Development of Treasury's new model of Australian retirement incomes and assets: MARIA

Bastian, Bruce et al. Working Paper Development of Treasury's new model of Australian retirement incomes and assets: MARIA Treasury Working Paper, No. 2017-02 Provided in Cooperation with: The Treasury, The Australian Government Suggested Citation: Bastian, Bruce et al. (2017) : Development of Treasury's new model of Australian retirement incomes and assets: MARIA, Treasury Working Paper, No. 2017-02, ISBN 978-1-925504-34-7, The Australian Government, The Treasury, Canberra This Version is available at: https://hdl.handle.net/10419/210394 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/au/legalcode DEVELOPMENT OF TREASURY’S NEW MODEL OF AUSTRALIAN RETIREMENT INCOMES AND ASSETS: MARIA Bruce Bastian, Matthew Smith, Belinda Cheong, Virginia Pineda, Martin Stevenson, Owen Hutchison, Madeleine Nargar, Susie Kluth 1 Treasury Working Paper 2 2017-02 Date created: February 2017 Date modified: February 2017 1 Tax Analysis Division, Revenue Group, The Treasury, Langton Crescent, Parkes ACT 2600, Australia. Correspondence: [email protected]. 2 The views expressed in this paper are those of the authors and do not necessarily reflect those of The Australian Treasury or the Australian Government. We thank Phil Gallagher, Matt Brine, Robert Ewing, Gareth Wett, Mark Bott and William Young for their valuable feedback on an early draft of this paper. © Commonwealth of Australia 2017 ISBN 978-1-925504-34-7 This publication is available for your use under a Creative Commons BY Attribution 3.0 Australia licence, with the exception of the Commonwealth Coat of Arms, the Treasury logo, photographs, images, signatures and where otherwise stated. The full licence terms are available from http://creativecommons.org/licenses/by/3.0/au/legalcode. Use of Treasury material under a Creative Commons BY Attribution 3.0 Australia licence requires you to attribute the work (but not in any way that suggests that the Treasury endorses you or your use of the work). Treasury material used 'as supplied' Provided you have not modified or transformed Treasury material in any way including, for example, by changing the Treasury text; calculating percentage changes; graphing or charting data; or deriving new statistics from published Treasury statistics — then Treasury prefers the following attribution: Source: The Australian Government the Treasury Derivative material If you have modified or transformed Treasury material, or derived new material from those of the Treasury in any way, then Treasury prefers the following attribution: Based on The Australian Government the Treasury data Use of the Coat of Arms The terms under which the Coat of Arms can be used are set out on the It’s an Honour website (see www.itsanhonour.gov.au). Other uses Enquiries regarding this licence and any other use of this document are welcome at: Manager Media Unit The Treasury Langton Crescent Parkes ACT 2600 Email: [email protected] Development of Treasury’s new Model of Retirement Incomes and Assets (MARIA) Bruce Bastian, Matthew Smith, Belinda Cheong, Virginia Pineda, Martin Stevenson, Owen Hutchison, Madeleine Nargar, Susie Kluth 2017-02 17 March 2017 ABSTRACT Treasury currently has a project underway to significantly enhance its capability to produce detailed modelling of retirement incomes in Australia. The Model of Australian Retirement Incomes and Assets (MARIA) is a long-term dynamic microsimulation model of Australia’s retirement income system. When fully operational, MARIA will provide whole-of-population projections beyond the medium term of: Age Pension expenditure and take-up; superannuation funds under management; and retirement income adequacy. It will also enable detailed distributional analysis of both current and hypothetical policy settings. This paper provides high-level details of the model and outlines a number of challenges and research questions for further consideration by Treasury and interested researchers. Keywords: Retirement incomes, superannuation, pensions, modelling, microsimulation, Tax Analysis Division, Revenue Group Matthew Smith Tax Analysis Division Revenue Group The Treasury Langton Crescent Parkes ACT 2600 Dr Susie Kluth Tax Analysis Division Revenue Group The Treasury Langton Crescent Parkes ACT 2600 1 1. INTRODUCTION A clear understanding of how retirement income policies affect both the Commonwealth budget and retirement incomes of individual Australians is essential for effective policy development. However, because retirement income policies take decades to mature, it is necessary to model 40 or more years into the future to fully understand their effects, and this is an intrinsically challenging task. Treasury has been producing detailed modelling of retirement incomes since the substantial reform of the superannuation system in the 1990s 3 and currently has a project underway to significantly enhance this capability. The new Model of Australian Retirement Incomes and Assets (MARIA) will provide detailed information regarding sustainability issues, such as long-term estimates of superannuation tax concessions, Age Pension expenditure and the savings held outside superannuation; and adequacy issues, such as distributional analysis of superannuation assets and retirement incomes. MARIA will be capable of projecting the current retirement income policy settings and generating forecasts for key variables, as well as the impacts of changes to these policies or economic conditions. Treasury’s previous retirement income model RIMGROUP was initially built in the early 1990s and served the department well, including through four Intergenerational Reports. 4 RIMGROUP was so named because it grouped the population into cohorts based on year of birth, gender and life-time income decile. Each group was further sub-divided into six accumulation account types with different numbers in each, but the accumulation account types were pooled at retirement. It was built at a time when computing power was limited and detailed administrative data and longitudinal data were not available to the Treasury. Instead RIMGROUP relied on ABS Survey of Income and Housing and Australian Prudential Regulation Authority information supplemented by Treasury analysis. The modelling had some limitations due to ‘pooling’ of work histories, account balances, voluntary saving levels and so on within each account type within each group. For example, if a group contained both voluntary savers and super guarantee-only cases their voluntary saving and outcomes were averaged at retirement, giving the whole group moderate levels of assets. Similarly migrants were pooled with others in the model and were assumed to have similar assets to the income group they joined. 5 In addition, the make-up of the retirement income system has changed fundamentally since the model was built, for example self-managed superannuation funds have become an increasingly large component of the sector and non-concessional savings have significantly increased for a group of high-saving individuals. 6 There have also been substantial changes in the diversity of income stream products like defined benefit pensions, annuities and term products. These changes were not straightforward to incorporate into the existing model structure. As the superannuation system has matured, a new broad range of administrative and longitudinal data on drivers of retirement have become available. This, combined with substantial advances in computing power and modelling software, allow for modelling of life paths at an individual level. In light of these developments, Treasury has decided to invest in the development of a new microsimulation model. Such a model can better capture the diversity of the population, and allows for greater versatility of output and analysis than cohort models like RIMGROUP. 3 This was the effective birth of the current retirement income system comprised of a means tested aged pension; superannuation and voluntary savings outside of superannuation. A more fulsome historical overview of the superannuation system was provided in Appendix B of the Australia’s Future Tax System Review. 4 Available at http://www.treasury.gov.au/igr. 5 Rothman, G.P.(2012). 6 Australian Prudential Regulation Authority (2007). 2 2. DYNAMIC MICROSIMULATION MODELS Dynamic microsimulation models are a key tool used by Governments and academics to analyse issues relating to distributional and intergenerational equity. They are used for analysing retirement incomes in a broad range of countries 7 including the UK 8 , 9 and the US. 10 A microsimulation model involves a record by record simulation of policy on a large, representative sample of an overall population. This can be contrasted with macroeconomic models, or even cohort models, which tend to group individuals into representative clusters. The advantage of microsimulation models is their ability to model outcomes for individuals in a broad variety of different circumstances while allowing the user to aggregate individual records to determine outcomes for sub-groups of the population or the population as a whole. There are two classes of microsimulation models: static and dynamic. The key difference is how they adjust the representative sample to account for changes in the population over time. A static model is based on a snapshot of the population taken at a given point in time with the characteristics of these people remaining constant over the simulated period, although the weight given to individuals within the sample may be adjusted to account for broad demographic trends. Dynamic models, on the other hand, simulate the characteristics of each individual every year based on their characteristics in the previous year. This method, which is used in MARIA, produces a lifepath for each individual and thereby seeks to capture the heterogeneity of Australian lifetimes and how they are expected to change in the future. The dynamic aspect is crucial for modelling Australia’s retirement income system as the characteristics of future retirees are likely to differ from current retirees as the superannuation system matures. There are a whole range of circumstances that can affect the retirement incomes of individuals. These include: their family composition; how much time they spend in the workforce and how much they get paid; how much they save for their retirement; any inherited wealth; the impact of disability or illness and whether they have a partner whose wealth they can draw upon. Simulating detailed distributions of retirement income and assets in the long-term requires predicting all of these factors for each of the individuals in MARIA. This is important when modelling retirement income policy, as its effects can take many years to be realised. In this way MARIA can capture drivers that are cumulative in nature – particularly in terms of contributions and the effect of the Superannuation Guarantee. Dynamic microsimulation models are notoriously resource intensive to build and maintain. 11 With this in mind Treasury has adopted a phased approach to the development of MARIA by building a basic functioning model in the initial phase to use as a base for future enhancements. The model will be refined and improved through further analysis, use and insight, and with newly available data and modelling techniques. 7 Li, J and O’Donoghue, C. (2012); ‘A methodological survey of dynamic microsimulation models’; UNU-MERIT Working Paper Series. 8 See Department of Work and Pensions (2012). 9 Emmerson, C. Reed, H. and Shephard, A. (2004). 10 Congressional Budget Office (2009). 11 Zaidi A, Rake K (2001). 3 3. MARIA MODEL DETAILS OVERVIEW MARIA begins with base data which captures the Australian population aged 25 and over at a set point in time. The model is run on a representative sample of this complete data set. MARIA then uses Treasury analysis and projections – the ‘input parameters’ – to model the lifepaths (including employment status, superannuation contributions and balance) of these individual records for one year. The output from the model becomes the input for the following year, and so on, year by year, as individuals age from working life to retirement and death. Each year new records are introduced to represent new 25 year olds in the population. Typically the model comprises 40 annual loops keeping track of each individual throughout his or her lifetime. The model structure is summarised in Figure 1. Figure 1: MARIA model structure THE BASE DATA The base data is constructed using the 2013-14 confidentialised full administrative datasets from personal income tax returns, superannuation member contribution statements, and the Department of Social Services and Department of Veterans Affairs pensions systems. Adjustments are made to the administrative datasets to remove overlaps and exclude non-residents. To ensure that the base data aligns with demographic benchmarks, the administrative datasets are augmented with records from the Australian Bureau of Statistics Survey of Income and Housing 12 to represent any individuals not captured in the administrative datasets. 12 Survey of Income and Housing, Australian Bureau of Statistics; Cat. no. 6553.0. 4 MARIA requires demographic characteristics for each individual such as age, education level, amount of work experience, whether the individual is in a couple and whether the individual has children. However, administrative data generally lacks some of this information. For instance, there may be little or no information on whether an individual works full- or part-time, or the number and age of their children. To fill out this picture, the base data estimates these characteristics using information from the Survey of Income and Housing and from the Household, Income and Labour Dynamics in Australia Survey (HILDA 13 ). Figure 2 shows groups of people typically captured by each of the base data sources. The result is a unique, representative snapshot of the Australian population aged over 25 at a given point in time. Figure 2: The MARIA base data OUTCOMES MODELLED A number of steps are performed sequentially to generate a picture of each individual for each year. 14 The output then becomes the input for the following year. In each model loop the steps are performed by stand-alone ‘modules’ in the model which operate in sequence. These modules for a single loop are summarised in Figure 3. 13 HILDA is a household based panel study consisting of 15 annual waves. It collects information about economic and subjective well-being, labour market dynamics and family dynamics. This type of longitudinal data, which capture individuals’ histories, is highly desirable for the development of a dynamic microsimulation model. 14 In reality some decisions, such as coupling and having children may occur concurrently, or in a different order than in the model. However, the sequential approach is typical in dynamic microsimulation models, see for instance Levell, Roantree and Shaw (2016) at page 7.