Who Bears the Burden of Social Security Contributions in the Netherlands? Evidence from Dutch Administrative Data

Abstract

This paper sheds light on the incidence of social security contributions (SSC) in the Netherlands. Our unique dataset on earnings inclusive and exclusive of these SSC enables us to apply the methodology by Alvaredo et al. (De Econ. doi:10.1007/s10645-017-9294-7, 2017) and draw clear conclusions on local incidence. First, our finding of a smooth distribution of gross earnings indicates that both employer and employee do not shift their contributions. This contradicts the standard incidence prediction of full shifting of SSC to employees but corroborates recent findings. Moreover, it is hard to reconcile with the standard static model of labour supply and demand where gross earnings are irrelevant. Second, our finding of a discontinuity in labour costs supports our conclusion of non-shifting and renders out measurement error as an alternative explanation. Overall, these findings suggest that the statutory split matters and that the burden of SSC close to thresholds is borne by employers.

Appendix

1.1 McCrary Test

Applying the local linear density estimator requires two steps (McCrary 2008). First, a histogram of the relative distance to the earnings threshold is calculated and bins are formed to the left and right of the threshold. This results in bin grid $X_{1}, X_{2}, \ldots , X_{J}$ of width b. The cellsize $Y_{j}$ is normalized and equals $Y_{j}=(1/nb)\sum \mathbf 1 (g(R_{i})=X_{j})$. The histogram is just the scatterplot of ($X_{j},Y_{j}$). Second, separate local linear smoothers are applied to bins to the left and to the right of the threshold. The midpoints of the bins are the regressors ($X_{j} - c$) and the normalized counts of the observations in each bin are the outcome variables. To be precise, the density estimate at r is given by $\hat{f}(r)=\hat{\phi _{1}}$. With ($\hat{\phi _{1}}, \hat{\phi _{2}}$ ) choosen to minimize $L(\hat{\phi _{1}}, \hat{\phi _{2}}, r) = \sum \{ Y_{j} - \phi _{1} - \phi _{2} (X_{j} - r) \}^{2} K((X_{j} - r)/h)\{ \mathbf 1 (X_{j} > c) \mathbf 1 (r \ge c) + \mathbf 1 (X_{j}< c) \mathbf 1 (r < c) \}$. Here K(.) denotes a kernel function and the number of observations used in the regression is given by h, which is the bandwidth. This step adopts a weighted regression where the height of the bins are explained by the midpoints of the bins and most weight is given to the bins nearest to the threshold $\theta$.

Formally, a density function, f(r) of running variable $R_{i}$ is evaluated. The parameter of interest is the log difference in height at the threshold $\theta$.

$$\begin{aligned} \theta = \ln \lim _{r \downarrow c} f(r) - \ln \lim _{r \uparrow c} f(r) \equiv \ln f^{+} - \ln f^{-} \end{aligned}$$

In practice, one estimates two separate local linear regressions for bin points to the left and to the right of the threshold c. The log difference of the coefficients on the intercepts then estimates $\theta$.

$$\begin{aligned} \theta= & {} \ln \left\{ \sum _{X_{j}>c} K \left( \frac{X_{j} - c}{h}\right) \frac{S_{n,2}^{+} - S_{n,1}^{+}(X_{j} - c)}{S_{n,2}^{+} S_{n,0}^{+} - (S_{n,1}^{+})^{2}} Y_{j} \right\} \nonumber \\&- \ln \left\{ \sum _{X_{j}<c} K \left( \frac{X_{j} - c}{h}\right) \frac{S_{n,2}^{-} - S_{n,1}^{-}(X_{j} - c)}{S_{n,2}^{-} S_{n,0}^{-} - (S_{n,1}^{-})^{2}} Y_{j} \right\} \end{aligned}$$

(2)

where $S_{n,k}^{+} = \sum _{X_{j} > c}K((X_{j} - c)/h)(X_{j} - c)^{k}$ and $S_{n,k}^{-} = \sum _{X_{j} < c}K((X_{j} - c)/h)(X_{j} - c)^{k}$.

1.2 Polynomial Test

In this paper the polynomial test is applied in two steps. The first step is equivalent to the McCrary test where individual observations are aggregated into bins. In the second step a polynomial is fitted to the counts in each bin and a dummy variable is included indicating bins to the right of the threshold. The following regression is estimated $C_{j} = \sum _{i=0}^{q} \beta _{i}^{}(X_{j} - c)^{i} + \gamma {}^{} \mathbf 1 [X_{j} - c \ge 0] + \epsilon _{j}^{}$. Where $C_{j}$ denotes the number of individuals in income bin j, q is the order of the polynomial and the indicator function $\mathbf{1}$ measures the difference before and after the threshold. Normalizing this coefficient with the average density slightly below the cap results in our estimate of the discontinuity.