3. What drives the gender pension gap? Case studies from the United States, Germany and Finland

Women generally receive a lower pension income than men in part because they tend to have earned less and have had shorter careers than men. This is a common explanation for the gender pension gap. However, it is less clear to what extent factors other than labour market factors may also contribute to the gender pension gap. Chapters 1 and 2 explore what some of these factors may be. This chapter complements their content by using case studies from three OECD countries to delve deeper into the question of what drives gender differences in some pension indicators.

The analysis in this chapter shows that while labour market factors remain a significant reason why women may end up with lower retirement income than men, that is only part of the story. Rather, delving deeper into case studies of different countries reveals that other factors are often at play, and that there is room for policy intervention to address the sources of the gender pension gap.

It explores how different features of policy design, demographic characteristics, incomes, workplace choices, and career trajectories contribute to gaps in pension coverage, assets, and entitlements in retirement savings arrangements. Understanding drivers of these gaps is important because individuals’ future retirement income is, at least in part, a function of whether someone has a pension plan and how much they save in defined contribution (DC) plans or accumulate in entitlements in defined benefit (DB) plans. Unpacking the contribution of different drivers of the pension gap makes it possible for policy makers to understand how much differences between men and women’s careers can affect their pensions. Conversely, if career disparities cannot explain differences in retirement income outcomes, a different explanation, such as pension policy or behavioural biases might better explain the gender pension gap.

This chapter considers these questions through case studies of three OECD countries: the United States, Germany, and Finland. It examines what drives gender disparities when it comes to coverage, assets, and entitlements accrued for both occupational pension arrangements and personal pension arrangements separately.

The analysis starts by using econometric analyses to explore why, in some countries, women may be less likely to have a pension plan through their workplace. For people who do have a pension plan linked to their employment, it also explores why assets or entitlements from those plans might differ between men and women. The analysis shows that the industry people work in and the number of hours they work can be related to whether or not a person has a pension plan from their workplace (for example, in the United States and Germany). Workplaces in some industries are significantly more likely to offer plans to their workers than other industries, and those industries happen to be male-dominated. A similar effect is seen in the type of work that people do. In the United States, some plans may only be offered to full time workers, which can disadvantage women. While in Germany such distinctions are not made within workplaces, it remains that workplaces that tend to hire workers with more standard and stable careers are also those that offer workplace pension plans to their employees. Again, this can disadvantage women. When coverage of occupational pension plans is universal, there is no meaningful difference in pension plan coverage between the genders (for example, Finland). However, in all the countries studied, when it comes to the assets or entitlements individuals may have from their occupational pension plans, labour market factors such as income and years of work appear to be the main drivers of a potential gender gap in accumulated assets or entitlements.

With respect to personal pension plans, the analysis shows that there is no clear sign that women are less likely to have taken out a personal pension plan than men. Instead, in Finland, the opposite appears true, that women in some age groups are more likely to have personal pension plans. While this is a positive sign, the same is not true of the amount of assets people accumulate in those plans. In Finland and Germany, the amounts women contribute to those plans are significantly less than men, as are the assets they manage to accumulate in those plans. Even after accounting for labour market factors, those differences are still evident, pointing to a potential behavioural bias between men and women.

This chapter proceeds as follows. Section 3.1 summarises the plans analysed in this chapter. Section 3.2 outlines the approach taken in the analysis. Sections 3.3 and 3.4 explore the drivers of the pension gaps in occupational pension plans and personal pension plans respectively in the three countries analysed. Section 3.5 discusses the results and concludes.

The analysis focusses on funded and private occupational or personal pension arrangements available in three countries, the United States, Germany, and Finland.1 Annex 3.A contains details of the data used in this analysis for the three countries.

The case study of the United States considers the voluntary occupational schemes, which can be DB or DC. In the United States, individual employers or groups of employers may voluntarily establish a complementary occupational pension plan for their employees. The plan sponsor decides what type of plan to establish. If a private employer offers a DB plan, participation is automatic and thus compulsory for covered employees. In the case of DC plans (such as 401(k) plans), participation may be automatic or voluntary depending on the plan type and its rules. In some cases, employees can choose whether to participate in a pension plan that is offered by their employers. Most occupational DC plans are 401(k) plans, where an employer can also make a matching contribution to the employee’s account. In the survey the data is based on, the question refers to these as a pension or retirement plan that an individual has through their job or their union. The question specifically excludes Social Security Railroad Retirement and individual retirement accounts (IRAs). This chapter does not consider drivers of gaps in personal pension plans in the United States.

For the case study on Germany, the analysis in this chapter focusses on private funded occupational and personal pension plans. Like in the United States, private occupational pension plans, or workplace pension plans, are voluntary in Germany. Access to the schemes is usually determined by collective agreements negotiated at the sector or company level, and are seen as a means of attracting and retaining staff.2 To collect this information, the German survey underlying the data used in this analysis asks respondents whether they have at least one contract for an occupational pension plan. The survey clarifies that occupational pension plans mean pension funds, pension schemes, retirement funds and direct pension commitments by the employer.3 This analysis also considers personal pension plans in Germany, which people take out for their own private saving purposes. In gathering this information, the survey questionnaire refers to all types of personal pension plans, providing examples such as the "Riester pension" or the private “Basic Pension” ("Rürup pension"), as well as non-government-subsidized private retirement pensions.

For the case study on Finland, this chapter explores occupational and personal pension plans. The Finnish occupational pension plan that is considered in this analysis is the statutory pension provided under the Employees Pensions Act (TyEL). It is a partially funded DB arrangement that is compulsory for all private sector employees in the country. It stands in contrast to the other case study countries, where the study considers voluntary occupational pension arrangements. Finally, the case study also considers personal pension plans in Finland, which do not form a large part of the pension system in the country, but are informative nonetheless in discerning people’s behavioural patterns.

The analysis considers two key factors that can lead to a gender gap in pensions: whether a person is covered by a pension plan, and the assets or entitlements they accumulate in that pension plan. This analysis considers the drivers of each of those factors separately.

The analysis relies on econometric modelling to examine the magnitude and statistical significance of different drivers of coverage and assets and entitlements. The approach used makes it possible to assess whether these drivers explain all the difference in pension outcomes between genders, or whether there is something else driving the gaps. Details of the methodology are available at Annex 3.B and full results of the econometric analysis at Annex 3.C. For brevity, only high-level summaries of the results are included in the main text of this chapter.

The analysis relies on two main models (which are explained in more detail in Annex 3.B). The first model uses logistic regressions to investigate drivers of pension plan coverage. In those regressions, the response variable is a relevant indicator (dummy) variable of whether an individual is covered by a pension plan. The approach is used to analyse coverage of occupational pension plans and personal pension plans. The second model relies on regressions using a two-part model to determine the drivers of the value of assets, entitlements, and contributions of people who were covered by a pension plan.

Readers should note that the analysis in this chapter relies on survey data. A key shortcoming of doing so is that people can under-report retirement plan coverage because people are often simply unaware that they have workplace pension arrangements. When they do have pension plans, whether occupational or personal, many also do not know the value of those plans. Therefore, the analysis in this chapter makes a key assumption that men and women do not differ in the way they under-report coverage or mis-report the value of their pension plans.

Many features of individuals’ working lives can cumulatively lead to gender gaps in pension income. The main ones are differences in wages, career lengths, type of work, and contribution levels. Chapter 1 contains a fuller discussion of these factors. This chapter explores which factors contribute to a gender gap in pension income and the effect their inclusion has on a gender indicator.

This section first considers drivers of a difference in coverage of occupational pension plans between men and women for the three countries. Then, for the people who are covered by an occupational pension plan, it explores why the assets, entitlements, and contributions in those plans may differ between men and women.

Women perform worse than men in almost all indicators of occupational plan coverage in the United States. Table 3.1 shows aggregate outcomes for coverage indicators. First, it shows that women are less likely than men to be covered by an occupational pension plan. The data shows that women are also generally less likely to be eligible for a plan through their work. This is related to the question of coverage in the United States since individuals may not be eligible for occupational plans through their employers, and if they are eligible, they may not participate in that plan. Table 3.1 shows that even among people who are eligible for an occupational plan through their workplace, women less commonly participate in that plan (by making contributions, for example). These results are evident across all age groups, although the differences are starker for the older groups. To delve further into whether these differences are significant enough to be statistically meaningful, what follows explains the results of a econometric model that analyses these factors.

The results of econometric modelling show that the differences in coverage, eligibility, and participation in occupational plans are statistically significant in most age groups (Table 3.2). The analysis shows that the odds of women having a pension plan through their workplace was 77.8% of the odds of males having a workplace pension plan (22.2% less) in 2017. This lower likelihood of women having an occupational pension plan (whether from a current or previous employer) persists across the age cohorts analysed, and to a similar magnitude, although the statistical significance of the result for the youngest cohort is weaker. Table 3.2 shows that there is also a statistically significant difference in eligibility and participation between the genders. However, the aggregate differences are mostly driven by older cohorts. While women have lower odds of being eligible for an occupational plan through their workplace, this overall result appears to be driven by workers aged 45 and older. The gap between men and women is greatest with respect to participation in plans, when workers are eligible. The modelling shows that overall, the odds of women participating in a plan when eligible are 77.2% of the odds of males doing so. This result is mostly driven by people aged between 30 and 59.

To unpack these aggregate results, the analysis then accounted for other characteristics that may be driving the gaps, but are somehow associated with gender. The econometric analysis added those characteristics gradually, to see whether gender continues to have a statistically significant relationship with eligibility and participation, even after accounting for other characteristics that may lead to gender gaps (Annex 3.C). This approach aims to account for the fact that it is often not gender alone that explains a gender pension gap, but rather, factors associated with gender. When the modelling accounts for more of these characteristics, the explanatory power of gender diminishes, which shows that the model successfully accounted for and measured drivers that relate to gender. When the model includes enough drivers such that gender loses all predictive power, it is possible to conclude that the model is accounting for most drivers associated with gender that have a bearing on eligibility and participation. Detailed results are available at Annex 3.C, but the tables that follow summarise the results when all explanatory variables are included alongside the gender indicator, for brevity.

The predictive power of gender diminishes entirely after accounting for other factors likely to be associated with gender but which have a bearing on occupational plan eligibility and participation, suggesting that gender alone is not what drives occupational plan eligibility and participation. Table 3.3 shows gender to be uncoloured (not statistically significant at the 90% confidence interval or higher) when other relevant drivers are included in the analysis. The same analysis could not be conducted for drivers of coverage of occupational pension plans from any employer, since most possible drivers available in the data refer to features of an individual’s current employment.

The results show that, broadly speaking, the following variables tend to drive eligibility for occupational plans in the United States:

  • Whether an individual is employed on a full-time or part-time basis. In the United States, it is not unusual for retirement plans to be restricted to workers with full-time or near full-time schedules (Kobe, 2010[1]). Furthermore, females are much more likely to be working part time than men (Table 3.4).4 This is therefore is a key driver of the eligibility gap.

  • Industry of work. The industries men and women work in are likely to explain much of the difference in occupational plan coverage between the genders. Men are more likely to work in industries that tend to have higher occupation plan coverage rates compared with women (such as manufacturing, mining, and technical services) (Table 3.4 and Figure 3.1). As such, they are more likely to be eligible for a workplace pension plan.

  • The length of time an individual has been at a workplace. In the United States, employers are more likely to offer pension plans to people who have worked there longer. Some plans also have minimum service requirements. The modelling results show that being in a job for less than a year significantly reduces the odds of an individual being eligible for a workplace plan. However, the data also show that the genders are about equally likely to have been in a job for under a year (Table 3.4). The fact that women and men do not have large differences in the likelihood of being in a job for less than a year suggests it is unlikely to explain much of the gender gap in coverage or eligibility for a pension plan.

  • Public / private sector mix. The data show that government workers are significantly more likely to have an occupational pension plan than private-sector workers. Women are over-represented in government work. As such, this can act to narrow the gender pension gap to a small degree.

  • Firm size. In the United States, workers in smaller firms are much less likely to have a plan available to them. This is borne out in the modelling, which shows that individuals employed in small or micro businesses are significantly less likely to be covered by or eligible for an occupational plan. However, the data show that employment in these smaller businesses is about evenly split between the genders, with a slight tilt towards males. As such, this factor is unlikely to drive the gender pension gap in the United States.

Women are less likely to be participating in a pension plan, even when they are eligible for one, and this may be related to behavioural factors. When it comes to participation in a pension plan, when an individual is eligible for that plan, Table 3.2 shows that overall, the odds of a woman participating in a workplace plan are about 77% of the odds of men doing so.5 However, this difference is only statistically significant for the two cohorts of people aged 30-59. In the youngest cohort (aged 15-29), no such difference in participation exists between the genders, suggesting that younger generations may be exhibiting different behavioural trends than their predecessors. Those differences persist even after the model accounts for whether a person attended college, was in job for a short time period, and is a government employee. Controlling for part-time work has a small impact on the odds that women participate in a pension plan if eligible, but a statistically significant difference persists even after controlling for this variable. Whether or not an individual works for a small business similarly appears to have little bearing on the statistical significance and odds of a woman, participating in a pension plan compared to a man. What ultimately appears to affect the statistical significance of gender as a predictor is controlling for the variables discussed as well as whether a person works in an industry with high pension coverage and is covered by a union contract (Table 3.3). Intuitively, these factors should not alone affect a person’s decision to participate in a pension plan if they are eligible, suggesting that there may be some related behavioural factors at play.

The aggregate results from the Household Finance and Consumption Survey (HFCS) data show that Germany has a gender gap in occupational plan coverage – 24% of men and 19% of women reported being covered by funded occupational pension plans through their workplace.6

The econometric modelling shows that these differences are statistically meaningful across most age groups. The odds of women having a pension plan through their workplace was 73.6% of the odds of men being covered by a workplace pension plan (26.4% less) in 2015 (Table 3.5). However, analysing age cohorts separately shows that individuals in the age cohorts over 45 and below 30 mostly drive this result. The gap in occupational plan coverage in Germany was particularly large in the cohort of individuals aged 45-59 and 60-64. For those individuals, the odds of women having occupational pension plans were 65% and 45.5% of the odds of men, respectively. For younger workers, on the other hand, there was no statistically significant difference between men and women’s occupational plan coverage for workers in the 30-45 age cohort, and statistically significant difference at a lower level of confidence for the 15-29 age cohort. This suggests that for the younger generation of workers, the coverage gap might be closing.

Of the cohorts of individuals with statistically significant gaps in occupational plan coverage in Germany, the employment gap appears to explain the difference. When the model is adjusted to control for whether or not an individual was an employee, it shows that a positive and statistically significant relationship exists between an individual having an occupational pension plan in Germany and being employed (as opposed to self-employed). Importantly, when that variable is included in the model, the statistical significance of gender disappears for most age groups (Table 3.6). This suggests that the coverage gap is linked to differences in likelihoods of men and women being in the labour market. The data on men and women’s employment accords with this finding. It shows that in the year the HFCS survey was conducted, 75% of men compared to 68% of women aged between 14 and 65 received employee income in the last 12 months. Similarly, data on participation rates by gender show a gap of about ten percentage points in favour of men in Germany.7 This gap is evident across all age cohorts. While men are slightly more likely to be self-employed in Germany, and therefore not covered by an occupation pension plan, this does not fully offset the effect the gender employment gap has on occupational plan coverage.

The number of hours an individual worked, which on average differs by gender, was also a statistically significant predictor of occupational plan coverage in Germany. Table 3.6 shows that the more hours an individual tended to work, the greater their odds of having an occupational pension plan. However, German law prohibits a fund’s rules from discriminating between full-time and part-time employment for the purpose of company pension plan coverage. As such, hours worked alone is unlikely to be the cause of lower coverage, but rather, it suggests that firms whose workers have more standard and stable hours are also those who are more likely to offer pension plans to their workers. This matters for the gender pension gap because the data show that women are more likely to be in jobs that generally involve fewer hours, suggesting they are in the latter category.

Another potential factor at play in explaining the coverage gap is the company size. Men are more likely to be employed in larger companies than women are, and analysis by the Germany Ministry of Labour and Social Affairs has shown that larger companies are much more likely to offer occupational plans to their employees (Bundesministeriums für Arbeit und Soziales, 2020[2]). However, due to a lack of relevant information in the data source used, the econometric analysis did not include firm size in the analysis of drivers of company pension plan coverage in Germany. Notwithstanding, but it is likely to be an important factor explaining the coverage gap.

Another important difference not included in the analysis of pension plan coverage in Germany is the industry of employment. In Germany, access to workplace pension plans is usually determined by collective agreements negotiated at sector or company level or by company agreements. There are relatively large disparities between coverage across different industries. Coverage of company pension plans is highest in industries such as credit and insurance, mining and quarrying, electricity and gas supply, and water supply, which tend to be male-dominated.8 Others, such as health, veterinary and social services, and education and teaching have slightly lower coverage rates and are female dominated. On the other end of the spectrum are industries such as accommodation and food services, and administrative and support services, which are female dominated. As such, the sector or company can have a bearing on gender differences in coverage. While the sample data was not large enough to permit this variable to be included in the modelling, a high-level analysis of gender distribution by industry confirmed that it was likely an important driver.

Gender is not a key factor explaining any differences in coverage of occupational plans in Finland. Occupational plans are mandatory in Finland, so there is no strong reason for there to exist a gender difference in occupational plan coverage. The overall figures in Chapter 1 show that the difference in occupational plan coverage is small – 85% of men and 86% of women are covered by occupational pension plans. The results in Table 3.7 confirm this expectation, as they show that the gender indicator is not a statistically significant predictor of occupational plan coverage for any age cohort. Table 3.8 shows that the main predictor of whether an individual has an occupational plan in Finland relates to whether they are working. In any event, there does not appear to be a strong enough gender difference between the likelihood of having an occupational pension plan by gender, as having one such plan only depends on an individual having had employment at some point in their lifetimes. As such, no more analysis was needed for this case study, and it could proceed to an analysis of the drivers of differences in asset values and entitlements in occupational pension plans.

For people who do have occupational pension plans, it is possible to further investigate the drivers of overall gender pension gaps by exploring what might explain differences in the values of assets or entitlements accrued in those plans. This section proceeds by exploring the drivers of DC account values, expected DB income, employer contributions, mandatory employee contributions, and voluntary employee contributions in the United States. It will then explore drivers of differences in DC account values or DB entitlements for Germany and DB entitlements in Finland. It shows that, for the most part, unlike drivers of plan coverage, differences in assets and entitlements are more linked to labour market outcomes.

Men generally have greater accumulated assets or entitlements in their occupational plans compared with women in the United States. The data from the United States contain information about asset balances, expected pay-outs from DB plans, and contributions behaviour for people who do have occupational plans. Table 3.9 shows that on average, across all age groups, men accumulated larger balances in occupational DC accounts than women. The same result is evident for expected income from DB plans. Furthermore, of people who do have occupational plans, men tend to contribute greater amounts of money both mandatorily and voluntarily, and their employers contributed more overall as well. The same pattern can be seen in results for median values, which can be a better indicator for when distributions are skewed.

The results of regression analyses confirm the overall findings in these descriptive statistics (Table 3.10). The first row shows that the gender differences for key indicators of plan assets, entitlements, and contributions are statistically significant for almost all cohorts of individuals. That is, the gap between men and women’s DC asset values, expected DB income, and contributions are large enough to be statistically meaningful. An exception is the level of mandatory contributions by employees, for which gender was not a statistically significant predictor.

The results show that years an individual participated in an occupational pension plan, as well as income, are key drivers of the level of assets, entitlements, and contributions to occupational pension plans in the United States. This is possible to discern once the analysis accounts for other factors that are related to gender but also have a bearing on the results for the indicators being analysed. Table 3.10 summarises the broad outcomes of analyses that account for such possible drivers of the gender gaps discussed. Indeed, these drivers have a relationship with gender. Women tend to work fewer years overall than men, since they are more likely to take time off for parenting or other caring responsibilities. This means they have fewer years during which they are building up their assets or entitlements to pensions. Similarly, women experience a well-documented gender pay gap that persists even in younger generations.9 These results are unsurprising, but an important finding from the analysis is that after accounting for all these factors in the model, gender was no longer a statistically significant predictor of the gaps in occupational plan assets, entitlements, and contributions when it comes to workplace pension plans. This suggests that differences in labour market outcomes, mainly years of work and income gaps, are likely to account for most of the observed differences.

There are some exceptions to these trends. Analysing the youngest age cohort (individuals aged 15-29) did not show statistically significant differences in account balances in DC plans. The same analysis was not possible for expected income from DB plans since the sample size was too small. However, this gap was also not evident for employer contributions to occupational plans for individuals in that age group. This is not to say that the gender gap in pension assets is disappearing. Rather, the 15-29 age bracket might simply align with years before career breaks (such as for parenting) and may simply reflect a period prior to the emergence of gender-based differences in income and employment. To illustrate, Figure 1.9 of Chapter 1 shows that in OECD countries, at the early stages of individuals’ careers, women and men have almost the same amount of assets, but this gap widens with time.

The analysis of the drivers of the level of occupational plan assets and entitlements for those individuals who do have occupational plans shows that the overall difference between genders is statistically significant (Table 3.11). However, most cohorts of individuals do not show a statistically significant difference in asset or entitlement values between men and women, with the exception of the cohort aged 45-59. The analysis therefore suggests that while there appears to be a gap in occupational plan assets and entitlements in Germany, it may not be wide enough to be statistically meaningful over most age cohorts or the data sample was not large enough to determine the drivers of a gender gap with certainty.

Considered together, the results of an analysis of coverage and assets/entitlements in German occupational pension schemes suggest that differences in the overall asset level or value of entitlements in occupational plans in Germany might be explained by a coverage gap due to men being more likely to be in paid employment or in workplaces that offer plans to their employees. Any differences in the value of assets/entitlements between people in occupational pension plans is less evident, but if it exists, would mostly be related to income and work experience and gender does not appear to be a strong driver.10

The results for Finland, on the other hand, show a significant gender gap in pension plan entitlements that persists across all the age cohorts analysed. The overall figures show that for individuals in the cohorts aged 30-45, 45-60 and over 60, women have systematically lower pension plan entitlements than men. Analysis of the cohort aged 15-29 reveals the opposite result: women have systematically higher pension entitlements than men.

Gender differences are still evident even after the analysis accounts for education, age, relationship status, and income. However, the data set used does not make it possible to control for years of work history, which can be a significant driver of differences in entitlements. A crude way to circumvent this issue is to assume that the primary reason for career breaks is parenting, and a crude way to isolate individuals who have not taken career breaks is to consider only single people. Interestingly, after restricting the population to single individuals, the statistical significance of the female coefficient disappears for the cohort aged older than 35. This adds some weight to the possibility that differences in career lengths explain the apparent gender gap in pension entitlements in Finland.

However, an analysis of the cohort aged 15-35 shows that women, including single women, have systematically higher pension entitlements than men even after controlling for age, education, and income. This result might signal a generational shift in the gender pension entitlements gap in Finland. However, it may also be due to this group being less likely to have experienced long career breaks than older age groups.11 As such, the finding may support the view that taking career breaks for parenting may be a key driver of gender gaps in occupational pensions in Finland.

This section considers the drivers of the pension gap in personal pension plans. Gaps in personal pension plan coverage have some similarities to occupational pension plans, because much of the gap can be related to labour market differences. But unlike occupational pension plans, differences in personal pension plan coverage and asset values can have a lot to do with personal qualities and individual behaviours, as discussed in Chapter 2. Those qualities and behaviours cannot be captured effectively using the data at hand. Notwithstanding, it is possible to draw some conclusions regarding other key drivers of why individuals might take out personal pension plans, and what drives the asset values in those plans. Readers should note that it was not possible to conduct the same analysis for personal pension plans for the United States, and this section will proceed with an analysis of Germany and Finland.

The analysis on Germany only shows a statistically significant gender gap in coverage of personal plans for individuals in the 45-59 age cohort (Table 3.14).12 In this age group, the odds of women having a personal pension plan were 77.4% of the odds of men having one. While there is no systematic gender coverage gap across all cohorts, it is important to bear in mind that the years leading to retirement are when people are more likely to voluntarily start saving for retirement. As such, the older age cohorts are particularly relevant when analysing personal pension plans.

Employment and education gaps are likely to explain the gender gap for personal pension plan coverage in Germany. After controlling for individuals’ incomes and whether they received a tertiary education, the gender predictor loses its statistical significance for the 45-59 age cohort (Table 3.15). This suggests that the relationships between gender and income and gender and education are likely to explain much of the difference. As such, the gender wage gap, which disadvantages women financially can also have a bearing on the likelihood that they would take out personal pension plans. Intuitively, this would suggest that people with lower financial means are less likely to take voluntary steps to start saving in personal pension plans. A similar effect is evident when it comes to education. People who have completed tertiary education are significantly more likely to take out personal pension plans. This is likely true both because of better financial knowledge but also because people with tertiary education tend to have higher incomes. There is also an important relationship between tertiary educational attainment and gender for the age cohort that exhibits a difference in personal pension plan coverage. Of people in the 45-59 age group, German women are significantly less likely to have attained a tertiary education, unlike younger age cohorts, where tertiary education attainment tends to be closer to gender parity.13

Another interesting outcome from the analysis is that for the cohort aged 30-44, after accounting for income and educational attainment, women were more likely to have a personal pension plan than men. This suggests that something else might be at play which may warrant further research.

In contrast to the case of Germany, the overall figures for Finland show that women are more likely to have personal pension plans than men. Table 3.16 shows that overall, the odds of women having a personal pension plan were about 22.8% higher than those of men. Women aged between 30 and 59 appear to be driving this result. While it should be noted that personal pension plans are a very small component of Finland’s retirement income system overall, the results remain interesting.

While income and education are associated with higher personal plan coverage in Finland, factors associated with gender are also likely to be driving the positive gender coverage gap in favour of women. Table 3.17 shows that this gender coverage gap persists even after controlling for key factors that might affect coverage. It shows that, in line with intuition on the matter, higher income levels and attaining a tertiary education are all associated with higher odds of having a personal pension plan in Finland. Finland can be contrasted with Germany, since women are more likely to have attained a tertiary education in Finland. However, controlling for these factors in the econometric analysis (see also Annex 3.C) leaves the relevance of the gender indicator virtually unchanged, suggesting that other factors associated with gender are likely to be driving the positive gender coverage gap.

This section considers divers of differences in personal plan assets and contributions for individuals that have personal pension plans.

Labour market factors appear to be the primary reason women accumulate less than men in personal pension plans in Germany. The figures from the modelling on Germany show that overall, of people who do have personal plans, women tend to accumulate less assets in those plans. But after controlling for total income, years spent in the workforce, and entitlements in occupational pension plans, the explanatory power of the gender coefficient disappears. This suggests that women’s career lengths and incomes are the primary drivers of their lower personal pension savings. However, for the cohort of people aged 30-45 there is no statistically significant difference between men and women’s personal plan assets. Notwithstanding, the sample size is small, making the result uncertain (and at odds with the results for contributions shown in Table 3.19). On the other hand, for the cohort of individuals aged 45-60, the reverse is evident: women have systematically lower personal plan assets than men, and this difference persists even after controlling for variables such as education, age, time in employment, marital status and income.

When considering contributions to personal pension plans, the analysis shows that women tend to contribute significantly less than men to their plans (Table 3.19). This result is evident across almost all cohorts, with the exception being the oldest age cohort. The result holds even after controlling for predictors such as income, age, education, years of work experience, and the level of entitlements in occupational plans. While including these predictors diminished the effect of the gender coefficient, it remained statistically significant and negative for the age groups 15-29 and 45-59. This suggests that after controlling for education, work history, income, and occupational plan balance, women still contribute to personal plans less than men. It points to something more than simply labour market outcomes, but possibly a behavioural effect leading women to save less than men.

One shortcoming of this analysis, which readers should bear in mind, is that Riester plans come with a government subsidy, which does not appear in the data on individual contributions. That government subsidy can have the effect of better equalising asset gaps between the genders and the data on contributions would not account for this. Notwithstanding, it would not change the key findings of what may drive gender gaps where they exist.

In Finland, an overall gender gap in assets accumulated in personal pension plans is mostly driven by the cohorts aged 30-59 (Table 3.20).

The results show that other factors, such as having a tertiary education, age, marital status, and income earned can also be related to the assets accumulated in a personal pension plan in Finland. Individuals with a tertiary education, higher incomes, and more time in employment tend to save more in their personal pension plans. Furthermore, people who are living in couple are more likely to have accumulated greater amounts in their plan than those who are no longer in couple. Notwithstanding, in the cohorts aged 30-44 and 45-59, a statistically significant difference in the assets accumulated between the genders persists even after controlling for these factors.

A similar pattern is evident when examining contributions to personal pension plans. Women overall contribute less than men, and this result is mainly driven by the 30-59 age group. The gender difference persists even after controlling for typical labour market and educational outcomes, which suggests, like in the case of Germany, a possible behavioural bias between men and women when it comes to saving in personal plans.

This chapter analysed drivers of the gender gap in potential pension income, for three OECD countries: the United States, Germany and Finland. The focus of the analysis was to unpack causes of differences in retirement savings outcomes by gender, other than the typical labour market explanations.

The chapter focussed on what might drive gaps between the genders when it comes to pension plan coverage, and for people who are covered by a pension plan, what drives differences in the assets or entitlements accumulated in those plans. There are, as always, some shortcomings when conducting this type of analysis. First, what this chapter explored was only some of the many potential pathways for working life factors to affect retirement incomes. Of course, many other factors, whose analysis was outside the scope of the analysis described in this chapter, could affect retirement incomes. These include different life expectancies between the genders, different behavioural biases, the fact that couples may pool income sources, the effect of relationship breakdowns, and so on.14 Notwithstanding, the chapter shows that it is possible to discern how some key features of the accumulation phase in some countries can lead to gender pension gaps in the future.

The analysis showed that for occupational pension plans, gender differences in coverage could be explained by factors other than labour market outcomes, but not for differences in assets or entitlements accumulated. The case studies for the United States and Germany showed that the industries women tend to be employed in and the type of work they do, which is more likely to be part time, can lead to a difference in the likelihood that they have a pension plan with their workplace. The same is not true for Finland, which does not have a gender gap in occupational plan coverage because funded pension plans are mandatory for all workers.

The findings suggest that there is room for policy interventions in the United States and Germany to encourage more employers to offer plans to all workers, particularly in instances where coverage rules may disproportionately impact women. The analysis showed that this is potentially the main area policy makers can use pension policy design to improve outcomes for women, since gender gaps in occupational pension plan assets and entitlements appear to be mostly explained by labour market differences.

When it came to personal plans in Germany and Finland, there was no clear sign that women were less likely to have a personal plan, but a behavioural disparity may lead women to contribute less to plans when they have one. The analysis showed that men and women were about equally likely to have a personal pension plan in Germany, and women were more likely to have one in Finland. Notwithstanding, in both case studies, women contributed less to their plans than men, and accumulated lower assets in their plans than men. This result proved true even after the modelling accounted for potential labour market differences between men and women. The results suggest that something else, such as a behavioural disparity, might be at play. Of course, the question about what behavioural traits might drive people to contribute more to their plans is not something that is easy to pinpoint using survey data. Notwithstanding, it is valuable to discern that something other than typical labour market factors may be leading to differences in gender outcomes, suggesting that there may be room for policy makers to address such shortcomings. For instance, upon further examination policy makers might conclude that behavioural biases such as risk aversion might explain different attitudes to retirement saving among the genders. As such, they may choose to tailor financial education programmes to counter such outcomes.

References

[3] Belotti, F. et al. (2015), “twopm: Two-part models”, The Stata Journal, Vol. 15/1, pp. 3-20, https://journals.sagepub.com/doi/pdf/10.1177/1536867X1501500102 (accessed on 1 February 2021).

[9] Bundesministerium für Arbeit und Soziales (2020), Ergänzender Bericht der Bundesregierung zum Rentenversicherungsbericht 2020 gemäß § 154 Abs. 2 SGB VI (Alterssicherungsbericht 2020), https://www.bmas.de/SharedDocs/Downloads/DE/Rente/alterssicherungsbericht-2020.pdf;jsessionid=50F4F18C7380CF9FD520DD4816F94A30.delivery2-replication?__blob=publicationFile&v=1 (accessed on 18 February 2021).

[6] Bundesministerium für Familie, Senioren, F. (2016), Atlas zur Gleichstellung von Frauen und Männern in Deutschland.

[2] Bundesministeriums für Arbeit und Soziales (2020), “Arbeitgeber- und Trägerbefragung”, https://www.bmas.de/SharedDocs/Downloads/DE/PDF-Publikationen/fb567-endbericht.pdf?__blob=publicationFile&v=1 (accessed on 25 January 2021).

[4] Cooper, D., K. Dynan and H. Rhodenhiser (2019), “Measuring Household Wealth in the Panel Study of Income Dynamics: The Role of Retirement Assets”, 6, No. 19, Federal Reserve Bank of Boston, https://psidonline.isr.umich.edu/Guide/Quality/DataComparisons.aspx.

[1] Kobe, K. (2010), Small Business Retirement Plan Availability and Worker Participation Research Paper, https://www.sba.gov/sites/default/files/rs361tot.pdf (accessed on 14 January 2021).

[5] O’Rand, A. and K. Shuey (2007), “Gender and the Devolution of Pension Risks in the US”, Current Sociology, Vol. 55/2, pp. 287–304, http://dx.doi.org/10.1177/0011392107073315.

[7] OECD (2021), Gender wage gap (indicator), https://dx.doi.org/10.1787/7cee77aa-en (accessed on 25 January 2021).

[8] OECD (2019), Pension Markets in Focus 2019, https://www.oecd.org/daf/fin/private-pensions/Pension-Markets-in-Focus-2019.pdf (accessed on 28 January 2021).

The analysis of the United States in this chapter relies on data from the Panel Study of Income Dynamics (PSID), a longitudinal household survey conducted by the Survey Research Center at the Institute for Social Research at the University of Michigan. The dataset is a rich source of demographic and financial information dating back to 1968.

The PSID dataset is useful because it provides both household-level and individual-level information on a range of demographic and financial variables. The family household file contains information on gender, marital status, education, income, housing, children, employment history, pension plan coverage, pension plan values, and contributions to pension plans. The PSID pension module has detailed information about the head of household and any spouse’s retirement accounts (DC and DB) at their current employers and at as many as two previous employers each. It collects information about the balance of retirement accounts (IRAs), although this information is only reported in total for the household.

The PSID has some shortcomings. For instance, it does not track the upper end of the wealth and income distribution. Another issue is that not all people who report having DC retirement accounts know the value of those accounts. In some years, the PSID survey asked these people whether the account value lay within a certain range. In this analysis, where individuals only reported that value as a range, the midpoint was used.15 The same issue arises for people who were asked their expected future income from a DB plan. If people did not know their potential income from their DB plan, the questioners asked them for an estimate as a percentage of their salary. The analysis in this chapter calculates the expected DB income in such cases, based on the stated percentage and the individual’s reported income. However, since this chapter focuses on gender-based differences, it is unlikely that the data shortcomings will affect the results (since it is unlikely that males and females systematically differ in their abilities to recall account information).

The analysis for Germany and Finland relies on Wave 2 of the Household Finance and Consumption Survey data. The data was published by the European Central Bank in 2016 and provides household-level data in 20 Euro area countries for the second wave. Those data relate to survey responses collected between 2013 and 2015 for European countries.

The HFCS data is useful in that it contains many variables that could have a strong relationship with coverage and asset/entitlement levels in the working-age population. The HFCS data include information about individuals’ age, income, education, marital status, employment history, work type, hours worked, and pension assets or entitlements for both personal and occupational plans.

The HFCS data on occupational plan assets or entitlements is different for Germany and Finland. The data for Germany refer to assets or entitlements in private occupational plans. The German survey which collects HFCS data (the German Panel on Household Finances (PHF)) refers to these plans as ‘company pension plans’. The occupational plan data for Finland refer to the present value of future entitlements from employer-provided occupational pension schemes.16 The personal pension plan data for both countries refer to the total value of all of an individual’s voluntary pension plan assets.

Readers should note that the HFCS data for Germany are based on the Panel on Household Finance of the Deutsche Bundesbank. This sample contains only around 4 500 households and is not collected specifically for gathering data on occupational or private pensions.

The analysis of the coverage and asset/entitlement gaps was done in two stages. The first uses logistic regressions to investigate drivers of pension plan coverage, with occupational and personal plan coverage analysed separately. In those regressions the response variable is an indicator (dummy) variable of whether an individual was covered by each type of pension plan (occupational and personal). In the case study of the United States, the indicator dummy referred separately to whether an individual was eligible for a pension plan through their workplace, and whether they participated in a pension plan if they were eligible. The second stage uses a two-part model to determine the drivers of the value of assets or entitlements in funded pensions for people with personal or occupational plans, which are analysed separately (following the approach outlined in Belotti et al. (2015[3])). Using a two-part model made it possible to cater to a situation where a regression is conditional on a positive outcome, that is, that an individual has a relevant pension plan. The second part of the two-part model estimated the log of the dependent variable using ordinary least squares regression. The log transformation helps overcome shortcomings that come with having right skewed dependent variables. For results that examine the level of assets, entitlements or contributions to pension plans where an individual has one such plan, the chapter only reports the results from the second part. Rather than report results for the first part, instead the analysis relies on logistic regression analysis, to account for there being more information on whether an individual has a particular plan, than the amounts in that plan. This is because individuals were more likely to answer questions about whether or not they had a plan than the amount of assets in a plan, since they are less likely to know the latter.

In both stages of the modelling, the econometric analysis explores the effect of different relevant drivers by including them in the regressions incrementally. This makes it possible to see how the explanatory power of the female indicator changed as the regressions featured more of the predictors.

These analyses consider the whole of the population as well as cohorts of individuals in the following age groups: 15-29; 30-44; 45-59; over 60. Splitting the population into cohorts makes it possible to judge whether and how results change as younger generations benefit from an equalising playing field across genders. Some factors typically associated with gender pension gaps, such as gender pay gaps, have trended downward in OECD countries in recent years. Splitting the sample into cohorts is one way to track how the drivers of the gender pension gap may be changing in statistical significance and magnitude over time.

Readers should note that the analysis presented in this chapter makes it possible to examine the factors that drive the gender gaps in pensions, but the approach of using logit and a two-part model does not make it possible to understand the relative contribution or importance of different factors to the gender gaps in pensions. In the academic literature, such an analysis is typically done using a Oaxaca-Blinder decomposition approach. The data did not permit such a decomposition technique for the analysis in this chapter.