The economics of financing higher education

This paper was presented at the 66th Economic Policy Panel Meeting in Brussels, Belgium. This paper has benefited from discussions with André Decoster, Bas Jacobs, Walter Nonneman, Dominik Sachs, Erik Schokkaert, Johannes Spinnewijn, Alain Trannoy, Tom Truyts, Dirk Van Damme, and Frank Vandenbroucke. We also thank audiences at Euroforum, Leuven 2013 and the Canazei 2014 winter school for their helpful comments. We further thank Andrea Ichino, Roberto Galbiati, Andrea Mattozzi, and multiple knowledgeable anonymous referees for their valuable recommendations.

The Managing Editor in charge of this paper was Andrea Ichino.

Economic Policy, Volume 33, Issue 94, April 2018, Pages 265–314, https://doi.org/10.1093/epolic/eiy003

Cite

Ron Diris, Erwin Ooghe, The economics of financing higher education, Economic Policy, Volume 33, Issue 94, April 2018, Pages 265–314, https://doi.org/10.1093/epolic/eiy003

SUMMARY

Different arguments exist pro and contra tax-financed subsidies in higher education. It has been argued that private incentives to study are sufficiently high, while the financing of those subsidies can be deemed regressive as they are co-financed by relatively poorer non-students. Alternatively, several arguments have been put forward why (higher) tax-financed subsidies are desirable, such as externalities and credit constraints. This study scrutinizes these different arguments and discusses the implications for the different ways in which higher education can be financed. Calculations of private returns across the OECD confirm that private incentives to invest in higher education are high. However, economic theory poses that it is the marginal social return that must guide policy, which will reflect both equity and efficiency considerations. We assess the potential regressivity of higher education subsidies using different perspectives, and assess the different efficiency arguments for government intervention in higher education. Tax-financed subsidies turn out to be regressive in most cases, but depending on the perspective and the country under examination. Discussion of the efficiency arguments is focused on those that are most relevant in light of the empirical evidence and their relevance for the different financing modes: externalities, uninsurable risk, credit constraints, and misprediction. We conclude that, to deal with the more credible failures in higher education, tax-financed subsidies are blunt instruments. For a large share of the countries under consideration, shifting towards income-contingent loans or graduate taxes appears more appropriate when taking into account both efficiency and equity considerations.

1. MOTIVATION

Higher education has expanded considerably over the last decades. Participation rates have strongly increased in virtually all developed countries, through both increases in the number of domestic students as well as increased inflows of international students ( OECD, 2016). As a consequence, public spending on higher education has increased since the beginning of the century, across the OECD ( Figure 1).

Trends in spending on higher education

Note: The figure shows public spending on tertiary education as a % of GDP across regions, over the period 1998–2014.

Source: UNESCO World Development Indicators.

The resulting budgetary pressure, reinforced by the recent financial crisis, has led to reforms in the financing of higher education in many developed countries. Tuition fees have been increased in several countries. 1 In response, grants often expanded as well, but have typically not kept up the pace. Consequentially, the take-up of private loans and the default rate on those loans are on the rise, especially in high tuition countries. In some countries, income-contingent loans (ICLs) – loans with repayments that are contingent on future income – have been introduced to provide students with the necessary resources, while limiting the risk of loan default. Finally, several countries have introduced merit or demerit fees, most prominently by introducing financial sanctions for study delays. Taken together, these changes have often shifted a substantial share of the costs and risks towards students and their parents. Yet, tax-financed subsidies to institutions (to keep tuition fees low) and to parents (through either universal or targeted grants) remain the most dominant financing mode.

The recent policy changes and the general expansion of higher education have further intensified the policy discussion on how higher education should be financed. Two main arguments have been provided in support of increases in private contributions. First, tax-financed subsidies are argued to be unnecessary because the high private rate of return on higher education gives sufficient incentives to study. Second, tax-financed subsidies are argued to be perverse because non-students and their families, who are poorer on average than students and their families, co-finance higher education by income taxes. Alternatively, several arguments have been put forward why state intervention is desirable. First, higher education can produce positive externalities, which implies that, compared to the laissez-faire, there is too little investment in higher education from a social point of view. Second, credit constraints can prevent poor but talented students to participate in higher education. Third, the risks that students face during and after higher education are often difficult to insure without government intervention and may therefore cause efficiency losses. Fourth, individuals can take suboptimal/irrational investment decisions in higher education. Such behavioural failures can lead to underinvestment and provide an additional justification for (higher) subsidization.

In this study, we scrutinize these different arguments and analyse what they imply for the desirability of different financing modes. Current policy changes in higher education are predominantly driven by budgetary pressures, and are often reverted again a few years later (e.g., the tuition policies in Austria and Germany). Although ad hoc reforms in the financing of higher education can alleviate some of the budgetary deficits in the short run, there is a definite need to design structural policies that are sustainable in the long run. In order to establish such policies, the different economic arguments spelled out in this study should guide our thinking.

Before discussing these arguments, Section 2 provides a short overview of the current financing of higher education across developed countries. We discuss how the current dominant financing mode – tax-funded subsidies – compares to four alternatives: no government intervention, (secured) classical loans, ICLs, and graduate taxes (GRTs).

In Section 3, we take a closer look at the incentives for investing in higher education by calculating the private rate of return, which combines the private costs and the private benefits of investing in higher education. We compute approximate figures for OECD member countries, which confirm that the rate of return is indeed high on average across the developed world.

Still, whether the private rate of return is high or low does not tell us whether more or less subsidies are needed. According to economic theory, policy should be based on the marginal social return. Adopting a comprehensive social welfare perspective, the marginal social welfare effect of subsidies encompasses equity and efficiency considerations. Sections 4 and 5 discuss the equity and efficiency implications of financing higher education. We argue that the properties of two of the alternative financing modes – ICLs and GRTs – make them especially fit to deal with these problems. As Sections 4 and 5 contain the core message of this paper, we briefly introduce the main findings.

Section 4 deals with equity. As the public costs of higher education are typically funded from general taxes, we compute the fiscal cost that is paid by both the low-skilled and the high-skilled to finance higher education in the OECD member countries. We take two viewpoints: the parental view (parents invest in the higher education of their children, paid by current taxes) and the student view (students invest in their own higher education, repaid by future taxes). The student view supports the claim that subsidies are perverse, because the educational costs of students, who will be richer over their lifetime, is co-financed by the taxes of non-students. In the parental view, tax-financed subsidies can be progressive: even though low-educated parents receive less subsidies (as their children are less likely to attend higher education), they also pay less taxes. Depending on the relative usage and the relative tax contributions of each group, a positive net transfer towards the poor may result in some countries. As countries with high levels of tax-financed subsidies turn out to be regressive in both the parental and student view, we argue that the two alternative financing modes – ICLs and GRTs – would be less perverse, as the costs are borne by students rather than the population at large.

In Section 5, we review efficiency reasons why government intervention might be justified. Although these different reasons are well-understood in theory, the empirical literature is the ultimate judge of their relevance. Findings from the last decades point to three major insights that are important for the financing of higher education.

First, although positive externalities are often invoked to justify (strong) intervention, tedious research shows that it is hard to (causally) identify externalities in higher education. A critical overview of the literature indicates that pecuniary externalities are likely small and cannot, in isolation, justify the current levels of subsidization of higher education. Additionally, the literature suggests that fiscal externalities are also not large enough to recover the subsidy cost.

Second, uninsurable risks and credit constraints, arguments that were largely discarded based on earlier research, turn out to be of increasing importance. This is especially the case in countries where tuition has increased markedly and where students have to resort to classical loans to pay for higher education. Tax-funded subsidies are one way to avoid credit constraints and fare better than classical loans as an insurance device. However, ICLs and GRTs offer credit and insurance as well. In addition, because the loan amount and the GRT rate can be made dependent on study duration, these modes are better fit to deal with moral hazard during higher education. 2

Third, behavioural economics may provide new arguments for intervention. Decision-makers are not always rational and students and their parents turn out to be no exception. In line with hyperbolic discounting, they tend to overestimate (current) costs and underestimate (future) benefits. Misprediction of the private rate of return may lead to too little investment from a social point of view. Tax-funded subsidies are again one way to deal with misprediction. Yet again, ICLs and GRTs provide the same advantages, as they postpone repayment of costs to the future. Moreover, these financing modes are less ‘wasteful’ towards rational students who would participate in higher education anyway.

Section 6 concludes and summarizes the pros and cons of the different financing modes. We discuss the advantages of ICLs and GRTs. Naturally, there are trade-offs involved in comparing the different financing modes. The key question in the financing of higher education can therefore be better described as finding a good balance between the pros and cons. Still, we argue that for a large group of countries – either those with high levels of tax-financed subsidies or those with high tuition levels and classical loans – this balance would become more favourable by introducing GRTs or ICLs. A natural question then arises: why are these modes not yet implemented in these countries? We discuss two complications, related to international student mobility and the political economy of higher education.

The usual caveat applies. We focus on the economics of financing higher education in developed countries, predominantly from a normative point of view. As such, other fields (e.g., sociology), other instruments (e.g., regulation and governance on the supply side), other education levels (e.g., compulsory education), other tasks of higher education institutes (e.g., research), and other countries (e.g., developing countries) are not included or enter the discussion only if deemed essential.

2. FINANCING MODES

In this section we briefly review how higher education is financed in OECD member countries. Because most real-world schemes can be seen as a mixture of different financing modes – taxes, savings, and loans – we describe the characteristics of these financing modes in detail. These characteristics are important when we discuss which financing modes are more appropriate to deal with specific problems in the financing of higher education.

2.1. Financing higher education

The financing of higher education has changed over the last 25 years. Tuition is on the rise in many countries. New financing modes, such as ICLs and (de)merit fees, have been introduced or reinforced in several countries. Altogether, these changes have shifted a larger share of the costs and risks towards students and their families in many countries. Yet, despite these changes, the current financing still relies heavily on public funding in most countries.

Figure 2 shows the total amount spent on institutions of higher education in OECD member countries as a percentage of GDP. The figure excludes R&D expenditures (as we focus on education) and also grants (as these are spent on families rather than institutions). The figures range between 0.6% of GDP in Italy and 2.4% of GDP in the United States. These differences partly reflect differences in the shares of students versus non-students, but are predominantly driven by differences in the amount spent per student. 3

Public and private spending on higher education

Notes: The figure shows public spending and private spending on higher education institutions as a percentage of GDP across OECD countries. Countries are ranked by the sum of the two. The percentage share of public spending is provided at the top of each bar.

Source: Own computations based on OECD (2016).

The total expenditures on institutions in Figure 2 are further subdivided in private expenditures (mainly tuition, but also books and lodging if paid to institutions) and public expenditures. Public spending (on institutions) ranges from 0.5% and 1% of GDP for the majority of countries. Private spending is more heterogeneous, ranging from near 0% up to 1.5% of GDP. The share of public spending (mentioned at the top of each bar) ranges from 32% in Korea to 96% in Finland and Norway. Countries with higher total expenditures tend to have lower public shares.

The share of public funding is at least 90% of total expenditures in Austria and the Nordic countries. In other countries (Australia, Chile, Israel, Japan, Korea, and the United States), it does not exceed 50%. Families have to provide substantial resources in the latter countries, using savings and loans.

Grants – tax-financed subsidies which are either universal or targeted towards poor but talented students and their families – could also help to pay for the costs of higher education. Figure 3 shows total public expenditures as a percentage of GDP, split into subsidies to institutions and subsidies to households. The total public expenditure ranges from 0.5% up to almost 2% of GDP. Spending on families ranges between close to 0% up to 1% and is more heterogeneous compared with spending on institutions. The share spent on families (mentioned at the the top of each bar) ranges from close to 0% (in France and Spain) to more than 40% (in New Zealand, Norway, and Sweden).

Public spending on higher education: on institutions and on households

Notes: The figure shows public spending on higher educational institutions (as also shown in Figure 2) and on households as a percentage of GDP, across OECD countries. Countries are ranked by the sum of the two. The percentage share of spending on households is provided at the top of each bar.

Source: Own computations based on OECD (2016).

2.2. The features of different financing modes

The previous figures show a rather continuous spectrum of how higher education is financed across the OECD. Essentially, the financing of higher education is a mixture of public and private funding; and within public funding there are different uses of public money (subsidies to institutions versus grants to families). If these public subsidies do not suffice for some students to afford higher education, loans can be an alternative. In recent years, multiple countries have introduced ICLs as a new financing mode. Under an ICL, the loan repayment is dependent on income. Such schemes are used in Australia, Hungary, the Netherlands, New Zealand, and the United Kingdom. 4

We have alluded to two other financing modes before: tax-financed subsidies and (secured) classical loans. We elaborate further on these modes and add an additional one: GRTs. For the sake of the discussion later on, the different modes are presented in a highly stylized way.

First, we describe the laissez-faire situation, where financing is based on private funding only. There is no government intervention and students are left to the private market to improve their skills and to take up credit if needed.

Second, the government can provide loans (or secure private market loans). Although loans are private in nature, they often require public expenditures. For example, in a so-called risk-sharing arrangement, loan default is covered by general taxes. The alternative is a risk-pooling arrangement in which borrowers together pay for the loan default, for example, by increasing the interest rate with a risk premium. 5 A key feature is how loans are paid back. If it is a classical loan, then the amount repaid is fixed; if the loan is income-contingent, then the amount is expressed as a percentage of income.

Third, the government can introduce taxes to subsidize higher education institutions and students and their families. Because public funding is usually paid by general taxes, all taxpayers – irrespective of participation in higher education – contribute to the cost of higher education. Although not used in reality so far, GRTs could be used as well: they would differ from general taxes because only participants pay for the cost of higher education. Using a GRT rather than a general tax to finance the same amount of subsidies implies that the total tax rate (the GRT rate to finance higher education plus the general tax rate to finance other expenditures) will be higher for students and lower for non-students.

Table 1 summarizes the different financing modes (in columns) and their most important features (in rows); the laissez-faire is essentially not a financing mode and therefore not mentioned in the table (it rather serves as a benchmark in the discussion later on). We choose loans to be risk-sharing in the default scenario, as they typically are in reality, but will discuss other settings later on when appropriate. 6 Because GRTs are not used so far in reality, there is no benchmark. We assume it has the features of a tax. In addition, it seems reasonable to assume that it is based on participation, rather than graduation. Although we stick to the name, the GRT rate is thus assumed to depend on study duration.

Default characteristics of selected financing modes

Note: GET, general tax; GRT, graduate tax; LOA, (secured) loan; and ICL, income-contingent loan.

Default characteristics of selected financing modes

Note: GET, general tax; GRT, graduate tax; LOA, (secured) loan; and ICL, income-contingent loan.

The two tax modes have some features in common: they are compulsory (row 1), not capped, that is, you can pay back more than your total study costs (row 2), expressed as a percentage of income (row 3), and paid during a lifetime (row 5). The key differences between general and GRTs is that the latter depend on participation (row 4) and are paid only by students (row 6). Consequently, the risks are pooled over students (row 7).

The loan modes are different from the tax modes in several respects: loans are contracted on a voluntary basis (row 1), the amount repaid is capped over the lifetime (row 2), 7 and loans are study-contingent (row 4), as they are taken up to cover yearly study costs. ICLs and classical loans are both loans, but the installment plans differ. Classical loans require repayment of a fixed amount (row 3) during a fixed time period (row 5). In the case of ICLs, the repayment amount varies with income (row 3) and therefore the duration is variable (row 5): shorter if your future income (and thus your repayment) is higher and longer if your future income is lower. Although the duration is thus variable in case of ICLs, there is usually a maximum duration, which implies that incomplete repayment of a loan may occur without breaching the contract. We assume that loans are risk-sharing (row 7), so loan default and incomplete repayment are covered by general tax means. This implies that students mainly, but not exclusively, pay for the costs of higher education (row 6).

3. THE PRIVATE RATE OF RETURN

In this section, we calculate the private costs, private benefits, and the internal rate of return (IRR) to higher education. Because of data availability we restrict our analysis to OECD countries. 8 A more detailed explanation of the computations, and the corresponding data links, can be found in Appendix A.

3.1. The internal rate of return

We compute an IRR, which combines the different (pecuniary) costs and benefits of investing in higher education. Costs consist of direct and upfront investment costs (mainly tuition fees) and the opportunity costs of forgoing earnings in the labour market during the period of study. Figure A1 in Appendix A shows these costs across OECD countries. There is strong variation in the direct (i.e., upfront) costs of higher education, as there are many countries where public subsidies keep tuition fees low, while others demand a substantial private contribution. This variation largely follows the size of the private shares in Figure 2. The direct costs are relatively small compared with the opportunity costs, which are largely a reflection of the standard of living in countries. As such, countries with low direct costs such as Austria, Germany, and Norway, still end up with relatively high total private costs, while a country with high tuition fees like Chile ends up with relatively low total costs. Nonetheless, most countries with high direct contributions also have a high total cost in Figure A1. The total private costs are especially large for the United States.

We use information on gross wages, taxes, unemployment probabilities, and replacement rates to compute the expected (gross and net) income benefit per (labour market) year from completing higher education. Figure A2 in Appendix A shows the figures across OECD countries. Gross benefits are below 40% in the Nordic countries and are also relatively low in Korea, New Zealand, and Australia. The net wage return exceeds 100% for Chile and is also high in most Central European countries, Turkey, Ireland, and the United Kingdom. On average, the highly educated countries with higher gross benefits also pay relatively more taxes, which somewhat reduces the differences across countries if we move from gross to net benefits.

We combine these figures to compute the IRR. The IRR is the rate of return that balances the net present value of current investment costs with the net present value of future income benefits. The formula is provided in Appendix A. Figure 4 shows the IRR across OECD countries. The ranking of countries by the size of the IRR largely follows that of the private benefits of higher education. Countries with high private costs in Figure A1, such as Australia, the Netherlands, and the United States, are situated more to the left compared with Figure A2, but even in these cases the differences are limited. The rate of return exceeds 4% in all countries. This is partly because countries with low net income benefits also have low costs of education (most notably the Nordic countries). Yet, even for a country with relatively high costs and relatively low benefits such as Australia, the IRR is around 7.5%. Hence, although the public and political debate is typically focused on differences in direct private costs across countries, these costs are only a limited share of the total costs, and these total costs are in turn only of limited importance with respect to the variation in the IRR across countries.

Internal rate of return from higher education

Note: The figure shows the internal rate of return on higher education across OECD countries.

Source: Own computations based on OECD data.

3.2. Discussion and implications

OECD (2016) provides similar computations of the rate of return on higher education. There are some differences in the approaches. Mainly, the OECD uses minimum wages for opportunity costs, which are lower than our measures. 9 The averages in our approach are therefore slightly lower (12% in our approach versus 13% for the OECD), but very comparable. 10

Either of these return calculations are based on raw comparisons of pecuniary costs and benefits by educational level. Differences in educational attainment are likely to be selective. In the econometrics literature, estimates of the returns to schooling were originally based on OLS regressions that control for age and experience, and potentially other background indicators. A seminal overview by Psacharopoulos (1994) shows that gross earnings increase by 8% per extra year of higher education in developed countries. This corrected rate of return is somewhat lower than the IRR provided in Figure 4. 11

More recent studies address selection bias more thoroughly and estimate causal returns to education. This causal evidence is predominantly focused on the margin of high school completion. Studies that estimate the causal returns of higher education are relatively scarce. A few exceptions exist, all based on data from the United States. These studies estimate marginal returns that are markedly below the average return to higher education; see, for example, Carneiro et al. (2011).

Education has also been linked to non-pecuniary private benefits. Positive relations have been identified with respect to fertility, occupational choice, and consumption and savings behaviour. 12 Evidence shows that at least a part of these links is causal. Direct causal evidence exists for the effect of higher education on health, in terms of lower obesity, better physical and mental health, and lower incidences of smoking.

To sum up, there is considerable evidence, both correlational and causal, of a substantial pecuniary and non-pecuniary private return on higher education. Students at the margin (i.e., those who are induced to invest more in higher education when, e.g., tuition fees are lowered) appear to have lower (but still positive) pecuniary returns than the average student in higher education.

The private return to higher education is relatively high in many countries. The more sophisticated econometric literature confirms the overall message: there are sufficient incentives to invest in higher education. High average private returns are often used as an argument to reduce tax-financed subsidies to higher education. Because the incentives are so high (and because direct costs are only a limited part of the total costs), it can be argued that reducing subsidies will not affect participation substantially and can reduce the budgetary pressure.

However, economic theory tells us that low or high average private returns do not indicate whether more or less subsidies in higher education are desirable from a societal point of view. What matters is the marginal social return, which includes, but is not restricted to, the private return. Using a comprehensive social welfare perspective, the marginal social welfare gain from subsidizing higher education can differ from the private gain if there are equity gains (by reducing income dispersion) or if there are efficiency gains (by correcting failures). The next two sections will look at both possibilities and discuss the implications for the financing of higher education.

4. PERVERSE REDISTRIBUTION

While the direct costs of higher education are low in many countries, as they heavily rely on public subsidization, someone must ultimately bear the cost of these subsidies as a private person. We call these costs ‘fiscal’ as we assume that the subsidies are financed through earnings taxation. The difference between the fiscal costs and the received subsidies is the so-called net fiscal costs. The net fiscal costs show who gains and who loses from tax-financed subsidies.

We compute the net fiscal costs from two different viewpoints. The parental view assumes that parents are the beneficiaries of the subsidies, but also pay the taxes to finance them. The student view, in contrast, assumes that the students benefit from these subsidies, but pay taxes in the future to finance them. These views are sometimes called the cross-sectional and longitudinal view in the literature. 13

In both approaches, the net fiscal costs are separately computed by educational background. In the parental view, high-skilled parents are more likely to have children who participate in higher education and thus, to receive subsidies. 14 They also pay more taxes because their gross earnings are higher and because taxes are progressive. If the fiscal cost is smaller than the received subsidy, then the net fiscal cost of high-skilled parents is negative, leading to a perverse redistributive effect: low-skilled (poorer) parents subsidize the children of high-skilled (richer) parents.

In the student view, the high-skilled automatically receive all subsidies from higher education, but all future taxpayers pay for these subsidies. While this implies that there is regressivity by construction, there are several reasons why the student view is still worth analysing. First, the degree of regressivity can still differ strongly between countries, depending on the relative tax contributions and on the size of the public subsidies. Second, because different factors are weighted differently, the relative ranking of countries can strongly differ in each view. Third, unless both views lead to the same result for a country, the implications for financing higher education will depend on the adopted view.

Two final issues are worth emphasizing. First, in the exercise, a progressive system means that the current financing of higher education leads to a more progressive distribution than in the absence of subsidization of higher education. If the idea is to use the financing of higher education to increase redistribution, then a more progressive scheme is better. However, if the idea is that the ‘user pays’, then neutral financing schemes are ideal because the subsidy is exactly covered by the taxes paid in each group. This potential contrast only applies to the parental view. In the student view, all schemes are regressive so that a more progressive financing system is automatically also more neutral.

Second, there are limitations to the data we use in this analysis. Most prominently, we have to assume that the subsidy is equal across students of different family background. Especially in countries with targeted financial aid towards disadvantaged students, the computed fiscal costs can overestimate the degree of regressivity (in the parental view). We discuss the sensitivity of our results to this assumption later on.

4.1. Parental view

The budget balances, that is, the total cost of the subsidies received by the cohort of current graduates must be equal to the total fiscal cost paid by the cohort of their parents:

The fiscal cost of the parents is entirely collected via earnings taxation and in such a way that the ratio of the fiscal costs is equal to the ratio of the earnings taxes for both types:

While the first assumption is natural, the second one essentially imposes that the (elasticity of) progressivity remains the same with and without the contribution (as shown in Appendix A). These two assumptions provide two equations that allow for computing the two unknowns: the fiscal costs for each type. The resulting fiscal cost formula has a simple interpretation: it is equal to the tax share of each type (L or H) multiplied with the total cost of higher education.

The net fiscal costs subtract the subsidy amount that is received, the latter depending on the size of the subsidy (s_i), and the number of children enroled in higher education (q_i). As such, the net fiscal costs are equal to:

Figure 5 reports the average net fiscal cost for low-skilled and high-skilled parents in the different countries, ranked by the net fiscal costs of the low skilled. Countries with positive net fiscal costs for the high-skilled automatically have negative fiscal costs for the low skilled. If this is the case, there is subsidization from the (richer) high-skilled towards the (poorer) low-skilled and therefore a progressive system results. This applies to the left of (and including) France. Turkey, France, and Germany are closest to ‘neutral financing’.

Net fiscal cost of higher education (parental view)

Notes: The figure shows the average net fiscal cost for a low-skilled parent and a high-skilled parent for the higher education of their children. Countries are ranked by the net fiscal cost of low-skilled parents.

Source: Own computations based on OECD data.

The tax ratio t_L/t_H and the usage ratio q_L/q_H determine whether the subsidization of higher education is regressive (if the tax ratio is largest) or progressive (if the usage ratio is largest). For example, Korea and the Czech Republic have very similar tax ratios, but the higher relative usage ratio of the low skilled makes subsidization in Korea decidedly more progressive. Chile and the Czech Republic, on the other hand, have very similar usage ratios, but the much lower tax contribution of the low-skilled in Chile leads to strong progressive subsidization. Even when tax ratios and usage ratios are both similar, countries can differ in the degree of progressivity or regressivity (i.e., the height of the bars). For example, Austria and Poland are highly comparable in both tax and usage ratio, but regressivity is stronger in Poland because the absolute levels of q_L and q_H are higher, that is, more people participate in higher education. Differences in the public subsidy level between countries can further add to differences in the height of the bars. Italy represents an interesting case, as the tax ratio greatly exceeds the usage ratio (mainly because of low q_L). However, because the cost per student is low in Italy, the fiscal cost for the average low-skilled parent is still relatively low.

The net fiscal cost of the high skilled is especially high in Chile, the Netherlands, and Ireland. This is mainly because there are large earnings differences between low and high skilled in these countries, which leads to a very low tax ratio t_L/t_H. Regressive subsidization is especially apparent in Scandinavian countries, predominantly because a relatively low wage return to higher education and low tax progressivity lead to a relatively high tax ratio t_L/t_H. This is only partly compensated by their comparatively stronger usage of higher education by children of low educated parents.

While variation in the private return to higher education is an important factor, the figure also shows that countries with higher private contributions, including those with ICLs such as the Netherlands, New Zealand, and the United Kingdom, tend to be more progressive, while countries with almost full tax financing are all regressive. Australia provides an exception here, as it also uses an ICL system but shows regressive subsidization. This is mainly because the private return to education is especially low in Australia, which leads to a relatively low tax contribution of the high skilled.

Because data limitations impose that subsidies are equally high for each student, Figure 5 portrays progressivity of the financing of higher education net of any progressivity in grants. As such, countries in which the targeting of grants to disadvantaged students is common would end up with a more progressive scheme if this would be taken into account. Data show that average grants are especially high in Nordic countries, and also substantial in New Zealand, the United Kingdom, and the United States ( OECD, 2011). Grants in Nordic countries are highly universal and therefore the assumption is reasonable there. For the aforementioned Anglo-Saxon countries, grants are typically targeted so we are likely underestimating progressivity. However, these countries are already progressive in Figure 5, so the general conclusions remain. Additionally, we assess how sensitive results are to targeting. Based on the information from Figure 3, we instead assume that all public funding distributed to households goes to low-skilled households. This is a rather extreme assumption, as not all high-skilled earn high incomes and grants can also be merit based. Therefore, this alternative assumption leads to an upper bound for progressivity. In this exercise, the United States would move virtually to where the Netherlands is situated, while the United Kingdom would become even slightly more progressive than that. If we assume two-thirds of the household funding goes to the low skilled, both countries would move to in between Israel and Canada. Hence, although sensitivity to targeting of grants is not negligible, it does not lead to major shifts in our analysis.

4.2. Student view

The student view assumes that the cohort of current graduates bears the fiscal costs of higher education, but also receives the subsidy. Because only the high skilled in this age cohort obtain a degree, the formulas are the same as in the parental view, but now with usage levels q_L = 0 (non-students) and q_H = 1 (students).

Figure 6 reports the average net fiscal cost for each group across countries. Because non-students do not use higher education, but partially pay for it via future earnings taxes, the net fiscal cost of a non-student is always positive. As non-students are on average poorer than students over their lifetime, subsidization is by definition regressive in the student view.

Net fiscal cost of higher education (student view)

Notes: The figure shows the average net fiscal cost for higher education for the low-skilled (non-students) and the high-skilled (students). Countries are ranked by the net fiscal cost of non-students.

Source: Own computations based on OECD data.

The net fiscal costs are zero for both groups in Chile because the average earnings of the non-students are too low to pay income taxes. 15 Hence, Chilean students pay back their subsidy completely. If we compare the country rankings in Figures 5 and 6, then a positive rank correlation emerges but it is far from perfect (the rank correlation equals around 0.35 for the low skilled and 0.42 for the high skilled). Part of the difference occurs because the usage indicators are equal across countries in the student view. This explains somewhat of the comparatively higher regressivity for countries such as Canada and Finland (who had a high value of q_L/q_H in the parental view) and of the comparatively lower regressivity for countries such as Italy and Turkey (who had a low value of q_L/q_H in the parental view). More importantly, the size of the subsidy becomes a major factor, as it is not spread anymore across both groups but fully incurred by the high skilled. As such, countries with low absolute levels of public spending are relatively less regressive (Italy, Korea, and Turkey), while countries with high public spending become more regressive relative to the previous figure. The shift is most prominent for the Netherlands, which is most progressive in Figure 5, but among the most regressive in Figure 6. The Scandinavian countries are high spenders as well, but are already on the far right in Figure 5. The difference in net fiscal costs between the Scandinavian countries and all other countries is, however, substantially higher in Figure 6.

4.3. Discussion and implications

The academic literature on the regressivity of subsidies to higher education points to several complicating factors that are not addressed in our exercise. First, a prominent argument in the theoretical literature is that the optimal allocation of resources in higher education can be input-regressive; see, for example, De Fraja (2002). We emphasize, however, that regressivity has a different meaning in our context as we do not only look at how public resources are allocated but also how they are financed through the tax system. 16 Second, the empirical literature points out that results can be vulnerable to the presence of general equilibrium effects and non-linear spillovers. These can lead to relative wage increases for the low skilled when participation in higher education expands, which may counter the regressivity of increasing subsidies to higher education. Empirical results on the size of these effects is mixed, however.

The results on perverse redistribution in the more sophisticated econometric literature are largely similar to our computations. A comprehensive overview by Barbaro (2005) concludes that both progressivity and regressivity occur in the parental view. Studies for Austria, Canada, Japan, and the United States find progressive effects, while studies for Quebec, Kenya, and Switzerland find regressive effects. Gruske (1994) is one of the few studies that looks at both views. He shows that the financing of higher education in Germany is progressive in the parental view and regressive in the student view. This is, according to our analysis, still the case 20 years later.

Although the theoretical and empirical literature show that distributional considerations in higher education are complex, the important message is that all interventions in higher education have distributional effects, which should be weighted against other considerations. All financing modes, except tax-financed subsidies, are student-centred. Shifting from tax-financed subsidies to one of the other modes will shift a larger share of the costs towards students and their families. The distributional consequences of such a shift depend therefore on the adopted view. Financing higher education is regressive in the student view, so a shift away from general taxes will make the distribution less perverse. In the parental view, the answer varies. In the case of progressive countries (i.e., countries on the left in Figure 5), the subsidies received by poorer families are larger than the taxes these families pay. Hence, shifting from tax-financed subsidies to one of the other modes will actually lead to a less progressive system in this case.

5. EXTERNALITIES AND BEYOND

In this section, we turn to efficiency arguments pro and contra government intervention in higher education. Such arguments are based on failures at the individual, market, or government level. The aim is not to review this extensive literature, but to focus on what we consider the most important messages of the empirical literature from the last decades: externalities are hard to identify, uninsurable risks and credit constraints are on the rise, and misprediction (of the costs and benefits of higher education) might pose new concerns. An elaborate review of the relevant literature can be found in Online Appendix B.

The different efficiency arguments have implications for how we should ideally finance higher education. We therefore discuss these failures in light of the different financing modes that we introduced in Section 2 and assess which type(s) of financing are most appropriate to counteract the discussed failures.

5.1. Classical externalities

5.1.1. Theory and empirics

A classical externality occurs if an individual’s decision directly affects other individuals in society, but this external effect has not been taken into account by the decision-maker. If positive externalities exist in higher education, then there is too little investment from an efficiency point of view in the laissez-faire. Encouraging higher education could be a desirable policy intervention. To do so, tax-financed subsidies and grants stand out among the different financing modes. Loans and GRTs would be less appropriate. As they still assign the cost of higher education to students, they do not strongly encourage participation. However, this argumentation is only valid if positive externalities in higher education can be identified.

The empirics have been focused on pecuniary externalities. The early macro-economic literature identifies very large externalities, but more recent reviews conclude that these are large overestimations, and that the private benefits outweigh the externalities by far ( Lange and Topel, 2006). These reviews cast doubt on the external benefits of especially higher education. A few studies report comparatively larger returns, but these rely on instrumental variable estimation and it is notoriously difficult to find truly exogenous instruments for the share of higher educated workers in a city or state.

Causal evidence on non-pecuniary externalities from higher education is relatively more robust; see, for example, Lochner (2011) for a focus on causal evidence. Such externalities have been found for voting and trust. Additionally, externalities can be intergenerational; more educated parents ‘produce’ more educated children, but the causal impacts are relatively small. With respect to crime, there is clear evidence of externalities for compulsory education, but there is no evidence for higher education.

Finally, there can be reasons for negative ‘externalities’ as well. 17 These can be caused by information asymmetries (signaling), interactions in preferences (social status and peer pressure), and interactions in constraints (academic peer effects). Because of the same difficulties of identification that plague the literature on positive externalities, the empirical literature in this area is mixed too. Nonetheless, there is credible evidence that signalling occurs in the United States; see, for example, Lang and Kropp (1986).

5.1.2. Implications

While the evidence on non-pecuniary externalities is comparatively robust, the magnitude and scope of such externalities are not overwhelming. Taken together with the limited evidence on pecuniary externalities and the potential for negative externalities, the case for a social return that far exceeds the private one is thin. Heckman and Klenow (1998) claim that, to justify the subsidy levels in the United States, the excess social return (i.e., the difference between the social return and the private return) of college education should be about 12%, which is markedly above even the higher end of the more robust estimates from the literature. Moreover, one can expect that this required excess social return must be even higher in Europe, as the subsidy levels are higher than in the United States. 18 Hence, externalities alone are not enough to justify high levels of tax-financed subsidies and grants for higher education.

5.2. Fiscal externalities

5.2.1. Theory and empirics

Subsidizing higher education is a cost for society. However, because subsidies encourage investment in human capital, they will lead to higher earnings and lower unemployment at the individual level and thus to higher tax revenues and lower welfare benefits at the government level. If the revenue gain is larger than the subsidy cost, then a Pareto improvement is possible. These budgetary windfalls are often referred to as fiscal externalities. 19

If fiscal externalities exist, then policies that encourage investment in higher education, such as tax-financed subsidies and grants, may generate Pareto improvements. Financing modes which do not lower the private cost of higher education, such as loans and GRTs, would be less appropriate. Again, the argument pro tax-financed subsidies is conditional on the empirical evidence.

Several studies have aimed to quantify fiscal externalities. Calculations by, for example, OECD (2016) are based on raw differences in earnings by educational attainment and therefore do not take selection into account. Moreover, they only refer to average rather than marginal effects. They do not tell us whether countries should, from a purely fiscal revenue point of view, subsidize higher education more or less. Some recent papers compute marginal fiscal externalities, using more sophisticated econometrical approaches to address selection issues as well. Some (partial equilibrium) studies indicate that Pareto improvements are possible, provided that subsidies are targeted towards the disadvantaged; see, for example, Lawson (2017). However, these results are very sensitive to allowing for general equilbrium effects, as increased subsidization leads to an increase in college supply, which in turn reduces (private and fiscal) returns to college. Such general equilibrium effects are highly relevant in light of the different financing modes, as each mode is expected to induce different levels of participation.

5.2.2. Implications

At this stage, the empirical evidence for fiscal externalities is limited. Further analysis is especially needed to assess whether the results are different outside the United States and to identify what remains of fiscal externalities in the long run, including general equilibrium effects. For now, fiscal externalities cannot be used as a strong argument to encourage participation by increasing tax-financed subsidies. It is more likely that, at the margin, the cost of increasing subsidies is larger than the resulting gain via higher taxes and lower benefits in most countries, and especially in Europe. If one turns the argument around, reducing tax-financed subsidies will be a revenue gain in most countries, which should be weighted against the higher costs that will accrue to students and their families.

5.3. Beyond externalities

There exist other arguments which could potentially warrant government intervention in higher education. We discuss those that are best supported by current evidence and are most relevant for the policy discussion on the financing of higher education. These failures are: uninsurable risk, credit constraints, and misprediction.

5.3.1. Uninsurable risk

5.3.1.1. Theory and empirics.

Students do not know the costs and benefits of higher education for sure. On the cost side, yearly expenses in higher education are generally observable, but study length is partly uncertain. On the benefit side, uncertainty is likely to be even stronger. Income risks include uncertainty about future wages, employment opportunities, and tax legislation.

Uninsurable risks may create two plausible inefficiency losses. First, there is always a direct loss of welfare if individuals are risk-averse. Second, risk can impede those with positive expected net benefits to enrol in higher education. In the latter case, there is an indirect loss of welfare as the actual level of participation does not coincide with the efficient level (if all risks could be insured). In light of the empirical evidence, we focus on the first inefficiency loss and discuss the (ambiguous) findings of the second in Online Appendix B.

It is important to understand why some risks are difficult to insure in free markets; if not, government intervention is not needed. First, shocks to earnings and unemployment during an economic downturn affect the whole labour market. This collective risk component is difficult to insure in a private market. Second, insuring against graduation risk could lead to moral hazard and adverse selection if there is asymmetric information between insurers and students. Moral hazard would occur if students exert less effort once they are insured against graduation risk. Adverse selection would occur if students with a high success probability find the insurance premium too high and withdraw from the insurance market. Moral hazard and adverse selection lead to inefficiently low insurance levels in private markets.

The key question is whether governments can do better. Given softer budget constraints, governments are better equipped to provide insurance against collective shocks at the cohort level. Adverse selection and moral hazard are caused by asymmetric information and remain problematic. To deal with adverse selection, a government could make the insurance compulsory. Although compulsory insurance schemes in higher education do not exist, some financing modes offer insurance against certain risks. The different financing modes furthermore have different implications for moral hazard, both during studies and during working life. We discuss the empirical evidence on each issue.

There is clear evidence that college graduates face substantial income risk. There are large differences in the returns to higher education, and a substantial part of this difference cannot be predicted by the individual based on private information ( Mazza et al., 2013). Moreover, studies show that students are risk-averse ( Belzil and Leonardi, 2007). Hence, in the presence of both risk and risk-aversion, insurance would provide direct efficiency gains.

The direct welfare gains from insurance could be substantial, but could come at the expense of moral hazard. Causal evidence shows that study effort and performance vary with incentives ( Lavecchia et al., 2016). There is especially strong evidence that delayed graduation reduces substantially if students are financially sanctioned for such delays. Additionally, higher taxes can discourage human capital investment. Direct empirical evidence on the effect of the tax rate and tax progresssivity on enrolment is relatively limited, but suggests a weak relation, especially when general equilibrium effects are taken into account; see, for example, Heckman et al. (1999).

Moral hazard can occur in the workplace as well. College graduates could exert less effort when repayments depend on income. Studies have found that graduates are induced to earn more when they have to rely on a classical loan. There is no evidence on such effects for ICLs. A large literature examines the relation between taxation and work effort. The general conclusion is that the elasticity is low, on average, with respect to labour supply, and modest with respect to taxable income ( Saez et al., 2012). Hence, although there is no consensus on the exact size, most economists would agree that there is a distortion cost of taxation.

5.3.1.2 Implications.

We can conclude from the evidence that insurance can lead to considerable direct efficiency gains, but will also increase moral hazard. The different financing modes provide different degrees of insurance for both income risk and study risk, and consequentially also differ in their vulnerability to moral hazard.

With respect to risk and moral hazard during studies in higher education, a general tax provides the highest insurance, because students do not carry the risk of study failure or delay (they do carry the opportunity costs, but that applies to all schemes). In all the other modes, students do bear the study risk as the costs or repayments depend on study length. Consequently, general taxes are sensitive to moral hazard during the study, while this is much less of an issue in the alternative financing modes. 20

With respect to risk during working life, classical loans require repayments irrespective of future earnings and therefore do not insure against income risks. General taxes provide more insurance than classical loans, as the tax rate is higher. GRTs offer even more insurance for students than general taxes, as they involve a higher total income tax rate for graduates. 21 Whether general taxes provide more insurance against income risk than ICLs depends on the considered time frame. Overall, the level of insurance offered by ICLs is somewhere in between the limited level offered by classical loans (secured or other) and the more generous level offered by general taxes. 22

Moral hazard during working life can be an automatic drawback. The general rule is simple: the more insurance is offered, the higher the potential for moral hazard. GRTs are therefore most likely to suffer from moral hazard, followed by general taxes, ICLs, and classical loans. The trade-off is exactly opposite for non-students. For them, moral hazard and income insurance are high under a general tax, and low under a GRT (as their total tax rate is lower in the latter case).

Two additional points are worth noting. First, we focused on risk, but benefits in terms of consumption smoothing are connected ( Chapman, 2006). The more insurance is offered by a scheme from an ex ante perspective, the more consumption smoothing occurs from an ex post perspective. So, GRTs will offer most smoothing, followed by general taxes, ICLs, and classical loans. Second, if an ICL is risk-pooling rather than risk-sharing, adverse selection may occur. The terminated Yale plan is a typical example. GRTs are not vulnerable to adverse selection no matter whether they are risk-pooling or risk-sharing, because they are compulsory.

To conclude, especially GRTs have favourable features with respect to uncertainty. They offer most insurance against income risk and are not sensitive to moral hazard during studies, two issues that are prominent according to the empirical literature. In addition, GRTs offer high consumption-smoothing and are not sensitive to adverse selection. We cannot say that GRTs are unambiguously optimal, because more insurance always comes at the expense of more moral hazard during working life. From a broader perspective, both GRTs and ICLs allow for more flexibility in dealing with the trade-off between risk and moral hazard, as they allow to differentiate repayment by income and study costs.

5.3.2. Credit constraints

5.3.2.1. Theory and empirics.

Income risk can lead to default risk, that is, the risk of not being able to repay a loan that is used to finance higher education. Default rates of students in higher education are substantial, especially in countries and institutions with high tuition fees and high student debt. Three-year default rates (i.e., the share of students facing repayments that default on their loan within 3 years) reach up to 15% in the United States in recent years ( Lochner and Monge-Naranjo, 2016). Figures from OECD (2016) on the prevalence of loan forgiveness provide suggestive evidence that repayment issues are less prominent in other OECD countries, but these different sources are difficult to compare.

Like income risk, default risk is difficult to insure in a private market. As before, asymmetric information about the default risk of students may cause adverse selection (low risk students are driven out of the market by high risk students) and moral hazard (students exert less work effort to pay back their loan). In addition, students have little collateral to offer to secure their loan in a non-slave society ( Friedman, 1955) and parents are often reluctant to provide security ( Mazzeo, 2007).

The lack of insurance may imply that capital markets do not provide sufficient credit. Especially poor, but otherwise talented students will be financially constrained and refrain from participating in higher education. An inefficiency results: poor and talented students would benefit from higher education, but credit constraints prevent them from participation. Apart from an efficiency loss this could also reduce equality and intergenerational mobility. The most obvious first-best policy intervention is to provide either security or credit, especially for poor and talented students. However, the necessary information on who is needy and talented is not perfectly available and may hinder implementation. We review the most important empirical evidence on credit constraints and discuss the implications for the different financing modes.

5.3.2.2. Empirics.

Correlations between family income and college attendance in the United States are found to be very strong, and increasing over time ( Carneiro and Heckman, 2002). However, these correlations are not necessarily caused by an inability to pay at the time of enrolment. The literature highlights that these gaps are mainly driven by ‘long-run credit constraints’; that is, a lack of means to make investments earlier in life that improve the child’s learning and therefore its future access to higher education.

More recent findings indicate that the share of those that are credit constrained is twice as high in the early 2000s compared with the 1980s in the United States ( Belley and Lochner, 2007), and also twice as large for the combined effect of income and wealth compared with the effect of income alone ( Lochner and Monge-Naranjo, 2011). Credit constraints are likely to be substantially lower in Europe, given the much lower private contributions. Empirical evidence for the United Kingdom (before the introduction of tuition fees) confirms this ( Chowdry et al., 2013). These results also indicate that credit constraints could become a more pressing issue if tuition increases – which are commonly observed among OECD countries in recent years – are not accompanied by complementary policies.

Finally, recent studies have analysed the participation effects of the joint introduction of ICLs with higher tuition fees. Although these policy changes increased the average costs for students, the participation rates of students from low income families in Australia and the United Kingdom have not decreased. 23 In fact, the gap in participation rates between the rich and the poor has even slightly narrowed since the reforms. 24 Hence, the claim that ICLs do not limit access for those that are credit constrained compared with alternative schemes with lower private contributions is confirmed by the empirics. Chapman and Ryan (2005) show that, despite the cost increase, the decrease in the IRR is small under the new ICL scheme in Australia. This confirms our earlier results that the direct costs are not a big driver of the return to higher education (hyperbolic discounting could be another reason for the low effect on participation; see the next subsection). Hence, there are still very large incentives to invest in higher education under these schemes. Azmat and Simion (2017) further show that the reforms in the United Kingdom also had no major impacts on other margins, such as the field of study, quality of the attended college, or dropout from higher education, across socio-economic groups.

5.3.2.3. Implications.

A non-negligible share of students is credit constrained in the short run, mainly when private contributions are high. Tax-financed grants targeted towards poor and talented students can alleviate short-term credit constraints, improving efficiency (and potentially also equity) compared with a system with laissez-faire and classical loans. However, other policies – securing private loans, providing (income-contingent) loans, or using GRTs – will have similar effects.

For addressing long-run credit constraints, the literature suggests that investing in the early life circumstances of the disadvantaged is more effective in raising the educational attainment of the disadvantaged than targeted grants at the time of enrolment ( Heckman and Masterov, 2007).

Finally, while empirical evaluations have shown that the recent introductions of ICLs have not hampered access for students from low-income families, concerns have been raised about the potentially high degrees of default. It is still too early to assess the default rates of these schemes, but they should in any case not be a reason to advocate tax-financed subsidies over an ICL scheme. In the very worst case, when every student would fully default on his loan, the scheme essentially reverts back to a general tax scheme.

5.3.3. Misprediction and psychological mechanisms

We implicitly assumed up to now that individuals behave like a homo economicus, a decision-maker that rationally weighs the costs and benefits of each action before choosing the best one. Over the past decades, behavioural economics has analysed how individuals make decisions in real-world settings. It shows that the rational homo economicus is often not a good model to describe decision-making ( DellaVigna, 2009). The investment decision in higher education made by students and their parents is no exception. 25

The evidence indicates that mispredictions occur. Students tend to overestimate the costs and underestimate the benefits of higher education, especially if they are disadvantaged. Students are also largely unaware of financial aid for which they are eligible. Too little investment in higher education results from an efficiency point of view, which, again, offers a possible reason for encouraging investment through tax-financed subsidies.

What are the underlying reasons for misprediction? First, misprediction could simply be a matter of wrong information or a lack of information. Yet, even if all information is available, bounded rationality may cause too little investment in higher education. Second, deeper psychological reasons could also underly misprediction. In order to design appropriate policies to counteract misprediction, it is important to understand the underlying mechanisms. We discuss the empirical evidence for each potential mechanism in turn.

5.3.3.1. Information and complexity.

If misprediction is a matter of wrong information or a lack of information, then the appropriate policy follows immediately: provide the correct information. However, information provision and updating alone does not seem to be effective for the majority of students, although there are small positive effects for disadvantaged students ( Bettinger et al., 2012).

The complexity of application procedures, in combination with bounded rationality, may explain why providing information only is not sufficient. One possible policy intervention is therefore to simplify the application procedure. Overviews suggest that complex programs are indeed less effective, but the evidence is indirect ( Dynarski and Scott-Clayton, 2013). Combining information provision with direct help in completing the application procedure does appear effective, as does reducing the paperwork load of applications.

5.3.3.2. Non-standard preferences.

Non-standard time preferences, for example, a preference for immediate over delayed utility, are another possible explanation for misprediction. Empirical research shows that discount rates decline over time ( Frederick et al., 2002). In addition, benefits are discounted more than costs. Thus, overweighting the current costs and underweighting the future benefits can irrationally reduce investment in higher education. Direct evidence exists that students indeed strongly underweight future (labour market) benefits relative to current costs ( Cohodes and Goodman, 2014). In relation to the different forms of financing, it can therefore matter greatly whether costs are paid upfront or postponed to the future. Hyperbolic discounting could also explain the earlier discussed results on the effect of the joint introduction of ICLs and higher private contributions in Australia. The fact that participation did not respond to the increase in costs could occur because these postponed costs are not incorporated in the participation decision by myopic students (the decrease in the IRR was small, but not negligible).

5.3.3.3. Non-standard decision-making.

The framing of different choice options can matter as well, even if the choices are otherwise equivalent. A combination of high tuition and high grants can be made financially equivalent to a combination of low tuition and low grants, but tuition (the sticker price) can be more salient for students. If so, too much weight is put on tuition relative to grants, and the second combination leads to more investment in higher education. For the United States, framing effects appear present, although the elasticities strongly differ across studies ( Heller, 1997). Framing effects appear absent in Europe ( Falch and Oosterbeek, 2011). This could be related to the lower complexity of aid compared with the United States. There is also evidence that especially students with lower cognitive ability and students from lower income families are sensitive to framing. This is suggestive evidence that framing is (partly) the result of bounded rationality.

Apart from responding differently to changes in tuition versus grants, students can also have deviating responses to loans because of loan aversion. Loan aversion implies that credit take-up by students is lower if credit is explicitly labelled as a loan. Evidence from experimental studies in hypothetical settings confirms this, but evidence outside of lab environments on elasticities of grants versus (classical) loans is often contradictory ( Heller, 1997). More recent studies in the context of ICLs also provide mixed findings (again, framing appears less strong in Europe compared with the United States). Hence, although the empirical evidence is inconsistent, there are some indications that debt aversion can influence enrolment decisions.

5.3.3.4. Implications.

Research on misprediction and complexity indicates that there is a trade-off between the targeting of financial aid and lack of take-up by those who are eligible. Although certain policy interventions appear effective, they would also be costly when administered on a large scale. This provides an argument against financing schemes that rely on a combination of high tuition and high targeted grants. This is underlined by the evidence in the previous section that other financing schemes such as ICLs and GRTs are at least as effective in addressing the short-run credit constraints of low-income students.

There also exist deeper psychological explanations for misprediction. Given the long-term nature of investment decisions in education, hyperbolic discounting is likely to be a crucial behavioural failure in this context. Postponing a larger share of the upfront costs of higher education to a later date can be effective for students with non-standard time preferences, while causing relatively little harm to rational students.

Hyperbolic discounting could justify financial encouragement of irrational students. Tax-financed subsidies could be used, but they will affect irrational and rational students alike. As such, potential efficiency gains from encouraging irrational students would coincide with efficiency losses for rational students because they are subsidized too much. Given that misprediction appears to be higher among the disadvantaged, providing targeted grants might fare better. However, again, complexity might limit the effectiveness of such an approach. ICLs and GRTs provide an alternative. As these schemes postpone upfront tuition costs to later, irrational students – who underestimate these future costs – will be encouraged to participate, while rational students will internalize these future costs and are therefore not affected. Although the evidence on loan aversion is mixed and such effects appear unlikely to be large, loan aversion can be a reason to prefer GRTs over ICLs in dealing with behavioural failures.

6. CONCLUDING DISCUSSION

This paper has reviewed the economics of financing higher education. We have confirmed that the private incentives to invest in higher education are high, across the OECD (Section 3). Because the social, rather than the private return must guide policy, we have further analysed the equity and efficiency implications of financing higher education. Subsidizing higher education can be regressive, depending on the country and the generational perspective (Section 4). We furthermore have scrutinized the different efficiency arguments for government intervention in the market for higher education (Section 5).

In this final section, we relate the financing modes discussed in Section 2 to the regressivity implications presented in Section 4 and to the main failures identified in Section 5. The previous discussion has highlighted that externalities in higher education as a potential justification for tax-financed subsidies are not convincing. In addition, to deal with the more credible failures in higher education (uninsurable risk, credit constraints, and misprediction), tax-financed subsidies are blunt instruments that do not target the underlying failure. Shifting towards ICLs or GRTs seems more appropriate from both an efficiency and equity point of view. The exact pros and cons of introducing these policy instruments depend on current financing approaches.

On the basis of the analysis in Section 4 on the regressivity of subsidies in higher education, we can distinguish two broad groups of OECD countries. Some countries are regressive in both the parental and student view, while other countries are regressive in the student view, but progressive in the parental view.

Among the first group, the Scandinavian countries are the most prominent example. These countries provide (almost) full subsidization through general taxes, and are highly regressive in both the student and the parental view. Shifting towards ICLs and GRTs would therefore provide insurance and consumption-smoothing for students (even more so in case of GRTs), would reduce regressivity, and would reduce moral hazard during studies. Moreover, suboptimal choices of irrational students caused by hyperbolic discounting would be corrected, without distorting the choices of rational students. The choice between ICL and GRT is a more subtle one. Given the centrality of taxation in this group of countries, GRTs could be more appropriate.

Among the second group of countries, of which the United States is the most prominent example (Canada, Israel, Japan, and Korea also belong to this group), private contributions are high. On the one hand, because tax contributions of the highly educated are very high in the parental view, introducing ICLs or GRTs would reduce progressivity (although not in the student view). It should be noted that it is largely a political question whether less progressivity is a disadvantage in this case, as a ‘neutral’ financing scheme can also be the desired outcome. On the other hand, introducing these financing modes in these countries would avoid the adverse efficiency and equity effects of both credit constraints and hyperbolic discounting by postponing upfront costs. Moreover, research shows that attempts in some of these countries to counteract high tuitions by high grants is often not effective. Finally, evidence from the United States shows that income risk is comparatively high and also increasing in recent years. As such, the benefits of ICLs and (especially) GRTs in terms of insurance and consumption-smoothing are likely to be large as well. To sum up, a shift towards ICLs or GRTs in the second group of countries may reduce progressivity (in the parental view), but also leads to a wide range of benefits. Given the centrality of loans in this group of countries, ICLs are closer to the current practice and therefore more logical to implement than GRTs.

Outside these two broad groups, other countries provide a more ambiguous picture, but the same general principles apply. A country such as Ireland is progressive in the parental view without large upfront costs. Hence, the advantages towards credit constraints and hyperbolic discounting would likely be smaller (but could still exist; concrete evidence on credit constraints in ‘medium tuition countries’ is still lacking). Such countries have to weigh progressivity considerations (including their perceived relevance of the parental versus the student view) against the trade-off in terms of risk and moral hazard of the different schemes. Additionally, most Eastern European countries also have relatively high public funding and are regressive, but less than the Scandinavian countries. Hence, the trade-offs are comparatively more nuanced in these countries.

To be clear, the message is not that ICLs and GRTs are optimal financing tools in every way. Higher insurance comes at the cost of higher moral hazard, and vice versa. The objective is to find a reasonable balance and ICLs and GRTs can contribute to fine-tuning this balance. Nonetheless, our discussion of the pros and cons of the different financing modes has indicated that there are several reasons for countries to adopt such schemes, and this is especially true for countries with either very high or very low shares of public subsidization of higher education.

We conclude by considering two practical issues: the international mobility of students and the political feasibility of implementing these schemes. First, postponed repayment can be complicated when students are internationally mobile. Although the share of international students is still rather low (6% in the OECD), it is rapidly increasing. Moreover, there is a large heterogeneity across countries and degree levels ( OECD, 2016). It should be emphasized that this concern applies to all the financing schemes. Loans, being contractual agreements, might fare better than taxes, but default risks are also likely to be especially high for internationally mobile students. An expected lack of repayment from international students could be tackled by requiring higher tuition fees from (or not offering ICL or GRT schemes to) international students. This can be complicated in certain settings. EU countries, for example, are not allowed to discriminate fees with respect to other EU citizens. Coordination can offer a solution, which seems especially feasible in those same EU countries. Rizzo and Ehrenberg (2004) report on tuition reciprocity agreements between public colleges in different states in the United States. When the flow of students between states is not balanced, then interstate transfers are used to compensate. For European countries, this can be a logical solution, especially when a GRT would be preferred over an ICL.

Second, if ICLs and GRTs could be useful in so many countries, then why are they not yet implemented? The political economy literature has examined which plausible coalitions can arise in favour or against different financing schemes in higher education. Our analysis in Section 4 has shown that, in the student view, (richer) participants in higher education benefit from subsidization, while (poorer) non-participants lose. In the parental view, the story is more ambiguous and depends on the relative usage of higher education versus the relative tax contribution in each group. The literature indicates additional complicating factors. Fernandez and Rogerson (1995) show that a coalition of middle- and high-income families may prefer to keep subsidies for higher education (positive but) low, so that credit constraints prevent the poor from studying, while they still contribute to subsidies via the tax system. They refer to this as ‘exploitation of the poor’. Epple and Romano (1996) show that, in the presence of both public and private higher education, there can be a coalition of the extremes against the middle. The poor would prefer lower subsidization because they participate little in all forms of higher education, while the rich would prefer lower subsidization because they attend private education. Additionally, Del Rey and Racionero (2012) emphasize the important role of risk in support for ICLs, and distinguish not only by wealth but also by ability.

Turning to the first group of (low tuition) countries defined before, these complications do not appear highly relevant. There is very little private higher education in Scandinavian countries, risk is comparatively low, and credit constraints are absent in either the tax-financed system or the ICL/GRT. Furthermore, the system is regressive in both the student and the parental view for these countries, so there is also no ambiguity there. The support for (regressive) tax-financed subsidies versus GRT or ICL will then largely revert back to the simple case of the rich (highly educated) versus the poor (lowly educated). The relatively high share of highly educated in these countries, and their comparatively larger political power, might explain why high levels of subsidization are still in place.

For the second group of (high tuition) countries, credit constraints and private education are present, and income risk is especially high. Additionally, the reference point is different, as the choice is about classical loans versus ICLs or GRTs. In contrast to Epple and Romano (1996)’s comparison between laissez-faire and tax-financed subsidies, private education appears less important here. However, ICLs could end potential exploitation of the poor by relieving credit constraints. 26 Additionally, it is likely that a shift from classical loans to ICLs would increase total take-up of loans (through both higher total participation and through a shift in participation to more expensive colleges), which would increase the total default cost carried by the average taxpayer. 27 For the very wealthy, who gain relatively little from the extra insurance of ICLs, this can be a reason to prefer classical loans over ICLs, even when they participate in higher education. Aside from this complication, (potential) participants in higher education (i.e., the talented) would prefer the ICL, and non-participants would prefer the classical loan. To sum up, although the mechanisms are complex in this group of countries, it is likely that support for an ICL would mainly come from (parents of) the talented poor and middle income families.

Given that support for ICLs and GRTs, especially when compared with tax-financed subsidies, should mainly come from the lower half of the income distribution, it is remarkable that they have little support among leftist parties. This appears mainly driven by concerns about falling enrolment rates and access for students from disadvantaged families in particular (see also the discussion in Murphy et al., 2017). Both these arguments are, as documented in this study, refuted by the empirical evidence.

As a final note, a democracy is not only about voting, but also about deliberation. The arguments in favour of ICLs and GRTs transcend the question which households will gain and lose, and thus, who will (selfishly) vote in favour of or against changes in the financing of higher education. We hope that this paper may help to shape the debate and (in)form opinions, beyond selfish gains and losses, and contribute to a deliberative democracy.

Discussion

Roberto Galbiati

Sciences Po-CNRS

After a decade of slow economic growth, developed economies are facing serious public budget restrictions imposing some control on public spending. At the same time, in most of these economies we observe an increase in the number of students accessing public education. The joint effect of the restrictions in public spending is the stagnation, or in some cases reduction, of per-student public spending in higher education. As far as this reduction in public spending is not followed by a reorganization of the service, the overall quality of higher education can be negatively affected thus implying a decrease in the quality of the human capital provided by higher education. Such a deterioration may itself have a further negative impact on growth. This phenomenon catches the attention of policymakers and academics triggering a debate on the reorganization of the financing modes of higher education. This paper timely addresses this issue and provides the reader with a rich set of information about the pros and cons of different financing modes.

There are two main modes of financing higher education. The first and most diffuse is through transfers to the users (students and families) and the providers (universities) of the service financed through general taxation. The second is based on the direct contribution of the users that bear part of the cost of the service through university fees. This second mode of financing can take various forms, from the direct payment of fees by the students through family support or private loans to income contingent loans and graduate taxes. Both these last tools share the common feature of conditioning the repayment of the debt contracted to finance higher education to the future (post-graduation) repayment capacity of prospective students. The authors address the pros and cons of these different financing modes providing the readers and the policymakers interested in the issue with a useful framework to think about the organization of the financing of higher education. The authors make it clear the reasons why the direct financing of higher education by the state (or the public more in general) is not necessarily a good approach. First, taxes and transfers can have perverse redistributive effects (the poor finance the education of the rich). Second, higher education is an investment with high private returns. Finally, direct public intervention to finance higher education is motivated only by market failures. Thus, the authors discuss in detail the potential market failures involved in higher education providing a rich set of results from the literature and comparing how different financing modes may perform to address the different market failures.

The main message of the paper, at least the one of greater policy relevance is that while both classic loans have a number of drawbacks, income-contingent loans and graduate taxes perform better than traditional tax-financed subsidies over different dimensions. Based on the broad literature surveyed by the authors, they are better suited to deal with market failures. Moreover, they have more favourable equity properties. In most countries, they will decrease the progressivity with respect to tax-financed subsidies while providing credit and insurance against specific risks. The results of the paper are in line with two recent empirical studies about United Kingdom ( Azmat and Simion, 2017; Murphy et al., 2017) assessing the impact of recent reforms. The empirical evidence from United Kingdom shows, in a nutshell, that private financing of higher education has not impacted on the number of enroled students or other relevant distributional dimension, if anything it has improved students' (in particular those from lower income families) conditions by providing them more liquidity.

While we learn a lot from this interesting paper, it also raises some further questions that future research might address and that I list hereafter. The big open question only partially addressed in this paper is why don’t we observe a more widespread diffusion of users’ direct contributions to higher education through university enrolment fees. This is ultimately a political economy question that the authors address in the revised version of the paper in relation of some classic papers (as suggested in the discussion of this paper by Andrea Mattozzi) and that could be further developed in future research. 28

A further aspect that remains unexplored in the paper is the possible effect on segregation of different financing modes. While people care about the exclusion from higher education of low-income students in privately financed systems, in purely publicly financed systems (where private provision of higher education is still possible) a dual system may emerge. In this case students having higher return to education can enrol in privately financed universities while students with lower return would stay in the publicly funded university. In such a scenario, students in publicly funded universities (and in the presence of budget restrictions) would suffer both from lower resources per capita and would not benefit from the interaction with higher quality peers. As shown in Figure 7 this dynamic can be observed in France. Given this possible dynamic, it would be very interesting comparing students’ outcomes in systems at the extreme of the spectrum of financing modes in Europe such as United Kingdom and France.

Governement spending and private/public students composition (France)

Notes: Trends in the per students budget in higher education in France. Author elaboration.

Sources: French Ministry of Budget (Extrait du Budget de la Recherche et Enseignement Supériour; Voarious Years) and French Ministry of Education (Atlas Régional Effectifs D’Etudiants Inscrits).

A final interesting aspect that further research in the field might address is how different financing models affect students’ majors or subject choices. This question is related to the discussion of the behavioural biases such as present biased preferences included in the panel version of the paper and should aim at understanding how fees can be used as a policy tool to promote enrolment in fields where we observe a suboptimal supply of students. While we know something on the effects of students’ fees on the time employed to complete a degree ( Garibaldi et al., 2012) we still know little about the impact of fees and financing modes on subject choices.

Andrea Mattozzi

European University Institute

Summary

This paper provides a comprehensive survey of alternative ways of financing higher education in developed countries. It extensively examines the pros and cons of a number of alternatives based on incentives to invest, equity considerations, and efficiency reasons to justify public intervention in terms of externalities, uninsurable risks, and credit constraints. In general, higher education in OECD countries is financed through a mixture of public and private funds. For a large part, it is financed through taxes and subsidies and with classical loans. There is also a limited use of ICLs and, although not used so far, GRTs could also be an option. Overall the topic is extremely relevant and this paper has the potential to become a classic reference in informing the economic and policy debate on the financing of higher education.

Comments

The comparison between the redistributive aspects of different financing modes of higher education is done from two different perspectives: the ‘parental’ or cross-sectional view, and the ‘student’ or longitudinal view. I found this distinction not quite informative, since the two approaches are not directly comparable. Furthermore, the student view is regressive by construction. The authors argue that the value of considering both approaches rests in that they can lead to very different conclusions and to a different ranking of countries in terms of whether and how much the system is progressive. Hence, in a sense, there is no way to exante choose between these approaches or prefer one to the other. As it is presented in the paper, this argument appears as non-sequitur. It would be useful to go one step further and elaborate on the policy perspective. For example, what are the political feasibility implications of the alternative views?

In discussing the merits of each alternative way of financing higher education, an often overlooked aspect of GRTs is that repayment is affected by students’ international mobility. While the paper mentions this issue, the narrative does not offer any real proposal on how to address it other than referring somewhat vaguely to the enforcement of some form of international agreement. Since in my opinion the main contribution of this paper is to foster a constructive policy debate on the ways in which higher education can and should be financed, advancing proposals to avoid a potentially disruptive erosion of the tax base is a crucial point.

A natural question arising from reading this paper is: If ICL and GRTs have so many upsides and could potentially lead to a welfare improvement in a large number of cases, why is it that no country is using such financing schemes? The obvious constraint that comes to mind is political feasibility. For example, a blocking coalition of middle- and high-income families can prevent increasing subsidies for higher education as in Fernandez and Rogerson (1995). In this way, relatively high-income families defacto exploit the less wealthy. Alternatively, if private and public educations coexist and low income are credit constrained, a coalition of the extremes against the middle can emerge where rich and poor vote for lower subsidies to public higher education (see, e.g., Epple and Romano, 1996). It would be interesting to further elaborate on the political feasibility of alternative ways of financing higher education. In particular, which income/social classes will be willing to vote in favour of ICL and GRTs, and is there a chance they can constitute a majority? It would also be interesting to comment on the historical aversion of left parties to support private financing of higher education and stress how ICL and GRTs would in fact not necessarily lead to a decrease in the enrolment rate.

Panel discussion

Replying to comments from the discussants, R.D. first argued that crime is an important externality that should be considered. He also acknowledged that they should better discuss political feasibility issues even if these are not the focus of the paper, a point that Andrea Ichino also highlighted. Regarding the choice between longitudinal and cross-sectional approaches when computing fiscal costs, R.D. claimed that both are relevant, that is, while the former assumes that parents are the beneficiaries of the subsidies but also pay the taxes to finance these subsidies, the latter considers that the students benefit from these subsidies but pay the taxes based on their future earnings.

In their comments, Uwe Sunde suggested that only examining lump-sum spending on higher education may be misleading, while Tommaso Monacelli wondered if it is surprising that income-contingent loans are a superior financing scheme in a market characterized by asymmetric information, unobserved private returns, and market failures. Related to the latter point, Tommaso Monacelli also asked what can explain the fact that there are only a few countries financing education using such type of loans.

Richard Portes highlighted the lack of discussion about debt accumulation, a severe problem in the United Kingdom and the United States. He also suggested it would be interesting to examine the effect of different funding systems on academic salaries. Mentioning the visionary but abolished experiment conducted at Yale University where students would give a share of their future income stream to the university instead of getting a loan, Saumitra Jha asked how feasible can different types of income-contingent loans be in the long run. In a related comment, George de Menil argued it is crucial to distinguish which form of income-contingent loans or graduate taxes may be the most efficient.

Finally, Ghazala Azmat asked if the authors have considered financial aid. She gave the example of the UK system that reduced financing constraints for the lower part of the social-economic distribution. Related to the first comment of Richard Portes, she also said that some projections suggest that default rates will rise sharply following the latest reform in the United Kingdom that started charging more tuition fees to university students.

APPENDIX A: FORMULAS AND DATA

This appendix elaborates on all the computations carried out in Sections 3 and 4 of this study. All data sources can be found at the end of the Appendix.

Private costs

• The average direct private expenditures per year of higher education (spent towards educational institutions) is computed as the product of two factors: the annual expenditure per student exclusive of R&D costs [Table B1.2 in OECD (2016)] and the share of household expenditure in tertiary education [Table B3.1b in OECD (2016)]. Books, transport, and lodging costs are included when they are provided through the institution, but not when they are provided through other channels.
• The net opportunity costs per year of higher education are based on extrapolated average earnings of those aged 20 years. To extrapolate age-earnings profiles, we rely on Eurostat, as they offer average earnings across a wide range of age categories. We fit the age-earnings profile using data for all available OECD countries in Eurostat and specifying a logarithmic function (which provides the best fit) for those aged below 60 years. We predict earnings at age 20 years and compute the ratio of average age 20 years earnings to total average earnings for the sample as a whole (we take the sample average since we only have European OECD countries available, but the ratios are remarkably consistent across countries). We apply this ratio to compute the age 20 years earnings from the average earnings reported by country by the OECD. We use the OECD tax-benefit calculator to subsequently compute net opportunity costs (using singles without children as family type). Finally, we incorporate the probability of employment by multiplying the net opportunity costs with the probability of not being unemployed, retrieved from Table A5.2 from OECD (2016), and by adding the product of the probability of unemployment and the replacement wage (retrieved from OECD Benefits and Wage Statistics). In other words, the expected net opportunity cost is equal to ( 1 − p ) ( y − t ) + p ρ ( y − t ) ⁠ , with p the probability of unemployment (for the medium-skilled), y − t the net opportunity cost, and ρ the net replacement rate.

The private costs across OECD countries, separately for direct upfront costs and opportunity costs, are portrayed in Figure A1.

Private benefits

• The gross earnings are computed by combining two data sources: the low-skilled to high-skilled and the medium-skilled to high-skilled gross earnings ratios for the 25–64 age cohort [reported in Table A6.1 in OECD (2015)] and the average gross earnings for the 25–64 age cohort (reported via OECD stats based on the Labour Force Survey). Combined with the numbers of individuals in each group (n_L, n_M, and n_H) we obtain three equations and three unknowns which allow us to solve for the gross earnings in each group (of which we only use y_M and y_H here).

• The earnings tax levels (consisting of both income tax and social security contributions) are computed by applying the online OECD tax and benefits calculator to the gross earnings of the medium skilled and the high skilled. As taxes depend on family situation, we compute a weighted average of the four available family forms (single without children, single with two children, married without children, and married with two children). 29 Data on the weights for the four family forms are retrieved from the OECD family database, Table SF1.1.A.

Figure A2 shows the average private gross income benefit from completing tertiary education by country. The figure splits up the gross benefit into the net benefit and the net taxes paid. Countries are ranked by the gross income benefit.

Internal rate of return

• Study duration d is retrieved from Table B1.3a in OECD (2014); ‘average duration of tertiary studies in 2011 (in years)’, for all tertiary education. We rely on information from Column (2). If information from Column (3) is missing, we take the duration from Column (2), correcting with the ratio of the OECD averages for each indicator. For countries with missing information in either column, we use the data from Table C1.6 (Expected years in education). We correct these figures with the ratio of the average duration from Table B1.3a to the average expected years from Table C1.6 for the OECD as a whole.

The IRR is the return that balances the net present value of (expected) net income streams with and without a degree in higher education. We use E as a prefix to denote expectations: the expected gross income of type i = L, H is equal to E y i = ( 1 − p i ) y i and the expected net tax (taxes minus benefits) becomes Ent i = ( 1 − p i ) t i − p i ρ ( y i − t i ) ⁠ . The difference E y i − Ent i is the expected net income. Using continuous discounting (with d study duration), the net present value with a degree is equal to:

with adpe the average direct private expenditures as described before. The net present value without a degree is equal to:

with Enoc the expected net opportunity cost as described before. Equating these net present values, we obtain:

In other words, the IRR ensures that the expected net present value of gains in net income (the left-hand side) is equal to the net present value of the total costs (the right-hand side). Solving for r leads to:

It is easy to verify that the IRR decreases with study duration d, increases with the expected net income gain ( ⁠ ( E y H − Ent H ) − ( E y M − Ent M ) ⁠ ), and decreases with the (expected) costs (adpe + Enoc). 30

Perverse redistribution

with n_i the number of individuals (parents or students) of each type, s_i the average public subsidy for higher education (in 2013-PPP-US$), and q_i as defined before. Similarly, the total fiscal cost of all individuals can be written as:

The budget balances, that is, the total cost of the subsidies received by the cohort of current graduates must be equal to the total fiscal cost paid by the cohort of the individuals under scrutiny:

The fiscal cost is entirely collected via earnings taxation and in such a way that the ratio of the fiscal costs is equal to the ratio of the earnings taxes for both types:

The elasticity of progressivity measures the percentage change in the average tax rate for a percentage change in gross income. Formally, let y_i be the gross income of type i = L , H ⁠ . The elasticity of progressivity without and with the contribution is equal to:

Both expressions are equal if and only if Equation (A.2) holds. For an individual of type i = L , H ⁠ , this results in the following fiscal costs:

The net fiscal cost are then equal to c i − s i q i ⁠ .

Net fiscal costs in the parental view

In the parental view, the cohort of parents bears the direct and net fiscal costs of the cohort of recent graduates. We approximate the cohort of recent graduates by the 25–34 age cohort and the cohort of their parents as the 45–54 age cohort when we compute the absolute sizes of each group (but look at lifetime income to compute fiscal cost shares in both cases).

We classify parents in two groups according to education level. We follow the OECD classification, which is based on the highest degree obtained. The low skilled (group L in the main text) are those without a tertiary degree. In OECD terminology they are the low skilled (below upper secondary education) and medium skilled (upper secondary or post-secondary non-tertiary education) together and we will therefore refer to this group as LM in the Appendix. The high skilled (group H) are those with a tertiary degree.

The net fiscal costs for each type ( ⁠ c i − s i q i , i = LM , H ⁠ ) follow immediately.

• The number of individuals per type ( ⁠ n LM = n L + n M and n_H) is obtained from the shares of low-, medium-, and high-skilled adults in the 45–54 age cohort multiplied by the number of adults in this age cohort. Table A1.3 in OECD (2016) reports the shares per age cohort and per country. Table S.6 in United Nations Department of Economic and Social Affairs (2015) reports the total number of adults per age cohort and per country.
• The average number of children with a bachelor degree per parent of either type (q_LM and q_H) is the product of two factors. We write q i = π i b i for each type, with π_i the probability that a child of an i-skilled parent obtains a bachelor degree in tertiary education and b_i the birth rate (the average number of children of an i-skilled parent). The conditional probabilities (π_LM and π_H) are reported per country in Table A4.3 of OECD (2016). 31 The birth rates (b_LM and b_H) are assumed to be the same for both groups (because of data limitations). We will come back to its computation when we link the parental and student view later on.
• The public subsidy for a bachelor degree (s_LM and s_H) is assumed to be the same for both types (because of data limitations). Table B4.1 in OECD (2016) reports the total public expenditure on higher education, in % of GDP (we use the figures exclusive of R&D). 32 We convert these numbers to average expenditure per student, by using the ratio of the numbers provided in Table B1.1 (annual expenditure per student by educational institutions) and Table B2.1 (expenditure on educational institutions as a % of GDP) to compute a ‘conversion factor’ for each country.
• The earnings tax amounts (t_LM and t_H) are computed by applying the online OECD tax and benefits calculator to the gross earnings of each type (computed as described before). We look at income taxes here, as we assume that public spending on higher education is funded from this source. We again weigh taxes by family situation as explained before. For the low skilled, we compute the average tax amount, that is, t LM = ( n L t L + n M t M ) / ( n L + n M ) ⁠ .

Net fiscal costs in the student view

• The number of individuals per type ( ⁠ n LM = n L + n M and n_H) is based on the shares of low-, medium-, and high-skilled adults in the 25–34 age cohort multiplied by the number of adults in this age cohort. Table A1.3 in OECD (2016) reports these shares per age cohort and per country. Table S.6 in United Nations Department of Economic and Social Affairs (2015) reports the total number of adults per age cohort and per country.
• The public subsidy per bachelor degree (s_LM and s_H) is assumed to be the same for both types (because of data limitations) and has been defined before.
• The earnings tax levels (t_LM and t_H) are computed as before. The tax levels are based on the population aged 25–64 years (to capture average earnings across the lifetime) in both the parental view and the student view. There are still differences in t_LM in each case, as the relative shares of low skilled and medium skilled differ in each cohort. Hence, the tax levels are the same for L, M, and H, but there is a different weighted average for t_LM.

In the student view, we by definition have q_LM = 0 and q_H = 1. Applying formulas A3 and A4, the fiscal costs in the student view are then equal to:

Linking the parental and student view

To make our results comparable, we impose that the total subsidy cost of higher education is the same in both approaches, in each country. The birth rate (i.e., the average number of children per adult) will provide the necessary link. To distinguish between the parental and student view, we introduce a superscript P and S where needed. Using the previous definitions and assumptions, we impose

Overview of the data

Tables A1 and A2 summarize all the data.

Data on costs of higher education

Notes: “DC/y” = direct cost of higher education per year, “OC/y” = expected net opportunity costs per year, “PE/y” = total public expenditure on higher education per year.

Data on costs of higher education

Notes: “DC/y” = direct cost of higher education per year, “OC/y” = expected net opportunity costs per year, “PE/y” = total public expenditure on higher education per year.

Data on population, education, income, and taxation

Notes: Superscript P refers to parental view, superscript S refers to student view, subscript LM refers to those without a higher education degree (low-skilled and medium-skilled), and subscript H refers to those with a higher education degree.

Data on population, education, income, and taxation

Notes: Superscript P refers to parental view, superscript S refers to student view, subscript LM refers to those without a higher education degree (low-skilled and medium-skilled), and subscript H refers to those with a higher education degree.

Data links

Total public expenditure on higher education: http://dx.doi.org/10.1787/888933397862.

Public and private shares in total expenditure: http://dx.doi.org/10.1787/888933397770.

Unemployment rates by educational attainment: http://dx.doi.org/10.1787/888933396971.

Private costs for higher education

Notes: The figure shows the average yearly private cost of higher education, comprised of direct private costs and opportunity costs. Countries are ranked by the sum of the two.

Source: Own computations based on OECD data.

Private pecuniary benefits of higher education

Notes: The figure shows the private income benefit from completing higher education across OECD countries. It reflects both the net income benefit and the net taxes paid. Countries are ranked by the gross income benefit.

Source: Own computations based on OECD data.

SUPPLEMENTARY DATA

Supplementary data are available at Economic Policy online.

Footnotes

See, for example, Eurydice (2014) and OECD (2016) for a more systematic overview of recent reforms in higher education.

As discussed in the same section, this may come at the expense of moral hazard in the labour market.

The ranking by and differences across countries are highly similar when they are based on expenditure per student instead, such as those provided in Table B1.2 in OECD (2016).

Other countries also use loans with income-contingent features, but these are either targeted towards specific students (e.g., in Korea and the United States), or loans that revert back to mortgage-style loans above a certain threshold (e.g., in Germany); see Chapman (2016).

Another source of public expenditure is that the interest rate on public loans is often subsidized.

There are, to the best of our knowledge, only two examples of risk-pooling ICLs: the (abolished) Yale plan and the Hungarian Diákhitel; see, for example, Chapman (2006) for an overview.

As such, one cannot pay back more than the loan plus interest payments. This can potentially be augmented to cover the cost of loan default, but the scheme is assumed to be risk-sharing here.

The analysis includes all OECD countries except Greece, Hungary, Iceland, Latvia, Luxembourg, Mexico, Portugal, and Switzerland, for which no reliable data on one or more indicators could be obtained.

We use age-earnings profiles to impute the average earnings of those aged 20 years, as a measure of opportunity costs; see Appendix A for more details.

On a country level, relatively larger differences occur for countries for which the OECD reports very large figures, mainly Poland and Czech Republic. In general, the dispersion in the IRR is smaller in our approach.

It is not standard in the econometric literature to subtract taxes and to include direct costs. If we recompute the IRR for gross income benefits without direct costs, the average return is around 14%. Hence, these differences cannot explain the lower return in the econometric literature, which are likely due to the inclusion of control variables.

We do not use this terminology because we approximate, in both approaches, the tax contribution on the basis of the taxes that individuals (parents or students) pay over their lifetime.

When we refer to the high skilled in this section, this pertains to those with high education (either parents or students). Similarly, the low-skilled refers to those with no higher education degree.

Naturally, some low skilled still pay income taxes in Chile, but because we rely on average income positions, this is not captured by the analysis. In any case, average tax contributions of the low skilled will be very low in Chile.

Moreover, there is no variation in the size of the subsidy across students in our exercise. Because q L < q H in all countries, there is always regressivity of inputs, but countries can still be progressive through their taxation.

These are not externalities in the classical sense, but they still are based on the premise that the take-up of higher education by one individual can impact the payoff of others.

The evidence from the literature is exclusively focused on the average social return to education. There is, to the best of our knowledge, no empirical study that estimates externalities for the marginal student, which is what is relevant for policy. Given that private returns are lower at the margin, it is plausible a priori that their social returns are comparatively smaller as well at the margin.

To understand why, one can imagine a government that consists of separate departments responsible for, for example, taxation, welfare, and education. Without sufficient coordination between these departments, the budgetary windfalls accrue to the departments of taxation and welfare and are therefore an externality of the decision, taken at the department of education, to increase subsidies. In other words, if the government would jointly set educational subsidies and earnings taxes, then there would be no fiscal externalities (these revenue effects would still be present, but internalized).

Sanctions can provide an additional policy tool for handling moral hazard during higher education (see also Online Appendix B). A benefit of ICLs and GRTs is that they allow for flexibility, for example, by letting repayments increase non-linearly with study length.

A related disadvantage of GRTs, being uncapped, is what is labelled by Barr (2001) as the Mick-Jagger effect. It refers to the lead singer of the Rolling Stones who briefly studied at LSE and who would therefore have faced massive repayments under a GRT, which would be far removed from either costs or returns.

For an elaborate explanation, see Online Appendix B5.

The UK reform also involved considerable increases in support for those from low-income families. Chowdry et al. (2012) find that lifetime repayments are lower for the poorest 29% under the new system. Hence, the UK reform is not that informative on the participation effects of the poor when increasing private contributions through an ICL, but the Australian reform is.

For reviews of behavioural economics in an educational context, see, for example, Lavecchia et al. (2016).

While the United States is progressive in the parental view when we compare the high educated with all others, additional analysis shows that the children of the very low educated (i.e., high school dropouts) have very low usage rates and therefore do not benefit from the current system. Credit constraints could be a potential reason.

This is especially the case if a large share of the current classical loans are risk-pooling, which they often are in this group of countries.

The general interest political economy question here is why do voters do not support economically desirable policies ( Dal Bo and Dal Bo, 2017).

In the two ‘married’ cases, taxes depend on the income of the partner. The OECD calculator allows four options for the wage of the partner relative to the average wage: 0%, 67%, 100%, and 167%. We take the average of two opposite cases: ‘random sorting’ (the partner always earns the average wage) and ‘perfect sorting’ (the partner earns 67% for lowly educated, 100% for medium educated, and 167% for highly educated).

The latter being true if, for example, the expected net income gain is positive, which is usually the case.

This table reports conditional probabilities according to the highest education level of one of the parents of a family.

For countries that only report this spending inclusive of R&D, we correct these numbers based on the ratio of R&D spending to total spending based on Table B1.2 in OECD (2016).