2.1 – The model

9We consider the after-tax earnings of an individual in full-time employment as an increasing function of schooling. To capture differences in the progressivity of the tax system across countries, we include both average and marginal rates of income tax. Moreover, if unemployed, individuals obtain an unemployment benefits that may be related to their previous earnings and/or to average earnings (it will depend on the unemployment system of the country). We consider the possible changes in the probability of being employed: The higher the education level, the lower the unemployment rate. This is confirmed in the data for all countries independently on the gender. We consider that, while in school, individuals devote a fraction of their time to studying and attending school. Therefore, the potential labour supply of students and their probability of being employed is lower then the one of a full time worker.

10The relationship between fertility rates and education is an important component of our analysis. The (presumed) negative relation is confirmed for most countries, with average fertility rates of 1.56%, 1.28% and 1.09% for low, medium and high levels of education. To evaluate "if" and "how" this affect the private returns to education of women, we explicitly introduce maternity and parental leave and child benefits as follows: We compute the fraction of the working life when the representative woman can spend in maternity leaves. It will depend upon the number of children, if any, and upon the length of (paid or unpaid) maternity leave allowed by law. During this fraction of her active life, a female member of the labour force can, legally, be on maternity leave. During this period, she can be either employed or unemployed. If employed, she will receive a fraction of her previous earning,, plus other benefits related to child care, and typically independent of personal income and depending instead on average income of the country, ?. If unemployed, her income will be determined by the unemployment benefits plus the maternity-related benefits which are, however, independent of employment, ?.

11Finally, schooling also implies direct private costs. Given that, we can compute the present value of expected net lifetime earning. Then, the private rate of return to education (RRE) is the value r such that the average level of education is the optimal solution to the problem of maximizing the present value of the expected net lifetime earnings, for the representative agent, men or women, who studies for S years and retires at time U:

12

13Equation (1) may be easily interpreted: The denominator can be seen as the sum of the marginal opportunity, ?OC, and direct costs of education, ?DC. Similarly, the numerator gives the marginal effect of education on earnings. For men, this effect can be decomposed into two components: One related to the wage profile, ?W, which is driven by the Mincerian parameter, and a second one related to the effect of education on the probability of employment and unemployment benefits on income, ?E. A relevant role in determining ?W is played by the tax system: The more progressive the tax system, the lower the impact of the wage premium on the RRE. In the case of women, there is a third component, ?F, due to the effect of education on fertility and the benefits that a woman obtains if she is out of work because she is on maternity leave. Once we have obtained the values of the right-hand sides of eq (1), the values of r_w and r_m can be estimated by numerical methods (Mendolicchio and Rhein (2011)).

2.2 – Comparing private returns and elasticities

14To calibrate the model and compute RRE, we use the values of the relevant parameters referring to the male and female population in 14 European countries. From Table 1 we can see that for 10 out of 14 countries, r_W is larger than r_M. For men, r_M range, for most countries, between 6.5% and 11%, with an average of 8.71%. The minimum value, 5.53%, is in Sweden, while r_M exceed 12% in Portugal and UK and are over 10% in France and Ireland.

15For most of the countries, r_W lie between 7% and 12%, with an average of 9.50%. They are lower than the average in Sweden (5.70%) and in Netherlands (5.36%). For Ireland, Portugal and UK, the rates are much higher than the average: 15.55%, 12.44% and 13.85%, respectively.

Table 1

Private rates of return on education by gender, and elasticities, E_? and E_?

Country rM% rW% E? E? AUSTRIA 8,02 7,64 - 0.0082 - 0.0301 BELGIUM* 6,43 6,64 - 0.0204 - 0.0252 DENMARK 9,27 6,88 - 0.0859 - 0.0211 FINLAND 8,51 8,30 - 0.0022 - 0.0380 FRANCE 10,29 10,92 - 0.0589 - 0.0290 GERMANY 8,29 10,16 - 0.0198 - 0.0193 GREECE 7,34 9,92 - 0.0076 - 0.0043 IRELAND 11,29 15,55 - 0.0012 - 0.0025 ITALY 6,62 7,95 - 0.0184 - 0.0021 NETHERLANDS 7,13 5.36 - 0.0448 - 0.0130 PORTUGAL 12,12 12,44 - 0.0111 - 0.0121 SPAIN 7,88 9,03 - 0.0075 - 0.0180 SWEDEN 5,53 5,70 - 0.0734 - 0.0407 UK 12,06 13,85 - 0.0017 - 0.0038 AVERAGE Coun. 8,71 9,50 - 0.0206 - 0.0133

Private rates of return on education by gender, and elasticities, E_? and E_?

16The maternity-related policy parameters are quite different across countries. Therefore, it is natural to ask what the effect of their changes on female RRE is. Increases in the values of ? and ?, have a direct effect on the RRE, because they decrease the opportunity cost of maternity, and they have also indirect effects. Here, we will just consider the (larger) direct effects [2]. From the model, by direct computation, we obtain that both derivatives have an undefined sign. About E_?, under weak conditions, the sign is negative, as one would expect, because increases in ? increase the opportunity cost of schooling. The estimated values of E_? are, indeed, always negative. It turns out that E_? is always negative too. This is somewhat counterintuitive, because one would expect a positive value for it, given that an increase in the value of E_? increases the expected future income. However, the opportunity cost of schooling also increases. The impact of a change in the opportunity costs dominates all the others.

17To interpret these results in a proper way, notice that the focus of our analysis is just on one of the possible channels for the effectiveness of these policies. In particular, they could possibly have a quantitatively relevant impact on the choice to participate or not to the labour force. Our approach allows us to correct only partially for this endogenous decision. Moreover, the elasticities computed here measure the impact of a monetary change in the policy parameters. A change in the length of maternity leaves can have different effect.

3.1 – Estimation of the wage equation

21The preliminary ingredient to evaluate the RRE is the estimation of the effect of education on the wage profile. The standard approach has been developed building on Mincer (1974). Our estimates of the Mincerian coefficients are based on the Survey on Household Income and Wealth of 2002 (SHIW), collected by the Bank of Italy. The dataset is based on a random sample of about 8,000 households composed by 22,148 individuals and contains information about households’ financial behavior and fundamental individual socio-demographic characteristics. We restrict the sample to men and women 25-65 years old, fulltime, full-year employees and such that information about earnings are available. The estimation of the wage equation faces a technical difficulty related to endogeneity in the schooling variable, i.e., this variable is not exogenous and it is correlated with the error term in the earning function. The use of Instrumental Variable (IV) estimation is considered the most appropriate solution to this problem. The main difficulty is to find instruments satisfying the requirements [3]. Given the information available in the data, we couldn’t find an IV satisfying the requirements (tests are available on request).

22A second approach to get rid of the endogeneity bias is to introduce a proxy of the unobserved variable in the original regression. The Proxy Variable (PV) should be redundant in the structural equation, i.e., it should be irrelevant for the dependent variable in a conditional mean sense. We use the number of years of schooling of parents, i.e. father and mother education, as PV of the individual’s ability. Given a sample of individuals, denoted by the subscript i, observed at time t, we proceed to estimate the econometric model

23

24where w_i is an earnings measure for individual i, and S_i represents a measure of his/her schooling while x_i is measure of his/her work experience, S_i^F is the education level of individual i’s father, S_i^M the one of his/her mother and D_i^A is a dummy variable (equal to 0 if the individual lives in a rural area, to 1 otherwise). As usual, e_i is a zero mean error term. Most of the estimation details are available on request, while in the text we just report and discuss the main results (see Table 2).

25The education effects evaluated using a PV approach are lower, or equal, to the ones obtained using the OLS. In each area the coefficient of schooling is highly significant. However, they are much less satisfactory at the level of single regions, maybe due to the (occasionally) small size of the selected samples. In order to test the sensitivity of the results, we will calibrate the model using the estimations obtained from both, OLS and PV, approaches.

26A second common problem is that there may be a selection bias related to the lower participation of female to the labour market. Indeed, in Italy the rates of participation are lower for women, especially in the South of Italy. To check for the presence of a selection bias problem, we have introduced a dummy variable, D_i^A, derived using the EU definition of rural area. The intuition is that differences in the structure of the labour markets in the two different kinds of areas should induce differences in the female rates of participation [4]. While the introduction of this additional variable cannot solve the selection bias problem, our results suggest that its actual relevance is probably small. To conclude, we have also estimated the Mincerian coefficients by age cohorts. The differences in the estimated values are quite large. This could probably be explained by the increasing job opportunities for women and the differences in the experience profile across school attainments.

3.2 – Calibration of the model and returns to education

27Calibrating the model, we introduce different policy variables. In particular, we consider two different kinds of benefits. The first one is related to the unemployment status. The rates of unemployment enter eq (1) contributing to determine the weights of ?E. Therefore, significant differences in the rates of unemployment for men and women and by areas could induce differences in the RRE. This problem is potentially important in Italy, due to the dis-homogeneity of the labour market conditions. Indeed, in all the regions, female rates of unemployment are larger than the male ones. In average the difference is 5%. In the regions of the South, it increases to 12%.

28The second group of policy variables is specific for women and include categories: Cash benefits, ?, and maternity leave benefits, ?. We compute the total and by education fertility rate using the SHIW (2002): A negative relationship is confirmed for most of the regions (but Trentino and Toscana), with an average fertility rate of 1.62%, 1.32% and 1.09% for low, medium and high levels of education.

29To include the tax system, we consider average and marginal, national and regional income taxes. We ignore the (fairly small) city additional taxes because of lack of data (see Mendolicchio (2006)).

Table 2

Wage equation estimations and private rate of returns to education^(**),(***)

MEN WOMEN REGIONE Wage premium rM% Wage premium rW% Piemonte 0,037*** 4,25 0,058*** 6,15 Lombardia 0,061*** 7,20 0,075*** 7,74 Trentino 0,056*** 6,62 0,045** 5,30 Veneto 0,053*** 6,28 0,070*** 8,36 Friuli 0,071*** 8,42 0,083*** 9,83 Liguria 0,054*** 6,46 0,060*** 6,33 Emilia 0,063*** 7,44 0,059*** 7,14 Toscana 0,062*** 7,18 0,073*** 8,61 Umbria 0,033*** 3,29 0,066*** 7,50 Marche 0,054*** 6,03 0,060*** 6,83 Lazio 0,057*** 7,02 0,070*** 7,70 Abruzzo 0,035*** 3,56 0,043*** 4,71 Molise 0,075*** 8,47 0,001 - Campania 0,046*** 5,30 0,060 - Puglia 0,043*** 4,76 0,096*** 10,59 Basilicata 0,061** 6,94 0,093** 10,43 Calabria 0,083*** 9,22 0,167*** 17,95 Sicilia 0,047*** 5,64 0,092*** 11,17 Sardegna 0,085*** 9,82 0,058*** 7,73 NORTH 0,055*** 6,59 0,066*** 7,89 CENTER 0,053*** 6,30 0,067*** 7,92 SOUTH 0,050*** 5,75 0,068*** 8,22 ITALY 0,053*** 6,27 0,065*** 7,81

Wage equation estimations and private rate of returns to education^(**),(***)

Significant at 10%
(**) significant at 5%
(***) significant at 1%
Note: The coefficients in italic are not statistically significant.
Valle d’Aosta is excluded because her coefficients are not significant for both genders.

30Table 2 presents the wage equation estimations (PV) and the regional returns, by gender. Our main conclusions are that r_W% are higher for most of the Italian regions; the minimum value for men is in the South, while in this area we observe the maximum return for women. The results show that there are large variations in the effects of education across regions and across genders. While it is beyond the scope of this work to provide an in depth analysis of the factors determining the regional differences in the RRE, and the opposite behavior of men and females rates, we may at least formulate some observations.

31Let’s consider men. To stay on the safe side, we have better to focus the analysis on the largest regions. Among them, the highest RRE are in Emilia, Lombardia and Toscana. In Italy as a whole, men with a lower level of education are more likely to be unemployed. The average rates of unemployment in Italy are 7%, 6% and 3% respectively for low, medium and high levels of education. However, looking at the regional values, this relationships fails to hold in several regions of the North and of the Center (basically, the North-East and the regions along the Adriatic coast), where highly educated individuals actually have a lower probability to be employed. This suggests the presence, in these regions, of a mismatch between labour supply and demand at the high level of education and the possibility of over education phenomena. This is a consequence of the fact that the private industry in Italy has been (and is) characterized by low intensity of education and this has limited the demand for highly educated worker. The picture is different when we consider women: the highest RRE are in the South. The effect of education on the probability of employment is generally positive: The unemployment rate decreases from 14% to 11% and to 7% when education increases. The obvious conjecture is that, for women, relatively high RRE are correlated with lower rates of employment. As already observed, women RRE are typically higher than the ones of men. Moreover, the difference between female and male rates is larger in the South of Italy. The rates of participation of women decrease "geographically" if we move from the North to the South: 55.7%, 50.8% and 36.8% respectively in the North, Center and South, with an average value of 47.9%. This is of interest for two reasons: First, these results could suggest the presence of some selection; second, the higher RRE for women could be due to fact that potentially less productive individuals are screened (or, given the labour market conditions, self-screen themselves) out of the labour force. However, the results that we have presented suggest that the selection bias (if any) in not strong. An alternative explanation could be the rate of employment in the public sector. Up to a very large degree, wages level and wages profiles in the public sector are determined by centralized bargaining, and are not necessarily related to the local labour market characteristics. Given that a significant share of the educated labour force is composed by public employees (for instance, teachers, health professionals, etc.), it is natural to ask if part of the differences in Mincerian coefficients (and, indirectly, in the RRE) can be explained by this factor. To explore this issue, we have run a last set of estimates introducing a dummy variable for the public sector. The relevance of the sector of employment is much higher for women than for men in all the macro areas, even more so in the South of Italy. The coefficient is still higher for women in the South of Italy but they become almost identical at the national level, 5.6% for men and 5.7% for women. his can be reinterpreted as suggesting that the selection bias problem is, in fact, not relevant. Indeed, what it appear to be more important is the composition (by sector of the employer, public vs private) of the labour demand.

3.3 – Counterfactual experiments

32A final exercise is to asses the impact of public policies on the rates of return to education. To do that, we run five "experiments". Firstly, we compute the RRE assuming as counterfactual the absence of unemployment benefits. Even if their effects are not very large, unemployment benefits, by increasing the opportunity cost to be employed, decrease the RRE of about 0.2% for men and 0.4% for women. Not surprisingly, this negative effect is stronger in the South. Secondly, we compute the impact of income taxes, by comparing the actual RRE to the ones obtained if there were no income taxes on wage. Income taxes turn out to decrease the RRE of about 1.2% for men and 1.1% for women. This result is intuitive and its quantitative magnitude is clearly related to the degree of progressivity of the tax system: The more progressive the tax system, the larger the negative impact on RRE. Next, we compute the impact of public financing of education, by comparing the actual RRE with the ones that would prevail if the individuals had to bear the total cost of education. Given the structure of financing of education in Italy, the impact is high: RRE would decrease of 1.7% for men and of 2.6% for women. Considering just women, we also compute the impact of childcare policies, by comparing the actual RRE to the ones computed setting equal to zero cash and maternity benefits: The decrease of the RRE is approximately - 0.064%. Finally, we compute the RRE if there were no public intervention at all. Comparing the RRE of this “basic scenario” with the actual ones, we obtain that the second ones are higher, suggesting that the positive effects of education spending is more important than the negative effect of taxes and unemployment benefits. The full results from the counterfactual experiments are available on request. To sum up, we can conclude that an increase in unemployment benefits has a negative impact on RRE, for both genders. Concerning the tax system variables, an increase in the marginal and average tax rates has a negative impact on RRE. Finally, increase in maternity leaves and childcare benefits imply a weak decrease in women’s RRE. Notice that our analysis considers just one dimension of maternity related policies: The effect on RRE and on their differences across gender. These policies may also have aims which are beyond the scope of this work, for instance to promote an increase in fertility. From this viewpoint, the small values of the elasticities presented above are reassuring in that since they suggest that they can be implemented at a fairly small cost in terms of returns to education, hence of investment in human capital.