Tax credits, labour participation and home production in the Netherlands

Abstract

We set up a dynamic reduced form model of labour market participation for women who balance career and motherhood. The model accounts for the occurrence of future child birth and early retirement, and includes home production; however, it does not require the estimation of a structural model. Careful implementation of pension institutions can return optimal life patterns of participation without the need of a structural approach. The weaker theoretical framework is compensated by the rich spectrum of possible policy simulations. As illustration, we simulate the effect of two tax credits policy options on the hazard rate out of work.

Acknowledgements

We thank Patricia Apps, Adriaan Kalwij, Marcel Lever, Rocus van Opstal and the participants of the European Society for Population Economics (ESPE) conference (Paris, 16–18 June 2005) for their useful comments.

Notes

¹ In the data available, only the cross-sectional evidence testifies of this partial return to the labour market. The panel shows no evidence of substantial re-entrances as it does not track women long enough over time to observe their return to the labour market.

² Full income is the sum of the value of hours supplied on the market (wage) plus the value of hours reserved for home production. We impute this second value on the base of the time spent in home production paid at the hourly wage of home assistants. We follow the so called ‘input approach’ to the evaluation of home production (see for instance Kerkhofs and Kooreman, 2003).

³ We refer to a companion study (Mastrogiacomo and Bosch, 2006) for further details on the tax credits, the imputation of the probability distribution of a birth event and early retirement, re-entrances into the labour market after child birth, our income incentive measure to participation and home production.

⁴ Kalwij (2005) has shown significant presence of excess sensitivity of consumption to income in the Netherlands, which makes our substitution of consumption with income less of a severe problem as it may seem. Extending the model up to including a budget constraint with savings could therefore complicate the computations without returning a better understanding of the underlying consumption dynamics. Note that the macro saving rates in the Netherlands is currently extremely low (Serres and Pelgrin, 2002) and that micro studies on household portfolio show that participation into financial assets is low, being most of the household wealth represented by home ownership (Schors et al., 2007). Further we would ideally model the budget constraint using consumption data (including for instance child expenditures), but no panel data exists for the Netherlands that contains these information. Introducing home production the way we do seems our best option in this sense.

⁵ How much less is established on the basis of the computations in Fig. 1. This means that we assume a strict proportionality between hours worked in t − 1 and t if in t child birth is expected. This is a very strong assumption since it assumes a (stochastic) drop in participation that is actually what the model may aim to explain. As will be clear later, we will introduce the presence of children in the initial condition equation that will be the exclusion restriction explaining observed participation at the first observation in time. The proportional drop in participation plays a role exclusively in determining the expected pattern of future income and participation, of course not the observed choices (that are indeed the revealed preferences in the model).

⁶ We can also change the specification of the intratemporal utility function a bit, e.g. in the following way:

Obviously, the variables y_{i , t} and y_{i , t −1} should not be perfectly collinear. If we write down the function corresponding to within period preferences, we can show that one identifies separately the impact of habit formation if transitions from nonwork to work are allowed. In his model Blau (1998) allows for habit formation. He is able to do that because in the United States a retired person regularly returns to the labour market, while in our panel re-entrances are negligible.

⁷ Notice that, since we assume retirement as an absorbing state (that implies y _{i,t −1} = 1 , t = 1, … ,T), j is always equal to 1.

⁸ We do not aim to estimate a full wage equation. We aim to get an estimate of the parameters that allows imputation of future wages (Burkhauser et al., 2004). As we are only interested in the fit the consistency of our results is not relevant.

⁹ To ensure that the fit is not affected by endogeneity issues (ability affects both income and education) we have also instrumented education using the age of entrance into the labour market (results available from the author). This instrument is definitely relevant as those with higher education enter the labour market later in life. Exogeneity to income is harder to defend. Early entrants may be unskilled and therefore earn less. However,they could also have more time to start up an own business and eventually earn more money. If the instrument is not entirely exogenous it is however unclear what the direction is of the relation between wage and age of entrance into the labour market. No other relevant instruments were available. The effect of education turns out to be underestimated relative to the Instrumental Variable (IV) estimates, nonetheless the imputed wage patterns based on both models (instrumented and not) hardly differ, as expected.

¹⁰ Notice that R^p does not vary with time since the early retirement schemes we describe in this article are nonactuarially fair.

¹¹ The duration of the unemployment benefit depends on the amount of years worked before t.