www.pse.ens.fr

This paper shows that utility di¤erences between the self-employed and the employees increase with nancial development. This e¤ect is not explained by increased prots but by an increased value of nonmonetary benets, in particular job independence. We interpret these ndings by building a simple occupational choice model in which nancial constraints may impede rmscreation and depress labor demand, thereby pushing some individuals into self-employment for lack of salaried jobs. In this setting, nancial development favors a better matching between individual motivation and occupation, thereby increasing entrepreneurial utility despite increasing competition and so reducing prots. Keywords: Financial development; entrepreneurship; job satisfaction. JEL codes: L26, J20, G20. Sincere thanks to Abhijit Banerjee, Erik Lindqvist and Jörgen Weibull for very helpful discussions. I have also beneted from comments by Cedric Argenton, Andrew Clark, Tore Ellingsen, Fabrice Etilé, Justina Fischer, Thibault Fally, Peter Haan, Andrew Oswald, Thomas Piketty, Asa Rosen, Giancarlo Spagnolo; and from presentations at Stockholm School of Economics, Paris School of Economics, Northwestern University, UNU-Wider, IFN. Financial support from the Knut and Alice Wallenberg Foundation and from Région Ile-de-France is gratefully acknowledged. yE-mail: bianchi@pse.ens.fr 1 ha ls hs -0 05 86 05 9, v er si on 1 14 A pr 2 01 1


Introduction
From a standard economic viewpoint, the choice of becoming an entrepreneur displays some puzzling features. First, it is on average unpro…table: returns to capital are too low and risk too high (Hamilton, 2000;Moskowitz and Vissing-Jorgensen, 2002). Second, it seems to deliver high utility: entrepreneurs often report higher levels of job satisfaction than employees with similar characteristics (Blanch ‡ower and Oswald, 1998;Hundley, 2001;Benz and Frey, 2004). A popular explanation of these puzzles posits that being an entrepreneur gives substantial non-monetary bene…ts and that, due to …nancial barriers to entry, entrepreneurs can enjoy utility above market clearing (Blanch ‡ower and Oswald, 1998).
In this paper, we examine the above argument by exploring both theoretically and empirically how utility di¤erences between entrepreneurs and employees respond to …nancial development. The aim is to contribute to a better understanding of occupational choices, as driven by these utility di¤erences, particularly in relation to market conditions.
Analyzing utility di¤erences is a way to highlight that individuals may become entrepreneurs for very di¤erent reasons. These latter may in turn signi…cantly a¤ect their market behaviors. For example, the type of entrepreneurs, particularly in terms of their motivations and aspirations, is a key predictor of their potential for job creation and growth. 1 Similarly, whether entrepreneurs are driven by push or pull factors is a central determinant of their entry and exit over the business cycle. 2 Hence, understanding entrepreneurs'motivations appears crucial for assessing their contribution to economic development and so ultimately for guiding policy interventions.
We start by building an occupational choice model in which individuals can choose between becoming an entrepreneur, which requires investing capital and hiring workers, or looking for a job as an employee. The model builds on two main ingredients. 3 First, in addition to pro…ts and wages, individuals value also non-monetary dimensions of their job. For example, entrepreneurs may derive utility from being their own boss. 4 In line with the evidence in Fuchs-Schündeln (2009), we assume that individuals may di¤er in how much they like (or dislike) not having a boss, and so more generally in their (intrinsic) motivation for becoming an entrepreneur.
The second key ingredient is that labor demand is determined by the amount of individuals who become entrepreneurs. If entrepreneurs are a few, labor demand is low and so is the probability of …nding a salaried job. This may push some individuals to become entrepreneurs through lack of better opportunities. 5 It follows that individuals may start their businesses with very di¤erent motivations. On the one hand, they may choose to be entrepreneurs, as it is typically the case in more developed countries. On the other, they may become entrepreneurs by necessity. A substantial fraction of entrepreneurs in developing countries falls into this category (Reynolds et al., 2002), and these individuals may be very happy to leave their businesses for a salaried job. 6 We then explore the e¤ects of …nancial development in this setting. While the relation between …nancial constraints and occupational choices has received signi…cant attention (see Du ‡o, 2005 andLevine, 2005 for recent surveys), we focus on the rather unexplored aspect of how …nancial development may a¤ect individual utility, and in particular the non-monetary returns from entrepreneurship. As mentioned above, and as our analysis also con…rms, such returns seem a crucial component of entrepreneurial choices.
In our model, …nancial development allows some poor individuals to access credit and set up a …rm, which in turn increases competition and the demand for labor. In this way, the poor and most motivated individuals can become entrepreneurs, while the rich and least motivated individuals are induced to look for a salaried job. It follows that higher levels of …nancial development are associated with more satis…ed entrepreneurs, and this is the case even if …nancial development increases competition and so reduces pro…ts. In fact, in more …nancially developed countries, individuals tend to have chosen to be entrepreneurs because of their particular motivation rather than for lack of a better job.
These predictions are tested by using individual data on job satisfaction taken from the World Value Surveys, which provide comprehensive household surveys for a large set of countries over two decades. We focus on self-reported levels of job satisfaction in order to account both for monetary and non-monetary returns from a job, which is crucial in our framework since pro…ts and utility need not move in the same direction. Furthermore, in addition to standard demographic variables, these data provide information on beliefs, personality and di¤erent dimensions of individual jobs, which permits to test whether …nancial development works through these channels. Finally, while most of the evidence on entrepreneurs'job satisfaction comes from OECD countries, these data cover a wide sample of developing and developed countries. This allows us to draw a broader picture of whether entrepreneurship has di¤erent meanings, and …nancial development has different e¤ects, according to a country's stage of development.
Our main …ndings lend support to the predictions of the model. First, descriptive statistics show that entrepreneurs report higher levels of job satisfaction than employees only in more …nancially developed countries; moreover, in these countries, entrepreneurs tend to report lower income than employees. These patterns are con…rmed in a more structured analysis in which we control for a set of individual variables and, most importantly, for country-year …xed e¤ects. It emerges that entrepreneurial utility relative to the utility of the employees increases with …nancial development. This result is robust to the inclusion of additional macroeconomic variables, accounting for example for better institutions or economic perspectives, as well as to the use of alternative measures of …nancial development. Moreover, this e¤ect appears stronger in less …nancially developed countries, where many individuals become entrepreneurs by necessity and so many would be happy to switch to a salaried employment.
Finally, we explore the question of which mechanisms may underlie this relation. We …rst note that adding income to the explanatory variables does not change our results. Income appears (as expected) a strong determinant of job satisfaction, but higher …nancial development does not increase entrepreneurs'utility by making them richer. On the other hand, the e¤ect of …nancial development becomes insigni…cant once we control for the degree of independence enjoyed in the job. This suggests that higher …nancial development allows entrepreneurs to enjoy higher non-monetary bene…ts, and in particular higher freedom in taking decisions in their job. We present our model and theoretical analysis in Sections 2 and 3, respectively; Section 4 describes our data and Section 5 reports the empirical results; Section 6 concludes by discussing some policy implications. Omitted proofs and tables are reported in the Appendix.

The Model
Consider an economy populated by a unitary mass of risk-neutral individuals. Each individual is characterized by a type (a; b); where a describes his initial wealth and b his taste for being an entrepreneur (which for now we simply call motivation). Wealth is drawn from a smooth cumulative distribution function F with density f ; motivation from a smooth cumulative distribution function G with density g. These draws are assumed to be statistically independent. In addition, each individual is endowed with one unit of labor, which he may employ either for setting up a …rm or to work as an employee. We now describe these options in further detail.

Options
First, an individual can set up a …rm. We assume that each …rm produces the same homogeneous good and it has the same size: it employs k units of capital, l workers, and it produces q units of output. The pro…t is then = pq wl rk; (1) where p denotes the price of the good, w denotes workers'wage, and r is the market interest rate. In addition, managing a …rm gives utility b. Hence, an individual who sets up a …rm enjoys utility These individuals are called entrepreneurs, and we denote their population share with x 1 . As a second option, an individual can look for a job in one of these …rms. If he is hired, he enjoys utility The population share of workers is denoted with x 2 : If he is not hired, he remains idle and enjoys some utility which we normalize to zero. 7

Markets
There are three markets in our economy: a labor market, a product market and a credit market. In the labor market, the wage w is bounded below by w, which implies that this market may display excess supply. In this case, each applicant has the same probability of getting a job. While the general spirit of the model would be unchanged if we had market clearing wages, we wish to capture the idea that some persons may be pushed into self-employment as a way to avoiding unemployment. As detailed in the Introduction, this appears a prominent case, especially in developing countries. The speci…c 7 Perhaps more sensibly, these individuals may be thought as turning to petty entrepreneurial activities, which require very little capital and no additional worker to operate and provide subsidence levels of production (see the references in footnote 6). This interpretation would reinforce our subsequent results. 5 modelling choice is meant to be minimal with respect to this goal; more sophisticated reasons for non-market clearing wages are given for example in Weiss (1980) and Shapiro and Stiglitz (1984). The number of workers equals …rms'demand, so we have The product market is described by a decreasing inverse demand function where Q = x 1 q denotes the total output produced in the economy. The product demand is here taken as exogenous (say, coming from abroad). Again, this is meant to be the simplest way to model a situation in which a larger share of entrepreneurs increases competition in the product market. While one may also think of positive externalities among …rms, we will show that, even disregarding them, entrepreneurs may report higher payo¤s when more …rms are created. 8 Entrepreneurs take the price p as given, and inelastically supply their output. The …nancial market is competitive, the interest rate r is …xed and exogenous, and we normalize it to one. An individual with wealth a can ask for a loan (k a) in order to set up a …rm: However, ex-post moral hazard limits the maximum size of this loan. Since, at cost c; an individual can renege on his loan contract and run away with the money, the required repayment (k a) cannot exceed c. Hence, only individuals with enough wealth can set up a …rm, and we de…ne this lower bound on wealth as a k c: The threshold a decreases with c, which measures how easy it is to enforce loan contracts and is therefore an indicator of …nancial development. 9

Equilibrium
In equilibrium, each individual, given his type, chooses an option in order to maximize his expected utility and the markets function according to equations (3), (4) and (5). In this equilibrium, an individual with wealth lower than a has no option other than to look for a job as worker. Instead, an individual with wealth greater than a and motivation b prefers to set up a …rm if and only if which implicitly de…nes a lower bound on b as b wl 1 x 1 + rk pq: Provided that an equilibrium exists, the share of entrepreneurs x 1 is implicitly de…ned by This equation also characterizes labor supply (1 x 1 ) and, by equation (3), the share of workers x 2 = lx 1 . We are then interested in identifying the conditions for the existence and uniqueness of an equilibrium in our economy.

Analysis
To show that an equilibrium exists and that it is unique, we …rst note that the right hand side of equation (7) decreases in b : In fact, a higher b leads to a lower share of entrepreneurs x 1 and so to a higher labor supply (1 x 1 ) and to a higher price p (since total output Q increases in x 1 ): This implies that equation (7) uniquely de…nes b : Moreover, the minimal motivation of those who prefer running a …rm increases with the share of entrepreneurs x 1 . In fact, a higher x 1 reduces the incentive to set up a …rm both because it reduces the price p and because it increases the demand for workers and so the probability of being hired. This is expressed in the next Lemma.
Lemma 1 The minimal entrepreneurial motivation b is increasing with the share of entrepreneurs x 1 .
It follows from Lemma 1 that the right hand side of equation (8) decreases in x 1 , and thus equation (8) uniquely de…nes the share of entrepreneurs x 1 : We summarize with the following Proposition.

Financial Development, Pro…ts and Job Satisfaction
We are then interested in analyzing how …nancial development a¤ects utility di¤erences between entrepreneurs and workers. In particular, we consider how these e¤ects may depend on a country's stage of development and how they may di¤er along monetary and non-monetary dimensions of individual utility. 10 The average utility of an entrepreneur can be decomposed into the sum of pro…t = pq wl rk; and average non-monetary bene…t Utility di¤erences are de…ned as Di¤erentiating equation (10) with respect to c; we write the e¤ects of …nancial development on utility di¤erences D as @D @c = @p @c q @w @c (1 + l) + @ b @c : In order to interpret equation (11), we …rst note that, by relaxing wealth constraints, …nancial development allows a higher fraction of individuals to pay the cost of setting up a …rm. The share of entrepreneurs then increases in …nancial development, up to the point at which everyone is employed either as a worker or as an entrepreneur, i.e. x 1 + lx 1 = 1. We show this in the following Lemma.
Lemma 2 There exists a level of …nancial development c such that the share of entrepreneurs x 1 increases in c for c < c and it is x 1 = 1=(1 + l) for all c c : It follows that utility di¤erences between entrepreneurs and workers tend to be higher in more …nancially developed countries. By equation (6) it must be that, for all those who become entrepreneurs, Given Lemma 1, the share of entrepreneurs is low when …nancial development is low. In this case, many individuals choose to be entrepreneurs even if they would prefer to be workers, since labor demand is low and so the probability of being hired is small. 11 Hence, in countries with low …nancial development, entrepreneurship may come from the necessity of …nding a job rather than from the choice of highly motivated individuals. In these countries, then, entrepreneurs need not be more satis…ed with their job than employees. When …nancial development is high, instead, x 1 = 1=(1 + l) and so U 1 U 2 for all those who become entrepreneurs. This implies that utility di¤erences between entrepreneurs and workers are positive.
We then turn to the e¤ect of …nancial development on pro…ts and wages. For c < c ; higher …nancial development increases labor demand, but the wage remains at its minimum w as there is still excess labor supply: Total production also increases (as less individuals end up idle), and this reduces the price p and so the pro…t: For c c ; the share of entrepreneurs is constant, and so is the price, while the wage increases as more people compete to attract workers. This is shown more formally in the next Lemma.
Lemma 3 For c < c ; the price p decreases with c and the wage w is constant at w; for c c ; the price p is constant and the wage w increases with c.
Lastly, we look at the e¤ects of …nancial development on b; which represents the non-monetary dimensions of individual utility. These e¤ects depend on how the minimal motivation b varies with c. For c < c ; b increases both as pro…ts decrease (via product market competition) and as the probability of being hired increases. For c c ; b still increases (though possibly less than for c < c ) since labor market competition increases the wage. Hence, …nancial development allows poor individuals with high motivation to become entrepreneurs and induces those with low motivation to exit and look for jobs as employees. The following Proposition summarizes these predictions, which we test in the next Section.
Proposition 2 a. Entrepreneurs enjoy higher utility than employees only in more …nancially developed countries. b. Entrepreneurial pro…ts decrease with …nancial development. c. Entrepreneurial non-monetary bene…ts b increase with …nancial development, and this e¤ ect may be stronger when …nancial development is low. …nancial development. From a theoretical viewpoint, a positive relation between …nancial and labor market development is what one would expect in models where moral hazard on the part of workers is the reason for imperfect labor markets (see the references in Section 2.2) and moral hazard on the part of borrowers is the reason for imperfect access to credit. From an empirical viewpoint, this is a common theme in the literature on labor market imperfections and self-employment (see Addison and Teixeira, 2003 and the references in footnote 6), and it will be con…rmed in our data too.

Testing the Model
We are interested in exploring the e¤ects of …nancial development on the utility of entrepreneurs relative to workers. In particular, in line with the interpretation suggested by the previous model, we look at the e¤ects of …nancial development both on income and on non-monetary components of individual utility, and we test whether these e¤ects depend on the country's stage of development.
It should already be noted, however, that we are going to estimate the changes in utility within the group of entrepreneurs relative to the group of workers, but the composition of these groups may change with …nancial development. In other words, we do not estimate the e¤ects on the same individuals, but rather the e¤ects on a representative individual within a group over time and across countries.

Data
In most of our analysis, the dependent variable is the self-reported level of job satisfaction. We consider a 1 to 10 index based on the answer to the question: "Overall, how satis…ed or dissatis…ed are you with your job?" This variable is taken from the World Value Surveys (WVS), and is available for 46 countries over the period 1981 2001. 12 We focus on job satisfaction since, as emphasized in the previous analysis, we need an indicator which includes both monetary and non-monetary returns from a job. Indeed, we will see that income is a major determinant of job satisfaction (which suggests we are not capturing purely non-monetary returns), but it is not the only determinant of job satisfaction (which suggests we are not capturing purely monetary returns either).
We are interested in exploring job satisfaction of the self-employed vs. employees. As common in survey studies, we classify an individual as selfemployed if he/she responded that self-employment represents his/her main activity, as opposed to salaried work. 13 The self-employed represent both own-account entrepreneurs and employers, the vast majority being very small businesses. 14 In relation to typical business statistics, which include only …rms beyond some size, this allows a more direct link with occupational choices: even in highly …nancially developed countries, the vast majority of new …rms are very small (Kerr and Nanda, 2009). Moreover, as argued in Blanch ‡ower, Oswald and Stutzer (2001), self-employment is de…ned fairly consistently across countries.
For each individual, information is also provided on demographic characteristics, income, employment status, and several variables describing beliefs, personality and di¤erent dimensions of his or her job. In total, we have 50978 individual observations for full time employees and 7010 for the self-employed, divided into 88 country-year groups.
While …nancial development has a rather precise theoretical de…nition 15 , its measure presents several challenges (see Levine, 2005 for a discussion). In our analysis, we employ the most commonly used indicator in the literature on …nance and growth: the level of domestic credit to the private sector, in percentage of GDP. The variable is taken from the World Development Indicators, published by the World Bank. In our sample, it displays a considerable variation both within and across countries, ranging from 1:68 (Poland, 1989) to 195:98 (Japan, 1990. This indicator seems well suited for our purposes. It re ‡ects the availability of bank credit, which is a fundamental ingredient to facilitate the access to credits for individuals or very small …rms. On the other hand, private credits represent an outcome of …nancial development; hence, we will check the robustness of our analysis with an indicator of inputs of the …nancial system, i.e. privately owned banks. In particular, following Aghion, Fally and Scarpetta (2007), we employ a measure of the percentage of bank deposits held in privately owned banks.
Finally, we use other macroeconomic variables such as per capita GDP, GDP growth, unemployment, minimum wage, regulation, legal origin and trust. A more detailed description and summary statistics of all our variables can be found in the Appendix.

Descriptive Evidence
As suggested by our model, the self-employed need not enjoy greater utility than the employees: in less …nancially developed countries, self-employment can simply be a way to avoid unemployment. To get a …rst idea of where the status of self-employed is a signi…cant determinant of job satisfaction, we estimate the following equation separately for each country and year: The dependent variable U i denotes individual job satisfaction, X i is a set of individual variables including gender, age, age-squared, education, marital status, and SE i is a dummy equal to one if i is self-employed. If the selfemployed enjoy higher utility in a given country and year, then the coe¢ cient should be positive. Table 2 reports the estimates of the coe¢ cient for each country and year. It is clear that the self-employed are not always more satis…ed than the employees, but this tends to be the case only in more developed countries. Moreover, the results remain basically unchanged if income is included in the set of controls X i (columns 4-6). In fact, the set of countries and years in which the self-employed enjoy higher utility becomes slightly larger, which already suggests that income di¤erentials are not the explanation behind di¤erences in job satisfaction.
In order to better highlight these relationships, we construct the following variables. The variable HAP P Y is equal to the estimated coe¢ cient in equation (13), weighted by the inverse of its standard error. We also run a similar regression with income as dependent variable in equation (13). Given this regression, we construct the dummy RICH which is again equal to the estimated coe¢ cient weighted by the inverse of its standard error. 16 As shown in Table 3, the variable HAP P Y is positively correlated with …nancial development, GDP per capita and it is negatively correlated with RICH; the level of unemployment (UNEMPL) and whether there is a mandatory minimum wage (MINWAGE): In accordance with our model, the self-employed enjoy higher utility than the employees in countries with high …nancial development and low labor market imperfections. Moreover, in these countries, the self-employed tend to have a lower income than the employees.

Job Satisfaction and Financial Development
The previous results suggest that utility di¤erences are not due to …nancial market imperfections. We now explore this argument more systematically. We …rst estimate the equation U i;c;t = + X i;c;t + I c;t + F D c;t SE i;c;t + " i;c;t ; where U i;c;t denotes the reported job satisfaction for an individual i in country c and year t; X i;c;t is a set of individual variables including gender, age, age-squared, education, marital status and employment status; I c;t is a country-year dummy, F D c;t is the level …nancial development and SE i;c;t is a dummy equal to one if i is self-employed; …nally, " i;c;t is the error term. 17 Equation (14) follows the spirit of Rajan and Zingales (1998), and it allows to estimate the e¤ect of …nancial development on a particular set of individuals, the self-employed, after having controlled for the e¤ect on the whole population and for country-year …xed e¤ects. Our main interest is in the coe¢ cient ; which describes how …nancial development a¤ects the job satisfaction of the self-employed relative to (full-time) employees. 18 When is positive, we say that …nancial development is positively correlated with entrepreneurial utility. Table 4 reports our estimates on the full sample. The …rst column includes only the controls X i;c;t . Self-employed, old, married and well-educated individuals tend to be more satis…ed with their jobs. The second column describes our most basic speci…cation, as reported in equation (14). The coe¢ cient is positive and statistically signi…cant. Financial development bene…ts the self-employed more than the employees.
In order to check the robustness of this result, we …rst try to identify whether …nancial development is capturing any e¤ect of better macroeconomic conditions, like better institutions or economic perspectives, which may have a di¤erential impact on the self-employed. When we include GDP per capita, interacted with the employment status dummy, the e¤ect of …nancial development becomes slightly weaker, but still highly signi…cant (column 3). Adding other macroeconomic variables like GDP growth (column 4), unemployment (column 5), and an index of regulatory pressure (column 6), always interacted with the self-employment dummy, does not change the estimate of . Hence, our preferred speci…cation, which serves as the baseline for the next analysis, is the one in column (3).
We then check whether this pattern is con…rmed when using a variable based on the percentage of bank deposits held in privately owned banks (BANK), which is a measure of the development of the banking sector. This variable can be employed either as an instrument for …nancial development, as in Aghion et al. (2007) (see Table 5); or as an alternative measure of …nancial development (reduced form). The results in column 7 show that this measure of …nancial development is also positively correlated with entrepreneurial utility.
In addition, despite endogeneity may be less of a concern in our spec-i…cations, in that we estimate how a macro variable a¤ects an individual variable and we include country-year …xed e¤ects, we investigate whether our estimates may be biased, e.g. due to omitted variables, by using instrumental variables. In line with the previous literature, we instrument …nancial development with legal origin (as in several papers, surveyed in Levine, 2005), with the level of trust (as in Guiso, Sapienza and Zingales, 2004) and, as already mentioned, with the variable BANK. While these re-sults should be interpreted with caution, due to well-known issues of …nding valid instruments for …nancial development, we nonetheless notice that they are consistent with the previous estimates (see Table 5). 19 Our next set of regressions estimates whether the e¤ect of …nancial development depends on the country's stage of development. We divide the sample into country-years with high and with low …nancial development, where this threshold is determined by the median value in our sample. 20 The results are in columns (1)-(2) of Table 6: the e¤ects of …nancial development on entrepreneurial utility are positive and signi…cant only in less developed countries.
Our model suggests a possible explanation for this result. In less developed countries, individuals become self-employed either because of their motivation or for lack of salaried jobs. As these countries develop their …nancial system, more jobs are created so only those who value it the most remain self-employed. This composition e¤ect is weaker in more developed countries, where labor demand is higher and so most individuals become self-employed by choice. Indeed, we get similar …ndings if we split the sample according to unemployment (UNEMPL; columns 3-4) or to whether the country has a mandatory minimum wage (MINWAGE; columns 5-6). The e¤ect of …nancial development on entrepreneurial utility is stronger in countries where unemployment is high and there is a minimum wage. 21 Finally, to highlight the nonlinearity in the e¤ects of …nancial development, column (7) includes the level of …nancial development squared and cube. The …rst appears to be negative, the second positive, and both are signi…cant.
From these results, it is evident that the self-employed enjoy higher utility than the employees only in countries with high …nancial development; in less developed countries, entrepreneurial utility increases with …nancial development. In highly developed countries, approximately those above the sample median, the e¤ect of …nancial development is U-shaped, and it appears not statistically signi…cant if one applies a linear model.

Mechanisms
We now explore the mechanisms underlying the relation between …nancial development and entrepreneurial utility. As stressed in our model, these mechanisms should not be evaluated only in monetary terms.
We start by enriching the set of regressors in equation (14). First, we control for income, both in the full sample and separating country-years 1 9 The only di¤erence is that the e¤ect of …nancial development does not appear statistically signi…cant when instrumented by the variable LEGAL alone. This is probably due to the poor …t of this variable in the …rst stage, as shown by the F-statistic.
2 0 Splitting the sample according to the mean gives the same qualitative results. 2 1 Similarly, we …nd that the e¤ects are stronger when GDP per capita is low and labor market regulation (as described by the variable LABOR) is high. according to their level of …nancial development. As shown in columns (1)-(3) of Table 7, if anything, the results are even stronger. Income appears to be a major determinant of job satisfaction; but, as Benz and Frey (2004) have also observed, higher income does not explain entrepreneurial utility. In addition to the existing literature, we document that the e¤ects of …nancial development on entrepreneurial utility are not only monetary. 22 To get further evidence along these lines, we estimate equation (14) with income as dependent variable. Results appear in Table 8. We observe that the self-employed are richer than the employees in less developed countries, while this is not the case in more developed countries (columns 1-2). Moreover, …nancial development decreases the income of the self-employed, relative to the income of the employees (column 3), and this e¤ect tends to be stronger in less developed countries (columns 4-5). The fact that …nancial development reduces pro…ts is consistent with our model in that …nancial development increases competition, either in the product or in the labor market.
The results in columns (4)-(5) of Table 8 and those in columns (1)-(2) of Table 6 are used to draw Figure 1, which summarizes our main results so far. It clearly emerges that the e¤ects of …nancial development on entrepreneurial utility may di¤er from those on pro…t; actually, in our case, these e¤ects go exactly in the opposite direction. Entrepreneurial utility increases with …nancial development, while pro…t decreases. Moreover, both e¤ects tend to be stronger in less developed countries.
The above results suggest that …nancial development works through nonmonetary aspects of job satisfaction. To better identify these mechanisms, we include in our regressions variables such as the degree of pride in one's work, the satisfaction with job security and the degree of independence enjoyed in the job. We also control for work-related beliefs such as how important work is in one's life, the main reason why one works, and so on. None of these variables signi…cantly a¤ects our results, with the exception of independence, which is an indicator derived from the question: "How free are you to make decisions in your job?" The importance of this variable in explaining entrepreneurial utility has already been pointed out in Benz and Frey (2004), and indeed, also in our sample, being self-employed becomes negatively related to job satisfaction when we add this control (Table 7, column 4).
We observe that, once independence has been controlled for, the e¤ect of …nancial development almost halves in magnitude and is not statistically signi…cant (column 5). Hence, our results add to the existing evidence by documenting that most of the e¤ects of …nancial development seem to work Estimates from Table 6, Columns (1)-(2) and Table 8, Columns (4)-(5).
through this channel. According to the model, this is the case because …nancial development o¤ers to the most motivated individuals the opportunity to become entrepreneurs. Indeed, these results suggest that what we have so far called motivation may be (broadly) de…ned in terms of taste for independence at work. Moreover, the coe¢ cient on independence in less developed countries is lower than in more developed countries (columns 5 and 6). 23 This suggests that, as in our model, in more developed countries independence is given to those who value it the most.

Conclusion and Policy Implications
We started our analysis by examining the argument that entrepreneurs enjoy higher utility than employees due to a lack of …nancial development. This argument has not found support in our data; on the contrary, we have shown that …nancial development increases utility di¤erences between the self-employed and employees. Moreover, this e¤ect is not explained by increased pro…ts; rather, it seems to work through non-monetary dimensions of job satisfaction, and in particular independence. We have interpreted these …ndings by building a simple occupational choice model in which …nancial development favors both job creation and a better matching between individual motivation and occupation. According to our results, the existence of utility di¤erences is not due to some market imperfection, and as such it does not in itself call for policy intervention. By highlighting how …nancial development a¤ects also nonmonetary dimensions of entrepreneurial utility, instead, the results point toward other policy implications. First, they bring an additional reason to promote an e¢ cient …nancial system. Second, from the viewpoint of promoting entrepreneurship, they suggest that recognizing the importance of entrepreneurs'intrinsic motivation does not imply that external conditions do not matter. It then appears that a broader investigation on how di¤erent markets and institutions a¤ect non-monetary returns from a job would be of great interest both for researchers and policy makers. Proof. With simple algebra, di¤erentiating equation (7), one can write This expression is positive since the …rst term is positive (note that x 1 may increase only if lx 1 + x 1 < 1; i.e. there is excess labor supply and w = w, which implies that w does not depend directly on x 1 ) and the second term is negative (Q increases in x 1 and so p decreases in x 1 ): Lemma 2 There exists a level of …nancial development c such that the share of entrepreneurs x 1 increases in c for c < c and it is x 1 = 1=(1+l) for all c c : Proof. Suppose …rst that lx 1 + x 1 < 1; i.e. there is excess labor supply and w = w: Implicitly di¤erentiating equation (8), we have The numerator measures the increment in individuals who can a¤ord to become entrepreneurs. The denominator indicates how the mass of individuals who are su¢ ciently motivated and so willing to be entrepreneurs changes as entry increases. Given Lemma 1, @b =@x 1 is positive and hence @x 1 =@c is also positive. Hence, x 1 is strictly increasing in c for lx 1 + x 1 < 1: Let c be the minimal c such that x 1 = 1=(1 + l): Beyond c ; x 1 cannot increase any further since everyone is employed either as a worker or as an entrepreneur.
Lemma 3 For c < c ; the price p decreases with c and the wage w is constant at w; for c c ; the price p is constant and the wage w increases with c.
Proof. Given Lemma 2, x 1 is strictly increasing in c for c < c : The total output produced Q = x 1 q depends positively on x 1 hence for equation (4) the price p decreases with x 1 : The wage w instead does not depend on x 1 ; since x 1 can increase only if there is excess labor supply and so w = w: For c c ; x 1 is …xed and so p is …xed; w instead is such that demand equals supply of labor, i.e. L = 1 x 1 lx 1 0: Implicitly di¤erentiating L, we have @w @c Proposition 2 a. Entrepreneurs enjoy higher utility than employees only in more …nancially developed countries. b. Entrepreneurial pro…ts decrease with …nancial development. c. Entrepreneurial non-monetary bene…ts b increase with …nancial development, and this e¤ ect may be stronger when …nancial development is low.
Proof. Recall from equation (12) that If c is low, then x 1 is low and so there are many entrepreneurs for which U 1 < U 2 : These are the individuals with motivation b 2 [b ; b ]; where b is such that + b = w: If c c ; then lx 1 = 1 x 1 and U 1 U 2 for all entrepreneurs. Part b. of the Proposition follows from Lemma 3: as c increases, either p decreases or w increases, hence the pro…t = pq wl k decreases with c. Finally, from equation (9), we have From equation (7), we see that b increases in x 1 and p and it decreases in w so, given Lemmas 1, 2 and 3, @b =@c > 0. This implies that b increases in c: Notice also that this e¤ect may be stronger for c < c ; when b increases in c both as the result of reduced pro…t and of an higher probability of being hired (an higher lx 1 =(1 x 1 )). For c c ; instead, only the …rst e¤ect is at play.

Description of variables
Individual variables: Job Satisfaction: 1-10 index based on the answer to "Overall, how sat-is…ed or dissatis…ed are you with your job?" 10 indicates "satis…ed", 1 indicates "dissatis…ed". Source: WVS, variable c033. Education: 1-10 index for the age at which the individual completed education. 1 indicates the individual was less than 13 years old, 10 indicates the individual was more than 20 years old. Source: WVS, variable x023r.
Independence in job: 1-10 index based on the answer to "How free are you to make decisions in your job?" 10 indicates "a great deal", 1 indicates "none at all". Source: WVS, variable c034. TRUST: Average response to the question "Generally speaking, would you say that most people can be trusted? " 1 indicates that "most people can be trusted", 0 indicates "Can't be too careful in dealing with people". Source: WVS, variable a165.
LEGAL: Dummy equal 1 if the country's legal origin is civil law. Source: La Porta, Lopez-de Silanes, Shleifer and Vishny (1998).   Note: The table reports partial correlation coe¢ cients. The star indicates sig-ni…cance at the 1% level. HAPPY is equal to the estimated coe¢ cient on selfemployment, weighted by the inverse of its standard error, in an ordered probit regression with job satisfaction as dependent variable. RICH is equal to the estimated coe¢ cient on self-employment, weighted by the inverse of its standard error, in an ordered probit regression with income as dependent variable. All regressions include gender, age, age squared, education, marital status as controls.   All regressions include country-year dummies. In columns (1)-(2), Low FD and High FD indicate that the sample is restricted to countries respectively below and above the median value of FD in our sample (equal to 71.78). Similarly, in columns (3)-(4), High UNEMPL and Low UNEMPL indicate that the sample is restricted to countries respectively above and below the median value of UNEMPL in our sample (equal to 8.2); and in columns (5)-(6), High MINWAGE and Low MINWAGE indicate that the sample is restricted to countries respectively with and without a mandatory minimum wage. The coe¢ cient estimates and the standard errors for FD*SE are multiplied by 100. The coe¢ cient estimates and the standard errors for GDP*SE are multiplied by 1000. Robust standard errors, clustered at the country-employment status level, are in brackets. , and denote rejection of the null hypothesis of the coe¢ cient being equal to 0 at 10%, 5% and 1% signi…cance level, respectively. Note: This table reports the results of OLS regressions with job satisfaction as dependent variable. All regressions include country-year dummies. Low FD and High FD indicate that the sample is restricted to countries respectively below and above the median value of FD in our sample (equal to 71.78). The coe¢ cient estimates and the standard errors for FD*SE are multiplied by 100. The coe¢ cient estimates and the standard errors for GDP*SE are multiplied by 1000. Robust standard errors, clustered at the country-employment status level, are in brackets. , and denote rejection of the null hypothesis of the coe¢ cient being equal to 0 at 10%, 5% and 1% signi…cance level, respectively. High FD indicate that the sample is restricted to countries respectively below and above the median value of FD in our sample (equal to 71.78). The co-e¢ cient estimates and the standard errors for FD*SE are multiplied by 100. The coe¢ cient estimates and the standard errors for GDP*SE are multiplied by 1000. Robust standard errors, clustered at the country-employment status level, are in brackets. , and denote rejection of the null hypothesis of the coe¢ cient being equal to 0 at 10%, 5% and 1% signi…cance level, respectively.