Firm Survival and Change in Ghana, 2003–2013

Firm Survival and Change in Ghana, 2003–2013 Abstract This paper explores the determinants of firm survival in Ghanaian manufacturing and the contributions of growth and selection to the evolution of the firm size distribution. For this analysis we created a two-wave panel spanning 10 years to study exit, growth and decline, by re-surveying 1000 firms randomly selected from the 2003 National Industrial Census. We find strong differences in exit patterns by region and firm size. Former owners and managers commonly cite personal circumstances as the reason for exit in the case of small firms and increasing costs in the case of large firms. We show that both growth and selection played only a small role in the evolution of the firm size distribution, contradicting earlier work on Ghana. Overall, the picture we paint of manufacturing in Ghana is not a positive one: total employment by firms operating before 2003 decreased from 134,863 in 2003 to 74,319 in 2013, although we cannot explore to what extent new employment in firms that entered after 2003—who were not surveyed—compensated for this decline. 1. Introduction The Ghanaian economy has been characterised by important changes over the last few decades: high levels of GDP growth, an IMF reform process that led to many changes in policy, increases in consumption expenditure, the discovery and production of oil and a rise of the service industry. More recently the contribution of the manufacturing sector to GDP has been declining. Research using firm-level data has similarly documented the lacklustre performance of manufacturing firms during the 1990s and early 2000s (Teal et al. 2006; Sandefur, 2010). In this paper we explore whether this trend continued in the last 10 years. To do this we use a follow-up survey conducted in 2013 on manufacturing firms first interviewed as part of the 2003 National Industrial Census. This allows us to create a two-wave panel dataset of 1000 manufacturing firms. The picture our analysis paints is not a positive one—the firms we surveyed have generally performed poorly over the last 10 years, with high rates of exit and shrinkage of surviving firms. We also use the data to explore whether the evolution of the firm size distribution in Ghana is explained by growth, selection or entry (Cabral and Mata, 2003; Sandefur, 2010) and we find no role for growth and only a small role for selection in explaining the evolution of the firm size distribution in Ghana. This suggests that entry may be important but our work cannot speak directly to the importance of entry since we did not collect data on new firms that were born between 2003 and 2013. Previous research has suggested, however, that entry may be an important and under-researched contributor to the evolution of the firm size distribution in Ghana (Sandefur, 2010). This paper makes a contribution to the literature by describing patterns of Ghanaian firm growth and survival between 2003 and 2013, and the factors that are correlated with firm survival. In particular, this paper contributes to the literature on the evolution of the firm size distribution (Cabral and Mata, 2003; Luttmer, 2007), and provides some insight as to whether Ghana is facing a ‘missing middle’ (Tybout, 2000; Hsieh and Olken, 2014). A recent literature has emphasised the role of management in firm survival and growth (Bloom and Van Reenen, 2010; Bloom et al. 2014). Our paper provides some evidence that ownership and management matters: personal circumstances of owners and managers can be crucial for the survival of firms, in particular small ones. The paper is structured as follows: Section 2 provides some background on the Ghanaian economic environment and discusses some earlier studies on Ghanaian manufacturing. Section 3 describes the survey and provides some descriptive statistics. Section 4 presents the evidence on firm exit and survival and explores the self-reported reasons for exit. Section 5 focuses on firm growth and decline and considers the role of selection and growth on the overall firm size distribution. Finally, Section 6 concludes. 2. The economic environment The Ghanaian economy has recently exhibited high growth: according World Bank figures, between 2003 and 2012 growth was 7.5% per annum on average. During this time the composition of the economy has also seen some considerable changes: the contribution to the gross domestic product by the service industry has been growing significantly, with an average annual growth rate of 12.9% between 2003 and 2012. Services constituted 50.0% of value added in 2012, while in 1990 this was only 38.1%. Industrial output has also been growing considerably, but this growth has mainly been achieved in other industrial sectors than the manufacturing sector. The manufacturing sector is estimated to have been growing at 3.3%, while other industrial sectors, such as mining, water production and construction have grown by 9.1% on average between 2003 and 2013. This means that the relative share of the contribution of the manufacturing to GDP has declined, from 9.8% in 1990 to 6.9% in 2012 (see Table 1). Most of this decline happened after 2007. Household and government consumption has risen by 5.6% on average in the same time, indicating that Ghanaian manufacturing has profited less from this increase than other sectors. Filling out the picture from macrodata, previous work on microdata from manufacturing firms has shown weak performance, despite several regulatory changes, such as trade liberalisation and exchange rate reforms, which should have made it easier to compete (see e.g., Sutton and Kpentey, 2012, for a discussion of sector-specific policy measures). Teal et al. (2006) describe results from the RPED firm surveys indicating that output by manufacturing firms fell between 2000 and 2003 and Teal (2016) uses the 1987 and 2003 NICs to show that by 2003 a much larger proportion of the manufacturing labour force was working in low productivity firms than in 1987. Table 1: Share of Sectors in Value Added (as a Percentage of GDP)   1990  1995  2000  2005  2010  2012  Agriculture  45.1  42.7  39.4  40.9  29.8  22.7  Manufacturing  9.8  10.3  10.1  9.5  6.8  6.9  Other industry  7.0  16.5  18.3  18.0  12.3  20.5  Services  38.1  30.6  32.2  31.6  51.1  50.0  Total  100.0  100.0  100.0  100.0  100.0  100.0    1990  1995  2000  2005  2010  2012  Agriculture  45.1  42.7  39.4  40.9  29.8  22.7  Manufacturing  9.8  10.3  10.1  9.5  6.8  6.9  Other industry  7.0  16.5  18.3  18.0  12.3  20.5  Services  38.1  30.6  32.2  31.6  51.1  50.0  Total  100.0  100.0  100.0  100.0  100.0  100.0  Source: World Bank Development Indicators. One important question when interrogating the lacklustre performance of manufacturing is whether competition is driving out low productivity firms or whether these firms can continue to operate. This question can only be answered using panel data. Frazer (2005) uses the RPED firm panel to study the exit of Ghanaian manufacturing firms and how this relates to firm productivity. Frazer finds that low firm productivity is a good predictor of firm exits. The size and age of the firm also seem relevant in predicting whether a firm continues to operate or not: large firms and older firms are less likely to exit. The former can be explained by a simple model where firm growth is dependent on success: a less successful firm is less likely to grow, but also more likely to fail. An explanation for the latter might be that characteristics that helped to prevent exit in the past help prevent exits in the present as well, causing older firms to be more likely to survive. Trade models of firm selection, such as Melitz (2003) predict that exporting firms are less likely to exit, as predominantly successful firms are able to become exporters, but Frazer does not find evidence for this in Ghana. The importance of firm productivity on firm exits seems to challenge earlier studies done in sub-Saharan Africa, such as Liedholm et al. (1994), who emphasise personal circumstances playing a role in at least a quarter of exits, but who do not consider productivity due to a lack of data. Söderbom et al. (2006) also focus on the relationship between productivity and selection. They find, on the basis of firm panel surveys in Ghana and several other African countries, that efficiency matters more for the survival of larger firms. Being productive does not prevent small firms from going out of business. This might indicate the important role of other considerations, such as personal circumstances, in the case of small enterprises. We explore the correlates of firm survival and exit below. A second important question is how the firm size distribution evolves. Sandefur (2010) examines this question in Ghana using the 1988 and 2003 National Industrial Census microdata and finds that selection, rather than growth, drives the evolution of the firm size distribution. This differs from results from for example firm studies done in other countries, e.g., Portugal (Cabral and Mata, 2003). Although Sandefur’s paper is unpublished his result was discussed in the 2013 World Development Report on Jobs (World Bank, 2013) as well as in a recently published summary paper on firm dynamics in developing countries (Li and Rama, 2015). We take up this issue in Section 5, arguing that the result that selection and not growth drives the evolution of the firm size distribution is only partially correct. 3. Survey description In 2013 a total of 1,000 firms located in five locations (Accra, Tema, Kumasi, Sekondi-Takoradi and Cape Coast) were sampled from the 2003 Ghana National Industrial Census (NIC), conducted by the Ghanaian Statistical Service and covering all establishments engaged in manufacturing in Ghana.1 Stratification was based on firm size, firm age, region and sector, to ensure that firms with a wide range of characteristics were included. In total 135 strata were used (see Table 2 for the breakdown of the factors determining the stratification). To account for the diversity of firms, sampling probabilities were adjusted on the basis of the variance of firm size in each stratum. This led to oversampling of certain strata, while others were undersampled. As can be seen in Table 2, large and old firms have a much higher probability of being included in the final sample, while young and small firms have a lower probability of being included. Practically all large firms (with more than 75 employees) in the regions sampled were included in the final sample. Table 2: Main Summary Statistics of the Firms from the 2003 National Industrial Census, the Sampled Area and the Sample of the Survey   In 2003 National Industrial Census  Sampling area  Sampled  No.  %  No.  %  No.  %  Region   Greater Accra (incl. Tema)  6,654  25.1  6,655  59.2  579  57.9   Kumasi  3,374  12.8  3,374  30.0  304  30.4   Sekondi-Takoradi1  855  3.2  855  7.6  90  9.0   Cape Coast1  355  1.3  355  3.2  27  2.7   Other  15,237  57.6  –  –  –  –  Sector   Food & beverages  4,257  16.1  913  8.1  132  13.2   Textiles, garments & footwear  11,620  43.9  5,359  47.7  299  29.9   Wood & furniture  6,085  23.0  2,416  21.5  215  21.5   Machinery & metal  2,133  8.1  1,266  11.3  135  13.5   Other  2,381  9.9  1,284  11.4  219  21.9  Age   Founded 1999–2003  13,499  51.0  5,942  52.9  344  34.4   Founded 1989–1998  9,432  35.6  3,923  34.9  377  37.7   Founded before 1988  3,348  12.7  1,330  11.8  254  25.4   Unknown  197  0.7  43  0.4  25  2.5  Size   Small (0–9 workers)  22,375  84.5  9,394  84.6  386  38.6   Medium (10–74 workers)  3,733  14.1  1,619  14.4  392  39.2   Large (more than 75)  367  1.4  225  2.0  222  22.2    In 2003 National Industrial Census  Sampling area  Sampled  No.  %  No.  %  No.  %  Region   Greater Accra (incl. Tema)  6,654  25.1  6,655  59.2  579  57.9   Kumasi  3,374  12.8  3,374  30.0  304  30.4   Sekondi-Takoradi1  855  3.2  855  7.6  90  9.0   Cape Coast1  355  1.3  355  3.2  27  2.7   Other  15,237  57.6  –  –  –  –  Sector   Food & beverages  4,257  16.1  913  8.1  132  13.2   Textiles, garments & footwear  11,620  43.9  5,359  47.7  299  29.9   Wood & furniture  6,085  23.0  2,416  21.5  215  21.5   Machinery & metal  2,133  8.1  1,266  11.3  135  13.5   Other  2,381  9.9  1,284  11.4  219  21.9  Age   Founded 1999–2003  13,499  51.0  5,942  52.9  344  34.4   Founded 1989–1998  9,432  35.6  3,923  34.9  377  37.7   Founded before 1988  3,348  12.7  1,330  11.8  254  25.4   Unknown  197  0.7  43  0.4  25  2.5  Size   Small (0–9 workers)  22,375  84.5  9,394  84.6  386  38.6   Medium (10–74 workers)  3,733  14.1  1,619  14.4  392  39.2   Large (more than 75)  367  1.4  225  2.0  222  22.2  Source: Own calculations. 1Sekondi-Takoradi and Cape Coast were treated as one region in the stratification. The survey was conducted between August and November 2013. Attempts were made to interview each firm from the sample. If the firm was operating, a questionnaire similar to the 2003 National Industrial Census was conducted, asking mainly about employment and firm productivity. If a firm was no longer operating, enumerators attempted to find a former manager or representative of the firm and conduct a questionnaire with exit-specific questions. In case no former manager or representative could be found, a family member, former worker or neighbour was interviewed instead.2 In all cases where the main firm questionnaire was not undertaken, the enumerator was asked to record basic information on the firm, such as whether the firms was still operating or not, whether a firm sign was still present, and in case no interview was undertaken, what the reason was for this. The 2013 survey allowed the creation of a two-wave panel of 1,000 firms, some of which survived and some of which either died or were untraced. Table 3 shows the results of our attempts to trace 1,000 firms. Forty-five per cent of the firms were found and interviewed while another 12% were found and were operating but refused to participate in the survey. Twenty-one per cent of the firms had exited while no trace was found of 22% of the firms. This last group is likely to be mainly exits but could also include firms that moved (although the enumerators did try and trace firms that were known to have moved within the city in which they were located). Table 3 also shows that the survey was less successful in finding and interviewing firms in Accra, small and young firms. Large firms, those in the ‘Other’ sector (which includes a variety of sectors, such as utility companies, chemical companies and printing firms) and those in Accra and Takoradi were more likely to refuse. Table 3: Firm Status in 2013 by 2003 Characteristics   Found and interviewed  Exit  Untraced  Operating but refusal   All Firms  42.8  21.2  22.2  13.8  Region   Greater Accra  37.3  21.4  24.2  17.1   Kumasi  51.0  17.4  23.7  7.9   Sekondi-Takoradi  46.7  28.9  8.9  15.6   Cape Coast  55.6  33.3  7.4  3.7  Sector   Food & beverages  35.6  25.8  16.7  22.0   Textiles, garments & footwear  44.1  20.7  31.8  3.3   Wood & furniture  42.3  23.7  27.0  7.0   Machinery & metal  52.6  14.8  16.3  16.3   Other  39.7  20.5  11.4  28.3  Age   Founded 1999–2003  39.1  21.8  37  2.1   Founded 1989–1998  50.0  24.2  15.6  10.2   Founded before 1988  36.5  14.9  8.1  40.5  Size in 2003   Small (0–9 workers)  35.5  24.7  28.5  11.4   Medium (10–74 workers)  48.0  18.8  21.8  11.4   Large (more than 75)  45.7  19.7  13.8  20.9    Found and interviewed  Exit  Untraced  Operating but refusal   All Firms  42.8  21.2  22.2  13.8  Region   Greater Accra  37.3  21.4  24.2  17.1   Kumasi  51.0  17.4  23.7  7.9   Sekondi-Takoradi  46.7  28.9  8.9  15.6   Cape Coast  55.6  33.3  7.4  3.7  Sector   Food & beverages  35.6  25.8  16.7  22.0   Textiles, garments & footwear  44.1  20.7  31.8  3.3   Wood & furniture  42.3  23.7  27.0  7.0   Machinery & metal  52.6  14.8  16.3  16.3   Other  39.7  20.5  11.4  28.3  Age   Founded 1999–2003  39.1  21.8  37  2.1   Founded 1989–1998  50.0  24.2  15.6  10.2   Founded before 1988  36.5  14.9  8.1  40.5  Size in 2003   Small (0–9 workers)  35.5  24.7  28.5  11.4   Medium (10–74 workers)  48.0  18.8  21.8  11.4   Large (more than 75)  45.7  19.7  13.8  20.9  Note: Numbers reported in this table are percentages. Row percentages sum to 100%. The data are unweighted. Source: own calculations. 4. Firm survival and exit In this section we discuss survival and exit patterns between 2003 and 2013, and show how these differ between regions and sectors. Furthermore, we describe the reasons for exits, as reported by the respondents. 4.1 Patterns of survival and exit Table 4 shows correlates of two measures of firm exit: exit measure 1 excludes firms that were not found while exit measure 2 assumes that firms that were not found actually exited. Weights increase the contributions of small firms to any statistics since these firms were less likely to be sampled and large firms had a selection probability close to one. Thus, when weighted, exit rates over 10 years were between 33% (assuming all untraced firms did not exit) and 56% (assuming all untraced firms exited). This is equivalent to an annualised compound exit rate of between 3.9% and 7.9%. Aga and Francis (2015) used the World Bank’s Enterprise Survey panels from 47 countries, including Ghana, and found that annualised compound exit rates were generally between 3% and 5% per year,3 when using panels that were an average of 4 years apart. In the Enterprise Survey panel for Ghana between 68% and 77% of the (manufacturing and services) firms interviewed in 2007 had exited in 2013 depending on how exit was measured. The annualised compound exit rate was thus between 15% and 19%. This exit rate was the highest of any country in the sample and exit rates for the other African countries in the sample (DRC, Kenya, Tanzania, Uganda, Zambia, Rwanda) were much lower. The exit rate for Ghana using the 2003–2013 panel is thus slightly above the exit rates for the other African countries in which enterprise surveys have been undertaken but markedly lower than in the Ghanaian enterprise survey conducted by the World Bank. Table 4: Correlates of Firm Exit   Exit measure 1 (excluding firms not found)  Exit measure 2 (treating not found as exit)  No exit  Exit  No exit  Exit  Region   Greater Accra (incl. Tema)  61.1  38.9  36.7  63.3   Kumasi  77.5  22.5  53.6  46.4   Sekondi-Takoradi  67.7  32.3  55.5  44.5   Cape Coast  53.1  46.9  52.3  47.7  Sector   Food & beverages  61.6  38.4  44.6  55.4   Textiles, garments & footwear  63.6  36.4  38.6  61.4   Wood & furniture  66.4  33.6  39.8  60.2   Machinery & metal  79.8  20.2  64.8  35.2   Other  66.6  33.4  51.3  48.7  Age   Founded 1999–2003  59.9  40.1  36.7  63.3   Founded 1989–1998  75.2  24.8  51.3  48.7   Founded before 1988  68.1  31.9  52.8  47.2  Size in 2003   Small (0–9 workers)  64.8  35.2  40.1  59.9   Medium (10–74 workers)  71.6  28.4  60.1  39.9   Large (more than 75)  83.9  16.1  77.2  22.8  All firms  66.6  33.4  43.7  56.3    Exit measure 1 (excluding firms not found)  Exit measure 2 (treating not found as exit)  No exit  Exit  No exit  Exit  Region   Greater Accra (incl. Tema)  61.1  38.9  36.7  63.3   Kumasi  77.5  22.5  53.6  46.4   Sekondi-Takoradi  67.7  32.3  55.5  44.5   Cape Coast  53.1  46.9  52.3  47.7  Sector   Food & beverages  61.6  38.4  44.6  55.4   Textiles, garments & footwear  63.6  36.4  38.6  61.4   Wood & furniture  66.4  33.6  39.8  60.2   Machinery & metal  79.8  20.2  64.8  35.2   Other  66.6  33.4  51.3  48.7  Age   Founded 1999–2003  59.9  40.1  36.7  63.3   Founded 1989–1998  75.2  24.8  51.3  48.7   Founded before 1988  68.1  31.9  52.8  47.2  Size in 2003   Small (0–9 workers)  64.8  35.2  40.1  59.9   Medium (10–74 workers)  71.6  28.4  60.1  39.9   Large (more than 75)  83.9  16.1  77.2  22.8  All firms  66.6  33.4  43.7  56.3  Note: Exit measure 1 excludes firms not found from the analysis whereas exit measure 2 assumes firms not found exited. The data are weighted, and hence the figures are different to those reported in Table 3. Source: own calculations. Exit rates do not necessarily or sufficiently indicate poor performance. Creative destruction suggests that exit can be positive if less efficient firms are replaced by more efficient ones. Aw et al. (2001) find that in a period of rapid growth in Taiwan between 70% and 87% of manufacturing firms that were alive in 1981 had exited by 1991. Shiferaw and Bedi (2013) find annual exit rates in the annual Ethiopian manufacturing census of between 14% and 17% for the period 1997 and 2007, during which real sales of manufacturing firms tripled and employment increased by around 40%. However, it should be noted that firms dropping below 10 employees are excluded from future waves, which means that exit rates are likely overstated. The annual average exit rate of between 3.9% and 7.8% of Ghanaian firms is not out of line with exit rates in other countries. Davis and Haltiwanger (1992) report an annual exit rate of 8% in the United States. Hallward-Driemeier (2009) reports an annual exit rate of 7.6% for a large group of Eastern European and Central Asian countries, for the time period 2002–2009. However, as we show later our other findings using our Ghanaian panel data do not give reason for optimism about the growth of Ghanaian manufacturing. We now explore correlates of firm exit. Using the first measure of exit firms located in Accra and Cape Coast, smaller firms and younger firms are more likely to have exited between 2003 and 2013. By this measure around one-third of the firms from the 2003 sample had exited 10 years later. However if we assume, as in the second measure of exit variable, that firms that could not be traced also exited then around 56% of firms had exited after 10 years. Using this second measure also changes some results obtained using the first measure. For example Accra firms are unambiguously more likely to have exited than firms from the other regions while the smallest firms were much more likely not to be traced and therefore have much higher rates of exit when using the second measure. Table 5 reports the results of a simple linear probability model of exit. Using the measure of exit that excludes firms not traced column 1 shows that firms located in Kumasi, the largest firms, firms in the middle age category (5–14 years old in 2003) and machinery and metals firms were less likely to exit than other firms. Column 2 assumes that firms that were not traced exited. Accra now stands out as being the location most likely to be correlated with exit, suggesting that tracing firms was more of an issue there than in Takoradi or Cape Coast. Small firms are now much more likely to have exited than large or medium-sized firms, implying that small firms were more likely not to be traced. This differential exit rates for small and medium or large firms is a pattern that has been documented more widely in sub-Saharan Africa (see e.g., Van Biesebroeck, 2005). Table 5: A Linear Probability Model of Exit Dependent variable  Exit measure 1 (excluding firms not found)  Exit measure 2 (treating not found as exit)  Kumasi  −0.157***  −0.156***  (0.0523)  (0.0478)  Sekondi-Takoradi  −0.0486  −0.180**  (0.0830)  (0.0777)  Cape Coast  0.0495  −0.162  (0.134)  (0.124)  Textiles & garments  −0.0270  0.00707  (0.0868)  (0.0754)  Wood & furniture  −0.0216  0.0588  (0.0914)  (0.0800)  Machinery & metal  −0.166*  −0.203**  (0.0913)  (0.0870)  Other sector  −0.0479  −0.0788  (0.0963)  (0.0876)  Medium size firm (10–74)  −0.0257  −0.137***  (0.0424)  (0.0391)  Large size firm (75+)  −0.184***  −0.316***  (0.0526)  (0.0491)  Founded 1989–1998  −0.138***  −0.110**  (0.0530)  (0.0468)  Founded before 1988  −0.0577  −0.0762  (0.0708)  (0.0632)  Constant  0.500***  0.718***  (0.0861)  (0.0738)  Observations  778  1,000  R2  0.063  0.089  Dependent variable  Exit measure 1 (excluding firms not found)  Exit measure 2 (treating not found as exit)  Kumasi  −0.157***  −0.156***  (0.0523)  (0.0478)  Sekondi-Takoradi  −0.0486  −0.180**  (0.0830)  (0.0777)  Cape Coast  0.0495  −0.162  (0.134)  (0.124)  Textiles & garments  −0.0270  0.00707  (0.0868)  (0.0754)  Wood & furniture  −0.0216  0.0588  (0.0914)  (0.0800)  Machinery & metal  −0.166*  −0.203**  (0.0913)  (0.0870)  Other sector  −0.0479  −0.0788  (0.0963)  (0.0876)  Medium size firm (10–74)  −0.0257  −0.137***  (0.0424)  (0.0391)  Large size firm (75+)  −0.184***  −0.316***  (0.0526)  (0.0491)  Founded 1989–1998  −0.138***  −0.110**  (0.0530)  (0.0468)  Founded before 1988  −0.0577  −0.0762  (0.0708)  (0.0632)  Constant  0.500***  0.718***  (0.0861)  (0.0738)  Observations  778  1,000  R2  0.063  0.089  Note: Reference (omitted) categories are firms located in Accra, in the food and beverage sector, with 0–9 employees and younger than 4 years old in 2003. Standard errors in parentheses. Significance levels are: ***p < 0.01, **p < 0.05, *p < 0.1. Source: own calculations. 4.2 Self-reported reasons for firm exit Table 6 shows reasons for exit amongst the 191 of 212 exiting firms that provided a reason for exit (firms that were not found are not included in this analysis). Circumstances of the owner accounted for around a third of all reasons for exit given—such as illness, retirement, moving to another region or country, etc. Falling demand and loss of land, buildings or equipment each account for just under 20% of exits while around 20% responded that they did not know why the firm had exited (if the owner, a manager or a worker could not be traced a neighbour was asked). Table 6: Reasons Given for Exits by Respondents Reasons for exit  Weighted percentage  Illness or death of owner or manager  12.3  Set up other business, merged firm or found wage employment elsewhere  7.1  Owner moved  13.6  Falling demand  19.4  Increased competition from imports  3.6  Increased competition from local competition  1.5  Increased costs  4.0  Loss of building, land or equipment  16.7  Debt or credit problems  1.5  Managerial problems  0.2  Don’t know  19.7  Refused  0.6  Total  100.0  Reasons for exit  Weighted percentage  Illness or death of owner or manager  12.3  Set up other business, merged firm or found wage employment elsewhere  7.1  Owner moved  13.6  Falling demand  19.4  Increased competition from imports  3.6  Increased competition from local competition  1.5  Increased costs  4.0  Loss of building, land or equipment  16.7  Debt or credit problems  1.5  Managerial problems  0.2  Don’t know  19.7  Refused  0.6  Total  100.0  Note: The respondents were either former managers, owners or workers of the firm, or if they could not be found, a neighbour or family member. Source: answers from the exit questionnaire. Table 7 breaks down reasons for exit by firm size. Circumstances of the owner, decreasing demand and loss of land, buildings and equipment are more likely to be given as a reason for exit amongst small firms compared to large firms. Large firms were much more likely to have exited due to increased costs than small firms and this was the most common reason for exit amongst large firms. That small firms are more likely to exit due to the circumstances of the owner accords with the work of Liedholm et al. (1994), who found this in their survey of smaller firms (fewer than 50 employees) in several African countries. Teal et al. (2006) find that selection on efficiency is a more important determinant of exit amongst large firms than small firms. If rising costs imply lower efficiency then the fact that Table 5 shows a high fraction of large firms exiting due to increased costs suggests that our results accord with those of Teal et al. (2006). Table 7: Breakdown of Exit Reasons by Firm Size Reason for exit  0–9 employees  10–74 employees  75+ employees  All  Illness or death of owner or manager  12.3  12.8  3.9  12.3  Set up other business, merged firm or found wage employment elsewhere  7.9  2.9  3.9  7.1  Owner moved  14.4  9.9  3.9  13.6  Falling demand  20.1  16.5  3.9  19.4  Increased competition from imports  3.3  4.5  7.9  3.6  Increased competition from local competition  1.3  2.3  3.9  1.5  Increased costs  3.7  4.1  25.0  4.0  Loss of building, land or equipment  16.6  18.4  3.9  16.7  Debt or credit problems  1.3  2.9  3.9  1.5  Managerial problems  0.0  0.9  3.9  0.2  Don’t know  19.3  21.3  31.6  19.7  Refused  0.0  3.4  3.9  0.6  Total  100.0  100.0  100.0  100.0  Reason for exit  0–9 employees  10–74 employees  75+ employees  All  Illness or death of owner or manager  12.3  12.8  3.9  12.3  Set up other business, merged firm or found wage employment elsewhere  7.9  2.9  3.9  7.1  Owner moved  14.4  9.9  3.9  13.6  Falling demand  20.1  16.5  3.9  19.4  Increased competition from imports  3.3  4.5  7.9  3.6  Increased competition from local competition  1.3  2.3  3.9  1.5  Increased costs  3.7  4.1  25.0  4.0  Loss of building, land or equipment  16.6  18.4  3.9  16.7  Debt or credit problems  1.3  2.9  3.9  1.5  Managerial problems  0.0  0.9  3.9  0.2  Don’t know  19.3  21.3  31.6  19.7  Refused  0.0  3.4  3.9  0.6  Total  100.0  100.0  100.0  100.0  Note: The data have been weighted. 5. Firm growth and decline This section presents evidence on firm growth and decline and shows that there are regional and sectoral differences in the evolution of surviving firms. We then explore what determines the evolution of the firm size distribution. Finally, we consider the changes in aggregate employment by the firms in our sample and show that total employment in the firms in our sample dropped by 45%. Due to data limitations, we cannot distinguish between paid apprentices and regular workers for most of the firms. Our measures of employment therefore include apprentices. This might not be the most reliable measure for firm performance, given that apprenticeships are often not paid very well or paid at all and therefore apprentices are less likely to be laid off when the firms is not performing. In the Appendix we apply a number of corrections, which suggest that the drop in total employment excluding apprentices is likely to be between 33% and 36%. 5.1 Patterns of growth and decline Table 8 explores the patterns of growth and shrinking amongst the firms that survived and were traced. The first part of the table shows that 54.0% of the surviving firms shrank, and that 34.9% grew.4 If we include exits, refusals and firms not found about 14% of the firms in the 1000 firm sample grew, 21.6% shrank, 21.1% exited, 33.1% were not traced and 5.7% were operating but refused to be interviewed. Table 8: Firm Status For continuing firms  Number  Percentage  Weighted percentage  Shrunk  256  62.9  54.0  Same Size  31  7.6  11.1  Grew  120  29.5  34.9  Including exits and refusals   Not found  222  22.2  33.1   Exit  212  21.2  21.1   Refusal, but still operating  159  15.9  5.7   Shrunk  256  25.6  21.6   Same Size  31  3.1  4.5   Grew  120  12.0  14.0  For continuing firms  Number  Percentage  Weighted percentage  Shrunk  256  62.9  54.0  Same Size  31  7.6  11.1  Grew  120  29.5  34.9  Including exits and refusals   Not found  222  22.2  33.1   Exit  212  21.2  21.1   Refusal, but still operating  159  15.9  5.7   Shrunk  256  25.6  21.6   Same Size  31  3.1  4.5   Grew  120  12.0  14.0  Using data from the World Bank’s Enterprise Surveys, including Ghana and 30 other sub-Saharan African countries, as well as 74 countries in lower, lower middle and upper income countries in other regions, Aterido and Hallward-Driemeier (2010) reported that ‘more than half’ of incumbent firms in sub-Saharan Africa increased employment while 20% shrank. This suggests that employment growth for survivors in our Ghanaian sample is lower than the average for these thirty one sub-Saharan countries. Mitra et al. (2014) use the Business Environment and Enterprise Performance Surveys (BEEPS), part of the World Bank Enterprise Surveys, to explore firm growth in transition economies and compare this to firm growth patterns in mature European economies. They report the balance between growing and shrinking firms—the proportion of firms that grew minus the proportion that shrank—and ignore firms that stayed the same size. They find that in 2005 the proportion of firms that grew was 20–22 percentage points higher than the proportion of firms that shrank in CIS countries, 15 percentage points higher in SEE countries, 6 percentage points higher in the eight countries that joined the EU in 2004 and 16 percentage points higher in the PIGS countries (Portugal, Ireland, Greece and Spain). Again this suggests that Ghanaian firms had lower growth than low income CIS countries, although the differences in sampling, discussed above, imply that growth rates in the enterprise surveys would be lower if a similar method was used to calculate growth rates. Focusing on the surviving firms Table 9 explores correlates of growth and decline. Firms in the ‘Other’ sector (e.g., chemical products) were more likely to grow than those in other sectors. Textiles and Wood were the two worst performing sectors. Conditional on survival firms in Accra and Takoradi were more likely to grow than firms in Kumasi. Firms in Cape Coast performed terribly with only six per cent of firms reporting more employment than in 2003. Younger firms were more likely to grow than older firms while the smallest and largest firms were more likely to grow. Table 9: Changes in Firm Size Sector  Shrunk  Same Size  Grew  Total  Food & beverages  46.8  12.6  40.5  100.0  Textiles, garments & footwear  61.9  11.0  27.2  100.0  Wood & furniture  56.4  7.6  36.0  100.0  Machinery & metal  45.1  13.8  41.1  100.0  Other manufacturing  41.9  12.5  45.6  100.0  Location   Greater Accra  49.6  10.1  40.3  100.0   Kumasi  62.6  11.2  26.2  100.0   Takoradi  50.7  1.2  48.1  100.0   Cape Coast  36.3  57.8  5.9  100.0  Age group   Founded 1999–2003  51.2  13.2  35.6  100.0   Founded 1989–1998  51.9  10.0  38.2  100.0   Founded before 1988  68.3  8.3  23.4  100.0  Size group   Small (0–9)  48.4  13.2  38.4  100.0   Medium (10–74)  74.1  4.9  20.9  100.0   Large (75+)  58.4  1.9  39.7  100.0  Total  54.0  11.1  34.9  100.0  Sector  Shrunk  Same Size  Grew  Total  Food & beverages  46.8  12.6  40.5  100.0  Textiles, garments & footwear  61.9  11.0  27.2  100.0  Wood & furniture  56.4  7.6  36.0  100.0  Machinery & metal  45.1  13.8  41.1  100.0  Other manufacturing  41.9  12.5  45.6  100.0  Location   Greater Accra  49.6  10.1  40.3  100.0   Kumasi  62.6  11.2  26.2  100.0   Takoradi  50.7  1.2  48.1  100.0   Cape Coast  36.3  57.8  5.9  100.0  Age group   Founded 1999–2003  51.2  13.2  35.6  100.0   Founded 1989–1998  51.9  10.0  38.2  100.0   Founded before 1988  68.3  8.3  23.4  100.0  Size group   Small (0–9)  48.4  13.2  38.4  100.0   Medium (10–74)  74.1  4.9  20.9  100.0   Large (75+)  58.4  1.9  39.7  100.0  Total  54.0  11.1  34.9  100.0  Note: The data is weighted. The ‘same size’ category includes all firms that experienced an increase or decrease in employment of less than 5%. Firm size is measured as the number of ‘persons engaged’ with the firm, and includes both paid and unpaid workers. Table 10 shows similar analysis but including exits, refusals and untraced firms. An important point to highlight is that large firms had a high refusal rate. Table 10: Firm Status Sector  Untraceable  Exit  Refusal  Shrunk  Same Size  Grew  Total  Food & beverages  27.6  27.8  7.1  17.0  5.4  15.1  100.0  Textiles, garments & footwear  39.4  22.1  3.9  21.3  3.8  9.5  100.0  Wood & furniture  40.1  20.2  3.8  21.1  2.9  11.9  100.0  Machinery & metal  18.8  16.4  9.5  23.7  7.7  23.8  100.0  Other  22.9  25.8  12.9  16.5  4.2  17.6  100.0  Location   Greater Accra  39.9  23.4  5.7  15.5  3.4  12.2  100.0   Kumasi  30.9  15.5  6.7  29.6  4.8  12.5  100.0   Takoradi  18.0  26.5  4.1  25.7  0.4  25.2  100.0   Cape Coast  1.4  46.3  4.4  17.5  27.9  2.5  100.0  Age group   Founded 1999–2003  38.7  24.6  4.9  16.3  4.4  11.1  100.0   Founded 1989–1998  31.7  16.9  6.5  24.0  4.1  16.8  100.0   Founded before 1988  22.5  24.8  7.9  30.3  3.9  10.7  100.0  Size group   Small (0–9)  38.1  21.7  4.1  17.7  4.7  13.6  100.0   Medium (10–74)  16.1  23.8  10.3  37.8  2.3  9.6  100.0   Large (75+)  8.0  14.8  43.5  21.3  0.4  12.0  100.0  Total  34.4  21.9  5.8  20.6  4.2  13.0  100.0  Sector  Untraceable  Exit  Refusal  Shrunk  Same Size  Grew  Total  Food & beverages  27.6  27.8  7.1  17.0  5.4  15.1  100.0  Textiles, garments & footwear  39.4  22.1  3.9  21.3  3.8  9.5  100.0  Wood & furniture  40.1  20.2  3.8  21.1  2.9  11.9  100.0  Machinery & metal  18.8  16.4  9.5  23.7  7.7  23.8  100.0  Other  22.9  25.8  12.9  16.5  4.2  17.6  100.0  Location   Greater Accra  39.9  23.4  5.7  15.5  3.4  12.2  100.0   Kumasi  30.9  15.5  6.7  29.6  4.8  12.5  100.0   Takoradi  18.0  26.5  4.1  25.7  0.4  25.2  100.0   Cape Coast  1.4  46.3  4.4  17.5  27.9  2.5  100.0  Age group   Founded 1999–2003  38.7  24.6  4.9  16.3  4.4  11.1  100.0   Founded 1989–1998  31.7  16.9  6.5  24.0  4.1  16.8  100.0   Founded before 1988  22.5  24.8  7.9  30.3  3.9  10.7  100.0  Size group   Small (0–9)  38.1  21.7  4.1  17.7  4.7  13.6  100.0   Medium (10–74)  16.1  23.8  10.3  37.8  2.3  9.6  100.0   Large (75+)  8.0  14.8  43.5  21.3  0.4  12.0  100.0  Total  34.4  21.9  5.8  20.6  4.2  13.0  100.0  Note: The data is weighted. Same size includes all firms that did not change employment by more than 5%. 5.2 The evolution of the firm size distribution and the ‘missing middle’ We now explore what these patterns of selection and growth mean for the evolution of the firm size distribution. The firm size distribution and its evolution can give us insights into which dynamics influence industrial change. Cabral and Mata (2003) examined the growth and exit patterns of a cohort of young Portuguese firms and showed that industrial evolution was mainly driven by within-firm growth rather than by selective exit of small firms. In a study on older firm-level data from Ghana, Sandefur (2010) argued that there was actually little evidence for within-firm growth: firms that were small in 1988 remained small in 2003, and the big firms that survived to 2003 were already big in 1988. But Sandefur argued that the firms that survived were much larger to start with and thus that changes in the firm size distribution were driven by selection. Hsieh and Klenow (2014) also showed that within-firm growth was negative in formal Indian firms and that selection effects were stronger even than in the USA. Constraints to firm growth have been argued to be one of the reasons why there is a ‘missing middle’ in developing countries, an underrepresentation of medium size firms (Tybout, 2000).5 In this section we will consider the overall firm size distribution, look at how firm age affects the firm size distribution and finally consider the growth and selection patterns of young firms (those founded between 1999 and 2003). Figure 1 shows the overall firm size distribution of the firms in the four cities surveyed. The figure also includes a best fit of the log-normal distribution, based on the firm size distribution in 2013. We find some evidence for a ‘missing middle’ in the sense that firms in the medium-sized categories are underrepresented compared to the log-normal distribution and there is an overrepresentation of larger firms. Like Hsieh and Olken (2014), we do not find evidence for bimodality in the firm size distribution. Figure 1: View largeDownload slide The Overall Firm Size Distribution of all the Firms in 2003, The Log-normal Best Fit Distribution and the Overall Firm Size Distribution of the Surviving Firms in 2013. Firm size is defined as the number of persons engaged with the firm. Note: 2013 includes only 401 surviving firms with positive employment. Weights are applied. Figure 1: View largeDownload slide The Overall Firm Size Distribution of all the Firms in 2003, The Log-normal Best Fit Distribution and the Overall Firm Size Distribution of the Surviving Firms in 2013. Firm size is defined as the number of persons engaged with the firm. Note: 2013 includes only 401 surviving firms with positive employment. Weights are applied. Figure 2 shows the firm size distributions of several cohorts of firms. This figure shows a positive relation between size and age: older firms are larger than younger firms. Some bimodality is present in firms more than 30 years old: there is a second mode at around 300 employees. Figure 2: View largeDownload slide Cross Sectional Relationship Between Age and Size in 2003. The data is weighted. The age refers to the age of the firm in 2003. Figure 2: View largeDownload slide Cross Sectional Relationship Between Age and Size in 2003. The data is weighted. The age refers to the age of the firm in 2003. 5.3 The evolution of entrants (the 2000–2003 cohort) Our survey did not collect information on entry of firms after 2003, since only firms from the 2003 industrial census were revisited. We can therefore not make any claims about what happened to entrant firms founded after 2003. However, we can have a look at young firms in 2003 and see how they fared in the time period 2003–2013 to explore whether growth or selection is responsible for the evolution of the firm size distribution. Figure 3 shows the firm size distribution of firms founded between 2000 and 2003, both in 2003 and 2013. The 2003 firms are further split into the firms that survived until 2013 (the survivors) and the firms that did not survive until 2013 (the non-survivors). Table 11 shows the average firm size of these groups of firms. Table 11: Comparison of the Mean Firm Size of Firms Founded Between 2000 and 2003 and all Firms   Firms founded between 2000–2003  All firms in sample  (1) All firms in 2003  5.37 (0.30)  12.00 (0.66)  (2) Surviving firms in 2003  7.25 (0.71)  17.91 (1.35)  (2*) Non-surviving firms in 2003  4.36 (0.26)  7.40 (0.58)  (3) Surviving firms in 2013  7.05 (0.95)  15.91 (1.85)  t-Test of the difference in means  p value:  p value:   Selection: (1) versus (2)  0.016**  0.000***        (2) versus (2*)  0.000***  0.000***   Growth: (2) versus (3)  0.864  0.384  Kolmogorov–Smirnov equality-of-distribution test   Selection: (1) versus (2)  0.248  0.019**       (2) versus (2*)  0.033**  0.000***   Growth: (2) versus (3)  0.450  0.002***    Firms founded between 2000–2003  All firms in sample  (1) All firms in 2003  5.37 (0.30)  12.00 (0.66)  (2) Surviving firms in 2003  7.25 (0.71)  17.91 (1.35)  (2*) Non-surviving firms in 2003  4.36 (0.26)  7.40 (0.58)  (3) Surviving firms in 2013  7.05 (0.95)  15.91 (1.85)  t-Test of the difference in means  p value:  p value:   Selection: (1) versus (2)  0.016**  0.000***        (2) versus (2*)  0.000***  0.000***   Growth: (2) versus (3)  0.864  0.384  Kolmogorov–Smirnov equality-of-distribution test   Selection: (1) versus (2)  0.248  0.019**       (2) versus (2*)  0.033**  0.000***   Growth: (2) versus (3)  0.450  0.002***  Note. Firm size is defined as the number of ‘persons engaged’ and includes apprentices. The p value is reported for a t-test of the difference between means. Figure 3: View largeDownload slide Selection or Decline? The weighted firm size distribution of the 2003 full distribution, the 2003 ‘destined to survive’ firms, the 2003 ‘destined to exit’ firms and the 2013 surviving firms. Only firms founded between 2000 and 2003 are included. Note. 2003 sample weights used and adjusted for refusals in 2013. Figure 3: View largeDownload slide Selection or Decline? The weighted firm size distribution of the 2003 full distribution, the 2003 ‘destined to survive’ firms, the 2003 ‘destined to exit’ firms and the 2013 surviving firms. Only firms founded between 2000 and 2003 are included. Note. 2003 sample weights used and adjusted for refusals in 2013. We can use Figure 3 to replicate the test for whether growth or selection drives the firm size distribution that was suggested and tested by Cabral and Mata (2003) for Portugal, and later implemented in Ghana by Sandefur (2010). This test involves comparing (1) the firm size distribution at the initial measure point with (2) the firm size distribution of firms ‘destined to survive’ at the initial measure point (i.e., excluding firms that exited) as well as (3) the firm size distribution of the surviving firms at the final measure point. By comparing these distributions, we can attribute changes in the firm size distribution within a cohort to growth and selection. Comparing distribution (1) with distribution (2) shows the role of selection, as both distributions relate to the same (initial) point in time, but to a different group of firms: distribution (1) includes both surviving and exiting firms, while distribution (2) only includes the firms that survived up to the final measure point. Comparing distribution (2) with distribution (3) shows the role of firm growth, as both distributions use the same group of firms that did not exit in the measured time period. Cabral and Mata (2003) found strong differences between distributions (2) and (3), but not between distributions (1) and (2), indicating that growth played a more important role than selection in the evolution of the Portuguese firm size distribution. Sandefur (2010) found the opposite pattern in Ghana: strong differences between (1) and (2), but little difference between (2) and (3), indicating a strong role of selection. However, the difference in the distributions in Figure 3 is not as clear as in the Portuguese case of Cabral and Mata (2003) or as in the previous Ghanaian study of Sandefur (2010). Selection. The 2003 distribution of firms that survived is very similar shaped to the 2003 distribution of all firms (including non-survivors). A Kolmogorov–Smirnov equality-of-distribution test, using weights, does not reject the null hypothesis that the distributions are the same (the p value is 0.248).6 However, the means of these distributions are significantly different. In Table 11 we can see that, in 2003, the surviving firms had on average 1.88 more workers than all firms together, and on average 2.89 workers more than firms that did not survive. These differences are statistically significant at respectively a 5% and 1% level of significance, but are much smaller than the differences found by Sandefur.7 Firm growth. From Figure 3 we see that the 2003 distribution of surviving firms is very similarly shaped to the 2013 distribution of surviving firms. The test of the differences shows a similar picture: the average firm size of the young cohort that survived was 7.25 in 2003 and 7.05 in 2013. This decline is not statistically significant and reveals that on average, firms stagnated. Unlike Cabral and Mata (2003) we find no strong evidence for growth driving the evolution of the firm size distribution. But unlike Sandefur (2010) we find little evidence of selection driving the evolution of the firm size distribution. Regional differences. In Section 4 and in Section 6 we saw strong regional differences between Accra and Kumasi: Accra firms were more likely to exit and Kumasi firms were less likely to grow. We see similar patterns when comparing the firm size distributions. Figure 4 shows the evolution of the firm size distribution for young firms for both Accra and Kumasi. In Kumasi we see little effect of differential selection, as the 2013 firm size distribution of surviving firms resembles the 2003 overall firm size distribution quite closely, but we see a strong pattern of decline: the firm size distribution of surviving firms has shifted to the left in 2013 compared to 2003. In Accra we find a stronger pattern of differential selection, but less evidence for a decline in firm growth: in fact, firms in Accra grew on average by 0.45 worker, even though this change is not statistically significant (the p value is 0.825). Figure 4: View largeDownload slide The Weighted Firm Size Distribution of the 2003 Full Distribution, the 2003 ‘Destined to Survive’ Firms and the 2013 Surviving Firms, for both Accra and Kumasi. Only firms founded between 2000 and 2003 are included. Note. The 2003 sample weights were used (adjusted for refusals). Figure 4: View largeDownload slide The Weighted Firm Size Distribution of the 2003 Full Distribution, the 2003 ‘Destined to Survive’ Firms and the 2013 Surviving Firms, for both Accra and Kumasi. Only firms founded between 2000 and 2003 are included. Note. The 2003 sample weights were used (adjusted for refusals). Compared to Sandefur’s study we find only a small role of selection in driving the evolution of the firm size distribution. Three factors could explain these differences. First of all, Sandefur’s analysis included all firms, unlike Cabral and Mata (2003) who only included firms from one particular cohort. This was done because of the lack of availability of age information in Sandefur’s data. Including all firms instead of a cohort will make the role of selection look stronger. This is because older firms tend to be larger and more likely to survive, as our regression in Table 5 showed, while younger firms are more likely to exit and tend to be smaller, as can be seen in Figure 2. The old large firms who are likely to be survivors will therefore dominate the distribution of survivors. When we repeat our analysis including all firms, we see that the selection effects become stronger and significant in all cases (see the right column of Table 11). Second, Sandefur relied on matching firms from the 1987 census to the 2003 census by ISIC code, region and firm name. This procedure most likely led to an overrepresentation of larger firms. Only 13% of the 2003 firms who said they were operating in 1987 were matched, and Sandeufur (2010: 7n8) noted that ‘average firm size is somewhat larger for those I was able to match’ but did not provide any further details. This means that Sandefur’s matching procedure sets up a matched sample of 1987 firms that survived to 2003, in which larger firms are overrepresented, for comparison with the full sample of firms in 1987. This does not affect his result on the unimportance of within-firm growth, but does overstate his result that selection is driving the firm size distribution. Third, Sandefur’s panel dealt with a different time period (1987–2003) than ours (2003–2013). The time period between 1987 and 2003 was characterised by substantial changes in policy in Ghana, including a structural adjustment programme and trade liberalisation. It is possible that the introduction of these policies made selection more important than in the more recent period that we studied. Cabral and Mata focus on either growth or selection as explanations for the positive age–size correlation. We find that they only played a small role. What does then explain the positive age–size correlation shown in Figure 2? Changes in the composition of entering firms is the most likely explanation. Sandefur showed that between 1987 and 2003 the total number of manufacturing firms increased threefold and the average firm size halved. Unfortunately we cannot investigate this further using our 2003–2013 panel because we did not sample new entrants. However, our analysis of the most recent entrants in our sample, the 2000–2003 cohort, is consistent with this conclusion. This suggests that studying entry is key to understanding the evolution of the firm size distribution in Ghana. 5.4 Changes in aggregate employment How did the patterns in growth and exit that we have described affect employment for the firms sampled? Table 12 gives the levels of aggregate employment in each sector, both in 2003 and 2013. The table shows how important large firms are in providing employment: in 2003 firms with more than 75 employees provided almost half of the total employment measured. We can see that aggregate weighted employment of the Ghanaian manufacturing firms sampled in 2003 (the last row of the table) has decreased from 134,863 in 2003 (column 1) to 74,319 in 2013 (column 2). This is a decrease of 45%. Part of this decrease has come from firms that exited by 2013 (column 3); these firms employed an estimated 20,643 workers in 2003. Another 24,107 workers were employed in 2003 in firms that we were not able to trace in 2013 (column 4). As mentioned earlier, due to data limitations, we cannot separate between paid apprentices and regular workers for the 2013 figures of a large group of firms. The employment figures therefore include apprentices. Table 12: Aggregate Weighted Employment in 2003 and 2013   Total 2003 employment  Total 2013 employment in surviving firms  2003 Employment in exiting firms  2003 Employment untraced firms  2003 Employment in surviving firms  Sector   Food & beverages  19,735  13,283  3,083  1,484  15,168   Textiles, garments & footwear  35,272  13,795  5,010  11,706  18,556   Wood & furniture  29,119  13,059  6,034  6,591  16,494   Machinery & metal  15,722  11,618  1,774  2,270  11,678   Other manufacturing  35,015  22,564  4,742  2,056  28,217  Size   Small (0–9 workers)  37,060  15,446  7,247  13,775  16,038   Medium (10–74 workers)  32,335  19,725  7,195  4,768  20,373   Large (more than 75)  65,468  39,148  6,201  5,565  53,702  Total  134,863  74,319  20,643  24,107  90,112    Total 2003 employment  Total 2013 employment in surviving firms  2003 Employment in exiting firms  2003 Employment untraced firms  2003 Employment in surviving firms  Sector   Food & beverages  19,735  13,283  3,083  1,484  15,168   Textiles, garments & footwear  35,272  13,795  5,010  11,706  18,556   Wood & furniture  29,119  13,059  6,034  6,591  16,494   Machinery & metal  15,722  11,618  1,774  2,270  11,678   Other manufacturing  35,015  22,564  4,742  2,056  28,217  Size   Small (0–9 workers)  37,060  15,446  7,247  13,775  16,038   Medium (10–74 workers)  32,335  19,725  7,195  4,768  20,373   Large (more than 75)  65,468  39,148  6,201  5,565  53,702  Total  134,863  74,319  20,643  24,107  90,112  Note. Employment here is measured as the number of ‘persons engaged’ with the firm, and includes apprentices. The 2013 weights are adjusted only for refusals by continuing firms. If we only consider the firms that survived, we also see that the total employment in these firms decreased by around 18%, which is 1.8% annually. Few studies report total employment changes for surviving firms, but a couple of studies report changes in employment averaged by firm. In the UK, Bryson and Nurmi (2011) showed that surviving firms shrunk on average by two per cent between 1998 and 2004. A large study on 85 developing countries conducted by Aterido et al. (2011) found positive employment growth in surviving firms, unlike the negative employment growth found in our sample.8 Many other studies focusing on job destruction do not only incorporate figures from surviving firms but also job losses due to firm exit. Davis and Haltiwanger (1999) summarize a large number of studies showing that in the United States, as well as in a wide range of OECD countries, about one in ten jobs in manufacturing disappears every year. However, for most countries, the loss in jobs is compensated by new jobs creation in other manufacturing firms.9 Mazumdar et al. (2001) used the RPED surveys conducted in several African countries in the early 1990s to investigate changes in total employment during the period from 1982 to 1992. The authors found that total employment in Ghanaian firms increased by about 30% over this period for firms that were alive when the RPED surveys were conducted. Table 12 also shows that the declines in total employment in surviving firms in Ghana between 2003 and 2013 were the largest in textiles and garments and wood and furniture making and the smallest in machinery and metals. Again, size matters: the decline was the highest in small firms and the lowest in medium-sized firms. Table 13 shows the declines amongst different kinds of workers. Declines in apprentices engaged were very high, around 80%, compared to only a 33% decline in paid production workers and a 31% decline in paid other workers. Declines were similar for men and women. Table 13: Aggregate Employment by Worker Category and Gender   Employment  2003  2013  Worker category   Apprentices*  35,276  7,016   Production workers  73,840  48,845   Other workers  19,893  15,564  Gender   Male  99,085  52,047   Female  34,369  17,501    Employment  2003  2013  Worker category   Apprentices*  35,276  7,016   Production workers  73,840  48,845   Other workers  19,893  15,564  Gender   Male  99,085  52,047   Female  34,369  17,501  Note. The 2013 weights are adjusted only for refusals by surviving firms. Firms report total persons engaged and are then asked about each kind of worker—thus totals do not add to exactly the totals reported in Table 12. *The 2003 figure on apprentices include paid apprentices. For the 2013 figure, paid apprentices are classified as production workers and not as apprentices. There are two potential complicating factors in our measurement of apprentices over time, which we discuss more extensively in the Appendix. First of all, the 2013 survey only included unpaid apprentices whereas the 2003 survey most likely included both paid and unpaid apprentices in the apprentice category.10 The exclusion of paid apprentices in the 2013 figures means that the decline is most likely overstated. Data from representative household surveys (the Ghana Living Standards Surveys) suggests that between 18% and 40% of the apprentices are paid. Using these shares, the 2013 figures including paid apprentices would be between 8556 and 11,693, corresponding to a decrease in apprentices of between 67% and 75% compared to the 2003 figure. Similarly, the classification of paid apprentices as production workers in 2013 means that the decline in non-apprentice production workers was most likely higher, between 36% and 40%. The decline in total employment excluding apprentices (but including paid other workers) is most likely between 33% and 36% (see also the Appendix). Second, apprenticeships are concentrated in younger and smaller firms and last only for a few years. The 2003 census suggests that, for Ghana as a whole, of the 75,319 apprentices 24,235 were working in firms older than 10 years. By construction, our 2013 data only include firms founded before 2003 and all firms are therefore older than 10 years. If the pattern of mainly young firms using apprentices as seen in the 2003 census continued between 2003 and 2013, we expect our apprentice figure in 2013 to be lower, simply because the firms in 2013 are older. An estimation of this impact is made in the Appendix. To sum up this section, our dataset only covers firms that were operating in 2003. Therefore, it should be emphasised that we cannot conclude that total employment in Ghanaian manufacturing declined between 2003 and 2013 since we do not have data on firms that entered after 2003 and the amount of employment in these new firms. We are thus only able to say that total employment in the weighted sample of firms decreased by around 45% between 2003 and 2013, and that this decline was larger for apprentices, although this is a result of apprenticeships being concentrated in young firms and the different definitions of apprentices used in 2003 and 2013. Average firm size actually increased slightly due to the exit of small firms: the average firm size was 12 in 2003 and 16 for those surviving in 2013, which is mainly caused by the differential exit rates of small and large firms. 6. Conclusion The manufacturing sector has been seen as a potential engine of growth and employment in the Ghanaian economy. But our research has shown that Ghanaian manufacturing firms that existed in 2003 performed poorly over the 10 years between 2003 and 2013, a continuation of poor performance that has been documented for the 1990s and early 2000s in other firm-level research. Around 21% of firms exited between 2003 and 2013 while another 22% were untraced. If the untraced firms are assumed to have exited then these figures are similar to those estimated in Ghana over the 5 years between 1993 and 1998 by Söderbom et al. (2006). Firms in Accra, young firms and small firms were more likely to have exited. Exploring the reasons for exit amongst those owners or managers of firms who could be found suggested that small firms were more likely to exit due to personal circumstances of the owner while the most cited reason for exit in large firms was increasing costs. We also explored the importance of selection in explaining the evolution of the firm size distribution in Ghana. Using the simple graphical test suggested by Cabral and Mata (2003) we found that selection only played a small role, contradicting earlier work by Sandefur (2010) who used an inappropriate sample due to data limitations. But unlike Cabral and Mata (2003) we also find little role for within-firm growth in explaining the evolution of the firm size distribution. Firm entry is likely to be a key factor driving this evolution, and would seem to be a crucial area for future research. Broadening our analysis to surviving firms we have shown that only about 35% of the surviving firms that were successfully interviewed grew their employment by more than 5%, while 54% shrunk by more than 5%. Aggregate weighted employment (including apprentices) fell by 45%, from 134,863 in 2003 to 74,319 in 2013, an estimate that includes adjustments for the non-response of surviving firms. Due to an inconsistent definition of apprenticeships used across the two waves, we cannot separate paid apprentices from other employees for the 2013 figures of a substantial number of firms. This means that we cannot provide a precise figure for the drop in employment excluding apprentices. However, if we make assumptions for the purpose of establishing bounds on the share of paid apprenticeships in total apprenticeships, the decrease in employment excluding apprentices is likely between 33 and 36% (see the Appendix for details). We cannot estimate growth in total manufacturing employment in Ghana without surveying new firms, and creative destruction is one possible positive interpretation of our result. But our work showing a lack of growth of young firms in 2003 up until 2013 does not paint a positive picture of the state of manufacturing in Ghana. Supplementary material Supplementary material is available at Journal of African Economies online. Acknowledgements We thank Marcel Fafchamps, Francis Teal and the anonymous reviewer for helpful comments and Sofia Monteiro and Václav Těhle for excellent research assistance. We thank Moses Awoonor-Williams for invaluable assistance with the fieldwork and the Ghanaian Statistical Service (GSO) for their cooperation. Furthermore, we would like to thank our team of enumerators for their dedication to this project. Footnotes 1 Unlike some firm surveys conducted in other countries (e.g., India), the only firms to be excluded from the census were household based enterprises that did not have a sign indicating that a firm was operating in the household. There was no explicit size based criterion for inclusion. 2 For 14.2% of the firms where this exit questionnaire was conducted, we interviewed the former owner or manager. In 23.6% of the cases we interviewed a former worker or relative and for the remaining firms (62.3%) we interviewed a neighbour. 3 Aga and Francis (2015) calculate annualised rates that are not compounded so the numbers reported in their paper are different to the ones we have calculated and reported using their results. 4 Firm size is measured as the number of persons engaged, and includes both paid and unpaid workers. This is the same definition as used by Sandefur (2010) for his analysis of firm growth and selection between 1989 and 2003. 5 As Hsieh and Olken (2014) argue, a ‘missing middle’ is an exaggeration, as there are indeed medium-sized firms operating and we do not see strong signs of bimodality. In a reply, Tybout (2014) argues that the term ‘missing middle’ does not relate to a bimodality in the firm size distribution, but to an underrepresentation compared to an ‘undistorted’ distribution. Theoretically it has been argued that this ‘undistorted distribution’ should resemble a Pareto distribution (see e.g., Luttmer, 2007). 6 Note that we are comparing the 2003 full distribution with the 2003 distribution of surviving firms, of firms founded between 2000 and 2003. The 2003 full distribution includes both the surviving and non-surviving firms. 7 Note that our methodology differs from the methodology used by Sandefur (2010). Sandefur (2010) was limited in the number of firms he could match: only 13% of the firms from 2003 that claimed to have been in existence before 1987 were matched with the 1987 observations of these firms. For his analysis he therefore used all firms, since his sample was too small to use only new firms. We follow the methodology of Cabral and Mata (2003) who only used a cohort of new firms for their analysis. For our figures we use the cohort of firms founded between 2000 and 2003 (i.e., at most 4 years old). If we follow Sandefur (2010) and use all firms, we find stronger differences, as can be seen in the right part of Table 11. 8 Aterido et al. (2011) report employment growth as the 3-year change in employment divided by the average employment in the first year and the third year, for 85 developing countries based on the World Business Environment Survey (WBES). By construction this measure is bounded between −2.0 and + 2.0 and therefore less dominated by outlying observations than a simple average of employment change rates. For their entire sample of firms this figure is 0.117 (taken over 3 years). If we use the same calculation method for our sample, the equivalent figure (taken over 11 years) is −0.229. 9 These studies do not only include job losses due to exits, also include job losses due to exits of young firms that entered after the first measurement point. Our sample only includes firms operating in 2003, which means that firms that entered after 2003 but exited before 2013 are not reported by us. However, such firms are included in the job destruction rates discussed by Davis and Haltiwanger (1999). The inclusion of these young firms exiting early will likely lead to higher job loss figures, because young firms are more likely to exit than older firms. 10 The 2003 Phase I questionnaire, which was used to calculate the number of apprentices, does not specify whether paid apprentices should be included as apprentices or as production workers. In the 2003 Phase II questionnaire and in the 2013 follow-up survey paid apprentices are classified as production workers. A comparison of 2003 Phase I and 2003 Phase II data for the limited number of firms who were interviewed in both phases shows that the number of apprentices reported was higher in the Phase I survey, which suggests that paid apprentices were classified as apprentices. See also the Appendix for a further discussion. References Aga G. A., Francis D. C. ( 2015). As the Market Churns: Estimates of Firm Exit and Job Loss Using the World Bank’s Enterprise Surveys, World Bank Policy Research Working Paper 7218. Aterido R., Hallward-Driemeier M. ( 2010). The Impact of the Investment Climate on Employment Growth. Does Sub-Saharan Africa Mirror Other Low-Income Regions? World Bank Policy Research Working Paper 5218. The World Bank. Aterido R., Hallward-Driemeier M., Pagés C. ( 2011) ‘ Big Constraints to Small Firms’ Growth? Business Environment and Employment Growth Across Firms’, Economic Development and Cultural Change , 59 ( 3): 609– 47. Google Scholar CrossRef Search ADS   Aw B., Chen X., Roberts M. J. ( 2001) ‘ Firm-level Evidence on Productivity Differentials and Turnover in Taiwanese Manufacturing’, Journal of Development Economics , 66: 51– 86. Google Scholar CrossRef Search ADS   Bloom N., Lemos R., Sadun R., Scur D., Van Reenen J. ( 2014) ‘ JEEA-FBBVA Lecture 2013: The New Empirical Economics of Management’, Journal of the European Economic Association , 12: 835– 76. doi:10.1111/jeea.12094. Google Scholar CrossRef Search ADS   Bloom N., Van Reenen J. ( 2010) ‘ Why Do Management Practices Differ across Firms and Countries?’, Journal of Economic Perspectives , 24 ( 1): 203– 24. Google Scholar CrossRef Search ADS   Bryson A., Nurmi S. ( 2011) ‘ Private Sector Employment Growth, 1998−2004: A Panel Analysis of British Workplaces’, Cambridge Journal of Economics , 35 ( 1): 85– 104. Google Scholar CrossRef Search ADS   Cabral L. M. B., Mata J. ( 2003) ‘ On the Evolution of the Firm Size Distribution: Facts and Theory’, American Economic Review , 93 ( 4): 1075– 90. Google Scholar CrossRef Search ADS   Davis S. J., Haltiwanger J ( 1992) ‘ Gross Job Creation, Gross Job Destruction, and Employment Reallocation’, Quarterly Journal of Economics , 107 ( 3): 819– 63. Google Scholar CrossRef Search ADS   Davis S. J., Haltiwanger J ( 1999) ‘Chapter 41 – Gross job flows’, in Ashenfelter O., Card D. (eds), Handbook of Labor Economics , Vol. 3. Elsevier, Amsterdam. Frazer G. ( 2005) ‘ Which Firms Die? A Look at Manufacturing Firm Exit in Ghana’, Economic Development and Cultural Change , 53 ( 3): 585– 617. Google Scholar CrossRef Search ADS   Hallward-Driemeier M. ( 2009). Who survives? The Impact of Corruption, Competition and Property Rights across Firms. World Bank Policy Research Working Paper 5084. Hsieh C.-T., Klenow P. ( 2014) ‘ The Life Cycle of Plants in India and Mexico’, Quarterly Journal of Economics , 129 ( 3): 1035– 84. Google Scholar CrossRef Search ADS   Hsieh C.-T., Olken B. A. ( 2014) ‘ The Missing ‘Missing Middle’, Journal of Economic Perspectives , 28 ( 3): 89– 108. doi:10.1257/jep.28.3.89. Google Scholar CrossRef Search ADS   Li Y., Rama M. ( 2015) ‘ Firm Dynamics, Productivity Growth, and Job Creation in Developing Countries: The Role of Micro- and Small Enterprises’, World Bank Research Observer , 30 ( 1): 3– 38. Google Scholar CrossRef Search ADS   Liedholm C., McPherson M., Chuta E. ( 1994) ‘ Small Enterprise Employment Growth in Rural Africa’, American Journal of Agricultural Economics , 76: 1177– 82. Google Scholar CrossRef Search ADS   Luttmer E. ( 2007) ‘ Selection, Growth, and the Size Distribution of Firms’, The Quarterly Journal of Economics , 122 ( 3): 1103– 44. doi:10.1162/qjec.122.3.1103. Google Scholar CrossRef Search ADS   Mazumdar D., Mazaheri A., Biggs T., Ramachandran V., Shah M. ( 2001). The Manufacturing Sector in Sub-Saharan Africa: An Analysis Based on Firm Surveys in Seven Countries. RPED Discussion Paper Series. Melitz M. ( 2003) ‘ The Impact of Trade on Intra-industry Reallocations and Aggregate Industry Productivity’, Econometrica: Journal of the Econometric Society , 71 ( 6): 1695– 1725. doi:10.1111/1468-0262.00467. Google Scholar CrossRef Search ADS   Mitra P., Muravyev A., Schaffer M. ( 2014) ‘ Labor Reallocation and Firm Growth: Benchmarking Transition Countries Against Mature Market Economies’, IZA Journal of Labor & Development , 3 ( 1): 1– 22. Google Scholar CrossRef Search ADS   Sandefur J. ( 2010). On the Evolution of the Firm Size Distribution in an African Economy. CSAE Working Paper, (5). Shiferaw A., Bedi A. ( 2013) ‘ The Dynamics of Job Creation and Job Destruction in an African Economy: Evidence from Ethiopia’, Journal of African Economies , 22 ( 5): 651– 92. Google Scholar CrossRef Search ADS   Söderbom M., Teal F., Harding A. ( 2006) ‘ The Determinants of Survival among African Manufacturing Firms’, Economic Development and Cultural Change , 54 ( 3): 533– 55. Google Scholar CrossRef Search ADS   Sutton J., Kpentey B. ( 2012) An Enterprise Map of Ghana . London: International Growth Centre. Teal F., Habyarimana J., Thiam P., Turner G. ( 2006) ‘Ghana: an analysis of firm productivity’, in Regional Program on Enterprise Development . Washington DC: The World Bank, pp. 1– 72. Teal F. ( 2016). Firm Size, Employment and Value-added in African Manufacturing Firms: Why Ghana needs its 1%. CSAE working Paper WPS/2016-07. Tybout J. R. ( 2000) ‘ Manufacturing Firms in Developing Countries: How Well Do They Do, and Why?’, Journal of Economic Literature . doi:10.1257/jel.38.1.11. Tybout J. ( 2014) ‘ The Missing Middle, Revisited’, Journal of Economic Perspectives , 28 ( 4): 235– 36. Google Scholar CrossRef Search ADS   Van Biesebroeck J. ( 2005) ‘ Firm Size Matters: Growth and Productivity Growth in African Manufacturing’, Economic Development and Cultural Change , 53 ( 3): 545– 83. doi:10.1086/426407. Google Scholar CrossRef Search ADS   World Bank ( 2013) World Development Report 2013: Jobs . Washington, DC: World Bank Publications. © The Author 2017. Published by Oxford University Press on behalf of the Centre for the Study of African Economies, all rights reserved. For Permissions, please email: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of African Economies Oxford University Press

Firm Survival and Change in Ghana, 2003–2013

Loading next page...
 
/lp/ou_press/firm-survival-and-change-in-ghana-2003-2013-7SwNRpU6VF
Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press on behalf of the Centre for the Study of African Economies, all rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
0963-8024
eISSN
1464-3723
D.O.I.
10.1093/jae/ejx023
Publisher site
See Article on Publisher Site

Abstract

Abstract This paper explores the determinants of firm survival in Ghanaian manufacturing and the contributions of growth and selection to the evolution of the firm size distribution. For this analysis we created a two-wave panel spanning 10 years to study exit, growth and decline, by re-surveying 1000 firms randomly selected from the 2003 National Industrial Census. We find strong differences in exit patterns by region and firm size. Former owners and managers commonly cite personal circumstances as the reason for exit in the case of small firms and increasing costs in the case of large firms. We show that both growth and selection played only a small role in the evolution of the firm size distribution, contradicting earlier work on Ghana. Overall, the picture we paint of manufacturing in Ghana is not a positive one: total employment by firms operating before 2003 decreased from 134,863 in 2003 to 74,319 in 2013, although we cannot explore to what extent new employment in firms that entered after 2003—who were not surveyed—compensated for this decline. 1. Introduction The Ghanaian economy has been characterised by important changes over the last few decades: high levels of GDP growth, an IMF reform process that led to many changes in policy, increases in consumption expenditure, the discovery and production of oil and a rise of the service industry. More recently the contribution of the manufacturing sector to GDP has been declining. Research using firm-level data has similarly documented the lacklustre performance of manufacturing firms during the 1990s and early 2000s (Teal et al. 2006; Sandefur, 2010). In this paper we explore whether this trend continued in the last 10 years. To do this we use a follow-up survey conducted in 2013 on manufacturing firms first interviewed as part of the 2003 National Industrial Census. This allows us to create a two-wave panel dataset of 1000 manufacturing firms. The picture our analysis paints is not a positive one—the firms we surveyed have generally performed poorly over the last 10 years, with high rates of exit and shrinkage of surviving firms. We also use the data to explore whether the evolution of the firm size distribution in Ghana is explained by growth, selection or entry (Cabral and Mata, 2003; Sandefur, 2010) and we find no role for growth and only a small role for selection in explaining the evolution of the firm size distribution in Ghana. This suggests that entry may be important but our work cannot speak directly to the importance of entry since we did not collect data on new firms that were born between 2003 and 2013. Previous research has suggested, however, that entry may be an important and under-researched contributor to the evolution of the firm size distribution in Ghana (Sandefur, 2010). This paper makes a contribution to the literature by describing patterns of Ghanaian firm growth and survival between 2003 and 2013, and the factors that are correlated with firm survival. In particular, this paper contributes to the literature on the evolution of the firm size distribution (Cabral and Mata, 2003; Luttmer, 2007), and provides some insight as to whether Ghana is facing a ‘missing middle’ (Tybout, 2000; Hsieh and Olken, 2014). A recent literature has emphasised the role of management in firm survival and growth (Bloom and Van Reenen, 2010; Bloom et al. 2014). Our paper provides some evidence that ownership and management matters: personal circumstances of owners and managers can be crucial for the survival of firms, in particular small ones. The paper is structured as follows: Section 2 provides some background on the Ghanaian economic environment and discusses some earlier studies on Ghanaian manufacturing. Section 3 describes the survey and provides some descriptive statistics. Section 4 presents the evidence on firm exit and survival and explores the self-reported reasons for exit. Section 5 focuses on firm growth and decline and considers the role of selection and growth on the overall firm size distribution. Finally, Section 6 concludes. 2. The economic environment The Ghanaian economy has recently exhibited high growth: according World Bank figures, between 2003 and 2012 growth was 7.5% per annum on average. During this time the composition of the economy has also seen some considerable changes: the contribution to the gross domestic product by the service industry has been growing significantly, with an average annual growth rate of 12.9% between 2003 and 2012. Services constituted 50.0% of value added in 2012, while in 1990 this was only 38.1%. Industrial output has also been growing considerably, but this growth has mainly been achieved in other industrial sectors than the manufacturing sector. The manufacturing sector is estimated to have been growing at 3.3%, while other industrial sectors, such as mining, water production and construction have grown by 9.1% on average between 2003 and 2013. This means that the relative share of the contribution of the manufacturing to GDP has declined, from 9.8% in 1990 to 6.9% in 2012 (see Table 1). Most of this decline happened after 2007. Household and government consumption has risen by 5.6% on average in the same time, indicating that Ghanaian manufacturing has profited less from this increase than other sectors. Filling out the picture from macrodata, previous work on microdata from manufacturing firms has shown weak performance, despite several regulatory changes, such as trade liberalisation and exchange rate reforms, which should have made it easier to compete (see e.g., Sutton and Kpentey, 2012, for a discussion of sector-specific policy measures). Teal et al. (2006) describe results from the RPED firm surveys indicating that output by manufacturing firms fell between 2000 and 2003 and Teal (2016) uses the 1987 and 2003 NICs to show that by 2003 a much larger proportion of the manufacturing labour force was working in low productivity firms than in 1987. Table 1: Share of Sectors in Value Added (as a Percentage of GDP)   1990  1995  2000  2005  2010  2012  Agriculture  45.1  42.7  39.4  40.9  29.8  22.7  Manufacturing  9.8  10.3  10.1  9.5  6.8  6.9  Other industry  7.0  16.5  18.3  18.0  12.3  20.5  Services  38.1  30.6  32.2  31.6  51.1  50.0  Total  100.0  100.0  100.0  100.0  100.0  100.0    1990  1995  2000  2005  2010  2012  Agriculture  45.1  42.7  39.4  40.9  29.8  22.7  Manufacturing  9.8  10.3  10.1  9.5  6.8  6.9  Other industry  7.0  16.5  18.3  18.0  12.3  20.5  Services  38.1  30.6  32.2  31.6  51.1  50.0  Total  100.0  100.0  100.0  100.0  100.0  100.0  Source: World Bank Development Indicators. One important question when interrogating the lacklustre performance of manufacturing is whether competition is driving out low productivity firms or whether these firms can continue to operate. This question can only be answered using panel data. Frazer (2005) uses the RPED firm panel to study the exit of Ghanaian manufacturing firms and how this relates to firm productivity. Frazer finds that low firm productivity is a good predictor of firm exits. The size and age of the firm also seem relevant in predicting whether a firm continues to operate or not: large firms and older firms are less likely to exit. The former can be explained by a simple model where firm growth is dependent on success: a less successful firm is less likely to grow, but also more likely to fail. An explanation for the latter might be that characteristics that helped to prevent exit in the past help prevent exits in the present as well, causing older firms to be more likely to survive. Trade models of firm selection, such as Melitz (2003) predict that exporting firms are less likely to exit, as predominantly successful firms are able to become exporters, but Frazer does not find evidence for this in Ghana. The importance of firm productivity on firm exits seems to challenge earlier studies done in sub-Saharan Africa, such as Liedholm et al. (1994), who emphasise personal circumstances playing a role in at least a quarter of exits, but who do not consider productivity due to a lack of data. Söderbom et al. (2006) also focus on the relationship between productivity and selection. They find, on the basis of firm panel surveys in Ghana and several other African countries, that efficiency matters more for the survival of larger firms. Being productive does not prevent small firms from going out of business. This might indicate the important role of other considerations, such as personal circumstances, in the case of small enterprises. We explore the correlates of firm survival and exit below. A second important question is how the firm size distribution evolves. Sandefur (2010) examines this question in Ghana using the 1988 and 2003 National Industrial Census microdata and finds that selection, rather than growth, drives the evolution of the firm size distribution. This differs from results from for example firm studies done in other countries, e.g., Portugal (Cabral and Mata, 2003). Although Sandefur’s paper is unpublished his result was discussed in the 2013 World Development Report on Jobs (World Bank, 2013) as well as in a recently published summary paper on firm dynamics in developing countries (Li and Rama, 2015). We take up this issue in Section 5, arguing that the result that selection and not growth drives the evolution of the firm size distribution is only partially correct. 3. Survey description In 2013 a total of 1,000 firms located in five locations (Accra, Tema, Kumasi, Sekondi-Takoradi and Cape Coast) were sampled from the 2003 Ghana National Industrial Census (NIC), conducted by the Ghanaian Statistical Service and covering all establishments engaged in manufacturing in Ghana.1 Stratification was based on firm size, firm age, region and sector, to ensure that firms with a wide range of characteristics were included. In total 135 strata were used (see Table 2 for the breakdown of the factors determining the stratification). To account for the diversity of firms, sampling probabilities were adjusted on the basis of the variance of firm size in each stratum. This led to oversampling of certain strata, while others were undersampled. As can be seen in Table 2, large and old firms have a much higher probability of being included in the final sample, while young and small firms have a lower probability of being included. Practically all large firms (with more than 75 employees) in the regions sampled were included in the final sample. Table 2: Main Summary Statistics of the Firms from the 2003 National Industrial Census, the Sampled Area and the Sample of the Survey   In 2003 National Industrial Census  Sampling area  Sampled  No.  %  No.  %  No.  %  Region   Greater Accra (incl. Tema)  6,654  25.1  6,655  59.2  579  57.9   Kumasi  3,374  12.8  3,374  30.0  304  30.4   Sekondi-Takoradi1  855  3.2  855  7.6  90  9.0   Cape Coast1  355  1.3  355  3.2  27  2.7   Other  15,237  57.6  –  –  –  –  Sector   Food & beverages  4,257  16.1  913  8.1  132  13.2   Textiles, garments & footwear  11,620  43.9  5,359  47.7  299  29.9   Wood & furniture  6,085  23.0  2,416  21.5  215  21.5   Machinery & metal  2,133  8.1  1,266  11.3  135  13.5   Other  2,381  9.9  1,284  11.4  219  21.9  Age   Founded 1999–2003  13,499  51.0  5,942  52.9  344  34.4   Founded 1989–1998  9,432  35.6  3,923  34.9  377  37.7   Founded before 1988  3,348  12.7  1,330  11.8  254  25.4   Unknown  197  0.7  43  0.4  25  2.5  Size   Small (0–9 workers)  22,375  84.5  9,394  84.6  386  38.6   Medium (10–74 workers)  3,733  14.1  1,619  14.4  392  39.2   Large (more than 75)  367  1.4  225  2.0  222  22.2    In 2003 National Industrial Census  Sampling area  Sampled  No.  %  No.  %  No.  %  Region   Greater Accra (incl. Tema)  6,654  25.1  6,655  59.2  579  57.9   Kumasi  3,374  12.8  3,374  30.0  304  30.4   Sekondi-Takoradi1  855  3.2  855  7.6  90  9.0   Cape Coast1  355  1.3  355  3.2  27  2.7   Other  15,237  57.6  –  –  –  –  Sector   Food & beverages  4,257  16.1  913  8.1  132  13.2   Textiles, garments & footwear  11,620  43.9  5,359  47.7  299  29.9   Wood & furniture  6,085  23.0  2,416  21.5  215  21.5   Machinery & metal  2,133  8.1  1,266  11.3  135  13.5   Other  2,381  9.9  1,284  11.4  219  21.9  Age   Founded 1999–2003  13,499  51.0  5,942  52.9  344  34.4   Founded 1989–1998  9,432  35.6  3,923  34.9  377  37.7   Founded before 1988  3,348  12.7  1,330  11.8  254  25.4   Unknown  197  0.7  43  0.4  25  2.5  Size   Small (0–9 workers)  22,375  84.5  9,394  84.6  386  38.6   Medium (10–74 workers)  3,733  14.1  1,619  14.4  392  39.2   Large (more than 75)  367  1.4  225  2.0  222  22.2  Source: Own calculations. 1Sekondi-Takoradi and Cape Coast were treated as one region in the stratification. The survey was conducted between August and November 2013. Attempts were made to interview each firm from the sample. If the firm was operating, a questionnaire similar to the 2003 National Industrial Census was conducted, asking mainly about employment and firm productivity. If a firm was no longer operating, enumerators attempted to find a former manager or representative of the firm and conduct a questionnaire with exit-specific questions. In case no former manager or representative could be found, a family member, former worker or neighbour was interviewed instead.2 In all cases where the main firm questionnaire was not undertaken, the enumerator was asked to record basic information on the firm, such as whether the firms was still operating or not, whether a firm sign was still present, and in case no interview was undertaken, what the reason was for this. The 2013 survey allowed the creation of a two-wave panel of 1,000 firms, some of which survived and some of which either died or were untraced. Table 3 shows the results of our attempts to trace 1,000 firms. Forty-five per cent of the firms were found and interviewed while another 12% were found and were operating but refused to participate in the survey. Twenty-one per cent of the firms had exited while no trace was found of 22% of the firms. This last group is likely to be mainly exits but could also include firms that moved (although the enumerators did try and trace firms that were known to have moved within the city in which they were located). Table 3 also shows that the survey was less successful in finding and interviewing firms in Accra, small and young firms. Large firms, those in the ‘Other’ sector (which includes a variety of sectors, such as utility companies, chemical companies and printing firms) and those in Accra and Takoradi were more likely to refuse. Table 3: Firm Status in 2013 by 2003 Characteristics   Found and interviewed  Exit  Untraced  Operating but refusal   All Firms  42.8  21.2  22.2  13.8  Region   Greater Accra  37.3  21.4  24.2  17.1   Kumasi  51.0  17.4  23.7  7.9   Sekondi-Takoradi  46.7  28.9  8.9  15.6   Cape Coast  55.6  33.3  7.4  3.7  Sector   Food & beverages  35.6  25.8  16.7  22.0   Textiles, garments & footwear  44.1  20.7  31.8  3.3   Wood & furniture  42.3  23.7  27.0  7.0   Machinery & metal  52.6  14.8  16.3  16.3   Other  39.7  20.5  11.4  28.3  Age   Founded 1999–2003  39.1  21.8  37  2.1   Founded 1989–1998  50.0  24.2  15.6  10.2   Founded before 1988  36.5  14.9  8.1  40.5  Size in 2003   Small (0–9 workers)  35.5  24.7  28.5  11.4   Medium (10–74 workers)  48.0  18.8  21.8  11.4   Large (more than 75)  45.7  19.7  13.8  20.9    Found and interviewed  Exit  Untraced  Operating but refusal   All Firms  42.8  21.2  22.2  13.8  Region   Greater Accra  37.3  21.4  24.2  17.1   Kumasi  51.0  17.4  23.7  7.9   Sekondi-Takoradi  46.7  28.9  8.9  15.6   Cape Coast  55.6  33.3  7.4  3.7  Sector   Food & beverages  35.6  25.8  16.7  22.0   Textiles, garments & footwear  44.1  20.7  31.8  3.3   Wood & furniture  42.3  23.7  27.0  7.0   Machinery & metal  52.6  14.8  16.3  16.3   Other  39.7  20.5  11.4  28.3  Age   Founded 1999–2003  39.1  21.8  37  2.1   Founded 1989–1998  50.0  24.2  15.6  10.2   Founded before 1988  36.5  14.9  8.1  40.5  Size in 2003   Small (0–9 workers)  35.5  24.7  28.5  11.4   Medium (10–74 workers)  48.0  18.8  21.8  11.4   Large (more than 75)  45.7  19.7  13.8  20.9  Note: Numbers reported in this table are percentages. Row percentages sum to 100%. The data are unweighted. Source: own calculations. 4. Firm survival and exit In this section we discuss survival and exit patterns between 2003 and 2013, and show how these differ between regions and sectors. Furthermore, we describe the reasons for exits, as reported by the respondents. 4.1 Patterns of survival and exit Table 4 shows correlates of two measures of firm exit: exit measure 1 excludes firms that were not found while exit measure 2 assumes that firms that were not found actually exited. Weights increase the contributions of small firms to any statistics since these firms were less likely to be sampled and large firms had a selection probability close to one. Thus, when weighted, exit rates over 10 years were between 33% (assuming all untraced firms did not exit) and 56% (assuming all untraced firms exited). This is equivalent to an annualised compound exit rate of between 3.9% and 7.9%. Aga and Francis (2015) used the World Bank’s Enterprise Survey panels from 47 countries, including Ghana, and found that annualised compound exit rates were generally between 3% and 5% per year,3 when using panels that were an average of 4 years apart. In the Enterprise Survey panel for Ghana between 68% and 77% of the (manufacturing and services) firms interviewed in 2007 had exited in 2013 depending on how exit was measured. The annualised compound exit rate was thus between 15% and 19%. This exit rate was the highest of any country in the sample and exit rates for the other African countries in the sample (DRC, Kenya, Tanzania, Uganda, Zambia, Rwanda) were much lower. The exit rate for Ghana using the 2003–2013 panel is thus slightly above the exit rates for the other African countries in which enterprise surveys have been undertaken but markedly lower than in the Ghanaian enterprise survey conducted by the World Bank. Table 4: Correlates of Firm Exit   Exit measure 1 (excluding firms not found)  Exit measure 2 (treating not found as exit)  No exit  Exit  No exit  Exit  Region   Greater Accra (incl. Tema)  61.1  38.9  36.7  63.3   Kumasi  77.5  22.5  53.6  46.4   Sekondi-Takoradi  67.7  32.3  55.5  44.5   Cape Coast  53.1  46.9  52.3  47.7  Sector   Food & beverages  61.6  38.4  44.6  55.4   Textiles, garments & footwear  63.6  36.4  38.6  61.4   Wood & furniture  66.4  33.6  39.8  60.2   Machinery & metal  79.8  20.2  64.8  35.2   Other  66.6  33.4  51.3  48.7  Age   Founded 1999–2003  59.9  40.1  36.7  63.3   Founded 1989–1998  75.2  24.8  51.3  48.7   Founded before 1988  68.1  31.9  52.8  47.2  Size in 2003   Small (0–9 workers)  64.8  35.2  40.1  59.9   Medium (10–74 workers)  71.6  28.4  60.1  39.9   Large (more than 75)  83.9  16.1  77.2  22.8  All firms  66.6  33.4  43.7  56.3    Exit measure 1 (excluding firms not found)  Exit measure 2 (treating not found as exit)  No exit  Exit  No exit  Exit  Region   Greater Accra (incl. Tema)  61.1  38.9  36.7  63.3   Kumasi  77.5  22.5  53.6  46.4   Sekondi-Takoradi  67.7  32.3  55.5  44.5   Cape Coast  53.1  46.9  52.3  47.7  Sector   Food & beverages  61.6  38.4  44.6  55.4   Textiles, garments & footwear  63.6  36.4  38.6  61.4   Wood & furniture  66.4  33.6  39.8  60.2   Machinery & metal  79.8  20.2  64.8  35.2   Other  66.6  33.4  51.3  48.7  Age   Founded 1999–2003  59.9  40.1  36.7  63.3   Founded 1989–1998  75.2  24.8  51.3  48.7   Founded before 1988  68.1  31.9  52.8  47.2  Size in 2003   Small (0–9 workers)  64.8  35.2  40.1  59.9   Medium (10–74 workers)  71.6  28.4  60.1  39.9   Large (more than 75)  83.9  16.1  77.2  22.8  All firms  66.6  33.4  43.7  56.3  Note: Exit measure 1 excludes firms not found from the analysis whereas exit measure 2 assumes firms not found exited. The data are weighted, and hence the figures are different to those reported in Table 3. Source: own calculations. Exit rates do not necessarily or sufficiently indicate poor performance. Creative destruction suggests that exit can be positive if less efficient firms are replaced by more efficient ones. Aw et al. (2001) find that in a period of rapid growth in Taiwan between 70% and 87% of manufacturing firms that were alive in 1981 had exited by 1991. Shiferaw and Bedi (2013) find annual exit rates in the annual Ethiopian manufacturing census of between 14% and 17% for the period 1997 and 2007, during which real sales of manufacturing firms tripled and employment increased by around 40%. However, it should be noted that firms dropping below 10 employees are excluded from future waves, which means that exit rates are likely overstated. The annual average exit rate of between 3.9% and 7.8% of Ghanaian firms is not out of line with exit rates in other countries. Davis and Haltiwanger (1992) report an annual exit rate of 8% in the United States. Hallward-Driemeier (2009) reports an annual exit rate of 7.6% for a large group of Eastern European and Central Asian countries, for the time period 2002–2009. However, as we show later our other findings using our Ghanaian panel data do not give reason for optimism about the growth of Ghanaian manufacturing. We now explore correlates of firm exit. Using the first measure of exit firms located in Accra and Cape Coast, smaller firms and younger firms are more likely to have exited between 2003 and 2013. By this measure around one-third of the firms from the 2003 sample had exited 10 years later. However if we assume, as in the second measure of exit variable, that firms that could not be traced also exited then around 56% of firms had exited after 10 years. Using this second measure also changes some results obtained using the first measure. For example Accra firms are unambiguously more likely to have exited than firms from the other regions while the smallest firms were much more likely not to be traced and therefore have much higher rates of exit when using the second measure. Table 5 reports the results of a simple linear probability model of exit. Using the measure of exit that excludes firms not traced column 1 shows that firms located in Kumasi, the largest firms, firms in the middle age category (5–14 years old in 2003) and machinery and metals firms were less likely to exit than other firms. Column 2 assumes that firms that were not traced exited. Accra now stands out as being the location most likely to be correlated with exit, suggesting that tracing firms was more of an issue there than in Takoradi or Cape Coast. Small firms are now much more likely to have exited than large or medium-sized firms, implying that small firms were more likely not to be traced. This differential exit rates for small and medium or large firms is a pattern that has been documented more widely in sub-Saharan Africa (see e.g., Van Biesebroeck, 2005). Table 5: A Linear Probability Model of Exit Dependent variable  Exit measure 1 (excluding firms not found)  Exit measure 2 (treating not found as exit)  Kumasi  −0.157***  −0.156***  (0.0523)  (0.0478)  Sekondi-Takoradi  −0.0486  −0.180**  (0.0830)  (0.0777)  Cape Coast  0.0495  −0.162  (0.134)  (0.124)  Textiles & garments  −0.0270  0.00707  (0.0868)  (0.0754)  Wood & furniture  −0.0216  0.0588  (0.0914)  (0.0800)  Machinery & metal  −0.166*  −0.203**  (0.0913)  (0.0870)  Other sector  −0.0479  −0.0788  (0.0963)  (0.0876)  Medium size firm (10–74)  −0.0257  −0.137***  (0.0424)  (0.0391)  Large size firm (75+)  −0.184***  −0.316***  (0.0526)  (0.0491)  Founded 1989–1998  −0.138***  −0.110**  (0.0530)  (0.0468)  Founded before 1988  −0.0577  −0.0762  (0.0708)  (0.0632)  Constant  0.500***  0.718***  (0.0861)  (0.0738)  Observations  778  1,000  R2  0.063  0.089  Dependent variable  Exit measure 1 (excluding firms not found)  Exit measure 2 (treating not found as exit)  Kumasi  −0.157***  −0.156***  (0.0523)  (0.0478)  Sekondi-Takoradi  −0.0486  −0.180**  (0.0830)  (0.0777)  Cape Coast  0.0495  −0.162  (0.134)  (0.124)  Textiles & garments  −0.0270  0.00707  (0.0868)  (0.0754)  Wood & furniture  −0.0216  0.0588  (0.0914)  (0.0800)  Machinery & metal  −0.166*  −0.203**  (0.0913)  (0.0870)  Other sector  −0.0479  −0.0788  (0.0963)  (0.0876)  Medium size firm (10–74)  −0.0257  −0.137***  (0.0424)  (0.0391)  Large size firm (75+)  −0.184***  −0.316***  (0.0526)  (0.0491)  Founded 1989–1998  −0.138***  −0.110**  (0.0530)  (0.0468)  Founded before 1988  −0.0577  −0.0762  (0.0708)  (0.0632)  Constant  0.500***  0.718***  (0.0861)  (0.0738)  Observations  778  1,000  R2  0.063  0.089  Note: Reference (omitted) categories are firms located in Accra, in the food and beverage sector, with 0–9 employees and younger than 4 years old in 2003. Standard errors in parentheses. Significance levels are: ***p < 0.01, **p < 0.05, *p < 0.1. Source: own calculations. 4.2 Self-reported reasons for firm exit Table 6 shows reasons for exit amongst the 191 of 212 exiting firms that provided a reason for exit (firms that were not found are not included in this analysis). Circumstances of the owner accounted for around a third of all reasons for exit given—such as illness, retirement, moving to another region or country, etc. Falling demand and loss of land, buildings or equipment each account for just under 20% of exits while around 20% responded that they did not know why the firm had exited (if the owner, a manager or a worker could not be traced a neighbour was asked). Table 6: Reasons Given for Exits by Respondents Reasons for exit  Weighted percentage  Illness or death of owner or manager  12.3  Set up other business, merged firm or found wage employment elsewhere  7.1  Owner moved  13.6  Falling demand  19.4  Increased competition from imports  3.6  Increased competition from local competition  1.5  Increased costs  4.0  Loss of building, land or equipment  16.7  Debt or credit problems  1.5  Managerial problems  0.2  Don’t know  19.7  Refused  0.6  Total  100.0  Reasons for exit  Weighted percentage  Illness or death of owner or manager  12.3  Set up other business, merged firm or found wage employment elsewhere  7.1  Owner moved  13.6  Falling demand  19.4  Increased competition from imports  3.6  Increased competition from local competition  1.5  Increased costs  4.0  Loss of building, land or equipment  16.7  Debt or credit problems  1.5  Managerial problems  0.2  Don’t know  19.7  Refused  0.6  Total  100.0  Note: The respondents were either former managers, owners or workers of the firm, or if they could not be found, a neighbour or family member. Source: answers from the exit questionnaire. Table 7 breaks down reasons for exit by firm size. Circumstances of the owner, decreasing demand and loss of land, buildings and equipment are more likely to be given as a reason for exit amongst small firms compared to large firms. Large firms were much more likely to have exited due to increased costs than small firms and this was the most common reason for exit amongst large firms. That small firms are more likely to exit due to the circumstances of the owner accords with the work of Liedholm et al. (1994), who found this in their survey of smaller firms (fewer than 50 employees) in several African countries. Teal et al. (2006) find that selection on efficiency is a more important determinant of exit amongst large firms than small firms. If rising costs imply lower efficiency then the fact that Table 5 shows a high fraction of large firms exiting due to increased costs suggests that our results accord with those of Teal et al. (2006). Table 7: Breakdown of Exit Reasons by Firm Size Reason for exit  0–9 employees  10–74 employees  75+ employees  All  Illness or death of owner or manager  12.3  12.8  3.9  12.3  Set up other business, merged firm or found wage employment elsewhere  7.9  2.9  3.9  7.1  Owner moved  14.4  9.9  3.9  13.6  Falling demand  20.1  16.5  3.9  19.4  Increased competition from imports  3.3  4.5  7.9  3.6  Increased competition from local competition  1.3  2.3  3.9  1.5  Increased costs  3.7  4.1  25.0  4.0  Loss of building, land or equipment  16.6  18.4  3.9  16.7  Debt or credit problems  1.3  2.9  3.9  1.5  Managerial problems  0.0  0.9  3.9  0.2  Don’t know  19.3  21.3  31.6  19.7  Refused  0.0  3.4  3.9  0.6  Total  100.0  100.0  100.0  100.0  Reason for exit  0–9 employees  10–74 employees  75+ employees  All  Illness or death of owner or manager  12.3  12.8  3.9  12.3  Set up other business, merged firm or found wage employment elsewhere  7.9  2.9  3.9  7.1  Owner moved  14.4  9.9  3.9  13.6  Falling demand  20.1  16.5  3.9  19.4  Increased competition from imports  3.3  4.5  7.9  3.6  Increased competition from local competition  1.3  2.3  3.9  1.5  Increased costs  3.7  4.1  25.0  4.0  Loss of building, land or equipment  16.6  18.4  3.9  16.7  Debt or credit problems  1.3  2.9  3.9  1.5  Managerial problems  0.0  0.9  3.9  0.2  Don’t know  19.3  21.3  31.6  19.7  Refused  0.0  3.4  3.9  0.6  Total  100.0  100.0  100.0  100.0  Note: The data have been weighted. 5. Firm growth and decline This section presents evidence on firm growth and decline and shows that there are regional and sectoral differences in the evolution of surviving firms. We then explore what determines the evolution of the firm size distribution. Finally, we consider the changes in aggregate employment by the firms in our sample and show that total employment in the firms in our sample dropped by 45%. Due to data limitations, we cannot distinguish between paid apprentices and regular workers for most of the firms. Our measures of employment therefore include apprentices. This might not be the most reliable measure for firm performance, given that apprenticeships are often not paid very well or paid at all and therefore apprentices are less likely to be laid off when the firms is not performing. In the Appendix we apply a number of corrections, which suggest that the drop in total employment excluding apprentices is likely to be between 33% and 36%. 5.1 Patterns of growth and decline Table 8 explores the patterns of growth and shrinking amongst the firms that survived and were traced. The first part of the table shows that 54.0% of the surviving firms shrank, and that 34.9% grew.4 If we include exits, refusals and firms not found about 14% of the firms in the 1000 firm sample grew, 21.6% shrank, 21.1% exited, 33.1% were not traced and 5.7% were operating but refused to be interviewed. Table 8: Firm Status For continuing firms  Number  Percentage  Weighted percentage  Shrunk  256  62.9  54.0  Same Size  31  7.6  11.1  Grew  120  29.5  34.9  Including exits and refusals   Not found  222  22.2  33.1   Exit  212  21.2  21.1   Refusal, but still operating  159  15.9  5.7   Shrunk  256  25.6  21.6   Same Size  31  3.1  4.5   Grew  120  12.0  14.0  For continuing firms  Number  Percentage  Weighted percentage  Shrunk  256  62.9  54.0  Same Size  31  7.6  11.1  Grew  120  29.5  34.9  Including exits and refusals   Not found  222  22.2  33.1   Exit  212  21.2  21.1   Refusal, but still operating  159  15.9  5.7   Shrunk  256  25.6  21.6   Same Size  31  3.1  4.5   Grew  120  12.0  14.0  Using data from the World Bank’s Enterprise Surveys, including Ghana and 30 other sub-Saharan African countries, as well as 74 countries in lower, lower middle and upper income countries in other regions, Aterido and Hallward-Driemeier (2010) reported that ‘more than half’ of incumbent firms in sub-Saharan Africa increased employment while 20% shrank. This suggests that employment growth for survivors in our Ghanaian sample is lower than the average for these thirty one sub-Saharan countries. Mitra et al. (2014) use the Business Environment and Enterprise Performance Surveys (BEEPS), part of the World Bank Enterprise Surveys, to explore firm growth in transition economies and compare this to firm growth patterns in mature European economies. They report the balance between growing and shrinking firms—the proportion of firms that grew minus the proportion that shrank—and ignore firms that stayed the same size. They find that in 2005 the proportion of firms that grew was 20–22 percentage points higher than the proportion of firms that shrank in CIS countries, 15 percentage points higher in SEE countries, 6 percentage points higher in the eight countries that joined the EU in 2004 and 16 percentage points higher in the PIGS countries (Portugal, Ireland, Greece and Spain). Again this suggests that Ghanaian firms had lower growth than low income CIS countries, although the differences in sampling, discussed above, imply that growth rates in the enterprise surveys would be lower if a similar method was used to calculate growth rates. Focusing on the surviving firms Table 9 explores correlates of growth and decline. Firms in the ‘Other’ sector (e.g., chemical products) were more likely to grow than those in other sectors. Textiles and Wood were the two worst performing sectors. Conditional on survival firms in Accra and Takoradi were more likely to grow than firms in Kumasi. Firms in Cape Coast performed terribly with only six per cent of firms reporting more employment than in 2003. Younger firms were more likely to grow than older firms while the smallest and largest firms were more likely to grow. Table 9: Changes in Firm Size Sector  Shrunk  Same Size  Grew  Total  Food & beverages  46.8  12.6  40.5  100.0  Textiles, garments & footwear  61.9  11.0  27.2  100.0  Wood & furniture  56.4  7.6  36.0  100.0  Machinery & metal  45.1  13.8  41.1  100.0  Other manufacturing  41.9  12.5  45.6  100.0  Location   Greater Accra  49.6  10.1  40.3  100.0   Kumasi  62.6  11.2  26.2  100.0   Takoradi  50.7  1.2  48.1  100.0   Cape Coast  36.3  57.8  5.9  100.0  Age group   Founded 1999–2003  51.2  13.2  35.6  100.0   Founded 1989–1998  51.9  10.0  38.2  100.0   Founded before 1988  68.3  8.3  23.4  100.0  Size group   Small (0–9)  48.4  13.2  38.4  100.0   Medium (10–74)  74.1  4.9  20.9  100.0   Large (75+)  58.4  1.9  39.7  100.0  Total  54.0  11.1  34.9  100.0  Sector  Shrunk  Same Size  Grew  Total  Food & beverages  46.8  12.6  40.5  100.0  Textiles, garments & footwear  61.9  11.0  27.2  100.0  Wood & furniture  56.4  7.6  36.0  100.0  Machinery & metal  45.1  13.8  41.1  100.0  Other manufacturing  41.9  12.5  45.6  100.0  Location   Greater Accra  49.6  10.1  40.3  100.0   Kumasi  62.6  11.2  26.2  100.0   Takoradi  50.7  1.2  48.1  100.0   Cape Coast  36.3  57.8  5.9  100.0  Age group   Founded 1999–2003  51.2  13.2  35.6  100.0   Founded 1989–1998  51.9  10.0  38.2  100.0   Founded before 1988  68.3  8.3  23.4  100.0  Size group   Small (0–9)  48.4  13.2  38.4  100.0   Medium (10–74)  74.1  4.9  20.9  100.0   Large (75+)  58.4  1.9  39.7  100.0  Total  54.0  11.1  34.9  100.0  Note: The data is weighted. The ‘same size’ category includes all firms that experienced an increase or decrease in employment of less than 5%. Firm size is measured as the number of ‘persons engaged’ with the firm, and includes both paid and unpaid workers. Table 10 shows similar analysis but including exits, refusals and untraced firms. An important point to highlight is that large firms had a high refusal rate. Table 10: Firm Status Sector  Untraceable  Exit  Refusal  Shrunk  Same Size  Grew  Total  Food & beverages  27.6  27.8  7.1  17.0  5.4  15.1  100.0  Textiles, garments & footwear  39.4  22.1  3.9  21.3  3.8  9.5  100.0  Wood & furniture  40.1  20.2  3.8  21.1  2.9  11.9  100.0  Machinery & metal  18.8  16.4  9.5  23.7  7.7  23.8  100.0  Other  22.9  25.8  12.9  16.5  4.2  17.6  100.0  Location   Greater Accra  39.9  23.4  5.7  15.5  3.4  12.2  100.0   Kumasi  30.9  15.5  6.7  29.6  4.8  12.5  100.0   Takoradi  18.0  26.5  4.1  25.7  0.4  25.2  100.0   Cape Coast  1.4  46.3  4.4  17.5  27.9  2.5  100.0  Age group   Founded 1999–2003  38.7  24.6  4.9  16.3  4.4  11.1  100.0   Founded 1989–1998  31.7  16.9  6.5  24.0  4.1  16.8  100.0   Founded before 1988  22.5  24.8  7.9  30.3  3.9  10.7  100.0  Size group   Small (0–9)  38.1  21.7  4.1  17.7  4.7  13.6  100.0   Medium (10–74)  16.1  23.8  10.3  37.8  2.3  9.6  100.0   Large (75+)  8.0  14.8  43.5  21.3  0.4  12.0  100.0  Total  34.4  21.9  5.8  20.6  4.2  13.0  100.0  Sector  Untraceable  Exit  Refusal  Shrunk  Same Size  Grew  Total  Food & beverages  27.6  27.8  7.1  17.0  5.4  15.1  100.0  Textiles, garments & footwear  39.4  22.1  3.9  21.3  3.8  9.5  100.0  Wood & furniture  40.1  20.2  3.8  21.1  2.9  11.9  100.0  Machinery & metal  18.8  16.4  9.5  23.7  7.7  23.8  100.0  Other  22.9  25.8  12.9  16.5  4.2  17.6  100.0  Location   Greater Accra  39.9  23.4  5.7  15.5  3.4  12.2  100.0   Kumasi  30.9  15.5  6.7  29.6  4.8  12.5  100.0   Takoradi  18.0  26.5  4.1  25.7  0.4  25.2  100.0   Cape Coast  1.4  46.3  4.4  17.5  27.9  2.5  100.0  Age group   Founded 1999–2003  38.7  24.6  4.9  16.3  4.4  11.1  100.0   Founded 1989–1998  31.7  16.9  6.5  24.0  4.1  16.8  100.0   Founded before 1988  22.5  24.8  7.9  30.3  3.9  10.7  100.0  Size group   Small (0–9)  38.1  21.7  4.1  17.7  4.7  13.6  100.0   Medium (10–74)  16.1  23.8  10.3  37.8  2.3  9.6  100.0   Large (75+)  8.0  14.8  43.5  21.3  0.4  12.0  100.0  Total  34.4  21.9  5.8  20.6  4.2  13.0  100.0  Note: The data is weighted. Same size includes all firms that did not change employment by more than 5%. 5.2 The evolution of the firm size distribution and the ‘missing middle’ We now explore what these patterns of selection and growth mean for the evolution of the firm size distribution. The firm size distribution and its evolution can give us insights into which dynamics influence industrial change. Cabral and Mata (2003) examined the growth and exit patterns of a cohort of young Portuguese firms and showed that industrial evolution was mainly driven by within-firm growth rather than by selective exit of small firms. In a study on older firm-level data from Ghana, Sandefur (2010) argued that there was actually little evidence for within-firm growth: firms that were small in 1988 remained small in 2003, and the big firms that survived to 2003 were already big in 1988. But Sandefur argued that the firms that survived were much larger to start with and thus that changes in the firm size distribution were driven by selection. Hsieh and Klenow (2014) also showed that within-firm growth was negative in formal Indian firms and that selection effects were stronger even than in the USA. Constraints to firm growth have been argued to be one of the reasons why there is a ‘missing middle’ in developing countries, an underrepresentation of medium size firms (Tybout, 2000).5 In this section we will consider the overall firm size distribution, look at how firm age affects the firm size distribution and finally consider the growth and selection patterns of young firms (those founded between 1999 and 2003). Figure 1 shows the overall firm size distribution of the firms in the four cities surveyed. The figure also includes a best fit of the log-normal distribution, based on the firm size distribution in 2013. We find some evidence for a ‘missing middle’ in the sense that firms in the medium-sized categories are underrepresented compared to the log-normal distribution and there is an overrepresentation of larger firms. Like Hsieh and Olken (2014), we do not find evidence for bimodality in the firm size distribution. Figure 1: View largeDownload slide The Overall Firm Size Distribution of all the Firms in 2003, The Log-normal Best Fit Distribution and the Overall Firm Size Distribution of the Surviving Firms in 2013. Firm size is defined as the number of persons engaged with the firm. Note: 2013 includes only 401 surviving firms with positive employment. Weights are applied. Figure 1: View largeDownload slide The Overall Firm Size Distribution of all the Firms in 2003, The Log-normal Best Fit Distribution and the Overall Firm Size Distribution of the Surviving Firms in 2013. Firm size is defined as the number of persons engaged with the firm. Note: 2013 includes only 401 surviving firms with positive employment. Weights are applied. Figure 2 shows the firm size distributions of several cohorts of firms. This figure shows a positive relation between size and age: older firms are larger than younger firms. Some bimodality is present in firms more than 30 years old: there is a second mode at around 300 employees. Figure 2: View largeDownload slide Cross Sectional Relationship Between Age and Size in 2003. The data is weighted. The age refers to the age of the firm in 2003. Figure 2: View largeDownload slide Cross Sectional Relationship Between Age and Size in 2003. The data is weighted. The age refers to the age of the firm in 2003. 5.3 The evolution of entrants (the 2000–2003 cohort) Our survey did not collect information on entry of firms after 2003, since only firms from the 2003 industrial census were revisited. We can therefore not make any claims about what happened to entrant firms founded after 2003. However, we can have a look at young firms in 2003 and see how they fared in the time period 2003–2013 to explore whether growth or selection is responsible for the evolution of the firm size distribution. Figure 3 shows the firm size distribution of firms founded between 2000 and 2003, both in 2003 and 2013. The 2003 firms are further split into the firms that survived until 2013 (the survivors) and the firms that did not survive until 2013 (the non-survivors). Table 11 shows the average firm size of these groups of firms. Table 11: Comparison of the Mean Firm Size of Firms Founded Between 2000 and 2003 and all Firms   Firms founded between 2000–2003  All firms in sample  (1) All firms in 2003  5.37 (0.30)  12.00 (0.66)  (2) Surviving firms in 2003  7.25 (0.71)  17.91 (1.35)  (2*) Non-surviving firms in 2003  4.36 (0.26)  7.40 (0.58)  (3) Surviving firms in 2013  7.05 (0.95)  15.91 (1.85)  t-Test of the difference in means  p value:  p value:   Selection: (1) versus (2)  0.016**  0.000***        (2) versus (2*)  0.000***  0.000***   Growth: (2) versus (3)  0.864  0.384  Kolmogorov–Smirnov equality-of-distribution test   Selection: (1) versus (2)  0.248  0.019**       (2) versus (2*)  0.033**  0.000***   Growth: (2) versus (3)  0.450  0.002***    Firms founded between 2000–2003  All firms in sample  (1) All firms in 2003  5.37 (0.30)  12.00 (0.66)  (2) Surviving firms in 2003  7.25 (0.71)  17.91 (1.35)  (2*) Non-surviving firms in 2003  4.36 (0.26)  7.40 (0.58)  (3) Surviving firms in 2013  7.05 (0.95)  15.91 (1.85)  t-Test of the difference in means  p value:  p value:   Selection: (1) versus (2)  0.016**  0.000***        (2) versus (2*)  0.000***  0.000***   Growth: (2) versus (3)  0.864  0.384  Kolmogorov–Smirnov equality-of-distribution test   Selection: (1) versus (2)  0.248  0.019**       (2) versus (2*)  0.033**  0.000***   Growth: (2) versus (3)  0.450  0.002***  Note. Firm size is defined as the number of ‘persons engaged’ and includes apprentices. The p value is reported for a t-test of the difference between means. Figure 3: View largeDownload slide Selection or Decline? The weighted firm size distribution of the 2003 full distribution, the 2003 ‘destined to survive’ firms, the 2003 ‘destined to exit’ firms and the 2013 surviving firms. Only firms founded between 2000 and 2003 are included. Note. 2003 sample weights used and adjusted for refusals in 2013. Figure 3: View largeDownload slide Selection or Decline? The weighted firm size distribution of the 2003 full distribution, the 2003 ‘destined to survive’ firms, the 2003 ‘destined to exit’ firms and the 2013 surviving firms. Only firms founded between 2000 and 2003 are included. Note. 2003 sample weights used and adjusted for refusals in 2013. We can use Figure 3 to replicate the test for whether growth or selection drives the firm size distribution that was suggested and tested by Cabral and Mata (2003) for Portugal, and later implemented in Ghana by Sandefur (2010). This test involves comparing (1) the firm size distribution at the initial measure point with (2) the firm size distribution of firms ‘destined to survive’ at the initial measure point (i.e., excluding firms that exited) as well as (3) the firm size distribution of the surviving firms at the final measure point. By comparing these distributions, we can attribute changes in the firm size distribution within a cohort to growth and selection. Comparing distribution (1) with distribution (2) shows the role of selection, as both distributions relate to the same (initial) point in time, but to a different group of firms: distribution (1) includes both surviving and exiting firms, while distribution (2) only includes the firms that survived up to the final measure point. Comparing distribution (2) with distribution (3) shows the role of firm growth, as both distributions use the same group of firms that did not exit in the measured time period. Cabral and Mata (2003) found strong differences between distributions (2) and (3), but not between distributions (1) and (2), indicating that growth played a more important role than selection in the evolution of the Portuguese firm size distribution. Sandefur (2010) found the opposite pattern in Ghana: strong differences between (1) and (2), but little difference between (2) and (3), indicating a strong role of selection. However, the difference in the distributions in Figure 3 is not as clear as in the Portuguese case of Cabral and Mata (2003) or as in the previous Ghanaian study of Sandefur (2010). Selection. The 2003 distribution of firms that survived is very similar shaped to the 2003 distribution of all firms (including non-survivors). A Kolmogorov–Smirnov equality-of-distribution test, using weights, does not reject the null hypothesis that the distributions are the same (the p value is 0.248).6 However, the means of these distributions are significantly different. In Table 11 we can see that, in 2003, the surviving firms had on average 1.88 more workers than all firms together, and on average 2.89 workers more than firms that did not survive. These differences are statistically significant at respectively a 5% and 1% level of significance, but are much smaller than the differences found by Sandefur.7 Firm growth. From Figure 3 we see that the 2003 distribution of surviving firms is very similarly shaped to the 2013 distribution of surviving firms. The test of the differences shows a similar picture: the average firm size of the young cohort that survived was 7.25 in 2003 and 7.05 in 2013. This decline is not statistically significant and reveals that on average, firms stagnated. Unlike Cabral and Mata (2003) we find no strong evidence for growth driving the evolution of the firm size distribution. But unlike Sandefur (2010) we find little evidence of selection driving the evolution of the firm size distribution. Regional differences. In Section 4 and in Section 6 we saw strong regional differences between Accra and Kumasi: Accra firms were more likely to exit and Kumasi firms were less likely to grow. We see similar patterns when comparing the firm size distributions. Figure 4 shows the evolution of the firm size distribution for young firms for both Accra and Kumasi. In Kumasi we see little effect of differential selection, as the 2013 firm size distribution of surviving firms resembles the 2003 overall firm size distribution quite closely, but we see a strong pattern of decline: the firm size distribution of surviving firms has shifted to the left in 2013 compared to 2003. In Accra we find a stronger pattern of differential selection, but less evidence for a decline in firm growth: in fact, firms in Accra grew on average by 0.45 worker, even though this change is not statistically significant (the p value is 0.825). Figure 4: View largeDownload slide The Weighted Firm Size Distribution of the 2003 Full Distribution, the 2003 ‘Destined to Survive’ Firms and the 2013 Surviving Firms, for both Accra and Kumasi. Only firms founded between 2000 and 2003 are included. Note. The 2003 sample weights were used (adjusted for refusals). Figure 4: View largeDownload slide The Weighted Firm Size Distribution of the 2003 Full Distribution, the 2003 ‘Destined to Survive’ Firms and the 2013 Surviving Firms, for both Accra and Kumasi. Only firms founded between 2000 and 2003 are included. Note. The 2003 sample weights were used (adjusted for refusals). Compared to Sandefur’s study we find only a small role of selection in driving the evolution of the firm size distribution. Three factors could explain these differences. First of all, Sandefur’s analysis included all firms, unlike Cabral and Mata (2003) who only included firms from one particular cohort. This was done because of the lack of availability of age information in Sandefur’s data. Including all firms instead of a cohort will make the role of selection look stronger. This is because older firms tend to be larger and more likely to survive, as our regression in Table 5 showed, while younger firms are more likely to exit and tend to be smaller, as can be seen in Figure 2. The old large firms who are likely to be survivors will therefore dominate the distribution of survivors. When we repeat our analysis including all firms, we see that the selection effects become stronger and significant in all cases (see the right column of Table 11). Second, Sandefur relied on matching firms from the 1987 census to the 2003 census by ISIC code, region and firm name. This procedure most likely led to an overrepresentation of larger firms. Only 13% of the 2003 firms who said they were operating in 1987 were matched, and Sandeufur (2010: 7n8) noted that ‘average firm size is somewhat larger for those I was able to match’ but did not provide any further details. This means that Sandefur’s matching procedure sets up a matched sample of 1987 firms that survived to 2003, in which larger firms are overrepresented, for comparison with the full sample of firms in 1987. This does not affect his result on the unimportance of within-firm growth, but does overstate his result that selection is driving the firm size distribution. Third, Sandefur’s panel dealt with a different time period (1987–2003) than ours (2003–2013). The time period between 1987 and 2003 was characterised by substantial changes in policy in Ghana, including a structural adjustment programme and trade liberalisation. It is possible that the introduction of these policies made selection more important than in the more recent period that we studied. Cabral and Mata focus on either growth or selection as explanations for the positive age–size correlation. We find that they only played a small role. What does then explain the positive age–size correlation shown in Figure 2? Changes in the composition of entering firms is the most likely explanation. Sandefur showed that between 1987 and 2003 the total number of manufacturing firms increased threefold and the average firm size halved. Unfortunately we cannot investigate this further using our 2003–2013 panel because we did not sample new entrants. However, our analysis of the most recent entrants in our sample, the 2000–2003 cohort, is consistent with this conclusion. This suggests that studying entry is key to understanding the evolution of the firm size distribution in Ghana. 5.4 Changes in aggregate employment How did the patterns in growth and exit that we have described affect employment for the firms sampled? Table 12 gives the levels of aggregate employment in each sector, both in 2003 and 2013. The table shows how important large firms are in providing employment: in 2003 firms with more than 75 employees provided almost half of the total employment measured. We can see that aggregate weighted employment of the Ghanaian manufacturing firms sampled in 2003 (the last row of the table) has decreased from 134,863 in 2003 (column 1) to 74,319 in 2013 (column 2). This is a decrease of 45%. Part of this decrease has come from firms that exited by 2013 (column 3); these firms employed an estimated 20,643 workers in 2003. Another 24,107 workers were employed in 2003 in firms that we were not able to trace in 2013 (column 4). As mentioned earlier, due to data limitations, we cannot separate between paid apprentices and regular workers for the 2013 figures of a large group of firms. The employment figures therefore include apprentices. Table 12: Aggregate Weighted Employment in 2003 and 2013   Total 2003 employment  Total 2013 employment in surviving firms  2003 Employment in exiting firms  2003 Employment untraced firms  2003 Employment in surviving firms  Sector   Food & beverages  19,735  13,283  3,083  1,484  15,168   Textiles, garments & footwear  35,272  13,795  5,010  11,706  18,556   Wood & furniture  29,119  13,059  6,034  6,591  16,494   Machinery & metal  15,722  11,618  1,774  2,270  11,678   Other manufacturing  35,015  22,564  4,742  2,056  28,217  Size   Small (0–9 workers)  37,060  15,446  7,247  13,775  16,038   Medium (10–74 workers)  32,335  19,725  7,195  4,768  20,373   Large (more than 75)  65,468  39,148  6,201  5,565  53,702  Total  134,863  74,319  20,643  24,107  90,112    Total 2003 employment  Total 2013 employment in surviving firms  2003 Employment in exiting firms  2003 Employment untraced firms  2003 Employment in surviving firms  Sector   Food & beverages  19,735  13,283  3,083  1,484  15,168   Textiles, garments & footwear  35,272  13,795  5,010  11,706  18,556   Wood & furniture  29,119  13,059  6,034  6,591  16,494   Machinery & metal  15,722  11,618  1,774  2,270  11,678   Other manufacturing  35,015  22,564  4,742  2,056  28,217  Size   Small (0–9 workers)  37,060  15,446  7,247  13,775  16,038   Medium (10–74 workers)  32,335  19,725  7,195  4,768  20,373   Large (more than 75)  65,468  39,148  6,201  5,565  53,702  Total  134,863  74,319  20,643  24,107  90,112  Note. Employment here is measured as the number of ‘persons engaged’ with the firm, and includes apprentices. The 2013 weights are adjusted only for refusals by continuing firms. If we only consider the firms that survived, we also see that the total employment in these firms decreased by around 18%, which is 1.8% annually. Few studies report total employment changes for surviving firms, but a couple of studies report changes in employment averaged by firm. In the UK, Bryson and Nurmi (2011) showed that surviving firms shrunk on average by two per cent between 1998 and 2004. A large study on 85 developing countries conducted by Aterido et al. (2011) found positive employment growth in surviving firms, unlike the negative employment growth found in our sample.8 Many other studies focusing on job destruction do not only incorporate figures from surviving firms but also job losses due to firm exit. Davis and Haltiwanger (1999) summarize a large number of studies showing that in the United States, as well as in a wide range of OECD countries, about one in ten jobs in manufacturing disappears every year. However, for most countries, the loss in jobs is compensated by new jobs creation in other manufacturing firms.9 Mazumdar et al. (2001) used the RPED surveys conducted in several African countries in the early 1990s to investigate changes in total employment during the period from 1982 to 1992. The authors found that total employment in Ghanaian firms increased by about 30% over this period for firms that were alive when the RPED surveys were conducted. Table 12 also shows that the declines in total employment in surviving firms in Ghana between 2003 and 2013 were the largest in textiles and garments and wood and furniture making and the smallest in machinery and metals. Again, size matters: the decline was the highest in small firms and the lowest in medium-sized firms. Table 13 shows the declines amongst different kinds of workers. Declines in apprentices engaged were very high, around 80%, compared to only a 33% decline in paid production workers and a 31% decline in paid other workers. Declines were similar for men and women. Table 13: Aggregate Employment by Worker Category and Gender   Employment  2003  2013  Worker category   Apprentices*  35,276  7,016   Production workers  73,840  48,845   Other workers  19,893  15,564  Gender   Male  99,085  52,047   Female  34,369  17,501    Employment  2003  2013  Worker category   Apprentices*  35,276  7,016   Production workers  73,840  48,845   Other workers  19,893  15,564  Gender   Male  99,085  52,047   Female  34,369  17,501  Note. The 2013 weights are adjusted only for refusals by surviving firms. Firms report total persons engaged and are then asked about each kind of worker—thus totals do not add to exactly the totals reported in Table 12. *The 2003 figure on apprentices include paid apprentices. For the 2013 figure, paid apprentices are classified as production workers and not as apprentices. There are two potential complicating factors in our measurement of apprentices over time, which we discuss more extensively in the Appendix. First of all, the 2013 survey only included unpaid apprentices whereas the 2003 survey most likely included both paid and unpaid apprentices in the apprentice category.10 The exclusion of paid apprentices in the 2013 figures means that the decline is most likely overstated. Data from representative household surveys (the Ghana Living Standards Surveys) suggests that between 18% and 40% of the apprentices are paid. Using these shares, the 2013 figures including paid apprentices would be between 8556 and 11,693, corresponding to a decrease in apprentices of between 67% and 75% compared to the 2003 figure. Similarly, the classification of paid apprentices as production workers in 2013 means that the decline in non-apprentice production workers was most likely higher, between 36% and 40%. The decline in total employment excluding apprentices (but including paid other workers) is most likely between 33% and 36% (see also the Appendix). Second, apprenticeships are concentrated in younger and smaller firms and last only for a few years. The 2003 census suggests that, for Ghana as a whole, of the 75,319 apprentices 24,235 were working in firms older than 10 years. By construction, our 2013 data only include firms founded before 2003 and all firms are therefore older than 10 years. If the pattern of mainly young firms using apprentices as seen in the 2003 census continued between 2003 and 2013, we expect our apprentice figure in 2013 to be lower, simply because the firms in 2013 are older. An estimation of this impact is made in the Appendix. To sum up this section, our dataset only covers firms that were operating in 2003. Therefore, it should be emphasised that we cannot conclude that total employment in Ghanaian manufacturing declined between 2003 and 2013 since we do not have data on firms that entered after 2003 and the amount of employment in these new firms. We are thus only able to say that total employment in the weighted sample of firms decreased by around 45% between 2003 and 2013, and that this decline was larger for apprentices, although this is a result of apprenticeships being concentrated in young firms and the different definitions of apprentices used in 2003 and 2013. Average firm size actually increased slightly due to the exit of small firms: the average firm size was 12 in 2003 and 16 for those surviving in 2013, which is mainly caused by the differential exit rates of small and large firms. 6. Conclusion The manufacturing sector has been seen as a potential engine of growth and employment in the Ghanaian economy. But our research has shown that Ghanaian manufacturing firms that existed in 2003 performed poorly over the 10 years between 2003 and 2013, a continuation of poor performance that has been documented for the 1990s and early 2000s in other firm-level research. Around 21% of firms exited between 2003 and 2013 while another 22% were untraced. If the untraced firms are assumed to have exited then these figures are similar to those estimated in Ghana over the 5 years between 1993 and 1998 by Söderbom et al. (2006). Firms in Accra, young firms and small firms were more likely to have exited. Exploring the reasons for exit amongst those owners or managers of firms who could be found suggested that small firms were more likely to exit due to personal circumstances of the owner while the most cited reason for exit in large firms was increasing costs. We also explored the importance of selection in explaining the evolution of the firm size distribution in Ghana. Using the simple graphical test suggested by Cabral and Mata (2003) we found that selection only played a small role, contradicting earlier work by Sandefur (2010) who used an inappropriate sample due to data limitations. But unlike Cabral and Mata (2003) we also find little role for within-firm growth in explaining the evolution of the firm size distribution. Firm entry is likely to be a key factor driving this evolution, and would seem to be a crucial area for future research. Broadening our analysis to surviving firms we have shown that only about 35% of the surviving firms that were successfully interviewed grew their employment by more than 5%, while 54% shrunk by more than 5%. Aggregate weighted employment (including apprentices) fell by 45%, from 134,863 in 2003 to 74,319 in 2013, an estimate that includes adjustments for the non-response of surviving firms. Due to an inconsistent definition of apprenticeships used across the two waves, we cannot separate paid apprentices from other employees for the 2013 figures of a substantial number of firms. This means that we cannot provide a precise figure for the drop in employment excluding apprentices. However, if we make assumptions for the purpose of establishing bounds on the share of paid apprenticeships in total apprenticeships, the decrease in employment excluding apprentices is likely between 33 and 36% (see the Appendix for details). We cannot estimate growth in total manufacturing employment in Ghana without surveying new firms, and creative destruction is one possible positive interpretation of our result. But our work showing a lack of growth of young firms in 2003 up until 2013 does not paint a positive picture of the state of manufacturing in Ghana. Supplementary material Supplementary material is available at Journal of African Economies online. Acknowledgements We thank Marcel Fafchamps, Francis Teal and the anonymous reviewer for helpful comments and Sofia Monteiro and Václav Těhle for excellent research assistance. We thank Moses Awoonor-Williams for invaluable assistance with the fieldwork and the Ghanaian Statistical Service (GSO) for their cooperation. Furthermore, we would like to thank our team of enumerators for their dedication to this project. Footnotes 1 Unlike some firm surveys conducted in other countries (e.g., India), the only firms to be excluded from the census were household based enterprises that did not have a sign indicating that a firm was operating in the household. There was no explicit size based criterion for inclusion. 2 For 14.2% of the firms where this exit questionnaire was conducted, we interviewed the former owner or manager. In 23.6% of the cases we interviewed a former worker or relative and for the remaining firms (62.3%) we interviewed a neighbour. 3 Aga and Francis (2015) calculate annualised rates that are not compounded so the numbers reported in their paper are different to the ones we have calculated and reported using their results. 4 Firm size is measured as the number of persons engaged, and includes both paid and unpaid workers. This is the same definition as used by Sandefur (2010) for his analysis of firm growth and selection between 1989 and 2003. 5 As Hsieh and Olken (2014) argue, a ‘missing middle’ is an exaggeration, as there are indeed medium-sized firms operating and we do not see strong signs of bimodality. In a reply, Tybout (2014) argues that the term ‘missing middle’ does not relate to a bimodality in the firm size distribution, but to an underrepresentation compared to an ‘undistorted’ distribution. Theoretically it has been argued that this ‘undistorted distribution’ should resemble a Pareto distribution (see e.g., Luttmer, 2007). 6 Note that we are comparing the 2003 full distribution with the 2003 distribution of surviving firms, of firms founded between 2000 and 2003. The 2003 full distribution includes both the surviving and non-surviving firms. 7 Note that our methodology differs from the methodology used by Sandefur (2010). Sandefur (2010) was limited in the number of firms he could match: only 13% of the firms from 2003 that claimed to have been in existence before 1987 were matched with the 1987 observations of these firms. For his analysis he therefore used all firms, since his sample was too small to use only new firms. We follow the methodology of Cabral and Mata (2003) who only used a cohort of new firms for their analysis. For our figures we use the cohort of firms founded between 2000 and 2003 (i.e., at most 4 years old). If we follow Sandefur (2010) and use all firms, we find stronger differences, as can be seen in the right part of Table 11. 8 Aterido et al. (2011) report employment growth as the 3-year change in employment divided by the average employment in the first year and the third year, for 85 developing countries based on the World Business Environment Survey (WBES). By construction this measure is bounded between −2.0 and + 2.0 and therefore less dominated by outlying observations than a simple average of employment change rates. For their entire sample of firms this figure is 0.117 (taken over 3 years). If we use the same calculation method for our sample, the equivalent figure (taken over 11 years) is −0.229. 9 These studies do not only include job losses due to exits, also include job losses due to exits of young firms that entered after the first measurement point. Our sample only includes firms operating in 2003, which means that firms that entered after 2003 but exited before 2013 are not reported by us. However, such firms are included in the job destruction rates discussed by Davis and Haltiwanger (1999). The inclusion of these young firms exiting early will likely lead to higher job loss figures, because young firms are more likely to exit than older firms. 10 The 2003 Phase I questionnaire, which was used to calculate the number of apprentices, does not specify whether paid apprentices should be included as apprentices or as production workers. In the 2003 Phase II questionnaire and in the 2013 follow-up survey paid apprentices are classified as production workers. A comparison of 2003 Phase I and 2003 Phase II data for the limited number of firms who were interviewed in both phases shows that the number of apprentices reported was higher in the Phase I survey, which suggests that paid apprentices were classified as apprentices. See also the Appendix for a further discussion. References Aga G. A., Francis D. C. ( 2015). As the Market Churns: Estimates of Firm Exit and Job Loss Using the World Bank’s Enterprise Surveys, World Bank Policy Research Working Paper 7218. Aterido R., Hallward-Driemeier M. ( 2010). The Impact of the Investment Climate on Employment Growth. Does Sub-Saharan Africa Mirror Other Low-Income Regions? World Bank Policy Research Working Paper 5218. The World Bank. Aterido R., Hallward-Driemeier M., Pagés C. ( 2011) ‘ Big Constraints to Small Firms’ Growth? Business Environment and Employment Growth Across Firms’, Economic Development and Cultural Change , 59 ( 3): 609– 47. Google Scholar CrossRef Search ADS   Aw B., Chen X., Roberts M. J. ( 2001) ‘ Firm-level Evidence on Productivity Differentials and Turnover in Taiwanese Manufacturing’, Journal of Development Economics , 66: 51– 86. Google Scholar CrossRef Search ADS   Bloom N., Lemos R., Sadun R., Scur D., Van Reenen J. ( 2014) ‘ JEEA-FBBVA Lecture 2013: The New Empirical Economics of Management’, Journal of the European Economic Association , 12: 835– 76. doi:10.1111/jeea.12094. Google Scholar CrossRef Search ADS   Bloom N., Van Reenen J. ( 2010) ‘ Why Do Management Practices Differ across Firms and Countries?’, Journal of Economic Perspectives , 24 ( 1): 203– 24. Google Scholar CrossRef Search ADS   Bryson A., Nurmi S. ( 2011) ‘ Private Sector Employment Growth, 1998−2004: A Panel Analysis of British Workplaces’, Cambridge Journal of Economics , 35 ( 1): 85– 104. Google Scholar CrossRef Search ADS   Cabral L. M. B., Mata J. ( 2003) ‘ On the Evolution of the Firm Size Distribution: Facts and Theory’, American Economic Review , 93 ( 4): 1075– 90. Google Scholar CrossRef Search ADS   Davis S. J., Haltiwanger J ( 1992) ‘ Gross Job Creation, Gross Job Destruction, and Employment Reallocation’, Quarterly Journal of Economics , 107 ( 3): 819– 63. Google Scholar CrossRef Search ADS   Davis S. J., Haltiwanger J ( 1999) ‘Chapter 41 – Gross job flows’, in Ashenfelter O., Card D. (eds), Handbook of Labor Economics , Vol. 3. Elsevier, Amsterdam. Frazer G. ( 2005) ‘ Which Firms Die? A Look at Manufacturing Firm Exit in Ghana’, Economic Development and Cultural Change , 53 ( 3): 585– 617. Google Scholar CrossRef Search ADS   Hallward-Driemeier M. ( 2009). Who survives? The Impact of Corruption, Competition and Property Rights across Firms. World Bank Policy Research Working Paper 5084. Hsieh C.-T., Klenow P. ( 2014) ‘ The Life Cycle of Plants in India and Mexico’, Quarterly Journal of Economics , 129 ( 3): 1035– 84. Google Scholar CrossRef Search ADS   Hsieh C.-T., Olken B. A. ( 2014) ‘ The Missing ‘Missing Middle’, Journal of Economic Perspectives , 28 ( 3): 89– 108. doi:10.1257/jep.28.3.89. Google Scholar CrossRef Search ADS   Li Y., Rama M. ( 2015) ‘ Firm Dynamics, Productivity Growth, and Job Creation in Developing Countries: The Role of Micro- and Small Enterprises’, World Bank Research Observer , 30 ( 1): 3– 38. Google Scholar CrossRef Search ADS   Liedholm C., McPherson M., Chuta E. ( 1994) ‘ Small Enterprise Employment Growth in Rural Africa’, American Journal of Agricultural Economics , 76: 1177– 82. Google Scholar CrossRef Search ADS   Luttmer E. ( 2007) ‘ Selection, Growth, and the Size Distribution of Firms’, The Quarterly Journal of Economics , 122 ( 3): 1103– 44. doi:10.1162/qjec.122.3.1103. Google Scholar CrossRef Search ADS   Mazumdar D., Mazaheri A., Biggs T., Ramachandran V., Shah M. ( 2001). The Manufacturing Sector in Sub-Saharan Africa: An Analysis Based on Firm Surveys in Seven Countries. RPED Discussion Paper Series. Melitz M. ( 2003) ‘ The Impact of Trade on Intra-industry Reallocations and Aggregate Industry Productivity’, Econometrica: Journal of the Econometric Society , 71 ( 6): 1695– 1725. doi:10.1111/1468-0262.00467. Google Scholar CrossRef Search ADS   Mitra P., Muravyev A., Schaffer M. ( 2014) ‘ Labor Reallocation and Firm Growth: Benchmarking Transition Countries Against Mature Market Economies’, IZA Journal of Labor & Development , 3 ( 1): 1– 22. Google Scholar CrossRef Search ADS   Sandefur J. ( 2010). On the Evolution of the Firm Size Distribution in an African Economy. CSAE Working Paper, (5). Shiferaw A., Bedi A. ( 2013) ‘ The Dynamics of Job Creation and Job Destruction in an African Economy: Evidence from Ethiopia’, Journal of African Economies , 22 ( 5): 651– 92. Google Scholar CrossRef Search ADS   Söderbom M., Teal F., Harding A. ( 2006) ‘ The Determinants of Survival among African Manufacturing Firms’, Economic Development and Cultural Change , 54 ( 3): 533– 55. Google Scholar CrossRef Search ADS   Sutton J., Kpentey B. ( 2012) An Enterprise Map of Ghana . London: International Growth Centre. Teal F., Habyarimana J., Thiam P., Turner G. ( 2006) ‘Ghana: an analysis of firm productivity’, in Regional Program on Enterprise Development . Washington DC: The World Bank, pp. 1– 72. Teal F. ( 2016). Firm Size, Employment and Value-added in African Manufacturing Firms: Why Ghana needs its 1%. CSAE working Paper WPS/2016-07. Tybout J. R. ( 2000) ‘ Manufacturing Firms in Developing Countries: How Well Do They Do, and Why?’, Journal of Economic Literature . doi:10.1257/jel.38.1.11. Tybout J. ( 2014) ‘ The Missing Middle, Revisited’, Journal of Economic Perspectives , 28 ( 4): 235– 36. Google Scholar CrossRef Search ADS   Van Biesebroeck J. ( 2005) ‘ Firm Size Matters: Growth and Productivity Growth in African Manufacturing’, Economic Development and Cultural Change , 53 ( 3): 545– 83. doi:10.1086/426407. Google Scholar CrossRef Search ADS   World Bank ( 2013) World Development Report 2013: Jobs . Washington, DC: World Bank Publications. © The Author 2017. Published by Oxford University Press on behalf of the Centre for the Study of African Economies, all rights reserved. For Permissions, please email: journals.permissions@oup.com

Journal

Journal of African EconomiesOxford University Press

Published: Mar 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 12 million articles from more than
10,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Unlimited reading

Read as many articles as you need. Full articles with original layout, charts and figures. Read online, from anywhere.

Stay up to date

Keep up with your field with Personalized Recommendations and Follow Journals to get automatic updates.

Organize your research

It’s easy to organize your research with our built-in tools.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve Freelancer

DeepDyve Pro

Price
FREE
$49/month

$360/year
Save searches from
Google Scholar,
PubMed
Create lists to
organize your research
Export lists, citations
Read DeepDyve articles
Abstract access only
Unlimited access to over
18 million full-text articles
Print
20 pages/month
PDF Discount
20% off