Policy Research Working Paper 8918

Informality, Harassment, and Corruption

Evidence from Informal Enterprise Data from Harare, Zimbabwe

David C. Francis

Development Economics Global Indicators GroupJune 2019

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Produced by the Research Support Team

Abstract

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.

Policy Research Working Paper 8918

New representative survey data for Harare, Zimbabwe are used to analyze the conditions under which informal businesses encounter requests for bribes. A simple model develops basic expectations of bribe exposure: those with higher opportunity costs of formalizing and with a higher ability to pay (resources) are expected to face a higher bribe risk. Empirical results find that informal businesses oper-ating out of necessity are likelier to have a bribe demanded

of them. Robustness checks are performed to account for the endogeneity of the main regressors, through a bivari-ate probit specification, using an instrumental variable of whether an informal business owner started in the area of activity before the country’s 2009 dollarization. Limited significant results are found for the effect of revenues (ability to pay).

This paper is a product of the Global Indicators Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at dfrancis@worldbank.org.

Informality, Harassment, and Corruption: Evidence from Informal Enterprise Data from Harare, Zimbabwe

David C. Francis1

JEL: D73, E26

Keywords: informality, corruption, bribery, harassment

1 World Bank, Development Economics, Enterprise Analysis dfrancis@worldbank.org

(2)

I. Introduction

In many economies, informality predominates: it provides an undeniably large source of

employment as well as economic activity. While a formal definition of informality can be hard to

come by, it is useful to distinguish between informal employment and those businesses that operate

informally. Informal employment can be defined as simply as work that is not covered by labor taxes

and public pension contributions. These jobs are almost certainly off-the-books but can exist in either

formal or informal firms. Somewhat apart, informal sector businesses operate outside of the official

channels by which formalized firms are registered, engage the legal system, pay taxes, and deal with

government officials. These businesses largely consist of self-employed operators, who occasionally

employ others, who are almost invariably informally employed. On its face, then, operating informally

can expose these businesses to frequent (but unstructured) encounters with police and government

officials. In some cases, these interactions leave informal businesses exposed to harassment and

demands for bribes. Yet while there are substantial literatures on both informality and corruption,

there are few attempts to join the two, particularly using micro-level data.2

One reason for this gap in the literature is a lack of adequate data sources. Those data that

are commonly available rely on household-level measurement, and in turn, focus on labor dynamics

between formal and informal employment. However, households do not necessarily encounter

bribery the same way that businesses do, even if those businesses are operated out of the home. This

paper leverages unique, survey data on informal sector businesses in Harare, Zimbabwe, collected in

2017. These data were collected using a geographically based, adaptive sampling approach and can

be considered representative. Observations are, by construction, centered on informal businesses, a

fact which makes the data well-suited to assess under what conditions those activities are exposed to

extractions of bribes and harassment at the hands of public officials. This is likely important when

assessing the informal sector’s exposure to corruption: operators of informal businesses may directly

face a demand for a bribe so that they can avoid a fine or being shut down. By contrast, the bribe

exposure associated with off-the-books labor is, on its face, more likely to be borne by the tax-evading

employer rather than the informal employee.

2 Lavallée and Roubaud (2018) provide one recent exception. Emerson (2006) models an informal-formal divide but is concerned with bribery at the entry of firms into the formal sector. Dutta et al (2011), similarly, discuss both informal and corruption, but they rely on generalized perceptions of corruption for Indian states.

(3)

While the existing literature on informal businesses’ exposure to bribery shows that business3

characteristics often matter, a clear framework is lacking. A first contribution of this paper is such a

framework, albeit a simple one. A second contribution is to provide evidence using a representative

and business-level data set; these survey-based data use a novel sampling approach, affording in-

depth, micro-level information on informal businesses, which are often absent from such firm-level

data sets. A third contribution is to provide empirical results of this exposure. Specifically, results

show that informal-sector operators with a higher opportunity cost of formalizing are more exposed

to an extortion of a bribe; a business’s ability to pay is found to have a limited effect. Two measures

of the opportunity cost of formalizing are used. First is a measure of whether the informal business

owner started the activity due to a lack of other income (NECESSITY); since this measure is self-

reported and opinion-based, an alternative measure is used. A second measure is if any member of

the owner’s household has employment under a contract (NOWAGE). Positive and significant

results for NECESSITY and NOWAGE are largely robust to methods adjusting for the potential

endogeneity of the regressors. These methods leverage instruments of whether the owner started in

the sector of activity before or after the 2009 dollarization of Zimbabwe’s economy.

The remainder of the paper is organized as follows: Section II summarizes the literature on

informality and corruption, section III discusses the background of these issues in Zimbabwe

specifically. Section IV presents a stylized model. Section V describes the data used, and section VI

gives the main econometric specification. Section VII includes the main results, and VIII concludes.

II. Literature on Informality and Corruption

Informality

A broad literature discusses the informal sector as a driving sector of economic activity and

employment. While this stylized fact remains largely undisputed, the characterization of the sector

itself is a matter of debate, particularly over the nature of informal employment. A first view is that

informality’s prominence is a result of market segmentation. From this perspective, informality

results from mostly insurmountable barriers to formal employment4 or to starting a business (De

Soto, 1989). In this sense, informality is not a free and willful decision, but a reflection of necessity

3 Stylistically the term “business” or “business activity” is used to describe informal “firms”; while the latter term “firms” is typical of some literature, it is generally avoided here, as it connotes a legal status more consistent with operating in the formal sector. 4 Amin (2009) summarizes several early works.

(4)

due to a lack of viable opportunities in the formal sector. For those in this camp, informality

represents a wealth of untapped and under-utilized resources, due to an inefficiently segmented

market.

A second view is that informality is an informed decision, undertaken for the avoidance of

taxes, fees, and regulations by operating outside of official channels. From this viewpoint, informality

presents a means to evade costs, and so it provides marginal benefits to employees and entrepreneurs

who would be unable (or unwilling) to conduct business formally. These informal operators will be

less likely to formalize if the benefits from formalizing are minimal (as in the case of a low-productive

formal sector or low expected payoffs to formality), and so the choice of informality may reflect a

lack of enticing opportunities in the formal sector (Maloney, 2004). This second view is consistent

with informality as an operational or individual choice. This choice implies some level of fluidity, and

thus informality can serve as a safety net (Loazya and Rigolini, 2011) or last resort (Günther and

Launov, 2012) in times of economic duress.

More recent work, along with more available household-level data, has shown the importance

of the heterogeneity of the informal sector (Cunningham and Maloney, 2001; Günther and Launov,

2012; Nguimkeu, 2014). Maloney (1999), for instance, finds that self-employed, informal business

operators are likely differentiated from those working informally; they are more akin to entrepreneurs

at the margin of the formal market and may represent a top “tier”. This is especially true if returns to

formal employment are particularly low relative to the off-the-books income from operating

informally (e.g. as shown by Falco and Haywood (2016) in Ghana). Various types of entrepreneurs

may be operating in the informal sector, demanding similar heterogeneity among informal sector

businesses (Amin, 2009). Businesses in the informal sector may also have varying levels of output,

rely on differing levels of human capital, and use public resources differently.

This heterogeneity can exist even within closely defined areas or neighborhoods. Indeed,

informality tends to be highly clustered, even within urban areas. A recent World Bank study using

city-based data for Kampala from the Uganda Bureau of Statistics found that the vast majority of

informal business activity was centered around three locations near the central business district, and

that 84% of informal businesses sold to customers thought to be within a 30-minute walk to their

business (World Bank, 2018). Informal businesses may also be differentiated by household-level

factors. Unsurprisingly, members of poorer households tend toward informality. At least some

(5)

evidence from Nairobi suggests that informal business operators from poorer households also have

lower performance (Gulyani and Talukdar, 2010).

Corruption and informality

A separate literature has also leveraged micro-level data to analyze the heterogenous

determinants and effects of corruption. Overall, corruption is expected to flourish in the absence of

good governance and to be ultimately distortionary (see, for instance, Schleifer and Vishny, 1993).

Nonetheless, there is less agreement under what circumstances firms must pay bribes, and the mirror

image of when officials demand those informal payments. Some analyses have found that rather than

‘grease’ transactions, bribes are extracted by officials who ‘sand the wheels’ of the provision of public

services, and thus use delays to extract rents from firms (Kaufmann and Wei, 1999). In fact, such a

delay-to-extract mechanism has been shown to operate across several types of transactions (Freund

et al, 2016). These models can be loosely grouped under a category of “bureaucrat as gatekeeper”,

where firms actively seek a government service, such as a license or permit, and are thus exposed

under certain conditions to a request for a bribe. Less frequently analyzed is the possibility that a

corrupt government official will actively seek out and harass firms to demand a bribe, what may be

termed as “bureaucrat as harasser”.5 Marjit et al (2000) propose a framework where such a mechanism

could occur in the context of tax auditors harassing tax-paying firms. It is not a far leap to extend

such a framework of harassment to extract a bribe to include the informal sector.

While discussions of both informality and corruption, separately, have been substantial, little

space has been dedicated to directly linking the two. This is somewhat surprising as informality and

corruption are two areas that seem sensibly related. Informal businesses, in all but the rarest of

circumstances, do not pay taxes. These businesses are not inspected nor audited in a conventional

sense. They do not obtain licenses or permits. If they do engage services—such as financial

intermediation—it is under their own personal auspices. It is, thus, reasonable to expect informal

businesses to be exposed to bribe seekers outside of the official channels where their formal-sector

counterparts face bribe requests. Yet nearly no studies—Lavallée and Roubaud (2018) are an

5 Kaufmann and Wei (1999) term their model as “harassment”. The distinction here is between a “gatekeeper” who maintains monopolistic control of a public good, and so must grant items such as a permit, license, etc. By contrast, a “harasser” actively approaches a business, possibly to extract a bribe. Harassment need not be the sole purview of officials in this sense—those businesses could just as easily be approached by entities such as a gang to be allowed to continue operating.

(6)

exception—discuss the direct experience of informal businesses dealing with bribery in these

situations.

Most existing work focuses on the interplay of the informal sector and the prevalence of

corruption. In a large cross-section of countries, Besley and Persson (2014) find that where the size

of the informal sector is larger, the associated share of tax revenues is also often lower. Smaller tax

coffers limit the money available for funding public administration and, thus, cut the money available

for public servant salaries. The link between reduced public salaries and corruption is both self-

evident and well-forged (see Besley and McLaren, 1993, for a discussion and more recently Olken

and Pande, 2012).

Choi and Thum (2005) develop a model where entrepreneurs move to the shadow economy

to avoid the extraction of bribes by public officials. In their model, corruption (at the point of

business registration) is a barrier to formal entry. They also consider corruption as depleting funding

to and provision of public services; in both ways, informality is freely chosen relative to the cost of

entry and expected payoffs from greater public good access. As such, graft is self-limiting: greater

corruption drives firms underground, which in turn, reduces the number of firms from which a rent

can be extracted. Counter to this prediction, some analyses, such as Buehn and Schneider (2009)

using multi-country data, find that higher levels of corruption and activity in the “shadow economy”

are complementary. Dutta et al (2011) provide similar complementarities between informality and

corruption levels in 20 states in India. Emerson (2006) adds a similar model, framing corruption as

reducing competition in the informal sector by relegating businesses to informality, providing

suggestive cross-country evidence of this relationship.

Most of the preceding analyses model or assume that corruption is largely only faced by firms

in the formal sector, or by extension, firms seeking to formalize. They can be grouped in the category

of “bureaucrat as gatekeeper”. However, the informal sector is also likely exposed to corrupt officials,

and such interactions can be considered functionally different than in the formal sector. Businesses

in the informal sector often lack redress if they are harassed for a bribe and, in a mirror image, corrupt

public officials are likely to encounter less punishment if they harass informal businesses for payoffs.

Alternatively, informal businesses can operate sporadically, over multiple locations and times, and are

often less visible, allowing them to avoid interactions with corrupt officials. Waller et al. (2002) and

Echazu and Bose (2008) model the behavior of bureaucrats (including corrupt and uncorrupt) who

have purview over both the formal and informal sectors. In these models, if the formal and informal

(7)

sectors are regulated jointly, bribery increases and social welfare declines. By contrast, if formal and

informal sectors are regulated separately (what Echazu and Bose term as “horizontal centralization”),

rent extraction cannot be coordinated and maximized, thus increasing social welfare. Both models,

however, do not include differential effects of bribery exposure within the informal sector.

Empirical evidence on the informal sector’s exposure to bribery is also lacking. The number

of studies using micro-level data to study corruption remains limited, and the share incorporating the

informal sector is smaller still. While not using data on the informal sector directly, Safavian et al.

(2001) study microenterprises and corruption in the Russian Federation and find that larger firms are

more likely to report constraints due to corruption. Svensson (2003) controls for firms in Uganda

paying taxes (a proxy for formality), finding that those that do are more likely to encounter a bribe

request. Rand and Tarp (2012) use data for Vietnam and include a sub-set of un-registered businesses

in their analysis of bribery incidence, which is found to be lower among informal businesses: they

speculate that this lower probability is due to the less visibility. A reading of the literature yields few

studies using micro-data for informality and bribery in Sub-Saharan Africa, specifically. Lavallée and

Roubaud (2018) are an exception. They use a sample of informal-sector businesses found through

household surveys6 in six West African countries. In line with results from analysis of the formal

sector, they find that a business’s ability to pay (proxied by sales per worker) is associated with a

higher likelihood of facing a bribe request.

III. Informality, Corruption, and Harassment in Zimbabwe

Informality and Employment in Zimbabwe

Zimbabwe’s statistical office (ZimStat) reports some figures on the informal sector in its 2014

labour force survey (LFCLS). According to the LFCLS, about 14% of employment is provided by

the informal sector (informal employment, by contrast, nears 90%, implying substantial off-the-

books employment elsewhere). ZimStat places unemployment in Zimbabwe at 11%, a contested

figure, as its definition of the labor force excludes informal operators and vendors not actively seeking

employment.7 Some accounts at the time of the LFCLS claimed that once this informal employment

was included, nearly 9 out of 10 Zimbabweans were unemployed.

6 They use data from surveys using what is known as the 1-2-3 method, which finds informal sector businesses by first interviewing a representative sample of households. 7 BBC, “Reality Check: Are 90% of Zimbabweans unemployed?” https://www.bbc.com/news/business-42116932

(8)

Trends in employment figures are also informative. The ILO8 offers a modeled estimate

(using labor force surveys) of the “vulnerable employment” rate, the share that are self-employed,

own-account workers, or contributing family members (ILOSTAT, 2019).9 Figure 1 reports this

measure in three series: for men, women, and overall, respectively. Each series reports vulnerable

employment as a share of each category’s total employment (i.e., as a share of female (male)

employment). These trends track with two salient breaks affecting Zimbabwe’s economy. First, in

2000, after a contentious election and political fallout—including widespread political violence—the

country’s economy entered freefall (Figure 2). Following a series of events over the next several years,

including the large withdrawal of international credit, the economy entered hyperinflation. In 2009,

the government abandoned its currency and dollarized, which contributed to stabilized prices and

was followed by rocky, but far less tumultuous periods of economic performance. Together, both

figures show a picture of substantial movement into informal employment after the 2009

dollarization, providing at least a partial pattern of the informal sector as a fallback source of income

under dire circumstances. Particularly relevant for Zimbabwe’s poor and vulnerable, coming in

between these two periods, was the widespread Operation Murambatsvina in 2005 (literally translated

as “trash removal”) program, which involved widespread and forceable clearing of several slums,

including in Harare. These forces, collectively, indicate this period as one of displacement and raised

insecurity among the country’s poor.

8 International Labour Organization. 9 Via the World Bank’s World Development Indicators.

(9)

Figure 1: Vulnerable Employment as Percentage of Total (1993-2017)

Source: ILO, via WDI. Each share is of the relevant total (Female to Female, e.g.)

Figure 2: Gross Domestic Product Per Capita (constant USD) (1993-2017)

Source: WDI

What characterizes Zimbabwe’s informal sector, given this context? Generally, informal

businesses have been found to be smaller and less productive than their formalized counterparts (La

Economic& Political

Crisis

Murambatsvina

Dollarization

Male

Female

Total

30

45

60

75

90

Per

cent

age

of E

mpl

oym

ent (

by to

tal a

nd g

ende

r)

1993 1999 2005 2011 2017

Year

Vulnerable Employment (%)

Economic& Political

CrisisMurambatsvina

Dollarization

0

500

1000

1500

1993 1999 2005 2011 2017

Year

GDP per capita (constant)

(10)

Porta and Shleifer, 2014). Yet study after study has found an overlap in the distributions of measures

of revenues and labor productivity, indicating that parts of both sectors are operating at the same

points in these distributions.10 While not definitive by any means, such over-lapping distributions

suggest that some portion of the informal sector is operating more freely out of choice (perhaps

avoiding a costly process of formalizing into a still-low-productive formal sector). The data for

Harare’s informal and formal sectors fit this pattern. Figure 3 shows kernel densities of the log of

labor productivity (in USD) for both formal and informal businesses. Since the values for informal

businesses are recorded as of the last completed month, this value is multiplied by 12.11 Values for

the formal sector are calculated from the World Bank’s Enterprise Surveys.12

Figure 3: The Relative Productivity of Harare’s Informal and Formal Sectors

Formal and informal sectors may vary in terms of the sectoral composition, which may

account for the differences shown in Figure 3. Table 1 does show that there is a notable difference

in the types of activities engaged in by the formal and informal sectors. Nearly half of Harare’s

informal sector is dedicated to food provision (sales of food, beverages, and bakeries). Only 10% of

10 Graphs in this vein have made frequent appearances in the literature on informality, including Ulyssea, 2018. Graphs produced in Stata using Daniel Bischof’s plotplain scheme. 11 The densities shown are for the log of revenues+1, as last month’s sales can be 0. 12 www.enterprisesurveys.org. The typical Enterprise Survey (ES) includes only enterprises with 5+ employees, but a supplemental survey for micro (1-4 employees) was conducted at the same time in 2017. Calculations use both micro and typical ES.

0

.1

.2

.3

0 5 10 15

Formal

Informal

Harare only. Log of revenues with adjustment. Informal monthly values multiplied by twelve to annualize value.

Labor Productivity, Sales per Worker (log)

(11)

the formal sector in Harare undertakes these same activities. The city’s formal establishments tend to

be involved in non-food retail. To test if these sector composition effects are driving the observed

productivity differences, survey-weighted regressions accounting for these sector fixed effects were

also run. OLS and quantile regression results confirm that on average, informal sector businesses are

88% or 86% less productive, respectively.13

Table 1: Formal and Informal Sector Activities, Share

Formal Informal Manufacturing 18% 17%

Food provision 10% 50% Retail 42% 17%

Other services 30% 16%

Note: weighted estimates. Food provision includes retail sale of food, restaurants, bakeries, and beverage production

Bribery in Zimbabwe

Using globally comparable indices, Zimbabwe fares poorly in terms of corruption.

Transparency International’s Corruption Perceptions Index for 2017 ranked Zimbabwe 157th out of

180 countries, tied with Burundi, Haiti, and Uzbekistan. Further surveys by Transparency

International, from 2015, report that roughly one in five respondents report encountering a bribe

request when soliciting public services. According to the same Transparency International survey,

nearly one in four respondents encounter a bribe request when dealing with police. What is more,

Harare itself fares poorly within the country: the 2017 ES report that, at 19%, the rate of formal firms

encountering at least one bribe request is the highest in the capital (World Bank, 2016).

A brief look at news articles from the first six months of 2017 provides anecdotal flavor to

these data.14 In the first half of the year, the economy was stagnant, facing widespread cash shortages,

as a result of government-introduced bond notes pegged to the US dollar the previous year.15 Reports

of corruption were pervasive in every-day life, from encounters with gas station attendants to

13 Informal-sector dummy coefficient values of -2.14 [.166] and -1.93 [.242], respectively. Robust standard errors in brackets. 14 Subsequently cited news articles are based on a Factiva search of (“Harare” OR “Zimbabwe”) AND (“BRIBE” OR “CORRUPTION”), October 2018. 15 Note that by the end of 2017 the economy again had entered all-out crisis, including the ouster of President Robert Mugabe on November 19.

(12)

skimmed charges from security guards.16 One particular locus of frequent, petty corruption

allegations was the city’s police force. At the time, the city’s informal vendors complained to the city

council about harassment and demands for bribes from municipal police officers.17 Several reports

of incidents involving physical clashes between vendors and police officers were reported, particularly

as the officers attempted to remove vendors from the city’s central business district.

Over a similar time period, the informal sector remained pervasive. Vendors and street sellers

were known to operate in various areas and times of the day. Though in some arrangements police

and officials allowed vendors to continue operating unimpeded,18 informal businesses remained

vulnerable. The treatment of these informal businesses resulted in protests under the Mugabe

government: their complaints were not limited to harassment but also extended to a need for

infrastructure, such as public-access toilets and water, used frequently by vendors.19

IV. A Simple Model: A Business and a Bureaucrat

An informal business

A simple model can help frame the two literatures on informality and corruption. While not

the main contribution of this paper, such a framework does have the benefit of combining two

literatures to give tractable predictions of which types of informal businesses face a bribe request.

Such a framework is largely missing in the literature. Those—such as Emerson (2006)—which do

combine these two do not focus on the exposure of the informal sector to bribes, or they do not

consider differential effects within the informal sector, e.g. Echazu and Bose (2008). The model here

assumes a bribe structure of rent-seeking interactions between informal businesses and public

officials: that is, the extraction of a bribe outside of official channels and under the threat of

repercussion for operating informally (“bureaucrat as harasser”).

Informal businesses (𝑖) are assumed to not pay wage taxes20 or fixed-cost administrative fees

paid by formal firms. All such taxes and fees are due each period. Informal businesses sell a

homogenous good and use a single input (labor), which for simplicity, does not require wages, either

because businesses are self-operators or employ unpaid workers (such as family). This assumption

16 “Harare Residents Find Ingenious New Ways to Survive”, All Africa, Feb. 22, 2017. 17 “Municipal Police Officers Fuel Graft”, Newsday, Feb. 1, 2016. 18 Trade on the Streets, and Off the Books, Keeps Zimbabwe Afloat, New York Times, March 4, 2017. 19 Protesters Fume as Zimbabwe Vice President Runs Up a Hotel Bill, New York Times, July 27, 2016. 20 For simplicity, wages are the only marginal tax levied.

(13)

appears reasonable: based on a representative estimate from the survey (described later), 81% of

Harare’s informal businesses are either self-operators or employed only unpaid labor in the last

month. All goods markets are competitive, and informal businesses are price-takers; revenues are

given by 𝑃𝑞 .

Informal businesses face a probability of detection by public officials 𝜀 . Assume that a

certain share 𝜃 of public officials are corrupt, and these will demand a bribe 𝐵 payment to allow

informal businesses to continue operating; for the sake of simplicity, honest public officials 1 𝜃

are indifferent to informal activity and let business continue without a fine or loss-inducing

interruption.21 Such bribes are encountered by unregistered businesses, and so this particular bribe

exposure is 0 for formal firms. An informal business’s profit is given by: 𝜋 𝑃𝑞 𝜀 𝜃 𝐵 𝑐,

where 𝑐 is a marginal cost and the same for both informal and formal sectors. Informal businesses

face an exogenous likelihood of exit κ, —like Melitz (2003)— which can be considered to encapsulate

a sector-specific discount rate. As such, the expected value of an informal business’s remaining in

operation is . The super-script 𝐼 indicates informal-sector businesses, differentiated from formal

firms, 𝐹. This follows Ulyssea (2018) and subsumes sector-specific discount rates into 𝜅 ,22 which

can also represent an exogenous risk term for remaining informal.

𝐶 represents formalizing costs; its counterpart 𝐶 is the cost of de-formalizing for formal

firms. Entry costs (𝐶 ) to the formal sector can include every-period fees to formalize and taxes.

Bribes demanded by gatekeepers (for instance, to receive an operating license or construction permit)

can also be regarded as entry costs. So, too, this will include associated costs, such as the time and

information needed to register (McKenzie and Sakho, 2010). These costs are, thus, foregone by

businesses that remain informal. De-formalizing costs 𝐶 , by contrast, represent an incurrence for

those continuing in the informal sector. These costs include such things as an inability to hire

additional workers for risk of being detected (Ulyssea, 2018) or unrealized payoffs to human capital

(Rausch, 1991). Varying both costs by individual characteristics (i.e. 𝐶 , 𝐶 ) introduces business-level

heterogeneity.

21 Of course, the honest official’s actions can be modeled to include a fine or shut-down action 𝐹, which would represent a cost of 𝜀 1 𝜃 𝐹 . 22 Ulyssea (2018) explicitly normalizes discount rates to 1 but notes that this term can be considered as a sector-specific discount term.

(14)

An informal business faces a two-sided exit decision. They will cease business altogether if

expected, discounted future profits fall below 0 (assuming no asset resale value), that is if 0.

Alternatively, they will formalize if the cost of operating informally exceeds the expected returns of

operating formally, net of entry costs. A business will remain informal therefore if: 𝐶

𝐶 . Re-arranging this equation in terms of present informal profits gives:

1 𝜋 θB 𝐶 𝐶 𝜅 𝐷

The right-hand side of this equation represents a net opportunity cost of remaining informal, denoted

𝐷 , and so the condition to remain informal is:

2 𝜋 θB 𝐷 ; 𝐷 𝜋𝜅

𝐶 𝐶 𝜅

Where 0 ;

0 ;

0 ;

0 ;

0. Put more plainly, the opportunity cost of

remaining informal increases with (foregone) formal-sector profits by remaining informal. 𝐷 also

increases with a rising risk factor of informality (𝜅 ) or with higher de-formalizing costs (𝐶 ). By

contrast, this opportunity cost decreases with a higher risk factor in the formal sector (𝜅 ) or with

greater associated entry costs to formalize (𝐶 ).

As noted above, informal businesses are frequently highly clustered, and this may partially be

due to localized factors. It is helpful, then, to consider informal business profits expressed relative to

discrete areas—roughly thought of as neighborhoods (Ellison and Glaeser, 1997). Informal business

profits (in area 𝑗) can be expressed as an average neighborhood-based term 𝜋 plus a business-

specific term 𝜖 , giving 𝜋 𝜋 𝜖 . The profit distribution 𝜋 is a random term, with location-

based endowments, such as infrastructure or foot traffic, pre-determined and exogenous to each

informal business. It is easy to see, then, that an informal business’s decision to remain informal

becomes 𝜋 𝜖 𝐷 . The informal business’s location decision, therefore, becomes a relative

expression of 𝑗 relative to all other locations 1 … 𝑗 1. This makes intuitive sense: an informal

business will remain in business and in the same location if the costs of remaining informal in the

same area are comparatively low to those elsewhere.

(15)

The bureaucrat

Corrupt public officials will demand a bribe, 𝐵, from a set of 𝑘 businesses, a sub-set of 𝑁,

the total number of informal businesses. The total take for a corrupt official is just 𝐵𝑘. Officials are

assumed to be segmented, and they thus deal with either the formal or informal sector, as the two

operate in a legally distinct way. An official encountering informal businesses can no longer demand

bribes from those formalized businesses, as their credible threat is punishment for operating without

registration. This is consistent with, for example, horizontal separation models (Echazu and Bose,

2008). Corrupt officials face a probability of detection by (honest) supervisors, and a complementary

probability of impunity, 𝑚 ∈ 0,1 . If corrupt activity is discovered, an official is fired. Informal

businesses lack legal redress; 𝑚 is therefore a decreasing function of 𝑘 (but not 𝐵). Put differently,

corrupt officials face a higher likelihood of detection and dismissal if they extract more bribes, but

not necessarily if they extract higher ones.23 An official’s time is finite, and dereliction of duties will

also result in eventual detection and dismissal: an official using all their time to solicit bribes will be

found out and removed. Therefore, the time to extract bribes is a function of the concentration of

informality, and 𝑚 increases with 𝑁.

There is a threshold level of impunity, 𝑚∗, below which a corrupt officer will be fired with

certainty. Note, then, that a corrupt official’s bribe take in area 𝑗 is bounded, such that 𝑓 𝑘 , 𝑁

𝑚∗ . Similarly, the amount of the bribe extracted, 𝐵 | 𝜀 , 𝜃 cannot exceed 𝑃𝑞 𝐷 .24 Bribes

exceeding this amount will cause informal businesses to exit the local market and, in turn, reduce the

pool of potential bribe payers. In this way, bribe extraction is mitigated by the option to formalize or

to cease business altogether; the model, thus complements others such as Choi and Thum (2005). A

corrupt official maximizes their bribe take, such that:

3 max 𝐵 𝑘 max 𝑓 𝑃𝑞 , 𝐷 , 𝜀 | 𝑚∗ ∗ 𝑘

An individual, informal business’s likelihood of facing a bribe is straightforwardly expressed as 𝑘 /𝑁

and a function of max 𝐵𝑘 .

23 This model is in many ways an extension of Emerson (2006), who models the corrupt official’s decision of extracting bribes from formalized firms. 24 For simplicity, since marginal costs are assumed the same across informal and formal sectors, this term is omitted.

(16)

Note that this yields a few, basic expectations. First, bribe extraction will increase with a

business’s revenues—that is, bribery tracks with a business’s ability to pay (e.g. Kaufmann and Wei,

1999). Secondly, the greater the cost to formalize, so too, the increase in bribes. That is, if a business

operator has opportunities outside informality, they will have little incentive to deal with corrupt

officials. The associated costs also vary by location, as opportunity costs are likely to be localized. So,

too, the likelihood of detection by a corrupt official will, of course, increase the chances of facing a

bribe, but this is also contingent on the tolerance for corruption and local circumstances.

V. Data: Collection and Methodology

One limitation to studying informal sector businesses is that reliable data are sparse. By most

operating definitions, informal businesses remain outside the bounds of many data sources. Informal

businesses do not appear on business registries nor other administrative data. Adequate sampling

frames for a robust survey methodology are, thus, nearly impossible to come by. Methods linked to

household and labor-force surveys (for instance, what is known as the 1-2-3 method) have offered

more systematic measurement of informal business activity (such as the data used by Lavallée and

Roubaud, 2018), however such data are based on initial sampling designed to be representative of

residential enumeration areas, and ultimately households. These methods are implemented over

multiple waves and are thus much more expensive and can lack a depth of information.

The analysis here uses a data set that was developed by the World Bank, which implemented

a survey in Harare, Zimbabwe over April–May 2017. The survey methodology uses a method called

“adaptive cluster sampling”, with stratification (SACS), which was developed in bio-statistics

(Thompson 1990, 1991) to efficiently measure elusive and clustered populations: the former

characteristic being mostly self-evident and the latter—that informality tends to be highly clustered—

being well-established in the literature (see, e.g., World Bank 2018, Harris, 2014, Gulyani and

Talukdar, 2010).

SACS is implemented through the selection of primary sampling units (PSUs) of square-block

areas. In the case of the Harare survey, the administrative area of the city was divided into 200-by-

200-meter squares, stratified by the classification of informal business activity (low, medium, high),

with a simple random sample of PSUs in a fourth stratum made up of known markets. The idea of

SACS is that the discovery of some target units (i.e., informal businesses) is useful and can make

fieldwork more efficient. Implementers set a threshold for expansion, where enumerators will move

(17)

to all adjacent squares,25 if that threshold is reached or exceeded. Enumerators are instructed to

approach all informal businesses in the area (regardless of structure, including households and

hawkers). Enumeration, in turn, develops, networks,26 which can be given survey weights and

considered as geographically representative (Appendix A.1). This method requires many informal

businesses to be enumerated, so many that a lengthy (termed here as “long-form”) survey would be

prohibitively costly and timely. As a result, a sub-set of businesses were selected randomly for a “long-

form” interview. In total, 3,687 businesses were enumerated in 129 networks27 and a total of 515

long-form surveys were completed (see Aga et al, 2017 for a full explanation of implementation).

VI. Econometric Specifications & Descriptive Statistics

Main Econometric Specification

The basic econometric specification uses a probit estimate of bribery, which follows from

Eq. 3:

4 𝑌∗ 𝛷 𝛽 𝛽 𝑅 𝛽 𝐷 ∑ 𝛽 𝑋′ 𝜇

Y* is a latent variable for the binary outcome of interest, Y. The main outcome measure is BRIBE,

which records if an informal business reports having to pay a bribe to remain informal.28 This measure

uses a ‘vague present’ time reference and is not linked to the “last completed month” as are several

other variables. Such a direct measure of bribery may be unreliable if respondents are reticent to

admit having to pay a bribe (e.g. Kraay and Murrell, 2016), as a result a “yes” value to BRIBE is likely

a lower bound.

The log of the last month’s sales (𝑅 ) is the main measure of having resources to pay. The

data include a monthly measure of revenues, as without formal accounting, businesses’ recall of the

last month’s sales is considered reliable and readily available in respondents’ minds. 𝐷 is a measure

of the opportunity cost of staying informal, though clearly difficult to measure. The survey instrument

25 Note that ACS with or without stratification can implement various types of expansion rules, including to squares on the cardinal (N,S,E,W) axes or diagonals. 26 In the literature on ACS, “network” is defined to mean all contiguous squares meeting or exceeding the threshold. Networks may be bordered by adjacent “edge units” where informal businesses may be encountered. While such accommodations were used in data collection, for the sake of not introducing further confusion, “network” is used here to include all squares enumerated from the selection of the same initial PSU. 27 The actual number of networks enumerated was higher as there were some non-unit networks. 28 The exact wording of questions is presented in Appendix A.II.

(18)

includes a wide array of information, and so various measures are possible. Two variables are used

separately. The first is a measure of whether an informal business was started because the owner was

unable to find another source of income, NECESSITY; this could be a direct opportunity cost of

informality, however it is self-reported and relies upon an assessment of a counter-factual (the owner

could not find another source of income). The NECESSITY measure is also likely downstream of

other facets of each individual’s situation, and this assessment is likely endogenous with other

unobserved characteristics (for example, reticence). A second proxy measure, NOWAGE, is a

dummy variable for whether any other household member of the operator has a job under contract.

An additional vector of controls is given by 𝑋′. These include business-level controls 𝐵′ .

There is little uniformity in the literature on firm-level bribery exposure, and so these most closely

follow Lavallée and Roubaud (2018). The controls include dummies of the education level

(completed or not) of the largest owner, including one dummy for ‘no or primary education’ and

another dummy for ‘greater than secondary education’. ‘Some or completed secondary education’ is

the reference category. These controls follow previous studies and are also called for by theoretical

models (e.g. Rausch, 1991) noting that greater human capital increases the likelihood of exiting the

informal sector. The (log) of the age of the owner is also included, as is a dummy (taking the value

of 1 if yes) if the operator is female, as gender may affect alternate opportunities and previous studies

have shown that females may be less likely to face a bribe request.29 The controls also include the

(log) number of employees (whether paid or unpaid) in the last month, as size may be an indicator

of ability to pay (as in Svensson (2003)) or affect the likelihood of detection (Fortin et al, 1997).

Other factors possibly affecting the likelihood of detection include the type of premises of

the informal business. It is unclear if such information has been available in previous studies; here,

these controls are included as they are presumed to affect visibility and are thus likely to be correlated

with the error term 𝜇 , justifying their inclusion. The estimations include dummies for the type of

location, specifically if it is a temporary (but fixed) structure, a household, or no fixed structure

(hawker). These are included in ∑ 𝑃′ , with the reference category for all regressions being a dummy

measure if a business is a hawker.

Still other factors affecting exposure merit inclusion: specifically, the threshold, 𝑚∗ cannot

be observed, but as the model suggests, is likely to vary by location. As a result, location is conditioned

29 See Lavallée and Roubaud (2018) and Rand and Tarp (2012).

(19)

on, including fixed effects by the administrative ward in Harare 𝐽′ . This is akin to claiming that

the tolerance for corruption can vary by neighborhood, ward, or precinct. As harassment by public

officials and police may be a primary mechanism for the extraction of bribes, HARASS is included

as a control in later specifications. However, as this certainly introduces simultaneity problems (if

harassment is a means of bribery extraction, businesses may be solicited for a bribe because they are

harassed or harassed to solicit a bribe), an additional measure is used. This measure averages the

harassment experiences of similar businesses—those located nearby—as a proxy for harassment

exposure, without including the direct experience of an individual business. Specifically, location data

are used to estimate the rate of harassment in the 500-meter radius around each location point,

exclusive of those points within 200 meters of each informal business. Other studies on firm-level

corruption have used “leave-out” averages, where an average of an inter-sectional stratum (e.g. size-

sector-location) is calculated omitting the firm-level input (see, e.g. Freund et al, 2016); the measure

of harass, called “harass donut”, uses GPS data to provide a location-based structure of these

measures. As such, it proxies an informal business’s exposure to harassment. Lastly, as proximity to

fixed premises may vary, controls for the measured location to eight municipal city offices are

included.

Descriptive Statistics

Table 2 presents the descriptive statistics of the main variables and controls; as described

above, the data are representative of the Harare administrative area. Nearly one in four informal

business operators reports having to pay bribes or give informal gifts to remain un-registered

(BRIBE=1). Similarly, roughly one-quarter of the informal sector reports being harassed by police or

government officials in the last month; the proxy, ‘donut’ measure is a lower rate of 10%. The average

informal business had between one and two employees in the last month and earned just under $318

in revenue from the business activity.30 Roughly 40% of informal businesses do not have fixed

premises, referred to as “hawkers”, this followed by just under 30% with fixed, but temporary

premises. Nineteen percent are household-based businesses, which on its face, may appear to be an

under-estimate, a concern if enumeration does not adequately reach in-home, informal activity.

30 The exponent of the mean shown in the table: exp(0.5) ≈1.6. The same value for sales last month (in USD) is exp(5.8)≈318. Due to the loss of precision by presenting the tables at one-digit values of logs, the levels in the text are calculated from the full, precise log estimates.

(20)

Nearly 60% of owners have at least some secondary education,31 while almost 30% has either a

vocational or university education beyond secondary. Four of ten of informal operators reached are

female.32

Table 2: Descriptive Statistics [95% Con. Interval] Mean Std. Dev. [Lower Upper] n(u)

BRIBE (Y=1) 23.5% 42.5% 13.5% 33.6% 506 Sales in last month (log) 5.8 1.4 5.4 6.2 500

Exp of log (USD) 317.7 4.1 213.5 472.5 NECESSITY (Y=1) 77.9% 41.5% 69.4% 86.4% 513 NOWAGE (Y=1) 64.4% 47.9% 53.9% 74.8% 515 Workers in a normal month (log) 0.45 0.55 0.30 0.60 515

Exp of log 1.6 1.7 1.4 1.8 Age of the owner (log) 3.59 0.03 3.53 3.65 515

Exp of log 36.3 1.0 34.2 38.5 Permanent, fixed premises (Y=1) 15.6% 36.3% 3.3% 27.8% 515 Temporary, fixed premises (Y=1) 28.7% 45.3% 18.2% 39.2% 515 Household (Y=1) 18.9% 39.2% 10.6% 27.2% 515 Hawker (Y=1) 36.8% 48.3% 25.5% 48.2% 515 No or only primary education (Y=1) 12.5% 33.1% 4.0% 21.1% 515 Secondary education (Y=1) 58.7% 49.3% 46.0% 71.5% 515 Beyond secondary education (Y=1) 28.7% 45.3% 16.2% 41.2% 515 Female (Y=1) 37.9% 48.6% 26.7% 49.2% 515 Harassed by police or officials (Y=1) 24.1% 42.8% 13.9% 34.3% 514 Harass 'donut' rate(d) 10.3% 19.2% 7.0% 13.6% 515

(d) Rate of harassment within 500m but excluding nearest 200m. (u) Observation counts are un-weighted.

       

The specific geo-location data available from the survey also allow for additional analysis

using GIS33software. Appendix A.IV shows heat maps that help visualize the spatial occurrence of

key variables. The first panel shows the occurrence of all enumerated informal businesses (whether

given the long-form survey or not): the map shows clear clustering, a construction of the data

collection process. The remaining three panels in red, purple, and orange show heat maps of the

occurrence of BRIBE and HARASS, respectively. One notes immediately that these are not an exact

31 ZimStat’s 2014 Labour Force Survey (LFCLS) reports that 68% of workers in the informal sector had their highest level of education in as some form of secondary school, completed or not (Form 1-6, with 46% terminal at Form 4) (ZimStat, p. 92) 32 This compares to a figure of 52% from the ZimStat LFCLS (p. 87). Discrepancies can likely be explained by composition effects (i.e. the LFCLS covers areas outside of Harare), but these values are within confidence ranges. 33 That is, Geographical Information Systems.

(21)

overlay of the concentration of informal businesses. In other words, certain clusters of informality

are also clusters of bribe exposure and harassment, while others are not.

VII. Results

Table 3 presents baseline, probit results for the two main opportunity-cost variables

(NECESSITY and NOWAGE). The odd-numbered columns exclude all controls except for stratum

and sector fixed effects as well as sales revenue; even-numbered columns include all co-variates on

location type, size, age of the respondent, a dummy for if the owner is a woman, harassment exposure

(‘donut’), and a location dummy for proximity to a municipal office. All specifications are survey-

weighted by both the PSU sampling block and the number of observations in each network,

clustering by stratum of selection to account for standard errors that are correlated by the type of

area. To understand the size of the shown coefficients, Table 4 includes the marginal effects of a

change from a “No” to “Yes” response for key binary variables and a 1 S.D. change in key continuous

variables. These effects are averages over the distribution and are in levels, and thus the level

percentage changes shown represent notably different relative changes in probabilities depending on

the underlying base probability of having to pay a bribe. That is, for the case of 𝐷 , this effect is given

by 𝛷 𝛽 𝛽 | 𝑋′ 𝛷 𝛽 | 𝑋′ , where 𝑋′ is the ceteris parabis vector of other controls.

(22)

Table 3: Probit Estimations Pr(BRIBE=1) (1) (2) (3) (4) Sales in last month (log) -0.066 0.319*** -0.000 0.236** (0.115) (0.117) (0.115) (0.106) Started due to lack of opportunity (NECESSITY) 1.107*** 0.858** (0.288) (0.334) No HH member w/contract (NOWAGE) 1.045*** 1.264*** (0.271) (0.262) Workers in a normal month (log) 0.139 0.246 (0.267) (0.256) Age (log, years) of owner -0.133 0.019 (0.482) (0.466) Permanent, fixed premises (Y=1) 1.581*** 1.471*** (0.471) (0.449) Household (Y=1) -0.270 -0.519 (0.414) (0.411) Temporary, fixed premises (Y=1) 0.795** 0.915*** (0.321) (0.315) Female (Y=1) -0.416* -0.288 (0.239) (0.227) Beyond secondary education (Y=1) -0.402 -0.046 (0.303) (0.290) No or only primary education (Y=1) -0.234 -0.136 (0.439) (0.415) Harassment 'Donut' Rate(d) 4.313*** 4.672*** (1.297) (0.865) Constant -1.674* -0.729 -1.835* -1.147 (1.011) (6.183) (0.948) (5.008) Observations 489 489 490 490 *** p<0.01, ** p<0.05, * p<0.1 (d) Rate of harassment within 500m but excluding nearest 200m. Survey-weighted (using Stata’s svy prefix). Standard errors clustered by stratum. All columns include stratum, sector, and ward (neighborhood) fixed effects as well as measured distances to eight municipal offices. Sectors are manufacturing, food provision (including bakeries), retail, and other services. Reference category for premises: no fixed premises (hawkers).

Table 4: Marginal Effects, Pr(BRIBE=1) NECESSITY NOWAGE (1) (2) (3) (4)

Opportunity Cost (D=1) 18.7%*** 10.1%*** 19.9%*** 16.5%*** S.E. (5.4%) (4.1%) (6.2%) (4.7%)

1 SD change (log) sales -2.0% 5.5%** 0.0% 3.8%** S.E. (3.4%) (2.2%) (3.4%) (1.8%)

1 SD change in harassment 10.1%*** 10.1%*** S.E. (3.2%) (2.2%)

*** p<0.01, ** p<0.05, * p<0.1 Note: the difference in predicted values from (mean -.5*SD) to (mean + .5*SD) of (log) monthly sales, an increase of about $483 in sales per month. The difference in harassment exposure is the nearby (donut) harassment rate is (mean -.5*SD) to (mean + .5*SD), or 1% to 20%. Marginal effects produced using margins and mlincom commands in Stata, using vce(unconditional)option as it is survey data.

In line with expectations, there are positive, significant signs for coefficients for NECESSITY

and NOWAGE; the marginal effect of the former is a 10% increased likelihood of being exposed to

(23)

a bribe, and 17% for the latter (from even-numbered columns). The coefficients on sales (i.e., ability

to pay) are positive, as expected, but only significant with full controls, and with comparably smaller

magnitudes. The marginal effects of an increase in one standard deviation of sales per month (a

magnitude of $483) are increases in the likelihood of facing a bribe of 6% and 4% in columns 2 and

4, respectively. The exposure to harassment (the ‘donut’ measure) is also notable; its coefficient is

consistently positive and significant, with marginal effects of 10% in Table 4. Additionally, those

having only a primary education are consistently less likely to face a bribe request.

Measures for dealing with endogeneity of main regressors

A natural endogeneity concern is that the measures of opportunity cost are themselves

functionally determined by other measures. This would be the case if, for instance, education level

affected the likelihood that an informal business was started out of lack of opportunity elsewhere. As

both measures of the opportunity cost 𝐷 are binary, they can be expressed as well with a latent

variable form, given by:

5 𝐷∗ 𝛷 𝛽 ∗ 𝛽 ∗𝑅 ∑ 𝛽 ∗𝑍′ 𝜀 ∗

Where the super-script D∗ indicates the functional form for the outcome of interest, here opportunity

cost. A bivariate probit (biprobit) model jointly estimates 4 and 5 and is generally appropriate in

the case of a binary outcome with a binary, endogenous regressor. It can be thought of as a

generalized case of the classic Heckman (1979) selection model, where outcome variables are

observed for all, not just selected cases. As such, four possible combinations of Y and D can be

observed.34 The error terms of 4 and 5 are correlated by a term denoted 𝜌, with endogeneity

indicated by 𝜌 𝑐𝑜𝑟𝑟 𝜇 , 𝜀 ∗ different than zero. The error terms are assumed to be bivariate

normal, with substantial bias resulting from violations of this assumption (Chiburis et al, 2011). The

biprobit, joint estimation is identified if 𝑍′ includes at least one excluded instrument that is considered

to affect D, but not Y; the vector 𝑍′ must also be longer than 𝑋′. The biprobit’s stricter assumption

of joint normality of the error terms allows for the calculation of an average treatment effect, given

by E 𝑌 | 𝐷 1, 𝑋′ E 𝑌 | 𝐷 0, 𝑋′ , where the effect of treatment is not differentiated within

sub-groups of the population.

34 That is, D=1, Y=1 ; D=1, Y=0 ; D=0, Y=1 ; D=0, Y=0.

(24)

A two-stage least squares regression (2SLS) additionally allows for an estimate of the local

average treatment effect (LATE), that is the effect of opportunity cost “treatment” on bribe

likelihood for compliers: that is, the effect of a higher opportunity cost for the group 𝐷 1.

However, 2SLS is generally regarded as inappropriate for binary outcomes with binary, endogenous

regressors. Termed the “forbidden regression”, this estimate will produce inconsistent and biased

estimates, as the linear projection of the regressor from the first stage will not align with the form of

the regressor itself (Wooldridge, 2002: 267–8). However, 2SLS has the advantage of being

straightforward, without requiring a further transformation. The procedure also yields substantial

information about the appropriateness of the instrument, which is unavailable for procedures for

endogenous, binary regressors (Nichols, 2011). Thus, it is often good enough to estimate the average

treatment effect by simply using 2SLS, conditional on controls, this is the local average treatment or

LATE (Angrist and Pischke, 2009). 2SLS estimates are presented, therefore, alongside probit and

biprobit estimates.

Choice of instrument

As noted in Section III, pre-2009 was a particularly tumultuous time for Zimbabwe’s

economy, including crippling hyperinflation. Though the country’s economy had been in freefall for

several years, rates of inflation peaked in the later half of 2008. In fact, though inflation had reached

a monthly rate of 50% in March 2007, more than a year later, in July 2008, this monthly rate was

2,600%.35 The abandonment of the Zimbabwean dollar by the first months 2009 stabilized prices,

and the economy showed some signs of recovery, with real GDP per capita even increasing by almost

5%. A cursory look at the movement of vulnerable employment before and after dollarization

suggests that employment opportunities outside the informal sector stabilized (at least relatively) after

2009. Under these conditions, then, it can be presumed that a selection effect occurred after

dollarization, when opportunities outside the informal sector were more available and viable. Those

taking up an activity with the purpose of earning income post-2009, then, would be assumed to do

so because they could not find formal-sector opportunities, even in a more stable environment. Put

35 Federal Reserve Bank of Dallas, 2011.

(25)

differently, those taking up new activities post-2009 may have found themselves particularly

segmented as the labor market stabilized.36

To understand this point, it is worth revisiting the expression of the opportunity cost of

remaining informal, 𝐷 𝐶 𝐶 𝜅 . Consider the risk factors for both operating in the

informal and formal sectors, 𝜅 and 𝜅 respectively, recalling that each subsumes a sector-specific

discount rate. A clear exogenous shock to such rates presents itself in the form of the post-

dollarization price stabilization, which would naturally reduce both the risk factors of operating

informally and formally. A palatable assumption is that the relative reduction in risk of operating

formally would be greater than that of the risk of operating informally. This assumption is intuitive:

exceptionally high-risk-rates during the hyperinflation crisis lead to a near-complete discounting of

all future returns, and this immediacy is consistent informality as a safety net. The relative stabilization

of the economy would have the effect of movement toward the formal sector after dollarization; this

is cursorily consistent with Figure 1. Note that this also implies that those taking up activities after

dollarization do so under less risky circumstances, thus implying that initiating these economic

activities is done even more so out of necessity. Thus, a first-stage expectation of taking up an activity

post-2009 is that these operators are more likely to be operating out of necessity.

A natural concern of such a measure as an instrument is that the pre- and post-2009 indicator

would also affect officials’ demands for bribes, violating the exclusion restriction. This would require

remarkable persistence in the effect on bribe demand, given that dollarization occurred roughly eight

years before the collection of the survey. Even if such an effect does persist, its directional effect is

likely in the opposite direction, meaning that the instrument is, if anything, underestimated. Take, for

instance, the public funding channel.37 An assumption in the literature is that public officials’ demands

for bribes will increase in the face of falling wages, which are reduced with shrunken public coffers.

Increasing general revenue, as was the case post-dollarization (Appendix A.VIII), would abate the

demand for bribes. In other words, if this effect is sufficiently persistent that it is still affecting bribe

36 Note that as of writing in 2018/19, Zimbabwe’s economy has entered into another period of economic crisis, which has continued. The point is, however, that the pitch of these crises was substantially less than those seen in the 2000-09 period. 37 I thank a colleague for this valuable suggestion.

(26)

demand eight years after hyperinflation, its result would be to reduce demand relative to pre-2009, just

as the effect is that those taking up sector activities are more likely to operate out of need.

The data include a measure of the years of experience of the largest owner in the specific

“industry of” activity—that is, for example, in making handicrafts or baking bread. This is a different

measure than from the age of the business, or for that matter, from the age of the owner. In fact, the

average informal business has been in operation for 4.9 years, while the average owner has been

active in that area of activity for 7 years.38 This is suggestive that informal business owners took up

their activity outside of a business context or possibly had a previous operation. A concern would be

if the measure of industry experience is simply reflecting age, savvy, or longevity in the informal

sector, all of which could plausibly be related to bribe exposure (through for example knowledge that

paying a bribe has certain known payoffs for continuing to operate or to use savvy to avoid corrupt

officials). In this case, it is important to control for age of the owner. For the instrument, a dummy

variable is created, taking a value of 1 if the owner started in that “sector” after 2009, 0 otherwise. If

this variable is sufficiently, strongly correlated with the measure of opportunity cost and sufficiently

exogenous to bribe exposure, it suggests a valid instrument; by contrast this measure would be

vulnerable to further endogeneity if there are other unobservables correlated with both the industry

experience as well as bribe exposure.

Results of Two-Stage Estimations

Table 5 shows the estimated effects (LATE for 2SLS and ATE for biprobit) from two-stage

estimations of the full models from Table 3, Columns 2 and 4; the single variate probit estimates are

re-presented here as well, for comparison. Each two-stage model uses only one additional, excluded

instrument in addition to all co-variates (which are in both stages). Appendix A.VI includes several

diagnostic statistics for both two-stage methods. From the 2SLS results, Olea and Pflueger’s (2013)

weak instrument test —which can handle the use of non i.i.d. standard errors, such as the clustered

standard errors required by the use of complex survey data—strongly rejects the null of a weak

instrument with one endogenous regressor.39 However, only the LM Kleibergen-Paap rank statistic

for NOWAGE suggests at any level of confidence (and even then weakly with a p-value=0.099) that

38 Survey-weighted; age of business 25th to 75th percentile is 1 to 6 years. Years in “industry” 25th to 75th range is 2 to 10 years. The two variables have a correlation of 76.7% (weighted), significant at 99% confidence. 39 Implemented using the user-written command weakivtest in Stata, using a tau, threshold value of 5% with 95% confidence.

(27)

the system is not under-identified. For birprobit, estimates for Murphy’s (2007) score test of

goodness-of-fit show strong evidence (all with p-values<0.01) of bivariate normality, indicating that

biprobit estimates are appropriate, as suggested by Chiburis et al (2011). The rho value for both

biprobit specifications are negative, suggesting that coefficients in a the singular probit are under-

estimated, although it is only significant in the NOWAGE specification; the latter point suggesting

that the two-stage NOWAGE estimates are likely the most appropriate.

Table 5: Effects on Pr(BRIBE=1)

   NECESSITY NOWAGE

Probit 10.1%*** 16.5%*** (S.E.) (4.10%) (4.70%)

LATE (2SLS) 51.4%*** 52.4%*** (S.E.) (13.3%) (13.0%)

ATE (Biprobit) 13.5%*** 26.1%*** (S.E.) (5.5%) (5.7%)

*** p<0.01, ** p<0.05, * p<0.1 Note: probit and biprobit estimates run using Stata’s svy: prefix, using stratum-clustered standard errors. 2SLS similarly use clustered errors on stratum. All estimates are survey-weighted. Probit estimates correspond to coefficients from Table 3, Cols. 2 and 4.

Turning to the results themselves, the effects of both NECESSITY and NOWAGE are

positive and significant for the biprobit estimations; though with the caveat that only the rho for

NOWAGE is significant, suggesting that an endogenous selection effect requires a two-stage

estimate. The ATE then bears some interpretation. This estimate shows that, all else equal, the bribery

rate faced by informal operators is 26% greater if no households had a member under contract

(NOWAGE=0 for all) compared to the counterfactual where all households included such a steadier

income source (NOWAGE=1 for all). The effect among those specifically without a contract income

in the household (that is the LATE from 2SLS) shows a magnitude of 52%. Such magnitudes are

substantial and have suggestive economic conclusions: that is the lack of an alternate, steady income

such as under-contract employment has a notably large and significant effect on the overall incidence

of corruption.

Appendix V shows the first- and second- stage results for these estimations, where the first

stage is a probit estimation of each opportunity cost measure (D). The biprobit estimates for

NOWAGE are most appropriate, as discussed above, though all estimates for all measures, including

2SLS and biprobit are given for completeness. The relative “uptake” probabilities implied by Eq. 5

(28)

merit some discussion. First, as a non-weak instrument, the dummy indicating that an owner was

active in an activity after 2009 is positive and strongly significant, indicating that all else equal, those

taking up activities after stabilization are much more likely to not have an under-contract income in

their household. Not surprisingly, those with beyond a secondary education are less likely to have

NOWAGE=1, as are those with more employees. Operators with a fixed, but temporary premises

(e.g. a market stall) are also less likely to be operating in a situation of few opportunities outside

informality.

As mentioned above, the biprobit specification is particularly susceptible to bias from mis-

specification, including from additional, omitted variables. The rich data from the survey allow for

additional co-variates to be included. Table 6 includes several additional controls (in both X’ and

therefore Z’), replicating results from Table 5, but including controls for dummies on whether there

are children (6 years old or younger) in the owner’s household, a dummy for whether the owner is

the primary income earner, a dummy for the marital status of the owner, a dummy if the business

uses public services (in the form of use of the electrical grid), and a dummy variable for whether the

business was enumerated in an isolated residential area (i.e. no industrial or retail area within 500m).40

With these additional controls, for NOWAGE, the 2SLS estimates are sufficiently identified with

non-weak instruments, and the biprobit estimates are bivariate normal with evidence that a two-stage

correction is needed; for NECESSITY, though 𝜌 is significant (and negative), suggesting the

appropriateness of the biprobit, the 2SLS fails to reject under-identification (though the instrument

is statistically not weak).

40 This was done using OpenSourceMaps, which include coded data on land usage, each informal business can be coded according to the near-area surrounding where they were interviewed. Using a 500m radius, an estimated 90% of informal businesses in Harare operate within 500m of a residential area. Only 27% of businesses are near a commercial or industrial area and a scant 16% are within half a kilometer of retail space, providing further support that the informal and formal sectors operate in segmented areas within Harare’s city limits.

(29)

Table 6: Effects using Additional Controls

   NECESSITY NOWAGE

Probit 7.8%*** 10.5%*** (S.E.) (0.9%) (3.6%)

LATE (2SLS) 51.6%*** 52.7%*** (S.E.) (12.1%) (13.5%)

ATE (Biprobit) 14.9%*** 19.4%*** (S.E.) (2.0%) (2.8%)

*** p<0.01, ** p<0.05, * p<0.1 Note: Controls include (log) sales, (log) employees, (log) age of owner, dummy for female=1, type of premises FE, education-level FE, stratum FE, sector FE, and harassment ‘donut’ rate. Results here add dummies for if married (Y=1), if there are children (6 y.o. or under) in the HH (Y=1), if the largest owner is the primary income owner in HH, dummy if the business is isolated residential (more than 500m from commercial or industrial area), and if the business uses the electrical grid (Y=1). HH: household, FE: fixed effect.

Results from the estimations with additional co-variates bolster previous findings. ATE

estimates for both NECESSITY and NOWAGE are positive and significant, with magnitudes of 15

and 19%, respectively. Significant effects from the second stage for those in isolated residential areas

(less likely to face a bribe request) and breadwinners as well as those with children in the household

(both positive, significant) suggest some justification for their inclusion. LATE estimates are

significant for both measures are positive and significant.

As noted above, the functional form of 2SLS may not be appropriate for binary outcomes

with binary endogenous regressors. Interpreting 2SLS effects as a local treatment effect requires that

the instrument (starting up in an area of activity post-2009) have a monotonic effect on NECESSITY

(NOWAGE). That is, that there are no cases where switching from a 0 to 1 value for the instrument

results in a decreased likelihood of NECESSITY(NOWAGE)=1. With a binary outcome measure, this

monotonicity assumption is not readily testable, and so the LATE magnitudes should be taken with

some caution.

However, two generally suggestive conclusions can be drawn. First, the LATE and ATE

effects are consistently positive and significant, establishing a relative robustness of their direction and

salience. Secondly, recalling that the LATE is estimated for so-called ‘compliers’—those operating out

of necessity or households de-linked from formalized employment—the 2SLS effects can be regarded

as the effect of operating under inelastic opportunities to formalize. That is, the effect on bribe

exposure among those most segmented from formalized opportunities. Such effects will be more

relevant if compliers are notably different than other informal sector businesses. By contrast, the

biprobit ATEs provide the difference in overall bribe rates if all operators started out of necessity (or

(30)

without a household link to formal employment) compared to the counterfactual scenario where all

operators are working out of free choice, all else equal based on observables. To the extent that

Harare’s informal sector is homogenous (given what can be observed), this effect is thus applicable—

it will be less generalizable in the absence of observing data on other characteristics, such as

entrepreneurship or, say, a preference to work informally.

VIII. Conclusions

Businesses in the informal sector may face differing risks of exposure to bribe requests than

formal firms; it is not unreasonable to argue that such exposure operates differently outside of the

formal economy. This paper puts forth a basic model where the informal sector’s risk of encountering

a bribe occurs in an interaction characterized as unstructured harassment, as opposed to a rent

extraction in the context of a formal firm soliciting an official service, such as a license or permit.

The expectations of this basic model are that an inelastic ability to formalize—in short, a lack of

opportunities outside informality—allows for more frequent extraction of bribe payments. Those

with a greater ability to pay (in terms of sales) are also expected to encounter demands from corrupt

officials. While evidence from a survey in Harare shows robust evidence for the former expectation,

there is limited evidence for the latter.

Further work will surely require more data. Such relationships may vary under different

circumstances, in different cities, and at different times. These data will hopefully be increasingly

available. While this study investigates bribe exposure among heterogenous informal businesses, it

does not have information on the size of those bribes; the model’s expectations are that businesses

more able to pay will have larger bribes demanded of them. Under constraints of the exposure of

corrupt officials, the expectation is that larger-sized bribes are initially preferable to more frequent,

pettier rents. Without data on the size of such bribes, these expectations are not testable. While not

the aim here, a similar model can be extended over repeated periods, where expectations of future

interactions can affect current behavior. For instance, extensions could model the incentives of

officials wishing to keep informal businesses informal (in order to extract bribes continuously) or a

model of competing and corrupt officials. Such extensions will require additional, micro-level

information under varying circumstances.

(31)

References

Aga, G., D. Francis, and M. Wild. (2017). Surveying Informal Enterprises: Applying Stratified Adaptive Cluster Sampling using CAPI with Implementation and Monitoring Tools. Presented to the United Nations Economic Commission for Europe (UNECE), October 12, 2017. Available here: https://www.unece.org/index.php?id=45491

Amin, M. (2009). “Labor productivity in the informal sector: Necessity vs. Opportunity firms.” World Bank Policy Working Paper 53031.

Angrist, Joshua D., and J-S Pischke. (2009). Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton: Princeton University Press.

Besley, T. and J. McLaren (1998). “Taxes and Bribery: The Role of Wage Incentives”. The Economic Journal. 103(416): 119-141. DOI: 10.2307/2234340

Besley, T. and T. Persson. (2014). “Why Do Developing Countries Tax So Little?” The Journal of Economic Perspectives. 28(3): 99–120. DOI: 10.1257/jep.28.4.99

Buehn, A. and F. Schneider. (2009). “Corruption and the Shadow Economy: A Structural Equation Model Approach.” IZA Discussion Paper No. 4182. https://ssrn.com/abstract=1409286

Choi, J. and M. Thum. (2005). “Corruption and the Shadow Economy.” International Economic Review, 46(3): 817–836. https://doi.org/10.1111/j.1468-2354.2005.00347.x

Cunningham, W. and W. Maloney. (2001). “Heterogeneity in the Mexican micro-enterprise sector: An application of factor and cluster analysis.” Economic Development and Cultural Change, 5(2001): 131–156. https://doi.org/10.1086/340012

De Soto, H. (1989). The Other Path: The Invisible Revolution in the Third World. New York: Harper and Row.

Dutta, N., S. Kar, and S. Roy. (2011). “Informal Sector and Corruption: An Empirical Investigation for India.” IZA Discussion Paper No. 5579.

Ellison, G. and E. Glaeser. (1997). “Geographic Concentration in U.S. Manufacturing Industries: A Dartboard Approach.” Journal of Political Economy, 105(5): 889–927. DOI: https://doi.org/10.1086/262098

Emerson, P.M. (2006). “Corruption, competition and democracy.” Journal of Development Economics, 81(1): 193–212. https://doi.org/10.1016/j.jdeveco.2005.04.005

Echazu, L and P. Bose. (2008). “Corruption, Centralization, and the Shadow Economy.” Southern Economic Journal, 75(2): 524–537.

Falco, P. and L. Haywood. (2016). “Entrepreneurship versus joblessness: Explaining the rise in self-employment.” Journal of Development Economics, 118: 245–265. https://doi.org/10.1016/j.jdeveco.2015.07.010

Federal Reserve Bank of Dallas (2011). Globalization and Monetary Policy Institute 2011 Annual Report.

(32)

Fortin, B., N. Marceau, and L. Savard (1997). Taxation, wage controls and the informal sector. Journal of Public Economics 66 (2), 293–312.

Freund, C., M. Hallward-Driemeier, and B. Rijkers. (2016). "Deals and Delays: Firm-level Evidence on Corruption and Policy Implementation Times," World Bank Economic Review, 30(2): 354-382. https://doi.org/10.1093/wber/lhv001

Gulyani, S. and D. Talukdar. (2010). “Inside Informality: The Links Between Poverty, Microenterprises, and Living Conditions in Nairobi’s Slums”. World Development, 38(12): 1710-1726.

Günther, I. and A. Launov. (2012). “Informal employment in developing countries: Opportunity or last resport?” Journal of Development Economics, 97(1): 88–98. https://doi.org/10.1016/j.jdeveco.2011.01.001

Harris, J. (2014). “The Messy Reality of Agglomeration Economies in Urban Informality: Evidence from Nairobi’s Handicraft Industry”. World Development, 61: 102–113.

Kaufmann, D. and S. Wei. (1999). “Does ‘Grease Money’ Speed up the Wheels of Commerce?” NBER Working Paper 7093.

Kraay, A. and P. Murrell. (2016). “Misunderestimating Corruption.” The Review of Economics and Statistics, 98(3): 455–466. https://doi.org/10.1162/REST_a_00536

La Porta, R. and A. Shleifer. (2014). “Informality and Development”. The Journal of Economic Perspectives, 28(3): 109-126. DOI: 10.1257/jep.28.3.109

Lavallée, E. and F. Roubaud. (2018). “Corruption in the Informal Sector: Evidence from West Africa”. The Journal of Development Studies. DOI: 10.1080/00220388.2018.1438597

Loayza, N. and J. Rigolini. (2011). “Informal Employment: Safety Net or Growth Engine?” World Development, 39(9): 1503-1515. DOI: https://doi.org/10.1016/j.worlddev.2011.02.003

Maloney, W. (2004). “Informality revisited.” World Development, 32(7),1159–1178. DOI: doi:10.1016/j.worlddev.2004.01.008

McKenzie, D., & Sakho, Y. S. (2010). “Does it pay firms to register for taxes? The impact of formality on firm profitability.” Journal of Development Economics, 91(1), 15-24. doi:10.1596/1813-9450-4449

Marjit, S., V. Mukherjee, and A. Mukherjee. (2000). “Harassment, Corruption, and Tax Policy.” European Journal of Political Economy, 16(1): 75-94. DOI: https://doi.org/10.1016/S0176-2680(99)00048-8.

Nguimkeu, P. (2013). “A structural econometric analysis of the informal sector heterogeneity.” Journal of Development Economics, 107: 175–191.

Olken, B. and R. Pande. (2012). “Corruption in Developing Countries.” Annual Review of Economics, 4: 479-509. DOI: https://doi.org/10.1146/annurev-economics-080511-110917.

(33)

Rand, J. and F. Tarp. (2012). “Firm-level Corruption in Vietnam.” Economic Development and Cultural Change, 60(3): 571-595. DOI: 10.1086/664022.

Rausch, J. (1991). “Modelling the Informal Sector Formally.” Journal of Development Economics, 35(1): 33-47. DOI: https://doi.org/10.1016/0304-3878(91)90065-4.

Roodman, D. (2011), “Fitting Fully Observed Recursive Mixed-Process Models with cmp,” Stata Journal, 11(2): 159–206.

Safavian, M., D. Graham, and C. Gonzalez-Vega. (2001). “Corruption and Microenterprises in Russia.” World Development, 29(7): 1215–1224. DOI: https://doi.org/10.1016/S0305-750X(01)00036-5.

Shleifer, A., and R. Vishny. (1993). “Corruption”. Quarterly Journal of Economics, 108:599-617.

Svensson, J. (2003). “Who Must Pay Bribes and How Much? Evidence from a Cross Section of Firms.” The Quarterly Journal of Economics, 118(1): 207–230. DOI: https://doi.org/10.1162/00335530360535180.

Thompson, S. K. (1990). “Adaptive cluster sampling.” Journal of the American Statistical Association, 85(412), 1050-1059.

---. (1991). “Stratified adaptive cluster sampling.” Biometrika, 78(2), 389–397.

Transparency International. Accessed October 17, 2018.

---. (2015). “People and Corruption: Africa Survey 2015.”

Ulyssea, G. (2018). “Firms, Informality, and Development: Theory and Evidence from Brazil.” American Economic Review, 108(8): 2015-2047. DOI: 10.1257/aer.20141745

Waller, C., T. Verdier, and R. Gardner. (2007). “Corruption: Top Down or Bottom Up?” Economic Inquiry, 40(4): 688–703. https://doi.org/10.1093/ei/40.4.688

Wooldridge, J. (2002). Econometric Analysis of Cross Section and Panel Data. MIT Press: Cambridge, MA.

---. (2015). “Control Function Methods in Applied Econometrics.” Journal of Human Resource, 50(2): 420–445.

World Bank. (2016). Enterprise Surveys. Via: www.enterprisesurveys.org.

---. (2018). The Role of City Governments in Economic Development of Greater Kampala. World Bank: Washington, DC.

Zimbabwe National Statistics Agency (Zimstat) (2014). Labour Force Survey (LFCLS).

(34)

Appendix

A.I. Survey Weighting Using Adaptive Cluster Sampling41

The adaptive cluster sampling approach is enabled by the creation of a sampling frame by

partitioning the urban area of Harare into primary sampling units (PSUs) of 200m by 200m squares.

As this approach includes areas, which with local knowledge, are known to have either a lower or

higher likelihood of finding informal-sector activity, a stratified approach was used. The sampling

frame (taken from Aga et al, 2017) is featured in Figure A1.

Eponymously, in ACS, PSUs are enumerated adaptively, meaning that when a certain

threshold of units (in this case informal businesses) are found, all adjacent squares are subsequently

enumerated. Due to this, sampling weights cannot be known at the beginning of fieldwork. In ACS,

and as noted throughout above, reference is made to networks. These can be of different sizes,

denoted by 𝑚 . The simplest network is the one with only a single node, namely the selected starting

square. Probabilities for these networks are calculated as in a stratified simple random sample setting:

41 This appendix borrows heavily from Aga et al (2017), with some parts taken verbatim.

(35)

𝜋 ,𝑛𝑁

with h indicating the corresponding stratum, n and N the corresponding sample and population size.

However, things become more complicated with networks of size 𝑚 1, which is the result if the

initial square exceeds the threshold. In this case 𝜋 , needs to account for the different selection

probabilities at the different stages:

𝜋 , 1

The final probability 𝜋 , 𝜋 , 𝜋 , can then be applied in the familiar way to calculate population

means, totals and standard errors. An additional adjustment needs to be made if a network crosses

the stratum boundaries, and the starting squares are different, as well as when networks overlap.

A.II. Key Questions from Long-Form Questionnaire

Question Wording BRIBE R5 Does this business or activity have to give gifts, informal

payments or bribes to remain unregistered? HARASS I6 In [insert last completed month], did this business or activity

experience harassment by government officials or police? NECESSITY B5d Please indicate if any of the following were reasons why the

largest owner started this business or activity: Unable to find another source of income

NOWAGE

=1 if B20=0

B20

How many people within the largest owner’s household premises have employment under a contract, including the largest owner?

(36)

A.III. Location of Municipal Offices

   Latitude Longitude Kambuzuma -17.86 30.97 Waterfalls -17.89 31.01 Greendale -17.81 31.13 Rowan Martin -17.83 31.04 Mt Pleasant -17.77 31.05 Borrowdale -17.75 31.12 Stodart Hall -17.86 31.03 Tafara -17.82 31.21 Note: Municipal office locations obtained via Google maps, search of Harare “municipal”, “city office”. Kambuzumua also includes Rugare Community Hall. Rowan Martin includes the City Planning and Development Office and Harare Town House. Mt. Pleasant includes Mt. Pleasant City Hall and Arundel Kewada Hall. Tafara includes Mabvuku District Council and Community Hall Old Tafara.

(37)

A.IV. Heat Maps of Outcome Variables (un-weighted)

A. All interviews B. BRIBE

B. HARASS

NOTE: heat maps rendered in QGIS, using an OpenStretMaps base map.

(38)

A.Va.: Biprobit Estimations (two-stage results)

D = NECESSITY NOWAGE Stage 2nd 1st 2nd 1st Pr( ) BRIBE D BRIBE D

Sales in last month (log) 0.296** 0.102 0.164 0.245** (0.117) (0.094) (0.107) (0.112)

D (indicated in col. Header) 1.150** 1.893*** (0.504) (0.377)

Workers in a normal month (log) 0.190 -0.011 0.384 -1.028*** (0.269) (0.248) (0.264) (0.312)

Age (log, years) of owner -0.091 -0.182 0.048 0.747 (0.484) (0.487) (0.455) (0.524)

Permanent, fixed premises (Y=1) 1.560*** -0.660 1.256*** 0.597 (0.468) (0.453) (0.439) (0.474)

Household (Y=1) -0.245 -0.626** -0.425 -0.479 (0.407) (0.308) (0.399) (0.338)

Temporary, fixed premises (Y=1) 0.765** -0.225 0.935*** -0.690** (0.320) (0.287) (0.302) (0.294)

Female (Y=1) -0.453* 0.723*** -0.224 -0.475** (0.247) (0.266) (0.236) (0.219)

Beyond secondary education (Y=1) -0.324 -0.937*** 0.170 -1.139*** (0.324) (0.255) (0.294) (0.305)

No or only primary education (Y=1) -0.215 -0.498 -0.127 0.013 (0.442) (0.420) (0.423) (0.394)

Harassment 'Donut' Rate(d) 4.356*** -0.848 4.584*** -1.122 (1.372) (0.960) (0.861) (0.918)

IV: started in 'industry' post 2009 (Y=1) 1.133*** 1.188*** (0.277) (0.307)

Constant 0.610 11.788* -1.539 2.564 (5.966) (6.109) (4.629) (3.889)

Ath rho -0.134 -0.389** (0.171) (0.187)

*** p<0.01, ** p<0.05, * p<0.1 (d) Rate of harassment within 500m but excluding nearest 200m. Survey-weighted (using Stata’s svy prefix). Standard errors clustered by stratum. All columns include stratum, sector, and ward (neighborhood) fixed effects as well as measured distances to eight municipal offices. Sectors are manufacturing, food provision (including bakeries), retail, and other services. Reference category for premises: no fixed premises (hawkers). All specifications have n=483 observations.

(39)

A.Vb.: 2SLS Estimations (two-stage results)

D = NECESSITY NOWAGE Stage 2nd 1st 2nd 1st Pr( ) BRIBE D BRIBE D

Sales in last month (log) 0.027*** 0.0184 0.004 0.064* (0.008) (0.0185) (0.023) (0.038)

D (indicated in col. Header) 0.514*** 0.524*** (0.133) (0.130)

Workers in a normal month (log) 0.065 -0.0807 0.141*** -0.221** (0.051) (0.0725) (0.049) (0.110)

Age (log, years) of owner 0.038*** -0.0277 -0.077 0.209 (0.010) (0.0405) (0.070) (0.142)

Permanent, fixed premises (Y=1) 0.398*** 0.0365 0.218*** 0.368*** (0.112) (0.137) (0.081) (0.140)

Household (Y=1) -0.015 0.0445 0.059 -0.102 (0.041) (0.0868) (0.109) (0.189)

Temporary, fixed premises (Y=1) 0.080 0.132 0.208*** -0.102 (0.110) (0.160) (0.069) (0.099)

Female (Y=1) -0.167** 0.255** 0.012 -0.105*** (0.070) (0.0996) (0.026) (0.039)

Beyond secondary education (Y=1) -0.119*** -0.121*** 0.004 -0.345*** (0.017) (0.0261) (0.075) (0.041)

No or only primary education (Y=1) 0.067** 0.00621 0.074 0.000 (0.027) (0.191) (0.125) (0.107)

Harassment 'Donut' Rate(d) 0.882*** 0.0479 0.979*** -0.133 (0.239) (0.303) (0.263) (0.182)

IV: started in 'industry' post 2009 (Y=1) 0.284* 0.292*** (0.155) (0.050)

Constant 0.464 1.403 0.697*** 0.751 (0.649) (1.160) (0.053) (1.066)

R-squared 0.380 0.476 *** p<0.01, ** p<0.05, * p<0.1 (d) Rate of harassment within 500m but excluding nearest 200m. Survey-weighted (using Stata’s svy prefix). Standard errors clustered by stratum. All columns include stratum, sector, and ward (neighborhood) fixed effects as well as measured distances to eight municipal offices. Sectors are manufacturing, food provision (including bakeries), retail, and other services. Reference category for premises: no fixed premises (hawkers). All specifications have n=483 observations.

(40)

A.VIa.: Biprobit Estimations, Supplemental Controls (two-stage results)

D = NECESSITY NOWAGE Stage 2nd 1st 2nd 1st Pr( ) BRIBE D BRIBE D

Sales in last month (log) 0.260** 0.042 0.197*** 0.148 (0.115) (0.071) (0.055) (0.145)

D (indicated in col. Header) 1.373*** 1.531*** (0.380) (0.390)

Workers in a normal month (log) 0.143 -0.120 0.336 -1.315*** (0.437) (0.254) (0.408) (0.485)

Age (log, years) of owner 0.143 -0.086 0.359 0.391 (0.345) (0.200) (0.376) (0.654)

Permanent, fixed premises (Y=1) 1.586** -0.896*** 1.212* 0.833** (0.673) (0.281) (0.687) (0.324)

Household (Y=1) 0.108 -0.811*** -0.162 -0.599 (0.449) (0.278) (0.515) (0.455)

Temporary, fixed premises (Y=1) 0.817** -0.205 0.912** -0.711* (0.351) (0.559) (0.371) (0.405)

Female (Y=1) -0.349*** 0.944** -0.224 0.247 (0.129) (0.367) (0.242) (0.235)

Beyond secondary education (Y=1) -0.047 -1.047*** 0.231* -1.442*** (0.049) (0.151) (0.131) (0.267)

No or only primary education (Y=1) -0.265 -0.528 -0.253 -0.168 (0.353) (0.326) (0.176) (0.620)

Harassment 'Donut' Rate(d) 4.698*** -1.032 4.808*** -0.527 (0.876) (1.325) (1.081) (1.000)

Married (Y=1) -1.421*** 0.358 -1.012** -0.959** (0.476) (0.431) (0.461) (0.488)

Isolated residential area (Y=1) 0.042 0.254** 0.356** -0.933*** (0.087) (0.120) (0.139) (0.353)

Owner is HH breadwinner (Y=1) 0.660*** 1.005** 0.310*** 2.511*** (0.233) (0.449) (0.099) (0.682)

Children in HH (Y=1) -1.050* 0.733*** -0.634 -0.643 (0.627) (0.182) (0.515) (0.574)

Uses electrical grid (Y=1) -0.198 -0.338 -0.203 -0.228 (0.369) (0.280) (0.390) (0.340)

IV: started in 'industry' post 2009 (Y=1) 1.241** 1.824*** (0.497) (0.477)

Constant -2.355 7.443*** -2.429 -1.210 (6.637) (2.782) (6.492) (3.904)

Observations 477 477 478 478 Ath rho -0.387** -0.402*

(0.166) (0.232) *** p<0.01, ** p<0.05, * p<0.1 (d) Rate of harassment within 500m but excluding nearest 200m. Survey-weighted (using Stata’s svy prefix). Standard errors clustered by stratum. All columns include stratum, sector, and ward (neighborhood) fixed effects. Sectors are manufacturing, food provision (including bakeries), retail, and other services. Reference category for premises: no fixed premises (hawkers).

(41)

A.VIa.: 2SLS Estimations, Supplemental Controls (two-stage results)

D = NECESSITY NOWAGE Stage 2nd 1st 2nd 1st Pr( ) BRIBE D BRIBE D

Sales in last month (log) 0.014** 0.009 -0.000 0.039 (0.007) (0.018) (0.022) (0.039)

D (indicated in col. Header) 0.516*** 0.527*** (0.121) (0.135)

Workers in a normal month (log) 0.050 -0.070 0.123** -0.200** (0.068) (0.066) (0.049) (0.089)

Age (log, years) of owner 0.078* -0.038 0.067* 0.003 (0.043) (0.032) (0.035) (0.098)

Permanent, fixed premises (Y=1) 0.331*** 0.057 0.181** 0.325*** (0.088) (0.175) (0.078) (0.098)

Household (Y=1) 0.018 -0.004 0.079 -0.129 (0.060) (0.114) (0.118) (0.162)

Temporary, fixed premises (Y=1) 0.061 0.131 0.189*** -0.101** (0.114) (0.186) (0.069) (0.046)

Female (Y=1) -0.147** 0.292*** -0.025 0.040 (0.066) (0.091) (0.043) (0.031)

Beyond secondary education (Y=1) 0.030 -0.181*** 0.088 -0.274*** (0.049) (0.049) (0.089) (0.037)

No or only primary education (Y=1) 0.044 -0.014 0.037 0.005 (0.039) (0.153) (0.128) (0.079)

Harassment 'Donut' Rate(d) 0.870*** 0.060 0.966*** -0.128* (0.192) (0.340) (0.252) (0.076)

Married (Y=1) -0.294*** 0.100** -0.135 -0.185*** (0.058) (0.040) (0.087) (0.048)

Isolated residential area (Y=1) 0.051 -0.021 0.143* -0.184*** (0.046) (0.040) (0.078) (0.055)

Owner is HH breadwinner (Y=1) 0.038 0.194** -0.087 0.432*** (0.034) (0.078) (0.098) (0.138)

Children in HH (Y=1) -0.347*** 0.184*** -0.162* -0.188 (0.097) (0.055) (0.094) (0.143)

Uses electrical grid (Y=1) -0.045 -0.060*** -0.111** 0.064 (0.060) (0.007) (0.054) (0.061)

IV: started in 'industry' post 2009 (Y=1) 0.288* 0.286*** (0.149) (0.046)

Constant 0.199 1.199 0.448** 0.589 (0.585) (0.997) (0.189) (0.758)

Observations 477 477 478 478 R-squared 0.479 0.517

*** p<0.01, ** p<0.05, * p<0.1 (d) Rate of harassment within 500m but excluding nearest 200m. Using sampling weights with standard errors clustered by stratum. All columns include stratum, sector, and ward (neighborhood) fixed effects. Sectors are manufacturing, food provision (including bakeries), retail, and other services. Reference category for premises: no fixed premises (hawkers).

(42)

A.VII: Diagnostics of 2SLS and Biprobit

Base controls (Tables 7 & A.V)

Biprobit NECESSITY NOWAGE Rho -0.13 -0.37

Ath-rho -0.13 -0.39 p-value 0.432 0.038

2SLS KP rk LM Stat 1.7 2.7 Chi-2, p-value 0.187 0.099

Effective F-stat 115.3 1,165.8 tau=5% (95% confidence) 37.4 37.4

Additional controls (Tables 7 & A.VI)

Biprobit NECESSITY NOWAGE

Rho -0.37 -0.38 Ath-rho -0.39 -0.40 p-value 0.020 0.083

2SLS KP rk LM Stat 1.8 2.7 Chi-2, p-value 0.185 0.0991

Effective F-stat 129.00 1,356.91 tau=5% (95% confidence) 37.4 37.4

Note: KP rk LM stat as reported by ivreg2, specifying standard errors clustered on the stratum level. Effective F-stat is obtained by running weakivtest from Pflueger and Wang (2015), implementing Olea and Pflueger’s (2013) weak instrument test for the case of clustered standard errors, as used here. A critical value threshold of tau=5% is shown, at a 95% confidence level.

(43)

A.VIII.: General Government Finances, Zimbabwe, 2005–2017

Source: IMF, WEO Accessed: March 21, 2019 2005 earliest available data

Dollarization

0

10

20

30

40

Per

cent

age

of G

DP

2005 2009 2013 2017

Year

Revenue

Total Expenditure

Note: 2016 and 2017, IMF estimates

General Government Finances