This paper aims to analyse how consumption and saving decisions depend upon uncertainty about employment and upon the employment status itself. The probability of being unemployed, and more precisely the dependence of this on the current employment status, should have an effect on precautionary saving. The paper is an empirical investigation, based on experimental data, of saving behaviour when the states of being employed or unemployed follow a first-order Markov process; that is, when the probability of being employed (and hence unemployed) in any one period depends upon the employment status in the immediately preceding period.

The innovative feature of the experiment presented in this paper is that the experiment was run on a sample of Dutch households forming part of the ongoing survey by CenTER at the University of Tilburg. This panel has been closely monitored by CenTER over a period of years and provides a rich body of information is concerning the characteristics of the households in the sample: in addition to demographic data there are the households’ responses to a variety of questions concerning their consumption and saving behaviour. Additionally, the fact that the households participating as subjects in the experiment have known characteristics, enables their representativeness in terms of the typical Dutch household to be assessed. In turn, this means that the experimental results can be used to give a picture of the responses of the typical Dutch household to the saving problem presented in the experiment.

The motivation of this study was to test how closely the predictions of the optimality theory fit the actual behaviour of subjects in an experimental setting. An important empirical question which this experiment enables us to answer is whether, in this setting, subjects follow the appropriate optimising rule, and particularly whether they correctly take into account their employment status, or whether, for example they over-consume in periods of employment and under-consume in periods of unemployment.

The paper uses the results of a simulation of the optimal behaviour (a simulation is necessary here because an explicit analytical solution is not available) and tests the predictions of the theory experimentally.

Furthermore, the paper presents a comparison of the results of this experiment (with a sample of households from the Dutch population) with the results of a previous experiment (with the same experiment design) on a sample of students from the University of York (the more normal kind of experimental subject). This comparison gives important insight into the representativeness of the usual kind of experimental subject pool.