Women and Superannuation in the 21st Century: Poverty or Plenty?

There are a number of modelling techniques that could be used to project the financial situation of retired women. These include hypothetical models, group models and dynamic microsimulation models.

NATSEM has developed a dynamic microsimulation model to show the implications of policy outcomes on individuals while being able to provide aggregate and average results. The dynamic model, DYNAMOD, begins with the 1986 Australian Bureau of Statistics (ABS) one per cent Census sample file. Onto the individuals on this file are imputed additional characteristics (such as more detail on the level and type of education being undertaken, the state in which the person lives, the value of the family home, etc). The result is a base population of 150,000 synthetic individuals with over 80 characteristics each. The model takes this base population and simulates their lives from 1986 onwards, by stepping month-by-month through time until the middle of the 21^st century. As the individuals step through life, they experience a range of life events, in line with Australian data about the probabilities of those events happening to real Australians. The life events include death; fertility; couple formation and dissolution; emigration and immigration; primary, secondary and tertiary education; labour force changes (full and part-time employment, unemployment and NILF), disability onset and recovery, and the earning of income.

These individuals are the “micro” part of the simulation. Analysis of any aspect of their lives can be done at this micro level. Additionally, individuals in the model are linked to simulate couple formation and the links broken to simulate couple dissolution. Links are maintained between parents and their children and between current and ex-spouses. In this manner, the model maintains the diversity in the population, copes very well with couple formation and dissolution, and, importantly, allows generation of a longitudinal history (or future) for individual cases in the model. This longitudinal element is essential in order to capture the accumulation processes that underlie growth in superannuation and financial assets. Alternative futures can be simulated by varying particular probabilities, such as the chance of becoming full-time employed or of having a first child, given certain characteristics. The model also contains a capacity to make the summed outcomes for all individuals within the model fit some exogenously imposed total. For example, the proportion of all women of a certain age range that are to be employed full-time in 2010 can be set by the user.

A wide range of technical papers provides detail of the processes used in DYNAMOD (detailed overviews are provided in Antcliff, 1993 and King et al., 1999). The latest version of the dynamic microsimulation software is DYNAMOD-3. This version adds life events relating to paying income tax, accumulating and diluting personal and family wealth assets and contributing to superannuation. It is this version and its outcomes that are discussed throughout the remainder of the paper.

Superannuation is a recent addition to DYNAMOD and this module promises to provide very detailed estimates of those who have superannuation and the amounts that are held. The accuracy of the modelling, naturally, depends on the availability of good quality data. For dynamic microsimulation there is an additional need and that is for very detailed and preferably longitudinal or panel data. Unfortunately the data available on superannuation holdings in Australia are limited. A significant step forward will be the release later this year of the confidentialised unit record or micro data from the 2000 ABS superannuation survey. At the present time, the major source of superannuation asset data is a survey of major private superannuation funds conducted by RIM in the early 1990s (Brown, 1994). This data collection, combined with ABS survey data, provides the foundation of superannuation asset modelling in DYNAMOD.

Superannuation is simulated from 1993 onwards. There is both RIM and ABS data available for that year. RIM, through its data collection and research, has estimated the average amounts held in superannuation in 1993 while ABS undertook a survey of superannuation coverage in that same year. The ABS data shows who had a superannuation account in 1993, while the RIM data indicates the balance of the account at that time. The simulation uses this coverage and account balance information to impute an initial superannuation amount onto individuals.

From 1993 onwards, a variety of transactions take place:

• Contributions equivalent to employer contributions are deposited into a person’s account as income is earned;

• A person may choose to make voluntary contributions into their superannuation account;

• Interest is paid on the account balance; and

• Accounts can be closed and the balance transferred into another type of account on retirement.

For reasons of practicality, the full complexity of superannuation could not be modelled and some areas were simplified. In some other areas assumptions had to be made as actual data were not available. The assumptions and simplifications include:

• The reduction of fund types to only one – the Accumulation Fund. This is the most common type of fund (87% of superannuation membership is of this type);

• Employer contributions are made for all employees at the superannuation guarantee charge (SGC) rate. In line with the legislation, the following are the only exceptions – individuals who are paid less than $450 per month; those employed but over 70 years old; and those employed but under 18 years old and not working full-time;

• Employee and voluntary contribution percentages are based on the 1995 ABS survey data for 45-74 year-old employees. For others in the labour force an employee contribution of either zero or 5% is assigned;

• Each person has only one account;

• Withdrawal of funds before retirement is not permitted; and

• All funds pay the same rate of return.

Each of these assumptions and their impact on the model is described in more detail in Appendix A.

Contributions to superannuation are made throughout the simulated working life of a person and interest is earned annually on funds invested in superannuation until retirement is reached, when the funds are transferred to the person’s cash account. The decision of a person to retire is done randomly but is based on probabilities derived from ABS data.

In the event of a divorce, the model currently includes superannuation in the division of family assets. After the division, the respective amounts are deposited back into each individual’s superannuation fund. In effect, DYNAMOD is enacting (a simplified version of) the legislation that is currently before the parliament.

Before using DYNAMOD to project outcomes in regard to superannuation, it may be appropriate to show that aggregate projections produced by the model agree with benchmark data. Unless a reasonable fit is achieved at this level, the credibility and usefulness of the model in regard to superannuation will be undermined.

Figure 1 is an example of the extent of concordance that is currently achieved between DYNAMOD projections for Australia’s population and ABS actual data and projections. The ABS projections are Series II.¹ As can be seen at this aggregate level, the net effect of the simulation of mortality, fertility and migration is to reproduce a population total that closely matches the ABS data.

Figure 1

Download asset Open asset

The alignment at this aggregate level is remarkable when you consider the different methodologies. ABS use a form of group modelling² while dynamic microsimulation looks at the circumstances of every individual, every month, and decides whether they will become pregnant or not; whether they will continue to live or not; and whether they will emigrate or not. These are very different approaches and the closeness of the simulated aggregates with the ABS data suggests that the underlying modelling is capturing individual behaviour. A major advantage of dynamic microsimulation modelling is that we can now “drill down” into this aggregate data or we can group individuals together to look at collective or group outcomes. We could look at the retirement assets of an individual who marries at 25, divorces at 35, is diploma qualified, has two children, works in part-time employment for four years, and is employed full-time for ten years; we could compare women and men at various ages; or we could project the number of marriages per 1000 population (Figure 2). It is this flexibility that gives dynamic microsimulation a significant advantage over other techniques.

Figure 2

Download asset Open asset

In a previous section the aggregate projected superannuation amounts were presented. In this section a more detailed look is taken at individual holdings. DYNAMOD begins simulating superannuation in 1993. In July of that year it allocated every person with superannuation cover an initial balance based on age and gender. In Table 1 this alignment by age and gender with the RIM data can be seen. The higher ‘All Ages’ value in DYNAMOD reflects the slight differences in spread of superannuation over the ages.

All values in the table are in real 1999 dollars. The final four columns of the table show the estimated superannuation assets in 2030 and the percentage change 1993-2030. While the average superannuation amounts for women are well below those for men, there is a marked improvement over the period 1993 – 2030. The average female superannuation assets increase from 45% of the average male in 1993 to 70% in 2030.

In the introduction the ageing of the population and the future fortunes of women were discussed. An analysis of ABS population projections highlights both the ageing of the population and its impact on women.

3.4.1 Ageing

In Figure 5 and 6 significant change to the Australia age profiles can be seen. In the two comparisons, the population in 1991 (dark colour) is compared with 30 and 60 years later, while the percentage of the population is presented in five year age groups from 0 to 85 and 85+. In 1991 the largest groups were the three groups covering ages 20-35 (ie born from 1956-1971), with a fairly clear decrease in the percentages above that age. The age group 85+ represented only 0.9% of the population (0.25% males and 0.63% females). By 2021 the peak age groups will have flattened and become broader. The 1991 Christmas tree shape has transformed into a more rectangular one. There is now a fairly even distribution of people between the ages of 20-65. The average age has increased and the proportion aged 85+ has more than doubled to 2.0%.

Figure 5

Download asset Open asset

Figure 6

Download asset Open asset

By 2051, as seen in Figure 6, the age groups over 40 contain the higher proportions of people and the rectangular shape is beginning to be replaced by an inverted pyramid. The 85+ population has now become quite significant, representing 4.5% of the population, with almost two out of three of those aged 85+ being women (64.2%).

In the space of only 60 years, from 1991 to 2051, a population with the majority aged below 40 years will be supplanted by a population dominated by people aged over 40 — while the proportion aged over 85 will have increased five fold.

The five-fold increase in people aged 85+, along with the significant increase in the proportion aged 65 and over (increasing from 11% in 1991 to 24% in 2051), will result in greatly increased medical costs and nursing requirements. These costs and requirements are going to continue to grow and place an increasing burden on people’s own financial resources, their relatives and/or on the government. The five fold increase in women aged 85+ makes them vulnerable to bearing a high proportion of this cost.

In summary, the proportion of people aged 85+ will increase five fold by 2051 and women will represent two-thirds of the people in this age group. The proportion aged 65+ will increase to 24 percent of the population.

3.4.2 Trends in Marital Status

The ageing of the population will impact on women but a number of other changes will also have an impact. One of these is the trend towards later marriage. Figure 7 shows the forecast decrease in the proportion of females married in the two age groups up to age 35 and the higher proportion of females married in their 40s and beyond. The ABS estimate that the median age for women at first marriage has increased from 24.0 years in 1988 to 26.2 in 1998. The median age for women at remarriage has also increased, from 35.3 to 38.4 over that period (ABS 4102.0). These trends are reflected in the declining proportion of younger women who are married.

Figure 7

Download asset Open asset

There are two possible reasons for the increased proportion of women married at ages 40 and above. The first is that since marriage started at an older age, it will finish at an older age. The assumption here is that marriages are of the same duration. In fact, the duration of marriages increased slightly over the period 1988-98 – by 0.5 years to 7.8 years (ABS 4102.0). The end result of getting married 2-3 years later and the marriage lasting 6 months longer would have the impact of moving the age distribution curve to the right as we see in Figure 7. A second or alternative reason for the movement in the curve is the ageing discussed in the previous paragraph. While on average women still outlive men, the ageing of both genders causes wives to become widows at a later age. This trend can be seen in Figure 8. The proportion of females who are widows decreases between 1990 and 2030/2050.

Figure 8

Download asset Open asset

An alternative view of the changing marital status of women around retirement age is presented in Figure 9. The figure shows the outcomes produced by DYNAMOD for women aged 55 – 64 over the period 1986 – 2030. Each colour represents the proportion of women in this age group with that particular status. The proportions are cumulative, with the proportion not shown being married women (de jure or de facto). As an example, in 1986 28% of women aged 55-64 have a marital status other rather “Married’, while the remaining proportion (72%) are married. The figure therefore shows an increased proportion married over the period 1986-2030. A decrease in the “widowed” proportion is also evident.

Figure 9

Download asset Open asset

The labour force participation of men and women is markedly different. Women have always been more likely to be employed part-time. In 1999 the proportion of women employed part-time was 43.5 percent, while for men it was only 12.5 percent (ABS 4102.0 pp.108). This wide disparity was evident throughout the 1990s, despite a significant growth overall in part-time employment for both genders (the overall proportion employed part-time rose from 20.9% in 1990 to 26.0% in 1999). Labour force participation rates are also quite different when compared on a gender basis. The female participation rate rose by to 2.0 percentage points to 53.9 percent in the 1990s while the male participation rate fell 2.7 to 72.8 percent over the same period. In some age groups the difference was even greater. For example, the participation rate trend for 55-59 year old men is slowly downwards (from 75.0% to 73.2% in the 1990s), while the trend for women of the same age is rapidly upwards (up from 32.4% at the start of the 1990s to 43.6% at the end) (ABS 4102.0 pp. 109).

Overall, female labour force participation and the type of employment are changing. The proportion not in the labour force is falling and the proportion in part-time employment is rising. Both DYNAMOD projections and ABS actual data suggest that some of the part-time rise may be at the expense of full-time but most is coming as a result of choosing to participate in the labour force on a part-time basis rather than not participate at all.

An advantage of dynamic microsimulation is that these differences can be easily handled. The decision to participate in the labour force or to change employment status is done at the individual level — and hence the influences on the decision are applied at this level and can vary over time.

A different way of looking at the labour forces changes is to consider a particular group over a period of time. This is done for females aged 45-54 in Figure 10. The decreasing proportion not in the labour force is evident, along with the increasing proportion in part-time employment. The growth in part-time comes mainly from females entering or re-entering the workforce (the decreasing proportion NILF) but there is a small impact on the full-time participation as well. A slight decrease in females employed full-time can be seen from 2000 onwards which is perhaps a reflection of the desire to be in the labour force yet still have time for other things.

Figure 10

Download asset Open asset

To accumulate superannuation assets a person must have superannuation coverage. As discussed in the section Coverage Projection (page 11) almost everyone will have had superannuation contributions made on their behalf by retirement because of the government’s SGC policy. The amount that will be held in this superannuation account will depend on three factors – the rate of return of the superannuation fund, the contributions made into the account, and taxation on the superannuation. The last of these, taxation, is not part of NATSEM’s modelling at this stage. The SGC employer contributions are a percentage of earnings, the higher the earnings and the longer the time in employment, the greater the superannuation assets. In most cases a working life spent in full-time, high salary employment and not interrupted by any periods of unemployment or NILF would produce the highest superannuation. This set of circumstances is most likely to achieved by males — and it is for this reason that average female superannuation amounts were only 45% of those of males in 1993. However, the increasing age of marriage and childbirth, reduced periods of NILF for child raising purposes, and the increased use of part-time employment are having a positive effect. Female superannuation is increasing as a proportion of male superannuation, and is estimated by DYNAMOD to reach 70 percent by 2030. Unless there is complete equality in the labour force roles, the female proportion will remain lower than the male, due partly to lower female earnings and different workforce participation patterns in the child bearing and rearing years.

The estimated amounts of superannuation held by various age groups of women are shown in Figure 11. The impact of compound growth is evident in the figure. In 1993 few women had contributed to superannuation in the early part of their working life and hence the balance in their superannuation account did not grow through interest being paid and added each year. In comparison, by 2030 most women will have made superannuation contributions at the start of their working life. For the next 30-40 years that contribution will have been added to each year through interest. This, combined with more years of employment than the previous generation’s, results in exponential growth in the value of the asset.

Figure 11

Download asset Open asset

As Table 1 (Page 13) suggests, women’s accumulated superannuation upon retirement is expected to increase very rapidly over the next three decades. For example, the average account balances of women aged 55 to 59 are forecast to increase to $196,930 by 2030, up from only $23,993 in 1993. This is an increase of more than 700 per cent. Similarly, women aged 60 to 64 are forecast to have average account balances of just over $190,000 in 1993, up from about $30,000 in 1993 – a 449% increase.

Table 2 shows the superannuation balances for females aged 55-64 at three points across the superannuation spectrum – the 10^th percentile, the median (or 50^th percentile) and the 90^th percentile. 80 percent of females with superannuation in this age range have a balance between the 10^th and 90^th percentiles.

A very encouraging feature of this table is the huge increase for the 10^th percentile. In real terms it increases by almost $90,000 over the 30 years. Restating this, by 2030 DYNAMOD estimates that 90 percent of women aged 55-64 will have approximately $93,000 or more in superannuation.

ABS Projection Series II most closely reflects prevailing trends. The Series II assumptions are (i) a declining mortality rate; (ii) fertility rates falling to 1.75 by 2005/6 and then constant; and (iii) a net annual immigration gain of 70,000.

‘The ABS uses the cohort-component method which begins with a base population for each sex by single years of age and advances it year by year by applying assumptions regarding future mortality and migration.’ (ABS 3222.0 pp.125)

Historically there have been three types of superannuation funds. The first is the “defined-benefit” scheme where the pension entitlement is a complex calculation involving the number of years in the scheme and the salary at the time of retirement. This type of fund is predominately found in the public sector and in large white collar industries. In general these funds are heavily subsidised by the employer and very generous in their benefits. However, most defined-benefit schemes are now closed to new entrants.

The second type of scheme is the ‘accumulation’ scheme (or defined-contribution scheme). With the accumulation scheme, contributions (by the individual and/or their employer) and interest build up in a personal account over time. The balance of this account at the time of retirement represents the entitlement. This is the most common type of scheme in the private sector. The introduction of the three percent industrial award superannuation, and then the Superannuation Guarantee Charge (SGC) in 1992, have resulted in accumulation funds becoming the dominant form of scheme. In June 2000, 87% of superannuation members were in an accumulation scheme (see Table A1).

A third type of fund is a ‘hybrid’ scheme, representing a combination of the defined-benefit and accumulated funds.

Given the predominance of the accumulation scheme, in DYNAMOD all superannuation entitlements are treated as if they were of the accumulated type. RIM noted that members of defined-benefit schemes in 1993 had an older age distribution and had a heavier concentration of higher salary members than members of accumulated schemes. This resulted in the average entitlement in defined-benefits accounts being 3.8 times higher than in accumulation schemes. The omission of defined-benefit funds may therefore underreport aggregate superannuation in the early years of the simulation.

Under industrial award superannuation and then under the SGC, employers have been required to make minimum contributions for most employees to a complying superannuation fund. The minimum contributions under SGC are shown in Table A2.

Superannuation contributions do not have to be made for some categories of employees. These categories include employees who are:

• paid less than $450 in a month;

• aged 70 or over;

• under 18 years of age and working for not more than 30 hours a week;

• performing work of a domestic or private nature for not more than 30 hours a week for a non-business employer; and

• who have vested benefits in excess of their pension RBL and who elect not to receive any more employer contributions.

Based on the exclusions above, the simulation assumes that everyone receives an employer contribution equal to the rate defined in Table 2. The only exceptions are those with gross annual earnings less than $5400 (450 x 12); those under 18 years and not working full-time; and those over the superannuation age limit (currently set at 70). Again there may be some underreporting in the early years as some organisations pay more than the SGC amount. This is particularly true in the public sector.

In addition to the employer contributions to an individual’s superannuation, a person may either be required to contribute part of their salary into a superannuation fund or may voluntarily contribute to their own superannuation fund. In DYNAMOD, employee contributions are added to the same account as the employer funds. Whether a personal contribution is made and, if so, the amount contributed, is based on two sources - a RIM conference paper (Tinnion, 1998) for non-employees and those aged under 45; and the 1995 ABS publication ‘Superannuation Australia’ for employees aged 45+.

For people who are currently unemployed, not in the labour force or doing unpaid work, it is assumed they make no contribution to their superannuation. If a person is employed then the chance of them contributing is based on a probability distribution. The distribution is shown in Table A3 below. The amount contributed is five percent of their salary. This rate is one of the most common default rates used by organisations that require employees to contribute to a superannuation scheme.

For employees aged 45 –76, a more complex algorithm is used based on ABS data. ABS Cat 6319.0 Superannuation Australia November 1995 “Table 13" shows employees aged 45-76 making various personal contributions (see Table A4 below). On the basis of this table, an annual employee contribution rate is imputed for all employees aged 45-76.

The reason for using this income-based employee data is to attempt to capture changing rates of contribution. There is evidence that the favourable taxation rates on superannuation contributions may induce some individuals to contribute large amounts to their superannuation fund. This is particularly true for high income individuals. Evidence of this can be seen in Figure A1. The figure shows a correlation between weekly income and the chance of giving 10-15% of income (a high rate).

The use of the ABS data shown in Table A4 attempts to model the relationship between increasing contributions and increasing income.

Figure A1

Download asset Open asset

The existence of multiple accounts for individuals is well known and is discussed in Rothman (1995). The reasons for the multiple accounts include:

• People holding more than one job;

• Requirements that some award payments be paid into a different account;

• Timing difficulties, that is, old accounts being reported as current when in reality they are inactive (for example the person may have changed jobs and started in a new scheme/account and data relating to the old account is still being reported); and

• Other unusual arrangements, such as where superannuation associated with performance pay for public servants was put in a separate account

DYNAMOD calculations are based on individual data rather than account data and DYNAMOD assumes there are no differences between superannuation funds. These assumptions mean that the there is no difference between a person having three accounts with $1,000 in each and a person having $3,000 in one account. The existence of multiple accounts does not impact on the model.

In general, funds contributed to superannuation schemes cannot be withdrawn until retirement (and being aged over 54). However, the superannuation legislation did allow access to funds in certain circumstances — namely times of severe financial hardship or emigration from Australia. There was some abuse of these guidelines, with the result that the legislation for gaining early access was tightened and the preservation age increased to 60 years for those born after June 1964. From 1 July 1999, all earnings and contributions can only be paid out if a person dies, becomes disabled, or APRA approves an early release. Permanent emigrates have to wait for their funds until retirement. These changes have had such an impact as to make early withdrawals a minor, short term case which can be ignored. Early withdrawals are not modelled in DYNAMOD.

An average rate of return for superannuation funds is very difficult to obtain. Firstly, there are some 213,747 superannuation funds in Australia (APRA June 2000). Of these, 211,175 have less than five members. This number of small funds makes calculation of an overall rate of return for superannuation funds difficult and perhaps meaningless, as it is possible that a number of these small funds are not motivated by maximising returns over the long term. Exclusion of these small funds reduces the total assets by only 15% (down $68B to $409B) and should allow the rate of return of the remaining major funds to be calculated. However, a second problem with measuring rates of returns then becomes evident - rates of return can still not be accurately obtained as the superannuation industry’s ‘reserving’ practice. Reserving involves paying a smaller rate of return in good years and putting the extra funds aside for use when the actual rate of return is negative or low. A number of superannuation funds use reserving to implement a policy of not passing on negative returns. In years when the actual return is negative or zero, the reserve is used to pay a small positive return. The practice of using reserves results in a smoothing of rates and makes measurement of actual real returns very difficult to obtain.

An InTech survey (2000) shows an asset weighted average for a range of major retail growth funds of 6.3% in real terms over the last seven years, and 8.4% over the last five. The RBA in Occasional Paper No 8, “Historical Economic Statistics”, show average ten year growth for a balanced fund of 13.0% to 1995, but inflation was an average 5.4% pa over the period giving a real rate of 7.6%. ARPA and Treasury use 8% nominal as an upper bound.

ARPA suggest that forward projections above 5% or 6% real are likely to be unsustainable.

The consensus view on these figures would suggest a real rate of 6% for 1986-94, 8% for 1995-99 and 5% for 2000 onwards. These are rates used in DYNAMOD.

No specific funding for this article is reported.

1. Version of Record published: August 31, 2023 (version 1)

© 2023, Kelly et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.