Returns to education in Greece: adjusting to large wage cuts

ABSTRACT

This paper estimates returns to education during a period of sharp wage cuts in Greece, considering both the endogenous nature of education and women’s self-selection. Findings suggest that dramatic wage declines were followed by sharp decreases in returns to education, while the documented convergence of returns between genders is an added benefit. Once endogeneity is examined, using parental education and number of siblings in the household as instruments, and self-selection is accounted for, returns to education almost double compared to OLS. These findings are verified using several robustness tests and alternative specifications.

KEYWORDS:

• Returns to education
• endogeneity
• self-selection
• Greece

JEL CLASSIFICATION:

• I26
• J21
• J31

Acknowledgements

Special thanks to the journal editor Professor Harry Antony Patrinos for his helpful comments, as well as two anonymous referees for their constructive comments. We are also grateful to Costas Kanellopoulos, Kyriakos Pierrakakis and Panos Tsakloglou for their productive feedback and constructive criticism on an earlier draft. We would also like to thank participants of the 2023 INOEK International Conference.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 These numbers from the Labour Force Survey data are available at the Hellenic Statistical Authority (https://www.statistics.gr/en/statistics/-/publication/SJO01/-).

2 Recently, Law 5094/2024 introduced another major reform, allowing, for the first time, the functioning of private higher education via the establishment of branches from foreign non-profit, non-public universities. This could potentially reduce future higher education students’ outflow to other countries.

3 In an over identified model, one can test the validity of over identifying instruments. Here we use the Hansen J statistic.

4 Another approach is to construct bounds for the effect of the endogenous variable using the methodology proposed by Nevo and Rosen (2012). To provide valid results, one key assumption that this approach requires is that the correlation between the years of schooling and the error term from the wage equation has the same sign as the correlation between the instruments and the error term from the wage equation. While this may be true for the paternal and maternal years of schooling, it does apply to the number of siblings in the household, leading to estimated lower and upper bounds crossing.

5 For more information see the original paper of Conley, Hansen, and Rossi (2012), and for an implementation see Clarke and Matta (2018).

6 https://ec.europa.eu/eurostat/databrowser/view/edat_lfse_03__custom_10842848/default/table?lang=en

7 A recent discussion and evidence regarding this phenomenon can be found in Lazaretou (2022).

8 Part of the explanation lies with the decrease in the dispersion of wages. This is clear when comparing the standard deviation of wages by level of education between 2011 and 2019, as reported in Tables A3 and A4.

9 In Greece, until 2012, all married employees were entitled by general national collective agreement to a marriage allowance of 10%. Since then, the marriage allowance is obligatory for those receiving the minimum wage.

10 This sizeable difference between OLS and IV-2SLS estimates is not uncommon. See, for example, Melianova et al. (2021) for Russia, and Heckman and Li (2004) for China.

11 The population density of the area of residence, the presence of children (under 12) or elderly people (75+) in the household, the existence of work experience, as well as the existence of a father with tertiary education, are used in the female participation equation (not shown here) as exclusion restrictions in the calculation of the IMR.

12 See the database for Greece in Montenegro and Patrinos (2023).

13 A third alternative is to use a full Bayesian approach like that of (Hoogerheide, Block, and Thurik (2012) and Hou, Tian and Wang (2020). Both find that when the strict validity assumption is relaxed, the size of the bias is negligible for small to moderate values of the ratio $\delta /\beta$. In our case, the ratio $\delta /\beta$ varies between 0.11 and 0.20, except for paternal education for men which has a value of 0.40.