To Formalize or Not to Formalize? Comparisons of microenterprise data from Southern and East Africa *

Transcription

1 1 Gelb Mengistae Ramachandran Shah DRAFT To Formalize or Not to Formalize? Comparisons o microenterprise data rom Southern and East Arica * Alan Gelb Taye Mengistae Vijaya Ramachandran Manju Kedia Shah January 2009 *An earlier drat o this paper was prepared or a workshop on The Regulatory Environment and its Impact on the Nature and Level o Economic Growth and Development in South Arica, hosted by the Development Policy Research Unit, University o Cape Town, South Arica, October The views presented in this paper are solely those o the authors. We are grateul to Robin Krat or research assistance.

2 2 Abstract Why do irms choose to locate in the inormal sector? Researchers oten argue that the high cost o regulation prevents inormal irms rom becoming ormal and productive. Our results point to a more nuanced story. Using data rom surveys o microenterprises in South Arica, Namibia, Botswana, Kenya, Uganda, Tanzania and Rwanda, we ind that the labor productivity o inormal irms is virtually indistinguishable rom that o ormal irms in East Arica but not in Southern Arica. We provide a theoretical model to explain this result, based on the key assumption that irms may evade taxes subject to a cost (or concealment cost) that is increasing and convex in the irm s employment size. Consequently, the productivity distributions relect the dierences in concealment costs and the opportunity cost o ormality in the two regions. Greater enorcement o laws and better provision o services such as inance and electricity to ormally registered irms in Southern Arica means that irms are more likely to register; those that do not are likely to be operating as survivalist irms. But in East Arica, weak enorcement o tax payment and no signiicant dierence in access to services between ormal and inormal irms means that these variables do not explain the allocation o irms across the inormal-ormal divide. We conclude that in countries with weaker business environments such as those in East Arica, inormal irms are just as likely as ormal irms to increase their productivity as they grow. Thus, interventions to increase productivity and lower the cost o ormality may be helpul. But in countries with strong business environments such as those in Southern Arica, owners o inormal irms are likely to be better o entering the labor market as wage labor. In the latter case, investment in education or vocational training is probably more important.

3 3 Introduction Inormal irms account or a large share o employment in most developing countries, oten a dominant share. What drives the choice o whether a irm is inormal or ormal? How is that decision impacted by existing Government regulations and policies towards the MSME sector, and by the characteristics o the irms themselves? How should inormal irms be considered as a mainspring o uture growth and employment generation or as a survivalist employment alternative? Are there implications or policymaking? This paper considers these questions or a set o countries in Southern and Eastern Arica, using data rom the World Bank s Enterprise Surveys carried out in these countries in recent years. These surveys have been completed or ormal businesses in some 20 Sub- Saharan Arican (Arican) countries since 2005, and have been complemented by surveys or inormal businesses in 10 o these countries. The content o the surveys has not been the same o course; ar more detail is asked o ormal irms in areas relating to production, costs, productivity and proitability. However, the surveys contain a number o similar questions, making it possible to derive some conclusions about the nature o ormalization. In this paper, we use data rom 7 o the countries in which surveys have been undertaken. To provide a ramework or the empirical analysis, Section 2 o the paper discusses a model o the determinants o the ormalization decision. Like other decisions, many researchers view this decision as relecting the balance o beneits and costs o ormalization--or example, access to credit versus tax liability. This balance will depend on the ormal regulatory ramework but also on the capability and eort made by the state to enorce regulations. I tax inspectors can be bribed and irms not registered or taxes can also settle their problems by bribing inspectors, there may be little real dierence whatever the ormal tax rates. In any given country, the balance can also depend on the characteristics o the irm (or the entrepreneur). Some entrepreneurs may ace dierent opportunity costs o being in business; some might also be better placed than others to achieve the potential access and productivity gains that come with ormalization. We ormulate a model that allows us to determine how irms sort into the ormal vs. the inormal sector. Section 3 picks up on the empirical analysis. One question, on which studies are not unanimous, is how to deine ormal. We use a uniorm deinition based on registration or taxes. We present some descriptive statistics or three sets o irms: ormal microenterprises (5 employees or less), inormal micro-enterprises (also 5 employees or less), and ormal small irms with 5 to 10 employees. These are compared across the seven countries: a Southern group consisting o South Arica, Namibia and Botswana and an Eastern group, Uganda, Tanzania and Kenya, and Rwanda. Each country is unique, and there are important dierences between countries in each group. Nevertheless, there appear to be systematic dierences between the two groups, and these suggest very dierent stories in Southern and Eastern Arica. In particular,

4 4 inormal irms seem to be more survivalist in the ormer region, and more likely to be potential sources o growth and employment generation in the latter one where the main dierences are between smaller and larger irms rather than ormal and inormal irms. Indeed, the decision o whether to ormally register or not may be quite an idiosyncratic one in the East Arican group, in contrast to the Southern Arica group where inormal irms are truly less productive than other types o irms. Why this should be so is an interesting question, and we hypothesize that it relects a combination o weak delivery o the services that are supposed to low rom ormalization combined with weak ability to enorce ormalization. Rwanda displays some distinctive characteristics relative to these two groups. Interestingly, some o the dierences implied by the micro-data appear to be consistent with country-level characteristics, including rom Doing Business and the Worldwide Governance Indicators. Section 4 looks at the role o human capital and the business environment in the sorting o irms into ormality and inormality. We consider the costs o inormality (avoidance o taxes and laws) and the beneits o ormality (access to electricity, water, sewage, and inance) across the seven countries in our sample. Section 5 summarizes, and draws some implications or policy. These dier between the Southern and Eastern groups. In particular, we argue that the dierences between ormal and inormal irms in the Southern group relect the characteristics o the entrepreneurs and labor orce. Most o the survivalist irms in the inormal sector (which is not to be conused with the small-scale sector) are unlikely to become ormal irms. While certain initiatives may ease the situation o these irms, the approach towards improving the living standards o those engaged in this sector are most likely to be ound in increasing access to employment in the ormal sector, whether through temporary wage subsidies or other mechanisms. On the other hand, improvements in the business environment in East Arica are potentially more valuable in changing the balance o beneits and costs rom ormalization, and so encouraging small irms to ormalize and grow.

5 5 2. A comparison o productivity in the inormal and ormal sector in seven Arican countries Enterprise surveys covering the ormal sector conducted or Arican countries cover only the ormal sector and irms with a minimum employment level o 5 employees, but surveys in several countries have included a separate survey o microenterprises-those with less than ive employees. For various reasons, notably the small size o establishments, their expected high rate o turnover, the high level o inormality o establishments in many activities and consequently the diiculty o obtaining trustworthy inormation rom oicial sources, an aerial sampling approach is used to estimate the population o establishments and select the sample in this stratum or all regions o the survey. Table 1 describes the sample o irms used in this analysis. 1 Appendix 1 describes the sampling methodology. Table 1: Sample Sizes Botswana Namibia South Arica Kenya Tanzania Uganda Rwanda Inormal micro Formal micro Total micro Total ormal While enterprises in very small size classes are commonly reerred to as inormal, a part o the irms with less than 5 employees in our sample is indeed classiied as ormal. This raises the question o how the boundary between ormal and inormal irms should be deined. Enterprises in the micro size class were asked about their registration status. Firms were asked whether they had: registered name with the Oice o the Registrar or other government institutions responsible or approving company names, registered with the Oice o the Registrar, the local courts, or other government institutions responsible or commercial registration, an operating or trade license or otherwise registered or a general business license with any municipal agency, 1 We know that there will be some irms with more than 5 employees that are not registered or tax purposes in our small-irm sample. But the scope o this paper is to look at the determinants o ormality amongst microenterprises; thus the analysis is mostly conined to this subset o irms.

6 6 obtained a tax identiication number rom the tax administration or other agency responsible or tax registration. In all the countries in our sample, only a subset o irms in the irst three categories are registered or tax purposes. We deine irms that are registered or taxes to be ormal microenterprises, those that are not registered are deined as inormal. This distinction o microenterprises into these two categories ollows earlier studies. It is clear that to grow into larger size classes, a irm must irst ormalize its operations by registering with tax authorities. As a irst step towards looking at the dierences between ormal and inormal irms, we look at a simple measure o productivity value added per worker o inormal microenterprises, ormal microenterprises and ormal small irms in each o the seven countries in our sample. Figure 1: Labor productivity o irms in the sample (median) Figure 1 presents the median productivity between microenterprises and ormal sector small irms in our sample. We see that, between Southern and Eastern Arica, dierences in ormal sector productivity are much greater than dierences in microenterprise productivity. There are other large gaps as well--ormal microenterprises are more productive in Southern Arica compared to their inormal counterparts, and much more productive than ormal microenterprises in Eastern Arica. However, median values present only part o the picture. The dispersions around the median provide urther inormation on the nature o enterprise perormance, and competitiveness o the sector. Figures 2-8 show kernel density estimations or labor productivity across the three types o irms in South Arica, Botswana and Namibia (Southern Arica) and in Kenya, Tanzania and Uganda (East Arica) and Rwanda. A striking eature o the igures below is that in some countries the probability density o labor productivity is bimodal along the ormal vs. inormal irms divide but not in others. Labor productivity is lower in inormal irms on average in every country sample, and the

7 7 density o labor productivity in those irms also signiicantly overlaps the density or ormal irms. But the degree o overlap also varies a great deal rom country sample to country sample, being much smaller in samples rom Southern Arica than it is in country samples rom East Arica. It is possible that this has something to do with dierences in survey sample designs. However, we also understand that the sample design is reasonably standardized across countries. It is thereore likely that the smaller extent o overlap in labor productivity across the ormal-inormal divide in Southern Arica relects dierences in the structure and workings o the economies o the two regions. Speciically we hypothesize that it may indicate that governments in Southern Arica enorce business tax laws and codes o regulation more than their counterparts do in Eastern Arica and that the ormal environment provides better services. Figure 2: South Arica Kernel Density Estimate o Value Added Per Worker South Arica Figure 3: Botswana Kernel Density Estimate o Value Added Per Worker Botswana Density Inormal microenterprises Formal small enterprises Formal microenterprises Density Formal small enterprises Inormal microenterprises Formal microenterprises Figure 4: Namibia Kernel Density Estimate o Value Added Per Worker Namibia Figure 5: Kenya Kernel Density Estimate o Value Added Per Worker Kenya Density Inormal microenterprises Formal microenterprises Formal small enterprises Density Formal microenterprises Inormal microenterprises Formal small enterprises

8 8 Figure 6: Tanzania Kernel Density Estimate o Value Added Per Worker Tanzania Figure 7: Uganda Kernel Density Estimate o Value Added Per Worker Uganda Density Inormal microenterprises Formal microenterprises Formal small enterprises Density Inormal microenterprises Formal microenterprises Formal small enterprises For the three countries in Southern Arica, the dierence in labor productivity between ormal and inormal micro-irms is especially stark or South Arica. There the density unctions barely overlap, and productivity in the modal ormal irm is several times higher than in the modal inormal irm. Indeed, in South Arica there almost no dierence between the productivity density unctions or micro ormal irms and small ormal irms. In Botswana and Namibia, the picture is muddier, yet with some similarities. In contrast, the East Arican pattern is one with a large productivity overlap between irms that are inormal versus those that are ormally registered microenterprises. In Kenya there is some dierence between ormal and inormal micro densities, but there is a great deal o overlap in Uganda and Tanzania; indeed, there is little or no dierence between the densities or ormal and inormal micro-irms in these countries. In Uganda the overlap is almost total across all three types o irms. In Tanzania and particularly Kenya, small irms tend to be somewhat more productive than micro irms. Figure 8: Rwanda Kernel Density Estimate o Value Added Per Worker Rwanda Density Inormal microenterprises Formal small enterprises Formal microenterprises

9 9 Finally, Rwanda is clearly a special case (Figure 8). As in Southern Arica, productivity o inormal micro-irms is lower than or the ormal micro and small irms. However, because o the stringent enorcement o registration requirements and punitive penalties or non-registration, there are almost no real inormal micro irms in Rwanda. The survey ound only 23 such irms, all very small and engaged in ringe activities. Alone among the countries, in this case the question o whether or not to register cannot be regarded as a practical choice; the question appears to be more one o whether or not to establish a business A Simple Framework or Understanding Productivity Dierences Perhaps the most visible proponent o the cost-beneit view o ormalization is Hernando de Soto. In his seminal analysis, The Mystery o Capital, he argues: "Extra legal businesses are taxed by the lack o good property law and continually having to hide their operations rom the authorities. Because they are not incorporated, extralegal entrepreneurs cannot lure investors by selling shares; they cannot get low interest ormal credit because they do not even have legal addresses. They cannot reduce risks by declaring limited liability or obtaining insurance coverage. In act, the only 'insurance' available to them is that provided by their neighbors and the protection that local bullies or maia are willing to sell them. Moreover, because extralegal entrepreneurs live in constant ear o government detection and extortion rom corrupt oicials, they are orced to split and compartmentalize their production acilities between many locations, thereby rarely achieving important economies o scale. With one eye always on the lookout or police, underground entrepreneurs cannot openly advertise to build up their clientele or make less costly bulk deliveries to customers." The question then is why irms choose to endure this situation. DeSoto concluded that inormal actors remained out o the ormal economy due to the cost imposed by the regulatory structure within the country (DeSoto, 2000). Increasing unemployment, either through loss o public sector jobs or a growing labor orce, has also been cited as a actor or the growth o the inormal economy, which serves as a social saety net or the 3 unemployed. Whatever the mix o reasons, the inormal economy in Arica is indeed large, both in terms o the number o enterprises as well as its contribution to GDP. Schneider estimates it to be about a third o total economic activity (Schneider, xx). The past ew decades have witnessed the emergence o a large volume o literature on the inormal sector. The literature is quite diverse, covering inormal labor (Almeida, 2005; Chen, 2004), tax policy (Ihrig and Moe, 2004; Emram and Stiglitz, 2004), as well as corruption (Johnson et al, 1998; Marcouiller and Young, 1995). Many researchers argue that irms locate in the inormal sector because the manager/entrepreneur believes that the 2 Appendix 2 shows the kernel densities, pooled by region. 3 Recent unemployment estimates or the seven countries in this study range rom 23% (or Botswana) to 56% (or Kenya).

10 10 beneits o inormality outweigh the costs (Djankov et al 2002; Loayza et al 2005; Ishengoma and Kappel 2006). Djankov et al enumerate beneits and costs o inormality and conclude that regulatory barriers must be lowered to encourage ormalization. Gatti and Honorati look at inormal irms in 49 developing countries to investigate the role o access to credit and external inance. The authors ind that more tax compliance is signiicantly associated with more access to credit and that the link between credit and ormality is stronger in high-ormality countries (Gatti and Honorati, 2008). Arterido et al ind that a weak business environment shits downward the size distribution o irms and that poor quality inance and business regulation reduces the employment growth o micro and small irms in particular. They argue that signiicant reorms are needed to or micro irms to cross over into the category o SMEs (Arterido et al, 2007). Loayza et al discuss the inormal economy as a result o excessive taxes and regulation. They study the determinants and eects in an endogenous growth model and ind that the size o the inormal sector depends positively on the proxies or tax burden and labor restrictions and negatively on a proxy o the quality o government institutions. This model was tested using country level data in Latin America (Loayza et al 2005). Ishengoma and Kappel (2006) set up a ormalization model around the costs and beneits in the business environment (ormal and inormal). They then assess approaches to encourage ormalization in this paradigm and make recommendations or speciic measures. Implicit is many o these analyses is the notion that inormality impedes growth and acts as a drag on productivity. 4 Are regulatory costs the driver o the decision to ormalize? I so, why do we observe so little dierence in productivity between inormal and ormal irms in East Arica relative to Southern Arica? In this section we spell out a partial equilibrium model o business inormality that provides an explanation o the dierence in labor productivity between inormal and ormal irms in Southern Arica and the contrast it bears to East Arica, where the overlap between the densities o the ormal and inormal sectors is almost complete. The Model We begin with a two actor model in which actor prices are assumed to be given and ixed, along with the sorting o the workorce between employees and entrepreneurs. Our ocus is on the sorting o observed entrepreneurs between the ormal and the inormal sectors rather than the question o selection into entrepreneurship which was the main 4 Fachamps (1994) discusses six actors that may govern this choice. Fachamps argues that the observed inormality is only a short-run disequilibrium phenomenon; however, given that the number o irms in each o these countries in this sector has grown rapidly, this is unlikely to be the case. High transport costs may limit a irm s market, it thereore produces on a small scale. However, this alone, cannot determine inormality. Market ailures, inormation asymmetries and management requirements play a role as well-- in each case, the demands are ewer on inormal enterprises. Government policies and regulations such as registration procedures, costs, tax laws, labor regulations, worker saety laws can be avoided by inormal irms. Inormal irms, with lexible technologies, can adjust more easily to market demand. And inally, large scale production requires managerial skills that entrepreneurs may not have.

11 11 ocus on the Lucas model. In the next ew paragraphs we describe our model and the key propositions that are derived rom it. The model is based on Lucas s theory o the size distribution o business irms (Lucas, 1978) and its extension in Rauch (1991) into a theory o inormality, and builds on Fortin et al. (1997), who introduce direct taxes and endogenous concealment costs into Rauch s model. 5 Appendix 3 describes each o the propositions in detail. As in Lucas (1978), society s workorce consists o a continuum o agents endowed with managerial talent level, x, drawn rom a ixed distribution Γ : R + [0,1]. A irm in the economy consists o a single entrepreneur, managing n identical employees, who use k units o homogenous capital to produce y units o net output by means o a two-level production unction given by y = xg[ ( n, k)], where, g : R + R + and each o and g is twice dierentiable, increasing and strictly concave, with g ( 0) = 0. We also assume that the production technology is one o constant returns to scale so that, k denoting actor proportion r =, ( n, k) = nφ( r), where φ : R + R + is also twice n dierentiable, increasing and strictly concave. Firm ace an identical wage rate, w, in the labor market and a uniorm rental price or capital services, u, and are assumed to maximize proits in their production and hiring decisions. Following Fortin et al. (1997), we assume that all irms are required to pay a proit tax at the uniorm rate o t π and a payroll taxt w. Firms may evade both taxes subject to a cost o evasion (or concealment cost) that is increasing and convex in the irm s employment ' '' size. The concealment cost can be described as c = c(n) such that c ( n), c ( n) > 0. We assume that all ormal irms pay both taxes while all inormal irms evade both. The degree o convexity varies with the overall strength with business environment. It is greater in countries where regulations and laws are better enorced, both because detection is more diicult to avoid and because the opportunity costs o inormality, in the orm o access to services enabled through ormalization, are higher. It ollows rom our assumptions that or any irm x, proits are given by 5 An even more recent extension o Rauch s model o inormality is de Paula and Scheinkman (2007) in which the distortions generating inormality consist o indirect taxes and credit rationing. Microeconomic models o inormality similar to these but outside o the Lucas ramework include Azuma and Grossman (2002), and Straub (2005), and Erosa and Cabrillana (2008). These are all models o selection on some endowment on a scare actor o production 5, the most abstract ormulation in these terms being Azuma and Grossman (2002), which does not even speciy what that actor is. In Straub (2005) selection is on initial wealth as a actor input as well as potential collateral or external inancing. In all the others selection is ultimately on total actor productivity, the sources o inter-irm productivity dierences being unspeciied in Erosa and Cabrillana (2008). See also Loayza (1996) or a model o inormality rom a perspective dierent than those described above.

12 12 (1a) = ( 1 t )[ xg( nφ( r)) (1 + t ) wn uk] π π i the irm operates ormally, and by w (1b) π i = xg( nφ( r)) c( n) wn uk i x is choose to be inormal instead. I x is in the ormal sector, its input demands would maximize (1a), or which the irst order conditions are ' (2a) xg [ n ( r )][ ( r ) ' φ φ rφ ( r )] = (1 + t ) w w and ' ' (3) xg [ nφ ( r)] φ ( r) = u, and rom which it ollows that (4a) ' φ( r) rφ ( r) (1 + t ) w w = ' φ ( r) u Equation (4a) deines the actor proportion, r as an implicit unction o actor prices and shows that ormal sector irms should exhibit uniorm capital intensity under our technological assumptions. The irm s labor demand unction is implicitly deined by each o equations (2a) and (3), and is o the orm n = n ( x, t.because o the uniormity o r, the proit maximizing demand or capital services is k = k ( x, t = rn ( x, t. Employment size is thereore a reliable indicator o the scale o output o ormal irms. Since the cost o concealment is a unction o irm size, or inormal irms (3) will still be one o the irst order conditions or its input demands, but equation (2a) will not be. The other irst order condition will be ' ' ' (2b) xg [ nφ ( r)][ φ( r) rφ ( r)] = w + c ( n) Upon division by (3) this leads to (4b) ' ' φ( r ) rφ ( r) w + c ( n) = ' φ ( r) u which shows that r is not uniorm or inormal irms since it depends on n and, hence, on x. Still, (2b) and (3) together implicitly deine the two actor demand unctions with x, u as the sole arguments when x is an inormal irm. The labor demand unction is now o orm ni = ni ( x,, to which corresponds an capital input demand unction o the orm k k ( x,, i = i

13 13 The labor demands n and n i have the standard property o strictly decreasing in w and u. They are also both strictly increasing in x. The result in Lucas (1978) that more talented entrepreneurs run larger businesses thus holds in this setting too, applying in the inormal sector as well as in the ormal sector. There may be an additional sorting mechanism through the labor market. In the Lucas model and its derivatives, labor is homogeneous with respect to productivity as an employee: the only inter-personal dierences are those relating to management capability. In practice however some attributes that will make an individual a more productive manager, such as education, will also make her a more productive employee. Especially with a steep ormal-sector skill gradient or wages, there will thereore be another selection process operating on managers through the labor market. I an inormal business cannot generate enough surplus to compete with ormal employment, there will be ew well-educated inormal entrepreneurs. This might be considered as another opportunity cost o inormalization, and the convexity o the cost unction will depend on the wage premium or an educated individual. In East Arica, it may be the case that the gradient is not as steep as in the South, possibly because the ormal sector is not big enough to absorb all educated applicants so that many o these become inormal entrepreneurs. Fox and Oviedo (2008) show or example, that the wage dierence between unskilled and skilled production workers is much higher in Namibia and Botswana versus Tanzania and Uganda. Clarke et al (2007) show that the median monthly wages in South Arica are high compared to Brazil, China and Poland in all employment categories and that managers in South Arica are particularly highly paid relative to other job categories when compared to their counterparts in the other three countries. Substituting or input demands into equation (1) gives entrepreneurial rent in ormal sector as π = π ( x, u, t tπ ).This is a standard proit unction except that managerial talent x (scaled down by proit taxes) plays the role that output price would in an ordinary proit unction, as we are using output as the numeraire. Proits are thereore strictly increasing in x by the envelope theorem and strictly decreasing in both actor prices and the two tax rates. Thus, not only do more talented managers run larger businesses, they also earn more entrepreneurial rent than those who run smaller irms. Obviously these properties also carry over to the entrepreneurial rent unction o the inormal sector. This is obtained by substituting or labor and capital input n i ( x, and k i ( x,, respectively, in equation (1b), which leads to π = ( x, u, c(.)) i π i Each o π and π i is deined or all entrepreneurs x : x x0 or each o which obviously only one o the two variables is realized. We assume that the other is nonetheless known to the entrepreneur as a notional value and that x is in the ormal sector i π ( x, u, t, tπ ) π ( x, and in the inormal sector otherwise. w i

14 14 This is the sorting rule motivating the econometric estimations that are presented in Section 4. The 8 key propositions o our model are as ollows: Proposition 1: The entrepreneurs with the greatest managerial talent are always in the ormal sector. Proposition 2 : Higher taxes would push more talented entrepreneurs into the inormal sector. Proposition 3: Inormal irms are always smaller than ormal irms. Proposition 4: There is discontinuity between the size distribution o ormal irms and inormal irms. Proposition 5: Inormal entrepreneurs earn less than ormal entrepreneurs. Proposition 6: Inormal irms are less productive than ormal irms. Proposition 7: The distribution o labor productivity o irms in the ormal and inormal sectors overlap with each other. Proposition 8: The extent o this overlap decreases in the convexity o concealment costs in the inormal sector. 4. Sorting Mechanisms Human Capital We now turn to empirical tests o the propositions identiied by our model. First, let us look at whether there is sorting by dierences in the level o managerial talent, as proxied by educational attainment. Figure 9 compares the educational attainments o the entrepreneurs in the inormal and ormal sectors across the six countries. (Appendix 4 shows educational attainment in all categories or inormal and ormal microenterprises). We are ar more likely to see owners with university degrees operating irms in the inormal sector in East Arica than in Southern Arica, where irms are clearly sorted according to managerial talent.

15 15 Figure 9: Pct o Firm Owners with Primary or University Education Inormal entrepreneurs have lower levels o education than those in ormal irms in all countries except Tanzania and Uganda where the distinction is insigniicant. The sharpest dierence by ar is in South Arica; it is astonishing that there, despite the relatively strong human capital base o the country, inormal entrepreneurs have lower levels o education than comparable businessmen in any other country. One possible actor is that the opportunity cost o being an entrepreneur in the inormal sector, in the orm o a high ormal sector wage, is particularly high in the South where skill-based wage gradients are relatively steep. Namibia comes second; there is also a sizeable gap in Botswana and Kenya. Race is not a direct sorting mechanism in our analysis because it is a proxy or capability-related variables, in particular education. Histories o race discrimination will have ed into the education system and may also aect the nature o the business environment (or example, access to inance or other services) across or within the countries in our sample--our model does allow us to investigate the direct eects o these variables on the decision to ormalize. Furthermore, the vast majority o microenterprises our sample are Black-owned (Table 2). South Arica has the smallest share o micro ormals (67 percent) that are Black-owned while Uganda has the highest at 100 percent.

16 16 Table 2: Share o Black-Owned Firms in the Sample Inormal micro Formal micro Formal small Botswana Kenya Rwanda South Arica Tanzania Uganda Namibia From other studies across Arica, education is a key driver o rates o irm growth, except or certain networked irms owned and operated by ethnic minorities (Ramachandran and Shah, 2006). As discussed urther belo there is also a strong relationship between educational attainment and productivity in Southern Arica, as well as between both o these variables and the likelihood o being ormally registered. In contrast, there is little dierence in education levels between ormal and inormal irms in Tanzania and Uganda, and also little productivity gap. The model described in the previous section assumes that concealment costs are convex in employment size, so that the average as well as the marginal concealment cost will increase as a irm gets larger. The degree o convexity will depend on the overall strength o the business environment, as relected in the enorcement o business tax laws and codes o regulation as well as in the higher opportunity cost o ormality (the provision o public services such as electricity etc). To the extent that enorcement and the provision o public services are characterized by a high degree o arbitrariness and variability, the concealment unction will be less sharply convex. Thereore, it would pay or an entrepreneur in one country to remain inormal while a similar entrepreneur in another country would be better o by operating ormally. The productivity and education comparisons suggest that the decision to register ormally in East Arica might be a somewhat random or idiosyncratic one, as costs and opportunity costs o ormality remain low across the size distribution o irms. The beneits o ormal registration might not be realized because o poor delivery o inancial or other services, or because the main concern o businesses such as reliable power is similar or ormal and inormal irms. Conversely, the costs o non-ormalization might not be so high either, perhaps because inspectors are equally likely to seek bribes whether a irm is ormal or inormal. Other work (Emery, 2003) has drawn attention to a high degree o arbitrariness in the business climate in many Arican countries. In contrast, i the business climates o the Southern Arican countries are relatively strong (taking into account enorcement as well as the de jure constraints tabulated in Doing Business); individuals with the capabilities to beneit rom the opportunities opened up by ormalization will tend to ormalize.

17 17 Beneits (or Opportunity Costs )o Formality The beneits o ormality lie in the access to public services and business support services, access to ormal banking sector or micro inance, and availability o basic inrastructure acilities such as electricity, telephone, public sewage, as well as premises. In countries where these services are well developed, we would expect the inormal sector to be smaller, and a sharper dierence between the perormance characteristics o inormal versus ormal microenterprises, ceteris paribus-only irms that have very low productivity/proitability will choose inormality. Examining reported inrastructure access (electricity, water and sewage), presented in Figures we see that or irms in all six countries, inrastructure constraints are in act greater or inormal irms, compared to ormal enterprises. Inrastructure access could provide incentives or irms to ormalize. The beneits to ormalization are clearly higher in terms o inrastructure or Southern Arican irms but much less so in Uganda and Tanzania. In Kenya, the dierence in access to inrastructure is airly large. Figure 10: Pct with electrical connections Figure 11: Pct with water connections Figure 12: Pct with public sewage acilities Figure 13: Pct with externally-inanced working capital Uganda Botswana Namibia Formal Inormal Tanzania South Arica Kenya

18 18 Figure 14: Pct with a bank account The data on access to inance also show some dierences between the two sets o countries. In all o the East Arican countries, access to a loan is actually higher or inormal irms than or such irms in any o the Southern Arican countries. Figure 13 shows that access to external sources o inance or working capital is higher or inormal irms in East Arica than in Southern Arica. Figure 14 shows that a large percentage o irms in Uganda and Tanzania have bank accounts even though they are in the inormal sector. Finally, Table 3 shows that inormal microenterprises inance a much higher share o working capital externally in East Arica than in Southern Arica. Table 3: External sources o working capital, as share o all working capital Percentages Formal microenterprises Formal small enterprises Inormal microenterprises Botswana Namibia S. Arica Kenya Tanzania Uganda Costs o Formality A key actor driving the decision to become ormal is the cost o ormality. A irm with a given level o productivity may choose inormality in a country which has higher tax rates and much greater compliance costs, compared to countries where such costs are low.

19 19 Figure 15: Pct reporting bribe payments Figure 16: Pct visited by tax inspectors Figure 15 shows that a very high share o both ormal and inormal irms in Kenya and Uganda pay bribes and a lower but equal share o irms in the ormal and inormal sector in Tanzania pay bribes as well. In South Arica, the share o irms making bribe payments in the inormal sector is higher than in the ormal sector, suggesting that the rule o law provides a degree o protection to ormal irms. 6 Figure 16 indicates that visits rom tax inspectors are likely to be higher or ormal irms in every country but the share o irms visited in East Arica is much higher than in Southern Arica. Kenyan irms are equally likely to be visited, whether or not they are in the inormal sector. In Southern Arica, harassment by tax inspectors is a lot lower, especially or irms in the inormal sector. Country indicators are consistent with our indings that countries within the Southern and Eastern Arican block dier considerably in their business environments. Botswana, Namibia and South Arica are middle income countries, with relatively good governance and regulatory environments, while countries in East Arica are lower income, with mixed regulatory and governance conditions. Table 4 shows a number o indicators that might be relevant to understanding the business environment across the seven countries in our sample. Overall, these external sources o data corroborate the country-level dierences that we identiy in our descriptive charts above. 6 Kaplinsky 1995 and Manning and Mashigo 1993 note that both small inormal irms and owneroperators are particularly likely to suer rom violence and insecurity in South Arica.

20 20 Table 4: Indicators o the Business Environment GDP per Capita Doing Business Rankings (2007) Institutional Investor Credit Rating Voice and Accountability (2007) Government Eectiveness (2007) Control o Corruption (2007) Botswana Namibia S. Arica Kenya Tanzania Uganda Rwanda Sources: World Bank Doing Business Indicators, Kaumann, Kraay and Mastruzzi Governance Indicators(scale o -2.5 to +2.5), Institutional Investor. South Arica, Botswana and Namibia consistently score well in terms o governance dimensions relative to the East Arican group. Rwanda emerges as a distinctive case high scores relative to East Arica in dimensions related to government eectiveness and control o corruption but low in the area o voice and accountability a pattern very consistent with the tight control over inormal business noted above. Similarly, the Doing Business indicators suggest a higher quality business environment in the Southern countries. 5. The probability o ormalizing In this section, we summarize the results o Probit models that estimate the likelihood o ormalization, (presented in Appendix 5). Inequality (6) rom section 3 provides a π D = π ( x, u, t tπ ) π i ( x, structure or the Probits, with the dierence providing the latent variable underlying the ormality dummy D such that D = 1 or irms or which π D 0 and D = 0 i π D < 0, where we assume that we can express π D as a stochastic unction o a vector o observable determinants or proxies, Z, o management talent, x, actor prices, w and u, and tax rates, t w and tπ such that π Di = γ ' Z + ε i, where ε i a random error tem orthogonal to Z and summing up a host o other unobservable determinants o π D. Table 5 shows the signiicance o size, education, and measures related to the business environment in determining ormality. Overall, the empirical results are consistent with the predictions o the model. The probability o being ormal is correlated with irm size in every country except Rwanda. Managerial talent (as proxied by educational attainment) is signiicant in determining likelihood o registering, ceteris paribus, or

21 21 irms in all the countries in Southern Arica and Kenya. But it does not play a role in determining ormality in Tanzania, Uganda or Rwanda. Female ownership is correlated with inormality in only two countries Namibia and Rwanda in all others, it is not a signiicant driver o irm status. The beneits o ormality (access to public services, external inancing) are clearly driving the decision to ormalize in Southern Arica and Kenya, while in Tanzania and Uganda, they are not likely to drive the decision. Interestingly, in Namibia, Uganda, Kenya, and Rwanda the likelihood o bribe payments increases the chances o ormalization suggesting that ormalization does not necessarily increase access to the protection o the law. In South Arica, however, the opposite is true. The regressions also conirm that access to inance is likely to be a signiicant driver o the decision to ormalize, except in Tanzania and Uganda. Table 5: Determinants o ormalization Botswana Namibia South Arica Tanzania Uganda Kenya Rwanda Firm size X X X X X X Inspector X X X visits Access to X X X X public services Owner has university education X X X X Owner is X(-) X(-) emale External X X XX X XX inancing Bank account X X X X Pays bribes X X(-) X X X Note: X notes signiicance at either the 5 percent or the 1 percent level. XX notes cases where very ew inormal irms have access to external inancing and a coeicient cannot thereore be estimated. 6 Conclusion This paper has shown that the relative characteristics o inormal and ormal irms can dier across countries, by showing the distinct patterns that emerge rom an analysis covering East and Southern Arica. Many studies simply assign irms to ormal and inormal categories based on their size. Here we are able to assign micro irms (those with less than 5 employees) into ormal and inormal categories on the basis o tax registration, and also to compare them with small ormal irms with between 5 and 10 employees.

22 22 In Southern Arica, and particularly in South Arica, there is a sharp biurcation o the micro irms. Formal micro-irms have high labor productivity, at very similar levels to the small ormal irms. This suggests that some sorting mechanism is at work, creating incentives or some entrepreneurs and their irms to operate in the ormal sector. These ormal micro-entrepreneurs are likely to have the capabilities to grow their irms, at least into small businesses; indeed, a survey suggests considerable upward size mobility within the South Arican ormal sector (Clarke et al, 2007). Inormal irms in Southern Arica on the other hand have very low labor productivity, as little as one-tenth that o the ormal sector. We believe that business environments are less likely to be the cause o this biurcation. Compliance costs tend to be lower and inrastructure and inancial services are more likely to be delivered in the South i irms become ormal. Indeed, vulnerability to bribe-taking oicials (which is sometimes cited as a disincentive to ormalization) is greater among inormal irms. Yet is seems that inormal orms are not ully denied services; a ew use banking services and other amenities. What does this mean or the inormal irms in Southern Arica? Our data suggest that, irst, it is vital to distinguish micro-irms rom inormal irms. Many o the latter ace internal barriers, in terms o management abilities and productivity, that will make it very diicult to cross over the productivity chasm that separates inormal rom ormal irms. In South Arica or example, not one single irm in the inormal sample was run by a university graduate. In addition to the other sorting actors, the opportunity cost o being a ormal sector employee will be high or such individuals. This is in marked contrast to East Arica, where many inormal irms are run by individuals with relatively high educational qualiications. For many irms in the inormal sector in Southern Arica, it is not clear that they would be able to take advantage o the potentially expanded range o opportunities that could be expected to come rom ully ormal status. While there will always be exceptions, the inormal sector in Southern Arica captured by the survey appears to be essentially composed o survivalist irms. What might help these irms? Access to training programs and simple technical assistance may help to some degree. One o the areas that the South Arican inormal irms complain about in the survey is lack o access to premises. Rogerson 1992 notes the legacy o slum clearance and other periodic attacks on the illegal spaces within which inormal enterprises thrive. While many restrictions have been progressively lited, there were still lingering controls and repressive bye-laws at the time o the survey. In South Arica, the survey data show that about one-ith o inormal irms see land (premises) as a serious problem while almost none o the ormal micro irms see this as problematic. Some provisions here might ease their survival. But expanding employment in the ormal sector, including in the many high-productivity micro-irms that are shown to exist by the survey, might be more useul than trying to grow inormal irms. Whether such eorts should include entry wage subsidies or other approaches is beyond the scope o this paper. Given the low educational status o most inormal entrepreneurs, it is

23 23 diicult to see a route to higher-productivity employment that does not include the building o human capital, though this will be a lengthy process. The picture is quite dierent in East Arica. There, especially in Uganda and Tanzania, there are indications that essentially similar irms make dierent decisions on whether to become ully ormal. The inormal sector includes many higher-productivity microenterprises with the capabilities to ormalize and grow; the entrepreneurs running these irms are oten well-educated and have access to a range o business services. Why these irms make the decisions they do plausibly relates to the business environment and its various weaknesses. Our analysis suggests that there may well be gains rom improving the business environment to crowd in these irms into the ormal sector and to widen the pool o ormal micro- irms that can serve as a seedbed or small ormal irms and beyond. This is not to say that eorts in these countries should not also include the building o human skills. From these eorts, we would also, however, expect duality to emerge. In a stronger business environment, the inormal sector will more likely be composed o survivalist irms. The case o Rwanda raises a distinctive question. How stringent should ormalization requirements be at an early stage o development? Does a stringent registration and tax policy lead to a well-regulated and increasingly productive micro-seedbed or growth? Or, will it stultiy the growth o the entire small-scale business sector? Only time will tell.

24 24 Appendices Appendix 1: A brie note on sampling Survey coverage The World Bank Enterprise Surveys in Sub-Saharan Arica cover establishments located in all major regions o a country, in the ollowing industries (according to ISIC Revision 3.1)--all manuacturing sectors (group D), construction (group F), retail and wholesale services (sub-groups 52 and 51 o group G), hotels and restaurants (group H), transport, storage, and communications (group I), and computer and related activities (sub-group 72 o group K). For establishments with ive or more ull-time permanent paid employees, this universe was stratiied according to the ollowing categories o industry: 1. Manuacturing: Food and Beverages (Group D, sub-group 15); 2. Manuacturing: Garment (Group D, sub group 18); 3. Manuacturing: Other Manuacturing (Group D excluding sub-groups 15 and 18); 4. Retail Trade: (Group G, sub-group 52); 5. Rest o the universe o irms, including: Construction (Group F); Wholesale trade (Group G, sub-group 51); Hotels, bars and restaurants (Group H); Transportation, storage and communications (Group I); Computer related activities (Group K, sub-group 72). The survey also sampled a selection o micro establishments (establishments with less than ive ull-time permanent paid employees) rom the targeted universe, without stratiication by industry. Sampling methodology The sampling methodology implemented varied by strata and region, depending on the characteristics o the population. With respect to establishments with ive or more employees, the ollowing approaches were applied (according to survey region and industrial stratum): For the three manuacturing strata, a required second level o stratiication (ater sector) was irm size, deined as: small (5-19 employees), medium (20-99 employees) and large (100 and more employees). In all surveys, the micro establishment stratum covers all establishments o the targeted categories o economic activity with less than 5 employees. For many reasons including the small size o establishments, their expected high rate o turnovers, the high level o

25 25 inormality o establishments in many activities and consequently the diiculty to obtain trustworthy inormation rom oicial sources, EEC Canada selected an aerial sampling approach to estimate the population o establishments and select the sample in this stratum or all regions o the survey. The inal dataset contains inormation on sample weights or this group, which approximates the number o micro-enterprises ound in the population. Appendix 2: Pooled Kernel Densities or East and Southern Arica Figures 1 and 2 show the kernel density estimations aggregated across East and Southern Arica. The picture is clear in East Arica, inormal and ormal irms have more or less the same level o productivity; in Southern Arica, the picture is starkly dierent. It is also interesting to note that inormal irms in Southern Arica look very similar to inormal irms in East Arica, but ormal irms are ar more productive in Southern Arica than in East Arica. Figure 1 Figure 2

26 26 Appendix 3: Propositions derived rom the model o ormality The rent unctions π ( x, u, t tπ ) and π i ( x, are deined or all entrepreneurs x : x x0. Although or each only one o π or π i is realized or any one entrepreneurs, we assume that the other is also known to the entrepreneur as a notional value, and that that x is in the ormal sector i π ( x, u, t tπ ) π i ( x, and in the inormal sector otherwise. Given this sorting rule and our other assumptions, the ollowing propositions can be proved about the size distributions, relative earnings, and TFP and labor productivity gaps o the ormal and inormal sectors o the economy. 7 Proposition 1: The most talented entrepreneurs are always in the ormal sector For any given set o actor prices and tax rates, there is a threshold management talent level, x = xˆ, deining the irm at the margin o ormality and inormality such that (5) π ( xˆ, u, t, tπ ) = π ( xˆ, Moreover, w (6) π ( x, u, t, tπ ) π ( x, i x xˆ and w π ( x, u, t, tπ ) < π ( x, i x < xˆ w i i i In other words, every entrepreneur in the inormal sector has less management talent than every entrepreneur in the ormal sector. Proposition 2 : Higher taxes would push more talented entrepreneurs into the inormal sector Equation (5) deines the talent level o the marginal ormal sector entrepreneur, x = xˆ, as an implicit unction o w, u, t and t π, which unction can be dierentiated to show that xˆ increases in each tax rate t w and t π. It ollows that, the higher is either o these tax rates, the higher the share o the inormal in the total population o irms, and the higher is the average entrepreneurial talent in the inormal sector. Proposition 3: Inormal irms are always smaller than ormal irms 7 Proos are available rom the authors.

27 27 This is in the sense that every inormal irm is smaller, in employment terms, than every ormal sector irm. The minimum irm size in the ormal sector is n ( xˆ, t while the largest inormal sector irm has employment size n i ( xˆ, so that n ( x, t n ( xˆ, t or every ormal irm x and ni ( x, < ni ( xˆ, or every inormal irm. Proposition 4: The missing- middle hypothesis There is discontinuity between the size distribution o ormal irms and the size distribution o inormal irms. This is in the sense that the employment size o the marginal ormal sector irm x = xˆ is smaller than what the same irm would employ in the inormal sector, that is, n ( xˆ, t > ni ( xˆ,. In other words, there is a gap between the minimum employment size o ormal irms and the maximum employment size o inormal irms. Proposition 5: Inormal entrepreneurs earn less than owners in the ormal sector This is in the sense that proits per irm are lower in the inormal sector. For any irm x = x 1 observed in the inormal sector and any irm x = x2 observed in the ormal sector, we have π ( x2, u, t tπ ) > π i ( x1,. This is in act a corollary o Proposition 1, which implies that x < xˆ and x xˆ 1 2 so that π (,,,, ) ( ˆ,,,, x2 w u t w tπ π x w u t w tπ ) and π ( xˆ, u, t, tπ ) > π ( x1,. w i Proposition 6: Inormal irms are less productive than ormal irms This is in the sense that total actor productivity any inormal irms is lower than total actor productivity in any ormal irm. This too ollows orm Proposition 1. The Lucas model should be interpreted as one o selection (into entrepreneurship or inormality) on total actor productivity, given that management talent can only be partially observable. For example, under the assumption o Gibrat s Law (that irms grow at a uniorm mean rate independent o initial size) made in much o the Lucas (1978), the production 1 2 unction is o the orm: y = A( x) k λ n λ, where A( x) = αx is TFP, α > 0 and 0 < λ 1, λ2 < 1. Underlying the size distribution o irms in the baseline model is thereore an inter-irm dispersion in TFP, which is increasing monotonically over the size spectrum starting rom lowest observation or the marginal irm. Thus or any two irms x = x1 and x = x2, and a given set o actor prices w and u, their employment size ratio would be a monotonic transormation o their relative total actor productivities: 1 λ1 λ2 A( x2 ) n( x2 ) (7) = ( 1) ( 1) A x n x

28 28 Proposition 7: The densities o labor productivity in the ormal and inormal sector overlap Total actor productivity is unambiguously lower in the inormal sector than in the ormal sector in the sense that the A(x) o every inormal irm is smaller than that o every ormal irm. And in general labor productivity is higher in the ormal sector or that reason. However, there is signiicant overlap between the probability density o labor productivity in the ormal sector and the density o labor productivity in ormal irms due to the convexity o concealment costs. Proposition 8: The extent o overlap in the densities o labor productivity decreases in the convexity o concealment costs in the inormal sector The range o x (or subset o irms) over which the density o labor productivity in the ormal sector overlaps with the density o labor productivity in inormal irms depends on the curvature o the concealment cost unction, the range being smaller the more convex '' is the cost o concealment, that is the greater is, c ( n( x)). Arguably the curvature o the concealment cost unction will be greater the higher is government s ability or willingness to enorce tax laws. I that is indeed the case, then the degree o overlap in the density o labor productivity between the ormal and inormal sectors will be smaller where the enorcement o laws is stronger.

29 Appendix 4: Educational Attainment o Firm Owners 29