microenterprises and the lure of wage work: theory and ... · microenterprises and the lure of wage...

85
* *

Upload: others

Post on 19-Mar-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

Microenterprises and the Lure of Wage Work: Theory

and Evidence from Mexican Export Manufacturing

Michael Koelle∗

18 August 2017

Abstract

This paper unites two perspectives on entrepreneurship in developing countries: the

entrepreneur as a �rm, and the entrepreneur as a self-employed worker. I analyse

theoretically and empirically the consequences of non-separability of occupational and

entrepreneurial choices. In my model of investment decisions and occupational choices

in the presence of search frictions, entrepreneurs face a trade-o�. They could either run

a �rm which ful�ls their entrepreneurial potential and that few job o�ers could com-

pete with, or run a small �rm while waiting to take the next attractive job opportunity.

Better opportunities in wage employment increase incentives to keep �rms small. For

an empirical test of the model's predictions, I exploit trade-induced changes in em-

ployment opportunities in large export manufacturing establishments across space and

time. I �nd that entrepreneurs reduce their �rm size in anticipation of local expansions

of export manufacturing industries. An event study of announcements of automotive

plant expansions, amongst other supplementary evidence, suggests that it is indeed

the expectation of job opportunities, and not other mechanisms, which drive the em-

pirical e�ects. The theoretical and empirical results support the notion that frequent

transitions out of entrepreneurship are a determinant of �rm size.

Keywords: Occupational choice, search frictions, entrepreneurship

JEL Codes: J24, J64, L26, O17.

∗Address: Department of Economics, University of Oxford, Manor Road Building, Oxford OX1 3UQ, e-mail: [email protected]. Thanks to Simon Quinn, and Margaret Stevens, for encouragementand advice; Orazio Attanasio, Ian Crawford, Elwyn Davies, Marcel Fafchamps, Paolo Falco, James Fenske,Markus Goldstein, Doug Gollin, Juan Carlos Hallak, Leander Heldring, Lukas Hensel, Clément Imbert,Michael Keane, Kalina Manova, Peter Neary, Chris Roth, Marlon Seror, Raúl Sánchez de la Sierra, DanielaScur, Chris Woodru�, the members of the Firms and Development Student Research Group at Oxford,and audiences at various conferences and seminars for helpful discussion and comments. I thank PavelLuengas-Sierra for help with some of the data.

1

Page 2: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

1 INTRODUCTION

1 Introduction

Small �rms, owned and operated by an entrepreneur, unite several economic elements. Theyare �rms, who generate pro�ts by turning inputs into output, and as employers provide for animportant share of jobs. Small �rms are also a vehicle of self-employment for entrepreneurs.This is particularly true in developing countries, where a large share of the urban workforceare self-employed entrepreneurs with small businesses, as Figure 1 documents. Depite theirindividual sizes, as a group, small-scale entrepreneurs create a signi�cant number of wagejobs. In Mexico, for instance, 28 of percent of the workforce are wage-employed in �rms often workers or fewer. Given this central role of entrepreneurs in labour markets of developingcountries, researchers and policymakers have experimented with many di�erent policies toalleviate potential constraints on �rms and entrepreneurs.1 Yet, an important puzzle prevails:microenterprises on average tend to stay small, in spite of apparently very high returns tocapital.2 While previous work has paid careful attention either to the entrepreneur and their�rm or to the self-employed worker, it has largely ignored the dual role of the entrepreneuras a �rm and as a self-employed worker.3

In this paper, I unite these two perspectives on entrepreneurship. I address the questionof whether entrepreneurs' decisions about their �rm, and their choices about their ownemployment, are separable or not. I propose a particular mechanism through which suchnon-separability matters: the anticipation of a better outside opportunity for entrepreneursin the future creates incentives for keeping �rms small. Many entrepreneurs in developingcountries lack attachment to their �rm, and most of those who leave their �rm do so totake up a wage job. In Mexico, about 20 percent of self-employed men are found in wageemployment a year later, and a similar pattern holds in other developing countries.4

My paper analyses this mechanism both theoretically and empirically. In my model, forward-looking entrepreneurs discount returns to investing into �rm-speci�c capital stock, with thediscount factor given by the probability that they eventually taken up a wage o�er. Theyunder-invest into their business relative to a world where entrepreneurial choices are madeindependently and only depend on the returns to capital. The comparative statics of thetheoretical model represent predictions that are testable with a regression model in �rstdi�erences. I implement this test for small �rms in Mexico. I exploit exogenous determinantsof job creation in large export manufacturing establishments across local labour markets

1See Blattman, Fiala, and Martinez (2014), Fafchamps and Quinn (2016), and McKenzie (2015) on capitalgrants experiments to foster high-growth entrepreneurship; Karlan and Valdivia (2011); Karlan, Knight,and Udry (2015) and for a review McKenzie and Woodru� (2014) on business training grants. Relatedly,Banerjee, Karlan, and Zinman (2015) summarise the �ndings from six �eld experiments on the introductionof commercial micro�nance.

2Experimental evidence suggests monthly returns to capital may be as high as 20 percent in Mexico(McKenzie and Woodru� (2008)), 15 percent in Ghana (Fafchamps, McKenzie, Quinn, and Woodru� (2014))or 5 percent in Sri Lanka (De Mel, McKenzie, and Woodru� (2008)).

3There is a literature that takes self-employment as an occupation, going back to Evans and Leighton(1989), Magnac (1991), Blanch�ower and Oswald (1998), and more recently Haywood and Falco (2016),Lopez-Garcia (2013). This literature has largely been silent about the endogenous determination of businesspro�ts. An exception are Koelle and Quinn (2016) who look at the life-cycle determinants of self-employmententry, exit and earnings.

4See table in Appendix C for details with data from Colombia, Mexico and Ghana.

2

Page 3: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

1 INTRODUCTION

to identify the e�ects of wage employment opportunities on entrepreneurs. I documentthat the model matches this setting very closely, and that my design isolates market-levelshocks to job postings. I �nd evidence that is consistent with the mechanism where morewage employment opportunities for entrepreneurs reduce the size of small �rms. I providesupplementary evidence to suggest that it is indeed the expectation of job opportunitites,and not other mechanisms, which drive the results.

In the theoretical part, I formulate a two-period model that captures how discrete occu-pational choice can a�ect investment decisions of entrepreneurs. I nest a standard modelof optimal investment under uncertainty (with convex preferences and technology) within atwo-sector labour market search model (building on Stigler (1962), Mortensen (1970) andsubsequent literature). Entrepreneurs can engage in on-the-job search for a wage job. Whena su�ciently attractive job o�er arrives, entrepreneurs will take it and shut down their �rm.E�ectively, the anticipated shutdown probability drives a wedge between the cost of capitaland the return to capital. I show that, viewed from the perspective of the microenterprise,it serves as an additional discount factor in the intertemporal asset allocation problem. Thisis the transmission channel between occupational and entrepreneurial choices. Four criticalassumptions let this non-separability arise. First, the choice between being a self-employedowner-manager of a small �rm (an entrepreneur) and holding a wage job is discrete. Second,management cannot be passed on to an external manager. Third, invested capital is speci�cto the �rm and cannot be (fully) liquidated. Fourth, there is ex-ante uncertainty about theavailability and attractiveness of wage jobs at the time of investment.

In the limit � when attractive wage employment opportunities are absent � the shutdownprobability goes to zero, and the model collapses to a neoclassical investment model. Insuch a case, the problem of the �rm is separable from occupational choices of the owner.Under non-separability of occupational and entrepreneurial choices, the schedule of optimalinvestment can be convex, implying the existence of multiple local equilibria in optimal�rm size. In a dynamic setting, initial asset endowment determines which equilibrium isreached, and a `poverty trap' may arise (see Kraay and McKenzie (2014)). Comparativestatics show that entrepreneurs reduce investment into their �rm when they anticipate wageemployment to be more attractive, either because of an increase in the o�er probability,or due to a stochastic improvement in the wage o�er distribution. The intuition is thatbetter outside opportunities increase the likelihood of leaving entrepreneurship, which inprobability reduces the states of the world in which the fruits from investment into the �rmcan be reaped. Consequently, entrepreneurs choose to lower their exposure.

The model is able to replicate several stylised facts from recent literature on �rms in devel-oping countries. It formalises how choices regarding �rm size can depend on occupationalchoices of entrepreneurs. The resulting wedge between the marginal product and the cost ofcapital rationalises the `puzzle of high returns'. The model can further replicate the �ndingsthat access to �nance has the biggest bene�ts for ex-ante more successful microenterprises,and that the size of start-up grants can determine �rm size of new entrant entrepreneurs.

In the second part of the paper, I test this mechanism empricially. A test of the theoryrequires variation in wage-employment opportunities available to entrepreneurs that is exo-genous to other determinants of the returns to capital. I exploit the di�erential impact

3

Page 4: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

1 INTRODUCTION

on wage employment opportunities of a trade shock: the boom of Mexican manufacturingexports from 2005-2014, and the change in the composition of Mexican exports to the USas more and more high-end manufacturing moved to the country. The boom was brie�yinterruped by a bust during the US �nancial crisis of 2008/9, which temporarily destroyedclose to half a million manufacturing jobs in Mexico. This bust and the subsequent recoveryconstitute additional sources of variation.

Self-employed entrepreneurs in Mexico and their exposure to export manufacturing jobsprovide an ideal testing ground for my theory. I show descriptively that all the key modelassumptions conform to the empirical setting: in Mexico, occupational choice is discrete, thevast majority of exiting entrepreneurs cannot hand over the management of their �rm tosomeone else, and �rm assets can only be sold at a large markdown. Furthermore, jobs inexport manufacturing are �nancially attractive to almost all entrepreneurs, yet the marketsfor small �rms and export establishments are mostly non-overlapping. I assemble a datasetthat includes almost one million small �rms embedded into more than 40,000 unique locallabour market constellations that span the decade from 2005-2014. Using the universe ofsocial security records, I am able to build a quarterly panel of export manufacturing jobs inevery municipality in Mexico, which I de�ne as a local labour market.

The key challenge to identi�cation is that export manufacturing employment might be cor-related with factors that directly impact the productivity (and hence the returns to capital)of small �rms. I isolate variation in employment that is exogenous to conditions of thelocal economy by interacting national-level variation in exports at the subsector level witha local labour market's pre-determined exposure to di�erent export industries. The chal-lenge which remains is that even purely exogenous labour market shocks can a�ect marginsother than occupational choice, for instance through equilibrium adjustments of wages andprices. I overcome these concerns in two ways. First, I directly look for evidence on localprice adjustments, and drop the segments of the market for which I �nd such evidence fromthe analysis. Second, closely guided by my theoretical model, I explore di�erences in myestimated e�ects over time. In particular, my theory of the mechanism predicts that onlyexpected changes in wage employment opportunities in the future should play a role in it. Bycontrast, equilibrium adjustments will play out in the aftermath of an export manufacturingshock.

I structure my main empirical �ndings into four steps. First, I show that predicted changesin export manufacturing employment at the level of a municipality are a good predictor ofactual changes in this variable. This con�rms that the variation that I isolate constitutesa signi�cant shock to local labour markets. I call the predicted local export manufacturingchanges `export manufacturing shocks'. I also show that it is really the interaction betweeninitial exposure to export industries and the national aggregate volume of exports per in-dustry that constitutes this shock, not the individual components. Second, I documentadjustments in local labour markets to export manufacturing shocks. I report how the shocka�ects transitions into export manufacturing, but not wages in the sector. This is consistentwith a model of search frictions where shocks increase the probability of obtaining o�ers, butnot the value of o�ers. Third, I �nd that export shocks in the local municipality increasethe propensity of entrepreneurs to leave self-employment. In terms of timing, this �ndingextends to quarters after an export manufacturing shock, but not before. Fourth, I �nd that

4

Page 5: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

1 INTRODUCTION

export shocks negatively a�ect the growth of small �rms: entrepreneurs exposed to locallabour markets where more new wage employment opportunities are created are less likelyto hire new workers, compared to �rms in labour markets without these opportunities. Im-portantly, these e�ects are present for up to a year before export manufacturing employmentshocks, but not after these shocks.

These �ndings are consistent with the predictions from my theoretical model. I providethree further pieces of complementary analysis to determine whether these �ndings arisethrough the model mechanism, or through alternative mechanisms with similar observableimplications. First, I explore the heterogeneity of e�ects among the main dimension that isrelevant for the model: whether an entrepreneur is really an `entrepreneurial type' or rathera `wage worker type'. I �nd that both e�ects on occupational choices and on entrepreneurialchoices are twice as large for the entrepreneurs who look like wage workers. Second, I directlyconsider alternative mechanisms. I focus on the two leading alternative explanations formy �ndings: local demand adjustments, and local wage adjustments. I document that formanufacturing entrepreneurs � who are more likely to operate in similar market segmentsas export manufacturing plants do � there is indeed evidence that alternative mechanismsare at play. I do not �nd such evidence for non-manufacturing entrepreneurs, for whom themain results uphold. Third, I provide direct evidence which suggests that the expectations ofemployment opportunities are what drives reductions in the growth of small �rms. I collecta dataset on announcements of large plant expansions in the auto manufacturing industryin Mexico, the largest among the export manufacturing industries. Using an event studyapproach, I document that small �rm growth only responds after expansions and the jobcreation associated with them have been made public.

Taken together, I interpret my �ndings as evidence that the mechanism I postulate is indeedrelevant. This has at least two important implications for policy. First, policies aimed atfostering growth of small �rms need to take into account the incentives created for entre-preneurs by the occupational choice channel, which can be an important mitigating factorfor the success of such policies. Second, from a macro perspective, the non-separability is achannel which causes formal employment in large �rms and informal employment to movein opposite directions. Whether this is a boon or a burden depends on the direction of theshock. For instance, a negative trade shock to formal �rms � such as during the U.S. crisis of2008/09 � can be mitigated by the fact that fewer wage employment opportunities increasethe incentives for local entrepreneurs to expand their �rms and hire workers, who are largelyinformal.

1.1 Related literature

By linking the perspectives of an entrepreneur as a �rm and the entrepreneur as a worker,this paper contributes to several strands of literature.

Occupational choice, entrepreneurship and �rm performance. First and foremost, Icontribute to the literature on entrepreneurship, occupational choice, and �rm performance(Lucas (1978), Kihlstrom and La�ont (1979), Jovanovic (1982), Hopenhayn (1992), Evansand Jovanovic (1989), Levine and Rubinstein (2017)). It has established a range of determ-

5

Page 6: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

1 INTRODUCTION

inants of �rm performance: capital constraints, human capital formation, character traits,learning and experimentation. It only considers occupational choice as far as entry and selec-tion into entrepreneurship is concerned. I add to this by showing how occupational choice canspill over into �rm decisions by in�uencing the �rm size of incumbent entrepreneurs, holdingany innate ability, traits or preferences �xed. One strand of literature asks how income op-portunities outside entrepreneurship in�uence risk taking of entrepreneurs (Polkovnichenko(2003) and Vereshchagina and Hopenhayn (2009); for empirical evidence Hombert, Schoar,Sraer, and Thesmar (2014) and Gottlieb, Townsend, and Xu (2016)). They assume a �xedwage or bene�t as the outside option. By contrast, I focus on the role of uncertainty thatentrepreneurs face in the wage labour market, not the risk of running a business. To thebest of my knowledge, this is the �rst contribution which considers this mechanism, togetherwith a paper that was developed in parallel by Humphries (2017). I di�er from Humphries inthat I focus on incentives for incumbent entrepreneurs instead of �rm size decisions by newentrants. This enables me to derive testable comparable statics that allow for separating therole of opportunities in wage employment from the one of innate characteristics. I furtherprovide a rigoruous empirical test of the model, and my �ndings provide evidence for theexistence of this novel mechanism from a large middle-income country.

Labour markets and microenterpises in developing countries. The mechanismthat I propose is particularly relevant for developing, where the employment share of self-employment is large compared to rich countries. A recurring theme in the literature on thenature of occupational choice in developing countries (going back to Lewis (1954), Harrisand Todaro (1970), and Fields (1975)) is the role of self-employment and the nature of en-trepreneurship (Magnac (1991), Bosch and Maloney (2010), Günther and Launov (2012),Haywood and Falco (2016)). My model nests both the `competitive' and `segmented' view ofself-employment in developing country labour markets. It takes agents as utility maximiserswho optimally choose from the set of employment opportunities available to them. At thesame time it acknowledges the presence of frictions which imply that not all potentially at-tractive opportunities are available at all times; and that there is ex ante uncertainty abouttheir availability.

Many of the conceptual ideas on constraints �rm performance have been recently put totest in developing countries through �eld experiments that randomly eased the constraints.McKenzie and Woodru� (2008), De Mel, McKenzie, and Woodru� (2008) and Fafchamps,McKenzie, Quinn, and Woodru� (2014) ease capital constraints for existing microentrepren-eurs; Blattman, Fiala, and Martinez (2014), Fafchamps and Quinn (2016), and McKenzie(2017) do the same for new prospective entrants into entrepreneurship, at di�erent scales.Karlan and Valdivia (2011) and Karlan, Knight, and Udry (2015) aim to shock entrepreneur-ial ability through business training and consulting. A very recent strand of this literatureexamines whether constraints in the form of labour market frictions prevent microenterprisesfrom hiring more workers (Hardy and McCasland (2015), Cohen (2016), de Mel, McKenzie,and Woodru� (2016)). I contribute to this discussion by considering a margin of choice bymicroentrepreneurs that has not received attention by this literature: the decision to leavethe �rm for a di�erent occupation, for example to take up a wage job. I show that this candirectly be a constraint to the size and growth of small �rms. It furthermore can explain anumber of stylised facts on investment from this experimental literature.

6

Page 7: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

1 INTRODUCTION

The model and mechanism in this paper can in principle be applied more generally thanjust in developing countries. Ir relies on a small number of institutional features and marketfailures that can be expected to apply in rich countries just as well as in developing countries.5

Entrepreneurship in rich countries has again received increased attention very recently, dueto the rise in self-employment and similar work arrangements, notably in the gig economy(documented by Katz and Krueger (2016) for the US and Adam, Miller, and Pope (2017) forthe UK). To some degree this constitutes a convergence of developing and developed countrylabour markets.

Local labour market e�ects of trade exposureWith my empirical analysis that exploitsa shock to the export manufacturing industry in Mexico, I contribute to the literature on thee�ects of trade shocks on local labour markets. Being mainly concerned with the e�ects ofimport competition from emerging economies on labour markets of advanced countries suchas the US (e.g., Autor, Dorn, and Hanson (2013)), Acemoglu, Autor, Dorn, Hanson, andPrice (2016)) or Germany (Dauth, Findeisen, and Suedekum (2014)) it has paid less atten-tion to the �opside of the coing: job creation in developing countries as a result of o�shoringand export-led growth. Notable exceptions are Atkin (2016) and Heath and Mobarak (2015)who show how local factory openings alter the opportunity cost of education in Mexico andBangladesh.6 McCaig and Pavcnik (2014)'s analysis of trade-induced labor reallocation inVietnam, unlike most previous work, pays explicit attention to the role of ment in devel-oping country labour markets. Methodologically, my paper joins Adão (2015) in exploitingpredicted local trade shocks as identifying variation in an occupational choice model.

Marginal product dispersion and `misallocation'. A regular empirical �nding in theliterature on �rms and productivity is that marginal products are not equalised across �rms,even within narrowly de�ned industries (e.g. Söderbom and Teal (2004), Bartelsman, Halti-wanger, and Scarpetta (2013)). Hsieh and Klenow (2009) interpret this as evidence for mis-allocation, in the sense that aggregate productivity and output should be higher if factorswere reallocated across �rms. David, Hopenhayn, and Venkateswaran (2016) argue thatinformation asymmetries are one such source of frictions, causing large ex-post losses in pro-ductivity and output. I show that uncertainty in the market for entrepreneurs' labour canbe one source of such information frictions. My model replicates the empirical �nding ofdispersed marginal products of capital across �rm. This arises because entrepreneurs withdi�erent option values in wage employment apply di�erent e�ective discount factors on thereturn to capital, even if their �rms are identical.

Non-separability in production and labour. On a more general level, I contribute tothe literature that uncovers non-separability of choices across markets. Such results generallyrely on missing markets because of market failures, here the missing markets for manageriallabour in small �rms. Prominent examples of non-separability results appear in development

5Indeed, Humphries (2017) documents that model similar to mine is consistent with a number of stylisedfacts on entrepreneurship in rich countries. He further shows that, inserted into a life-cycle framework, sucha model can replicate the rich patterns of self-employment found in Swedish data. He does not attempt ata proper test of the model, though.

6Also, Dix-Carneiro (2014) and Costa, Garred, and Pessoa (2016) analyse labour market adjustment totrade shocks in Brazil, and Blattman and Dercon (2016) study occupational choice responses to randomallocations of factory jobs in Ethiopia.

7

Page 8: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

economics for production and consumption of household farms (De Janvry, Fafchamps, andSadoulet (1991), Benjamin (1992); LaFave and Thomas (2016))7 or in labour economics forfertility and female labour force participation (Killingsworth and Heckman (1986), Ecksteinand Wolpin (1989), Blundell, Costa Dias, Meghir, and Shaw (2016)). Poverty traps are aparticular kind of non-separability between initial wealth and the stable equilibrium thatis eventually reached in a dynamic model. Poverty traps in a production economy usuallyrequire a deviation from neoclassical assumptions about the production technology (Baner-jee and Newman (1993), Piketty (1997), Aghion and Bolton (1997)) or about preferences(Banerjee and Mullainathan (2010)), in addition to missing credit markets. In this paper,I show how poverty traps can arise in a model with neoclassical (convex) technology andpreferences. Instead, the poverty trap is created by a discrete-continuous choice model withirreversible investments.

2 Model

2.1 Conceptual framework

The framework starts with the idea that owners of small businesses in urban areas have accessto an active market for wage jobs. Most of their daily decisions about the business may beon the intensive margin � decisions about buying, selling, and investments. Entrepreneurswill periodically consider the extensive margin too � that is, whether to continue or to shutdown their business. For large �rms, exit decisions are based on alternative uses of capital,and may follow a learning story (Evans (1987), Jovanovic (1982)): �rms exit the industrywhen they have observed their performance su�ciently to conclude that capital is not usedin the �rm more e�ciently than in the market. For small owner-managed �rms and self-employed entrepreneurs, however, the opportunity cost of their own labour is perhaps reallythe binding constraint to �rm survival.

There is a crucial di�erence in how information is revealed in markets for capital and labour,which determine the respective value of alternative uses of production factors. Financialmarkets make anonymous o�ers of contracts to the general public, and investors are able toaccept such contracts at any point in time. Employers, on the other hand, make job o�ersonly to speci�c applicants. At the pre-contracting stage, workers only know the distributionof wage contracts in the market, and learn about the value of a speci�c contract only when anpersonal o�er has been made to them. This is formalised in the canonical partial equilibriumsearch model (Stigler (1962), Mortensen (1970), McCall (1970)) where workers go aroundsearching for the optimal contract, never knowing precisely at which moment in time anacceptable o�er will be made to them. This uncertainty about the opportunity cost of theirown labour can have crucial implications for how entrepreneurs conduct their business.

I convey this idea in a simple but generalisable model of an individual entrepreneur. I nestan otherwise standard model of investment behaviour under uncertainty within the canonicaltwo-sector labour market search model.8 The self-employed entrepreneur engages in on-the-

7LaFave and Thomas (2016) refer to non-separabiliy of farm households' production and consumption,where there is a clear order to the timing of decisions, as `non-recursivity'.

8Both of these basic models are widely used textbook models, but to my knowledge have not yet been

8

Page 9: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

job search for a wage job, meaning that with a certain probability, a wage o�er arrives.9 Theagent faces a distribution of wages and only learns about the value of a particular possiblewage o�er if and when it arrives.10 Entrepreneurs have to decide about how much to investinto their business before an eventual job o�er is made. The invested capital is speci�c tothe �rm and cannot be liquidated. The reservation wage summarises entrepreneurs' optimalown future occupational choice. A static two-stage model of an individual entrepreneurembedded in a perfect capital market is su�cient to express all the key insights into themechanism. It gives a closed-form solution for the reservation wage, which allows me tokeep the algebra simple and focus on the intuition of how occupational choice interacts withinvestment behaviour. I provide some extensions to the model after the discussion.

The model is able to generate two key insights. First, it can rationalise why marginalproducts of capital di�er in otherwise identical �rms, and why they are higher than themarket interest rate. Individuals with better prospects in wage employment will tend towardsa smaller capital stock, in anticipation of the possibility that they may shut down the �rm� and lose their investment � when an attractive job o�er arrives. The model's comparativestatics are directly useful as an empirical test of this prediction. An improvement in eitherthe o�er probability or the wage o�er distribution shifts incentives towards a smaller �rmsize. Second, the model can generate multiple local solutions to the conditions for optimal�rm size. Changes in the economic environment may then cause agents into select to adi�erent, distant equilibrium.

2.2 A model of investment under on-the-job-search

There are two occupations: self-employment and wage-employment. The agent is initiallyself-employed as the owner of a business (an entrepreneur) with access to a productiontechnology π(k). The timing is as following. She rents capital k on a perfectly competitivemarket at rental rate r. Capital is contracted and then, before production is realised, a wageo�er is drawn from a compound lottery. There is a probability λ that the agent receivesa wage o�er. This allows for the presence of search frictions. If an o�er is made, it isitself drawn from a distribution F (w). The stochastic distributions are known to the agent.The agent decides to accept or to reject the wage o�er. Production, wage payments, andconsumption are realised.

Four assumptions, which I maintain throughout, are critical for the occupational choice

combined in this form in the literature. The model of investment behaviour is identical to the Ramsey (1928)neoclassical growth model with a representative agent. The search model follows the standard textbookversion of the Mortensen (1970) model.

9Here I model search as an exogenous process. In Chapter ?? I will include both a costly search ef-fort choice and capital accumulation into an occupational choice model between self-employment, wageemployment, and non-employment. In such a model, agents have to trade o� two potential investments:investments into capital to increase future earning capacity in self-employment, and investment into costlysearch to increase the probability of receiving a job o�er.

10For simplicity, I do not model uncertainty in business pro�ts. Instead, incomes from self-employmentare assumed to be deterministic. This implies that agents know about their entrepreneurial ability, and thatthere is no market risk to running a business. Since I model entrepreneurs as risk-neutral, this assumptionis not restrictive. It allows me to focus on the mechanism of occupational choices.

9

Page 10: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

mechanism that I model:

Assumption 1. Market environment

1a. Discrete occupational choice: An individual can either be self-employed or wage-employed at a given point in time.

1b. Missing market for managerial labour: Production in a �rm only takes placewhen the �rm owner is self-employed.

1c. Firm-specific capital: Capital invested in the �rm cannot be liquidated.

1d. Search frictions and wage dispersion: A wage o�er arrives with probability 0 ≤λ < 1. F (w; a) is the continous distribution function of wage o�ers w parameterisedby a that admits �rst order stochastic dominance in a:

F (w, a) =

ˆ w

−∞f(x, a) dx with

ˆ +∞

−∞f(x, a) dx = 1;

Fa(w, a) =∂F (w, a)

∂a≤ 0.

Together, the assumption of a missing market for managerial labour and discreteness ofoccupational choices embody the principle that ownership and control of small �rms cannotbe separated.11 Firm-speci�city of capital implies that occupational choices are not costless.12

The assumption of search frictions and wage dispersion introduces uncertainty about thearrival and value of wage-employment opportunities, which is crucial for occupational choicesto have any e�ect on input choices. The parameterisation in Assumption 1d lets me discuss animprovement in the wage distribution by means of comparative statics calculus on parametera. My notation omits this parameter where not required.

I further assume a neoclassical production function which ful�ls the Inada conditions:

Assumption 2. Neoclassical production function:

The production function π(k) is twice continuously di�erentiable and obeys the Inada condi-tions:

π(0) = 0; limk→0

π′(k) = ∞;

π′(k) > 0; limk→∞

π′(k) = 0;

π′′(k) < 0.

11Fama and Jensen (1983) in their essay �Separation of ownership and control� justify this as following (p.307):

[C]ombining decision and risk-bearing functions is e�cient in small noncomplex organizationsbecause the bene�ts of unrestricted risk sharing and specialization of decision functions areless than the costs that would be incurred to control the resulting agency problems. Theproprietorships, partnerships, and closed corporations observed in small scale production andservice activities are the best examples of classical entrepreneurial �rms in which the majordecision makers are also the major residual risk bearers.

12This cost is borne upon liquidation, at the end of life of a �rm. Alternatively, one could assume one-timeentry costs that are proportional to the capital stock of a �rm, and obtain a very similar solution. What isimportant is that this cost is independent of the lifetime of the �rm.

10

Page 11: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

For the static model, I make the following assumption on preferences:

Assumption 3. Preferences

Risk neutrality: The agent's utility function over consumption is given by u(c) = c.

The assumption of risk neutrality isolates the pro�t motives for capital investment fromother motives such as consumption smoothing.13 Given risk neutrality, expected utilityequals expected income. At the stage when capital is chosen, occupational choices can onlybe known in expectations. Expected utility at this stage can be written as

EU(k) ={

(1− λ) · [π(k)− rk] + λ ·(Emax(w, π(k))− rk

)}, (1)

where the expectation is taken over the wage distribution. Given that w is monotonic aswe integrate along its distribution, there will be a unique reservation wage z = π(k) abovewhich all wage o�ers are accepted. We can rewrite the problem (1) as:

EU(k) ={π(k) ·

[1− λ(1− F (π(k))

]+ λ

ˆπ(k)

w · f(w) dw − rk}. (2)

In this representation, there is a single choice variable, capital, which determines both �rmsize and occupational choice, given technology and the economic environment. The cost ofcapital has to be borne in all states of the world, whereas returns are only realised when theagent is self-employed.

I de�ne �rm size to be synonymous with the capital stock of a �rm. Given the assumptions,there is a determininstic one-to-one mapping between kapital stock and output.

2.3 Solution and characterisation

I start by characterising the solution to optimal �rm size in the absence of changes in theeconomic environment. I show that a solution to optimal �rm size is guaranteed to exist,but that local optima need not be unique. Next, I show how �rm size in the neighbourhoodof an optimum responds to changes in the economic environment. I derive local comparativestatics with respect to a �rst-order stochastic improvement in the wage o�er distribution,and to an increase in the probability that a wage o�er is made. I then discuss equilibriumselection and show how that even under multiple local equilibria, the model gives clear globalcomparative statics prediction.

2.3.1 Optimal �rm size

The agent chooses capital to maximise her expected utility. Occupational choice is discrete,and depends on incomes in the two sectors (wage employment, self-employment) as well as

13Since accepted wage o�ers will always be higher than pro�ts, and debt repayments are independent ofthe state of the world, a risk-averse agent might use capital to make a transfer in expectation from thehigher-income (wage-employment) to the lower-income (self-employment) state. These incentives are ruledout by assuming risk neutrality.

11

Page 12: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

on the degree of search frictions. The necessary �rst-order condition for optimal capitalchoice is given by:

EUk = π′(k) ·[1− λ(1− F (π(k)))

]− r = 0. (3)

The returns to capital are multiplied by the probability of staying in self-employment, whichI denote δ(k):

δ(k) = 1− λ(1− F (π(k))). (4)

If this probability δ(k) is less then one � i.e. if entrepreneurs take up a wage job with apositive probability � then the �rm chooses an optimal size where returns to capital canbe substantially higher than the rental rate of capital r. In other words, the returns fromcapital are discounted by the endougenous probability of leaving the �rm. This is the keyinsight of this model.

The slope of the marginal utility (i.e. the second derivative of the objective function) is givenby

EUkk = π′′(k) · δ(k) + π′(k) · δ′(k). (5)

The �rst-order condition is necessary for a solution to the entrepreneur's problem. Thesu�cient second-order condition for a local maximum requires (5) to be negative at a solutionto the �rst-order condition (3). The next section deals with existence and uniqueness of localoptima.

2.3.2 Existence and uniqueness

The following proposition summarises the results on existence and uniqueness of solutionsto the entrepreneur's problem.

Proposition 1. (Existence and uniqueness) There exists at least one solution to the �rst-order condition (3). There may exist more than one solution. If there is exactly one solution,it is a local maximum. If there is more than one solution, there may be multiple local maxima.

To prove existence, �rst note the limits of the marginal product of capital and of the prob-ability of staying in entrepreneurship, which are given by:

limk→0

π′k =∞; limk→0

δ(k) = 1− λ;

limk→∞

π′k = 0; limk→∞

δ(k) = 1.

Together, this implies that EUk approaches in�nity as capital reaches zero, and that itapproaches a negative bound (−r) for very large values of capital. By continuity, there mustbe a crossing point where EUk = 0. This establishes existence of a solution to (3). Giventhat expected utility increases in the lower limit and decreases in the upper limit, therefurther must be a solution which is a maximum.

To consider uniqueness, note that although the marginal utility in (3) needs to eventuallyturn negative somewhere in order to satisfy the limit conditions, it does not need to decreasefor every level of capital. This follows from the sign of (5), which is in general ambiguous. Ifit is possible that marginal utility increases locally, expected utility is not necessarily globally

12

Page 13: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

concave. Therefore, multiple solutions to the �rst-order condition (3) may exist. Togetherwith what we established about the slope of expected utility in the limit, this means that ifthere are multiple solutions to the �rst-order condition, there must be multiple local maxima(unless all but one of the solutions are saddle points).

If there are multiple local maxima, only one of these will be a global optimum which satis�esutility maximisation. I will return to the problem of selecting among multiple solutionsafter discussing how changes to the economic environment a�ect the choice of capital in theneighbourhood of local solutions.

2.3.3 Local comparative statics

I use comparative statics calculus to describe how local optima change with changes in theeconomic environment. In particular, I evaluate the e�ects of changes to the attractivenessand availability of wage-employment. I model an improvement in the wage distribution as a�rst order stochastically-dominant shift in the wage distribution by increasing the parametera (see Assumption (1d)). I obtain the partial derivative

dk

da= −EUka

EUkk= −

λ · ∂F (π(k);a)∂a

· π′(k)

EUkk≤ 0. (6)

Since EUkk < 0 at a local maximum, the sign of (6) is negative. A stochastic improvementin the wage o�er distribution reduces optimal �rm size.14 A shock to the wage distributionwill transmit into a shock to δ(k). For a given reservation wage, it is now more likely thata wage o�er is accepted. This brings down the optimal capital investments and in turn thereservation wage, until a new optimum is reached.

The same qualitative result obtains when the o�er probablity λ is hit with a shock, insteadof the wage distribution, since:

dk

dλ=

(1− F (π(k); a)) · π′(k)

EUkk≤ 0. (7)

I summarise these local comparative statics in Lemma 1:

Lemma 1. (Local comparative statics) (i) An improvement in the wage o�er distributionor (ii) an increase in the o�er probability reduces �rm size in the neighbourhood of a localmaximum, if it changes the probability of staying in self-employment δ(k).

2.3.4 Equilibrium selection and global comparative statics

The possibility of multiple local optima requires us to examine not only local, but alsoglobal, comparative statics. With multiplicity, a small change in the economic environment

14More precisely, it reduces k where ∂F (w;a)∂a |w=π(k) < 0; so at levels of k where π(k) overlaps with the

range of the wage distribution F (w; a). In other words, changes in wage opportunities matter only as faras they change the mass of wages above the reservation wage. In regions where pro�ts are lower or higherthan any wage o�ered in the market, this condition is not true. Similarly, equation (7) is only true forF (π(k), a) < 1. More wage o�ers only matter if there is some wage o�er that could be acceptable.

13

Page 14: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

can change the order in which local optima are ranked. This would then imply large changesin chosen capital as the agent jumps between local optima.

Ranks are a function of the expected utility associated with each of the local optima. Thereis no analytical way to express a ranking of solutions in a general; we would need to knowall the primitives of the model in order to numerically evaluate expected utility. However,we can use comparative statics on the expected utility function to analytically establish thedirection and relative magnitude in which expected utility, and therefore the rank evaluationof local maxima, are being shifted.

First, we can show that both an increase in a and an increase in λ increase expected utilityas a function of capital:

EUa(k) = −λ ·ˆπ(k)

∂F (w; a)

∂adw ≥ 0; (8)

EUλ(k) = −π(k) · [1− F (π(k))] +

ˆπ(k)

w · f(w) dw

=

ˆπ(k)

(1− F (w)) dw ≥ 0. (9)

where the last step involves performing an integration by parts. We can once more di�erenti-ate the partial derivatives (8) and (9) to show that the magnitude of the change in expectedutilities decreases with k:

EUak = λ · ∂F (π(k); a)

∂a· π′(k) ≤ 0; (10)

EUλk = − (1− F (π(k))) · π′(k) ≤ 0. (11)

I summarise the results in Lemma 2.

Lemma 2. (Comparative statics of expected utility) (i) An improvement in the wage o�erdistribution and (ii) an increase in the o�er probability weakly increase expected utility at alllevels of capital. This increase is stronger the lower the level of capital.

These results allow us to make clear predictions about comparative statics without havingto evaluate the expected utility, nor having to know which of potentially multiple solutionswas selected before the change in the economic environment. Our results all point in thesame direction: improvements in wage employment translate into a lower �rm size. This canoccur either through a local marginal change to the new optimum (Lemma 1) or througha jump from a higher to a lower optimum (a direct implication of Lemma 2). We cannotanalytically distinguish between which of these two mechanisms is at play. The globalcomparatives statics that I summmarise in Proposition 2, however, do not require us tomake this distinction:

Proposition 2. (Global comparative statics) (i) An improvement in the wage o�er distri-bution and (ii) an increase in the o�er probability weakly reduce optimal �rm size.

In Figure 2, I illustrate the characterisations of the model which we obtained so far, using aparametric form for the functions π and f . I use a Cobb-Douglas production function, and

14

Page 15: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

a lognormal wage distribution.15 The left-hand �gure depicts the marginal utility EUk andthe �rst-order condition (the crossing point of EUk with the vertical axis). I illustrate twoscenarios: a baseline scenario; and an alternative scenario where a small upward shift in thewage o�er distribution has taken place.

I �rst focus on the baseline scenario, depicted in blue dash-dotted curves. The left-hand�gure graphs marginal utility. There are three crossing points with the x-axis, marked withthinly dotted vertical lines. These are the three solutions to the �rst-order conditions. Theright-hand side depicts expected utility. The level of capital associated with each of thecrossing points is marked in the same way as before. The odd crossing points (A and C)are local maxima, the crossing point in the middle B is a local minimum. If there aremultiple local maxima, only one of these will be a global optimum which satis�es utilitymaximisation. In this example, C is the global maximum. The thin, dotted horizontal linemarks the maximum expected utility in the baseline case, for reference.

Consider what happens after a wage o�er improvement (graphed in red solid curves). Mar-ginal utility shifts by a small amount � while this is only visible in parts of the distribution,it can be veri�ed that this is true everywhere. As a result, both local �rm size optima shiftdownward (although only the downward shift of C is visible, with the new solution markedby an unlabeled grey dashed line). This illustrates the local comparative statics result inLemma 1. The shift in marginal utility is related to the shift in δ(k). At point A, most wageo�ers are already accepted at baseline. The wage improvement therefore does little to alterδ(k), the probability of staying. The next section provides a more extensive discussion onthe intuition.

The shift in expected utility, shown in the right-hand side of Figure 2, is much more dramaticthan the shifts in marginal utility and in local optima. Expected utility improves visibly as aresult of better wages. Moreover, these improvements are larger for smaller values of capital.This illustrates Lemma 2. As a result, the new solution close to C no longer is a globaloptimum. The change in the environment leads the entrepreneur to choose a much smaller�rm size at A.

2.4 Discussion

In this section, I o�er additional discussion on mechanisms, implications and predictions ofthe model. I start by providing intuition for why a either a low or a high-capital equilibriumcan be potentially optimal, and in what sense they constitute underinvestment. Second,I discuss the implications of introducing cross-sectional di�erences in entrepreneurial abil-ity. Finally, I show how the model is consistent with a number of stylised facts on theheterogeneity of microenterprises.

2.4.1 Intuition

How does non-separability between entrepreneurial and occupational choices arise in thismodel? Inspecting the �rst-order condition (3), we see that the di�erence to the neoclassical

15Appendix ?? lists the details of parameteric assumption for each �gure.

15

Page 16: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

model arises from introducing the probability of staying, δ(k), into the �rm's problem. Thee�ective discounting of pro�ts drives a wedge between the marginal product of capital andthe interest rate. This is how the model expresses that when entrepreneurs consider theirown occupational choices, they may decide to run a smaller �rm than their potential wouldallow them to. The fact that e�ective discounting varies with �rm size also leads to thepossibility that both A and C are self-sustaining optima. I discuss the intuition behindthese observations with Figure 3.

In the right-hand panel, I plot the model in the rates-of-return space. Any equilibriumrequires the marginal product of capital π′ to equal the e�ective cost of capital r/δ. Thisis clear from the �rst order condition. Potential equilibria are given by the local optima Aand C from above. At the low-capital solution A, the level of pro�ts that the entrepreneurmakes from the small capital stock kA is low. This directly results in a low reservation wagezA = π(kA). Most wage o�ers will be above zA, and the entrepreneur will accept them. Exante, the probability of leaving the �rm is high, reaching λ if all potential o�ers would beaccepted. With a high probability of leaving the �rm, a forward-looking entrepreneur appliesa high e�ective discount factor, and will choose a small capital stock. Hence, the choices ofcapital stock and reservation wage at A are mutually consistent. Similarly, at point C, ahigh capital stock implies a high reservation wage zC , which means that few wage o�ers willbe attractive to the entrepreneur. She stays in self-employment with a high probability � soit makes sense for her to choose a large capital stock.

The right-hand panel of Figure 3 also shows the neoclassical benchmark case, in which the�rm's decision are separable from the entrepreneur's occupational choices. My model reducesto the neoclassical model if λ = 0 � if there are no o�ers of wage-employment � or if pro�tsstrictly dominate wages. In the neoclassical model, δ(k) is always equal to one: entrepreneurscompare the return to capital in the �rm only to the opportunity cost of capital r. Optimal�rm size under that model is reached at CR.

Whenever the e�ective discount factor δ(k) is smaller than one, optimal �rm size is smallerthan the potentially achievable maximum CR. Again, we can verify this from the �rst-ordercondition (3). The degree of underinvestment is particularly strong when the low-capitalsolution A is selected as an equilibrium. As Lemma 2 has shown, this is more likely whenthere are many attractive wage employment opportunities for entrepreneurs. In such a case,the model solution is far from the benchmark. By contrast, when the high-capital equilibriumC is selected, the model solution is closer to the benchmark. A su�cient condition forachieving a unique optimum is π′(k)δ(k) decreasing in k; which is equivalent to δ−1(k)/π′(k)satisfying the monotone likelihood ratio property. Intuitively, if the rise in the probabilityof staying in entrepreneurship is less steep than the fall in the marginal product of capital,then discounted (expected) returns fall monotonically. If expected returns are monotonic,then only one solution to the return to capital satis�es the optimality conditions.

16

Page 17: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

2 MODEL

2.4.2 Heterogeneous entrepreneurial ability

I explore the e�ects of heterogeneous entrepreneurial ability with Figure 4. I �rst varyentrepreneurial ability while keeping all other parameters constant.16 This changes therelative location of entrepreneurial pro�ts and the wage o�er distribution in the cross-section.I then contrast comparative statics of a given wage increase at di�erent levels of ability.

The middle panels take the previous case as a baseline, where C is the optimal �rm size, asbefore. Lower ability entrepreneurs (left panels) choose a low-capital solution A1. Higherability entrepreneurs (right panels) choose a high-capital solution C2. In the illustratedexamples, the optima for both low and high ability entrepreneurs are unique; whereas baselineability entrepreneurs select among two local maxima.

Now consider again an increase in the wage o�er distribution through an increase in a. Notethat the discount factor depends on both ability (through the production function) and onthe state of the labour market (through f and λ). Therefore, a di�erent δ applies in eachcase. For ease of readibility, my notation continues to simply use δ; understand as beingparameter-speci�c. In each of the upper panels, the desired rate of return (the e�ective costof capital) changes to a new level; graphed as solid red curves. The equilibrium adjustmentoccurs through a mix in changes to local optima and in changes to equilibrium selection, ascan be seen from the lower panels. For lower ability entrepreneurs, optimal �rm size changeslocally to A′1. For higher ability entrepreneurs, a low-capital local maximum begins to exist,but it is still optimal to maintain a high-capital �rm; although a slightly lower one thanbefore (at C ′2 < C2). Entrepreneurs with baseline ability switch between local optima (fromC to A′).

Importantly, in each of the three cases, optimal �rm size decreases as a result of the wageshock: Proposition 2 holds for any admissible production function. While entrepreneurialability in itself is an important determinant of �rm size, it does not alter the qualitativeconclusions we draw from our comparative statics.

2.4.3 Consistency with stylised facts

Before I proceed to �nding an empirical setting that allows me to carry out an empirical testof these comparative statics, I will discuss a number of stylised facts from recent literatureon �rms in developing countries which the model is able to replicate.

High returns to capital. The model is able to rationalise the observation that despitereturns to capital that are much higher than the prevailing interest rate for loans, microen-terprises do not grow (McKenzie and Woodru� (2008), De Mel, McKenzie, and Woodru�(2008) and Fafchamps, McKenzie, Quinn, and Woodru� (2014)). This is particularly salientin a low-capital equilibrium (point A in Figure 2). The �gure illustrates how the return tocapital (π′) is much higher than the interest rate r at this point. Even when a high-capitalequilibrium is selected, a wedge exists between the returns to capital in the �rm and theinterest rate as long as there is a positive probability of receiving a wage o�er higher thanthe reservation wage.

16Here, higher entrepreneurial ability implies a Hicks-neutral upward productivity shift in the productionfunction.

17

Page 18: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

3 EMPIRICAL STRATEGY

The introduction of micro�nance has lasting impacts only for the ex-ante largest

microenterprises. Randomised evaluation have found that, while transformative e�ects ofimproved access to �nance are absent for the average microentrepreneur, there can be largepositive e�ects on business pro�ts for the ex ante most sucessful entrepreneurs (Angelucci,Karlan, and Zinman (2015), Banerjee, Du�o, Glennerster, and Kinnan (2015), Banerjee,Breza, Du�o, and Kinnan (2015), Fafchamps, McKenzie, Quinn, and Woodru� (2014)). Iillustrate this prediction in panels C of Figure 4. I take the situation from the top panelsof the Figure and assume that because of existing credit constraints, only low equilibria (atA1, A and A2) are selected. Now I introduce microcredit, which allows each entrepreneur toborrow an additional amount ∆k. As the �gure illustrates, this is not enough to bring thelower-ability entrepreneurs to a �rm size which would lead to a higher expected utility. Forexample, the mid-ability entrepreneur would be able to achieve a �rm size B with credit,which is actually a utility-minimising choice. Therefore, these entrepreneurs will not borrowfor productive purposes. The lower-ability entrepreneur would not prefer to run a larger�rm even if she faced no credit constraints at all. The high-ability entrepreneur could builda larger �rm by borrowing ∆k which would increase her expected utility; she therefore willchoose to take out a loan.

Large capital grants can help establish large and lasting �rms. This is a corollaryof the micro�nance case. Potential entrepreneurs that are given a large start-up grant canimmediately jump to a high equilibrium (such as C or C2), provided they have su�ciententrepreneurial ability. Entrepreneurs with a small start-up grant will start small and remainsmall, absent any further large shock to access to �nance. This matches evidence from start-up grants in several sub-Saharan countries (Fafchamps and Quinn (2016), McKenzie (2017)).

3 Empirical Strategy

Testing for separability between entrepreneurial and occupational decisions choices requiresvariation in wage employment opportunities available to entrepreneurs. For an empiricalcounterpart to the comparative static (??), we need to �nd a situation where wage employ-ment opportunities should in�uence entrepreneurial decisions only through their e�ect onthe probability entrepreneurs place on leaving self-employment in future periods, and notdirectly through their e�ect on future pro�ts made by the �rm. Having found such a situ-ation, we can then test whether variation in wage employment opportunities has an e�ect onthe size of entrepreneurs' �rms, and entrepreneurs' decision to stay in business or to leaveself-employment.

Going from the model to the data requires us to think about heterogeneity across individualsnot only in wage employment opportunities, but also in other dimensions. The model hasbeen set up as a model of an individual entrepreneur. We can think of all of constituentfunctions as individuals-speci�c. In any cross-section of individuals, we would expect to�nd di�erences across entrepreneurs in dimensions that are known to a�ect �rm size: adistribution of absolute and comparative advantages in a Roy-model sense; di�erences in riskpreferences; di�erences in the severity of credit constraints. Fortunately, the comparativestatics are robust, in the sense that they hold under for any of these types. As long as

18

Page 19: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

3 EMPIRICAL STRATEGY

individual heterogeneity is time-invariant (or, if time-varying, orthogonal to local shocksto wage-employment opportunities), we can run a test in �rst di�erences and be agonosticabout unobserved heterogeneity in ability, preferences or credit constraints.

The approach I take focusses on variation in wage employment opportunities across locallabour markets. I consider wage employment opportunities that are in principle available toany worker because they do not require a narrow set of skills. In particular, I exploit changesin employment created by a set of manufacturing industries dominated by blue-collar jobsin large �rms that produce for export to international markets.

There are three main concerns for identi�cation if one were to use observed variation inlocal export manufacturing jobs. First and foremost, omitted factors at the level of the locallabour market could drive both the growth in export manufacturing �rms, and the growthof small �rms. Second, there could be a link between export manufacturing and small �rmsthat operates through the direct pro�t channel, instead of through the occupational choicechannel. For instance, linkages in the product markets or competition in the demand forlabour, could directly in�uence net pro�ts of small �rms. Third, if small �rms themselvesare export manufacturers, a mechanical reverse causal e�ect could be at play.

My starting point to circumvent these identi�cation problems is to isolate the variation inexport manufacturing employment that occurs because of shocks to foreign demand for goodsproduced by Mexican export manufacturers. I construct a measure of job creation in theexport manufacturing sector in the local labour market that is purely due to the variation inforeign demand for the goods that the local labour markets specialised in during an initialperiod. In particular, I express predicted job creation in export manufacturing (XME), perthousand people of working-age, in local labour market m, as:(

∆X̂ME

WAPop

)m,t

=∑

g∈{XM}

Em,0,gWAPopm,t−1

· ∆xt,gx0,g

(12)

whereby I create the predicted number of jobs per industry g as the interaction of initial localemployment in this industry Em,0,g with the change of national Mexican exports of goods

from this industry between the current period t and the previous period ∆xt,gx0,g

, expressed

as an index normalised to initial period exports. Recent empirical work has used similarmeasures to allocate trade shocks to local labour markets (e.g. Adão (2015), Autor, Dorn,and Hanson (2013), Topalova (2010)). In particular, I use the variation in bilateral exportsto the US, by far the biggest export market, which I allocate to local labour markets de�nedas municipalities. This allow me to exploit the negative shocks to exports following theUS �nancial crisis of 2008/09, and the recovery period in subsequent years. I normaliseeverything by the lagged working-age population (in thousand) and sum up over all exportmanufacturing industries to create a single job creation �gure for a local labour market andperiod. I usually take a period to be a trimester.

Using predicted instead of actual local job creation provides for variation that is independ-ent of economic forces operating at the local level other than initial exposure to exportmanufacturing, which can be controlled for. Focussing on a sector dominated by large, inter-nationally integrated �rms, I e�ectively separate the spectrum of �rms in two groups: small

19

Page 20: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

3 EMPIRICAL STRATEGY

�rms run by entrepreneurs who may have an outside option in local wage employment � thegroup for whom I expect my model to apply � and large export manufacturers whose choicesare in�uenced only by international demand. However, this strategy is still exposed to thepotential problem that e�ects on small �rm growth operate through the pro�t channel, andnot through occupational choice.

To address this remaining issue, I begin by noting that the fact that I exploit variation atfairly high frequency � a quarter � should mitigate the in�uence of price adjustment in localmarkets, which take time to play out.17 Further, we expect some direct e�ects on pro�tsto in�uence small �rm growth in the opposite direction to wage employment opportunities:while increased job creation should reduce small �rm growth through the occupational choicechannel, local demand spillovers should increase it. I overcome any remaining concerns intwo ways. First, spillovers to local goods or labour demand are testable as long as they aremanifested in observed equilibrium prices. This allows me to detect such spillovers, and toidentify market segment for which spillovers are less of a concern. Second, any future exportmanufacturing shocks should not in�uence current local price adjustments, thereby shuttingdown the direct channel. But they should in�uence entrepreneurial decisions through theanticipated occupational choice channel, according to what I posit in the model. I thereforeexploit the timing of entrepreneurial decisions and export manufacturing shocks.

Self-employed entrepreneurs in Mexico and their exposure to export manufacturing jobsprovide an ideal testing ground for my theory. I will present this argument in a numberof steps. I start by documenting descriptive evidence that supports the key assumptions ofmy theory � that there is a discrete choice between self-employment and wage work, thatsmall �rms cease to operate when the entrepreneur leaves self-employment, and that �rmassets are hard to liquidate. Then I show that my isolated variation in export manufacturingemployment creation is a strong predictor of observed job creation. Furthermore, I documentthe impact of XME shocks at the level of the local labour market. To previews results for thisintermediate step, I �nd no evidence for short-term adjustments in export manufacturingwages. Rather, the main response seems to happen through the number of job postings,manifested in worker transitions into wage-employment, including export manufacturingemployment. I then show at the individual level that entrepreneurs do take up jobs inexport manufacturing when they are created in their local labour market. This body ofevidence suggests that the business environment for small �rms in Mexico closely mirrorsmy theoretical setup. In this context, my estimate of the e�ect of future XME opportunitieson small �rm growth then provides a test of the model. I expect to �nd that small �rm growthis negatively a�ected by shocks to future employment opportunities of entrepreneurs. Thiswould imply that entrepreneurial choices cannot be separated from occupational choices.Having documented the result of this test, I will then look into cases where the theoreticalmechanism is particularly salient, and into cases where some of the key assumptions fail tohold. This will provide further con�dence in that �ndings speak to the model I am actually

17Shocks to manufacturing industries need not necessarily propagate through equilibrium adjustments atthe local market level, but can rather spread through industrial backward and forward linkages. For the US,Acemoglu, Autor, Dorn, Hanson, and Price (2016) �nd that only 10-15 percent of job losses due to Chineseimport competition can be explained by local adjustments. The most important channel are backward(upstream) industrial linkages with suppliers of directly exposed industries.

20

Page 21: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

4 DATA

testing, and help to distinguish to which degree they speak to alternative models.

3.1 Empirical speci�cations

To understand the e�ects on export job growth on local labour markets, I regress the pre-dicted quarterly change of predicted export manufacturing employment (XME) as de�nedin equation (12) on the quarterly change of outcomes at the level of the local labour market,de�ned by a municipality m:

∆ym,t = α + β

(∆X̂ME

WAPop

)m,t

+ x′m,tβ1 + ∆εm,t (13)

By specifying the regression model in the �rst di�erence, I remove any time-invariant factorsat the level of the municipality, which could in�uence both outcomes and the level of ex-port manufacturing employment. I also control for the sources of variation that are usedas inputs in the construction of predicted XME, and which could directly shift trends inoutcomes. In the vector xm,t I control for trimester �xed e�ects to pick up any aggregatetime series variation contained in the time series of exports; the initial XME share to capturedi�erential trends by initial condition of the municipality; and state �xed e�ects to avoidspurious correlation due to di�erent trends in regions where export manufacturing happensto be located. Per default, I always include these three sets of controls; and then may addadditional controls discussed later.

I estimate the e�ects of wage employment opportunities in export manufacturing on indi-vidual decisions of entrepreneur i using the following regression model:

∆yi,m,t = α + β

(∆X̂ME

WAPop

)m,t−s

+ x′m,tβ1 + x′

i,m,tβ2 + ∆εi,m,t (14)

where the coe�cient of interest is the e�ect of predicted XME on individual changes in out-comes between period t and t−1. I again control separately for the ingredients for the XMEprediction. In xi,m,t I include controls for individual demographics, �rm characteristics, andsize of the locality of residence. I furthermore allow for the possibility of shifting the changein predicted XME by s quarters into the future or the past. This allows for picking upanticipation of, and adjustment to, XME shocks.

4 Data

I build a dataset that links individual survey data on the Mexican working-age population toadministrative records covering the universe of formal employment in Mexican local labourmarkets. In order for my test to have power, I require a large sample covering many di�erentlocal labour market constellations. Individual data come from 40 trimestral waves of theENOE labour force survey that span the years 2005 to 2014. The ENOE is collected bythe national statistical institute, INEGI, and is targeted to cover around 120,000 households

21

Page 22: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

4 DATA

every trimester. The survey has a rotating panel structure, and individuals remain in thesurvey for 15 months (5 waves) if they remain resident in the same dwelling.

In all of my analysis, I restrict the sample to the working-age population aged 15 to 64.Over the 2005-2014 decade, I obtain more than 10 milllion individual observations. I payparticular attention to the trajectories of individuals who initially own and operate a non-agricultural business with a maximum of 10 workers as their main occupation. I retainaround 918,000 spells that run over two quarters, and where individuals start in own-accountwork in the �rst of the two periods. Information in the ENOE encompasses the kind ofvariables typically found in labour force surveys, including sociodemographics, employmentinformation, and income. A smaller subset of the entrepreneurs surveyed for the ENOEunderwent an additional, cross-sectional enterprise module called ENAMIN, carried out inthe fourth quarter of 2008, 2010 and 2012.18

The local labour markets I consider in this analysis are de�ned by municipalities. At the timeof the 2005 Census, Mexico was divided into 2,454 municipalities, and I use these municipalityboundaries throughout. I identify municipalities by the unique identi�er assigned by INEGI,and merge di�erent data sources by municipality. With this, I am able to capture variationin 41,484 unique local labour market constellations across space (municipalities) and time(quarters) that each overlap with at least one observation from the ENOE.

Social security records come from IMSS, a government institution tasked both with adminis-tering the public social insurance scheme, and with providing services, in particular health-care and pensions. Enrolment with IMSS is mandatory for all workers, thus by de�nitionthe set of all IMSS-a�liated workers is equivalent to the universe of formal employment inMexico. From disaggregated end-of-month records provided by IMSS, I construct a datasetcontaining the average number of IMSS a�liates in each trimester, for each municipality andindustry, identi�ed by 3-digit NAICS codes.19 I exlude the Federal District of Mexico City(D.F.) from my analysis because I am not able to map IMSS records to municipality-levelsub-units of the D.F.

For the construction of predicted export manufacturing employment per thousand workingage population, I obtain monthly time series of Mexican exports to the US at the levelof 3-digit NAICS industries from the US Census Bureau. To construct the working-agepopulation for each municipality and quarter, I combine population counts from the 2005and 2010 Census, and o�cial population predictions 2010-2050 from CONAPO, the NationalPopulation Council. My measure of working-age population is hence based on a model oflong-term population trends, and not exposed to short-term �uctuations.

18From the ENAMIN I obtain information on capital and investment for around 54,500 small �rms witha maximum of 10 workers. However, I refrain from using this data for my analysis. The measuring tool forcapital and investment changes between 2008 and 2010, so I cannot pool the three waves. This limits thepower of my tests. More crucially, patterns in the data suggest that there might be misclassi�cation in theentrepreneur status for the ENAMIN waves. There is a large spike in `transitions' out of self-employmentinto wage-employment for the ENAMIN waves, and back into self-employment in the following ENOE wave.

19IMSS maintain their own system of geographical and industry classi�cation. I match IMSS sectors to 3-digit NAICS sectors by hand, and perform the crosswalk from IMSS geographical units to INEGI municipalityboundaries constructed by IMSS.

22

Page 23: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

4 DATA

4.1 Mexican export manufacturing

Mexico has seen a boom in export-oriented industrialisation in recent decades, in particularsince the creation of NAFTA, the free-trade zone between Mexico, the US, and Canada, in1994. Sharing a 2000 mile land border with the largest economy in the world unsurprisinglyimplies that the lion's share of Mexican exports (more than 80 percent) goes to the US. Whilethe initial phases of export manufacturing were characterised by a prevalence of low-skilledjobs, many of whom were held by young women (Atkin (2016)), by 2005 employment in thissector was roughly equally distributed between industries in the garment trade with taskson the lower end of the skill spectrum, and high-skill intensive industries in electrical andmachinery manufacturing, including automotive.20

I de�ne the export manufacturing sector to consist of the ten 3-digit NAICS industries wheremore than 50 percent of output was exported to the US for all months in 2005-2007. This isthe period immediately prior to the drop in Mexican manufacturing exports that came withthe U.S. �nancial crisis.21 22 The ten industries that fall within this de�nition of export man-ufacturing are: Textile products mills; Apparel manufacturing; Leather and allied productsmanufacturing; Fabricated metal products manufacturing; Machinery and equipment man-ufacturing; Computer and electronic product manufacturing; Electrial equipment, applianceand component manufacturing; Transportation equipment manufacturing; Furniture andrelated products manufacturing; Miscellaneous manufacturing.

Taken together, the export manufacturing sector de�ned this way accounts for two thirdsof all bilateral exports from Mexico to the US; around USD 13 billion of a total of USD 19billion per month.23 It provided for an average of 2 million jobs during 2005-2014. Hereand for the rest of the paper, I de�ne export manufacturing employment as formal (IMSS-a�liated) jobs in any of the ten industries. Export manufacturing employment amounts toclose to 13 percent of all formal jobs, and 5.7 percent of all paid employment in Mexico.Within export manufacturing, the largest industry by employment and export volume istransportation equipment manufacturing: the automotive sector accounts for 25 percent ofall export manufacturing jobs, and 37 percent of exports.

The vast majority of export manufacturing jobs - 85 percent - are blue-collar jobs andrequire full-time employment.24 25 This is mirrored by the characteristics of workers in

20See Verhoogen (2008) on how in the mid-1990s Mexican manufacturing plants responded to negativetrade shocks with quality upgrading.

21This de�nition is in spirit similar to the way Atkin (2016) de�nes export manufacturing industries.However, I work with the 3-digit NAICS classi�cation system that is applied in both Mexican and USnational statistics, whereas Atkin harmonises to the international ISIC classi�cation. I also focus on exportindustries after 2005, while Atkin looks at the 1985-2000 period.

22I obtain information on manufacturing industry production from the INEGI Monthly ManufacturingIndustries Survey (EMIM).

23The largest exporting sector industry that falls not under my de�nition is oil and gas, which exportedUSD 2.4 bn per month, on average.

24Data on job and worker characteristics come from the ENOE and IMSS data, and data on small �rmcharacteristics come from the matched ENOE-ENAMIN. Sampling weights are used. Table 2 compares thesecharacteristics to those of other wage workers, and those of entrepreneurs.

25Full-time employment means working at least 35 hours per week during the survey reference period (thelast 7 days).

23

Page 24: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

4 DATA

export manufacturing: around 70 percent hold either an upper or lower secondary degree(which corresponds to 9-12 years of schooling), and they are on average younger than thegeneral working-age population. Almost all (92 percent) export manufacturing �rms areincorporated enterprises, and they are much larger than the average Mexican employer:three quarters of the jobs are in large �rms with more than 50 employees, 90 percent in �rmswith more than 20 employees, and 95 percent in �rms with more than 10 employees. As apoint of comparison, half of all wage jobs in Mexico are in �rms with 10 workers or less.

By visual inspection of the respective time series displayed in Figure 5, there is a closecorrelation between the real volume of exports and employment in the export manufacturingsector. Exports and employment were on a slightly increasing path until the end of 2007,then started decreasing in 2008 and fell dramatically until the middle of 2009, when theyagain recovered. From peak to bottom during the US �nancial crisis, seasonally adjustedexports fell by about 25 percent, and export manufacturing employment by about 20 percent.This collapse of manufacturing exports, and the subsequent recovery, amplify the variationin wage employment opportunities across local labour markets over time and space that Iwill exploit.

4.2 Self-employment in the Mexican labour market

Around 67 percent of the Mexican working-age population � individuals between 15 and64 percent of age � are gainfully employed. Of those in paid employment, a quarter areself-employed, 40 percent are in formal wage employment, and a third are in informal wage-employment. Around 11 million Mexicans were self-employed during 2005-2014, on average.The remainder of this section puts self-employment into the context of the Mexican labourmarket, with a particular emphasis on the �rm size distribution.

Most �rms owned by self-employed entrepreneurs are very small, as Table 1 attests. In fact,80 percent of the self-employed do not have a single paid worker in their �rm.26 15 percentof small �rms have between one and four paid employees. Only 1 percent of small �rms have10 or more paid employees. Nonwithstanding that most have few or no workers, small �rmsas a group are very important for job creation. About 40 percent of all wage jobs are createdin �rms with 9 employees and less. Together with the self-employment of entrepreneurs, thisimplies that 55 percent of all jobs in Mexico are created in small �rms with at most tenworkers.

Three quarters of all workers in small �rms subsist in informal employment relationships.Conversely, 80 percent of all informal wage jobs are created in �rms with 10 workers or less.This contrasts with export manufacturing employment, which only consists of formal jobs byde�nition; and which almost exclusively is created by large �rms. Table 2 summarises someadditional characteristics of entrepreneurs, workers in export manufacturing, and other wageworkers. Entrepreneurs tend to be older and less educated than wage workers. They are lesslikely to work full time. The commerce sector in particular is dominated by independentbusinesses. There are no big di�erences in gender or migration status across the di�erent

26Of these 80 percent, at least 20 percent have an unpaid worker, often a family member who helps out inthe business.

24

Page 25: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

4 DATA

occupations.

An important assumption in this paper is that workers are attracted to export manufacturingby attractive job opportunities, yet that there is uncertainty about these opportunities.Figure 6 plots the distribution of entrepreneurial pro�ts and wages in export manufacturing.Export manufacturing pay on average higher wages; these should be attractive especiallyfor the sizable group of very low-paid entrepreneurs. Nevertheless, the distributions overlapright to the top, suggesting that there exist �nancially attractive export manufacturing jobsfor almost all entrepreneurs.

4.3 Transitions out of self-employment

There is a high degree of labour churning and employment transitions in the Mexican labourmarket. Table 3 tabulates quarterly transition rates. Of all workers who were found self-employed in a given period, only two thirds remain self-employed when interviewed in thefollowing trimester. 15 percent will have taken up a wage job, and a similar share willhave left employment. Wage-employment is somewhat more stable: 83 percent of the wage-employed stay in the sector from quarter to quarter, and 93 percent of those in formal wageemployment stay (not reported in Table 3).27

In this paper I am mainly interested in transitions out of self-employment. Table 4 providessome correlates of the decision to leave self-employment to either non-employment or to awage job. Entrepreneurs who take up a wage job are predominantly male � as opposed toentrepreneurs who transit to non-employment, three quarters of whom are female. Retailshop owners make up almost half of those leaving work altogether, whereas construction �rmowners are largely overrepresented in the group who leave to wage employment. Finally, thosewho take up wage jobs are younger than other entrepreneurs. Migration status, hours ofwork, and even education seem to play close to no role in the propensity to taking up a wagejob. One should bear in mind, though, that these aggregates hide substantial heterogeneity:those who take up manufacturing jobs, for example, are much more likely to have left schoolwith secondary education.

4.4 How well does the model match the setting?

The section concludes the descriptive analysis by assessing some of the key model assump-tions about the economic environment for small �rms in Mexico against the evidence in thedata.

The descriptive statistics above already shed some lights on these questions. A substantialfraction of entrepreneurs leave their occupation, and taking up a wage job is the mostimportant reason. Wage jobs in the export manufacturing sector appear to be an almostdisjoint subset of the labour market from entrepreneurship: they are formal jobs in large�rms, whereas entrepreneurs run small �rms with a very high degree of labour informality.Export manufacturing jobs pay relatively well, and judging by the income distributions,

27Workers in export manufacturing stay in wage employment at the same rate as formal workers in otherindustries.

25

Page 26: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

4 DATA

there seems to be some �nancially attractive wage job for almost every entrepreneur. Whilethere are some di�erences in demographics across occupations and switching behaviour in away that we would expect, no single variable seems to be a strong predictor.

The correlates of the choice to leave self-employment are in line with what the model predicts.Entrepreneurs who take up a wage job experience an average quarter-on-quarter income gainof 3 percent, compared to stagnating incomes for those who stay. Relative growth in �rmsize in the previous quarter is lower for those who leave self-employment than those whostay.

My model relies on a set of four key assumptions about entrepreneurs, formally statedin Assumption 1. First, the choice between self-employment and wage-employment is adiscrete one - it is not possible for workers to simulatenously hold a full-time wage job andto run a small business. Second, management of small �rms cannot be transfered to anotherindividual. One motivation for this is a missing market for managerial labour. Third,investments into the �rm are speci�c, and cannot be liquidated for some time. Fourth, thereis uncertainty about the arrival of wage o�ers; yet the distributions are known.

I use descriptive data in order to assess whether entrepreneurs in Mexico conform to theseassumptions. Secondary activities are rare for Mexican workers: Only 6 percent of those whoare self-employed in their main occupation, and 5 percent of those who are wage employed,report any secondary activity. To asses the second assumption, I look at evidence whethersmall �rms continue to exist even though the initial owner has changed their main occupationto wage employment. The data allows us to examine two possible ways through which thiscould happen: entrepreneurs might put the �rm into the trusted hands of a family member,or they might be able to continue managing their �rm as a secondary, part-time activity. Inorder to quantify the magnitude of each of these two channels, I look at whether a �rm withinthe same 2-digit NAICS sector continues to exist in the trimester after the entrepreneurhas switched from self-employment to wage-employment, either with a di�erent member ofthe same household, or as a secondary activity of the original entrepreneur. I �nd that10.8 percent of households of previous entrepreneurs continue to operate a business in thesame sector, and that 2.5 percent of entrepreneurs previously self-employed as their primaryactivity continue to operate a business in the same sector as their own secondary activity.Conversely, however, this implies that in 88 percent of all instances, there is no evidencethat a �rm will be continued after the entrepreneur leaves self-employment as their mainactivity.28

To assess the ability of exiting entrepreneurs to liquidate assets, I need to observe occupa-tional histories and microdata on personal �nance of the same individual. To my knowledge,the Mexican Family Life Survey (MxFLS) is the only dataset that ful�ls these criteria. Icompare the di�erences in incomes from major asset sales between individuals who recently

28It is not possible to link the ownership of a particular business over time, so these �gures should beconsidered an estimate. They could constitute an overestimate if household or secondary activity �rms inthe same sector are in fact di�erence enterprises. Or they could be an underestimate if important ways ofcontinuing the �rm are not accounted for in the data, for example installing a manager who does not live inthe same household. However, the �gure I report can be corroborated for the subset of �rms surveyed in theENAMIN, where 87.5 percent of small business owners declare to run �rms that they established themselves.

26

Page 27: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

5 EMPIRICAL RESULTS

left self-employment and entrepreneurs who stayed, controlling for industry and pro�ts ofentrepreneurs.29 The results will be included in a future revision of this paper. Finally, thelarge dispersion in export manufacturing wages documented in Figure 6 is consistent withuncertainty in the job search process. I will revisit the assumption that (changes in) thedistributions governing wage o�ers are known in section 6.3.

5 Empirical Results

In this section I document the main empirical �ndings. I proceed in four steps. First, I showthat my constructed variation in export manufacturing employment is a good predictor ofthe observed variation. Then I document the impact of export manufacturing shocks onthe local labour market. Third, I show how occupational choices of entrepreneurs respondto the creation of jobs in export manufacturing. Finally, I show the e�ect of future exportmanufacturing opportunities on entrepreneurial choices.

5.1 The predictive power of exports for employment

I construct the predicted change in export manufacturing jobs per thousand working-age pop-ulation in each municipality according to equation (12). To remove the short-term cyclicaland seasonal variation in exports that are evident from Figure 5, I apply a moving-averagesmoother with a bandwidth of half a year, and use the smoothed export volumes for pre-diction.30 For comparison, I also construct predicted export manufacturing employmentlevels.

The regression reported in column (5) of Table 5 demonstrates the predictive power of myBartik-style instruments. The change in export manufacturing employment (XME) predictedfrom initial exposure and changes in Mexican-US exports alone explains about a quarter ofthe variation in observed XME changes. The coe�cient of 0.696 is highly signi�cant. Thefollowing columns reveal that this does not change by much when the individual ingredientsof the prediction and state trends are separately controlled for. This con�rms that it is reallythe interaction of export volumes and initial exposure that provides for a strong instrument.The prediction in levels reported in columns (1)-(4) achieves an almost perfect �t, with anR2 of basically unity. This is driven by the fact that predicted XME levels are akin tomunicipality �xed e�ects: some municipalities always have a high number of export jobs,while other municipalities always have zero export jobs. For this reason, I disregard the levelinstruments in my further analysis.

5.2 Export shocks and local labour markets

By construction, export-driven variation in manufacturing employment constitutes a shockat the level of the local labour market. I document market-level adjustments to these shocks

29The MxFLS has no information on business assets or investments.30I tried di�erent methods for smoothing, and moving averages turned out to be most e�ective in removing

seasonal variation, and resulted in the instrument with the highest predictive power.

27

Page 28: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

5 EMPIRICAL RESULTS

by regressing average quantities (transition rates) and prices (incomes and wages) on pre-dicted export manufacturing employment changes, using the speci�cation from equation (13).Usually, we would expect labour demand to move up wages and thus create incentives forentrepreneurs to move to wage employment. But in a labour market with search frictions,the demand elasticity of wages may be small or even zero, at least in the short run. Higherlabour demand might simply manifest itself in a higher o�er probability. This is why I lookat both margins of adjustment. I further document e�ects on wages and employment inother sectors to gauge the potential for spillovers arising from market adjustments. I reportresults on broad outcomes here, and relegate further discussion of general equilibrium e�ectsto section 6.

I start by reporting the e�ect of export manufacturing shocks on municipality-level incomesin Table 6. I am interested in the e�ects on incomes in two sectors: pro�ts for entrepren-eurs, and total wages for employees. I report three speci�cations for these two outcomevariables: a baseline speci�cation with the standard set of instrument controls; a speci�ca-tion with municipality �xed e�ects; and a speci�cation controlling for demographic and �rmcharacteristics of the municipality at baseline (i.e. at t − 1). The latter is my preferredspeci�cation. Municipality controls allow for di�erential trends depending on time-varyingmarket structures. The estimates are small in economic magnitude and mostly insigni�cant:coe�cients from columns (5) and (6) suggest that an additional export manufacturing jobper 1000 workers in the local municipality increases entrepreneurs' pro�ts by 0.05 percentand wages by 0.02 percent.In Table 7, I report the e�ect of export manufacturing shocks on employment transitionrates. I divide the labour force into three mutually exclusive groups: those not in paidemployment (NPE), the wage-employed (WE), and the entrepreneurs/self-employed (SE).As expected, transitions to wage-employment from other sectors increase when an exportemployment shock occurs. The e�ect on transitions out of self-employment is twice as high ason transitions out of non-employment. Conversely, transitions into NPE and SE are reducedby export shocks; but none of the individual e�ects are statistically signi�cant. The nete�ect of all of these transition patterns is that employment shares rise for WE and fall forNPE, by a similar magnitude (see Appendix C). Overall, the �ndings in this section implythat local labour markets adjust employment rather than wages when faced with an exportmanufacturing shock.

5.3 Export shocks and occupational choice

A driving force of the model is the notion that entrepreneurs respond to the incentivesthat wage employment opportunities provide. Descriptively, we have seen signi�cant incomegains for switchers from self-employment to wage-employment. Here I provide more causalevidence how export manufacturing employment opportunities a�ect occupational choices ofentrepreneurs.

In Table 8, I report the e�ect of an export shock in the local municipality on the propensityof entrepreneurs to leave self-employment (to non-employment and/or wage-employment)or to stay self-employed, using the regression model (14). In the model, the only occu-pational choice is between self-employment and wage employment. In the real world , the

28

Page 29: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

5 EMPIRICAL RESULTS

choice between non-employment and employment constitutes another important margin thatI ought to examine empirically. The base speci�cation in columns (1) and (3) includes thestandard controls: trimester �xed e�ects, initial municipality XME exposure, and state �xede�ects. In my preferred speci�cation in columns (2) and (4), I add vectors of controls at theindividual, �rm and locality levels to capture potential di�erential trends. The coe�cientestimate implies that an additional export manufacturing wage job per thousand working-age population in the local municipality � a 0.1 percent increase in jobs per adult� raises theprobability that an individual leaves self-employment by 0.12 percent. There is virtually nodi�erence when we look at leaving to any sector versus leaving only to wage employment,suggesting that the choice to leave to non-employment is orthogonal to export manufacturingemployment shocks.

In a next step, I repeat the exercise from Table 8, but successively shift the export shock tothe past and to the future, respectively. That is, I run separate regressions of model (14),each for a di�erent value of s. How occupational choice reacts to future export shocks tellsus about how individuals respond to changes in expectations; and how choices react to pastexport shocks is informative about the speed of labour market adjustment mechanisms. InFigure 7a, I graphically report the results from these individual regressions. The coe�cientat zero on the horizontal axis corresponds the estimate in column (4) of Table 8. To itsleft are coe�cients for the e�ects of future shocks (s < 0), and to its right for past shocks(s > 0). The asymmetric �gure shows that occupational choice of entrepreneurs does notrespond to future, anticipated export manufacturing employement shocks. Instead, theyleave self-employment in the quarter when or after US demand shocks predict that jobs arecreated in Mexico � the e�ect at the �rst lag is statistically signi�cant at 1 %, and evenslightly larger than the contemporaneous e�ect.

Taken together, the results from this section show that entrepreneurs are more likely to shutdown their businesses at the time when job opportunities in export manufacturing arrive �but not before. This adjustment continues for another quarter or two, but not beyond.

5.4 Export shocks and entrepreneurial choice

Armed with evidence on what export shocks do to local labour markets, and how individualoccupational choices of entrepreneurs react to these shocks, I now proceed to testing forwhether entrepreneurial choices, too, respond to local export manufacturing employment(XME) shocks. In particular, I test for whether the growth of small �rms is a�ected byexport manufacturing shocks.

My main measure of �rm growth is the change in �rm size of small �rms, expressed by thenumber of workers that small �rms employ. I use two variants of this measure. Absolute�rm growth refers to the raw change in number of workers, for all �rms where employment isobserved. This measure includes both the intensive margin of �rm growth, and the extensivemargin for �rms who hire their �rst employee. Furthermore, �rm growth is restricted to bepositive for the majority of �rms that initially employ no workers. Therefore my preferredmeasure, relative �rm growth, expresses growth as the percent change in the number of

29

Page 30: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

6 MECHANISMS AND HETEROGENEITY

workers.31 This variable is only de�ned for �rms that initially employ workers.

In the model from section 2.1, �rm growth is a�ected by future wage employment oppor-tunities. In the regression I therefore shift the export shock by one quarter into the future.Table 9 reports the results for the base and the full speci�cation. The e�ect of future em-ployment opportunities on relative �rm growth is negative and statistically signi�cant. Thee�ect on absolute �rm growth is negative but insigni�cant. My preferred estimate in column(2) implies that an that an additional future export manufacturing wage job per thousandadults in the local municipality reduces growth of small �rms by about 4 percent of thesample mean.

The observation that future export employment opportunities matter for small �rm growthholds more generally when we again examine the e�ects of di�erent lags and leads of theexport shock. I document this pattern in Figure 7b. XME shocks up to three quarters of ayear into the future have a signi�cant and negative e�ect on small �rms job growth. PastXME shocks, however, do not appear to have any e�ect. This is in contrast to the �ndingon occupational choice, which is primarily a�ected by XME shocks of the recent past.

6 Mechanisms and Heterogeneity

These �ndings provide evidence that is consistent with the setup and the predictions of mytheoretical model. In this section I present further evidence which suggests that the e�ectson entrepreneurial choices � reduced growth and investment � arise through the channel ofanticipated occupational choice. First, I document that not only the main e�ects but alsotheir heterogeneity exhibits a pattern that is consistent with the model. In a second step, Idirectly address possible alternative explanations. I show that the results hold for subsampleswhere the alternative explanations are not likely to apply. Third, I provide direct evidenceon the role of expectations with an event study on announcements of large automotive plantexpansions. I relegate other robustness checks to the appendix.

6.1 Testing for heterogeneous e�ects

If the main e�ects I document in Tables 8 and 9 are due the mechanism from the model,then the coe�cients on leaving and on �rm growth should be stronger for those who expectto bene�t more from job opportunities.

From the discussion of descriptive statistics in section 4.1, education emerged as the variablethat most di�erentiates workers in export manufacturing from workers in other occupations.I split the sample according to the level of education workers left school with: low-skilledworkers with only primary schooling, mid-skilled workers with secondary schooling, and high-skilled workers with tertiary education. Table 10 reveals that the main results are drivenby workers with secondary education � the group which is overrepresented in the blue collarjobs that export manufacturing o�ers. Coe�cients are smaller and insigni�cant for othergroups.

31So relative �rm growth is 100 for a change from 1 to 2 workers, 50 for a change from 2 to 3 workers, etc.

30

Page 31: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

6 MECHANISMS AND HETEROGENEITY

I generalise this classi�cation that separates the type of entrepreneur who is likely to becomea wage worker (not just in export manufacturing) from the type who is likely to remainan entrepreneur. I conduct a species analysis similar to De Mel, McKenzie, and Woodru�(2010). To implement this, I construct the predicted probability from a logit model thatregresses a dummy for leaving to wage-employment on a rich set of baseline variables in a�exible speci�cation.32 In Table 11, I split the sample at the median predicted probability.I �nd that both coe�cients are twice as large for the entrepreneurs with a high predictedprobability of leaving. This suggests that type of entrepreneurs who are most susceptibleto wage o�ers are also the ones that respond in a much stronger way to changes in exportemployment opportunities.

In Appendix D, I additionally test for heterogeneous e�ects according to gender, age, migra-tion status, full-time versus part-time self-employment, the size of the locality and the signof the XME change. No clear pattern arises from these distinctions.

6.2 Alternative Explanations

I consider various alternative explanations for my �ndings. These alternative explanationsall operate through the mechanism that local export employment manufacturing shockscould a�ect margins of entrepreneurial choice other than through the expectation of leavingself-employment.

First, local export manufacturing shocks could a�ect the equilibrium of the local economy, forexample through their e�ects on local wages.33 This is a concern because higher labour costs� just like a higher probability of leaving self-employment � reduce the incentives for hiringworkers. In section 5.2, I documented that the e�ect on average wages is small (see Table6). However, this average e�ect may hide some important underlying heterogeneity, and thetype of small �rms that closely compete for workers with export manufacturers could indeedbe a�ected. Table 12 reports the e�ect on wages separately for the manufacturing and non-manufacturing sector, and for small and large �rms.34 The results suggest that an additionallocal export manufacturing job bids up wages in small manufacturing �rms by 0.16 percent;this is signi�cant at the 10 percent level. The e�ect is twice as large for starting wages ofnew workers (see Table A11 in Appendix E). These are precisely the workers that small �rmsneed to compete the hardest for. The coe�cients for wages paid by non-manufacturing �rmsare much smaller and not statistically signi�cant.

A second possibility is that XME shocks lead to local demand spillovers, through supply chainlinkages, goods market competition, or Keynesian multiplier e�ects. Positive spillovers workin the opposite direction to the occupational choice channel: positive demand e�ects increase

32The variables included in the regression are age categories, education levels, marital status, whether theentrepreneur works full time, the number of children, and a detailled classi�cation of business types (e.g.corner shop with family helpers; construction �rm with employees; sole professional working from home); allinteracted with gender.

33This is the mechanism how shocks to the exporting sector transmit to small domestic �rms in theeconomy in the standard trade model with heterogeneous �rms (Melitz (2003)).

34In the appendix I provide an alternative split by formal and informal employment, instead of employmentin large �rms and in small �rms.

31

Page 32: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

6 MECHANISMS AND HETEROGENEITY

the incentives for small �rm growth, whereas employment opportunities decrease them. Yetnegative spillovers work in the same direction as employment opportunities. They couldtherefore constitute a confounding mechanism. In Table 12, we observe a negative (but notstatistically signi�cant) coe�cient on pro�ts of small manufacturing �rms, which is suggestiveof negative demand spillovers.35

This evidence implies that these two alternative mechanisms should especially be of concernfor small manufacturing �rms. In Table 13, I report separate regressions for entrepreneursoperating manufacturing �rms, and those with non-manufacturing �rms. Manufacturingentrepreneurs reduce their �rm size when faced with an XME shock. But they are, if any-thing, less likely to take up a wage job. This suggests that indeed mechanisms other thanoccupational choice are driving the decline in �rm size for manufacturers. The results fornon-manufacturing entrepreneurs � for whom the data show no evidence that labour orproduct market competition are at play � are similar to the main results, and remain highlysigni�cant.

The timing of e�ects supports this interpretation. Figure 8 graphs the timing separately formanufacturing and non-manufacturing �rms. Manufacturing �rms reduce their size in thesame quarter when export jobs arrive, as well as in the preceding quarter. This supportsthe interpretation that labour and product market competition � which appear togetherwith the export jobs � drive the results for entrepreneurs in manufacturing. If the samechannels were responsible for the decrease in �rm size among non-manufacturing �rms, thenwe should see the same timing. Instead, non-manufacturing �rms reduce their �rm size threeto one quarters before the creation of export jobs, and before the entrepreneurs themselveseventually take up opportunities in wage employment.

Finally, a third alternative mechnism originates in the fact that XME shocks increase thevalue of employment in occupations other than self-employment. This implicitly providesan insurance for entrepreneurs against the loss of income when their business venture failsand they are forced to shut it down and leave self-employment.36 This implicit insurancecan in turn encourage entrepreneurs to take greater risks, and we should expect to see anincrease in the probability of leaving self-employment together with stronger growth in thesurviving �rms.37 However, I �nd that leaving self-employment and �rm growth move inopposite directions. This rules out the insurance mechanism.

6.3 The role of expectations

In order to have any e�ect on entrepreneurs' behaviour, future local export manufactur-ing employment opportunities must be anticipated by entrepreneurs. While comprehensive

35The coe�cient on pro�ts for new entrants to self-employment is positive; and additionally large and signi-�cant for non-manufacturing �rm. This is consistent with a story of increased reservation wages (which shiftthe minimum pro�t entrants are willing to accept) and/or positive demand spillovers for non-manufacturing�rms, most of which operate retail and service businesses. Either way, this is not a concern.

36See Gottlieb, Townsend, and Xu (2016) for evidence on the implicit insurance of job-protected leave forwomen giving birth in Canada.

37However, increased entry might also create stronger competition and lead to job destruction in incumbent�rms; see Hombert, Schoar, Sraer, and Thesmar (2014) for evidence from France in the context of a welfarereform aimed at insuring unemployed entrepreneurs if they enter new business ventures.

32

Page 33: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

6 MECHANISMS AND HETEROGENEITY

administrative data can tell us precisely when and where export manufacturing jobs werecreated, they are silent about when information about these jobs was revealed. Fortunately,publicly listed multinational companies centrally reveal information about major investmentprojects, targeted mainly at shareholders. This revelation of information is well-documentedin the �nancial and special interest industry press. I conduct an event study for the universeof Mexican car manufacturing plants, exploiting such central revelation of information onmajor plant expansions. Transport equipment manufacturing (NAICS sub-sector 336) is thelargest among Mexican export manufacturing industries in terms of employment and exportvolume; and the one with the fastest growth rate (see section 4.1).

I construct a database on announcements of large plant expansions in Mexico from the lead-ing special interest publication for the automotive industry, Automotive News. By the endof 2014, Mexico had 18 automotive manufacturing plants. Most of these were �nal assemblyplants for cars and light trucks; a few establishments were dedicated to intermediate pro-cesses such as transmissions production, engine assembly, or stamping. In total, I identify 56public announcements of major plant expansions: introductions of new product lines, capa-city expansions, installation of major �rst tier suppliers, and two green�eld plant openings.Most of these expansions tend to take three quarters to a year, between the �rst announce-ment and the completed expansion. Within each local labour market indexed by z, I assigndummy variables γz,q for each quarter since the announcement in period t, which itself isassigned to event time q = 0.38 I also assign dummies for quarters before the announce-ment. I corrobate these dates with municipality-level time-series of transport equipmentmanufacturing employment constructed from IMSS records.

My basic event study speci�cation at the individual level is the regression model

∆yi,z,t = α +

Qe∑q=Qb

γz,q + x′z,tβ1 + x′

i,z,tβ2 + ∆εi,z,t (15)

where control variables are de�ned as before, and where quarterly event dummies capture anon-parametric trend of arbitrary shape. I additionally estimate a parametric form of theevolution of outcomes following an announcement. That is, I place restrictions of a particularparametric shape on the time dummies, together with an assumption on the average e�ectbefore the announcement.39 I do not place any restriction on the announcement period itself,since announcements can occur at any point during a quarter. For instance, in the case ofrestricting the shape to a third-order polynomial, I estimate equation (15) subject to:

γz,q =

{0 if q < 0

φ0 + φ1 · q + φ2 · q2 + φ3 · q3 if q > 0(16)

38In the case of green�eld assembly plants � which tend to exhibit a lag of around 3 years between the �rstpublic announcement and the start of production � I work backward from the start of production. The �rstquarter generally seems to coincide with the start of major construction works. I allocate four quarters (q = 0through q = 3) exclusively to a single expansion. That is, I allow for subsequent expansion announcementsin q > 3.

39This follows the estimation of trajectories of treatment e�ects of randomised control trials (Abebe, Caria,Fafchamps, Falco, Franklin, and Quinn (2016)).

33

Page 34: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

6 MECHANISMS AND HETEROGENEITY

Most auto plants are located in suburbs of mid-sized metropolitan areas. Many such suburbsconstitute their own municipality. In order to adequately capture the exposure area for majorplant expansions, I de�ne commuting zones (CZ) as the relevant local labour market for thepurpose of this event study.40

I describe the evolution of auto jobs in local labour markets using an adapted version ofmodel (15), without controls. In Figure A1, I report the change in CZ-level auto jobs forthe commuting zones with an auto manufacturing plant. Each coe�cient can be interpretedas the change of auto jobs in quarter q following a major expansion announcement, relativeto other periods in these commuting zones. I also plot a quadratic trend estimate of thepost-announcement change in auto jobs. Expansion announcements do not occur entirelyrandomly. As the �gure shows, announcements are preceeded by an above-average increasein auto jobs in the commuting zone. There are at least two possibilities for this pattern.First, the auto plant starts its expansion already before the public announcement. Second,local suppliers (some of which fall within the same NAICS industry code) are driving thisjob creation. In order to distinguish between these explanations, I additionally report theevolution of auto jobs only for the municipality where the auto plant is located. Thereis no pre-announcement trend within the municipality itself. This suggests that this pre-announcement growth occurs in suppliers located elsewhere in the CZ. It is consistent withanecdotal evidence that local supply chains are a major determinant of plant location choices.

Auto job creation rises strongly following an announcement. It peaks a year after the an-noucement; which is when most expansions would be completed. During the �rst post-announcement year, on average 2,200 additional auto jobs are created in the local commutingzone. This combines jobs created directly in the announcing plant and indirectly in localsuppliers.41 Excess local job creation continues for another two or three quarters beforefalling back to zero.

In order to test for whether this revelation of information on future auto job opportunitieshas an e�ect on the �rm size choices of entrepreneurs, I estimate model (15) for my preferredoutcome variable, relative �rm growth. My comparison group are observations on entrepren-eurs within auto plant CZ that are outside the event study window, and all observations onentrepreneurs not residing in auto plant CZ. I report the regression results in Table 14. Inmy preferred speci�cation in column (2), I include the full set of control variables from earlierspeci�cations. I plot the coe�cients and their 95 percent con�dence interval in Figure 10. Ialso plot the estimated cubic trend.

First, note that, unlike for auto jobs, there is no pre-announcement trend in the growth ofsmall �rms, conditional on controls. Even though expansion announcements are not random,

40I use two pieces of information to aggregate municipalities up to commuting zones. First, I use the 2005de�nition of Zonas Metropolitanas maintained by a working group of Mexican public agencies, and publishedby INEGI. Second, I apply a further aggregation from Atkin (2016) based on cross-municipality commutingpatterns in the 2000 Census. These aggregation steps seem su�cient to properly delineate the urban areaaround an auto plant.

41This �gure is descriptive; it does not constitute an estimate of the employment impact. The overall jobcreation of an expansion is likely larger than the subsector-level employment impact. Casual evidence fromAutomotive News suggests employment multipliers in the order of 5-6 for auto plants. Many suppliers donot share the same NAICS subsector code with auto plants.

34

Page 35: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

6 MECHANISMS AND HETEROGENEITY

this parallel trend provides some con�dence that any subsequent changes in small �rm growthwill be a result of the announcement. Second, small �rm growth dimishes strongly in the�rst and second quarter after the announcement. These short-term e�ects are economicallylarge. The coe�cient at Q2 implies a 50 percent reduction in �rm growth relative to themean. Third, small �rm growth reverts back to normal a year after the announcement. Iteven increases in the post-expansion phase, once the expanded plant is established. This isconsistent with positive local demand spillovers that now come in as the dominant channeldriving small �rm growth.

If plant expansion announcements contain all the relevant information on changes to jobopportunities over the course of the next quarters, then the e�ect of any further variablesexpressing changes to export manufacturing employment should be zero once we conditionon announcements. Furthermore, if the lack of an announcement in my data implies thatother episodes of job growth are unanticipated, then other XME changes in these speci�clocal labour markets should not have any e�ect either. I test these hypotheses in column(4) of Table 14. I augment the event study speci�cation with predicted future XME changesand their interactions with auto plant CZ and annoucement indicators. I cannot reject thehypotheses that (i) predicted future XME changes have no e�ect in auto plant commutingzones (p = 0.67) and (ii) predicted XME changes have no e�ect in the immediate aftermathof an announcement (p = 0.31). However, predicted future XME changes in municipalitiesthat are not included in the event study continue to have a negative e�ect on small �rmgrowth. The estimated coe�cient is very similar to the main e�ect reported in column (2)of Table 9.

Taken together, this provides strong suggestive evidence that it is really the informationabout employment opportunities in the immediate future which links export manufacturingjob creation to reduced growth in entrepreneur-run small �rms. This holds true in the caseof the auto industry, the largest export manufacturing sector. More speculatively, we cancompare the graphical pattern in Figure 10 from this case study to the more general patternof timing from Figure 7b. Relative to the peak of auto job creation in Q3, we see �rm growthreduced in the two or three quarters that preceed this peak. This mirrors the general patternthat small �rm growth reduces in the three quarters prior to an XME shock.

6.4 Robustness

I relegate other robustness checks to Appendix F. There, I report OLS and 2SLS results.The OLS bias seems to work in the opposite direction of my model predictions: a downward(or attenuation) bias for leaving, and an upward bias for �rm growth. I re-run the mainregressions at the commuting zone level. The results hold overall � some coe�cients becomestronger, some weaker. Finally, I check whether local production shocks in industries witha strong agglomeration of production a�ect the results. If one or few municipalities cana�ect aggregate industry exports, then exports are no longer driven purely by US demand,an external factor. This may bias results. To implement this check, I in turn leave out eachconstituent export manufacturing industry. Results suggest that, if anything, the magnitudesare larger when I leave out highly concentrated industries. This suggests that any such biasworks again in opposite direction to my model predictions.

35

Page 36: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

7 CONCLUSION

7 Conclusion

Small-scale entrepreneurship creates self-employment opportunities for a large fraction ofworkers in many developing countries. As a group, these entrepreneurs employ a sizeablefraction of workers who hold a wage job. Yet individually, most of their �rms are verysmall, seldom employ any worker, and can rarely be found on transformative paths towardsdeveloping into a larger enterprise. This is despite vast support programmes targeted atalleviating a numbers of constraints known to policymakers and researchers. Recent changesin labour demand and in information technology have made these issues salient again evenin many advanced economies.

Our traditional understanding of small-scale entrepreneurship draws a stark distinctionbetween the `transformative' entrepreneur as a �rm, and the `subsistence' entrepreneur asa self-employed worker. In this paper, I unite these two prespectives. Occupational andentrepreneurial choices of entrepreneurs are not separable in my framework. There is atension between investment into future pro�table business opportunities in the �rm, andinvestment into search for a future wage job. By providing a simple but general model, Iwork out the assumptions that are crucial for this tension to arise: discrete occupationalchoice, a missing market for managerial labour, �rm-speci�c investments, and uncertaintyabout wage employment opportunities at the investment stage.

I provide an empirical test of this mechanism for small �rms in Mexico over the 2005-2014 decade. I exploit shocks to employment in the Mexican export manufacturing sectorbrought by the bust-and-boom of exports to the US during the �nancial crisis of 2008/09 andthe subsequent recovery. I document how these shocks transmit into local labour marketadjustments. They increase the transitions of entrepreneurs out of self-employment intowage employment, and they induce entrepreneurs to reduce their �rm size in anticipation. Iuse a species analysis, a case study of large plant expansion announcements in the Mexicanauto industry, and di�erential timing of e�ects in order to distinguish these �ndings fromcompeting mechanisms.

By construction, changes in wage employment opportunities � available to those who arenot yet wage-employed � constitute market-level shocks. My empirical analysis distinguishesthe occupational choice mechanism from other market adjustments that work in same dir-ection, and therefore could confound the mechanism. This is su�cient for establishing theexistence of the mechanism. However, determining the magnitude, and assessing the welfareconsequences of non-separability, require knowledge of the counterfactual: what �rm sizewould entrepreneurs choose under separability of occupational and entrepreneurial choices?Such a counterfactual would move local economies to a completely di�erent equilibrium, ifit applies to all entrepreneurs simultaneously. More structure would be needed to establishthe full counterfactual, and to answer these questions.

My empirical and theoretical results point to frequent transitions out of self-employed entre-preneurship as a novel determinant of �rm size that works in addition to many well-researchedmechanism; at least in the lower part of the size distribution. Jobs in the export sector are onesource for incentives to leave self-employment. This points at a tension between job creationin large, `formal' establishments and in small informal businesses. The �ipside, however, is

36

Page 37: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

REFERENCES

the implication that small �rms can act as a sort of automatic employment stabiliser againstnegative shocks to large, exporting �rms. The 2016 US presidential election, and the newpotential shock to Mexican exports it constitutes, may provide a setting for future empiricalresearch on this implication. It would also be useful to determine to which degree this modelcan explain heterogenous �ndings in the recent experimental microenterprise literature.

References

Abebe, G., S. Caria, M. Fafchamps, P. Falco, S. Franklin, and S. Quinn (2016):�Curse of Anonymity or Tyranny of Distance? The Impacts of Job-Search Support inUrban Ethiopia,� Working Paper 22409, National Bureau of Economic Research.

Acemoglu, D., D. Autor, D. Dorn, G. H. Hanson, and B. Price (2016): �ImportCompetition and the Great US Employment Sag of the 2000s,� Journal of Labor Econom-ics, 34(S1 Part 2), S141�S198.

Adam, S., H. Miller, and T. Pope (2017): �Tax, legal form and the gig economy,� inThe IFS Green Budget: February 2017, ed. by C. Emmerson, P. Johnson, and R. Joyce,pp. 203�238. Institute for Fiscal Studies.

Adão, R. (2015): �Worker Heterogeneity, Wage Inequality, and International Trade: Theoryand Evidence from Brazil,� Discussion paper, mimeo.

Aghion, P., and P. Bolton (1997): �A theory of trickle-down growth and development,�The Review of Economic Studies, 64(2), 151�172.

Angelucci, M., D. Karlan, and J. Zinman (2015): �Microcredit Impacts: Evid-ence from a Randomized Microcredit Program Placement Experiment by CompartamosBanco,� American Economic Journal: Applied Economics, 7(1), 151�82.

Atkin, D. (2016): �Endogenous Skill Acquisition and Export Manufacturing in Mexico,�American Economic Review, 106(8), 2046�85.

Autor, D., D. Dorn, and G. H. Hanson (2013): �The China syndrome: Local labormarket e�ects of import competition in the United States,� The American EconomicReview, 103(6), 2121�2168.

Banerjee, A., E. Breza, E. Duflo, and C. Kinnan (2015): �Do credit constraints limitentrepreneurship? Heterogeneity in the returns to micro�nance,� Mimeo.

Banerjee, A., E. Duflo, R. Glennerster, and C. Kinnan (2015): �The Miracle ofMicro�nance? Evidence from a Randomized Evaluation,� American Economic Journal:Applied Economics, 7(1), 22�53.

Banerjee, A., D. Karlan, and J. Zinman (2015): �Six Randomized Evaluations ofMicrocredit: Introduction and Further Steps,� American Economic Journal: Applied cE-conomics, 7(1), 1�21.

37

Page 38: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

REFERENCES

Banerjee, A., and S. Mullainathan (2010): �The shape of temptation: Implications forthe economic lives of the poor,� Discussion paper, National Bureau of Economic ResearchWorking Paper 15973.

Banerjee, A. V., and A. F. Newman (1993): �Occupational choice and the process ofdevelopment,� Journal of political economy, pp. 274�298.

Bartelsman, E., J. Haltiwanger, and S. Scarpetta (2013): �Cross-country di�er-ences in productivity: The role of allocation and selection,� The American EconomicReview, 103(1), 305�334.

Benjamin, D. (1992): �Household composition, labor markets, and labor demand: testingfor separation in agricultural household models,� Econometrica: Journal of the Economet-ric Society, pp. 287�322.

Blanchflower, D., and A. Oswald (1998): �What Makes an Entrepreneur?,� Journalof Labor Economics, 16(1), 26�60.

Blattman, C., and S. Dercon (2016): �Occupational Choice in Early IndustrializingSocieties: Experimental Evidence on the Income and Health E�ects of Industrial andEntrepreneurial Work,� Discussion paper, National Bureau of Economic Research WorkingPaper 22683.

Blattman, C., N. Fiala, and S. Martinez (2014): �Generating Skilled Self-Employmentin Developing Countries: Experimental Evidence from Uganda,� Quarterly Journal ofEconomics, 129(2).

Blundell, R., M. Costa Dias, C. Meghir, and J. Shaw (2016): �Female labor supply,human capital, and welfare reform,� Econometrica, 84(5), 1705�1753.

Bosch, M., and W. Maloney (2010): �Comparative analysis of labor market dynamicsusing markov processes: An application to informality,� Labour Economics, 17(4), 621�631.

Cohen, A. (2016): �Constraints on Labor and Land and the Return to Microcredit andMicroinsurance,� Mimeo.

Costa, F., J. Garred, and J. P. Pessoa (2016): �Winners and losers from a commodities-for-manufactures trade boom,� Journal of International Economics.

Dauth, W., S. Findeisen, and J. Suedekum (2014): �The rise of the East and the FarEast: German labor markets and trade integration,� Journal of the European EconomicAssociation, 12(6), 1643�1675.

David, J. M., H. A. Hopenhayn, and V. Venkateswaran (2016): �Information, misal-location and aggregate productivity,� The Quarterly Journal of Economics, pp. 943�1005.

De Janvry, A., M. Fafchamps, and E. Sadoulet (1991): �Peasant household behaviourwith missing markets: some paradoxes explained,� The Economic Journal, 101(409), 1400�1417.

38

Page 39: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

REFERENCES

De Mel, S., D. McKenzie, and C. Woodruff (2008): �Returns to capital in microen-terprises: Evidence from a �eld experiment,� The Quarterly Journal of Economics, 123(4),1329�1372.

(2010): �Who are the microenterprise owners? Evidence from Sri Lanka on Tok-man v. de Soto,� in International Di�erences in Entrepreneurship, ed. by J. Lerner, andA. Schoar, pp. 63�87. University of Chicago Press.

de Mel, S., D. McKenzie, and C. Woodruff (2016): �Labor Drops: ExperimentalEvidence on the Return to Additional Labor in Microenterprises,� World bank policyresearch working paper no. 7924.

Dix-Carneiro, R. (2014): �Trade liberalization and labor market dynamics,� Economet-rica, 82(3), 825�885.

Eckstein, Z., and K. I. Wolpin (1989): �Dynamic Labour Force Participation of MarriedWomen and Endogenous Work Experience,� The Review of Economic Studies, 56(3), 375�390.

Evans, D., and B. Jovanovic (1989): �An Estimated Model of Entrepreneurial Choiceunder Liquidity Constraints,� Journal of Political Economy, 97(4), 808�827.

Evans, D., and L. Leighton (1989): �Some Empirical Aspects of Entrepreneurship,�American Economic Review, 79(3), 519�535.

Evans, D. S. (1987): �Tests of alternative theories of �rm growth,� Journal of PoliticalEconomy, 95(4), 657�674.

Fafchamps, M., D. McKenzie, S. Quinn, and C. Woodruff (2014): �MicroenterpriseGrowth and the Flypaper E�ect: Evidence from a Randomized Experiment in Ghana,�Journal of Development Economics, 116, 211�226.

Fafchamps, M., and S. Quinn (2016): �Aspire,� Journal of Development Studies, forth-coming.

Fama, E. F., and M. C. Jensen (1983): �Separation of ownership and control,� Thejournal of Law and Economics, 26(2), 301�325.

Fields, G. (1975): �Rural-urban Migration, Urban Unemployment and Underemployment,and Job-Search Activity in LDCs,� Journal of Development Economics, 2(2), 165�187.

Gottlieb, J. D., R. R. Townsend, and T. Xu (2016): �Experimenting with entrepren-eurship: the e�ect of job-protected leave,� .

Günther, I., and A. Launov (2012): �Informal Employment in Developing Countries:Opportunity or Last Resort?,� Journal of Development Economics, 97(1), 88�98.

Hardy, M., and J. McCasland (2015): �Are small �rms labor constrained? Experimentalevidence from Ghana,� Mimeo.

39

Page 40: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

REFERENCES

Harris, J., and M. Todaro (1970): �Migration, Unemployment and Development: ATwo-Sector Analysis,� American Economic Review, 60(1), 126�142.

Haywood, L., and P. Falco (2016): �Entrepreneurship vs Joblessness: Explaining theRise in Self-Employment,� Journal of Development Economics, 118, 245�265.

Heath, R., and A. M. Mobarak (2015): �Manufacturing growth and the lives ofBangladeshi women,� Journal of Development Economics, 115, 1�15.

Hombert, J., A. Schoar, D. Sraer, and D. Thesmar (2014): �Can Unemployment In-surance Spur Entrepreneurial Activity?,� Discussion paper, National Bureau of EconomicResearch Working Paper 20717.

Hopenhayn, H. (1992): �Entry, Exit, and Firm Dynamics in Long Run Equilibrium,�Econometrica, 60(5), 1127�1150.

Hsieh, C.-T., and P. J. Klenow (2009): �Misallocation and Manufacturing TFP in Chinaand India,� The Quarterly Journal of Economics, 124(4), 1403�1448.

Humphries, J. E. (2017): �The Causes and Consequences of Self-Employment over the LifeCycle,� Job Market Paper, University of Chicago.

Jovanovic, B. (1982): �Selection and the Evolution of Industry,� Econometrica, 50(3),649�670.

Karlan, D., R. Knight, and C. Udry (2015): �Consulting and Capital Experiments withMicroenterprise Tailors in Ghana,� Journal of Economic Behaviour and Organization, 118,281�302.

Karlan, D., and M. Valdivia (2011): �Teaching Entrepreneurship: Impact of BusinessTraining on Micro�nance Clients and Institutions,� Review of Economics and Statistics,93(2), 510�527.

Katz, L. F., and A. B. Krueger (2016): �The Rise and Nature of Alternative WorkArrangements in the United States, 1995-2015,� Working Paper 22667, National Bureauof Economic Research.

Kihlstrom, R., and J.-J. Laffont (1979): �A General Equilibrium Entrepreneurial The-ory of Firm Formation Based on Risk Aversion,� Journal of Political Economy, 87(4),719�748.

Killingsworth, M. R., and J. J. Heckman (1986): �Female labor supply: A survey,�Handbook of labor economics, 1, 103�204.

Koelle, M., and S. Quinn (2016): �Self-Employment, Capital and Job Search in UrbanGhana: Evidence from a Structural Model,� Discussion paper, mimeo.

Kraay, A., and D. McKenzie (2014): �Do Poverty Traps Exist? Assessing the Evidence,�Journal of Economic Perspectives, 28(3), 127�148.

40

Page 41: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

REFERENCES

LaFave, D., and D. Thomas (2016): �Farms, Families, and Markets: New Evidence onCompleteness of Markets in Agricultural Settings,� Econometrica, 84(5), 1917�1960.

Levine, R., and Y. Rubinstein (2017): �Smart and Illicit: Who Becomes an Entrepreneurand Do They Earn More?,� The Quarterly Journal of Economics, 132(2), 963.

Lewis, W. A. (1954): �Economic Development with Unlimited Supplies of Labour,� TheManchester School, 22(2), 139�191.

Lopez-Garcia, I. (2013): �Human Capital and Labor Informality in Developing Countries:A Life-Cycle Approach,� Job Market Paper.

Lucas, R. (1978): �On the Size Distribution of Business Firms,� The Bell Journal of Eco-nomics, 9(2), 508�523.

Magnac, T. (1991): �Segmented or Competitive Labor Markets,� Econometrica, 59(1),165�187.

McCaig, B., and N. Pavcnik (2014): �Export markets and labor allocation in a low-incomecountry,� Discussion paper, National Bureau of Economic Research Working Paper 20455.

McCall, J. (1970): �Economics of Information and Job Search,� The Quarterly Journal ofEconomics, 84, 113�126.

McKenzie, D. (2015): �Identifying and Spurring High-Growth Entrepreneurship: Experi-mental Evidence from a Business Plan Competition,� .

(2017): �Identifying and Spurring High-Growth Entrepreneurship: ExperimentalEvidence from a Business Plan Competition,� American Economic Review, forthcoming.

McKenzie, D., and C. Woodruff (2008): �Experimental evidence on returns to capitaland access to �nance in Mexico,� The World Bank Economic Review, 22(3), 457�482.

(2014): �What Are We Learning from Business Training and EntrepreneurshipEvaluations around the Developing World?,� The World Bank Research Observer, 29(1),48�82.

Melitz, M. (2003): �The Impact of Trade on Intra-Industry Reallocations and AggregateIndustry Productivity,� Econometrica, 71, 1695�1725.

Mortensen, D. (1970): �Job search, the Duration of Unemployment, and the PhillipsCurve,� American Economic Review, 60(5), 847�862.

Piketty, T. (1997): �The dynamics of the wealth distribution and the interest rate withcredit rationing,� The Review of Economic Studies, 64(2), 173�189.

Polkovnichenko, V. (2003): �Human Capital and the Private Equity Premium,� Reviewof Economic Dynamics, 6(4), 831�845.

Ramsey, F. (1928): �A Mathematical Theory of Saving,� The Economic Journal, 38, 543�559.

41

Page 42: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

REFERENCES

Söderbom, M., and F. Teal (2004): �Size and e�ciency in African manufacturing �rms:evidence from �rm-level panel data,� Journal of Development Economics, 73(1), 369�394.

Stigler, G. (1962): �Information in the Labor Market,� Journal of Political Economy, 70,94�105.

Topalova, P. (2010): �Factor Immobility and Regional Impacts of Trade Liberalization:Evidence on Poverty from India,� American Economic Journal: Applied Economics, 2(4),1�41.

Vereshchagina, G., and H. A. Hopenhayn (2009): �Risk taking by entrepreneurs,� TheAmerican Economic Review, 99(5), 1808�1830.

Verhoogen, E. A. (2008): �Trade, Quality Upgrading, andWage Inequality in the MexicanManufacturing Sector,� The Quarterly Journal of Economics, 123(2), 489�530.

42

Page 43: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

Tables and Figures

FIGURE 1

International comparison of non-rural non-agricultural self-employment

0 .2 .4 .6 .8 1SE rate as % of urban non-agricultural employment

Mali

Burkina Faso

Senegal

Peru

Vietnam

Colombia

Mexico

United Kingdom

Germany

United States

Notes: Own calculation. Data restricted to capital cities or metropolitan areas with at least 1M population for thefollowing sources: 2007 ENOE (Mexico), 2007 ESLF (Colombia) and WIEGO (2009) uni�ed dataset compiled from:2003 1-2-3 survey (Mali, Burkina Faso, Senegal); 2006 ENAHO (Peru), 2007 LFS (Vietnam), and 2007 LFS tabulations(UK). Data restricted to non-agricultural self-employment for 2004 Microcensus tabulations (Germany) and 2007 CPStabulations (US).

43

Page 44: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

FIGURE 2

Multiple solutions, local and global comparative statics

k

M

argi

nal U

tility

A B C

0

EUk baseline

EUk alternative

(a) Marginal utility and the FOC

k

Exp

ecte

d U

tility

A B C

EU baselineEU alternative

(b) Expected utility, local and global optima

FIGURE 3

Intuition: Effective discount factor as a transmission channel

log wage

Den

sity

zA

zC

(a) Wage o�er distribution and reservation wages

k

Rat

e of

Ret

urn

CR

A

C

:0

r//r

(b) Rates of return representation

44

Page 45: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

FIGURE 4

Heterogeneous entreprenurial ability

A. Marginal products and discounted returns

k

Rat

e of

Ret

urn

A10

A1

MPK :0

ROR r / /New ROR r / /

(a) Lower ability

k

Rat

e of

Ret

urn

A0

C

MPK :0

ROR r / /New RoR r / /

(b) Baseline ability

k

Rat

e of

Ret

urn

C2

C20

MPK :0

ROR r / /New ROR r / /

(c) Higher ability

B. Expected utility and global comparative statics

k

Exp

ecte

d U

tility

A1

A10

EUEU

1

(d) Lower ability

k

Exp

ecte

d U

tility A0

C

EUEU

1

(e) Baseline ability

k

Exp

ecte

d U

tility

C2

C20

EUEU

1

(f) Higher ability

C. The heterogeneous e�ect of microcredit

k

Rat

e of

Ret

urn A

1

" K

MPK :0

ROR r//

(g) Lower ability

k

Rat

e of

Ret

urn

CB

A

" K

MPK :0

ROR r//

(h) Baseline scenario

k

Rat

e of

Ret

urn

C2

A2

" K

MPK :0

ROR r//

(i) Higher ability

45

Page 46: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

FIGURE 5

Employment and exports in Mexican export manufacturing

7080

9010

011

012

013

014

015

016

017

0In

dex

(200

5 =

100)

05M1 06M1 07M1 08M1 09M1 10M1 11M1 12M1 13M1 14M1Month

Employment in export manufacturingMEX-US exports [real USD]

Source: IMSS end-of-month social security records on formal employment; US Census Bureau monthly volume of importsfrom Mexico in nominal USD, de�ated with US CPI. Each time series normalised to their respective average annual levelin 2005.

FIGURE 6

The distribution of entrepreneurial profits and export manufacturing wages

0.2

.4.6

.81

Den

sity

1000 3000 9000 20000 60000Monthly income (2011 MXN, log scale)

Export manufacturing Self employment

Source: 2005-2014 ENOE. The graph shows the distribution of monthly income in self-employment/entrepreneurship andin formal wage work in the export manufacturing sector. Nominal incomes are de�ated to 2011 Mexican Pesos using theconsumer price index (INPC) provided by the Mexican Central Bank.

46

Page 47: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

FIGURE 7

Timing of entrepreneur's responses to export shocks

-.2-.1

0.1

.2E

ffect

siz

e: le

ave

SE

to W

E

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(a) Occupational choice: leave to WE

-2-1

.5-1

-.50

.51

1.5

2E

ffect

siz

e: re

lativ

e fir

m g

row

th

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(b) Entrepreneurial choice: relative �rm growth

Source: Results from regression of model (14), using the speci�cation reported in Column (4) of Table 8 (left) andTable 9 (right). Horizontal axis documents the number of quarters s that the dependent variable has been shifted. Eachcoe�cient corresponds to a separate regression. Reported 95 % con�dence intervals use two-way clustered standard errorsby municipality and quarter.

47

Page 48: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

FIGURE 8

Timing for manufacturing and non-manufacturing firms

Manufacturing �rms

-.4-.2

0.2

.4E

ffect

siz

e: le

ave

SE

to W

E

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(a) Occupational choice: leave to WE

-3-2

-10

12

3E

ffect

siz

e: re

lativ

e fir

m g

row

th

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(b) Entrepreneurial choice: relative �rm growth

Non-manufacturing �rms

-.4-.2

0.2

.4E

ffect

siz

e: le

ave

SE

to W

E

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(c) Occupational choice: leave to WE

-2-1

.5-1

-.50

.51

1.5

2E

ffect

siz

e: re

lativ

e fir

m g

row

th

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(d) Entrepreneurial choice: relative �rm growth

Source: Results from regression of model 14, using the speci�cation reported in Column (4) of Table 8 (left) and Table9 (right). Horizontal axis documents the number of quarters s that the dependent variable has been shifted. Eachcoe�cient corresponds to a separate regression. Reported 95 % con�dence intervals use two-way clustered standarderrors by municipality and quarter.

48

Page 49: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

FIGURE 9

Automotive case study: large plant expansions and job creation

-500

050

010

00C

hang

e in

aut

o jo

bs

-4 -3 -2 -1 0 1 2 3 4 5 6 7 8Quarter

Plant municipality Commuting Zone

Source: The graph shows the coe�cients from a regression of labour-market level changes in auto jobs on quarterlydummies indicating the time from the announcement of a major plant expansion. Con�dence intervals at 95 percent.The quadratic trend corresponds to a constrained CZ-level regression.

FIGURE 10

Automotive case study: large plant expansions and small firm growth

-20

-10

010

2030

Coe

ffici

ent o

n re

lativ

e fir

m g

row

th

-4 -3 -2 -1 0 1 2 3 4 5 6 7 8Quarter

Source: The dots plot the coe�cients from model (2) of Table 14 together with their 95 percent con�dence intervals. Theline shows a cubic trend, imposing the additional restriction (16) on the regression model, with 95 percent con�denceintervals. The hollow circle shows the coe�cient on the announcement period from the constrained model.

49

Page 50: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 1

Employment by firm size in Mexico

Firm Size (incl.entrepreneur)

# Entre-preneurs

# Em-ployees

%�rms

% wage em-ployment

1 8.97M 1.16M 82.5 4.12-5 1.62M 8.47M 14.9 30.16-10 0.16M 2.84M 1.5 10.1

11+ 0.12M 15.71M 1.1 55.7

Total 10.88M 28.19M 100.0 100.0

Source: ENOE, 2005-2014. Figures are obtained using sampling weights.Reported are averages across all 10 years. Employees in �rms of size 1 aredomestic workers.

TABLE 2

Individual and job characteristics by occupation

Entrepreneur XM worker Other worker

Employment share 0.25 0.06 0.69

Individual characteristicsMale 0.57 0.63 0.61Inter-state migrant 0.25 0.28 0.25

EducationPrimary 0.41 0.19 0.24Secondary 0.45 0.67 0.54Tertiary 0.14 0.14 0.23

Age15-29 0.15 0.43 0.3830-39 0.26 0.30 0.2740-49 0.29 0.19 0.2150-64 0.28 0.08 0.13

Job characteristicsFormal (IMSS-a�liated) 0.00 1.00 0.54Full time (min 35 hrs/week) 0.65 0.96 0.81

IndustryManufacturing 0.15 1.00 0.15Construction & Transport 0.15 0.00 0.17Commerce 0.37 0.00 0.17Services 0.32 0.00 0.50

Firm size (# workers)1-5 0.98 0.02 0.356-10 0.02 0.03 0.1111-50 0.00 0.13 0.2351-250 0.00 0.25 0.14> 250 0.00 0.58 0.17

Observations 1,272,771 285,370 3,825,407

Source: ENOE, 2005-2014. Figures are obtained using sampling weights. Entrepreneursare employers and self-employed; export manufacturing (XM) workers are formal wageemployees in the export manufacturing industries; other workers are all other employees.

50

Page 51: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 3

Transition rates in the Mexican labour market

Employment status next quarter

Employment status OLF WE SE OE UE Total

Out of LF % 69.8 8.0 5.4 3.2 13.6 100.0Wage-employed % 5.7 82.8 5.7 1.2 4.5 100.0Self-employed % 10.5 15.4 67.5 2.7 3.9 100.0Other Employment % 24.4 14.2 11.0 43.4 6.9 100.0Unemployed % 43.9 20.8 6.7 2.9 25.7 100.0

Total % 30.2 41.2 15.5 3.8 9.2 100.0

Source: ENOE, 2004-2014. Quarterly transitions for all respondents aged 15-64 with non-missing employment status inadjacent periods. OLF = Out of labour force; SE = self-employed; WE = wage employed; OE = unpaid employment;UE = unemployed.

TABLE 4

Correlates of the decision to leave entrepreneurship

Stay entrepreneur To non-employment To wage employment

Individual characteristicsMale 0.60 0.28 0.75Inter-state migrant 0.25 0.23 0.25

EducationPrimary 0.40 0.46 0.36Secondary 0.45 0.45 0.49Tertiary 0.15 0.10 0.15

Age15-29 0.11 0.20 0.2230-39 0.26 0.24 0.3140-49 0.31 0.24 0.2750-64 0.30 0.31 0.20Full time (min 35 hrs/week) 0.70 0.43 0.69

IndustryManufacturing 0.15 0.17 0.11Construction & Transport 0.13 0.08 0.30Commerce 0.38 0.44 0.24Services 0.33 0.29 0.32

Observations 613,736 170,567 132,782

Source: ENOE, 2005-2014. Figures are obtained using sampling weights.

51

Page 52: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 5

Goodness of fit of predicted export manufacturing employment

(1) (2) (3) (4) (5) (6) (7)Levels Levels Levels Levels FD FD FD

Pred. XME/WAPop 0.869*** 0.875*** 0.642*** 0.670***(0.030) (0.030) (0.063) (0.057)[29.247] [29.183] [10.131] [11.728]

∆ Pred. XME/WAPop 0.696*** 0.723*** 0.761***(0.092) (0.105) (0.116)[7.547] [6.913] [6.577]

Constant 2.467*** 17.043*** 2.144*** 16.398*** -0.036 0.290 0.492***(0.671) (3.347) (0.593) (3.243) (0.070) (0.193) (0.179)[3.675] [5.092] [3.615] [5.057] [-0.510] [1.505] [2.752]

Initial XME No No Yes Yes No No YesTrimester FE No Yes No Yes No Yes YesState FE No Yes No Yes No Yes Yes

Observations 97520 97520 97520 97520 95082 95082 95082R2 0.930 0.938 0.934 0.941 0.235 0.265 0.269

This table reports regression of predicted on actual export manufacturing employment for all 97,520 unique local labourmarket constellations in the cross-section / 95,082 in the �rst di�erence (2,438 non-D.F. municipalities over 40/39 tri-mesters). Municipal employment variables are normalised per 1000 working-age population of the corresponding trimesterand municipality, and winsorised at the top and bottom 0.1%. Two-way clustered standard errors by municipality andquarter in parentheses, t-statistics in square brackets.∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

52

Page 53: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 6

Effect of export shock on wages

Dependent Variable: 100×∆ log(average income)

(1) (2) (3) (4) (5) (6)Type of income: Pro�ts Wages Pro�ts Wages Pro�ts Wages

∆ Pred. XME/WAPop 0.0446 0.0429 0.0593 0.0440 0.0675 0.0207(0.0678) (0.0273) (0.0721) (0.0288) (0.0749) (0.0269)

Standard controls Yes Yes Yes Yes Yes Yes

Individual controls No No No No Yes Yes

Firm controls No No No No Yes Yes

Locality controls No No No No Yes Yes

Municipality �xed e�ects No No Yes Yes No No

Dep. var. mean -0.66 -0.18 -0.66 -0.18 -0.66 -0.18

R2 0.01 0.01 0.01 0.01 0.02 0.03# municipalities 1,482 1,491 1,482 1,491 1,481 1,483Observations 37,240 38,321 37,240 38,321 37,002 37,543

Standard errors in parenthesesThis table reports regression of the log average income di�erence on predicted quarterly export manufacturingemployment di�erence. Unit of observation is an unbalanced panel of Mexican municipalities (apart fromD.F.) over 40 quarters from 2005 to 2010. All employment variables are normalised per initial 1000 working-age-population of the corresponding trimester and municipality, and employment and outcome variables arewinsorised at the top and bottom 0.05%. All speci�cations include a constant, and per standard controlfor state, trimester, 2005 export employment. Individual controls are municipal average levels of education,experience, and share of men. Firm controls are municipal baseline industry shares and �rm size distributions.Locality controls are locality size distribution within municipality. Observations are weighted by working-agepopulation. Standard errors clustered by municipality.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

53

Page 54: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 7

Effect of export shock on transition rates

(1) (2) (3)To NPE To WE To SE

From NPE

∆ Pred. XME/WAP -0.0526∗∗ 0.0540∗∗∗ -0.000603(0.0263) (0.0199) (0.0173)

Dep. var. mean 83.38 10.92 5.70Observations 27,021 27,021 27,021

From WE

∆ Pred. XME/WAP -0.0227 0.00174 0.0212(0.0280) (0.0407) (0.0246)

Dep. var. mean 11.48 83.67 4.85Observations 26,570 26,584 26,584

From SE

∆ Pred. XME/WAP 0.00569 0.103∗∗ -0.108(0.0681) (0.0403) (0.0663)

Dep. var. mean 19.14 14.61 66.33Observations 26,736 26,752 26,752

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Standard errors in parentheses,clustered by municipality.Note: This table reports regression of transitions (in percentage points, row-wise) on predicted quarterly export manufacturing employment di�erence.Sample is an unbalanced panel of Mexican municipalities (apart from D.F.)over 40 quarters from 2005 to 2010. Export employment is normalised perinitial 1000 working-age-population of the corresponding trimester and mu-nicipality, and employment and outcome variables are winsorised at the topand bottom 0.05%. All speci�cations include a constant, and per standardcontrol for state, trimester, 2005 export employment. Individual controls aremunicipal average levels of education, experience, and share of men. Firmcontrols are municipal baseline industry shares and �rm size distributions.Locality controls are locality size distribution within municipality. Observa-tions are weighted by working-age population.

54

Page 55: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 8

Effect of export shock on propensity to stay self-employed

(1) (2) (3) (4)Dependent Variable: Leave Leave Leave to WE Leave to WE

∆ Pred. XME/WAP 0.0846∗∗∗ 0.105∗ 0.0931∗∗∗ 0.108∗∗∗

(0.0258) (0.0580) (0.0340) (0.0359)

Municipality controls Yes Yes Yes Yes

Individual controls No Yes No Yes

Firm controls No Yes No Yes

Locality controls No Yes No Yes

Dep. var. mean 33.46 33.57 17.68 17.41

R2 0.01 0.07 0.01 0.08# distinct LLM 25,875 25,522 25,460 24,985# municipalities 1,491 1,490 1,479 1,477# periods 39 39 39 39Observations 918,160 727,539 747,411 590,176

Standard errors in parenthesesThis table reports regression of probability of leaving self-employment onpredicted quarterly export manufacturing employment di�erence in the localmunicipality. Unit of observation are entrepreneurs with at most 9 employeesfrom 40 quarterly ENOE waves 2005-2014 who are observed in the followingperiod. Municipal export employment variables are normalised per 1000working-age-population of the corresponding month and municipality, andemployment and outcome variables are winsorised at the top and bottom0.01%. Two-way clustered standard errors by municipality and trimester.Municipality controls include wave dummies, state dummies, and 2005 exportmanufacturing employment. Individual controls include gender, educationlevel, and experience quadratic. Firm control include �rm size, industry,and pro�t quintiles. Locality controls include locality size category. Allspeci�cations include a constant, and ENOE sampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

55

Page 56: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 9

Effect of export shock on small firm growth

(1) (2) (3) (4)

Dependent Variable:Relative

Firm GrowthRelative

Firm GrowthAbsolute

Firm GrowthAbsolute

Firm Growth

F.∆ Pred. XME/WAP -0.618∗∗ -0.990∗∗∗ -0.170 -0.225(0.304) (0.221) (0.192) (0.199)

Municipality controls Yes Yes Yes Yes

Individual controls No Yes No Yes

Firm controls No Yes No Yes

Locality controls No Yes No Yes

Dep. var. mean -17.32 -22.17 6.56 3.88

R2 0.00 0.02 0.00 0.05# distinct LLM 16,274 14,759 24,507 23,899# municipalities 1,272 1,255 1,458 1,456# periods 38 38 38 38Observations 132,998 97,524 599,408 475,005

Standard errors in parenthesesThis table reports regression of �rm growth on predicted quarterly export manufactur-ing employment di�erence in the local municipality. Unit of observation are entrepren-eurs with at most 9 employees from 40 quarterly ENOE waves 2005-2014 who are ob-served in the following period. Municipal export employment variables are normalisedper 1000 working-age-population of the corresponding month and municipality, andemployment and outcome variables are winsorised at the top and bottom 0.01%. Two-way clustered standard errors by municipality and trimester. Municipality controlsinclude wave dummies, state dummies, and 2005 export manufacturing employment.Individual controls include gender, education level, and experience quadratic. Firmcontrol include �rm size, industry, and pro�t quintiles. Locality controls include loc-ality size category. All speci�cations include a constant, and ENOE sampling weightsare used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

56

Page 57: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 10

Heterogeneous effects by education level

Low skilled Mid-skilled High skilled

(1) (2) (3) (4) (5) (6)

Dependent Variable: Leave to WERelative

Firm GrowthLeave to WE

RelativeFirm Growth

Leave to WERelative

Firm Growth

∆ Pred. XME/WAP 0.0851 0.109∗∗ 0.118(0.0887) (0.0470) (0.129)

F.∆ Pred. XME/WAP -0.307 -1.608∗∗∗ -0.329(0.407) (0.489) (0.762)

Municipality controls Yes Yes Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes Yes Yes

Dep. var. mean 16.12 -32.66 18.58 -24.55 17.38 -3.59

R2 0.10 0.02 0.08 0.02 0.04 0.02# distinct LLM 22,046 8,979 21,880 10,416 11,591 5,988# municipalities 1,452 1,119 1,405 1,067 976 705# periods 39 38 39 38 39 38Observations 227,059 28,042 276,389 43,317 86,514 26,125

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export manufacturing employment di�erence in the localmunicipality. Unit of observation are entrepreneurs with at most 9 employees from 40 quarterly ENOE waves 2005-2014who are observed in the following period. Low-skilled: up to primary education; mid-skilled: lower and upper secondaryeducation; high-skilled: tertiary education. Municipal export employment variables are normalised per 1000 working-age-population of the corresponding month and municipality, and employment and outcome variables are winsorised atthe top and bottom 0.01%. Two-way clustered standard errors by municipality and trimester. Municipality controlsinclude wave dummies, state dummies, and 2005 export manufacturing employment. Individual controls include gender,education level, and experience quadratic. Firm control include �rm size, industry, and pro�t quintiles. Locality controlsinclude locality size category. All speci�cations include a constant, and ENOE sampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

57

Page 58: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 11

Heterogeneous effects - species analysis

Below median Pr(Leave) Above median Pr(Leave)

(1) (2) (3) (4)

Dependent Variable: Leave to WERelative

Firm GrowthLeave to WE

RelativeFirm Growth

∆ Pred. XME/WAP 0.0617 0.130∗∗∗

(0.0389) (0.0432)

F.∆ Pred. XME/WAP -0.730 -1.309∗∗∗

(0.485) (0.333)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 8.84 -22.00 25.46 -22.34

R2 0.01 0.03 0.05 0.02# distinct LLM 22,814 10,476 22,286 10,956# municipalities 1,460 1,090 1,435 1,143# periods 39 38 39 38Observations 274,290 46,350 305,071 49,556

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export manufacturingemployment di�erence in the local municipality. Unit of observation are entrepreneurswith at most 9 employees from 40 quarterly ENOE waves 2005-2014 who are ob-served in the following period. Municipal export employment variables are normalisedper 1000 working-age-population of the corresponding month and municipality, andemployment and outcome variables are winsorised at the top and bottom 0.01%. Two-way clustered standard errors by municipality and trimester. Municipality controlsinclude wave dummies, state dummies, and 2005 export manufacturing employment.Individual controls include gender, education level, and experience quadratic. Firmcontrol include �rm size, industry, and pro�t quintiles. Locality controls include loc-ality size category. All speci�cations include a constant, and ENOE sampling weightsare used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

58

Page 59: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 12

Effects on incomes in manufacturing and non-manufacturing firms

Dependent Variable: 100×∆ log(average income)

Sector: Manufacturing Non-manufacturing

(1) (2) (3) (4) (5) (6)

Type of income: Pro�tsSmall �rmwages

Large �rmwages

Pro�tsSmall �rmwages

Large �rmwages

∆ Pred. XME/WAPop -0.113 0.211∗∗ -0.0435 0.0630 0.0465 0.0781∗∗

(0.183) (0.0901) (0.0586) (0.0734) (0.0606) (0.0388)

Standard controls Yes Yes Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes Yes Yes

Municipality �xed e�ects

Dep. var. mean -0.34 -0.12 -0.11 -0.68 -0.24 -0.14

R2 0.01 0.01 0.01 0.02 0.02 0.01# municipalities 1,322 1,172 1,026 1,464 1,475 1,369Observations 21,753 19,437 20,183 35,714 36,069 30,996

Standard errors in parenthesesThis table reports regression of the log average income di�erence on predicted quarterly export manu-facturing employment di�erence. Unit of observation is an unbalanced panel of Mexican municipalities(apart from D.F.) over 40 quarters from 2005 to 2010. All employment variables are normalised perinitial 1000 working-age-population of the corresponding trimester and municipality, and employmentand outcome variables are winsorised at the top and bottom 0.05%. All speci�cations include a con-stant, and per standard control for state, trimester, 2005 export employment. Individual controlsare municipal average levels of education, experience, and share of men. Firm controls are municipalbaseline industry shares and �rm size distributions. Locality controls are locality size distributionwithin municipality. Observations are weighted by working-age population. Standard errors clusteredby municipality.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

59

Page 60: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 13

Effects on occupational and entrepreneurial choices: manufacturing vsnon-manufacturing

Manufacturing Not Manufacturing

(1) (2) (3) (4)

Dependent Variable: Leave to WERelative

Firm GrowthLeave to WE

RelativeFirm Growth

∆ Pred. XME/WAP -0.105 0.131∗∗∗

(0.103) (0.0435)

F.∆ Pred. XME/WAP -1.461∗∗∗ -0.869∗∗∗

(0.555) (0.142)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 13.70 -19.16 18.05 -22.74

R2 0.05 0.03 0.08 0.02# distinct LLM 15,583 5,907 24,405 13,455# municipalities 1,353 911 1,467 1,205# periods 39 38 39 38Observations 83,834 14,700 506,342 82,824

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export manufac-turing employment di�erence in the local municipality. Unit of observation areentrepreneurs with at most 9 employees from 40 quarterly ENOE waves 2005-2014who are observed in the following period. Municipal export employment variablesare normalised per 1000 working-age-population of the corresponding month andmunicipality, and employment and outcome variables are winsorised at the topand bottom 0.01%. Two-way clustered standard errors by municipality and tri-mester. Municipality controls include wave dummies, state dummies, and 2005export manufacturing employment. Individual controls include gender, educationlevel, and experience quadratic. Firm control include �rm size, industry, and pro�tquintiles. Locality controls include locality size category. All speci�cations includea constant, and ENOE sampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

60

Page 61: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE 14

Automotive case study: large plant expansions and small firm growth

Dependent variable: Relative �rm growth

(1) (2) (3) (4)

1Q to announcement 0.287 -3.852 -2.334 -2.164(7.479) (6.996) (7.342) (7.864)

2Q to announcement -2.598 -4.304∗∗ -2.784 -3.336∗

(2.830) (2.072) (2.021) (1.999)

3Q to announcement -1.429 -0.983 0.546 -0.165(2.877) (3.559) (3.342) (3.213)

4Q to announcement 0.125 0.525 2.138 -2.807(8.594) (5.534) (5.582) (2.720)

Expansion announcement -4.401∗ -2.214 -0.788 -0.663(2.278) (2.913) (2.399) (1.999)

Q1 after announcement -6.719∗∗∗ -6.587∗∗ -5.060∗∗ -5.278∗∗

(2.269) (2.629) (2.266) (2.112)

Q2 after announcement -2.501 -11.09∗∗ -9.614∗ -9.426∗

(3.702) (4.675) (5.130) (5.541)

Q3 after announcement 0.365 5.157 6.787 6.531(6.225) (9.887) (9.375) (10.64)

Q4 after announcement 11.08 -1.782 -0.379 0.635(8.929) (3.847) (4.017) (4.085)

Q5 after announcement 10.80∗∗∗ 12.27∗∗∗ 13.65∗∗∗ 12.76∗∗∗

(3.470) (4.066) (3.999) (4.017)

Q6 after announcement 1.775 8.065∗∗∗ 9.417∗∗∗ 9.336∗∗∗

(2.963) (2.395) (2.243) (2.101)

Q7 after announcement 12.50 1.402 2.697 3.120(9.938) (5.064) (5.159) (5.483)

Q8 after announcement 6.836∗ 7.171 8.555 2.355(3.972) (5.978) (6.050) (3.209)

Auto CZ -4.526∗∗ -4.933∗∗

(2.007) (2.243)

F.∆ Pred. XME/WAP -0.903∗∗∗

(0.310)

Auto CZ × F.∆ Pred. XME/WAP 1.316(1.015)

Expansion Q0-Q2 × F.∆ Pred. XME/WAP 0.368(1.010)

CZ controls No Yes Yes YesIndividual controls No Yes Yes YesFirm controls No Yes Yes YesLocality controls No Yes Yes Yes

Dep. var. mean -17.02 -22.09 -22.09 -21.96

R2 0.00 0.02 0.02 0.02# distinct LLM 11,874 10,871 10,871 10,595# CZ 1,038 1,024 1,024 1,008# periods 39 39 39 38Observations 139,394 101,674 101,674 99,622

Standard errors in parenthesesIn this table, I explore the e�ects of an announcement of a major auto manufacturingplant event in the local commuting zone (CZ). Unit of observation are entrepreneurs withat most 9 employees from 40 quarterly ENOE waves 2005-2014 who are observed in thefollowing period. CZ export employment variables are normalised per 1000 working-age-population of the corresponding month and CZ, and employment and outcome variablesare winsorised at the top and bottom 0.01%. Two-way clustered standard errors by CZand trimester. CZ controls include wave dummies, state dummies, and 2005 export manu-facturing employment. Individual controls include gender, education level, and experiencequadratic. Firm control include �rm size, industry, and pro�t quintiles. Locality controlsinclude locality size category. All speci�cations include a constant, and ENOE samplingweights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

61

Page 62: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

APPENDIX

Appendix

A Parametric assumptions in model simulations

I assume very standard functional forms. The production function is assumed to be Cobb-Douglas, withonly capital as an input. The utility function is CRRA. The wage distribution is log-normal. Search frictionsare binary variables that necessarily follow a Bernoulli distribution with a unique parameter, the mean λ. Imaintain these functional forms throughout:

π(k) = A · kα (A1)

u(c) =u1−θ

1− θ(A2)

lnw ∼ N(µw, σ2w) (A3)

I also maintain the parameter values chosen for the initial simulation, and only vary the mean of the log-wagedistribution:

TABLE A1

Parametric assumptions in model simulations

Figure (1) (2) (3) (4) (5) (6) (7)

??: RoR ??a ??b3: Intuition 34: Ability 4b 4a 4c4: Comp. Stat. 4d 4e 4f 4d 4e 4f??: Micro�nance ??b ??a ??c

β 0.95 0.95 0.95 0.95 0.95 0.95 0.95θ 1.5 1.5 1.5 1.5 1.5 1.5 1.5A 1.6 1.6 1.5 1.7 1.6 1.5 1.7α 0.5 0.5 0.5 0.5 0.5 0.5 0.5µw � 2 2 2 2.2 2.2 2.2σw � 0.2 0.2 0.2 0.2 0.2 0.2λ 0 0.7 0.7 0.7 0.7 0.7 0.

62

Page 63: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

APPENDIX

B Map of Mexican export manufacturing in 2005

FIGURE A1

Export manufacturing employment across Mexican municipalities in 2005

Source: The map shows the penetration (per 1,000 working-age population) of export manufacturing jobs at the municip-ality level in 2005. Employment data from IMSS, population data from 2005 Census, municipality and state boundariesfrom INEGI Marco Geoestadístico Nacional.

63

Page 64: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

APPENDIX

C Additional results on transitions and sector shares

This appendix documents additionall results on transitions in the Mexican labour market, and the e�ects ofXME shocks on transitions and sector shares. In Table A2 I show a comparable �gure of annual transitionrates out of entrepreneurship for working-age men. The transition rates are remarkably similar acrosscountries, and transitions to wage-employment are by far the most common type. In Table A3 I showthe e�ect of XME shocks on employment shares. I split the former category `not in paid employment'into its components: out of the labour force (not employed, not searching for a job), other employment(mostly unpaid) and unemployment. The results in the table indicate that the signi�cant increase in wage-employment which results from an XME is mostly (to 80 %) o�set by a reduction in unemployment. Thebulk of the remainder is driven by a reduction in entrepreneurship / self-employment, although this e�ect iseconomically small and statistically not signi�cant. In Table A4 I split up the increase in wage-employment.Unlike in table A3, the columns do not need to add up. Column (1) reproduces the estimate on overall wageemployment. In columns (2) and (3) I split up wage-employment into large �rms (more than 10 workers)and small �rms. All of the employment e�ects are driven by large �rms. In columns (4) and (5) I o�er analternative split into formal and informal wage-employment. Again, most of the e�ects are driven by formalemployment. In column (6) I separately estimate the e�ects on export manufacturing employment, a subsetof formal wage-employment. The magnitude of the estimate indicates that while constituting the largestpart, not all wage employment e�ects are created by the export manufacturing sector. In other words, some(around 0.02 percentage points) of the overall job creation happens in other sectors.

TABLE A2

Annual transition rates out of entrepreneurship, men [in %]

Country NEt+1 WEt+1 SEt+1 Total

Colombia 9.32 22.61 68.07 100.00Ghana 9.20 20.25 70.55 100.00Mexico 13.11 19.74 67.15 100.00

Notes: Working-age (15-64) men self-employed in period t and observed intwo adjacent periods. WE=Wage employed, SE=Self-employed (includesentrepreneurs), NE=Not gainfully employed (includes unpaid workers). Se-lected developing countries with available LFS panel data. Own calculations.Sources: 2005-2014 ENOE (Mexico), 2007-2010 ESLF (Colombia), 2005-2013GHUPS (Ghana).

64

Page 65: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A3

Effect of export shock on employment shares

Dependent Variable: 100×∆ sectoral employment share

(1) (2) (3) (4) (5)

Type of employment:Out of

labour forceWage

EmploymentEntre-

preneurshipOther

employmentUnempl-oyment

∆ Pred. XME/WAP -0.00431 0.0651∗∗∗ -0.00912 -0.000870 -0.0515∗∗∗

(0.0122) (0.0120) (0.00895) (0.00480) (0.0118)

Municipality controls Yes Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes Yes

R2 0.02 0.04 0.06 0.03 0.01# municipalities 1,496 1,496 1,496 1,496 1,496Observations 38,331 38,331 38,331 38,331 38,331

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Standard errors in parentheses, clustered by municipality.Note: This table reports regression of employment shares on predicted quarterly export manufac-turing employment di�erence. Sample is an unbalanced panel of Mexican municipalities (apartfrom D.F.) over 40 quarters from 2005 to 2010. Export employment is normalised per initial1000 working-age-population of the corresponding trimester and municipality, and employmentand outcome variables are winsorised at the top and bottom 0.05%. All speci�cations includea constant, and per standard control for state, trimester, 2005 export employment. Individualcontrols are municipal average levels of education, experience, and share of men. Firm controlsare municipal baseline industry shares and �rm size distributions. Locality controls are localitysize distribution within municipality. Observations are weighted by working-age population.

TABLE A4

Effect of export shock on types of wage employment

Dependent Variable: 100×∆ sectoral employment share

(1) (2) (3) (4) (5) (6)Type of wage-empl.: Any Large Firm Small �rm Formal Informal XME

∆ Pred. XME/WAP 0.0651∗∗∗ 0.0659∗∗∗ -0.0000653 0.0632∗∗∗ 0.00221 0.0477∗∗∗

(0.0120) (0.0105) (0.0162) (0.0127) (0.00727) (0.00564)

Municipality controls Yes Yes Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes Yes Yes

R2 0.04 0.12 0.12 0.03 0.04 0.04# municipalities 1,496 1,495 1,495 1,496 1,496 1,496Observations 38,331 37,186 37,186 38,331 38,331 38,331

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Standard errors in parentheses, clustered by municipality.Note: This table reports regression of employment shares on predicted quarterly export manufacturingemployment di�erence. Sample is an unbalanced panel of Mexican municipalities (apart from D.F.)over 40 quarters from 2005 to 2010. Export employment is normalised per initial 1000 working-age-population of the corresponding trimester and municipality, and employment and outcome variablesare winsorised at the top and bottom 0.05%. All speci�cations include a constant, and per standardcontrol for state, trimester, 2005 export employment. Individual controls are municipal averagelevels of education, experience, and share of men. Firm controls are municipal baseline industryshares and �rm size distributions. Locality controls are locality size distribution within municipality.Observations are weighted by working-age population.

65

Page 66: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

APPENDIX

D Additional heterogeneity results

This appendix documents additional heterogeneity analysis. Overall, no clear pattern emerges. I speculatethat the underlying reason is that each subgroup di�ers not only in their propensity to take a wage job (andone in export manufacturing). But they also di�er in the type of business they run, etc. Therefore therethe predictions from theory for di�erences in sub-groups are ambiguous. This makes interpretation of theseresults quite hard. Furthermore, subsamples can be of quite di�erent size, which a�ects the power of thetest. I nevertheless report results for completeness.

When splitting by gender (Table A5) or age (Table A6) no clear pattern or conclusion emerges at all. Forsplits by hours (A8), size of locality (Table A9), or migrant status of the entrepreneur (Table A7), coe�cientsare similar to the main results (in size and statistical signi�cance) for full-time entrepreneurs, those in urbanlocalities, and non-migrants. The mechanism should be expected to hold especially for these sub-samples.In particular, the fact that results uphold for non-migrants alleviate some concerns that selective migrationconfounds results. Results for part-time, rural and migrant entrepreneurs are more mixed and not veryconclusive. In any case, the di�erences across subgroups are not signi�cant.

Finally, I split the sample not according to personal or locality characteristics, but according to the signof the shock. For this split, theory provides some guidance. In order to have any e�ect on �rm growth,according to the model, future wage employment opportunities need to be anticipated, at least a few quartersin advance. This seems more likely in the case of expansions � new factories, or enlargements of existingplants, take time to build. The automobile case study in section 6.3 is a point in case. Redundancies andhiring freezes tend to be known rather suddenly. Table A10 documents the baseline results when the sampleis split according to the sign of the XME shock. For positive shocks, the e�ect on leaving is twice as large,but not individually signi�cant. The coe�cient estimate on growth is basically the same as the main result,and signi�cant at the 5 % level. However, the coe�cients for negative shocks are of similar magnitude and/ornot signi�cantly di�erent.

I also repeat the timing analysis for positive and negative shocks, reported in Figure A2. The results pointin the direction that we would expect, but they are not very precise or clear-cut. For positive XME shocks,leaving to WE has a positive coe�cient for a number of quarters before and after the XME shock; althoughonly the �rst lead is individually signi�cant at the 5 % level. This is consistent with a gradual hiringprocess and a time to build. Firm growth is signi�cantly negative in the three quarters before an XMEshock. However, the coe�cients continue to be negative after the XME shock. For negative XME shocks,switches to WE are only occur one quarter after the predicted shock; and this coe�cient is signi�cant. Thisis consistent with an interpretation that negative employment adjustments are sudden, and likely to be notanticipated. For negative XME shocks, we do not see an e�ect on small �rm growth that moves in theopposite direction to XME changes; neither before nor after the negative shock. Rather, small �rm growthmoves in the same direction than XME shocks, in particular in Q3 and Q4 after the negative shock. This isconsistent with a local demand channel being at play.

66

Page 67: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A5

Heterogeneous effects - by gender

Male entrepreneurs Female entrepreneurs

(1) (2) (3) (4)

Dependent Variable: Leave to WERelative

Firm GrowthLeave to WE

RelativeFirm Growth

∆ Pred. XME/WAP 0.0449 0.255∗∗

(0.0500) (0.101)

F.∆ Pred. XME/WAP -1.236∗∗∗ 0.0537(0.299) (0.534)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 21.35 -19.41 11.38 -33.12

R2 0.08 0.02 0.04 0.05# distinct LLM 23,039 13,508 21,881 6,709# municipalities 1,427 . 1,451 .# periods 39 . 39 .Observations 361,395 77,023 228,781 20,501

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export manufac-turing employment di�erence in the local municipality. Unit of observation areentrepreneurs with at most 9 employees from 40 quarterly ENOE waves 2005-2014who are observed in the following period. Municipal export employment variablesare normalised per 1000 working-age-population of the corresponding month andmunicipality, and employment and outcome variables are winsorised at the topand bottom 0.01%. Two-way clustered standard errors by municipality and tri-mester. Municipality controls include wave dummies, state dummies, and 2005export manufacturing employment. Individual controls include gender, educationlevel, and experience quadratic. Firm control include �rm size, industry, and pro�tquintiles. Locality controls include locality size category. All speci�cations includea constant, and ENOE sampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

67

Page 68: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A6

Heterogeneous effects - by age group

Age 15-29 Age 30-39 Age 40-49 Age 50-64

(1) (2) (3) (4) (5) (6) (7) (8)

Dependent Variable:Leaveto WE

RelativeFirm Growth

Leaveto WE

RelativeFirm Growth

Leaveto WE

RelativeFirm Growth

Leaveto WE

RelativeFirm Growth

∆ Pred. XME/WAP 0.380∗ 0.129∗∗ -0.00364 0.0917(0.205) (0.0576) (.) (0.0826)

F.∆ Pred. XME/WAP -0.226 -0.424 -1.512∗∗ -1.132∗∗

(1.161) (0.587) (0.641) (0.532)

Municipality controls Yes Yes Yes Yes Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes Yes Yes Yes Yes

Dep. var. mean 29.59 -32.34 19.49 -23.15 15.27 -20.12 11.84 -20.58

R2 0.10 0.04 0.06 0.03 0.05 0.03 0.05 0.02# distinct LLM 15,106 3,897 19,709 8,020 20,088 8,932 19,279 7,842# municipalities 1,321 727 1,401 1,003 1,396 1,032 1,393 995# periods 39 38 39 38 39 38 39 38Observations 79,125 8,316 162,604 26,993 178,899 33,721 165,058 27,932

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export manufacturing employment di�erence in the localmunicipality. Unit of observation are entrepreneurs with at most 9 employees from 40 quarterly ENOE waves 2005-2014 whoare observed in the following period. Municipal export employment variables are normalised per 1000 working-age-populationof the corresponding month and municipality, and employment and outcome variables are winsorised at the top and bottom0.01%. Two-way clustered standard errors by municipality and trimester. Municipality controls include wave dummies, statedummies, and 2005 export manufacturing employment. Individual controls include gender, education level, and experiencequadratic. Firm control include �rm size, industry, and pro�t quintiles. Locality controls include locality size category. Allspeci�cations include a constant, and ENOE sampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

68

Page 69: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A7

Heterogeneous effects - by migration status

Migrant Non-migrant

(1) (2) (3) (4)

Dependent Variable: Leave to WERelative

Firm GrowthLeave to WE

RelativeFirm Growth

∆ Pred. XME/WAP 0.166 0.0979∗∗∗

(0.114) (0.0380)

F.∆ Pred. XME/WAP -0.537 -1.185∗∗∗

(0.566) (0.389)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 17.61 -19.19 17.35 -23.20

R2 0.06 0.03 0.09 0.03# distinct LLM 13,702 6,047 24,521 13,492# municipalities 1,066 . 1,475 .# periods 39 . 39 .Observations 133,573 24,420 456,603 73,104

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export manufac-turing employment di�erence in the local municipality. Unit of observation areentrepreneurs with at most 9 employees from 40 quarterly ENOE waves 2005-2014who are observed in the following period. Migrants are those entrepreneurs whoreside in a federal entity which is di�erent from the state they were born in. Muni-cipal export employment variables are normalised per 1000 working-age-populationof the corresponding month and municipality, and employment and outcome vari-ables are winsorised at the top and bottom 0.01%. Two-way clustered standarderrors by municipality and trimester. Municipality controls include wave dummies,state dummies, and 2005 export manufacturing employment. Individual controlsinclude gender, education level, and experience quadratic. Firm control include �rmsize, industry, and pro�t quintiles. Locality controls include locality size category.All speci�cations include a constant, and ENOE sampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

69

Page 70: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A8

Heterogeneous effects - full-time vs part-time entrepreneurs

Full-time Part-time

(1) (2) (3) (4)

Dependent Variable: Leave to WERelative

Firm GrowthLeave to WE

RelativeFirm Growth

∆ Pred. XME/WAP 0.0990∗∗ 0.117(0.0411) (0.0759)

F.∆ Pred. XME/WAP -1.062∗∗∗ -0.583(0.313) (0.682)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 17.23 -20.14 17.70 -33.18

R2 0.09 0.02 0.06 0.04# distinct LLM 23,755 13,677 20,226 5,578# municipalities 1,464 . 1,431 .# periods 39 . 39 .Observations 400,391 82,688 184,097 14,089

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export man-ufacturing employment di�erence in the local municipality. Unit of observa-tion are entrepreneurs with at most 9 employees from 40 quarterly ENOEwaves 2005-2014 who are observed in the following period. Full-time areentrepreneurs working at least 35 hours per week; entrepreneurs with fewerweekly hours are classi�ed as part-time. Municipal export employment vari-ables are normalised per 1000 working-age-population of the correspondingmonth and municipality, and employment and outcome variables are win-sorised at the top and bottom 0.01%. Two-way clustered standard errorsby municipality and trimester. Municipality controls include wave dummies,state dummies, and 2005 export manufacturing employment. Individual con-trols include gender, education level, and experience quadratic. Firm controlinclude �rm size, industry, and pro�t quintiles. Locality controls include loc-ality size category. All speci�cations include a constant, and ENOE samplingweights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

70

Page 71: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A9

Heterogeneous effects - entrepreneurs in urban vs rural localities

Rural localities(<15k inhabitants)

Urban localities(>15k inhabitants)

(1) (2) (3) (4)

Dependent Variable: Leave to WERelative

Firm GrowthLeave to WE

RelativeFirm Growth

∆ Pred. XME/WAP 0.163∗ 0.116∗∗∗

(0.0906) (0.0352)

F.∆ Pred. XME/WAP 0.183 -1.276∗∗∗

(0.805) (0.314)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 15.75 -30.27 18.20 -19.43

R2 0.11 0.03 0.07 0.02# distinct LLM 18,676 8,256 12,868 9,085# municipalities 1,383 . 441 426# periods 39 . 39 38Observations 149,030 18,664 441,146 78,860

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export manufac-turing employment di�erence in the local municipality. Unit of observation areentrepreneurs with at most 9 employees from 40 quarterly ENOE waves 2005-2014who are observed in the following period. Municipal export employment variablesare normalised per 1000 working-age-population of the corresponding month andmunicipality, and employment and outcome variables are winsorised at the topand bottom 0.01%. Two-way clustered standard errors by municipality and tri-mester. Municipality controls include wave dummies, state dummies, and 2005export manufacturing employment. Individual controls include gender, educationlevel, and experience quadratic. Firm control include �rm size, industry, and pro�tquintiles. Locality controls include locality size category. All speci�cations includea constant, and ENOE sampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

71

Page 72: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A10

Differential effects for XME contractions and expansions

Positive XME shock Negative XME shock

(1) (2) (3) (4)

Dependent Variable: Leave to WERelative

Firm GrowthLeave to WE

RelativeFirm Growth

∆ Pred. XME/WAP 0.200 0.0971(0.138) (0.140)

F.∆ Pred. XME/WAP -0.983∗∗ -0.890(0.422) (0.866)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 18.09 -21.94 17.49 -19.89

R2 0.07 0.03 0.08 0.02# distinct LLM 14,890 8,871 10,095 5,888# municipalities 888 815 810 733# periods 36 35 34 32Observations 354,760 57,982 200,084 35,920

Standard errors in parenthesesThis table reports regression of �rm growth and upgrading on predicted quarterly exportmanufacturing employment di�erence in the local municipality. Unit of observation areentrepreneurs with at most 9 employees from 40 quarterly ENOE waves 2005-2014 whoare observed in the following period. Low-skilled: up to primary education; mid-skilled:lower and upper secondary education; high-skilled: tertiary education. Municipal exportemployment variables are normalised per 1000 working-age-population of the correspond-ing month and municipality, and employment and outcome variables are winsorised at thetop and bottom 0.01%. Two-way clustered standard errors by municipality and trimester.Municipality controls include wave dummies, state dummies, and 2005 export manufac-turing employment. Individual controls include gender, education level, and experiencequadratic. Firm control include �rm size, industry, and pro�t quintiles. Locality controlsinclude locality size category. All speci�cations include a constant, and ENOE samplingweights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

72

Page 73: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

FIGURE A2

Timing by sign of XME shock

Positive XME shocks

-.4-.2

0.2

.4E

ffect

siz

e: le

ave

SE

to W

E

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(a) Occupational choice: leave to WE

-4-3

-2-1

01

23

4E

ffect

siz

e: re

lativ

e fir

m g

row

th

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(b) Entrepreneurial choice: relative �rm growth

Negative XME shocks

-.4-.2

0.2

.4E

ffect

siz

e: le

ave

SE

to W

E

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(c) Occupational choice: leave to WE

-3-2

-10

12

3E

ffect

siz

e: re

lativ

e fir

m g

row

th

-4 -3 -2 -1 0 1 2 3 4Quarter after export manufacturing shock

(d) Entrepreneurial choice: relative �rm growth

Source: Results from regression of model 14, using the speci�cation reported in Column (4) of Table 8 (left) and Table9 (right). Horizontal axis documents the number of quarters s that the dependent variable has been shifted. Eachcoe�cient corresponds to a separate regression. Reported 95 % con�dence intervals use two-way clustered standarderrors by municipality and quarter.

73

Page 74: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

APPENDIX

E Additional results on wages

This appendix documents two further results on wages. Table A11 repeats the analysis from Table 6,substituting wages of all workers with wages paid to new workers (those not observed in the same sector in theprevious quarter) upon entry. The e�ects are more pronounced: the coe�cient on small �rm manufacturingwages has doubled, the one on small-�rm wages not in manufacturing has increased by 1.5. However, thelatter e�ect continues to be non-signi�cant, and much smaller than any other e�ect. What is new here is thatpro�ts of new entrants into entrepreneurship increase with an export manufacturing shock. This is consistentwith a selection e�ect: given more attractive alternatives in the labour market, potential entrepreneurs onlystart relatively more pro�table projects. Table A12 splits wage employment into formal (IMSS-a�liated)and informal. Around 80 percent of small �rm jobs are informal, and conversely around 80 percent oflarge-�rm jobs are formal. Given this imperfect overlap, the results are not very conclusive. There is noe�ect on average formal or informal wages either in manufacturing or elsewhere. Formal entry wages inmanufacturing reduce, whereas informal entry wages in non-manufacturing increase. Both of these e�ectsare only signi�cant at the 10% level, not having adjusted for multiple hypothesis testing.

TABLE A11

Effects on starting incomes in manufacturing and non-manufacturing firms

Dependent Variable: 100×∆ log(average entry income)

Sector: Manufacturing Non-manufacturing

(1) (2) (3) (4) (5) (6)

Type of income: Pro�tsSmall �rmwages

Large �rmwages

Pro�tsSmall �rmwages

Large �rmwages

∆ Pred. XME/WAPop 0.104 0.381∗∗∗ -0.0593 0.270∗∗∗ 0.0781 0.123(0.438) (0.118) (0.0908) (0.0974) (0.0722) (0.0944)

Standard controls Yes Yes Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes Yes Yes

Municipality �xed e�ects

Dep. var. mean -0.87 -0.28 -0.26 -0.79 -0.54 -0.32

R2 0.01 0.01 0.01 0.01 0.02 0.01# municipalities 1,188 1,025 862 1,426 1,449 1,202Observations 14,634 12,526 13,052 29,465 31,445 19,396

Standard errors in parenthesesThis table reports regression of the log average income di�erence on predicted quarterly export manu-facturing employment di�erence. Unit of observation is an unbalanced panel of Mexican municipalities(apart from D.F.) over 40 quarters from 2005 to 2010. All employment variables are normalised perinitial 1000 working-age-population of the corresponding trimester and municipality, and employmentand outcome variables are winsorised at the top and bottom 0.05%. All speci�cations include a con-stant, and per standard control for state, trimester, 2005 export employment. Individual controlsare municipal average levels of education, experience, and share of men. Firm controls are municipalbaseline industry shares and �rm size distributions. Locality controls are locality size distributionwithin municipality. Observations are weighted by working-age population. Standard errors clusteredby municipality.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

74

Page 75: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

APPENDIX

TABLE A12

Effects of of export shocks on wages - by formal/informal split

Dependent Variable: 100×∆ log(average income)

Sector: Manufacturing Non-manufacturing

Subsample: All Entrant All Entrant

(1) (2) (3) (4) (5) (6) (7) (8)

Type of income:Informalwages

Formalwages

Informalwages

Formalwages

Informalwages

Formalwages

Informalwages

Formalwages

∆ Pred. XME/WAPop 0.0714 0.00460 0.136 -0.136 0.0480 0.0429 0.0477 0.0241(0.107) (0.0592) (0.160) (0.0833) (0.0487) (0.0295) (0.0489) (0.0607)

Standard controls Yes Yes Yes Yes Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes Yes Yes Yes Yes

Municipality �xed e�ects

Dep. var. mean -0.05 -0.20 -0.21 -0.37 -0.12 -0.23 -0.19 -0.42

R2 0.01 0.01 0.01 0.01 0.02 0.01 0.02 0.01# municipalities 1,203 963 1,097 814 1,473 1,351 1,473 1,176Observations 20,858 19,543 15,082 12,998 36,277 30,576 36,126 20,958

Standard errors in parenthesesThis table reports regression of the log average income di�erence on predicted quarterly export manufacturingemployment di�erence. Unit of observation is an unbalanced panel of Mexican municipalities (apart from D.F.) over40 quarters from 2005 to 2010. All employment variables are normalised per initial 1000 working-age-populationof the corresponding trimester and municipality, and employment and outcome variables are winsorised at thetop and bottom 0.05%. All speci�cations include a constant, and per standard control for state, trimester, 2005export employment. Individual controls are municipal average levels of education, experience, and share of men.Firm controls are municipal baseline industry shares and �rm size distributions. Locality controls are locality sizedistribution within municipality. Observations are weighted by working-age population. Standard errors clusteredby municipality.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

F Robustness checks

This appendix documents a number of robustness checks on the main results. In Table A13 I repeat thespeci�cations from Table 8 and from columns (3) and (4) of Table 9, but I restrict the sample to entrepreneurswith at least one paid worker. This is the same restriction which is applied in order to create the relative�rm growth variable. Descriptively, these entrepreneurs are just as likely to take a wage job than are theself-employed with no workers, but they are much less likely to leave employment altogether. An XMEshock has the same e�ect on the transition to wage work than it does for the full sample, but the e�ect onleaving to any occupation becomes zero in column (2) with the full vector of controls. For this subsample ofemployer-entrepreneurs, the e�ect on absolute �rm growth is large and signi�cant. This suggests that thenull e�ect for the whole sample is driven by a failure of solo entrepreneurs to reduce their �rm size, which islikely due to the censoring of �rm size at zero paid workers.

Table A14 reports OLS results using actual instead of predicted export manufacturing employment changes.The coe�cients on �rm growth are now positive, but small and not signi�cant. The coe�cients on leavingare still positive, but reduced in size. This suggests a downward bias of OLS for leaving, and an upwardbias for growth. We would expect a bias in these directions if small and large (export manufacturing) �rmgrowth are driven by a common, unobservable variable. Then, small �rms and large �rms growth together.Episodes of export manufacturing growth also provide less incentives for taking up wage jobs if the relativeearnings are less responsive.

75

Page 76: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

APPENDIX

In Table A15, I report results from 2SLS instrumental variables regression. Predicted XME changes serve asan instrument for actual XME changes. I restrict the sample to non-manufacturing entrepreneurs. Rememberthat for this sub-sample, alternative mechanisms are less of a concern. Predicted XME changes are a stronginstrument for actual XME changes in a sample of individual entrepreneurs. This mirrors the result I obtainin Table 5 at the local labour market level. The Kleibergen-Papp heteroscedasticity-robust test statisticis well over the the Stock and Yogo (2005) critical values for a 10% maximal IV size. The second stageregression coe�cients are about 1.7 times as large as the associated reduced-form results in table 13. This isto be expected, given that the �rst-stage coe�cient is around 0.585 � the 2SLS coe�cient equals the reducedform divided by the �rst stage coe�cient. Caution needs to be taken when interpreting the IV results.In particular, they should not be interpreted as an `average treatment e�ect' estimate for the magnitudeof the occupational choice mechanism on �rm size. First, even though caution has been taken to removedirect mechanisms of XME changes on �rm pro�ts which work in the same direction as the occupationalchoice channel, some direct mechanisms may still have a quantitative in�uence, for example if they go in theopposite direction. Second, as I allude to in the conclusion, a simple linear extrapolation will likely not bea good approximation to the magnitude of the e�ects.

I �nally report robustness checks to alternative de�nitions of XME changes. In Table A16, I apply thecommuting-zone aggregation from section 6.3 and rede�ne local labour markets as commuting zones, insteadof municipalities. The coe�cients on leaving are similar to the main results. The coe�cient on �rm growthis reduced in magnitude and now only signi�cant at the 10 % level. On the other hand, the coe�cient onabsolute �rm growth is now large and also signi�cant. Next, I disaggregate my measure of XME changes,and change the composition of the industries that are included in my de�nition of export manufacturing.One potential concern with the identi�cation strategy could be that production of exports is concentratedin a small number of local labour markets. Then, a local market level shock transmits to aggregate Mexicanexports to the US. The ideal check to alleviate this concern would be to predict export employment in muni-cipality m with exports from all other municipalities apart from m. However, municipality-level export dataare not available. Instead, I check whether results are driven by exports and employment in any particularindustry (de�ned by a NAICS sub-sector). In Tables A17-A26, I report results obtained with predicted XMEchanges, leaving out each of the 10 constituent export industries in turn. None of these changes a�ects thesign or signi�cance of any result. Most of these leave-one-out results are very similar in magnitude. Whenthe magnitude is di�erent � such as when leaving out large industries such as transportation equipment orelectrical and appliances � it appears to be even larger. The auto industry is one of the more concentratedexport industries. If aggregate exports are not immune to local production shocks, this seems to attenuaterather than magnify the results (see also the OLS table).

76

Page 77: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A13

Robustness: Effects on leaving entrepreneurship for entrepreneurs with paid workers

(1) (2) (3) (4) (5) (6)

Dependent Variable: Leave Leave Leave to WE Leave to WEAbsolute

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.0723∗ 0.000563 0.159∗∗∗ 0.0966∗∗∗

(0.0420) (0.0734) (0.0471) (0.00733)

F.∆ Pred. XME/WAP -0.981∗∗ -1.627∗∗∗

(0.450) (0.486)

Municipality controls Yes Yes Yes Yes Yes Yes

Individual controls No Yes No Yes No Yes

Firm controls No Yes No Yes No Yes

Locality controls No Yes No Yes No Yes

Dep. var. mean 25.69 25.90 17.98 18.20 -34.38 -42.54

R2 0.01 0.07 0.01 0.07 0.00 0.03# distinct LLM 18,475 17,005 17,914 16,419 16,274 14,759# municipalities 1,340 1,320 1,325 1,307 1,272 1,255# periods 39 39 39 39 38 38Observations 182,645 133,993 165,530 121,512 132,998 97,524

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export manufacturing employment di�erencein the local municipality. Unit of observation are entrepreneurs with at least 1 and at most 9 employeesfrom 40 quarterly ENOE waves 2005-2014 who are observed in the following period. Municipal exportemployment variables are normalised per 1000 working-age-population of the corresponding month andmunicipality, and employment and outcome variables are winsorised at the top and bottom 0.01%. Two-wayclustered standard errors by municipality and trimester. Municipality controls include wave dummies, statedummies, and 2005 export manufacturing employment. Individual controls include gender, education level,and experience quadratic. Firm control include �rm size, industry, and pro�t quintiles. Locality controlsinclude locality size category. All speci�cations include a constant, and ENOE sampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

77

Page 78: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A14

Robustness: OLS regression

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ XME/WAP 0.0467 0.0496(0.0290) (0.0306)

F.∆ XME/WAP 0.101 0.0830(0.265) (0.137)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.61 17.39 -22.27 3.82

R2 0.07 0.08 0.02 0.05# distinct LLM 20,397 20,024 12,976 20,166# municipalities 1,489 1,476 1,274 1,471# periods 38 38 39 39Observations 707,913 573,675 99,540 485,832

Standard errors in parenthesesThis table reports OLS regression of outcomes on observed quarterly export manu-facturing employment di�erence in the local municipality. Unit of observation areentrepreneurs with at most 9 employees from 40 quarterly ENOE waves 2005-2014who are observed in the following period. Municipal export employment variables arenormalised per 1000 working-age-population of the corresponding month and muni-cipality, and employment and outcome variables are winsorised at the top and bottom0.01%. Two-way clustered standard errors by municipality and trimester. Municip-ality controls include wave dummies, state dummies, and 2005 export manufactur-ing employment. Individual controls include gender, education level, and experiencequadratic. Firm control include �rm size, industry, and pro�t quintiles. Locality con-trols include locality size category. All speci�cations include a constant, and ENOEsampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

78

Page 79: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A15

Robustness: IV-2SLS regression resultsIV Second Stage results

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ XME/WAP 0.323*** 0.243***(0.107) (0.0714)

F.∆ XME/WAP -1.628*** -0.319(0.367) (0.350)

Standard controls Y Y Y Y

Individual controls Y Y Y Y

Firm controls Y Y Y Y

Locality controls Y Y Y Y

Dep. var. mean 33.52 18.02 -22.74 4.191K-P Wald F-statistic 26.42 26.03 24.49 24.17

IV First Stage results

∆ Pred. XME/WAP 0.585*** 0.590***(0.114) (0.116)

F.∆ Pred. XME/WAP 0.585*** 0.586***(0.118) (0.119)

R2 . . . .# distinct LLM 23,729 23,099 13,044 22,543# municipalities 1,481 1,466 1,205 1,442# periods 38 38 38 38Observations 602,838 492,126 82,824 404,120

Source: This table reports IV-2SLS regression of outcomes on quarterly export manufacturing em-ployment di�erence in the local municipality. Unit of observation are entrepreneurs not in manu-facturing with at most 9 employees from 40 quarterly ENOE waves 2005-2014 who are observed inthe following period. Municipal export employment variables are normalised per 1000 working-age-population of the corresponding month and municipality, and employment and outcome variables arewinsorised at the top and bottom 0.01%. Two-way clustered standard errors by municipality andquarter. Municipality controls include wave dummies, state dummies, and 2005 export manufacturingemployment. Individual controls include gender, education level, and experience quadratic. Firm con-trols include �rm size, industry, and pro�t quintiles. Locality controls include locality size category.All speci�cations include a constant, and ENOE sampling weights are used. Stock-Yogo (2005) Crit-ical Values: 16.38 (at 10% maximal IV size), 8.96 (at 15 % maximal IV size), 5.53 (at 20% maximalIV size).

79

Page 80: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A16

Robustness: CZ-level regression

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.138∗∗ 0.134∗∗∗

(0.0650) (0.0458)

F.∆ Pred. XME/WAP -0.656∗ -0.324∗

(0.364) (0.181)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.63 17.68 -21.96 4.27

R2 0.07 0.08 0.02 0.04# distinct LLM 18,187 17,815 10,596 17,062# CZ 1,230 1,217 1,008 1,197# periods 39 39 38 38Observations 746,596 605,922 99,622 487,264

Standard errors in parenthesesThis table reports regression of outcomes on predicted quarterly export manufac-turing employment di�erence in the commuting zone (CZ). Unit of observationare entrepreneurs with at most 9 employees from 40 quarterly ENOE waves2005-2014 who are observed in the following period. CZ export employmentvariables are normalised per 1000 working-age-population of the correspondingmonth and CZ, and employment and outcome variables are winsorised at thetop and bottom 0.01%. Two-way clustered standard errors by CZ and trimester.CZ controls include wave dummies, state dummies, and 2005 export manufac-turing employment. Individual controls include gender, education level, andexperience quadratic. Firm control include �rm size, industry, and pro�t quin-tiles. Locality controls include locality size category. All speci�cations includea constant, and ENOE sampling weights are used.

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

80

Page 81: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A17

Robustness: Leaving out textile products manufacturing (NAICS 314)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.109∗ 0.111∗∗∗

(0.0621) (0.0374)

∆ Pred. XME/WAP -0.973∗∗∗ -0.207(0.233) (0.206)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 25,378 24,843 14,700 23,763# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving out tex-tile product manufacturing (NAICS sub-sector 314).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

TABLE A18

Robustness: Leaving out apparel manufacturing (NAICS 315)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.113∗ 0.109∗∗∗

(0.0584) (0.0373)

∆ Pred. XME/WAP -0.961∗∗∗ -0.231(0.255) (0.213)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 23,962 23,491 14,245 22,480# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving out ap-parel manufacturing (NAICS sub-sector 315).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

81

Page 82: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A19

Robustness: Leaving out leather, hide and allied products manufacturing (NAICS 314)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.102∗ 0.104∗∗∗

(0.0538) (0.0365)

∆ Pred. XME/WAP -0.981∗∗∗ -0.214(0.241) (0.204)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 25,360 24,827 14,696 23,749# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving outleather, allied and hide manufacturing (NAICS sub-sector 316).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

TABLE A20

Robustness: Leaving out metal products manufacturing (NAICS 332)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.106∗ 0.106∗∗∗

(0.0617) (0.0390)

∆ Pred. XME/WAP -0.991∗∗∗ -0.217(0.236) (0.212)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 23,576 23,118 14,000 22,123# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving out metalproducts manufacturing (NAICS sub-sector 332).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

82

Page 83: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A21

Robustness: Leaving out machinery and equipment manufacturing (NAICS 333)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.0992∗ 0.108∗∗∗

(0.0603) (0.0359)

∆ Pred. XME/WAP -1.010∗∗∗ -0.225(0.228) (0.211)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 25,483 24,948 14,746 23,864# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving out ma-chinery and equipment manufacturing (NAICS sub-sector 333).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

TABLE A22

Robustness: Leaving out computer equipment manufacturing (NAICS 334)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.109∗ 0.111∗∗∗

(0.0618) (0.0384)

∆ Pred. XME/WAP -1.162∗∗∗ -0.226(0.246) (0.227)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 25,520 24,983 14,758 23,898# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving out com-puter equipment manufacturing (NAICS sub-sector 334).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

83

Page 84: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A23

Robustness: Leaving out electric appliances, accesories, etc. manufacturing (NAICS 335)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.177∗∗ 0.150∗∗∗

(0.0758) (0.0434)

∆ Pred. XME/WAP -1.121∗∗∗ -0.128(0.380) (0.296)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 25,391 24,856 14,694 23,777# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving out elec-tric appliances, accessories, etc. manufacturing (NAICS sub-sector 335).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

TABLE A24

Robustness: Leaving out transport equipment manufacturing (NAICS 336)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.0664 0.127∗

(0.0949) (0.0661)

∆ Pred. XME/WAP -1.540∗∗∗ -0.704∗∗∗

(0.347) (0.250)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 25,410 24,884 14,731 23,801# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving out trans-port equipment manufacturing (NAICS sub-sector 336).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

84

Page 85: Microenterprises and the Lure of Wage Work: Theory and ... · Microenterprises and the Lure of Wage Work: Theory and Evidence from Mexican Export Manufacturing Michael Koelle 18 August

TABLE A25

Robustness: Leaving out furniture, mattress and blinds manufacturing (NAICS 337)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.113∗∗ 0.118∗∗∗

(0.0564) (0.0347)

∆ Pred. XME/WAP -0.972∗∗∗ -0.216(0.232) (0.206)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 24,719 24,214 14,479 23,181# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving out fur-niture, mattresses and blinds manufacturing (NAICS sub-sector 337).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

TABLE A26

Robustness: Leaving out other manufacturing industries (NAICS 339)

(1) (2) (3) (4)

Dependent Variable: Leave Leave to WERelative

Firm GrowthAbsolute

Firm Growth

∆ Pred. XME/WAP 0.108∗ 0.112∗∗∗

(0.0582) (0.0375)

∆ Pred. XME/WAP -1.002∗∗∗ -0.220(0.207) (0.205)

Municipality controls Yes Yes Yes Yes

Individual controls Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes

Locality controls Yes Yes Yes Yes

Dep. var. mean 33.57 17.41 -22.17 3.88

R2 0.07 0.08 0.02 0.05# distinct LLM 25,283 24,749 14,621 23,670# municipalities 1,490 1,477 1,255 1,456# periods 39 39 38 38Observations 727,539 590,176 97,524 475,005

Standard errors in parenthesesIn this table, predicted local export employment is constructed leaving out othermanufacturing industries (NAICS sub-sector 339).

∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

85