VoxEU Column Labour Markets Welfare state and social Europe

Family and government insurance: Wages, earnings, and income risks in the Netherlands and the US

Understanding the source of fluctuations in earnings, and how workers insure themselves against those fluctuations, is key to evaluating labour laws. This column uses administrative data from the Netherlands to compare the role played by households to the tax and transfer system in mitigating shocks to individual earnings. It then compares those findings to data from the US – a country with a substantially smaller welfare state – and finds that hours, not wages, account for most of the variability in earnings for workers in the bottom two deciles of the earnings distribution.

Earnings changes are an important source of uncertainty during one’s working life. Workers may experience positive changes, such as a promotion, or negative changes, such as an illness or job displacement. Understanding the sources of these fluctuations and the ability of workers to insure themselves against them is key to evaluating potential reforms of labour laws and the welfare state. For instance, if earnings fluctuations are mostly related to changes in hours worked, job protection policies will have a larger insurance value than wage regulations.

Increased availability of administrative data has made it possible to study earnings risk in more detail. Guvenen et al. (2015) use US Social Security data to document that the dynamics of earnings changes do not satisfy the canonical assumptions of linearity and log-normality (e.g. Abowd and Card 1989). This means that in the data, and unlike in the canonical earnings process, negative earnings changes are more likely than positive earnings changes of the same magnitude (earnings changes are negatively skewed). Moreover, earnings vary little year on year for most workers, but display very large fluctuations for a small number of workers (earnings changes are leptokurtic – i.e. they are more concentrated around the mean and in the tails relative to a normal distribution). Unfortunately, the nature of the data set that they use implies that they can only study the dynamics of individual, pre-tax earnings and cannot separate the role of hours and wages.

In De Nardi et al. (2020), we make use of very rich administrative data from the Netherlands (2001–2014) that allows us to decompose the dynamics of earnings into the contribution of hours and wages and to study the role of the household and the tax and transfer system in mitigating shocks to individual, pre-tax earnings. Our analysis combines tax records and data from the Dutch payroll administration on the number of hours worked. We also compare our findings for the Netherlands with those from the Panel Study of Income Dynamics (PSID) in the US, a country that differs substantially in the size of its welfare state.

What are the properties of male earnings changes in the Netherlands? 

For the Netherlands, the distribution of earnings changes differs substantially by age and position in the income distribution, in line with what has been documented by Guvenen et al. (2015) and Arellano et al. (2017) for the US. 

The left panel of Figure 1 shows the standard deviation of (log) earnings changes. It is larger for workers in the lower and upper parts of the income distribution, and also for those at the beginning and end of their working lives. Flexible contracts are common among young workers in the Netherlands and might generate more variability in their earnings. Older workers are more likely to become eligible for partial disability benefits, which are recorded in the data as a reduction in earnings.

The middle panel shows that earnings changes are negatively skewed – negative changes are more likely than positive ones – for most age groups and across most of the earnings distribution. This is particularly true for workers at or above the median of the distribution of previous earnings. The right panel reports kurtosis: values larger than three imply that the probability mass is more concentrated around the mean and in the tails compared to the normal distribution. Here, values in excess of 20 and almost reaching 100 indicate that, for most workers, earnings shocks are infrequent, but tend to be of a large magnitude when they happen. This is particularly true for older workers, for whom employment protection is strongest in the Netherlands.

Figure 1 Standard deviation, skewness, and kurtosis of male earnings in the Netherlands

What are the drivers of male earnings changes in the Netherlands?

To understand the drivers of earnings changes, we decompose them into changes in wages and hours. Figure 2 decomposes the variance (left) and skewness (right) of earnings changes into the variances (skewness) of hours and wages and a residual (covariance/co-skewness). The left panel shows that for workers with relatively low previous earnings, hours changes are the major driver of the variance of earnings changes. The opposite is true for workers in the top decile of previous earnings, for which wages account for the variance of most earnings changes. The people in the latter group are mostly full-time workers who remain in full-time employment and whose hours, therefore, vary much less. For this group, changes in earnings likely reflect variable or performance-related components of earnings such as bonuses.

The right panel in Figure 3 shows that negative skewness is also driven by hours rather than wage fluctuations. This means that, for most workers, the predominance of negative earnings changes is driven by falls in hours.

Figure 2 Male earnings changes versus hours and wage changes in the Netherlands

What is the role of household and government insurance in the Netherlands?

To understand the roles of household and government insurance, we determine the pass-through of changes of male earnings to before- and after-tax household income. The three panels in Figure 3 shows, for three ranks of the income distribution, how a change in male earnings (x-axis) transmits into before-tax household income and after-tax household income (y-axis). Labour income pooling within the household implies a lower variability of household earnings compared with male earnings changes (turquoise line) for individuals in the lowest income decile. 

Conversely, the tax and transfer system effectively smooths negative fluctuations for workers across the income distribution and even more so for those in the lowest income decile (red line). However, it also cushions positive shocks for this income group much more than for the other two. For low-income individuals, increases in earnings result in reduced benefit eligibility. In contrast, when high-income individuals earn more, the smoothing mechanism that they face is progressive taxation. 

Figure 3 Household before- versus after-tax income in the Netherlands 

There are two main reasons why a second earner can reduce the impact of shocks to male earnings onto household earnings. The first reason is income pooling: a second earner being present implies that a share of household earnings is not affected by a change in male earnings. The second reason, often called the added worker effect, implies that the second earner might react to positive or negative shocks to her partner’s earnings by changing participation or the number of hours worked. Figure 4 disentangles the role of income pooling and added worker effects in generating within-household insurance. It reports the average change in women’s hours between years t and t+2, for those who were working in both years, as a response to changes in male earnings between t and t+1. The figure shows that there is no association between changes in male earnings and changes in women’s hours worked, indicating that it is mostly income pooling which explains the reduction in earnings risk at the household level that we have observed in the previous set of graphs.

Figure 4 Male earnings changes and female labour supply in the Netherlands

Figure 5 breaks down the role of various government programs for the 55-59 age group by sequentially adding specific transfer programs and taxes. It shows that disability insurance reduces the standard deviation of household earnings changes, particularly at low percentiles of previous earnings, while unemployment insurance (UI) reduces it even at higher levels of previous earnings. It also shows that, for this age group, (early) retirement transfers associated with early access to occupational pensions play a much larger role in reducing variation in household income than progressive taxes. The right-hand-side graph of Figure 4 shows that negative skewness is largely offset by taxes and transfers in the Netherlands.

Figure 5 Relative contribution of transfers and taxes to the standard deviation and skewness of household income in the Netherlands, ages 55-59

The Netherlands versus the US

When comparing risks and insurance in the Netherlands and the US, the first noticeable feature is that the standard deviations of wages, hours, earnings, household income, and disposable income are much larger in the US than in the Netherlands (Figure 6). In addition, the standard deviation of wages is much closer to that of hours in the US, suggesting that wage adjustments are more frequent in the US than in the Netherlands, but also that the PSID survey data may be more subject to measurement error in hours and, therefore, in wages. 

Comparing male to household earnings reveals a larger role for spousal insurance in the US (in line with Pruitt and Turner 2018, who use administrative data from the US). Finally, comparing before- and after-tax income reveals that, while government insurance reduces the variability of earnings changes in both countries, its role is much larger in the Netherlands, particularly for households at the bottom of the (previous) earnings distribution.

Figure 6 Standard deviation for various income definitions by previous earnings in the Netherlands versus the US


We document that male earnings dynamics in the Netherlands display important deviations from the typical assumptions of linearity and normality made by the canonical earnings model. These findings are important because, as shown by De Nardi et al. (2019), the earnings process that we use is crucial to determining to what extent a household can self-insure by using saving and, therefore, to what extent households value government insurance.

More specifically, we find that individual-level male earnings risk is relatively high at the beginning and end of the working life, and for those in the lower and upper parts of the income distribution. Hours, rather than wages, account for most of the earnings variability for workers in the bottom two deciles of the earnings distribution. Hours are also the main driver of the negative skewness and, to a lesser extent, the high kurtosis of earnings changes. Our findings on the drivers of the negative skewness of earnings are consistent with the evidence presented in Hoffman and Malacrino (2019) using Italian administrative data, but at odds with the findings of Halvorsen et al. (2019) using Norwegian data. These international differences suggest that the institutional framework that governs the labour market is crucial to determining the sources of earnings fluctuations and whether adjustment occurs at the hour or wage margin. 

Comparing family and government insurance, we show that the government plays a much larger role in reducing wage risk in the Netherlands compared to the US This is similar to the results of Halvorsen et al. (2019) for Norway.  A breakdown in government programmes for older workers in the Netherlands shows that disability insurance and UI programmes reduce income risk, especially for the lowest quarter of the male earnings distribution. Pensions and taxes (to a lower extent) reduce earnings risk across the whole distribution. Instead, in the US, the role that the family plays is much more important. The results suggest that taxes and transfers may crowd out insurance that could be generated within the family.


Abowd, J and C David (1989), “On the Covariance Structure of Earnings and Hours Changes”, Econometrica 57: 411–45. 

Arellano, M, R Blundell and S Bonhomme (2017), “Earnings and consumption dynamics: A non-linear panel data framework”, Econometrica 85(3): 693–734.

De Nardi, M, G Fella and G Paz-Pardo (2019), “Nonlinear Household Earnings Dynamics, Self-Insurance, and Welfare”, Journal of the European Economic Association 18(2): 890–926.

De Nardi, M, G Fella, G Paz-Pardo, M Knoef, R Van Ooijen (2020), “Family and Government Insurance: Wage, Earnings, and Income Risks in the Netherlands and the US”, NBER Working Paper 25832.

Guvenen, F, F Karahan, S Ozkan and J Song (2015), “What do data on millions of US workers reveal about life-cycle earnings risk?”, NBER Working Paper 20913.

Halvorsen, E, H Holter, S Ozkan and K Storesletten (2019), “Dissecting idiosyncratic income risk”, mimeo.

Hoffman, E B and D Malacrino (2019), “Employment time and the cyclicality of earnings growth”, Journal of Public Economics 169: 160–171.

Pruitt, S and N Turner (2018), “The nature of household labour income risk”, Finance and Economics Discussion series working paper 2018-034.

840 Reads