An investigation of the decline in the returns to higher education in Vietnam

ABSTRACT

Vietnam’s economy and education system have had remarkable success in recent decades, yet there are concerns about the declining returns to higher education since 2008. We document this decline in returns to higher education and propose four hypotheses to explain it. Analysis of the VLSS/VHLSS and LFS data provides little evidence for three of four hypotheses. The fourth hypothesis is that changes in the demand for highly educated labor in Vietnam, perhaps due to recent changes in foreign direct investment inflows, rather than the labor supply, are perhaps the most important determinant of the returns to education across different levels.

KEYWORDS:

• Education
• Vietnam
• returns to education
• higher education

Acknowledgements

We would like to thank Nguyet Thi Anh Tran and Harry Patrinos for their thoughtful discussions and comments on early draft manuscripts of this paper. We gratefully acknowledge financial support under the World Bank’s ‘Viet Nam 2045 – Towards a Modern High-Income Economy’ Programmatic Advisory Services and Analytics. The views expressed herein are those of the authors and do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent.

Disclosure statement

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

Notes

1 Vietnam’s most recent data for GNI per capita was $4,010 in current US$ in 2022 (World Bank’s Open Data Catalog). With a growth rate of 5.1% (Vietnam’s average growth rate in real terms of the GNI per capita from 2000 to 2022), Vietnam’s per capita GNI will be $4,655 in 2025, which would make it an upper-middle-income country according to the World Bank’s classifications for FY24 (which define upper-middle-income countries as having a GNI between $4,466 and $13,845).

2 Vietnam’s performance on the PISA in 2015, 2018 and 2022 was still impressive, but not as strong as in 2012. For example, on the 2022 PISA it still outperformed the US in math, but not in reading, and it did not outperform the UK in either subject; on the other hand, it still had the highest performance in both subjects among the 25 developing countries that participated in the PISA in 2022, despite almost all of those countries having a higher per capita GDP. For a detailed analysis of Vietnam’s performance on the PISA in 2012, 2015 and 2018, see Dang et al. (2023).

3 Decree 121/2007/QD-TTg dated July 27, 2007, sets forth the government’s master plan for the network of universities and colleges during the 2006–2020 period. This plan is a component of the comprehensive reform agenda promulgated in the resolution on the Higher Education Reform Agenda (HERA) for the same timeframe, aimed at addressing several challenges within that sector.

4 The total enrolment targets by 2020 was revised downward from 4.5 million to 2.2 million (Decision 37/2013/QD-TTg dated June 26, 2013 on adjusting the master plan on the university and college network during 2006–2020).

5 The LFS was not implemented in 2008. Due to problems with the data (for example, the 2009 LFS is missing data on level of education for 76% of the working age population), we do not use the 2007 and 2009 LFS data.

6 Vietnam’s current (2022) official retirement age is 55 for women and 60 for men. Henceforth, ‘working age population’ refers to women 20–55 years old and men 20–60 years old.

7 The figure drops to about 19% in 1993 when defined as a wage job in the past 7 days, which is what one would expect because everyone who works for a wage in the past 7 days also works for a wage in the past 12 months, but not everyone who works for a wage in the past 12 months also works for a wage in the past 7 days. A similar, though smaller, gap exists when comparing the past 12 months to the past month for the VHLSS data from 2010 to 2020.

8 Most college degrees are three years, but some are two years (Circular No. 04/2022/TT-BLDTBXH dated March 30, 2022).

9 Another commonly used specification replaces Experience with Age. This specification estimates the impact of another year of schooling that includes the indirect impact of reducing work experience by one year. In contrast, the specification in equation (1) estimates the impact of one more year of schooling holding years of work experience constant. The results in this paper on changes in the return to education are very similar for these two specifications.

10 These (private) rates of return to an additional year of education are based on estimates of equation (2). More specifically, Moock, Patrinos, and Venkataraman (2003) estimate the impact on (log) wages of having completed a given level of schooling, relative to having no education. The return to an additional year at any level is the impact of completing a level (relative to the next lowest level), divided by the number of years at that level (e.g., dividing by seven years for secondary).

11 Other studies for this time period have focused on the difference between the returns to education in public sector jobs and private sector jobs; see Imbert (2013) and Phan and Coxhead (2013),

12 The 1993 and 1998 VLSS also asked about employment and (for wage earners only) wages in the past seven days. We do not show estimates based on these data because they are not comparable to the LFS data.

13 One possibility is that higher levels of education lead to more stable employment over one-year periods, but this is less true for one-month periods (the ‘wage’ data are actually employment earnings for the relevant reference period, not wages per hour or per day).

14 The larger coefficient estimates for college and university degrees from the VHLSS data relative to the LFS data reflect the fact that the former are based on annual wage calculations in order to have a consistent reference period for wages for all years of that survey, while the LFS data use wages for the past month. Recall from Figure 4 that the two surveys provide very similar results when the monthly reference period is used. See also footnote 13 for one possible reason why using different reference periods can affect estimates of (relative) returns to education.

15 There is no coefficient for college education because no one in the 1993 VLSS sample had a college level education as the highest level of education.

16 In 2015, a policy shift was introduced that stipulated a mechanism for public universities to decrease their dependence on state funding by adopting cost-sharing measures in exchange for academic autonomy and freedom. This policy has been applied to a select group of research-oriented universities offering training programs appealing to prospective students, allowing them to charge higher tuition fees.

17 One difference is that the much larger sample size leads to significant differences in the declines in the university × experience groups, but they are not widely different. For example, the decline for the 0–5 years of experience interaction is 0.095, while the decline for the 6–10 years interaction is 0.065.