2.1. Dutch pension system

The Dutch pension system consists of three pillars. The first one – known as AOW (Algemene Ouderdomswet) – operates as a state-funded old-age pension under a pay-as-you-go (PAYG) scheme and is financed by income-related taxes. The primary goal of this pillar is to prevent poverty among the elderly by providing a uniform pension benefit to all residents aged 65 and above.Footnote ³ The benefit is linked to the statutory minimum wage and depends on the length of legal residence in the Netherlands and cohabitation.

The second pillar involves occupational pensions managed by pension funds. This pillar serves as a supplement to the uniform public benefit, targeting workers who earn more than the minimum wage (Bovenberg and Nijman, Reference Bovenberg and Nijman2019). Occupational pension schemes are typically associated with a single employer (although there are several large industry pension funds). These funds are governed by boards whose members are appointed by or on behalf of employers, employees, and pensioners (Chen and Beetsma, Reference Chen and Beetsma2015). They are organised as defined benefit (DB) plans, where the benefit is determined according to the number of years worked and a reference wage (either final pay or an average of previous earnings). However, since the participants themselves bear the residual risk of any shortfall of pension funds, this system can also be considered a hybrid, having characteristics of both DB and defined contribution (DC) plans. While the accrued pension rights are specified as in a common DB plan, they are also partially DC since the yearly indexation is linked to the financial position of the pension funds, and therefore to investment returns (Ponds and van Riel, Reference Ponds and van Riel2009; Bikker et al., Reference Bikker, Broeders, Hollanders and Ponds2012). Participation is mandatory for almost all employees (coverage is 90%), but it excludes self-employed workers.

The third pillar consists of voluntary personal pension provisions such as complementary pensions, annuities, and life insurance policies. These voluntary savings are eligible for almost the same tax benefits (Exempted premiums with Exempted returns and Taxed annuity stream or EET) as those applicable to occupational pension contributions, albeit sharing similar limitations, such as reduced liquidity. As in many other countries, the third pillar is the least popular (Knoef et al., Reference Knoef, Been, Alessie, Caminada, Goudswaard and Kalwij2016). Only a small fraction of workers voluntarily tops up their mandatory pension wealth with additional voluntary pension schemes.

2.2. Supervisory framework

The Dutch supervisory framework (Financieel Toetsingskader, FTK in Dutch) for the second pillar of the occupational pension fund sector was introduced in 2007. Under the FTK, the financial position of a pension fund is determined by the funding ratio (dekkingsgraad) which represents the ratio between its assets and liabilities.Footnote ⁴ If this funding ratio falls below the minimum required threshold, it indicates that total assets are insufficient to cover expected future liabilities, including the pension benefits of the members of the pension plan. In such a scenario, the pension fund is required to submit a recovery plan to DNB, which serves as the regulatory body. This specific plan sets out the strategy that the fund will adopt to enhance its financial position, aiming to surpass the mandated funding ratio.

In 2015, the FTK was replaced by the new financial assessment framework (nFTK). In the FTK, liabilities were discounted using adjusted market rates, but in the nFTK this has been replaced by the ultimate forward rate (UFR). This is relevant to our analysis since the period that we investigate witnessed a persistent decline in interest rates, which contributed to reducing the funding ratio. In an environment with unfavourable interest rates, investment returns and funding ratios became closely linked to the interest hedging strategies employed by pension funds. The revised framework was introduced with the goal of avoiding widespread curtailments by pension funds, as well as excessively long recovery periods. Under nFTK, the recovery period has been shortened to ten years, and pension funds are obliged to apply pension curtailments if the funding ratio remains below the minimum required funding ratio (104.3% at the time) for five consecutive years. The nFTK introduced new reporting obligations for pension funds, enabling DNB to monitor their financial health and intervene with recovery plans if financial positions deteriorate, thereby mitigating underfunding issues.

4.1. Pension funds’ funding ratios and household savings

This section investigates whether shocks to the financial position of occupational pension funds impact the individual saving decisions of pension fund members. To answer this question, the identification strategy exploits the cross-sectional and time variations in the financial position of pension funds, as well as the heterogeneity in the type and size of the shocks to their financial position. We estimate the following equation:

(1)\begin{equation}{s_{i,p,t}} = {\text{ }}{c_p} + {\beta _1}{\text{ }}shock{{\text{ }}_{p,t}} + \delta '{X_{i,t}} + {\text{ }}\gamma '{Z_t} + {\varepsilon _{i,p,t}}\end{equation}

where ${s_{i,p,t}}{\text{ }}$represents active savingsFootnote ⁶ of individual $i$, member of pension fund $p$, at time $t$. Active saving is a flow measured in EUR and represents a measure of savings that is not attributable to capital gains. This information is retrieved from the DHS, where respondents are asked how much money they had put aside in the last 12 months and whether they are able to make ends meet.Footnote ⁷ The main independent variable $shock{{\text{ }}_{p,t}}$ represents the shock to the financial position of pension fund $p$ at time $t$, proxied by the level and change of the funding ratio.

Equation (1) controls for a set of variables ${X_{i,t}}$ potentially affecting individual saving rates. In particular, we include the net disposable income, a polynomial in age, marital status dummies, employment dummies, a gender dummy, a homeownership dummy, and health-related characteristics such as expected life-expectancy and self-assessed health condition. Moreover, we include the self-reported measures of risk-aversion and risk-attitude as binary indicators equal to 1 for respondents who ‘agree’ or ‘largely agree’ with the statements ‘it is important to have safe investments and guaranteed returns’, ‘I would never consider investments in shares because I find this too risky’ and ‘I want to be certain that my investments are safe’. Finally, ${c_p}{\text{ }}$represents a set of pension-fund fixed effects and ${Z_t}$ contains macroeconomic variables that jointly affect the development of funding ratios as well as individual savings, and ${\varepsilon _{i,p,t}}$ is the error term. The macroeconomic variables used are the long-term interest rate, the growth rate of equity prices and the percentage growth rate of employment in the population, all available from FRED .Footnote ⁸ The long-term interest rate determines the base market interest rate for long-term investors such as pension funds. The growth rate of equity prices, together with the long-term interest rate, explains a large fraction of the variation of the funding ratios (Kroon et al., Reference Kroon, Wouters and Leote de Carvalho2017) as pension funds invest heavily in equities and long-term government bonds given their long-term investment horizon. The percentage growth rate in employment of the population accounts for the fact that pension contributions – and thus pension funds liabilities – are not constant over the business cycle, but also depend on the number of active workers. The specification also controls for an index measuring the average pension contribution amount as a share of total wage, over time. This is used to account for potential feedback effects, that is, the impact of increasing pension contributions on the funding ratio and, indirectly, on household savings. This is because pension contributions have been increased over time (by about 40% over the 2007–2020 period) in an attempt to cope with the long-term effects of ageing population on pension sustainability.Footnote ⁹

Results are displayed in Table 2 (see Table A2 in the Appendix for a detailed specification, including all control variables). Specification (a) controls for the set of macroeconomic variables. Specification (b) further controls for the self-reported measures of risk-aversion and risk-attitude. The advantage of controlling for the level of interest rate, inflation and employment growth is to account for time-specific effects that impact both the level of household saving and the funding ratio of pension funds. The advantage of controlling for these soft measures of risk and savings preference is to account for potential increases in saving rates explained by individual attitudes but correlated with drops in funding ratios: when uncertainty increases households typically increase their savings (precautionary saving) while financial markets usually feature asset price corrections, affecting the investment value of institutional investors such as pension funds and, thus, their funding ratio.

Table 2. Main results

Notes: The dependent and main independent variables are transformed using the inverse hyperbolic sine (i.h.s.) All specifications are estimated via OLS. The control variables include net disposable income, a polynomial in age, year of birth cohort dummies, marital status dummies, employment dummies, a gender dummy, a homeownership dummy, and health-related characteristics such as expected life-expectancy and self-assessed health condition, the set of macroeconomic variables (pension premiums as percentage of wage, long-term interest rate, employment growth and growth rate of equity prices) and self-reported measures of risk-aversion and risk-attitude. Clustered standard errors by pension fund, in parentheses. Significance level:

*** p < .01, **p < .05, *p < .1.

Results show that when the funding ratio of a pension fund declines, savings increase among households that are members of that pension fund. The estimates suggest that a higher value of the funding ratio (one percentage point) translates, on average, into a 0.04 percent lower private active savings. Instead, a 1 percent decline in the funding ratio over time is associated with a 0.10 percent average increase in household savings. The main interpretation is that when the funding ratio declines, members of that pension fund anticipate a lower expected pension wealth. Lower pension wealth, for any levesl of pension contributions, should lead to an increase in private voluntary savings to maintain higher levels of consumption at the retirement age. In terms of size, the effect is stable across specifications. The first coefficient suggests that on average, a one percentage point decline in the funding ratio (from the average of 102% to 101%) increases mean active savings by 4.5 percent.Footnote ¹⁰ A back of the envelope calculation suggests that additional active savings amount to about 200 euros per year. The size of the effect is relatively large but suggests no full compensation, as a one percentage point shock in pension wealth (proxied by a 1 percentage point decrease in the funding ration) corresponds to about 1.500 euros, given that the average pension wealth in the Netherlands is 150.000 euros.

4.2. Funding deficits and household savings

This section presents an in-depth analysis of the types of shocks affecting the financial position of pension funds and their impact on household savings. It explores the connection between a pension fund’s deficit and the saving behaviour of its members, as shown in Table 3.

Table 3. Other type of shocks

Notes: The dependent variable is transformed using the inverse hyperbolic sine (i.h.s.) All specifications are estimated via OLS. The control variables include net disposable income, a polynomial in age, year of birth cohort dummies, marital status dummies, employment dummies, a gender dummy, a homeownership dummy, and health-related characteristics such as expected life-expectancy and self-assessed health condition, the set of macroeconomic variables (pension premiums as percentage of wage, long-term interest rate, employment growth and growth rate of equity prices) and self-reported measures of risk-aversion and risk-attitude. Clustered standard errors by pension fund, in parentheses. Reference number of observations, column (b) of Table 2. The drop of observations in columns (c) and (d) is due to missing values in the corresponding independent variables. Significance level:

*** p < .01, **p < .05, *p < .1.

Pension funds are in a funding deficit when the level of the funding ratio goes below certain minimum thresholds. A first important boundary is the required coverage ratio. If the funding ratio drops below this boundary, pension funds must submit a recovery plan to DNB. This recovery plan sets out how a pension fund intends to ensure that the funding ratio returns to a level above the required funding ratio. Two types of minimum requirements exist: the so-called strategic and the current minimum requirement. The difference between the two is that the first is based on the strategic (or target) asset allocation, while a the second is based on the actual asset allocation at one particular moment in time. A second important boundary is the absolute minimum required coverage ratio, set as 104.3 percent. If the funding ratio drops below this level, the pension fund must take actions to improve its financial position. Possible actions include a stop to pension indexation (to inflation), increases in pension contributions, or, in extreme cases, pension curtailments. If the funding ratio goes below 100 percent, the value of current assets is lower than the value of current liabilities, so a fund will almost certainly have to reduce pensions in order to restore a sound financial position. Results are reported in Table 3. These are obtained by re-estimating eq. (1) using, as the main independent variable ${shock_{p,t}}$, a binary indicator equal to one for pension funds in a funding deficit, based on different values of the funding ratio and the recovery measures imposed on the pension fund. In detail, specifications (a) and (b) investigate the impact of pension funds going underfunded, using the current minimum requirement and the strategic thresholds, respectively. Specification (c) and (d) investigate the worst-case situations of a funding ratio going below the absolute minimum requirement and the 100 percent level, respectively. Finally, specification (e) and (f) investigate the impact on household savings of the follow-up actions imposed on the pension fund (stops to conditional indexation and pension curtailment, respectively) after a situation of underfunding has occurred and a recovery plan has been set out.

The evidence reported in Table 3 suggests that a household’s savings response to a situation of underfunding is statistically significant only when the financial position is compromised, and a follow-up action is more likely. Specifically, households do not increase their savings when the funding ratio of their pension fund goes below the minimum requirement (either strategic or actual), and the pension fund is asked to come up with a recovery plan. Instead, households increase their savings when the funding ratio reaches the absolute minimum requirement. This situation, unlike from the case where the funding ratio goes below the current or strategic minimum requirement, corresponds to a much more deteriorated financial position and implies the need for a follow-up action such as a stop to pensions indexation, an increase in pension contributions or a pension curtailment. Therefore, a funding ratio below these levels likely anticipates shocks to the pension wealth of their members. One important consideration is that, while not statistically significant, the point estimate of the impact of pension curtailments is negative, suggesting a decrease in savings, differently from all other shocks. The explanation of a negative effect likely relates to budget constraints effects. While shocks such as changes in funding ratio, stops to pension indexation or underfunding situations impact the expectation of future pension benefits, pension cuts lead to current budget constrain shocks and higher liquidity constraints on current pensioners,Footnote ¹¹ which limit the extent to which they can save. Pension curtailments are a last resort solution in case of large financial imbalances and in fact have a low incidence in our sample, relative to the other shocks considered.

4.3. Transmission channels

The previous sections document a robust relationship between household savings and shocks to the financial position of pension funds. This section builds upon the previous results by further investigating potential explanations and channels involved. We do this by re-estimating eq. (1) using a different set of dependent variables. Descriptive statistics on these variables are reported in Figure 4, while empirical results are provided in Table 4.

Notes: The survey questions refer to adjusting behaviour as a consequence of pension cuts. The replacement rate refers to the expected ratio between last (gross) income and first pension benefit.

Figure 4. Description of variables representing potential transmission channels.

Table 4. Transmission channels

Notes: All specifications are estimated via OLS. The control variables include net disposable income, a polynomial in age, year of birth cohort dummies, marital status dummies, employment dummies, a gender dummy, a homeownership dummy, and health-related characteristics such as expected life-expectancy and self-assessed health condition, the set of macroeconomic variables (pension premiums as percentage of wage, long-term interest rate, employment growth and growth rate of equity prices) and self-reported measures of risk-aversion and risk-attitude. Clustered standard errors by pension fund, in parentheses. Reference number of observations, column (b) of Table 2. The drop of observations is due to missing values in the new dependent variables. Significance level:

*** p < .01, **p < .05, *p < .1.

First, we investigate hypothetical behaviour if pensions were reduced (column ‘a’ in Table 4). For this, we use the elicited responses to a question in the DHS data, where respondents are asked: ‘Will you adjust your behaviour if pensions are reduced, for example through an adjustment on the indexation, postponement of the retirement age, or in case of a different pension system?’. Here, we distinguish the cases where the respondents reply ‘Yes, I will put more money aside for my pension’ from all other possible answers (‘No, I will see what I’ll do when it happens’; ‘No, I think I can make ends meet fairly easily with the pension I will have’ or ‘Otherwise’). Contrary to our main specification relating pension fund shocks to actual household behaviour, this specification tests whether there is any effect on household intentions. Results in column (a) of Table 4 indicate that changes in the funding ratio do not affect people’s responses regarding their intentions in case of a pension curtailment.

Second, we test whether shocks to the funding ratio affect the expectations of pension fund members regarding the amount of their pension wealth. To do this, we use the expected replacement rate as a dependent variable in a regression specification analogous to eq. (1). Results in column (b) of Table 4 confirm that higher funding ratios directly affect people’s expected replacement rates.Footnote ¹² In particular, a one percentage point increase in the funding ratio is associated with an increase in the expected replacement rate by 0.06 percent, and the effect is statistically significant. The replacement rate is defined as the expected pension benefit, as a percentage of the final year’s net income. This result is in line with Van Santen (Reference Van Santen2019) for the 2009–2011 period.

Third, we test whether shocks to pension funds affect people’s willingness to top up their pension through other pension arrangements, such as annuities, life-insurance policies, or via extra pension rights acquired from the employer.Footnote ¹³ The results of Table 2 suggest that negative changes to the funding ratio are associated with higher voluntary savings by pension fund members, who may also want to put money aside to supplement their pension using retirement saving products. For instance, life insurance policies require holders to make periodic payments that respondents may use as commitment devices. Results in column (c) exclude this hypothesis, as lower funding ratios are not associated with higher willingness by respondents to top up their pension through other arrangements.

In addition, we test whether shocks to funding ratios affect people’s expectations regarding their retirement age. If people associate lower funding ratios with lower expected pension benefits, as shown in column (b) of Table 4, they might postpone the retirement age to maintain higher levels of pension benefits when retired. For instance, they might decide not to take advantage of early-retirement options. Results in column (d) also exclude this possibility, as we find no statistically significant association between the level of the funding ratio and the self-reported expectation about the retirement age.

Lastly, we examine respondents’ replies regarding their perceived importance of retirement savings depending on the financial position of their pension funds.Footnote ¹⁴ Results in column (e) suggest that a higher funding ratio is not associated with a higher perceived importance of savings for retirement. This suggests that workers with retirement savings preferences do not select into occupations and sectors covered by pension funds with higher levels of the funding ratio. This is also confirmed by the results of specification (f) which tests whether lower levels of the funding ratios lead respondents to shift to another pension fund the following year. This potential selection concern is further ruled out by the non-significant estimate in specification (f) of Table 4.

Overall, the results of this section suggest that lower funding ratios are mainly perceived as negative shocks to expected future wealth, given the evidence of lower expected replacement rates, which induce higher active savings. This is consistent with the fact that all respondents in the sample are currently active members of a pension fund, and report a current labour income, so lower funding ratios do not necessarily imply an immediate shock to their budget constraint and saving possibilities. With regards to their actual behaviour, the findings from Table 2 indicate higher active savings associated with lower funding ratios. Together with the evidence from Table 4, showing a non-statistically significant effect on the expected retirement age, the take-up of other pension arrangements and on the probability to switch fund, it suggests that workers mainly respond to a shock to their future wealth via voluntary private savings. They do so without changing their plans about retirement, (mandatory) retirement saving, or even employment (changing occupation allows worker to shift the already-accumulated pension wealth to other occupational funds). This in turn suggests a consumption-smoothing motive of households, as lower future wealth is replaced by higher current savings. This also suggests a shift away from the three-pillar pension system, as workers accumulate additional private voluntary savings held in current accounts or other financial assets, but do not take-up pension savings product such as annuity insurances, tax-efficient blocked savings accounts or complementary pension plans. Regarding households’ perceptions and hypothetical behaviour, the lack of correlation between the level of the funding ratios and the importance of retirement savings or need to deal with possible pension curtailments suggests that results are not driven by selection, whereby individuals with stronger preferences for retirement savings select into pension funds with higher funding ratios.

4.4. Displacement effect

Section 4.1 provides evidence that the relation between pension fund funding ratios and individual active savings can be explained by the impact of the financial position of pension funds on the expected pension wealth of their members. This section builds further on this result by providing a direct estimate of the displacement effect implied by the negative relation between funding ratios. In fact, the result of Table 2 cannot be interpreted as an estimate of the displacement effect directly, as there is no clear correspondence between changes in funding ratios and changes in individual pension wealth. To do this, we estimate a specification similar to Van Santen (Reference Van Santen2019). In particular, we regress active savings on the expected replacement rate elicited by the respondents in the DHS, and we use the pension fund funding ratio as an instrument for this measure. In other words, we use specification (b) in Table 4 as a first-stage regression of a 2SLS estimate of the displacement effect. Results are provided in Table 5. Specifications (a) and (b) show the baseline estimates of the displacement effect. Specifications (c) and (d) further account for the variance of the funding ratio as part of the regressors to account for the role of uncertainty. The variance of the replacement rate, here proxied by the variance of individual funding ratios, has been found to be negatively correlated with the level of the funding ratio in Van Santen (Reference Van Santen2019), thus potentially causing a downward bias in the estimate of the displacement effect.

Table 5. Displacement effect

Notes: The dependent variable is transformed using the inverse hyperbolic sine (i.h.s.) All specifications are estimated via 2SLS. The control variables include net disposable income, a polynomial in age, year of birth cohort dummies, marital status dummies, employment dummies, a gender dummy, a homeownership dummy, and health-related characteristics such as expected life-expectancy and self-assessed health condition, the set of macroeconomic variables (pension premiums as percentage of wage, long-term interest rate, employment growth and growth rate of equity prices) and self-reported measures of risk-aversion and risk-attitude. Robust standard errors between parentheses. Reference number of observations, column (b) of Table 2. The drop of observations is due to missing values in the independent variables. Significance level:

*** p < .01, **p < .05, *p < .1.

The results of Table 5 suggest a displacement effect of up to 63 percent or, when the variance of the funding ratio is accounted for, up to 42 percent. This suggests that a one percentage point increase in the expected pension benefit (as proxied by the replacement rate) is associated with a decrease in private voluntary savings of 0.42–0.63 percentage points. The complete specifications that include individual risk-aversion variables (specifications b and d) suggest estimates that are mildly significant, with a 90 percent confidence interval. The finding of a lower point estimate of the displacement effect in the specifications that include the variance of the funding ratio (specifications c and d) suggest evidence of a downward bias in the baseline estimates. It should be noted that, unlike in Van Santen (Reference Van Santen2019), the estimates obtained from the same specification but on a larger sample period covering three major financial crises suggest a potential weak instrument issue when using the funding ratio as instrument for the expected replacement rate. This is suggested by the first-stage F-statistic falling below the rule-of-thumb threshold of 10 (Stock and Yogo, Reference Stock and Yogo2005) despite the significant relationship between the funding ratio and the replacement rate found in Table 4. Additionally, the size of this estimate is higher than in Van Santen (Reference Van Santen2019), which identified a 32 percent displacement effect for the 2009–2011 period and, more broadly, it is well aligned with the range of estimates found in the literature in the Dutch (Alessie et al. (Reference Alessie, Angelini and Van Santen2013), Mastrogiacomo et al. (Reference Mastrogiacomo, Dillingh and Li2023)) and the international context (Attanasio and Rohwedder (Reference Attanasio and Rohwedder2003), Attanasio and Brugiavini (Reference Attanasio and Brugiavini2003), Hurd et al. (Reference Hurd, Michaud and Rohwedder2012), Chetty et al. (Reference Chetty, Friedman, Leth-Petersen, Heien Nielsen and Olsen2014), Börsch-Supan et al. (Reference Börsch-Supan, Reil-Held and Schunk2008), and Bottazzi et al., Reference Bottazzi, Jappelli and Padula2011).