How do female entrepreneurs perform? Evidence from three developing regions

The 27 ECA countries include 16 from Central and Eastern Europe (Albania, Bosnia and Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Former Yugoslavia, the former Yugoslav Republic of Macedonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovak Republic, Slovenia, and Turkey) and 11 from the Commonwealth of Independent States (Armenia, Azerbaijan, Belarus, Georgia, Kazakhstan, Kyrgyzstan, Moldova, Russia, Tajikistan, Ukraine and Uzbekistan). The 13 countries in LA are: Argentina, Bolivia, Chile, Colombia, Ecuador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru and Uruguay. The 22 SSA countries are: Angola, Botswana, Burundi, Burkina Faso, Cameroon, Cape Verde, Gambia, Ghana, Guinea, Guinea Bissau, DRC, Ethiopia, Kenya, Mali, Namibia, Nigeria, Rwanda, Senegal, Swaziland, Tanzania, Uganda, and Zambia.

We acknowledge that there are important differences across countries within each region and that ECA, in particular, is a rather artificial construct. However, these regions are those formally recognized by the World Bank Group and data, analytical work, and policy recommendations are routinely collected, produced, and disseminated for each region as a whole, as well as for individual countries (including, for example, in the Doing Business report).

There is a question as to whether the sample should be weighted to be representative of the relative sizes of these economies or populations. Since the distribution of firms across countries in the pooled analytical sample corresponds quite closely to the countries’ relative size within the region in terms of GDP and population, we do not re-weight our dataset.

In the case of Sub-Saharan Africa, two observations with extremely high values for fixed assets are also dropped.

We decided to discard the oldest, ‘historical’ firms because their performance may be disproportionately affected by brand recognition rather than by the ability of the (current) entrepreneur. This selection, moreover, increases the samples’ comparability across regions.

We are not too concerned about some of the sample selections made, which purely affect the interpretation of the coefficients—for example, our results are only valid for domestic firms that are privately held shareholding companies, partnerships and sole proprietorships, and are younger than 75 years of age. In Sect. 4, we discuss the implications of not being able to determine the owner’s level of involvement in managerial decisions, and we propose a robustness test to explore this issue. In the conclusions we discuss the implications of pooling countries at the regional level. More of concern are the missing data on sex and capital in the relevant sample, whose presence can bias the results in ways that are hard to gauge, a priori. Out of the relevant sample (firms with privately held shareholding companies, sole proprietorship and partnership), those firms with missing data on the gender of the entrepreneur are 1.2% in LA, 1.5% in SSA, and 17.6% in ECA. Firms with missing data on capital are generally a larger portion of the sample, 33.9% in ECA, 37% in LA, and 5% in SSA. Missing data are of concern if the process that generates them is correlated with the dependent variables of interest. We run descriptive and basic regressions (not including capital among the explanatory variables) for the broader sample, and we run the same analysis for the restricted sample with full information on capital. The results remain qualitatively unchanged.

Notice that the percentage change in a dependent variable that has been log-transformed due to a dummy switching from 0 to 1 is 100*[exp(β) − 1].

The latter formulation allows for the inclusion of new firms (those for which X_t−3 = 0); moreover, it “compresses” the very large growth rates that tend to be associated with small firms (because they typically start from a very low amount of sales). Notice that g varies between −2 and +2 and it is monotonically related to G. The two growth rate measures are linked by the identity G = 2g/(2 − g) and are approximately equal for small growth rates. When using g instead of G, the results are essentially the same in terms of significance, but the gap is smaller in LA, −0.067 for sales growth and −0.025 for employment growth, both significant at the 1% confidence level.

Equation 1 can also be interpreted as a first order approximation of more complicated revenue (production) functions. Notice that the production function can be alternatively specified with the dependent variable expressed as total revenue (as in 1) or as value added (i.e. ln(V _i) = ln(Y _i − M _i)), with labor and capital (and not intermediate material) included on the right-hand side.

This argument is reminiscent of the “occupational crowding” literature, arguing that women being crowded into a relatively small number of occupations results in them earning lower wages than men.

In the case of LA, industrial sectors, such as garments, food, and textiles drive this result.

Some researchers suggest that women entrepreneurs choose sectors with either of the following characteristics: (1) where it is easier to combine work with household responsibilities; (2) where women can utilize skills they have mastered as part of their socialization process; (3) which require a small initial investment; (4) where women can easily get credit from suppliers, and (5) sectors for which there is a readily tested, and large market.

Some authors have suggested that entrepreneurs from ethnic minorities and other demographic groups are discouraged from applying for a loan because they fear that their application will be turned down because of poor credit history, prejudice, etc. (Cavalluzzo et al. 2002). Our results for ECA are also coherent with this explanation.

Robustness tests have been carried out running richer specifications of the regressions presented in Table 5, including also variables for the degree of competition in the industrial sector and capacity utilization for ECA and manager experience for SSA; since the results are virtually the same, the more parsimonious specification is presented in Table 5.