Retirement behaviour is the theme of increasing interest in the recent labour economic studies due to its growing importance. In most countries which adopted pay-as-you-go public pension system the relative number of pensioners is increasing together with the level of life. This puts additional fiscal weight on the taxpayers who are financing current pensions. In order to compensate this demographic dynamics the governments have to take certain steps to improve labour force participation and stimulate additional supplementary pensions. To test possible policies in this field different models can be estimated to describe people s behaviour on the labour market.Current study presents several structural forward-looking quasi-dynamic models of households to serve the purpose of policy simulations. The models are set up in the random utility framework used by Thurstone (1927), McFadden (1973), Ben-Akiva, Lerman (1985) and spread widely in the discrete choice analysis. This approach suggests that the rational decision maker maximizes over the set of choices an utility function with determined and stochastic components. Stochastic part of the utility takes care of unobserved by the researcher factors and is assumed to be perfectly known to the agent. The determined part depends on the characteristics of either the agent or the alternatives, or both, and allows any specifications. Several utility specifications make up several models to be estimated and compared.
Since the utility is specified as dependent on the fundamental attributes of the available options such as disposable income and leisure, the model appears structural other than reduced form. Structural models yield much better, more accurate results in simulations since they reflect all direct and indirect effects the policy may have on the taken choices.
In the current study two revisions of the standard multinomial logit model are developed. It is taken into account that not only the current period utility is considered by the choice maker, but also the discounted sum of future periods utility. In other words, the agent plans the utility flow for several years ahead. Correspondingly, the two models reflect two and tree period setups in the first one the agent is allowed to consider the next period consequences of the current choice, and in the second one the planning horizon rises to two periods. The future periods considerations incorporated into the models make them forward-looking. These forward-looking issues are quite substantial when studying the retirement behaviour due to the fundamental property of being retired once a person has retired it is hardly possible to start working again. In the models we assume the retirement state to be strictly absorbing. Thus, once people choose retirement, they can not go back to any other state. Keeping this in mind, they will correct their retirement decisions compared to simple one period logic.
The models are quasi-dynamic due to the fact that even though the time scale is present, only one static choice is modelled. Namely, at the eligibility point new option (to retire) becomes available to an individual and the choice is made whether to retire or to stay in the labour force.
Finally, the sample is organized by households to take care of probable coordination between husband and wives as to retire simultaneously. Even if wife for example does not have retirement option available for her, she may change her labour force participation at the point when husband retires to take advantage of common leisure time. Thus, we assume some coordination within the household with respect to behaviour on the labour market and concentrate on studying coordinated choices of spouses.
Once the models are estimated and the best one chosen, some simple simulation is performed. We try altering taxation by rising the total amount of tax paid by households by 10%. These policy results in the change of distribution of households among states.
The thesis is organized as follows. Chapter 1 briefly describes the pension system in Norway and gives overview of the occurring rules. More detailed description with the mathematical formulas can be found in Appendix A. Chapter 2 is devoted to developing and presenting the models, but also a lot of attention is paid to describing the possible choice sets. One of the objectives of the current study was to keep the sample as vast as possible and not to exclude households with unusual wife s state. This resulted is several case divisions which are also discussed in chapter 2. Chapter 3 contains the sample data description used for estimating the models and concludes model specifications by introducing the deterministic part of the utility function. In chapter 4 the sequence of estimated models is followed, the models are compared and the best one is picked for simulations. Finally, chapter 5 presents the results of the simulation procedures. Appendix A contains accurate definitions of the procedures used for data construction while Appendix B has a set of table and graphs to visualize the sample.