t-1 and t is given by,
1
10.5ijt ijt ijt
ijtijt ijt ijt
X X Xg
m X X
, (1)
9
where X is the number of employees and, i and j index establishments and industries,
respectively. Equation (1) shows a growth rate calculation in which change in employment
is divided by average establishment size between two periods ijtm rather than by initial
size as conventionally calculated. This approach is widely used in the literature on job
flows and offers a number of advantages. It minimizes measurement problems in growth
rate due to transitory low/high initial and end of period establishment sizes that may lead to
overestimation of the expansion of small establishments or the contraction of large
establishments — a bias that could generate a negative association between
establishment size and growth. The formula yields a symmetric distribution of growth rates
centered about zero and is bounded in the interval [-2, 2] which corresponds, respectively,
to establishment exit and entry. In contrast, growth rates calculated in the traditional
manner range between zero and infinity, and do not capture entry and exit. Thus, a clear
advantage of the measure is that it accommodates a combined treatment of establishment
births (entry), deaths (exit) and continuing establishments on employment growth. At the
same time it is monotonically related to the standard growth calculation up to a second-
order Taylor series expansion (Davis et al., 1996).
Based on (1), several measures of gross job flows may be defined. Gross job creation is
obtained by summing the jobs created by new establishments and expanding incumbents
within an industry. In turn, the gross job creation rate (GJCR), a size weighted average
growth rate of all establishments in an industry with a positive growth rate, is written as,
ijtjt ijt
i J jt
mG JC R g
M
, (2)
where ijtg is positive employment growth rate, ijtm is average establishment size and jtM
is average industry size. Similarly, gross job destruction is obtained by summing the job
losses in an industry due to the closure and contraction of establishments and the gross
job destruction rate is written as,
ijtjt ijt
i J jt
mGJDR g
M
, (3)
10
where ijtg is the absolute value of negative employment growth rates. The weights in
equations (2) and (3) reflect the size of an establishment relative to the size of the industry
to which it belongs and both establishment and industry size are expressed as the average
employment in periods t-1 and t. Net employment growth rate (NEGR) is the difference
between GJCR and GJDR while the sum of GJCR and GJDR is termed the Gross Job
Reallocation Rate (GJRR). To elaborate, the GJRR which is the sum of the total jobs
created and destroyed relative to the size of an industry captures the extent of reshuffling
of jobs across employers within an industry associated with a given net employment
growth rate.4 Finally, the excess job reallocation rate (EJRR) refers to gross job
reallocation rate that is in excess of net employment change. This is calculated as the
difference between GJRR and the absolute value of NEGR. To illustrate, since in principle,
a 5 percent NEGR can be achieved with a 5 percent GJRR (i.e., 5 percent GJCR and 0
percent GJDR), the EJRR is a measure of the depth of adjustment beyond that needed to
accommodate a certain NEGR. To sum up the relationship between the various rates may
be written as follows,
j t j t j t
j t j t j t
j t j t j t
N E G R G J C R G J D R
G J R R G J C R G J D R
E J R R G J R R N E G R
.
5. Patterns of job flows
A. Temporal patterns
Based on the preceding discussion, we begin our empirical analysis by using equation (1)
to calculate establishment level growth rates. Unweighted and size weighted frequency
distributions of these growth rates for the entire period, 1997 to 2007, and for the two sub-
periods, 1997-2001 and 2002 to 2007, are provided in Figures 2a and 2b, respectively.
These figures provide an assessment of the degree of heterogeneity in the data. Figure 2a
exhibits wide variation across manufacturing establishments in terms of employment
growth rates. The bars labeled ‘entry’ and ‘exit’ indicate that Ethiopian manufacturing has 4 GJRR also represents that part of the total movement of workers triggered by employers’ decisions to create and destroy jobs, the other part of worker flows being explained by search and match processes and movements in and out of the labor force.
11
an 18-20 percent establishment entry rate and a 14-17 percent exit rate per annum. A
large percentage of the continuing establishments, (about 36-40 percent) experience
employment growth in the neighborhood of zero (±1 percent) growth rates. Figure 2b,
depicts the same distribution weighted by establishment size. The collapse in the mass of
the distribution corresponding to entry and exit reveals that new and dying establishments
are rather small in size and together account for about 10 percent of all size-weighted
growth rate observations. The increased concentration (about 68-70 percent) around zero
(±1 percent) growth rates shows that large incumbents expand or contract rather slowly as
compared to small establishments. Such an inverse relationship between firm size and
employment growth is a widely recognized empirical regularity (Evans, 1987; Bigsten and
Gebreeyesus, 2007; Gunning & Mengistae, 2001; Van Biesebroeck, 2005).
Before discussing and considering industry-level job flows, a concern is whether
establishment level employment changes, which underlie the industry-level job flows
discussed below, represent transitory fluctuations in size or adjustments towards a desired
level of employment. Table 1 offers a one year transition probability in firm growth regimes
as a measure of persistence. It shows that about 54 percent of firms which have created
jobs this year will continue to create jobs next year (41.4 percent) or maintain their current
size (12.5 percent). Similarly, about 55 percent of establishments that shed jobs in the
current period will either continue to cut jobs in the next period (44.5 percent) or maintain
their current size (10.3 percent). These patterns suggest that most of the jobs created or
destroyed are relatively persistent reflecting changes in desired employment rather than
temporary layoffs and rehires. While the degree of persistence in our sample is far less
than that experienced in US manufacturing where Davis and Haltiwanger (1992) report
persistence rates of 67 percent for job creation and 81 percent for job destruction, the point
remains that even in the Ethiopian case the bulk of job flows may be thought of as the
result of firms trying to adjust jobs to achieve a desired employment level.
Based on equations (1), (2) and (3), Table 2 presents annual job flows for the
manufacturing sector over the period 1996-2007. Over the entire period, the average
annual rate of job creation is 17.3 percent while the average annual rate of job destruction
12
is 10.3 percent, translating into an annual net employment growth rate of 7 percent. A look
at the year-specific figures shows that every year there is substantial job creation and job
destruction and even during periods of high net employment growth, job destruction never
falls below 7 percent. This observation highlights the point that the weak aggregate
performance of manufacturing employment during 1996-2001, as depicted in Figure 1, was
not the result of a particularly low job creation rate but due to a simultaneous process of
job creation and destruction. The average GJCR during 1997-2001 was 12.4 percent and it
never fell below 10 percent at any point over that period. However, this was matched by a
job destruction rate of about 11.7 percent leading to a low net employment growth (about
0.7 percent) during 1997-2001. The strong expansion of manufacturing employment during
2002-2007 was the result of a 9 percentage point increase in GJCR relative to 1997-2001,
coupled with a modest decline (of 2.5 percentage points) in GJDR, translating into a
remarkable 12.2 percent average annual increase in employment growth between 2002
and 2007.
Turning to the remaining columns in the table, we see that over the entire sample period
the average gross job reallocation rate is very high. The rate ranges from a low of 19.7
percent in 1999 to a high of 42.9 percent in 2007 with an average of 27.6 percent. These
figures imply that more than a quarter of all manufacturing jobs were either created or
destroyed each year to accommodate the 7 percent average net employment growth rate
recorded between 1996 and 2007. This amounts to an excess job reallocation rate of
about 20 percent. The rates of excess job reallocation remain high regardless of net
employment growth. Throughout the period there is greater variation in job creation as
compared to job destruction and gross job reallocation increases from 24 percent during
the first period (1996-2001) to 30.6 percent during the upswing (i.e. 2002-2007) implying a
pro-cyclical nature of job reallocation.
How do these observations compare with job flows in other parts of the world? We provide
a comparison with the cross country evidence in Haltiwanger et al. (2008) for the 1990s.
The simultaneous occurrence of high rates of job creation and destruction in Ethiopian
manufacturing is quite similar to the patterns that have been observed in other developed
13
and emerging economies. The 17 percent GJCR in our data is, however, well above the
12.7 percent and 14.8 percent average GJCR in OECD and Latin American countries,
respectively, while being almost equal to the average job creation rate for transition
economies. In terms of job destruction, the 10 percent average for Ethiopian
manufacturing is slightly below the 12.7 percent GJDR for both the OECD and transition
economies, and the 14 percent average for Latin American countries; it is however
comparable to that of the US during the 1970s and 1980s (Davis and Haltiwanger, 1992).
In terms of gross job reallocation rates, for the entire period the figures for Ethiopian
manufacturing (28 percent) are close to the magnitudes observed in other regions. Even
though the reallocation rate accelerated during the upswing (2002-2007), it matches the
job reallocation rates observed in transition economies, a region with the highest job
reallocation rate in the Haltiwager et al. (2008) sample.
The key difference between the time-series patterns observed in Ethiopia and developed
counties is the cyclical nature of job reallocation. In developed countries such as the
United States and Canada, as shown by Baldwin et al. (1998), the temporal variance of job
destruction is higher than that of job creation which implies that gross job reallocation is
countercyclical. In the case of Ethiopia the pattern is the opposite. Throughout the period,
the variation in the rate of job creation is higher than rates of job destruction indicating that
gross job reallocation is pro-cyclical. The shift from a sluggish performance in
manufacturing employment during 1996-2001 to a strong expansion during 2002-2007 is
characterized by a sharp increase in gross job creation with a modest change in job
destruction. While the Ethiopian case is different from the United States and Canada, it is
not unique. As shown in Table 2, similar patterns are observed in transition economies,
where rates of job creation outstrip rates of job destruction. One explanation for the
different cyclical pattern of job flows in our sample as compared to the US is the incipient
nature of the manufacturing sector where the skill mix of workers is likely to be simple and
less specific to an industry or to a firm. This is more likely to be the case among small firms
that dominate the industrial landscape in developing countries. Assuming that the level and
asymmetry of adjustment costs increases with the skill level of workers, as jobs that
demand specialized skills are harder to fill, small manufacturers in countries like Ethiopia
14
may have a relatively high elasticity of job creation with respect to demand than job
destruction – resulting in a different outcome than predicted by Campbell and Fisher
(2000).
B. Job flows across industries
While the manufacturing sector as a whole exhibits high rates of job creation and
destruction, it is likely that there are variations across industries due to differences in
industry specific technologies and market structure. To investigate such variations this
section analyses job-flow rates at the two-digit industry level. Figures 3 and 4 depict
average annual rates of GJCR and GJDR, respectively, for eight two-digit industries. The
industries are sorted in ascending order of average job flows during 1997-2001 for easy
comparison across industries and over business cycles. More information on industry level
annual job flows is provided in Table A1.
As shown in Figure 3, the rapid increase in job creation during the second half of the
sample period is experienced by all industries, albeit at different rates. The food and
beverage industry represents the average job creation rate for the entire manufacturing
sector while the textile, leather and printing industries have below average job creation
rates and the chemical, non-metal, metal and wood industries record above average
performance. The ranking of industries remains essentially the same during periods of
slow and rapid change in aggregate employment, suggesting that cross industry variation
in job creation is not randomly distributed but reflects systematic differences in technology
and market structure. At the same time, there is evidence of convergence in job creation
rates during the upswing as the gain in job creation rate since 2002 has been more
pronounced in industries with below average performance.5
Figure 4 shows that all industries, except food and beverage, have experienced a
reduction in gross job destruction rate in the second sub-period. In comparison with
GJCR, there is less disparity across industries in GJDR; the standard deviations are 8
5 Until 2001, the coefficient of variation for GJCR was about 0.54. This has declined to 0.34 since 2002.
15
percent and 4.5 percent, respectively. The spread in job destruction has also narrowed
since 2002 as job destruction rates declined sharply for industries with above average
GJDR. While the ranking of industries is not very distinct particularly for industries with job
destruction rates close to the sector average, there is enough variation to suggest that
certain industries have relatively higher job destruction rates than others irrespective of
business cycles. It is worth noticing that the industries with above average job creation
rates also feature above average job destruction rates, with the exception of the chemical
industry. This implies that job losses are on average higher in industries that created more
job opportunities and net employment growth is associated with sizable readjustment of
employment positions across firms. This point is further supported by Figure 5.
Figure 5 shows an interesting aspect of job flows where net employment growth is
positively correlated with gross job reallocation rate. This suggests that fast growing
industries in Ethiopian manufacturing are characterized by massive reallocation of labor
across establishments. In fact, despite the country’s low ranking in terms of the ease of
hiring and firing of workers, the job reallocation rate in Ethiopia is comparable to that of
developed and emerging economies suggesting that labor market regulations are not
exceptionally restrictive. A likely explanation is that while on-paper labor laws are
restrictive they are not strictly adhered to due to weak law enforcement mechanisms.
Haltiwanger et al. (2008) find a strong rank correlation between industry level job flows in
OECD, Latin American and transition economies and US manufacturing industries –
meaning that industries with higher/lower job creation rates in the US also tend to have
higher/lower job creation rates in comparator countries. Looking at the rank correlation of
GJRR for the eight industries in Ethiopian manufacturing with that of the US and Canada
during 1972-1992 as reported in Baldwin et al. (1998), we find very small positive
correlation coefficients (0.14 with Canada and 0.10 with the US) which are not statistically
significant. This outcome may not be surprising given the differences in the composition
and degree of sophistication of Ethiopian manufacturing as compared to advanced and
emerging economies. We also find that the industries with better than average net
employment growth (chemical and plastic, wood and furniture, metal) in our data are not
16
the priority areas indicated in the Ethiopian government’s industrial policy which include
the food processing, textile and leather industries. The latter are given priority mainly
because they fit well with the government’s overall development strategy which is widely
known as ADLI (Agricultural Development Led Industrialization) and seeks to promote
backward linkages of manufacturing with agriculture.
C. Decomposing job flows As discussed above, similar to other parts of the world, Ethiopian manufacturing firms
simultaneously generate and destroy jobs. To characterize these patterns and to obtain
additional insights on the process of job creation and destruction, this section pushes the
analysis in several directions. In turn, we examine the link between job flows and firm
demographics, job flows and firm size, job flows and production technology, and finally the
extent to which job reallocation may be attributed to within sector reallocation as opposed
to across sectors.
To examine the link between the life-cycle of establishments and job creation and
destruction we decompose gross job creation into the fraction of jobs created by the
expansion of incumbents and by the entry of new establishments. Similarly, gross job
destruction is decomposed into jobs lost due to downsizing and bankruptcy of incumbents.
Figure 6 shows that the majority of new jobs, about 55 percent, are created by new
establishments while the expansion of incumbents accounts for the remainder. This
relative importance is not sensitive to business cycles. In terms of international
comparisons, the 55 percent contribution of firm entry to job creation in our data is higher
than the 40 percent contribution of entrants in transition economies which in turn is higher
than the 35 percent contribution in OECD countries as documented in Haltiwanger et al.
(2008). These patterns clearly suggest that in more developed economies new entrants
play a much smaller role in generating new jobs as compared to developing countries.
Since most entrants are likely to be relatively small in size, these figures also corroborate
the well recognized fact that small firms play a disproportionately larger role in gross job
creation (relative to their share in total employment). The figure shows that job destruction
17
occurs mainly through the contraction of incumbents accounting for 60 percent and 52
percent of job losses during 1997-2001 and 2002-2007, respectively. The slowdown in
sector-wide job destruction during the upswing is thus entirely due to a decline in the rate
of contraction of incumbents, as the share of job losses due to firm closure increases from
40 to about 48 percent between the two periods. It is interesting to note that both the rate
of establishment exit (Figure 2a) and its contribution to job destruction (Figure 6) went up
rather than down during the rapid aggregate expansion since 2002 pointing to the
relentless pressure of competitive market selection. Since it is likely that a number of the
exiting establishments are small producers and recent market entrants, the overall effect is
a fall in sector-wide job destruction rate during the upswing, albeit by only 2.5 percentage
points.
A more explicit examination of the link between firm size and job flows is provided in Table
3. Manufacturers that employ at least 50 workers are classified as large establishments.
Although small establishments account for about 15 percent of total employment in our
data, the table shows that they contribute to one-third of new jobs both during periods of
slow and rapid aggregate employment growth. Small producers contribute to job creation
mainly at the point of market entry (22.5 percent), about twice their contribution through
post-entry expansion (10.2 percent). Large establishments account for the remaining two-
thirds of new jobs. The relative importance of entry and expansion for job creation among
large establishments is nearly symmetric with entry having a slight edge. The patterns in
terms of job destruction are similar with small firms accounting for about a third of job loss.
The figures for Ethiopia match the patterns observed in other settings. Haltiwanger et al.
(2008) show that large firms that employ at least 50 workers account for about 60-70
percent of total job creation and destruction in a number of countries in the OECD (Italy,
Portugal, France, UK, US), in Latin American (Argentina, Chile, Columbia and Mexico) as
well as in emerging economies (Latvia, Slovenia and Hungary).
Figure 6 has shown that the decline in GJDR is due to a slowdown in the degree of
contraction of continuing establishments rather than a reduction in the rate of exit. The
lower panel of Table 3 reveals that the slowdown in job losses due to establishment
18
contraction is evident only among large producers where its contribution has dropped from
47 percent of all job losses during 1997-2001 to 41.4 percent during 2002-2007. At the
same time, job losses by small producers have gone up from about 27 percent of total job
losses during 1997-2001 to nearly one-third during 2002-2007, an increase both in terms
of contraction as well as exit. The expansion of manufacturing employment in the second
sub-period is therefore accompanied by a reallocation of labor from small to large
establishments. Since large firms typically offer better working conditions than small firms,
other things being equal, the upswing may have led to a tight labor market and made it
difficult for small firms to retain their trained and experienced workers.6
Heterogeneity in job flows across establishments may partly be traced to the choice of
production technology, an important aspect of which is the choice of input proportions.
Accordingly, Table 4 presents the results of a decomposition of job flows conditional on
capital intensity. Capital intensity is defined in terms of the capital-labor ratio and
establishments with above average capital-labor ratio are treated as capital intensive. The
analysis shows that capital intensive establishments account for nearly 60 percent of job
creation and that the rise in gross job creation during 2002-2007 was driven mainly by the
expansion of capital intensive establishments. While the latter created most of the new
jobs, they also account for the bulk of job destruction mainly through contraction. The
increase in net employment growth since 2002 is therefore the result of a higher rate of job
creation among capital intensive establishments – through expansion, coupled with a lower
rate of job destruction among labor intensive establishments. However, since the reduction
in overall GJDR during 2002-2007 (2 percentage points) is much less than the gain in
GJCR (9 percentage points), there has been a reallocation of jobs in favor of more capital
intensive firms in the Ethiopian manufacturing sector.
The sizable excess reallocation of jobs raises the issue as to whether excess churning is
mainly due to an intra- or inter-industry reallocation of jobs. In the latter case jobs will be
6 The average wage rate among small Ethiopian firms, i.e., total wage bill as a ratio of the number of employees, is
about 50% of the average wage rate among large firms. This figure does not take into account differences in the
human capital of the work force of large and small firms and hence is likely to overestimate the wage gap.
19
reallocated from shrinking to expanding industries while in the former the adjustment is
within industries. A formulation to decompose excess job reallocation into these
constituents, as suggested by Davis and Haltiwager (1992) is,
St St St jt jt St jt St Stj S j S
GJRR NEGR M GJRR NEGR M NEGR NEGR M
, (4)
where S stands for the manufacturing sector, M is average size and, while j and t index
industry and time. The left hand side of equation (4) represents the volume of excess job
reallocation for the manufacturing sector expressed as the product of average
manufacturing sector employment and the excess job reallocation rate (the difference
between gross job reallocation rate and the absolute value of net employment growth
rate). The first term on the right hand side captures intra-industry excess job reallocation
measured as the sum over all industries of the product of excess job reallocation rate and
average size of the manufacturing sector at time t. The second term on the right hand side
is the proportion of excess job reallocation which may be attributed to inter-industry
reallocation of jobs and is the sum of the weighted product of the deviation of absolute net
employment growth for the industry level from the absolute net employment growth rate for
the manufacturing sector as a whole.
On average, over time, we find that 86 percent of excess job reallocation takes place
within industries. Overwhelmingly, excess job reallocation is an intra-industry phenomenon
reflecting the reshuffling of jobs across establishments producing broadly similar products.
There is some variation over time with inter-industry reallocation of jobs accounting for
about 20 percent of excess job reallocation during 1997-2001 which falls to 10 percent
over the period 2002-2007.7 The dominant role of intra-industry movement is consistent
with the patterns found in US manufacturing (Davis and Halitwanger, 1992) and in
transition economies (Faggio and Konings, 2003) where the share of between-industry
movements is even less than in the Ethiopian case.
7 This is mainly the result of a sharp decline in the employment share of the textile sector during the late 1990s, a decline which has abated since 2004. The textile industry is dominated by public enterprises and was until recently the single most important employer in the manufacturing sector.
20
6. Econometric analysis of job flows
The preceding section has shown that job flows vary over time and across groups of
establishments defined in terms of industries, and are sensitive to firm age, size and
capital intensity. The objective of this section is to consolidate the analysis and to assess
the relative importance of these sources of variation by simultaneously analyzing their
effects on industry-level job flows. We estimate a set of econometric models with gross job
reallocation as a dependent variable and time varying industry level characteristics, and
industry and time fixed effects as regressors. The time varying covariates include average
age of industries in a firm, the share of small establishments (less than 50 employees) to
capture size effects, and capital intensity. While this section focuses on job reallocation,
since it encompasses job creation and destruction, as discussed below we also estimate
models with job creation, job destruction and net job growth as dependent variables.
The choice of covariates is motivated by theory and existing empirical evidence. Models of
passive learning in the industrial evolution literature suggest that as compared to young
and small firms, older and larger firms grow at a slower pace as they are more likely to
have approached the scale of operation dictated by their innate relative efficiency
(Jovanovic, 1982; Lippman and Rumelt, 1982). This is essentially the result of market
selection based on time invariant initial conditions generating simultaneous job creation
and destruction within narrowly defined industries, an effect that is expected to taper off as
the industry matures. One would therefore expect a negative firm size and age effect in a
model of job reallocation. Theories of active learning however suggest that firms can
change their fate by engaging in productivity enhancing activities. In this case the basis for
selection and job reallocation is the degree of success in productivity enhancing activities,
i.e., establishments that succeed in improving productivity expand while establishments
that fail to do so lose jobs (Ericson and Pakes, 1995; Pakes and Ericson, 1998). We intend
to capture the latter effect using the capital-labor ratio.8 The model has the following
structure:
8 Given the interest in the literature about the cyclical nature of job reallocation, we include net employment growth
as an additional regressor in one of the specifications to capture expansion or contraction at the industry level.
21
1j t j t j t
j t j t
G J R R X u
u j t e
, (5)
where jtGJRR is gross job reallocation rate in industry j at time t, 1jtX stands for industry
level covariates lagged by one period, and jtu is a composite error term with industry and
time fixed effects as well as a time-varying error term ( jte ).
To isolate the sources of variation in GJRR, we start the analysis by controlling only for
industry fixed effects, this is followed by the inclusion of time fixed effects and finally the
model is expanded to include the time-varying industry specific factors discussed above.
We estimate these specifications using OLS. Since industry level job flows tend to be
persistent, we also estimate (5) using a feasible generalized least squares (FGLS)
technique which provides efficient estimates in the presence of autocorrelated and
heteroscedastic errors ( jte ). All variables enter the estimation models with a one period
lag as the contemporaneous values will obviously be influenced by current job flows.
Estimates of these various specifications are presented in Table 5.
The first column of Table 5 shows that industry specific effects play a large role (57
percent) in accounting for variations in gross job reallocation. As shown in column 2, the
inclusion of time fixed effects raises the proportion of the explained variation to 79 percent.
Consistent with the descriptive statistics presented earlier (Figures 3, 4 and 5), industries
like metal and, wood and furniture feature very high rates of job reallocation relative to the
food industry (omitted industry) which represents the average job reallocation rate for the
manufacturing sector.
Column 3 displays estimates which include the time varying covariates. The inclusion of
these variables leads to a small (4 percentage point) increase in the proportion of
explained variation in GJRR. Turning to the individual effects, the results indicate that job
reallocation increases with the fraction of small firms in an industry confirming the patterns
observed in the descriptive analysis. This result is also consistent with models of passive
learning. While the linear and quadratic terms of age have the expected signs, where job
22
reallocation decreases with age in a non-linear fashion, they are statistically insignificant.
Firms in capital intensive industries tend to have higher rates of labor adjustment although
the size of the coefficient is small. In column 4 we take into account heteroscedasticity and
panel specific autocorrelation using the FGLS estimator. These results show that the OLS
estimates are largely immune to potential problems which may be associated with the error
term. Finally, in the spirit of establishing the story revealed in Figure 5 we include NEGR
as a regressor to capture the correlation between job reallocation and net employment
expansion. The positive and statistically significant link between NEGR and GJRR
underscores the previous finding that, in our data, job reallocation is pro-cyclical and that
fast growing industries in Ethiopia manufacturing are characterized by high rates of intra-
industry labor reallocation. While the results in this section confirm a number of the
bivariate patterns identified earlier, the main insight which emerges is that similar to the
patterns observed in other countries (for example, see Baldwin, 1998) industry specific
factors account for the bulk of cross- industry differences in job reallocation rates. While
we do not explore the underlying sources of these industry-specific differences these
probably include differences in technology, the skill-mix required in such industries, the
costs associated with creating a new job, and demand conditions.
To complete the analysis, Tables A2 to A4 provide estimates of job creation, job
destruction and net employment growth. Similar to the patterns observed for job
reallocation, Tables A2 and A3 show that industry and time fixed effects explain large
shares of the variation in job creation (industry effect - 44 percent and time effect - 30
percent) and job destruction (industry – 32 percent and time effect – 31 percent). Thus,
industry-specific technology and market structure is relatively less important in explaining
job destruction than job creation, a feature already indicated in Figures 3 & 4. With regard
to net job growth the estimates show that there is very little cross-industry variation and
only 12 percent of the variation in job growth is explained by industry fixed effects (Table
A4). This reaffirms the previous observation that industries with high job creation rates also
tend to have high job destruction rates. As shown in Figures 3 and 4 all industries have
enjoyed an increase in job creation in the second sub-period while all except the food
industry have experienced a reduction in the job destruction rate. Since there has been a
convergence among industries in terms of job creation and job destruction rates during the
23
upswing, the overall difference in net employment changes across industries tends to be
lower. Most of the variation in net employment growth is therefore associated with
business cycles as time fixed effects explain over 40 percent of the variation.9
Turning to other regressors, according to the FGLS estimates, job creation increases with
the share of small firms in an industry. Similarly, job destruction decreases with the
dominance of large firms in an industry as most of the ‘creative destruction’ seems to take
place at the lower end of the firm size distribution. Capital intensity has a statistically
significant but weak positive association with job destruction. It is notable that both the
fraction of small firms in an industry and capital intensity do not exert a significant effect on
net employment change although they exert an influence on job creation, job destruction
and job reallocation rates. This raises an important question about the relative importance
of ‘creative destruction’ and resource reallocation for overall employment and economic
growth. While most of the firm churning and job reallocation takes place at the lower tail of
the firm size distribution, industries dominated by small firms do not seem to show a
significantly higher net employment growth rate.
7. Conclusion
This paper represents the first attempt at a detailed analysis of gross job flows in Sub-
Saharan Africa using establishment level data. Key findings include the existence of high
rates of simultaneous job creation and destruction behind the seemingly unimpressive
contribution of manufacturing to overall employment in the region. Furthermore, the rates
and patterns of job flows reported in this paper are very similar to the findings from studies
based on developed and emerging economies. As in other countries we find that job
reallocation is a predominantly intra-industry phenomenon and is relatively higher in
industries dominated by smaller and younger establishments. Somewhat different from the
existing work is that while there is strong rank correlation between industry level job
creation and job destruction rates between the US and other developed and emerging
economies, such a correlation does not exist between US and Ethiopian manufacturing. 9 The importance of industry fixed effects in explaining job creation, destruction and reallocation while playing a
minimal role in explaining net job growth is similar to patterns observed in Canada and the US (Baldwin et al. 1998).
24
Notwithstanding this difference, the upshot of the empirical work reported here is that
African firms adjust their work force in a manner broadly similar to firms elsewhere and
African labor markets are not uniquely restrictive in terms of undermining job reallocation
across firms. While in general, labor laws that make it difficult to hire and fire workers may
be expected to reduce labor market flexibility, there is little evidence that such restrictions
have a bearing on job flows in the current case. This might however be the result of
inadequate law enforcement rather than a reflection of labor market reforms as seems to
be the case in Ethiopia.
Some of the other findings are also quite informative. The growth of manufacturing
employment in Ethiopia during the second half of the study period was driven mainly by
new establishments joining the sector - a degree of contribution well above that of new
firms in transition economies which in turn is higher than that observed in OECD countries.
Thus, while there is no shortage of business startups which create jobs at the moment of
entry, such small establishments contribute less to job creation through subsequent
expansion. A similar observation has been made by other firm level studies in Africa which
show a lack of graduation of small firms into medium and large size categories. While it is
true that small firms grow faster than large firms, as documented in a number of firm level
studies, the growth rate does not seem to be strong enough to catapult them into a
different size category at least in the African context.
The similarity in the micro-dynamics of employment generation in an African setting as
compared to other countries suggests that poor employment generation in African
manufacturing may not be attributed to any peculiar features of African labor markets. This
is similar to the finding from recent firm-level studies (Frazer, 2005; Shiferaw, 2007) which
show that African manufacturing firms are subjected to the same market scrutiny and
competition as firms in other parts of the worlds. Thus, answers to the inability of African
firms in terms of creating and sustaining jobs probably lies in the broader political and
economic risk that they face rather than labor market regulations and restrictions.
25
References
Baldwin, J., T. Dunne, and J. Haltiwanger. 1998. “A Comparison of Job Creation and
Job Destruction in Canada and the United States.” Review of Economics and
Statistics 80, 3, 347-356.
Bigsten, A., and M. Gebreeyesus. 2007. “The Small, the Young, and the Productive:
Determinants of Manufacturing Firm Growth in Ethiopia.” Economic Development
and Cultural Change 55, 4, 813-840.
Bigsten, A., P. Collier, S. Dercon, M. Fafchamps, B. Gauthier, J.W. Gunning, R.
Oostendorp. C. Pattillo, M. Soderbom, and F. Teal. 2005. “Adjustment Costs and
Irreversibility as Determinants of Investment: Evidence from African Manufacturing,”
The B.E. Journals in Economic Analysis and Policy 4, 1, 1-29.
Bigsten, A., P. Collier, S. Dercon, M. Fafchamps, B. Gauthier, J.W. Gunning, A.
Isaksson, A. Oduro, R. Oostendorp. C. Pattillo, M. Soderbom, M. Sylvain, F. Teal,
and A. Zeufack. 1999. “Investment in Africa’s Manufacturing Sector: A Four Country
Panel Data Analysis,” Oxford Bulletin of Economics and Statistics 61, 4, 489-512.
Campbell, J., and J. Fisher. 2000. “Aggregate Employment Fluctuations with
Microeconomic Asymmetries.” American Economic Review 90, 5, 1323-1345.
Collier, P., and J. W. Gunning, 1999. “Explaining African Economic Performance”,
Journal of Economic Literature 37, 1, 64-111.
Davis, S., and J. Haltiwanger. 1990. “Gross Job Creation and Destruction:
Microeconomic Evidence and Macroeconomic Implications.” NBER Macroeconomic
Annuals 5, 123-168.
Davis, S., and J. Haltiwanger. 1992. “Gross Job Creation, Gross Job Destruction, and
Employment Reallocation.” The Quarterly Journal of Economics 107, 3, 819-863.
Davis, S., J. Haltiwanger, and S. Schuh. 1996. Job Creation and Destruction. The MIT
Press, Cambridge, Massachusetts
Denu, B., A. Tekeste and H van der Deijl. 2005. “Characteristics and
Determinants of youth unemployment, underemployment and inadequate
employment in Ethiopia.” Employment Strategy Papers No. 7, ILO, Geneva.
Ericson, R. and A. Pakes. 1995. “Markov-Perfect Industry Dynamics: A Framework for
Empirical Work,” Review of Economic Studies 62, 1, 53-82.
26
Evans, S.D. 1987. “The Relationship Between Firm Growth, Size and Age: Estimates
for 100 Manufacturing Industries.” Journal of Industrial Economics 35, 4, 567-81.
Faggio, G., and J. Konings. 2003. “Job Creation, Job Destruction, and Employment
Growth in Transition Countries in the 90s.” Economic Systems 27, 2, 129-154.
Frazer, G. 2005. “Which Firms Die? A Look at Manufacturing Firm Exit in Ghana.”
Economic Development and Cultural Change 53, 3, 585-617.
Gunning, J.W., and T. Mengistae. 2001. “Determinants of African Manufacturing
Investment: The Microeconomic Evidence,” Journal of African Economies 10, 2, 48-
80.
Haltiwanger, J., S. Scarpetta, and H. Schweiger. 2008. “Assessing Job Flows Across
Countries: The Role of Industry, Firm Size and Regulations.” NBER Working Paper
No. 13920.
ILO.2009. Database on Labor Statistics: www.laborstat.ilo.org
IMF.2008. “The Federal Democratic Republic of Ethiopia: Statistical Appendix”
International Monetary Fund Country Report No. 08/260.
Jovanovic, B. 1982. “Selection and the Evolution of Industry.” Econometrica 50, 3, 649-
70.
Lippman, S., and R. Rumelt. 1982. “Uncertain Imitability: An Analysis of Interfirm
Differences Under Competition.” Bell Journal of Economics, 13, 2, 418-438.
Pakes, A. and R. Ericson.1998. “Empirical Implications of Alternative Models of Firm
Dynamics,” Journal of Economic Theory 79, 1, 1-46.
Shiferaw, A. 2007. “Firm Heterogeneity and Market Selection in Sub-Saharan Africa:
Does it Spur Industrial Progress?” Economic Development and Cultural Change
55, 2, 393-423.
Shiferaw, A. 2009. “The Survival of Private Sector Manufacturing Establishments in
Africa: The Role of Productivity and Ownership?” World Development 37, 3, 572-
584.
Söderbom, M., F. Teal, and A. Harding. 2006. “The Determinants of Survival Among
African manufacturing Firms,” Economic Development and Cultural Change 54, 3,
533-555.
Tybout, J. 2000. “Manufacturing Firms in Developing Countries: How Well Do They Do,
27
and Why?” Journal of Economic Literature 38, 1, 11-44.
Van Biesebroeck, J. 2005. “Firm Size Matters: Growth and Productivity Growth in
African Manufacturing.” Economic Development and Cultural Change 53, 3, 545-
583
28
Table 1: One period transition probabilities in firm growth regimes (%) Growth Regimes
Positive Zero Negative
Positive 41.4 12.5 46.1
Zero 38.9 28.9 32.2
Negative 45.2 10.3 44.5
Source: Authors’ calculation based on CSA’s manufacturing census. Figures exclude entry and exit.
Table 2: Job Flows in Ethiopian Manufacturing GJCR GJDR NEGR GJRR EJRR
1997 0.1342 0.1861 ‐0.0519 0.3202 0.2683
1998 0.1248 0.1049 0.0199 0.2297 0.2098
1999 0.1042 0.0929 0.0112 0.1971 0.1859
2000 0.1335 0.1019 0.0316 0.2354 0.2038
2001 0.1230 0.0973 0.0256 0.2203 0.1946
2002 0.2238 0.0849 0.1389 0.3088 0.1698
2003 0.1306 0.0834 0.0473 0.2140 0.1667
2004 0.1463 0.0736 0.0727 0.2199 0.1472
2005 0.1966 0.0882 0.1085 0.2848 0.1763
2006 0.2980 0.0789 0.2192 0.3769 0.1577
2007 0.2871 0.1418 0.1453 0.4289 0.2836
Standard Deviation 0.1146 0.0827 0.1213 0.1589 0.1323
Period Averages
1997‐2001 0.1239 0.1166 0.0073 0.2406 0.2125
2002‐2007 0.2138 0.0918 0.1220 0.3055 0.1836
1997‐2007 0.1729 0.1031 0.0698 0.2760 0.1967
Other Regions (1990s)
OECD 0.127 0.127 0.000 0.254 0.223
Latin America 0.148 0.140 0.008 0.288 0.248
Transition Economies 0.174 0.128 0.046 0.303 0.227
Source: Authors’ calculations based on CSA’s manufacturing census data for Ethiopia and Table 3 of Haltiwanger et al. (2008) for other regions.
29
Table 3: Decomposition of Job Creation and Destruction by Firm Size Small Firms Large Firms
Job Creation Expansion Entry Total Expansion Entry Total
1997‐2001 0.0980 0.2353 0.3332 0.2773 0.3894 0.6668
2002‐2007 0.1052 0.2157 0.3209 0.3484 0.3306 0.6791
1997‐2007 0.1019 0.2246 0.3265 0.3161 0.3574 0.6735
Job Destruction Contraction Exit Total Contraction Exit Total
1997‐2001 0.0941 0.1747 0.2688 0.4701 0.2611 0.7312
2002‐2007 0.1097 0.2140 0.3237 0.4144 0.2619 0.6763
1997‐2007 0.1026 0.1961 0.2987 0.4398 0.2615 0.7013
Source: Authors’ computations based on CSA’s manufacturing census
Note: Job creation and destruction rates in the two size groups add up to 1.
Table 4: Decomposition of Job Creation by Factor Intensity of Firms Labor Intensive Capital Intensive
Job Creation
Expansion Entry Total Expansion Entry Total
1997‐2001 0.2558 0.1780 0.4339 0.1882 0.3689 0.5571
2002‐2007 0.1676 0.2239 0.3915 0.2855 0.3120 0.5976
1997‐2007 0.2077 0.2030 0.4107 0.2413 0.3379 0.5792
Job Destruction
Contraction Exit Total Contraction Exit Total
1997‐2001 0.2489 0.2376 0.4865 0.3436 0.1608 0.5045
2002‐2007 0.1926 0.1985 0.3912 0.3267 0.2686 0.5953
1997‐2007 0.2182 0.2163 0.4345 0.3344 0.2196 0.5540
Source: Authors’ computations based on CSA’s manufacturing census
Note: Job creation and destruction rates in the two groups add up to 1.
30
Table 5: Gross Job Reallocation OLS
(1) OLS (2)
OLS (3)
FGLS (4)
FGLS (5)
Textile & Garments ‐0.0992**(0.0466)
‐0.0992***(0.0345)
‐0.0342(0.0787)
0.0040(0.0595)
0.0552 (0.0476)
Leather & Footwear ‐0.1058**(0.0466)
‐0.1058***(0.0345)
‐0.0929**(0.0455)
‐0.0801*** (0.0279)
‐0.0616*** (0.0219)
Printing & Paper ‐0.0901*(0.0466)
‐0.0901**(0.0345)
‐0.0902**(0.0378)
‐0.0888*** (0.0254)
‐0.0751*** (0.0200)
Chemical & Plastic ‐0.0320 (0.0466)
‐0.0320(0.0345)
0.0349(0.0663)
0.0492(0.0456)
0.0303 (0.0381)
Non‐metal 0.0324 (0.0466)
0.0324(0.0345)
‐0.0031(0.0434)
‐0.0171(0.0308)
‐0.0376 (0.0248)
Metal 0.1191**(0.0466)
0.1191***(0.0345)
0.0875**(0.0403)
0.0790** (0.0335)
0.0508* (0.0304)
Wood & Furniture 0.2617***(0.0466)
0.2617***(0.0345)
0.1953***(0.0527)
0.1746*** (0.0387)
0.1375*** (0.0336)
Time Dummies No Yes Yes Yes Yes
Aget‐1 ‐0.0217(0.0421)
‐0.0161(0.0324)
‐0.0208 (0.0270)
Age2t‐1 0.0009(0.0012)
0.0007(0.0009)
0.0007 (0.0008)
Small Firmst‐1 0.4917*(0.2932)
0.5775*** (0.2139)
0.5766*** (0.1720)
Capital Intensityt‐1 0.0009**(0.0004)
0.0008*** (0.0002)
0.0008*** (0.0002)
NEGR 0.4633*** (0.0688)
Constant 0.3000***(0.0329)
0.4063***(0.0366)
0.0206(0.4671)
‐0.0746(0.3523)
‐0.0389 (0.2963)
Observations 88 88 80 80 80 R‐squared 0.57 0.79 0.83
Source: Authors’ computations based on CSA’s manufacturing census
Notes: Standard errors in parentheses. *, ** and *** represent significance at 10%, 5% and 1%,
respectively.
31
Figure 1: Manufacturing Employment and Sales in Ethiopia
Note: Sales values are on the right‐hand‐side y‐axis in millions of Ethiopian Birr and employment is in terms of the number of employees in the manufacturing sector.
32
Exit
Entry
0.05
.1.15
.2
Fra
ction
-2-1.5-1 -.5 0 .5 1 1.5 2
1997-2001
Exit
Entry
0.05
.1.15
.2
Fra
ction
-2-1.5-1 -.5 0 .5 1 1.5 2
2002-2007
Exit
Entry
0.05
.1.15
.2
Fra
ction
-2-1.5-1 -.5 0 .5 1 1.5 2
1997-2007
Figure 2a: Distribution of Unweighted Establishment Level Employment Growth Rates
33
ExitEntry
0.1
.2.3
.4
Fra
ction
-2-1.5-1 -.5 0 .5 1 1.5 2
1997-2001
ExitEntry
0.1
.2.3
Fra
ction
-2-1.5-1 -.5 0 .5 1 1.5 2
2002-2007
ExitEntry
0.1
.2.3
Fra
ction
-2-1.5-1 -.5 0 .5 1 1.5 2
1997-2007
Figure 2b: Distribution of Size Weighted Establishment Level Employment Growth Rates
34
Figure 3: Gross Job Creation Rate by Industry
Figure 4: Gross Job Destruction Rate by Industry
35
Figure 5: Net Employment Growth and Gross Job Reallocation (1997‐2007)
Figure 6: Decomposition of Gross Job Creation and Destruction Rates
36
Table A1: Job Flows by Industry
year Flows Food &
Beverage
Textile &
Garments
Leather &
Footwear
Wood &
Furniture
Printing &
Paper
Chemical &
PlasticNon‐Metal Metal
1997 GJCR 0.1797 0.0413 0.0702 0.4238 0.0661 0.2890 0.1581 0.2381
GJDR 0.1271 0.1632 0.1124 0.5722 0.2602 0.0720 0.2589 0.3043
GJRR 0.3068 0.2045 0.1826 0.9960 0.3263 0.3610 0.4170 0.5424
NEGR 0.0526 ‐0.1219 ‐0.0422 ‐0.1484 ‐0.1941 0.2170 ‐0.1008 ‐0.0662
EJRR 0.2542 0.0826 0.1404 0.8476 0.1322 0.1440 0.3162 0.4762
1998 GJCR 0.1066 0.0541 0.0717 0.3866 0.1387 0.2401 0.2481 0.1605
GJDR 0.0739 0.1157 0.0790 0.2910 0.0292 0.0586 0.1079 0.1669
GJRR 0.1805 0.1698 0.1507 0.6776 0.1679 0.2987 0.3560 0.3274
NEGR 0.0327 ‐0.0616 ‐0.0073 0.0956 0.1095 0.1815 0.1402 ‐0.0064
EJRR 0.1478 0.1082 0.1434 0.5820 0.0584 0.1172 0.2158 0.3210
1999 GJCR 0.1844 0.0549 0.0635 0.1939 0.0628 0.0618 0.0921 0.1499
GJDR 0.1478 0.0570 0.0715 0.1199 0.0751 0.0915 0.1017 0.0948
GJRR 0.3322 0.1119 0.1350 0.3138 0.1379 0.1533 0.1938 0.2447
NEGR 0.0366 ‐0.0021 ‐0.0080 0.0740 ‐0.0123 ‐0.0297 ‐0.0096 0.0551
EJRR 0.2956 0.1098 0.1270 0.2398 0.1256 0.1236 0.1842 0.1896
2000 GJCR 0.1663 0.0510 0.0402 0.2797 0.1220 0.1137 0.2567 0.3337
GJDR 0.0724 0.1630 0.0612 0.1220 0.0639 0.0525 0.0711 0.0777
GJRR 0.2387 0.2140 0.1014 0.4017 0.1859 0.1662 0.3278 0.4114
NEGR 0.0939 ‐0.1120 ‐0.0210 0.1577 0.0581 0.0612 0.1856 0.2560
EJRR 0.1448 0.1020 0.0804 0.2440 0.1278 0.1050 0.1422 0.1554
2001 GJCR 0.1214 0.1068 0.0850 0.2668 0.0543 0.1528 0.0924 0.2031
GJDR 0.0865 0.0534 0.0776 0.2283 0.0812 0.0708 0.1378 0.3333
GJRR 0.2079 0.1602 0.1626 0.4951 0.1355 0.2236 0.2302 0.5364
NEGR 0.0349 0.0534 0.0074 0.0385 ‐0.0269 0.0820 ‐0.0454 ‐0.1302
EJRR 0.1730 0.1068 0.1552 0.4566 0.1086 0.1416 0.1848 0.4062
2002 GJCR 0.2492 0.0770 0.2245 0.4919 0.3153 0.3214 0.1848 0.3855
GJDR 0.1228 0.0606 0.1098 0.1524 0.0285 0.0503 0.0803 0.0710
GJRR 0.3720 0.1376 0.3343 0.6443 0.3438 0.3717 0.2651 0.4565
NEGR 0.1264 0.0164 0.1147 0.3395 0.2868 0.2711 0.1045 0.3145
37
EJRR 0.2456 0.1212 0.2196 0.3048 0.0570 0.1006 0.1606 0.1420
2003 GJCR 0.1506 0.0479 0.1260 0.3127 0.0787 0.1295 0.2287 0.2170
GJDR 0.1305 0.0414 0.0601 0.1525 0.0736 0.0619 0.0939 0.0864
GJRR 0.2811 0.0893 0.1861 0.4652 0.1523 0.1914 0.3226 0.3034
NEGR 0.0201 0.0065 0.0659 0.1602 0.0051 0.0676 0.1348 0.1306
EJRR 0.2610 0.0828 0.1202 0.3050 0.1472 0.1238 0.1878 0.1728
2004 GJCR 0.1574 0.0970 0.1220 0.3085 0.0807 0.1266 0.2419 0.1944
GJDR 0.0855 0.0775 0.0267 0.1340 0.0260 0.0423 0.0984 0.0752
GJRR 0.2429 0.1745 0.1487 0.4425 0.1067 0.1689 0.3403 0.2696
NEGR 0.0719 0.0195 0.0953 0.1745 0.0547 0.0843 0.1435 0.1192
EJRR 0.1710 0.1550 0.0534 0.2680 0.0520 0.0846 0.1968 0.1504
2005 GJCR 0.1477 0.1545 0.2085 0.1935 0.1763 0.2986 0.2406 0.3067
GJDR 0.1051 0.0924 0.0346 0.1999 0.0471 0.0544 0.1164 0.0720
GJRR 0.2528 0.2469 0.2431 0.3934 0.2234 0.3530 0.3570 0.3787
NEGR 0.0426 0.0621 0.1739 ‐0.0064 0.1292 0.2442 0.1242 0.2347
EJRR 0.2102 0.1848 0.0692 0.3870 0.0942 0.1088 0.2328 0.1440
2006 GJCR 0.3157 0.2570 0.1440 0.6382 0.2005 0.1974 0.3028 0.4504
GJDR 0.0829 0.0730 0.0731 0.0990 0.0523 0.0746 0.0900 0.0914
GJRR 0.3986 0.3300 0.2171 0.7372 0.2528 0.2720 0.3928 0.5418
NEGR 0.2328 0.1840 0.0709 0.5392 0.1482 0.1228 0.2128 0.3590
EJRR 0.1658 0.1460 0.1462 0.1980 0.1046 0.1492 0.1800 0.1828
2007 GJCR 0.3440 0.2532 0.1831 0.3267 0.1375 0.3095 0.3399 0.3040
GJDR 0.1429 0.1176 0.0915 0.2861 0.1389 0.0788 0.1146 0.2938
GJRR 0.4869 0.3708 0.2746 0.6128 0.2764 0.3883 0.4545 0.5978
NEGR 0.2011 0.1356 0.0916 0.0406 ‐0.0014 0.2307 0.2253 0.0102
EJRR 0.2858 0.2352 0.1830 0.5722 0.2750 0.1576 0.2292 0.5876
Source: Authors’ computation based on CSA’s manufacturing census for Ethiopia
38
Table A2: Gross Job Creation
OLS OLS OLS FGLS
(1) (2) (3) (4)
Textile & Garment ‐0.0844**(0.0381)
‐0.0844***(0.0279)
‐0.1242(0.0774)
‐0.0694 (0.0495)
Leather & Footwear ‐0.0713*(0.0381)
‐0.0713**(0.0279)
‐0.0773*(0.0448)
‐0.0502* (0.0267)
Printing & paper ‐0.0627(0.0381)
‐0.0627**(0.0279)
‐0.0722*(0.0372)
‐0.0713*** (0.0236)
Chemical & Plastic 0.0107(0.0381)
0.0107(0.0279)
0.0188(0.0653)
0.0595 (0.0407)
Non‐Metal 0.0239(0.0381)
0.0239(0.0279)
0.0339(0.0427)
0.0061 (0.0287)
Metal 0.0746*(0.0381)
0.0746***(0.0279)
0.0922**(0.0397)
0.0857*** (0.0239)
Wood & Furniture 0.1545***(0.0381)
0.1545***(0.0279)
0.1490***(0.0519)
0.1161*** (0.0332)
Time Dummies No YES YES YES
Aget‐1 0.0060(0.0414)
0.0088 (0.0266)
Age2t‐1 0.0002(0.0012)
0.0001 (0.0007)
Small Firmst‐1 0.1094(0.2885)
0.3462* (0.1907)
Capital Intensityt‐1 0.0003(0.0004)
0.0003 (0.0002)
Constant 0.1930***(0.0269)
0.1776***(0.0296)
‐0.0368(0.4595)
‐0.2441 (0.3044)
Observations 88 88 80 80
R‐squared 0.44 0.74 0.74
Source: Authors’ computations based on CSA’s manufacturing census
Notes: Standard errors in parentheses. *, ** and *** represent significance at 10%, 5%
and 1%, respectively.
39
Table A3: Gross Job Destruction
OLS OLS OLS FGLS
(1) (2) (3) (4)
Textile & Garment ‐0.0148(0.0303)
‐0.0148(0.0239)
0.0900*(0.0470)
0.0608** (0.0249)
Leather & Footwear ‐0.0345(0.0303)
‐0.0345(0.0239)
‐0.0156(0.0272)
‐0.0184 (0.0133)
Printing & paper ‐0.0274(0.0303)
‐0.0274(0.0239)
‐0.0180(0.0225)
‐0.0315*** (0.0112)
Chemical & Plastic ‐0.0427(0.0303)
‐0.0427*(0.0239)
0.0162(0.0396)
0.0082 (0.0218)
Non‐Metal 0.0085(0.0303)
0.0085(0.0239)
‐0.0370(0.0259)
‐0.0298** (0.0130)
Metal 0.0445(0.0303)
0.0445*(0.0239)
‐0.0047(0.0241)
0.0072 (0.0180)
Wood & Furniture 0.1073***(0.0303)
0.1073***(0.0239)
0.0463(0.0315)
0.0439** (0.0186)
Time Dummies NO YES YES YES
Aget‐1 ‐0.0276(0.0251)
‐0.0088 (0.0141)
Age2t‐1 0.0007(0.0007)
0.0003 (0.0004)
Small Firmst‐1 0.3823**(0.1751)
0.3069*** (0.0976)
Capital Intensityt‐1 0.0006**(0.0002)
0.0004*** (0.0001)
Constant 0.1070***(0.0214)
0.2287***(0.0253)
0.0574(0.2788)
‐0.0591 (0.1603)
Observations 88 88 80 80
R‐squared 0.32 0.63 0.66
Source: Authors’ computations based on CSA’s manufacturing census
Notes: Standard errors in parentheses. *, ** and *** represent significance at 10%, 5%
and 1%, respectively.
40
Table A4: Net Employment Growth
OLS OLS OLS FGLS
(1) (2) (3) (4)
Textile & Garment ‐0.0696(0.0506)
‐0.0696*(0.0388)
‐0.2141**(0.1010)
‐0.1158** (0.0526)
Leather & Footwear ‐0.0368(0.0506)
‐0.0368(0.0388)
‐0.0616(0.0584)
‐0.0273 (0.0316)
Printing & paper ‐0.0353(0.0506)
‐0.0353(0.0388)
‐0.0541(0.0485)
‐0.0377 (0.0276)
Chemical & Plastic 0.0534(0.0506)
0.0534(0.0388)
0.0026(0.0852)
0.0586 (0.0451)
Non‐Metal 0.0154(0.0506)
0.0154(0.0388)
0.0708(0.0558)
0.0297 (0.0314)
Metal 0.0301(0.0506)
0.0301(0.0388)
0.0968*(0.0518)
0.0798*** (0.0298)
Wood & Furniture 0.0472 (0.0506)
0.0472 (0.0388)
0.1026 (0.0677)
0.0715* (0.0367)
Time Dummies NO YES YES YES
Aget‐1 0.0336(0.0541)
0.0158 (0.0288)
Age2t‐1 ‐0.0006(0.0015)
‐0.0001 (0.0007)
Small Firmst‐1 ‐0.2729(0.3765)
0.0871 (0.2065)
Capital Intensityt‐1 ‐0.0004(0.0005)
‐0.0001 (0.0003)
Constant 0.0860**(0.0358)
‐0.0510(0.0412)
‐0.0941(0.5997)
‐0.2034 (0.3394)
Observations 88 88 80 80
R‐squared 0.12 0.55 0.55
Source: Authors’ computations based on CSA’s manufacturing census
Notes: Standard errors in parentheses. *, ** and *** represent significance at 10%, 5% and
1%, respectively.