5 Self-employment and gender
5.2 The model
the empirical analysis is also explained in detail in Section 5.2.2. Section 5.3 illustrates the data-set and provides a descriptive analysis of the main variables that will be used in the econometric analysis. Finally, Section 5.4 reports the main results, while Section 5.5 draws the main conclusions and gives an indica- tion of potential caveats to the empirical analysis.
a successful entrepreneurial project, by the entrepreneurial effort (e) and by the amount of financial capital available to the individual (K). The higher the ability, the larger the expected income, all else being equal; in the aggregate a large number of self-employed individuals will then be observed. However, the indi- vidual has to consider also the costs incurred when making the self-employment choice. There are three types of costs: (a) the foregone wage income: the higher the current wage income, the less likely the individual is to become an entrepre- neur. Current wage income is influenced by the level of education, so that an increase in education reduces self-employment though it may improve the performance of those who do choose self-employment; (b) the costs of gathering funds to finance the project; and (c) the disutility cost of running a project that may be captured by a negative attitude towards self-employment. It is well doc- umented that non-pecuniary lifestyle preferences (like the desire to be one’s own boss or the desire to pursue not-for-profit objectives) play a significant role in influencing the self-employment decision (Blanchflower and Oswald, 1998).
Let me focus on (b). Under this heading, I group all the costs an individual has to bear to access external finance, among which the costs of persuading the external funders about the viability of the entrepreneurial project. Evans and Jovanovic (1989) argue that a liquidity constraint may occur when there is asymmetric information in the credit market, in the spirit of Stiglitz and Weiss (1981). In this case, entrepreneurs (borrowers) are more informed about both the profitability of the project and their own entrepreneurial ability than lenders, giving rise to problems of moral hazard and adverse selection. Indeed, external funders cannot observe the individual entrepreneurial ability and may attach to the project a different probability of success; therefore they will prefer to ration external funds on the basis of external indicators (gender, ethnic background and area of residence of the applicant), all assumed to be affecting the probability of entrepreneurial success. This implies that indi- viduals applying for external funds either will receive less financial resources than they ask for or will not receive any financial support at all. However, con- sider the case where the potential borrower cannot observe completely his entrepreneurial ability or the profitability of his project (probably because it is his first entrepreneurial project). In this instance, the individual will try to infer the probability of being rationed by looking at aggregate data on the number of individuals that can successfully gather the financial capital for their project. If the proportion of credit-rationed women and individuals of minority back- ground is high, then the potential borrower will anticipate that (s)he will be financially rationed; so (s)he will decide to self-select himself and therefore will not seek external funding.
These considerations suggest three potential models that relate the self- employment choice to both gender and financial constraints. Notice that rather than working with a highly structured model, I estimate reduced-form equations based on a linearization of the assumed probability function. In the first model I will test whether the individual’s probability of becoming self-employed is affected by both (a) gender/ethnic background and (b) by the interaction terms
between finance constraints and gender/ethnicity variables. In other words, the estimation of this model will help to quantify the extent to which finance con- straints are compounded by both the gender and the ethnicity of the respondents.
The empirical specification for this model is quite simple: the dependent variable is a dummy variable (DEP2ST) taking the value of 1 if the respondent is self- employed and 0 otherwise. Among the regressors I include the gender (sex) and/or the ethnic background of the respondent (White/Black/Asian) and an inter- action term between the indicator of finance constraints – here proxied by a dummy variable (Finance Constraints) taking the value 1 if the respondent has experienced finance constraints – and the gender/ethnicity variables. I also control for the respondent’s foregone wage income (proxied by either a variable that indicates whether the respondent has a degree2or his/her employment status), attitudes towards entrepreneurship, location and previous experience. I assume that ceteris paribus the probability of an individual becoming self-employed is affected positively by a subjective positive attitude towards entrepreneurship;
equally the past experience of self-employment will enhance the individual’s probability of becoming self-employed. Econometrically, this model will be esti- mated as a probit model. In formal terms, the equation that considers only the gender is the following:
where xi is the set of control variables. The specifications that focus on the ethnic background are the same with the only difference that instead of gender I control for the ethnic background (White/Black/Asian). One problem with this type of specification is that the probability of experiencing financial constraints may be endogenous and therefore I need to model its determinants. Therefore, the second model I will estimate allows us to test whether the individual’s probability of going for self-employment is jointly conditioned by the respon- dent’s gender and by the applicant’s probability of experiencing finance con- straints (that in turn is affected by an additional set of variables like the availability of collateral, education, location and so on). In other words I want to test the extent to which the probability of being financially constrained affects the individual’s self-employment choice. Econometrically, this is equivalent to controlling for the fact that the probability of experiencing financial constraints is not random but may be dependent on specific factors like the individual characteristics of the applicants (gender, ethnic background and so on) and the availability of one’s own private resources. I will therefore estimate a two-stage Heckman model: in the first stage the respondent’s probability of experiencing finance constraints (Finance Constraints) is affected simultaneously by gender (Sex), ethnic background (White/Black/Asian), location (here measured by the region of residence – Region), education (Degree) and the collateral (Collateral) availability (here proxied by a dummy variable that controls for whether or not the applicant owns a house); in the second stage the probability of becoming self-employed is modelled as a function of gender, the respondent’s foregone
i i i i
i i
i Sex FinCon Sex FinCon x u,
ST
DEP2 =α+β1 +β2 +β3 * +β4 +
wage income (proxied by a variable indicating whether the respondent has a degree),3attitudes towards entrepreneurship, location and previous experience.
In formal terms:
DEP2STi
where DEP2STi1 if FinCon0.
Finally, in the third model, I will test whether a self-selection mechanism is at work in our data so that individuals from either a specific ethnic background or gender may decide not to apply for external funds; also I will estimate the impact that this decision has on the self-employment choice. Again, this model will be estimated by using the Heckman two-stage procedure (see Appendix B for more details on the Heckman estimation procedure) where two equations are estimated: the first equation models the self-selection mechanism where I try to identify the factors that affect the individual’s choice of going for exter- nal finance; so I will model the probability of asking for external funds (DEP1ST) as a function of gender, education and location. The second equation models the self-employment choice and it is estimated only on the sample that
“survives” the self-selection mechanism identified in the first equation. In this equation, I will model the probability of becoming self-employed as a function (among the others) of the previous experience (proxing for the entrepreneurial ability), the employment status (Employment Status) (measuring indirectly the foregone wage income) and the attitudes towards self-employment (as a measure of the disutility cost attached to running an entrepreneurial project). In formal terms:
,