How Costly are Labor Gender Gaps? Estimates by Age Group for the Balkans and Turkey

In this paper, survey data are used to document the presence of gender gaps in self-employment, employership, and labor force participation in seven Balkan countries and Turkey. The paper examines the quantitative effects of the gender gaps on aggregate productivity and income per capita in these countries. In the model used to carry out this calculation, agents choose between being workers, self-employed, or employers, and women face several restrictions in the labor market. The data display very large gaps in labor force participation and in the percentage of employers and self-employed in the labor force. In almost all cases, these gaps reveal a clear underrepresentation of women. The calculations show that, on average, the loss associated with these gaps is about 17 percent of income per capita. One-third of this loss is due to distortions in the choice of occupations between men and women. The remaining two-thirds corresponds to the costs associated with gaps in labor force participation. The dimensions of these gender gaps and their associated costs vary considerably across ages groups, with the age bracket 36–50 years being responsible for most of the losses.


Introduction
There is a vast literature documenting the presence of gender gaps in the labor market in both developing and developed countries. Women are often underrepresented in the labor force in general but especially in high-earnings occupations. 1 In this paper we provide some descriptive statistics on the degree of gender inequality in participation the labor market and then use the framework in Cuberes and Teignier (2016) together with survey data to calculate the impact of these gender gaps on income and productivity in several countries of the Balkan Peninsula (Albania, Bosnia-Herzegovina, Croatia, Kosovo, Macedonia, Montenegro, and Serbia) and Turkey. We carry out these calculations for our entire sample and also for specific age groups. The aforementioned countries represent an interesting group to study gender inequality in the labor market because they all belong in the category of middle-income countries and yet they have sizable labor gender gaps. Turkey, for instance, had a labor force gender gap of about 59% in 2007, i.e. out of every 100 men that participated in the labor market, only 41 women did so. We calculate that, for this group of countries, the average gender gap in labor force participation is around 41% in the latest available year, much higher than the one calculated in Cuberes and Teignier (2016) for the rest of European countries (around 16%).
To our knowledge, there are very few papers that quantify the macroeconomic effects of gender gaps in the labor market. 2 The International Labor Organization provides some estimates of the output costs associated with labor gender gaps in the Middle East and Northern Africa but without proposing any specific theoretical model (ILO, 2014). Another shortcoming of this exercise is that it does not allow one to shed light on the mechanisms through which gender gaps in the labor market may affect aggregate efficiency. Cavalcanti and Tavares (2016) construct a growth model based on Galor and Weil (1996) in which there is exogenous wage discrimination against women. Calibrating their model using U.S. data, they find very large effects associated with these wage gaps: a 50 percent increase in the gender wage gap in their model leads to a decrease in income per capita of a quarter of the original output. Their results also suggest that a large fraction of the actual difference in output per capita between the U.S. and other countries is indeed generated by the presence of gender inequality in wages. Hsieh et al. (2013) use a Roy model to estimate the effect of the changing occupational allocation of white women, black men, and black women between 1960 and 2008 on U.S. economic growth and find that the improved allocation of talent within the United States accounts for 17 to 20 percent of growth over this period.
Finally, in the model summarized in Section 3.1 of this paper, Cuberes and Teignier (2016) calculate the macroeconomic effects of gender inequality in the labor market using data from the International Labor Organization for a large sample of countries, but they do not include any of the Balkan countries and they study the case of Turkey only in 2010.
The rest of the paper is organized as follows. Section 2 documents several gender gaps in the labor market of the eight countries considered here. In Section 3 we sketch the general equilibrium occupational choice model proposed by Cuberes and Teignier (2016) and calculate the static costs associated with the gaps in two different years, which vary across countries due to data availability. Section 4 carries out this exercise by different age groups and, finally Section 5 concludes the paper.

Gender Gaps in the Balkans and Turkey Data Description
In this section we provide some descriptive statistics of the labor market in Albania, Bosnia-Herzegovina, Croatia, Kosovo, Macedonia, Montenegro, Serbia, and Turkey for two recent years 3 . Table 1 lists the countries and years used in the paper, along with the number of available observations and their data sources. 4 The data come from household surveys and we only consider individuals at working age i.e. between the ages of 15 and 65. 5 Figures 1 and 2 and tables 2-4 display the evolution of the gender gaps in labor force participation rates, self-employed, and employers for each country over the two years of available data. In terms of Labor Force Participation, the most salient feature is that there is a very large gender gap in labor force participation, with women being underrepresented in the labor market in all countries. This gap is largest in Kosovo, where only 31 women participate in the labor market for every 100 men that do so in 2012. Second, this gap increases 3 We choose two years that are four to six years apart to have some sense of how thse gaps have changed over time. We acknowledge the fact that this time interval overlaps in some cases with the so-called Great Recession that started in 2007 and that these may somewhat distort our calculations. 4 All the variables used in the analysis contain data for the two years specified in this table. The only exception is Montenegro, for which we do not have information on self-employment for the year 2011. LFS stands for "Labor force survey" and ECAPOV stands for World Bank Database of household survey for Eastern Europe and Central Asia. 5 In some cases we also consider the population above 65 years-old, although they represent a very small fraction of the observations. 3    3 Aggregate effects of the gender gaps in the Balkans and Turkey

Theoretical benchmark
In order to help interpret the numerical results of Section 4, we present here a brief summary of the model developed in Cuberes and Teignier (2016). Their paper develops a general equilibrium occupational choice model where agents are endowed with a random entrepreneurship skill, based on which they decide to work as either employers, self-employed, or workers. An employer in this model can produce goods using a span-of-control technology that combines his or her entrepreneurship skills, capital, and workers. This span-ofcontrol element implies that more talented agents run larger firms, as in Lucas (1978). On the other hand, a self-employed agent can produce goods using a similar technology -adjusted by a productivity parameter -but without hiring any workers. Figure 3 displays the payoff of the three occupations at each talent level and shows that in this model agents with the highest entrepreneurship skill (those with a talent equal or larger than z 2 ) optimally choose to become employers, whereas those with the least skill become workers (with a talent lower than z 1 ), leaving the self-employed occupation to agents with intermediate skill 7 levels. Selfemployed

Employers
The model assumes that men and women are identical in terms of their managerial skills. 6 However, women are subject to several exogenous constraints in the labor market.
The first two are that a fraction of women who would like to be employers or self-employed are excluded from these occupations. On top of that, those women who end up becoming workers may receive a lower wage than men. Finally, the model also assumes that a fraction of women are entirely excluded from participating in the labor market. The first three restrictions alter the occupational choice and have general equilibrium implications that allow us to calculate the corresponding fall in income per capita. On the other hand, the restriction on the percentage of women who can participate in the labor market results in a reduction in output per capita. 7 The intuition behind the output loss is as follows. Assume a woman with very good management skills happens to be banned from becoming an employer. The model then implies that a less skilled man will take her position and become the manager of a firm.
But note that, if this man has a lower managerial skill than the woman who is not allowed to become a manager, he will run a smaller firm -due to the nature of the span-of-control technology. This would then have general equilibrium implications in terms of the amount of output produced, wages and firms' profits. In particular, it is easy to show that output/income per worker would be lower in this economy as a result of this restriction (See Cuberes and Teignier (2016) for more details on this.). We measure this loss by taking the ratio between output or income per worker in a country with no gender gaps and the corresponding output or income per worker given the gender gaps that we observe in the data for that country. Table 5 displays the estimated gender gaps and the corresponding income losses (with respect to a counterfactual scenario without gender gaps) for each of the countries in our sample. The third column shows the fraction of females who are excluded from the labor force, while columns 4 and 5 show the fraction of women excluded from employership and self-employment respectively. Column 6 shows the estimated loss in income per capita due to the three frictions, while in column 7 we can see the income loss due to the gender frictions in employership and self-employment. The difference between these last two columns -not displayed here -would correspond to the income loss generated by the gaps in the labor force participation. 8

Numerical results
As we can see in the table, Kosovo in 2012 has the highest income loss, 28.2%, followed by Turkey with a loss of 25.3% in 2007 and 22% in 2013. Croatia is the country with the smallest income loss in both years. In column 7 we can see that Kosovo and Turkey are also the countries displaying the highest income loss due to gender frictions in entrepreneurship. 9 Figure 4 plots the last two columns of table 5 for the first and the last available years.

Costs of gender gaps by age group
The purpose of this section is to to decompose the total income loss into the different age group gaps. As argued in World Bank (2014), gender gaps are likely to vary across age groups and this may have important implications for the future evolution of the labor mar-8 Our paper focuses on gender gaps in entrepreneurship and labor force participation. We don't have very good data on gender wage gaps and therefore we only calibrate the implied wage gap suggested by our model, without using actual data on this gap. In particular, whenever our model predicts too few self-employed women relative to the data, we impose a wage gap between men and women that pushes women towards becoming self-employed rather than workers. As it turns out, however, we only need to impose this wage gap in Croatia in 2009, and the gap is less than 3%. 9 In our exercise we assume that countries do not have wage gaps unless we need to introduce them to match the observed fraction of female self-employed. For the set of countries studied here, only Croatia in 2009 requires the introduction of a wage gap, which is equal to about 2.5% of the wage rate that year. ket and, in particular, the distortions associated with these gaps. Table 6 displays the number of observations by country and age group. 10

Data description
Tables 7-9 report the existing gender gaps in the labor force participation, employers and self-employed for different age groups. We split our sample in four age groups: 15-24, 25-35, 36-50, and 51-65. Table 7 shows that there exist significant gender gaps against women in labor force participation in the majority of age groups. Interestingly, the largest gaps 10 For a study of the changing demographic profile of women in Turkey, see also Dayıoğlu and Kırdar (2010). concentrate in the youngest population (ages 15-24) and the oldest one (ages 51-65   The gender gaps for employers by age are shown in Table 9. As before, most countries, years and age groups show women being underrepresented in this occupation, especially in ages between 15 and 24. In several instances the gaps are 100%, indicating that in our sample there are no female employers at all in these cases. Turkey shows the largest gaps in most groups and years. 11

Methodology: aggregate costs by age group
To carry out the age group decomposition of the total income loss, we first need to calculate a counterfactual set of gender gaps for each age group. In particular, to get the fraction of the total income loss due to a particular age group, we compute what would the aggregate gender gap be if all the gender gaps were zero except for that particular age group.
For example, if we had two age groups, the aggregate employer's gap, denoted by µ, would be defined as follows: where E stands for number of employers and L represents the labor force. The subindexes f, m denote females and males, respectively, while the superindexes 1, 2 represent the first and second age groups, respectively. Finally, the superindex all is the sum of the two groups. We can then rewrite µ in terms of the employers' gap of each group: To compute the income loss due to the age group 1, for instance, we would set µ 2 = 1 in equation (1), while to compute the income loss due to age group 2, we would set µ 1 = 1.
For n groups we would have:  Table 10 shows the fraction of the total income loss due to gender gaps of each age group. In the first row, for instance, we can see the decomposition of the total income loss in Albania for the latest available year: 11% of it is due to gender gaps in the age group 15-24, 23% is due to gender gaps in the age group 25-35, 28% is caused by gender gaps in the age group 36-50 and 38% to gender gaps in the age group 51-65. For most countries and years, the largest fraction of the total income loss comes from the third and fourth age groups, that is from individuals above age 35. The largest average loss is about 41%, which corresponds to the age group between 36 and 50 years-old, followed by a 29% average loss in the age group 51-65, a 21% average loss in the age group 25-35 and a 12% average loss in the age group 15-24.

Numerical results
Figure 5 plots for each country the income loss generated by each of the four age groups in the first and latest available year. The total height is the total income loss due to gender gaps and, as we can also see in Figure 4, Turkey has the largest losses in the first year while Kosovo displays the largest loss in the second year. In the first year, the area for age group 4 is the largest one in Serbia, while the area of age group 3 is the largest one for the rest of countries. In the last year, group 4 represents the largest area in Albania and Croatia, while group 3 is the largest for the rest of the countries.

Concluding remarks
In this paper we calculate gender gaps in labor force participation, employers and selfemployed for several Balkan countries (Albania, Bosnia and Herzegovina, Croatia, Kosovo, Macedonia, Montenegro, and Serbia) and Turkey. Our survey data reveal very large gaps in most countries in the two years studied, which reflect a puzzling underrepresentation of women in these countries given their relatively high income per capita. We then proceed to provide an estimate of the aggregate losses in terms of income per capita associated with this degree of gender inequality using the occupational choice model proposed in Cuberes and Teignier (2016).
Our results suggest sizable income per capita losses. On average, about 17% of income per capita is lost due to the gender gaps studied here. Of this loss, about a third can be attributed to the low numbers of women working as employers and/or self-employed. For comparison purposes, Cuberes and Teignier (2016), using ILO data, obtain an average income loss of 10.5% in Europe, while only Middle East and North Africa, as well as South Asia, have larger average income losses. When we conduct our analysis for different age groups, we find that the small role played in the labor market by women between the ages of 36 and 50 explains the largest fraction of the income losses due to gender gaps in the labor market.
As stated above, in our calculations we assume that men and women have the same distribution of managerial skills. Moreover, we assume that there are no restrictions in the capital market and so capital is assigned to firms in a competitive way. If women had lower skills than men, and/or had less access to capital than men we may be overestimating the effects of the entrepreneurship gender gaps. On the other hand, we may be underestimating them if, for example, gender gaps are larger in sectors with larger span-of-control parameters, i.e., sectors in which managerial ability is more important for production. Quantifying the importance of these biases is left for future research.