1  Introduction

The fiscal costs of meeting public social expenditures are widely expected to rise in most industrialised countries as their populations age.[1] Policies are already being implemented to deal with the increased share of the Gross Domestic Product (GDP) expected to be devoted to meeting social expenditures. Some countries have reduced eligibility for publicly provided services and transfers, while others have reduced the level of benefits. In the case of the universal superannuation scheme in New Zealand, the government, based on expected increases in the fiscal cost, has introduced a prefunding mechanism. In this way, today’s taxpayers contribute to a common fund which can subsequently be drawn upon to reduce the tax burden on future generations. Expectations about future fiscal costs are reflected in policy initiatives taken today. This paper presents projections for social expenditures in New Zealand over the next 50 years, as a contribution to the stock of information on which such policies can be based. It is precisely on the basis of projections of future social expenditures that policy decisions are being taken by governments today.[2]

Projecting social expenditures requires a range of assumptions about future paths of fertility, mortality, migration, labour force participation, unemployment and productivity growth, together with assumptions about the social policies governing future expenditures. There is substantial uncertainty surrounding projections of these underlying variables. Demographic projections typically recognise this uncertainty by conducting a sensitivity analysis using, for example, high medium and low values for some of the key variables.[3] In contrast, this paper examines the statistical properties of social expenditure projections. By specifying distributions of the relevant variables, simulation methods are used to translate the inherent variability of the component variables into variability of the projected social expenditures. Scenario based approaches, while providing a general sense of the possible range of outcomes, do not offer these distributional insights. A principal contribution of this paper is to provide such distributions through stochastic simulation.

A US Congressional Budget Office study of the financial balances of the Social Security trust Fund described stochastic simulation as follows:[4]

The ideal approach would be to assign a probability to every possible combination of paths for input assumptions, solve for the system’s finances under each set of paths, and then use the probabilities associated with each set of inputs to assign probabilities to every set of outcomes. Although it is impossible technically to assign probabilities to every set of outcomes, it is feasible to create an arbitrarily large sample of input combinations, solving each time for system finances, and then evaluate how finances vary within that sample and draw conclusions about the probability distribution of the outcomes…that technique (is) called stochastic simulation.

(Congressional Budget Office 2001):53).

Other studies for New Zealand which project the fiscal costs of population ageing include Bagrie who used a simplified model with labour productivity growing at 1.9%, and per capita social expenditure for all ages growing at an equivalent rate (Bagrie 1997). His study focused on the implications for revenue and expenditure and the sustainability of the fiscal settings. Polackova used a similar approach to examine the public sector balance (Polackova 1997), while Cook and Savage included an exploration of the net debt position (Cook and Savage 1995).[5] While the overall results for social spending are in line with the findings of the present study, uncertainty was handled through the conventional approach of alternative scenarios.[6]

Section 2 provides an overview of the key relationships together with a formal statement of the model. In Section 3, details of the data are set out, while Section 4 presents the benchmark results. The effect of varying some key assumptions is explored in Section 5. Conclusions are in Section 6.