The Treasury

Global Navigation

Personal tools

2 Social expenditure

The basic structure of the CM model is based on Creedy and Scobie (2005) and is shown in Figure 1.

Figure 1 – The Creedy-Scobie Social Expenditure Model
Figure 1 - The Creedy-Scobie Social Expenditure Model.

Using historical evidence on the mean and variance of per capita expenditures for a set of 13 different social expenditure categories, the CM model then applies various labour productivity, participation rate, unemployment rate, and so on, assumptions to derive stochastic projections of aggregate social expenditures and GDP. Variables for which input data are required are shown in shaded boxes, while the model's outputs are shown in white boxes. The model uses input data on fertility and migration from Statistics New Zealand (SNZ), together with data on mortality and initial (2010) population levels from to derive stochastic population projections to 2060. Data on unemployment rates and labour force participation rates (separately for males and females) from SNZ yield numbers of[1] workers, which are combined with initial (LTFM) productivity levels, and assumed 1.5% per annum growth, to yield GDP projections.

Total social expenditures are obtained from data on initial (2010) social expenditure per capita, by age and gender, for 13 separate categories.[2] These may be grouped as follows: health (personal; public; mental; DSS 65+; DSS <65)[3]; education (primary/secondary; tertiary); NZ Superannuation, and welfare payments (Domestic Purposes and Widow's Benefit; Invalid's and Sickness Benefit; Family Assistance; Accommodation Supplement; Unemployment Benefit).

The categories of total public spending that are excluded from the model comprise mainly of Law & Order spending and debt interest payments. According to an update of Treasury (2012a: External Panel 1 Summary, Table 2), in 2010 these four social spending categories listed above represented 24.3% of GDP with a further 9.5% of GDP accounted for by remaining spending. Social spending represents over 70% of total Crown spending.

The model combines per capita social expenditures with population and participation rate data and projects forward using the population and labour force estimates. Confidence intervals around the arithmetic mean projections for total social expenditure categories are obtained from observed variances for the main per capita social expenditure categories, and unemployment, over the period 1960-2000 (1980-2000 for unemployment). This permits examination of trends over time in age- and gender-specific social expenditure-to-GDP ratios, with associated confidence intervals.

Figures 2 and 3 show two of the important input data components of the CM model: respectively, labour force participation rates, and per capita expenditures for the four expenditure sub-aggregates (health, education, NZS, welfare payments), by age.

Figure 2 – Labour force participation rates by age
Figure 2 – Labour force participation rates by age.
Figure 3 – Per capita social expenditures by age (males)
Figure 3 - Per capita social expenditures by age (males).

It can be seen that the participation rates follow the typical pattern of rapidly rising rates in the 15-25 age range and rapidly falling rates in the 55-65 age range, with a plateau between. Female participation rates are generally lower than males, especially but not exclusively during the child-rearing age range of around 25-45. The distribution of social spending by age in Figure 3 unsurprisingly reveals the sharp rise in NZS expenditures at age 65 and also reveals the initially gradual rise in per capita health spending from childhood, but with an accelerating rise, especially from around age 55. Welfare benefit payments, by contrast, taper off from a maximum around age 35 to small amounts by age 65, when NZS generally becomes available.

An alternative approach to projecting social spending is Treasury's Long-Term Fiscal Model, LTFM; see Bell et al. (2010) and Rodway (2012). This projects social expenditures forward using a somewhat different, and more aggregate, procedure than the CM approach. In particular, the LTFM's social expenditures are less disaggregated by gender and are based on fewer social welfare payment categories, and the model does not incorporate stochastic aspects.

Furthermore, the LTFM uses the Treasury short-term forecasting model to forecast relevant variables over the first five (or so) years and adopts its longer-term projection assumptions thereafter. Thus, for example, whereas CM use data for 2010 to provide the starting point for projections, the LTFM involves a “return to trend” process during the forecast period (2012-16) and then projects forward from the final forecast year. One consequence is that the CM benchmark results are based on a starting unemployment rate of 6.6% whereas the LTFM projections start from a lower projected unemployment rate of 4.7% in 2016. The latter might be expected to be closer to an on-trend value.

Figure 4 shows a benchmark simulation for total social expenditures as a ratio of GDP over 2010-60 using the CM model.[4]

Figure 4 – Social expenditure/GDP projections: Benchmark case
Figure 4 – Social expenditure/GDP projections: Benchmark case

The central estimate (black unbroken line) and the 50% (grey unbroken lines) and 95% (black dotted lines) confidence intervals are shown. Beginning at around 24.6% in 2010, the central estimate projects a value of 28.2% of GDP in 2060, but with this rise largely being achieved over the period 2020-40, with a relatively flat profile otherwise. However, applying the estimated variances over such a long period generates a wide range of possible outcomes: by 2060 the 50% confidence interval is 24-42% and the 95% confidence interval is 20-37%. Importantly it appears that the period when baby-boom effects dominate, during 2020-40, accounts for almost all of the projected increase in social expenditures. This aspect is discussed further below when comparing results across models.

Figure 5 compares the mean outcome for this benchmark simulation with two other simulation model outcomes: the Treasury LTFM projection (Treasury, 2012a, Table 2, p.5) and the equivalent projection made by Creedy and Scobie (2005).

Figure 5 – Alternative social expenditure models
Figure 5 – Alternative social expenditure models

This last projection was based on a 2001 starting point with projections to 2051. As a result it has a confidence interval around its central estimate for 2011 which provides a convenient comparator against which to assess the latest observed (2010) values.

Two features stand out from Figure 5. Firstly, the LTFM and CM benchmark simulations start (2010) and finish (2060) at almost identical ratios but follow quite different paths in between. Secondly, the Creedy and Scobie (2005) projections for 2011-2051 reveal a similar pattern to the benchmark case but, in effect, with a delay of 10 years. For example, Creedy and Scobie (2005:27) projected an almost unchanged social expenditure-to-GDP ratio from 2001-2011 (not shown in Figure 5) and this is repeated by CM for 2010-2020. Similarly both models project a rapid rise in the ratio after the first decade of projections and a relatively flat profile for the last decade.

Considering 2010, Figure 5 also suggests that the Creedy and Scobie (2005) central projection of an unchanged social expenditure/GDP ratio at around 23% over 2001-11 appears to have been below the actual value, but is within the 50% confidence band. The period in fact saw substantial increases in social expenditures, for example associated with policy initiative such as Working for Families. Nevertheless, the evidence of a rising ratio when previous projections expected this to be constant suggests caution in interpreting projections of a constant or declining (LTFM) value over the next decade.

The Treasury LTFM projects a decline in the social expenditure ratio over 2010-2020 and a steady rise thereafter to 2060. The initial decline appears to result mainly from the LTFM's use of the Treasury forecasting model over the first five years. Since the New Zealand economy in 2011-16 is expected to transition from below-trend towards on-trend values of key macroeconomic variables (such as unemployment, social welfare payments and GDP levels), this largely explains the decline in social expenditure/GDP ratios over the next decade from their 2010 values. By abstracting from short-term fluctuations, especially the 2010 below-trend values, the benchmark projection may over-estimate the 2020 social expenditure/GDP ratio.

A more detailed breakdown of the LTFM-CM benchmark differences reveals that, apart from the 2010-20 period, social expenditure trends are almost identical in the two models. However, GDP trends differ. Figure 6 shows an index of GDP for each model (2010 = 1.0) on the left-hand axis, and the percentage difference between the two model GDPs on the right-hand axis.

Figure 6 – Differences between the Treasury LTFM and benchmark projections
Figure 6 – Differences between the Treasury LTFM and benchmark projections

This shows clearly that differences between the two models are minor except for the last two decades when the benchmark GDP is around 5-10% higher than the LTFM case. GDP projections so far into the future, by either model, embody large margins of error which could well substantially exceed the 5-10% difference between the models such that these differences do not carry much statistical confidence. Nevertheless, it is interesting that such relatively small trend GDP differences can generate such different trends in social expenditure/GDP ratios over the 2040-60 period; one model projects a rising ratio while the other projects a constant, or declining, ratio. It also suggests that model results may be quite sensitive to assumptions about future labour force participation rates (especially among the elderly), the main source of difference between the LTFM and benchmark GDP trajectories.

Figure 7A provides a decomposition of the total social expenditure/GDP ratio into the four main sub-aggregates (health, education, NZS, and non-NZS welfare benefits), while the right-hand panel (7B) shows the equivalent ratios produced by the LTFM.

Figure 7 – Decomposing social expenditure trends

Figure 7A shows that the flattening of the total social expenditure profile after 2040 is largely caused by a flattening of the health and NZS expenditure tracks. These in turn reflect the model projections of reduced ageing effects on these spending categories, as ratios of GDP, due to the projected increase in participation rates for older individuals. The LTFM, on the other hand, shows continuing upward trends in those two spending categories after 2040.

To examine the sensitivity of the benchmark simulations to assumptions regarding spending trajectories, Figure 8 shows alternative scenarios for health and welfare benefit spending.

Figure 8 – Testing growth assumptions

A plausible argument suggests that per capita real health costs may grow faster than the assumed 1.5% a year. productivity increase. Treasury (2012b), for example, assumes that health costs rise over time owing to slower public sector productivity growth compared to the economy as a whole, and “volume growth” due to non-demographic factors between 0.8% and 2.0% each year. Second, the assumed 1.5% real growth in (non-NZS) welfare benefits exceeds that which is written into current legislation, and historical patterns. The legislation specifies benefit levels indexed to prices only, while evidence on actual total welfare benefit payments suggests that over the last 20 years these have growth faster than consumer prices but slower than nominal incomes, that is, some real increase but less than real productivity/wages. Simulations in Figure 8 report the effect of reducing the real annual welfare payment growth rate from 1.5% to 1.0%.

The increased health cost assumption generates a pattern that begins to resemble the LTFM case: the post-2040 pattern now shows an upward trend, albeit less than during the 2020-40 period. The trend in welfare payments is now downward over 2020-60 rather than approximately constant, being about 1 percentage point of GDP lower by 2060 compared with the benchmark. These results suggest that identifying the most relevant policy settings and the profile of health costs could be important for conclusions regarding the expected future cost of providing these social services via the public budget. In the case of health, for example, the benchmark rise of only around 1.5 percentage points, 2010-60 (6.9% to 8.4%) becomes a rise of almost 4 percentage points (6.9% to 10.7%) over the same period, when health costs are assumed to rise faster than economy-wide productivity by 0.5% per year.

The impact of NZ Superannuation settings (for example, age of eligibility, indexation of NZS levels) continues to be a controversial issue in debates over the future pension costs of ageing. The CM model can be used to examine this by, for example, allowing labour force participation rates to change in response to a higher NZS eligibility age. Of course, increasing the NZS eligibility age could be implemented in several ways involving, for example, a shorter or longer transition period. Figure 9 shows the impact of two hypothetical simulations. In 9A & B, the model is run assuming: (i) the NZS age is raised in 2010 from 65 to 70 (for males and females) and; (ii) the new age of eligibility of 70 is introduced 2020. Neither of these scenarios is realistic, of course, but they provide a sense of the boundary impacts that a large NZS age-related change, over a short time frame, could produce.

Figure 9A shows NZS expenditure and 9B shows total social expenditure: non-NZS spending and GDP are affected by assumed participation rate changes in response to the age change. In particular, the model assumes that the participation rates for males and females rise especially in the 65-69 age range, with smaller increases in the 55-64 and 70-74 age ranges; see Figure 10.

Figure 9 – Effects of increasing NZS age of eligibility by five years
Figure 9 – Effects of increasing NZS age of eligibility by five years
Figure 10 – Participation rate changes with NZS eligibility at age 70
Figure 10 – Participation rate changes with NZS eligibility at age 70

Figure 9A shows that, had the age change to NZS been introduced in 2010 this would have reduced NZS spending by about 2 percentage points of GDP in 2010 (from 5% to 3%), rising to almost 3 percentage points by 2060. Introducing the change in 2020 has a smaller immediate effect in that year (to 5.3% instead of 5.9%). But the longer-term effect is also smaller: by 2060 NZS expenditure is around 6.6% instead of 7.4%. This small effect, compared with the case of a 2010 NZS change, largely reflects the fact that many of the baby-boomer retirees retain retirement eligibility under the 2020 option and the associated participation rate and productivity gains are lost. Hence, NZS costs relative to GDP are noticeably higher when the increased retirement age is delayed by a decade.[5]

The effects of these NZS change scenarios on total social expenditures are shown in Figure 9B. This indicates a fall in the total social expenditure to GDP ratio in 2020 by around 3 percentage points (from 25% to 22%) when NZS age is increased in that year. This 3-percentage-point difference is generally maintained throughout 2020-60 so that social expenditure is projected to be approximately 25% of GDP instead of 28% by 2060. That is, these mean estimates suggest that the NZS increase in the age of eligibility largely compensates for the increase in social expenditures that is otherwise projected to occur. However, as the comparison of 2010 and 2020 implementation of a 5-year increase demonstrates, delays in implementing an NZS age increase could have a large effect on this conclusion, as the potential near-term cost savings and participation improvements are foregone.

Notes

  • [1]The benefit system has more recently undergone a number of changes, including name changes.
  • [2]Data on the four main categories are obtained from: health (LTFM); education (schools - Treasury, based on Ministry of Education administrative data; tertiary - SNZ); NZS and welfare payments (Household Economic Survey (HES)), and Treasury's personal tax/transfer simulator, Taxwell).
  • [3]DSS = Disability Support Services.
  • [4]Each single year is necessarily projected in the CM model, but for convenience the results after each ten year period are plotted (and hence connected by straight lines in the diagram)
  • [5]These cost simulations are based on “steady state” or “full implementation” assumptions and hence should not be interpreted as capturing expected actual NZS cost changes under more likely implementation scenarios.
Page top