Back to top anchor
Policy perspectives paper

Modelling New Zealand's Long-term Fiscal Position (PP 06/01)

Acknowledgements

We would like to thank our colleagues in Treasury’s Financial Policy and Tax Modelling team, especially Dasha Leonova and Nathan McLellan for all the work they did producing the probabilistic projections in this paper.

Bob Buckle, Patrick Conway, John Janssen, Dasha Leonova, Struan Little, Michele Lloyd, Nathan McLellan and Brendon Riches provided numerous helpful comments.

We are also grateful for comments by external readers of early drafts: John Bryant, Kim Dunstan, Frank Eich and Paal Ulla.

Disclaimer

The views, opinions, findings, and conclusions or recommendations expressed in this Policy Perspectives Paper are strictly those of the authors. They do not necessarily reflect the views of the New Zealand Treasury.  The Treasury takes no responsibility for any errors or omissions in, or for the correctness of, the information contained in these Policy Perspectives Papers. The paper is presented not as policy, but to inform and stimulate wider debate.

Summary

The Public Finance Act (PFA) was amended in 2004 to require the Treasury to publish at least every four years a Statement on the long-term fiscal position looking out at least 40 years. The first Statement must be presented to Parliament by 30 June 2006.

This Policy Perspectives Paper sets out the key judgements that are needed to prepare the Statement, along with some illustrations of how the results could be displayed (for instance, confidence intervals around the median to convey uncertainty). The paper does not contain aggregate results, but discusses proposed methods and seeks to elicit comments and discussion.

The horizon to be covered by the Statement means that it has a different character from the existing PFA publications, such as the Economic and Fiscal Updates and the Fiscal Strategy Report, as it will extend across the terms of many more governments, with potentially many different policy regimes.

This paper uses the policy settings at the end of the official forecast in 2010 (“current policy”) as the base for forward projections of spending and revenue. The durability of “current policy” over a very long horizon is open to question. For instance, there may be a tension between top-down approaches determined by fiscal objectives and bottom-up extrapolations of existing trends in particular areas. Another example might be tensions over the long term between price and wage linkages for different social welfare benefits.

We have adopted a three-stage approach to projecting the long-term fiscal position.

First, we use Statistics New Zealand’s National Population Projections to derive a series of projections of the future structure of the population

We then use these projections of the population to generate projections of Gross Domestic Product (GDP).

Finally, we add projections of government spending and revenue.

The long-term nature of the Statement also means that its final results will be sensitive to small changes in some assumptions, while others will have a minor effect. We demonstrate the effects of changes in the some of the key demographic and economic assumptions.

We welcome feedback on the proposed model. Comments should be directed to either of the authors, before the end of March. Contact details are on page ii.

1.  Introduction

Mehemea kua otia e ahau te here atu i taku waka ki tōna tauranga, ā, kua tau taku hinengaro ki taku ahunga ki whea, he mea hanga i taku tūnga i runga ki ngā pakahiwi o te ao o nehe.[1]

The Public Finance Act requires the Treasury to publish at least every four years a Statement on the government’s fiscal position looking out at least 40 years.[2] The first such Statement must be tabled in Parliament by 30 June 2006.

The main focus of the paper is the judgements and assumptions, the risks and uncertainty that surround developing a set of long-term fiscal projections.

This Policy Perspectives Paper is the first step in meeting this new legislative requirement. It outlines the model we propose to use to prepare the fiscal projections to be included in the Statement.[3] The main areas of focus of the paper are the judgements and assumptions, the risks and uncertainty that surround developing a set of long-term fiscal projections. Exposing the modelling early is an important part of the process in producing the Statement. We are particularly interested in feedback on the degree of detail proposed and the assumptions we have made.

Using this model, the Statement will outline a set of fiscal projections going forward 40 or so years. The Statement will also discuss the origins of our present policy settings and the types of policy choices we could make in the next few years to increase the likelihood that the fiscal position remains strong over the decades ahead.

A brief summary of our model

We have adopted a three-stage approach to projecting the long-term fiscal position.

First, we use Statistics New Zealand’s National Population Projections to project the future structure of the population.

We then use these projections of the population to generate projections of gross domestic product (GDP).

Finally, we add the central government sector, in the form of projections of spending and revenue.

Outline of this paper

The next section of this paper summarises the work that we judge to be required to produce the Statement and also summarises our modelling strategy as well as our approach to dealing with uncertainty. The third section considers the assumptions made for the demographic projections, while the fourth does the same for GDP projections

The fifth section canvasses the necessary assumptions about projecting government spending and revenue and how these are modelled, and then uses several examples to illustrate the issues. This section also illustrates the effects of different choices for various parameters. The sixth and concluding section looks ahead to the Statement itself and how it will move beyond what has been developed in the present paper.

Feedback

We welcome feedback on the proposed model. Comments should be directed to either of the authors before the end of March. Our contact details are on page ii.

Notes

  • [1]This saying is quoted in the Whanganui Iwi Exhibition at Te Papa, 2003-06. In English: “If I am comfortable with where I am in the present and confident with where I will be in the future, it is only because I am standing on the shoulders of the past.”
  • [2]The Treasury publication A Guide to the Public Finance Act provides more details. It is available on our website at: http://www.treasury.govt.nz/pfa/default.asp. The “fiscal position” refers to a mix of indicators of the state of the Crown accounts: debt, net worth, operating balance, expenses, revenue, often stated as a ratio of nominal GDP. The Government’s fiscal objectives are expressed in terms of goals for these indicators.
  • [3]The choice of the term “projections” over the alternative “forecasts” is deliberate. A projection provides information about what could happen if no changes were made to present behaviour or policy. A forecast considers the likely changes and their interactions in assessing possible futures.

2. How we see the task at hand

The fiscal position at any point in time is the result of a series of policy choices made by governments over a long period.

Those choices are driven by a complex set of inter-connected influences. The state of the world (including the economic and social situation, in the past, the present and in the future), the government’s desired outcomes and the effectiveness or otherwise of outputs selected to achieve the desired outcomes, all combine to produce the fiscal outcome.

Given the vast numbers of variables that can and will affect future fiscal outcomes, the task of modelling 40 years ahead is no easy task.

A very simple approach, for example, might be to use some kind of trend GDP growth over the next 40 years, and take the average aggregate government spending-to-GDP and revenue-to-GDP ratios for the past decade and then use GDP to arrive at projections of spending and revenue. A clear weakness of this kind of approach is that it would miss out on capturing effects of the various drivers of more detailed spending (other than GDP).

It would, however, demonstrate a particular difficulty of long-term fiscal modelling: how to model the variability of past patterns of spending and revenue into the future. Figure 1 below is an example of one of the components of government spending that we need to model: core government services, which includes the cost of running government departments, Statistics New Zealand, the provision of Overseas Development Assistance, etc. Over the past 50 years, spending under this heading has fluctuated quite widely. Best practice fiscal modelling is unable to capture this degree of variability and produce projections that might mirror past performance (although probabilistic projections of expenditure might address this variability – see the later section on demography for more on this).

In this case, we simply grow out expenditure by nominal wage growth (the largest cost driver of these services) and assume that the share of public servants remains a constant proportion of total employment. This produces a track where long-term spending is a fixed proportion of GDP (Figure 1). If all components of spending and revenue are modelled on the same basis, then the end result becomes simply a “battle of the exponentials”: all the components of spending and revenue are growing at one rate or another and the outcome will be determined solely as a result of the differences in the rates of growth. Because of the long-term nature of the modelling, small changes in parameters could have large effects after 50 years.

Figure 1 Core government services expenditure projected to grow with GDP
Figure 1 Core government services expenditure projected to grow with GDP.
Source: The Treasury

Fortunately, there are some aspects of government spending that can be modelled with more precision. An example is New Zealand Superannuation, where we are able to combine demographic developments with other projections to develop a spending track such as that shown in Figure 2.

Two possible approaches

Other countries that prepare long-term fiscal statements use a variety of approaches to presenting potential outcomes. While there are differences in the details, the two major approaches used can be termed “top-down” and “bottom-up.”

The top-down approach would start with the current set of fiscal aggregate objectives (most notably long-run spending-to-GDP, tax-to-GDP and debt-to-GDP ratios) and determine what spending or revenue track would be required to continue to meet these objectives, given likely demographic changes.

The bottom-up approach involves modelling the effect on the aggregate fiscal results of current policy in individual spending programmes and the current tax system projected forward on the basis of demographic and other assumptions.

The top-down approach would start with the current set of fiscal aggregate objectives and determine what spending or revenue track would be required to continue to meet these objectives.

Three recent examples of countries producing long-term fiscal projections are Australia, the United Kingdom and the United States. The Australian Commonwealth Government produced an “Intergenerational Report” in 2002. This uses a bottom-up approach.[4] HM Treasury in the UK produces an annual “Long-term public finance report” which contains a mix of bottom-up and top-down projections.[5] The United States’ Congressional Budget Office prepares a Long-Term Budget Outlook.[6] This Outlook uses a bottom-up approach to model the effect of different scenarios of spending and revenue on the federal government’s fiscal balance and thus levels of national debt. In addition, the European Commission and the OECD periodically do bottom-up projections of their members’ fiscal positions. The EC published a set of such projections of age-related expenditure for its 25 member states in February 2006.[7]

The bottom-up approach involves modelling the effect on the aggregate fiscal results of current policy projected forward.

The difference between the two approaches should not be exaggerated. Both involve projecting forward individual elements of spending or revenue. The difference is that in the bottom-up approach, individual programmes are allowed to develop in isolation from the government’s overall fiscal objectives. With a top-down approach, some measure of the overall objectives acts as a constraint. In practice, a top-down approach usually involves allowing some programme to develop alone and then to use the overall objectives to derive a constraint that has to apply to all other programmes. So, for example, we could apply demographic projections to New Zealand Superannuation, thus deriving a long-term track for spending on that item, apply the government’s current debt and revenue objectives and see what would have to happen to all other levels of expenditure.

Notes

Our proposed approach

Both bottom-up and top-down approaches have much to commend them. A particular strength of the top-down approach is that it starts from the proposition that governments will operate in a fiscally prudent manner. There are certainly instances in New Zealand’s past where this has not been the case, but we believe that legislation such as the Public Finance Act and the Reserve Bank Act make it much less likely that this will occur in the future. While there will always be demographic and other pressures on the government to increase spending faster than taxes, there are limits. This is known in economics as Stein's Law: if something cannot go on for ever, it will stop.[8] A top-down approach assumes that Stein’s Law applies.

One particular advantage of a bottom-up approach is that it allows richer details of the individual drivers of spending and revenue to be examined. The disadvantage is that by construction, a bottom-up approach looks at individual spending and taxation items in isolation from everything else the government is doing, whereas top-down considers them as a lump with internal trade-offs. There can thus be an element of unreality in the combined picture of all spending and revenue.

We, therefore, propose to include both bottom-up and top-down fiscal projections in the final Statement and believe that this will meet the spirit of the Public Finance Act.

Types of spending programmes

To make projections of future spending and revenue, one place to start is to model the effects of a known set of policies in a projected world, namely “current policy.”

To derive both bottom-up and top-down projections will require us to make projections of future spending and revenue. One place to start is to model the effects of a known set of policies in a projected world, namely “current policy.” The definition of what is current policy, however, is not always straightforward. When this cannot be determined readily, we make what we estimate to be a plausible assumption about what is driving the expenditure or revenue category.

In respect of major spending types, current policy can be sub-divided into two types of programmes, parametric and non-parametric.

Parametric programmes

Parametric programmes are those where all (or at least most of the material) features of spending are driven by factors that are exogenous to the programme. The largest example is New Zealand Superannuation, the public, universal pension, where all scheme features are set in legislation and can be applied to a projected population to derive a projection of spending.

In terms of the Statement of the long-term fiscal position, parametric programmes can be modelled by applying the current parameters to a projected future world. Parameters can, of course, change, but it is possible to model the future fiscal impact of a set of parameters and construct scenarios around changes in parameters.

Non-parametric programmes

Non-parametric programmes are those where spending is the result of discrete decisions made by governments. Some of these may endure for many years, with spending being rolled over in out-years. Some examples are health, education and transportation. Some non-parametric programmes will remain in place for a number of years and can become at least “semi-parametric.” An example is a formula-driven funding system for tertiary education: providers will receive a fixed amount per student of a certain type. Projections of future numbers of students by age group can thus be used to derive projections of spending on education by levels.

Non-parametric programmes are more difficult to model as the parameters are not clearly specified in the design of the programme. Modelling, after all, is the process of finding suitable parameters that produce the important aspects of a spending programme, for example. The approach followed here involves using a level of past expenditure (or expenditure per capita) as a starting point, and then growing that in line with some indexes (CPI, wages, GDP, or a population group). For example, it might be thought that governments are likely to see defence spending as a proportion of GDP as an important consideration, and thus defence expenditure should be projected forward using the fixed ratio to GDP as the parameter.[9]

Taxes

Current tax policy can be defined as the current set of tax laws, applied to a projected tax base (income such as corporate profits or salary and wages, or consumer spending).

In constructing a set of bottom-up projections, one key issue is so-called “fiscal drag.” Fiscal drag is the term used to describe the situation where the tax on an individual's income grows at a faster rate than the income. This occurs when you have a progressive tax scale where the tax rate rises with income.

Our projections of individual (personal) income tax do not include any fiscal drag. Rather, all tax revenues, except tax on benefits, remain at their end-of-forecast (2010) ratio to nominal GDP. Tax on benefits is allowed to rise with the growth of benefit payments.

Notes

  • [8]Herbert Stein was Chairman of the US Council of Economic Advisers during the Nixon Administration. This particular quote comes from The Public Interest 97, Fall 1989.
  • [9]The modelling of many of our “non-parametric” spending programmes uses the equivalent of nominal wage growth (3.5% a year) as one of the growth factors. A reviewer has suggested that in the overall economy labour costs make up about 60% of the cost of production and that the rest, capital and other inputs, would have a smaller deflator. Hence, these spending categories would be growing at less than the growth of nominal wages. However, labour costs make up 80%, or more, of the costs of government services and so the difference in the deflation between a weighted sum of labour and capital deflators and that of labour alone would be relatively small. We have, therefore, decided to continue using wage growth as the per capita growth index for these spending programmes.

Adapting our existing long-term model

We are proposing to adopt a gradualist approach and initially build on the Treasury’s existing model, the Long-Term Fiscal Model (LTFM), used over the past decade to assess the effects of the proposed budget spending against the fiscal objectives over a period of 10 years or more.[10]

In effect, the LTFM adopts a three-stage approach to projecting the long-term fiscal position.

First, we adopt Statistics New Zealand’s Series 5 projections of the New Zealand population over the next 50 or so years, which take into account possible changes in demographic features (such as life expectancy).

We then take these projections of the population and use them to generate projections of GDP.

Finally, we add projections of government spending and revenue.

The projections in this paper use the Half Year Economic and Fiscal Update 2005 forecasts to June 2010 as a base (Treasury, 2005). By construction, in 2010 the economy is on its long-term growth path. The projections after that point follow demographic and economic trends.

The modelling methodology of the LTFM is a partial-equilibrium approach. By design, there are no explicit feedback loops from the government balance and debt back to the macroeconomy.[11] This is a common approach in long-term fiscal sustainability work, even though we might be accused of pushing the wrong Barro.[12] It has, however, the virtue of simplicity, and for us, familiarity. We have added to the LTFM by taking up some of the modelling techniques used by the Australian Productivity Commission, the OECD and the European Union’s Economic Policy Committee in making our long-term projections.

We are proposing to keep the modelling simple for three main reasons.

First, and most pragmatically, we have decided that it is better to base our work on an existing model (the Long-Term Fiscal Model), rather than to try to build an entirely new model. While the LTFM is a relatively simple model, it does produce detailed projections of the government’s GAAP[13] tables (expenditure and revenue items and the balance sheet) and it has a decade-long track record of producing long-term projections for the New Zealand Government.

Second, but related to this, there are few examples of complex, general equilibrium models on which we could base projections of the New Zealand fiscal position.

There is a risk that a more complex model will shift the debate from the drivers of fiscal policy onto issues of economic modelling, thus severely reducing the benefits of the Statement. 

Third, and perhaps most importantly, it is our judgement that trying to develop a more sophisticated model could be counter-productive. There is a risk that a more complex model will shift the debate from the drivers of fiscal policy onto issues of economic modelling, thus severely reducing the benefits of the Statement. We also think – but would welcome feedback on this point especially – that the LTFM is “fit for purpose”: it allows us to project likely outcomes with sufficient accuracy to better inform the policy debate about the best set of policies to achieve government objectives.

We are also mindful that the Public Finance Act requires the Treasury to produce a Statement on the long-term fiscal position, not on the long-term economic position, although they are related. Thus, we have directed our attention to fiscal matters, rather than trying to develop highly sophisticated models of the economic future. Of course, the issues raised by ageing go beyond their effects on the macroeconomy and the fiscal position to deal with social change, gender differences, ethnic and occupational effects, and the growth or reduction of regional communities. These issues were canvassed in the 1996/97 Task Force on Positive Ageing and are being carried forward by a team led by the Ministry of Social Development, working with many other agencies.

Our approach will undoubtedly evolve as we work on subsequent long-term reports. More broadly, as research into the economics of ageing progresses, we will incorporate more of this into our work.

Modelling uncertainty

Projections for half a century are subject to uncertainty which tends to grow with time. The accuracy of long-term demographic projections has not been great in the past. A study of world population undertaken in 1963 projected that by 2000, 9.2% of the total population in North America would be aged 65 and older. The actual result was 12.5%.[14]

Another example comes from the UK’s 2005 Long-term public finance report. This cites a study of population estimate made in 1891 where the projected combined population of Australia and New Zealand in 1981 would be 94 million, five times greater than the actual outcome.[15]

There are two main ways of handling this uncertainty. One is to display a handful of scenarios showing the results of a range of plausible values for key assumptions. The other is to run thousands of stochastic (or probabilistic) projections drawing on distributions of the input assumptions to produce a distribution of projections: this has the advantage of assigning a probability that some outcome could happen. We are experimenting with this probabilistic approach to uncertainty for the first Statement and will report on early work in section three of this paper.

The remaining sections of this paper discuss the three aspects of the model – demography, output and government activity.

Notes

  • [10]Further details on the LTFM can be found on the Treasury’s website at: http://www.treasury.govt.nz/ltfm/default.asp. We hope that using the same name for a model that is used in two different ways for two different purposes and has some different equations for some types of expenditure will not cause confusion. These differences will be spelled out later in this paper.
  • [11]Some of the macro-economic feedback loops that could be modelled in future are:
  • [12]Robert Barro (1990) pointed out that the government’s fiscal position had a strong influence on the economy and should be included in a general modelling framework.
  • [13]Generally Accepted Accounting Practice is an independent set of rules that governs the recognition and measurement of financial concepts such as assets, liabilities, revenues and expenses adopted by the Zealand Government for its own accounts. It is based on private sector commercial accounting standards.
  • [14]See Figure 3.5, Chapter III, IMF (2004).
  • [15]Chris Shaw, “Accuracy and uncertainty of the national population projections for the United Kingdom,” Population Trends No 77 (1994), page 24.

3. Who will populate New Zealand?

The starting point of our modelling is to look at the issue of the size of the New Zealand population.

From any given starting point, three variables drive population: fertility (how many children are born), mortality (how many people die each year, and importantly, at what age) and migration (how many people leave and arrive in New Zealand).

Demographic change: the big picture

In common with many other OECD nations, New Zealand is experiencing a shift in the structure of the population:  a transition from a high fertility/high mortality state to a low fertility/low mortality state.

In common with many other OECD nations, New Zealand is experiencing a shift in the structure of the population.[16] The developed world (and increasingly the developing world) has completed a transition from a high fertility/high mortality state to a low fertility/low mortality state.

This transition is not a demographic “bulge” that will correct itself at some time in the future. In particular, it is not just the result of the post-World War Two baby boom. Rather, a demographic transition from the high fertility and high mortality rates of a century or more ago to the present and projected low fertility and low mortality is what has been driving this ageing of the population. Figures 3 and 4 show the long-term trends in fertility and mortality.

Figure 3: The long-run demographic transition in New Zealand - birth and death rates have fallen . . .
 
Figure 4: . . . and death rates have fallen across all age groups
Source: For Figures 3 and 4, Treasury Long-term Data Series and Statistics New Zealand. Total crude birth and death rates. In the charts and tables in the demographics section, most of the historical data points are averaged from around census years such as 1991 or 1996. For clarity and consistency with other charts, we have moved the data to the nearest “rounded” year.

The reduction in fertility is, of itself, likely to lead to a lower population (absent migration), while lower mortality has the opposite effect.

The combined effect of lower fertility and lower mortality rates is seen in the resulting median age of the population: this is the age which divides the population exactly in half.

Figure 5: The median age has been rising in New Zealand (apart from the arrival of the baby boom) and is expected to continue rising
Source: Statistics New Zealand’s historical population and Series 5 projection

Statistics New Zealand, in the official National Population Projections, published in December 2004, presents a range of different scenarios for fertility, mortality (or the converse, survival rates) and net migration. They have produced nine separate projections. Of these nine, Statistics New Zealand considers that the “medium” projection (known as Series 5) is the most suitable for assessing future population changes. Because Statistics New Zealand prefers it, we have adopted Series 5 as the basis of future demographic profiles, but we will also illustrate the uncertainty around this series by use of alternative scenarios and probabilistic projections.

Fertility

The total fertility rate is assumed to fall slightly from 2.01 in 2005 to 1.85 live births per woman in 2016 and then remain constant. This is the level favoured by the United Nations in its long-term work for world population. The rate that replaces the population (with zero net migration) is around 2.1 births per woman. Hence we are assuming a sub-replacement birth rate, although one that is at present high compared with most of the OECD.

New Zealand’s experience of fertility rates falling to below replacement levels is not an isolated one.  Currently, 65 countries have fertility rates at, or below, replacement levels.

New Zealand’s experience of fertility rates falling to below replacement levels is not an isolated one. Currently, 65 countries (with a combined population of over 2.8 billion people) have fertility rates at, or below, replacement levels (UN 2005). The UN is predicting that the international trend towards sub-replacement fertility rates will continue into the future.

Alternative fertility paths

Statistics New Zealand has produced projections based on two alternative assumptions of the future course of fertility: low fertility, where fertility falls more sharply to 1.60 by 2016 and high fertility scenario, where fertility actually increases from the base year rate (2004) of 2.01 to 2.10 in 2016, after which it remains constant.

Figure 6: After the baby boom, fertility has stayed flat for 30 years
Source: Total fertility rate from Statistics New Zealand, medium fertility assumption

The effect of the low fertility assumption is to reduce the proportion of young New Zealanders in the population in 2050, and raise the proportion of people 65 and above compared with the base case. Overall population is smaller by 7% in 2050 and the number of people between 15 and 65 will be fewer. Assuming SNZ’s high fertility case will result in the opposite: a larger population (by 7% in 2050), with a greater proportion of youth and smaller relative numbers of the elderly.

Notes

  • [16]For a summary of the demographic transition, see Carter (2004).

Mortality

Life expectancy at birth in a particular year is a way of summarising age-specific mortality rates in that year.[17] Under Statistics New Zealand’s medium assumptions, the median male life expectancy at birth rises from 76.3 years in 2000 to 83.5 years in 2050, while median female expected longevity grows from 81.1 years to 87.0 years.

Figure 7: Statistics NZ assumes gains to life expectancy slow
Source: Statistics New Zealand, life expectancy at birth, history and medium mortality assumptions

This projected rate of gain is slower than we have seen in the past half century (for example, female longevity grew by 9.8 years from 1950 to 2000, but is expected to grow by only 5.9 years in the next half century). For those aged 65 and 85, the assumed life expectancy gains between 2000 and 2050 are generally greater than those between 1950 and 2000. In the extended projections to 2100, Statistics New Zealand holds the life expectancy constant after 2050.

Table 1: Median life expectancy at birth, at 65, at 85, in the stated year
Years 1900 1950 2000 2025 2050 1950-2000 2000-2025 2025-2050 2000-2050
Males Age Gain
Birth 57.4 67.2 76.3 81.4 83.5 9.1 5.1 2.1 7.2
Age 65   12.8 16.7 20.2 21.8 3.9 3.5 1.6 5.1
Age 85   3.9 5.2 7.3 8.3 1.3 2.1 1.0 3.1

 

Table 1: Median life expectancy at birth, at 65, at 85, in the stated year (continued)
Years 1900 1950 2000 2025 2050 1950-2000 2000-2025 2025-2050 2000-2050
Females Age Gain
Birth 60.0 71.3 81.1 85.3 87.0 9.8 4.2 1.7 5.9
Age 65   14.8 20.0 23.2 24.5 5.2 3.2 1.3 4.5
Age 85   4.2 6.5 8.5 9.4 2.3 2.0 0.9 2.9

Source: Life expectancy at birth from Statistics New Zealand, medium mortality assumption.

Alternative mortality paths

As with fertility, Statistics New Zealand has produced projections based on two alternative assumptions of the future course of mortality: high mortality, where life expectancy in 2050 is 81.0 and 85.0 years for men and women respectively; and low mortality, where life expectancy is 86.0 for men and 89.0 for men and women respectively.

We discuss this issue in some detail below, because future trends in life expectancy, and their underlying causes, have a significant impact on our long-term fiscal projections.

The world’s demographic transition started in North West Europe about 1800, with mortality rates generally trending downwards ever since (Lee 2003).

In New Zealand, there has also been a fall in mortality (a rise in survival), and thus a lift in life expectancy (see Figure 7 above). There are two particular features of this increase in life expectancy.

In 1937, 4% of children would be expected to die before their first birthday.  By 2003, this had fallen by almost a factor of 10 to 0.49%. 

First, there has been a substantial reduction in infant mortality. In 1937, for example, 4% of children (born alive) would be expected to die before their first birthday. By 2003, this had fallen by almost a factor of 10 to 0.49%. Similar reductions have occurred at all ages up to 10.

Second, death rates have also reduced substantially during the middle stages of life. Although the reduction is not as dramatic as in the early years of life, it is still substantial and is two to four times lower in 2003 than in 1937.

The combined impact of the lower death rates in early and middle age and a continuing rise in the oldest age to which people live results in what demographers refer to as a “rectangularisation” of the survival chart: far more people survive into old age, and indeed into very old age. This is happening at a faster rate than the increase in the age of the oldest.

While current mortality trends are clear, as yet we do not have full knowledge of what is causing this decline in death rates, what sorts of lives people are leading, especially in later life and whether the trends of the recent past will continue, or reverse. A recent (economics) paper suggests that the decline in mortality rates is ultimately determined by the application of scientific advance and technical progress (some of which is induced by income and facilitated by education).[18]

“Mortality improvements result from the intricate interplay of advances in income, salubrity, nutrition, education, sanitation, and medicine, with the mix varying over age, period, cohort, place and disease.”

There are many different theories and a consensus is yet to emerge, especially about what is to happen in the future. Quoting Riley, Oeppen and Vaupel (2002) note:

Mortality improvements result from the intricate interplay of advances in income, salubrity, nutrition, education, sanitation, and medicine, with the mix varying over age, period, cohort, place and disease.

Lee (2003) cites two separate stages to the decline in mortality. The first stage, starting in around 1800 in Europe, involved reductions in contagious diseases and infectious diseases spread by air or water. Personal hygiene improved (boosted by increases in income), as the germ theory of disease became more widely accepted. Improvements in nutrition were also helpful. Most of the developed world has probably attained most of the potential reductions in mortality due to reductions in infectious diseases and improved nutrition.

Thus, the continuing reductions in mortality seen in the developed world in recent years are the result of reductions in chronic and degenerative diseases, such as heart disease and cancer.

What of the future? Oeppen and Vaupel are at the optimistic end: they predict no decline in the rate of increase in life expectancy for the foreseeable future, with a continuation of a rate of increase of about 2.4 years per decade. This would see life expectancy at birth reach 97.5 years by the middle of the 21st century and a remarkable 109 years by 2100.

More conservative predictions, such as those of Lee and Carter (1992), still see life expectancy reaching 90 years by 2100.

Studies undertaken for the Australian Productivity Commission in its work on the economics of population ageing in Australia produced projections of life expectancy that are in excess of the official estimates of the Australian Bureau of Statistics (ABS).[19]

There are, however, other more pessimistic views about the future prospects for mortality and morbidity. Olshansky et al (2005) are critical of those studies that predict life expectancy on the basis of extrapolating the past. As they put it:

Given that past gains in life expectancy have largely been a product of saving the young, and since future gains must result from extending life among the old, another quantum leap in life expectancy can occur only if the future is different from the past.[20]

They prefer an approach that relies on trends in health and mortality that can be observed in the current adult population. On this basis, they examine the effect of one health condition – obesity – on future trends in life expectancy in the United States. They find that current levels of obesity are likely to increase mortality substantially across the age spectrum, which will, in turn, lead demographers to revise downwards their estimates of life expectancy at birth.

Statistics New Zealand uses its median mortality assumption to drive the official Series 5.  This may be on the conservative side when compared with some assumptions being used by other agencies.

Statistics New Zealand uses its median mortality assumption to drive the official Series 5. This may be on the conservative side (relatively low longevity outcomes) when compared with some assumptions being used by other agencies in their long-term work.

Statistics New Zealand’s low mortality (higher longevity) assumption has a greater proportion of elderly, a relatively smaller labour force, and a larger population by 2050 (up by 2.3%).

Assuming high mortality (lower life expectancy) will reverse these differences from the base case. The labour force would be proportionately larger, while the population in 2050 would be smaller.

Net migration

Finally, Statistics New Zealand assumes that migration settles at a net figure of 10,000 immigrants from 2009 onwards (0.24% of the population in that year). Typically, we have a net inflow of people in their late teens and 30s and 40s, but a net outflow of people aged in their 20s. The horizontal lines in Figure 8 are the averages for the periods covered by the lines and show that this assumption is plausible given recent trends.

Figure 8: Projected long-term net migration follows recent averages
Figure 8: Projected long-term net migration follows recent averages.
Source: Statistics New Zealand. Annual net permanent and long-term migration.

Notes

  • [17]Life expectancy at birth (or at age x, more generally) estimates the number of years a person can expect to live beyond birth (age x), based on the age-specific mortality rates of the population in a given year. Only if these mortality rates remained constant over time would the life expectancy at birth match the actual experience of people born in that year.
  • [18]Cutler, Deaton and Lleras-Muney (2006)
  • [19]Booth and Tickle (2003). For example, they estimate female life expectancy at birth in 2027 in Australia to be 88.1 years, compared to an ABS projection of 85.4 years.
  • [20]Note, however, that in New Zealand the fastest absolute and relative declines in mortality in the last 20-30 years have occurred in the oldest ages.

Resulting demographic projections

These assumptions produce projections of our population growing from 4 million today to 5 million in 2050, a 25% increase in 45 years. The projections also show changing shares in the total population of the young, those of the traditional working age,[21] the old (65 and over) and the “oldest” old (85 and over).

The number of people over 65 is projected to grow almost three-fold, while those 85 and over will grow six-fold by 2050.

The number of people over 65 is projected to grow almost three-fold, while those 85 and over will grow six-fold by 2050. Under this scenario, the working-age population grows until the mid-2020s and then contracts (it shrinks from 66% of the total now to 58% by 2050). Overall population growth slows down until the mid-century and then falls below zero.

Figure 9: The changing shares in the total population
Figure 9: The changing shares in the total population.
Source: Statistics New Zealand, history and Series 5 projection.

The ageing of the population is starkly shown in the following chart of the change in the ratio of people aged 65 and over in the population.

Figure 10: Yearly change in share of people 65 and older
Figure 10: Yearly change in share of people 65 and older.
Source: Statistics New Zealand, history and Series 5 projection. This chart graphs 100 times the difference in the share of 65 and older to the total population between years. If this is above zero, share of those 65 and older in the population is growing. If less than zero, the share of the old is falling.

One way of looking at the changing demographics is to chart the so-called dependency ratios of people 65 and over and those under 15 to working-age people aged 15 to 64 (see Figure 11 below). The aged-dependency ratio climbs from 0.18 in 2005 to 0.45 in 2050. Put another way, in 1900 there were 15 people of working age for every person over 65. Today this has shifted to five people of working age for every person 65 and over, while by mid-century there are projected to be only two.

Figure 11 also shows that the demographic change is not a “bulge” but rather a structural change in the population. Unlike the earlier baby boom, the “ageing boom” (which is partly due to the earlier baby boom) won’t be followed by an ageing bust under these demographic projections and under the probabilistic variants of them. These probabilistic variants are discussed in more detail in Box 1.

Unlike the earlier baby boom, the “ageing boom” (which is partly due to the earlier baby boom) won’t be followed by an ageing bust under these demographic projections. 

Even with Statistics New Zealand’s conservative assumption of fixed mortality rates after the mid-century, the aged-dependency ratio continues to rise until the 2070s when the ratio stabilises around 0.5. Past trends suggest that life expectancy could continue to rise strongly after the 2030s and this would mean that the aged-dependency ratio would be even higher.

Figure 11: Aged-dependency ratio doubles between 2000 and 2050
Figure 11: Aged-dependency ratio doubles between 2000 and 2050.
Source: Statistics New Zealand, history and Series 5 projection

Some people tend to downplay the effect of ageing by pointing to the total dependency ratio which was almost as high in 1960 (pushed up by the youth side) as it is likely to be in 2050 (pushed up by the elderly). The problem with this for the fiscal position is that the young are more likely to be supported privately by families and by relatively small amounts of public spending on schooling, while the elderly in the recent past have tended to use a far greater proportion of public resources in pensions and health care.

Immigration is not a long-term solution to population ageing, although selective immigration is useful for adding to the economy’s skills base.

Changing net migration might be seen as a way of maintaining a low aged-dependency ratio. Let’s assume that we can find large numbers of potential immigrants (this is doubtful, when most countries sooner or later will be facing rising dependency ratios and be competing for immigrants). If we double the number of net immigrants from the assumed 10,000 to 20,000 each year from 2010 to 2050, then the percentage of people aged 65 and over in the population in 2050 would fall from 26.2% to only 24.7%. While the bulk of new immigrants are of working age, they too grow old and eventually make demands on public resources, like the rest of us.

From another point of view, one way of keeping the aged-dependency ratio under 20% (where it was in 2005) all the way out to 2050 would require 300,000 net immigrants each year from 2020 onwards (4.9% of the population in 2020).[22] Immigration, in short, is not a long-term solution to population ageing, although selective immigration is useful for adding to the economy’s skills base.

Box 1: Capturing uncertainty with probabilistic projections

Statistics New Zealand and the Treasury are experimenting with probabilistic (or stochastic) modelling as a way of expressing the uncertainty that surrounds the demographic variables of fertility, life expectancy and migration.[23] In the future, we may extend this approach to some of the economic and fiscal variables used in the LTFM.

Probabilistic modelling usually uses historical information to calculate variability in the demographic data. It uses this variability to construct a probability distribution of outcomes. Probabilistic modelling randomly draws samples from probability distributions when projecting variables forward. This is repeated thousands of times to construct a plot showing the likelihood that certain scenarios will eventuate.

Probabilistic modelling is more informative than scenario analysis. It is less arbitrary and harder to manipulate. Moreover, representing uncertainty as a range of possible outcomes rather than a single number gives a more meaningful picture of the uncertainty arising from demographics. This ability to quantify and represent uncertainty is a major benefit of probabilistic modelling.

Extending probabilistic modelling to fiscal and economic variables provides an additional tool to help judge how much policy adjustment might be necessary to provide a high degree of confidence that fiscal sustainability will be achieved. It would also allow policy makers to identify and gauge the key sources of uncertainty that matter at different points in the future for particular fiscal variables. Finally, it would enable policy makers to evaluate how different policies perform in the context of uncertainty.[24]

In Figure 12 below, the black line is the median projection of the aged-dependency ratio; the dark shaded area indicates the 25% to 75% probability interval; and the total shaded area the 5% to 95% probability interval. Notice that uncertainty about the aged-dependency ratio increases significantly only after 20-25 years. This is because for the next two decades uncertainty around mortality is mainly associated with people whose births have already happened. After that, uncertainty around the aged-dependency ratio increases significantly as there is uncertainty about the births as well as the deaths of people.

Figure 12: The aged-dependency ratio will almost certainly double
Figure 12: The aged-dependency ratio will almost certainly double.
Source: Statistics New Zealand, indicative Series 5 probabilistic projections
It is also important to note that these projections show how likely the rise in the aged-dependency ratio will be. From the figure, we can say with reasonable confidence that the aged-dependency ratio will increase and we are 95% certain that it will increase from around 0.18 now to more than 0.40 in 2050 (that is, at least double in size).

Notes

  • [21]We use “working age” as a convenient label for describing people aged between 15 and 64. There are people outside these limits who work and people inside who don’t. Also, the (demographic) aged-dependency ratio classifies people solely by their age, not by their degree of independence (say, by degree of labour force attachment): there are those older than 65 and under 65 who are dependent and independent. In the future, we are likely to see a greater proportion of older people continuing to work after 65 and hence be “independent,” if present trends continue. The official definition of working age (as used in the Household Labour Force Survey) is the civilian, non-institutionalised population 15 and older (and so has no upper limit) and this definition is used to produce projections of the labour force.
  • [22]The average net migration over the past decade has been 12,000 per year.
  • [23]Dunstan and Speirs (2005)
  • [24]See Bryant (2003), Lee (2004), Heller (2002), and Lee, Anderson and Tuljapurkar (2003)

4.  What will people be doing?

The Long-Term Fiscal Model (LTFM) builds projections of real GDP from the end of the latest macroeconomic forecast by using working-age population, labour participation rates, and assumptions about long-run unemployment and labour productivity.

Under a growth-accounting framework, real GDP is based on the three Ps of Population, Participation and Productivity. Thus, real GDP is equal to:

Population: The total number of people available for work working-age population (15 and older)
  multiplied by
Participation:  The number actually working and how much they work participation rates  x  (1 – unemployment rate)  x  average hours worked
  multiplied by
Productivity: How much each person produces each hour that she or he works GDP per hour worked

In the LTFM, we use the growth form rather than the levels form of this identity to project GDP.

Participation

Once the working-age population (in the larger sense of all people 15 and above) has been calculated, the next step in constructing the projection of GDP is to calculate participation rates.

Labour-force participation has been changing in New Zealand since the Second World War. This means that we have had to take a view on whether there will be further changes in this pattern over time.

The pattern of labour-force participation has been changing in New Zealand since the Second World War.[25] This means that we have had to take a view on whether there will be further changes in this pattern over time.

Women’s participation rates have been rising since the Second World War, and women aged 25-54 have had a greater level of participation than their predecessors.[26]

While New Zealand does not have an obligatory retirement age, labour market participation at present drops from 60% or so for people aged 60-64 to 13% or so for those over 65 (both of these rates have risen over the past 20 years, particularly for females, because of the lift in the age of eligibility for New Zealand Superannuation in the 1990s).

Over the last 20 years, participation rates of young men and women have been falling, reflecting the greater enrolment in tertiary education. For the prime working ages (25-54), male participation has fallen as men were displaced by structural change and lacked the skills required by the changing market place. Through this period, prime-aged female participation rose as women moved into new areas of work, adapted to change, worked longer before having children, or decided to remain childless.

These patterns are projected into the future. The result is similar to the high participation projection of the labour force by Statistics New Zealand. Eventually all age-group participation rates stabilise and from that point labour force growth is driven completely by the underlying demographics.

Even though participation of the open-ended 65-plus cohort is expected to rise, aggregate participation falls, as a greater proportion of people spills into the older age groups where participation is lower.

The aggregate participation rate falls from 66% now to 59% by mid-century as the ageing effect outweighs behavioural changes. The labour force grows to mid-century and then begins to decline. Even though participation of the open-ended 65-plus cohort is expected to rise, aggregate participation falls, as a greater proportion of people spills into the older age groups where participation is lower than the average.

Figure 13: Rising female and largely falling male labour participation
Figure 13: Rising female and largely falling male labour participation.
Source: The Treasury

Alternative participation scenarios

New Zealand’s labour force participation rates are high relative to the OECD, and similar OECD countries. However, there is scope for increasing participation, particularly among young women. A Treasury study[27] has calculated the effect on GDP of hypothetical increases in employment from increased participation, taking into account the differences in productivity between new and existing workers. The results suggest that increasing the labour force participation of women aged 25-34 to the average, adjusted for paid maternity leave, of the top five OECD nations increases employment by 28,800, making GDP 1% higher than it actually was in the baseline year 2001. Raising participation overall to the average of the top five OECD countries increases employment by 142,600 and generates an increase of 5.1% in the level of GDP.

Employment and unemployment

The Half Year Economic and Fiscal Update 2005 economic forecast assumes that by 2010 the trend unemployment rate is 4.5% of the labour force and this is assumed to remain constant throughout the projection period. Along with this, average hours worked per employee are also assumed not to change after 2010. This ratio has remained relatively constant for the past decade. Employment grows from 2.05 million in 2005 to 2.45 million in 2050.

Figure 14: Employment projection sees steady but decreasing growth
Figure 14: Employment projection sees steady but decreasing growth.
Source: Statistics New Zealand and the Treasury

Notes

  • [25]See Hurnard (2005), Callister (2005) and Johnston (2005).
  • [26]These observations are based on participation rates derived from the Census. See Hurnard (2005). We created probabilities of entry and exit for each 5-year cohort (except the youngest) and then used these to project participation rates into the future. The lifting of the age of eligibility for New Zealand Superannuation through the 1990s has made it difficult to get “policy-free” exit and entry probabilities for the older age groups, given the shortness of the HLFS dataset. For these age groups, we have been guided by the Bell and Bryant (2004) results and the dynamic cohort results in the Australian Productivity Commission report (2005).
  • [27]Bryant, Jacobsen, Bell and Garrett (2004).

Productivity

Empirical estimates suggest that productivity rises with age up to middle age, before declining (Australian Productivity Commission, 2005). This could mean that ageing could produce a (small) decline in average productivity with the effects of a greater proportion of older workers being largely offset by relatively fewer younger ones. The present modelling, however, assumes that average labour productivity (real output per hour worked) grows by 1.5% annually for everyone over the projection period. This reflects the median growth in the output per hour worked between 1980 and 2003.

In section five, we will examine the effects of changing this productivity assumption on the fiscal position as measured by ratios to GDP. Labour productivity growth (which in the LTFM is assumed to equal the real wage growth) of 1.5% a year means that by mid-century real incomes will have doubled.

Labour productivity growth (which in the LTFM is assumed to equal the real wage growth) of 1.5% a year means that by mid-century real incomes will have doubled. 

Work done at the Australian Treasury after the release of the first Intergenerational Report in 2002 has considered the effect of having a lower productivity growth in the government-funded service sectors relative to the rest of the economy.[28] We may investigate such an approach for the final Statement.

Inflation and bond rates

The final major assumptions in the modelling are that annual inflation over the projection period is assumed at 2%, the middle of the present Reserve Bank target range, and that the real government 10-year bond rate is 4%.

Table 2: Summary of key economic assumptions, fixed from 2010 on
Variable Value Source of the assumption
Labour productivity growth 1.5% Historical average over 1980-2004
Inflation 2.0% Mid-point of Reserve Bank’s target band
10-year real gov’t bond rate 4.0% Historical average
LT unemployment rate 4.5% Based on an assessment of NZ’s LT rate
Average hours per week 38.4
hours
Based on an assessment of recent trends

Resulting GDP projections

In growth terms, nominal GDP in any one year (Yt) grows as follows from 2011 onwards:

where Yt-1 = GDP in the previous year, g = growth of labour force, p = labour productivity growth and i = the inflation rate.

In other words, growth of nominal GDP is roughly the sum of the labour force growth, labour productivity growth, and the inflation rate. This formula implicitly assumes that the employment rate and average weekly hours worked are constant after 2010 and so they drop out of the growth equation.

Combined with these assumptions, the demographic projections translate into weakening labour force growth and real GDP growth lowering from a 3.2% annual average over the past decade to a 1.6% average through the 2040s. Real GDP growth per capita is closer to labour productivity growth, but it falls below when population growth is larger than labour force growth from 2020 onwards.

Figure 15: Ageing reduces GDP growth, but not per capita GDP growth from 2030 onwards
Figure 15: Ageing reduces GDP growth, but not per capita GDP growth from 2030 onwards.
Source: Statistics New Zealand and Treasury projections

Notes

  • [28]Gruen and Garbutt (2004).

5.   What will governments be doing?

This section examines the issues we face in modelling government spending and revenue.

Over the next 50 years, as nominal GDP and government spending grow, it makes sense to report spending and revenue as ratios of (nominal) GDP. If a spending category rises as a share of GDP, this means it is growing faster than GDP.

In this section, we use four functional spending categories and total tax revenue to illustrate the issues behind modelling for the Long-term Fiscal Statement.

We start our examples of spending models with New Zealand Superannuation (NZS) which is governed by set rules and parameters. The second example has less explicit parameters, such as education (parameters are spending per student at various levels). A third example is where we simply use growth of an aggregate variable such as labour force or nominal GDP to generate spending projections. This is the approach we use for core government spending and largely for tax revenue.

Particularly challenging is the fourth example: health spending, where we model spending from past patterns and projected demographics (in other words, it has no explicit policy parameters).

We tend to use the same real growth driver for both spending and GDP (and hence revenue).  For some spending, however, this modelling assumption might signal a departure from recent growth trends and needs to be tested.

Most of the spending categories are modelled with demographic drivers, an indicator of real growth (such as labour productivity or equivalently real wages), and inflation. We tend to use the same real growth driver for both spending and GDP (and hence revenue). This allows spending ratios to GDP to throw any demographic shifts into relief. For some spending, however, this modelling assumption might signal a departure from recent growth trends (it could be higher or lower) and needs to be tested.

At the level of the functional spending categories, the choice of indexation can make a large difference over half a century. Over such a long period of time, we generally assume that most spending categories would grow by more than just inflation. In the model, following the prescribed rules, NZS payments grow by the average net weekly wage. The current policy is that social welfare benefits are indexed to CPI inflation and that will be the way it is treated in the modelling. But it could be argued on equity grounds that all beneficiaries should share in the labour productivity gains and their benefits would grow in line with the average net weekly wage, rather than just inflation.

Productivity growth is important for improving the living standards of all New Zealanders, but because most spending is indexed to labour productivity growth (or the real wage), changing the labour productivity growth assumption doesn’t make much difference to ratios of spending to GDP or projected operating balance to GDP. This is illustrated for all social welfare transfers at the end of this section.

Parametric programmes - NZ Superannuation

The parameters defining New Zealand Superannuation (NZS), the country’s tax-funded, universal pension, are indexation and the age of eligibility. The indexation rule is that the married benefit is indexed to CPI inflation (adjusted each April), provided that the net rate of NZS for a married couple is no less than 65% or more than 72.5% of the net average wage.[29] Otherwise it grows with the average wage. Single people receive a fixed proportion of the married rate.

For over 100 years, New Zealand has provided a public pension of one sort or another to its older citizens. NZS began in 1977. The most recent change in its parameters occurred when the age of eligibility for men and women was raised progressively from 60 in 1992 to 65 in 2001.

In the Long-Term Fiscal Model, payments of NZS are projected from 2011 onwards by the growth of the wage-indexed individual payment and by the growth in the numbers of people 65 and over. This is the same treatment used in the Fiscal Strategy Report projections.

This modelling takes no account of payments to spouses under the age of eligibility, or for people who have yet to satisfy a time-of-residency rule. While these details are important to the programme, these have little effect on the growth of superannuation payments in the long run because implicitly we are assuming that the proportions of these payment categories do not change significantly.

NZS is modelled in the Long-Term Fiscal Model as follows:

where B = the married benefit and n = nominal wage growth (3.53% per annum).

If Et is spending on NZS in year t, then

where b = the growth of B (i.e. nominal wages) and

= the growth of population aged 65 and over.

In line with doubling the numbers of people aged 65 and older in the population between now and 2050, it is not surprising that spending on NZS relative to GDP grows by 2¼ times.

In line with the doubling of the numbers of people aged 65 and older in the population between now and 2050, it is not surprising that spending on NZS relative to GDP grows by 2¼ times as the growth

of eligible population between 2005 and 2050 is divided by the growth
of the labour force (the demographic driver of GDP) with both nominal wage growth b and nominal labour productivity growth (equals b) terms cancelling out of the numerator and denominator. Roughly,

.

In the figure below, the solid line is payments of NZS indexed by nominal wage growth (current policy, given the assumptions that nominal wage growth is 3.53%, while inflation is 2%). Other countries, such as the United Kingdom, index their basic public pensions to prices. The lighter line shows the effect of assuming CPI indexation on NZS and illustrates the large effects a different indexation parameter has on the projected cost of superannuation (similar differences occur with many other transfer programmes). By 2050, the gap has grown to 4.1 percentage points of GDP for NZS indexation.

Figure 16: New Zealand Superannuation and its predecessors – effects of different indexation
Figure 16: New Zealand Superannuation and its predecessors – effects of different indexation.
Source: The Treasury

As a counter-factual comparison, the following figure illustrates the projected evolution of NZS spending if the age of eligibility had not been raised from 60 to 65 years during the 1990s. The difference is that payments of NZS would be 1.5 percentage points of GDP higher in 2005 and have grown to about 2 percentage points higher by 2050 if 60 years had been maintained as the age of eligibility for NZS.

Figure 17: New Zealand Superannuation and its predecessors - age of eligibility
Figure 17: New Zealand Superannuation and its predecessors - age of eligibility.
Source: The Treasury

The objective of the NZ Superannuation Fund is to build up assets for partially pre-funding future NZS expenses in the face of the expected rise in NZS costs. Under the projections presented here, the start of the drawdown of the Fund is in 2028.

The size of contributions to the Fund is calculated over a 40-year rolling horizon to ensure that superannuation obligations over the next 40 years can be met.

Establishing the Fund has not involved any significant changes to the parameters of NZS. The Fund shifts contributions through time and alters the track of net debt (representing a form of tax smoothing), but does not change the amount of benefits expected to be paid.

Notes

  • [29]The confidence and supply agreement between the Labour Party and the New Zealand First Party specifies that for the term of the current Parliament, the floor will be 66% of the (net) average weekly wage.

Non-parametric programmes - Education

Over the past decade, total education spending has grown by an average of 6.2% a year. 

An area of recent, rapid growth is spending on education. Over the past decade, total education spending has grown by an average of 6.2% a year. Nominal growth over the past half century has averaged 10.9% a year (4.5% real growth a year), about 1.7 percentage points faster than nominal GDP growth over this period. At present, spending on primary and secondary schooling takes about half of the education sector spending, while tertiary has about a third.

Here we assume that education spending depends in general on the wages per teacher, the student-teacher ratios, the enrolment rates and the population of potential students. The parameter is spending per full-time equivalent student which is the average wage per teacher increased by the standard real growth factor - 1.5% a year as we are assuming that the other ratios remain constant through the projection period.

We use this simple modelling approach to project forward all levels of education, apart from tertiary, using the growth of the population base, inflation and a real per-student growth factor of 1.5% (based on teachers’ wages) each year:

where E = expenditure, d = growth of the appropriate demographic group, i = the inflation rate and s = real growth per student.

The demographic groups are: ages 1-4 for early childhood education, 5-17 for primary and secondary, and 18-30 for tertiary. These age groups are projected to reduce in size over the next half century. The last two terms in the equation allow for the growth of nominal expenditure per student.

Tertiary has an extra growth driver. In this case, E is tertiary spending (plus student loan write-offs) and the extra growth driver is the growth of (1 - participation18-30). This is intended to capture the offsetting effects of changes in labour market participation on enrolment rates and the growth of the main tertiary cohort of 18-30 year olds.

The number of equivalent full-time tertiary students rose rapidly from 169,000 in 1999 to 245,000 in 2003 (the ratio of EFTs to the 18-30 cohort rose from 24% to 35% over this period).[30] Our base case projection implicitly assumes that the tertiary enrolment rate remains at the current rate through the projection period, apart from the small movement due to changes in the labour market participation.

The projection also assumes that we capture savings from the falling numbers of students at all levels as the population ages (see Figure 18). There are, however, several reasons why this may not happen. As a society, we may demand higher teacher-to-student ratios or we might see a greater trend towards publicly-funded “life-long learning.” However, the cost of extra training on the job (of older workers, for example) may possibly be paid for largely by the private sector.

Figure 18: Demographics reduce the projected GDP share of education spending
Figure 18: Demographics reduce the projected GDP share of education spending.
Source: The Treasury

Non-parametric programmes - Core government services

Core government services expenditure includes the costs of running departments such as Inland Revenue, State Services Commission, Ministry of Foreign Affairs, Treasury, Statistics, and Overseas Development Assistance, etc. This category of spending has no clear relation to ageing of the population.

Figure 19: Core government services projected to grow with GDP
Figure 19: Core government services projected to grow with GDP.
Source: The Treasury

This is modelled as follows:

where Et = spending,

= growth in employment, i = the inflation rate and g = wage growth, which is fixed at 1.5% per annum. This is the growth form of an equation where spending depends on the number of public servants times the average nominal wage and we assume that the number of public servants is a fixed proportion of total employment. This growth tracks that of the labour force and hence the ratio of spending to GDP is fixed.

Notes

  • [30]The 2005 Speech from the Throne signalled a move away from increasing participation in some parts of the tertiary sector to a greater focus on quality and relevance in those parts of the sector.

Non-parametric programmes – Health

Total health care is a major component of government spending in New Zealand and has been rising both in real terms and as a proportion of GDP.

Figure 20: Government health expenditure, in constant 2000 dollars, and as a percentage of GDP
Figure 20: Government health expenditure, in constant 2000 dollars, and as a percentage of GDP.
Source: The Treasury

In contrast to the case of NZS, there is not a single parameter-driven scheme in place for health spending.  Rather, there is a complex set of policies, which is usually described as “the public health system.” 

Modelling the future course of health spending is a particular challenge. In contrast to the case of NZS, there is not a single parameter-driven scheme in place. Rather, there is a complex set of policies, which is usually described as “the public health system.”

The Government currently funds large amounts of health care provided by private suppliers. For example, doctors visits are subsidised via the Primary Health Organisation system; pharmaceuticals are subsidised via the Pharmaceuticals Benefits Schedule operated by Pharmac; treatment of personal injury from accidents is reimbursed - sometimes in part, sometimes in whole - by the ACC scheme. It also supplies heath care services in kind (principally though the hospitals operated by District Health Boards).

Overall health spending is thus the result of a myriad of individual purchase decisions made by successive governments about what to fund and what to provide.

For the purposes of the Statement, and consistent with the “current policy” approach, we are assuming that the broad features of the existing health system remain in place and that future governments will continue to make purchase decisions much as they have in the past. Rates of growth in expenditure are extrapolated from the past.

Using this approach, the result is that government decisions are driven by a combination of demography, cost and policy decisions. The model does not, however, separately define the effects of each of these three elements. Rather, it projects current expenditure forward to 2050, applying:

  • cost weights to a demographic projection (we allow these to change through time - see below)
  • an income (GDP) effect, plus
  • a residual growth factor.

The result is a health-cost projection that incorporates demographic and non-demographic growth factors.

Given the fiscal and social importance of health spending, we have set out below an extended discussion of some of the underlying drivers of that spending.

Demography

Demographic drivers are changes in the composition of the population that result in changes in the pattern of medical treatment and, thus, government spending.

The amount of public spending on a 70-79 year old is four times the spending on a 20-29 year old.   However, just because the population is ageing, it does not necessarily follow that health expenditure will increase.

This can be thought of as a system where governments decide what treatments will be available to people in various categories (neonates, the young, the elderly, etc) and sufficient funding is made available to meet the required expenditure. For example, a government might decide that everyone over age 65 presenting to a DHB with pulmonary embolism (blood clots in the lung) will receive a course of Warfarin (an anticoagulant)free of charge. If the number of people over 65 increases by 15% and the frequency of pulmonary embolism remains constant, then funding for Warfarin treatments will increase by 15%.

Figure 21 below shows that per capita spending on personal health rises steeply with age. Personal health includes primary, secondary and tertiary medical care – about 70% of total public health spending (the rest consists of spending on disabilities, health education, mental health). The chart shows that the amount of public spending on a 70-79 year old is four times the spending on a 20-29 year old. However, just because the population is ageing, it does not necessarily follow that health expenditure will increase.

Figure 21: Per capita personal health costs by age/gender in 2003/04
Figure 21: Per capita personal health costs by age/gender in 2003/04.
Source: Ministry of Health

The upward slope of these curves reflects the fact that the older you get, the greater are your chances of ill health and dying. The reduction at the older ages is due to the cost of dying which tends to fall in very old age (April, 2004). The actual impact of an increase in life expectancy on health spending depends of what happens to people’s health status as they get older.

The fact of an increase in life expectancy means that there has been a change in the health status of the population. There is an unsettled debate in the medical literature on what is happening, and is likely to happen in the future, to health status. The first bar in Figure 22 represents a life before the increase in life expectancy. There are three possibilities for changes in health status, which are illustrated in a stylised form in the lower three bars. In each case, they take as given an increase in life expectancy: people are, on average, living longer. The question they seek to answer is whether those extra years of life are, to put it crudely, lived in “good” or “bad” health.[31]

The most optimistic scenario is that health is improving across the board.  This is known as a “compression of morbidity.” 

The first, and most optimistic, scenario is that health is improving across the board. This is known as a “compression of morbidity”:people both live longer and have fewer years of bad health.

The second is a “shift to right” or “dynamic equilibrium”: the absolute period of bad health stays the same, but falls in relative terms as the absolute period of good health increases.

The final and most pessimistic scenario is known as an “expansion of morbidity”: the absolute period of good health stays the same, with all the increased years of life expectancy being in poor health. A severe expansion of morbidity would see the absolute period of good health reducing.

Notes

  • [31]In Figure 22, we have neatly divided a person’s life into discrete periods of good and bad health. For many people, this is clearly not the case.

 

 

Figure 22: Possible future health states in years of life
Figure 22: Possible future health states in years of life.

A joint study by the Treasury and the Ministry of Health (Bryant and Teasdale et al., 2004), investigates the effect of improvements in mortality and morbidity on health spending. Bryant and Teasdale et al. assume that across both genders and all ages, mortality declines by 1.5% per year and that prevalence of disability within each age group declines by 0.5% per year. This produces a set of results that involve a compression of morbidity.

At first sight, it might be thought that these combined assumptions would see expenditure fall through time: fewer people are getting sick in any year and they are living longer. But we are still mortal. Everyone dies eventually.

It is a stark fact that our most extensive and expensive experience of the health system often occurs as it tries, and eventually fails, to cure us of our last illness or injury. It may be the case that the costs of treating this last event seem to decrease with age.

Thus, it is likely that demographic changes will see average health care costs increase. But the effect is not likely to be great. Extrapolations from Bryant and Teasdale et al. (see Figure 23) suggest that up to about age 55, there is a 3%-5% reduction in the average annual health care costs, while after that age, there is an increase of about 5%, with a marked increased (over 15%) for those 95 years and older.

Figure 23: Compression of morbidity has a small impact on average health costs (males) across most age groups
Figure 23: Compression of morbidity has a small impact on average health costs (males) across most age groups.
Source: Derived from Bryant and Teasdale et al, 2004

Cost increases

Advances in medical science are allowing more conditions to be treated, but in increasingly costly ways.  This is not a universal trend. Overall, however, health is becoming more expensive.

Less benign are the assumptions about costs. Internationally, the average cost of health treatments is increasing. Advances in medical science are allowing more conditions to be treated, but in increasingly costly ways. For example, advances in immuno-suppressant drugs have allowed more people to undergo organ transplants. New drug treatments for formerly fatal conditions sometimes come at a very high cost.

This rise in average costs is not a universal trend: for example, advances in techniques such as keyhole surgery are allowing people to be treated much more cheaply (and, often, more effectively). Overall, however, health treatments are becoming more expensive.

International experience is also that health expenditure increases with income, both at the individual level (in the case of private provision) and nationally (for public health systems). Our modelling assumes an income elasticity of demand for public health services around unity. Unit elasticity means that average growth of per capita demand for health services is the same as the growth of nominal GDP per capita (a measure of aggregate income).

Details of the modelling approach to health spending

We introduce a different approach to modelling health spending from the one traditionally used in the LTFM and will devote some space to describing this.

The assumption that the per capita health costs by age (such as those depicted in Figure 21) will remain the same over the next half century is commonly made, although it has been challenged.[32]

As population ageing accelerates over the next two decades and the age distribution moves to the right along the cost profiles (Figure 21), we would expect rising health spending (if these profiles don’t also change). Other increases in health spending come from a (non-demographic) shift upwards in the profiles so that cost rises are independent of the age of the recipient.[33]

As population ageing accelerates over the next two decades, we would expect rising health spending.   Other increases in health spending come from a (non-demographic) shift upwards in the profiles.

In analysing the drivers of total real per capita health spending over the period 1950 to 2005, we make two simplifying assumptions: that the implicit proportions of spending by age and gender in the cost profiles (such as those in Figure 21) have remained the same over the past three or four decades; and that aggregate health spending covers roughly the same bundle of services throughout this period (as we don’t have data indicating how this may have changed through the decades).

Running historical demographic changes through the fixed-cost profiles suggests average annual growth purely from ageing (the change in age structure) has been around 0.2-0.4 percentage points a year through the period, a relatively minor contributor to the annual growth of total real per capita health costs since the 1950s (3.0%). This growth analysis is highly dependent on the deflator (CPI is used here) and the period covered.

Estimates in the literature of the income elasticity of demand for public health services range from 0.9 to 1.2 and centre on unity. Using New Zealand data for the period 1950 to 2005 produced an estimate of 1.16.

Notes

  • [32]Richardson (2004): “. . . drawing time-series conclusions from cross sectional data is problematical,” p1. This is linked with assumptions about whether our expected longer lives are lived on average in relatively longer periods of good health (compression of morbidity), or the reverse. Little is known in New Zealand about healthy life expectancy (see Graham, Blakeley, Davis et al, 2005, for one study), but the modelling needs to take a position on this. Our work suggests that the effects of ageing on health expenditure are likely to be relatively small.
  • [33]This idea of splitting cost growth due to shifts along a curve and shifts upward of the curve dates from some researchers in the 1970s and has been recently taken up again by the Australian Productivity Commission (2005) and the OECD (Bjornerud and Martins, 2005, Martins and de la Maisonneuve, 2005). In fact, it may be impossible to decompose the effects of shifting along and shifts up of these curves as there is likely to be a complex interaction.

 

 

Table 4: Decomposition of historical real per capita health spending
Average real annual per capita growth to 2005 from 1950 1960 1970 1980 1985 1990 1995 2000
Total health 3.0 2.9 3.0 2.4 3.5 3.8 4.2 3.8
Pure age effect 0.2 0.3 0.4 0.4 0.4 0.4 0.3 0.4
Age-adjusted growth 2.8 2.7 2.7 2.0 3.1 3.4 3.8 3.4
Income elasticity = 1.00                
Income effect 1.6 1.6 1.5 1.8 1.9 2.7 2.4 2.7
Residual 1.2 1.1 1.2 0.2 1.2 0.7 1.5 0.7
Income elasticity = 1.16                
Income effect 1.8 1.8 1.7 2.1 2.2 3.1 2.7 3.1
Residual 1.0 0.9 0.9 -0.1 0.9 0.3 1.1 0.3

Source: Treasury

Based on history, the average annual residual growth factor over the period since 1950 is 1.2% (for unit income elasticity) or 1.0% (when income elasticity is equal to 1.16). Note that in the early 1980s, real health spending growth slowed, forcing the residual to be negative. For the projection period, the paper assumes an income elasticity of 1.0, but also runs a scenario for 1.16. Note that a higher estimate for the income elasticity requires a smaller estimate for the residual. The residual captures the effect of technology, preferences, and so on.

Over the projection period to 2050, we derive the growth of personal health spending from three effects:

  • the effects of rising population and the changing mix of ages are derived by applying the cost weights to the demographic projection[34]
  • the growth due to increases in income, and
  • the residual growth factor.

In other words,

where E is health spending, cw = growth of ∑acost weightsa x pop groupsa (summing over ages and genders, a), ε = income elasticity of demand for health services, g = nominal GDP growth (as a proxy for income), and r = the residual growth factor, tapering down to 0 in 2050.

By 2050 the life expectancy of a 65-year-old is five years more than it is now. Effectively we assume that a 65-year-old in 2050 has the same health status - and will require the same public spending - as a 60-year-old does now. The cost curves are therefore moved out to capture this compression of morbidity.

This equation differs from the standard LTFM equation in that it has the residual growth term and it tracks nominal GDP growth (as we are assuming that ε = 1 in the base case). Aside from the labour force effects in GDP growth, the major differences are the residual growth factor and the shifts to the right in the cost profiles.

For long-term care (proxied here by disability support, which includes home support, residential care and equipment), following Bryant, Teasdale et al., we assume that the incidence of disability decreases over time (paralleling the reduction in personal health costs with rising longevity). The provision of informal care is negatively related to labour participation of 50-64 year olds to capture choices between the provision of care to a relative at home and a job in the labour market.

where E = public spending on disability support, cw is the growth of cost weights times the age and gender groups, g = growth of nominal GDP, r = growth of the residual, which tapers down to 0 in 2050, h = growth in the participation of the 50-64 age group and p = is the rate of change in the prevalence of disability (falling by 0.5% a year).

The pure ageing effect (where we look at the effect of the changing mix in the age structure of the population – the first bracketed term in the above equations) contributes a rising amount to the growth of health expenditure until it reaches 1.5% in the mid-2020s and then it slows to 0.4% by 2050.

We illustrate in Figure 24 the effect of changing the income elasticity from the base level of 1 (with the accompanying residual 1.2 tapering to 0 in 2050 – the solid black line) to 1.16 (where the residual growth falls from 1.0 to 0 in 2050 – dotted line).

The figure shows that the effects of changes in the elasticity and residual are not offsetting: the elasticity change has more effect than the residual change. The chart also shows the sensitivity of the assumption about reductions in the prevalence of disability in the population. The base case assumes disability rates fall by 0.5% a year. The dashed line reflects the assumption that disability rates do not change.

Figure 24: The GDP share of Core Crown Health spending continues to grow
Figure 24: The GDP share of Core Crown Health spending continues to grow.
Source: The Treasury

Notes

  • [34]Changes in the age-specific patterns of disease or treatment could mean that future age-specific expenditures no longer follow the pattern of Figure 21. We have attempted to model these, but believe they are a relatively minor determinant of overall expenditures (Bryant, Teasdale, et al., 2004).

Taxation

Individual income tax is likely to vary by age and by gender of the taxpayer, but for simplicity we assume here that tax revenue largely remains a constant proportion of nominal GDP throughout the projection period.

We have chosen in these long-term projections not to model fiscal drag on individual income tax (fiscal drag is the rise in the average tax rate as income spills into higher brackets) as the government has announced it intends to index personal tax rate thresholds every three years from April 2008 onwards. In addition, fiscal drag is not modelled in this type of long-term work by other finance ministries, and the LTFM does not have the level of detail to model it properly.

The revenue-to-GDP ratio does move up slightly through the projection period, but that is because of taxation on the growing payout for NZ Superannuation.

Source deductions tax (or PAYE, tax on individual income) has the following form for tax on benefits:

where Tt= tax and β = growth of benefit payments. In aggregate, this rises through the projection period, driven by payments of NZS.

Benefits here include NZS, unemployment benefit, domestic purposes benefit, invalids benefit and sickness benefit.

Source deductions tax on all other income sources is

where g = growth of nominal GDP. Then, total source deductions tax is the sum of these two:

All other tax types (such as corporate tax and GST) are modelled as follows:

where g = growth of nominal GDP. In other words, the tax-to-GDP ratio for all taxes other than source deductions on benefits remains constant from 2010 onwards.

Although the base case does not include fiscal drag, we show the estimated effects of fiscal drag in the figure below, where a tax elasticity of 1.13 produces the lighter line,[35] which grows to a 2.4 percentage point difference with the baseline tax-to-GDP ratio.

Figure 25: Tax revenue as a share of GDP rises only slightly through the projection period
Figure 25: Tax revenue as a share of GDP rises only slightly through the projection period.
Source: Treasury projections

Effect of changing productivity growth

In the above modelling, we have assumed a constant real labour productivity (or real wage) growth of 1.5% from 2011 onwards. For benefit payments, this additional growth above inflation could be assumed on the grounds that the same gains enjoyed by workers should also apply to benefits (but as mentioned at the start of section 5, we are assuming the current policy of CPI indexation). Otherwise, a growing gap would open up between workers and beneficiaries. For spending without a strong demographic linkage (such as defence or core government services), we have assumed that annual spending would grow overall by 1.5% in real terms (this is the wage growth of people providing those services).

If most spending programmes are indexed to labour productivity, then changing the productivity growth assumption will have only a small effect on the share of total spending to GDP.

This assumption, therefore, pervades most of the modelling. It is used to construct projections of GDP and shows up in most, but not all, of the spending categories. It is not surprising, then, that having a larger productivity growth of, say, 2% per annum makes only a little difference to the ratios to GDP. With a higher growth in productivity, as a society we would be wealthier (nominal GDP would be larger by 22% in 2050), and so have a larger tax base, but on the other hand we assume we would be paying larger benefits and more in wages, and the costs of other non-demographic programmes would also rise. This alternative assumption is illustrated below for the core Crown social security and welfare benefits where the rise in GDP largely matches the rise in expenditure.

So while labour productivity growth is important for many reasons (such as lifting the living standards of New Zealanders), if most spending programmes were indexed to labour productivity (or equivalently to real wages), then changing the productivity growth assumption will have only a small effect on the share of total spending to GDP.

Figure 26: Higher labour productivity growth has little effect on the ratio of social security and welfare payments to GDP
Figure 26: Higher labour productivity growth has little effect on the ratio of social security and welfare payments to GDP.
Source: The Treasury
Table 5: Summary of major modelling approaches
Model type Example Drivers
Parametric NZ Superannuation
  • Demographics: 65 and over
  • Wage growth
Non-parametric Education (by level)
  • Demographics: appropriate age groups
  • Cost per student (driven by teaching staff wage growth)
Non-parametric Core government services
  • The equivalent of GDP
Non-parametric Health (by service group, such as public health and disability support)
  • Demographics
  • Age-dependent cost curves shifting to reflect contraction of morbidity and falling disability prevalence
  • Demand (GDP via income elasticity)
  • Residual growth (historical to capture technology, input cost growth)

Notes

  • [35]In other words, a 1% growth in personal income produces a 1.13% increase in tax on that income. This comes from an estimate of the tax elasticity on personal income using a microsimulation model.

6.   Conclusions

This paper has been written to explain our modelling strategy for the Statement on the long-term fiscal position. The first Statement must be tabled by 30 June 2006.

Our modelling starts with demographic projections of the population (adopting the official Statistics New Zealand projections), uses these to produce projections of GDP and then adds the central government sector.

The population modelling is based on continuing the present low fertility/low mortality state, following the transition from a high fertility/high mortality state over the past century. This transition is not unique to New Zealand. Importantly, it is not a demographic “bulge” that will reverse in time.

The economic model is based on the three Ps of Population, Participation and Productivity. The population component is derived from the Statistics New Zealand demographic projections, while for participation and productivity we are projecting a continuation of recent trends.

Modelling government spending for over 40 years into the future presents considerable challenges. We illustrate these by presenting examples of four different types of expenditure programmes:

  • New Zealand Superannuation: Projected expenditure is driven by existing scheme parameters and our population and economic projections. The result is that spending on this programme increases as a proportion of GDP.
  • Education: Future expenditure is driven off a combination of demographic factors (numbers of students) and assumptions about real spending per student driven by wages, resulting in a declining level of spending relative to GDP.
  • Core government services: Spending is projected at a fixed proportion of GDP.
  • Health: We combine a set of projections around demography, costs and policy choices to derive a level of spending that is increasing as a share of GDP.

We also illustrate our approach to projecting tax revenue, where we expect the tax-to-GDP ratio to increase slightly over time (the reason for the increase is that we are projecting that most tax bases grow in line with GDP, but that New Zealand Superannuation and other benefits, which are both taxed, will increase faster than GDP).

These examples also show how different parameters affect the projections in different ways. For example, our estimate of labour productivity has little impact on spending on New Zealand Superannuation as a proportion of GDP. This is because growth in GDP enters both the numerator and denominator of the spending-to-GDP ratio. In the numerator (spending), the level of benefits is linked to wage levels, which in turn is linked to productivity. GDP, the denominator, is similarly linked to productivity. Thus any change in productivity has an equal, and thus off-setting, impact on both spending and GDP.

Demographic projections can, however, affect the final results. Using Statistics NZ’s low mortality (higher longevity) assumption, rather than its preferred medium assumption would, for example, mean a greater proportion of elderly, a relatively smaller labour force, and a larger population by 2050. This is likely to place more pressure on the fiscal position.

The final Statement will include projections of all government spending programmes. Our plan is to present the results of the full model in two different ways.

We will start with a top-down approach, where the current set of fiscal aggregate objectives (most notably long-run spending-to-GDP, tax-to-GDP and debt-to-GDP ratios) is projected to continue. We will use the model to determine what spending or revenue track would be required to continue to meet these objectives, given likely demographic changes.

There are a number of different ways the top-down results could be modelled. For example:

  • With tax rates fixed, we could allow debt to move and see how it measures up to the long-term objectives.
  • With debt fixed, we could allow tax rates to move to accommodate the objectives.
  • With tax and debt ratios fixed, we could see just how much current spending ratios need to be pared back to satisfy our present long-term fiscal objectives.

We will then present a set of bottom-up projections, where we model the effect on the aggregate fiscal results of current policy in individual spending programmes and the current tax system projected forward on the basis of demographic and other assumptions.

Because the fiscal position is the result of policy choices, we also plan to accompany our projections with a discussion of how the current policy settings have evolved over the past few decades and where the spending share is likely to be heading. This will allow us to describe how different policy settings might impact on the fiscal position.

References

Aprile, R. (2004), “How to take account of death-related costs in projecting health care expenditure – updated version.” Ragioneria Generale Dello Stato.

Australian Productivity Commission (2005), Economic Implications of an Ageing Australia, Research Report, Canberra, April.

Barro, Robert (1990) “Government Spending in a Simple Model of Endogenous Growth.” The Journal of Political Economy 98(5) Part 2.

Bjornerud, Simen, and Joaquim Oliveira Martins (2005) “Disentangling demographic and non-demographic drivers of health spending: a possible methodology and data requirements.” OECD Economics Department.

Bryant, John (2003) "Can population projections be used for sensitivity tests on policy models?" New Zealand Treasury Working Paper 03/07.

Bryant, John (2003) “The Ageing of the New Zealand Population, 1881-2051.” New Zealand Treasury Working Paper 03/27.

Bryant, John (2003) “Demographic Change and New Zealand’s Economic Growth.” New Zealand Treasury Working Paper 03/04.

Bryant, John, Audrey Teasdale, Martin Tobias, Jit Cheung and Mhairi McHugh (2004) “Population Ageing and Government Health Expenditures in New Zealand, 1951-2051.” New Zealand Treasury Working Paper 04/14.

Burniaux, Jean-Marc, Romain Duval and Florence Jaumotte (2003) “Coping with Ageing: A Dynamic Approach to Quantifying the Impact of Alternative Policy Options on Future Labour Supply in OECD Countries.” OECD Economics Department Working Paper No. 371.

Callister, Paul (2005) “The Changing Gender Distribution of Paid and Unpaid Work in New Zealand.” New Zealand Treasury Working Paper 05/06.

Commonwealth of Australia (2002) “Intergenerational Report 2002-03.” 2002-03 Budget Paper No. 5, May.

Congressional Budget Office (2005), “The Long-Term Budget Outlook.” The Congress of the United States, December.

Cutler, David, Angus Deaton and Adriana Lleras-Muney (2006) “The Determinants of Mortality.” NBER Working Paper Series 11963, January.

Dang, Thai Than, Pablo Antolin and Howard Oxley (2001), “Fiscal Implications of Ageing: Projections of Age-Related Spending.” Economics Department Working Paper No. 305.

Davis, Nick and Richard Fabling (2002) “Population Ageing and the Efficiency of Fiscal Policy in New Zealand New Zealand.” New Zealand Treasury Working Paper 02/11.

Dunstan, Kim and Richard Speirs (2005) “Experimental Stochastic Projections of New Zealand’s Population.” Statistics New Zealand, December.

Economic Policy Committee (2001) “Budgetary challenges posed by ageing populations.” EPC/ECFIN/EPC/655//01-EN, European Commission.

Economic Policy Committee (2004) “Illustrative Long Run Projections for the EU15 and AC10 Countries up to 2050.” ECFIN/214/04-EN.

Economic Policy Committee (2006), “The impact of ageing on public expenditure: projections for the EU25 Member States on pensions, health care, long-term care, education and unemployment transfers (2004-2050).” European Economy Special Report No. 1/2006, February.

Graham, Patrick, Tony Blakeley, Peter Davis, Andrew Sporle and Neil Pearce (2004) “Compression, expansion, or dynamic equilibrium? The evolution of health expectancy in New Zealand.” J. Epidemiology and Community Health, 58, 659-666.

Gruen, David, and Matthew Garbutt (2004) “The long term fiscal implications of raising Australian labour force participation or productivity growth.” Australian Treasury Working Paper 2004-01.

Guest, Ross, Grant Scobie and John Bryant (2003) “Population Ageing in New Zealand: The Impact on Living Standards and the Optimal Rate of Saving with a Flexible Real Exchange Rate.” New Zealand Treasury Working Paper 03/34.

Guest, Ross, Grant Scobie and John Bryant (2003) “Population Ageing in New Zealand - Implications for Living Standards and the Optimal Rate of Saving.” New Zealand Treasury Working Paper 03/10.

Heller, Peter S (2002) "Considering the long-term in formulating fiscal policy." Internal paper for the New Zealand Treasury.

HM Treasury (2005) “Long-term public finance report: an analysis of fiscal sustainability.” This has been an annual report since 2002.

Hurnard, Roger (2005) “The Effect of New Zealand Superannuation Eligibility Age on the Labour Force Participation of Older People.” New Zealand Treasury Working Paper 05/09.

IMF (2004) “How will demographic change affect the global economy?” Chapter III in World Economic Outlook: The Global Demographic Transition, IMF Washington, September.

Janssen, John (2002) “Long-term Fiscal Projections and their Relationship with the Intertemporal Budget Constraint: An Application to New Zealand.” New Zealand Treasury Working Paper 02/04.

Johnston, Grant, (2005) “Women’s Participation in the Labour Force.” New Zealand Treasury Working Paper 05/06.

Lee, Ronald and Lawrence Carter (1992) “Modelling and Forecasting the Time Series of U.S. Mortality.” Journal of the American Statistical Association, 87(419) 659-71, September.

Lee, Ronald (2003) “The Demographic Transition: Three Centuries of Fundamental Change.” Journal of Economic Perspectives 17(4), 167–190.

Lee, Ronald (2004) "Quantifying our ignorance: Stochastic forecasts of population and public budgets." University of California.

Lee, Ronald, Michael Anderson and Shripad Tuljapurkar (2003) "Stochastic forecasts of the social security trust fund." Report for the Social Security Administration.

Martins, Joaquim Oliveira, Christine de la Maisonneuve (2005) “Projecting OECD health and long-term care expenditures: What are the main drivers?” OECD Economics Department, ECO/CPE/WP1(2005)14.

Montanino, Andrea, Bartosz Przywara and David Young (2004) “Investment in education: the implications for economic growth and public finances.” European Commission.

OECD (2001) “OECD Best Practices for Budget Transparency.” OECD, May.

Oeppen, Jim and James W. Vaupel (2002) “Broken limits to life expectancy.” Science.

Olshansky, S. Jay, Douglas J. Passaro, Ronald C. Hershow, Jennifer Layden, Bruce A. Carnes, Jacob Brody, Leonard Hayflick, Robert N. Butler, David B. Allison, and David S. Ludwig (2005) “A Potential Decline in Life Expectancy in the United States in the 21st Century.” The New England Journal of Medicine 352(11), March.

Stephenson, John and Grant Scobie (2002) “The Economics of Population Ageing.” New Zealand Treasury Working Paper 02/05.

The New Zealand Treasury (2005), “Half-Year Economic and Fiscal Update 2005.” B6, December.