Social Policy in Australia: Recent directions and the use of microsimulation models in the policy reform process

Ann Harding
Australia has a comprehensive income support system for those who cannot find work or cannot be expected to work. This system has evolved over time with evolving values regarding who should be working. Unlike social insurance systems, payments in the Australian system are income- and asset-tested, targeting those most in need. Roughly one-quarter of the Australian population receives income support cash transfers. Cash payments to families with children have expanded over the past ten years, as have "baby bonuses" and increased subsidies for childcare payments. There are also a number of minor cash transfers that often get introduced just before elections, including rent assistance, telephone allowances and pharmaceutical allowances.

Welfare payments as a percentage of GDP have doubled in the last 40 years, but have remained constant over the last decade. Japan and Australia have similar welfare state expenditures as a percentage of GDP. Australia's welfare system sees good redistributional outcomes without spending enormous amounts, primarily due to its targeted welfare system. Australia has a progressive tax schedule and the value of special tax concessions has increased in the last decade.

The welfare state faces the problems of population ageing and high effective tax rates. Although the projected increase in Australia's elderly population is dwarfed by that of Japan, in the next 40 years health and aged care payments by the Australian government are expected to double. Because the pension system is not based on earnings, social security payments are not projected to increase significantly as a percentage of GDP. The gap between income tax revenue and government outlays is projected to increase over the next 40 years, with revenues falling short of outlays by 3.5% of GDP by 2047. This can be expected to lead to higher taxes or reduced social security spending in the future.

Microsimulation models are based on microdata sets. When microdata were first released by the Australian Bureau of Statistics, it revolutionized the quality of information that could be given to politicians regarding the winners and losers of prospective policy options. Microdata are not just aggregated data, but also provide detailed distributional outcomes. Static tax-transfer models, which are widely used across the developed world, look at the morning-after impact of changing taxes or social security.

The Australian government has been using the STINMOD model since 1994. This model allows users to change different elements of the tax and welfare payment schemes, and then generates clear information on the effects of the changes on families, ministries, etc. One of the outputs of the model is the effective marginal tax rate.

Due to the Australian means-tested welfare system, any increase in income leads to increased tax rates and reduced welfare payments. A big issue in Australian politics is the amount of extra earnings that taxpayers will be able to keep after their taxes are increased and benefit payments reduced. The STINMOD model showed that about 1 million Australians, more than 7.1% of the working population, would lose more than half of each extra dollar they earn. An increasing number of families are seeing increases in effective marginal tax rates of over 50% due primarily to increased family payments (such as childcare subsidies). To combat this, work incentives that target working mothers have been implemented.

In the last ten years, health microsimulation models have come into greater use. Many governments now use these models in tax and social security planning. The big push in Australia is to try to extend this technology to the policymakers who work in the Health Ministry.

Dynamic models are structurally very different from the static model (STINMOD) mentioned previously. These models look decades forward and project the future impact of policy changes. The Treasury Intergenerational Report's projection of a coming budget crisis helped to foster the development of dynamic models. The Treasury Department wants to be able to see if the distribution of burdens and benefits between the young and the old is fair given the changes that are now occurring in the tax system. The model, APPSIM, is being built through partnerships by NATSEM with 12 different government agencies that will all share it once it is completed. Dynamic models start with a very large sample, like a 1% sample of the census in Australia's case. Each sample goes through each one of the model's processes, and the result is longitudinal data on how people in the sample will behave. One of the lessons learned from the first attempt at this type of model in the 1990s was that ease of use was very important for the model's success.

Spatial microsimulation modeling is a field in which NATSEM is a world leader. Census microdata for each street or block are not available in Australia, as national microdata do not include information on where surveyed persons live. The challenge is to fuse national survey sample microdata with census data on who lives in each area. Whereas the STINMOD system allows policymakers to look at the effects of tax changes on the entire nation, spatial microsimulation modeling allows policymakers to see the effects of tax changes on residents of a specific geographical area. Information on who lives where has been input into the STINMOD system, which allows policymakers to see not only how many winners and losers a tax change garners, but also where these winners and losers are distributed geographically. These results are made available over the web through NATSEM.

In conclusion, Australia has made enormous progress in providing policymakers with detailed quantitative estimates of the distributional impact of their policy changes. Starting at the national level and next-day impact projections, the system moved on to make multi-decade and local area specific projections. Given the high risks in policymaking, government officials are keen on using this system to highlight what impacts proposed policy changes will have on the citizens of Australia.

Q: Please explain the effects of marginal tax rates on families.

Ann Harding
The previously displayed chart shows what proportion of families faced effective marginal tax rates above 50%. Essentially, the issue is whether a family earning an extra dollar would be able to keep more than half of that dollar after taxes.

Q: Do the results shown in the graph point to an increase in the number of high-income families?

Ann Harding
Most of the families surveyed were not high-income families. Previously, the poor in Australia faced the highest effective tax rates because effective tax includes changes in the amount of welfare and social security payments. As poor families' income increased, their social security and family payments decreased. Changes in the social security system that have occurred over the last ten years have moved the burden of high effective tax rates from poor families to middle-income families. This has been a popular media topic in Australia and was listed as an important issue to be considered as part of the Henry Tax Review.

Q: What is the coverage of the microdata used to construct such simulations? Are the data based on family or individual data? How many elements are embedded in the original data? Please discuss more about the structure of the microdata.

Ann Harding
The STINMOD model uses a very rich microdata set. The Australian Bureau of Statistics runs a national income survey with 14,000 households, and 500-600 pieces of information are collected from each. This includes such things as number of people in the household, their ages, their incomes, the kind of house they live in, how much rent or mortgage they pay, their income tax, their education, etc. The rules of the tax and social security system are then replicated and the results are added together to produce national sample results. Sample size problems occur when the sample is less than 5,000 households. Many government agencies now build models using their administrative data.

Q: The dynamic model was presented without admitting to any possible errors or flaws, almost as if using the model could say in a God-like way what will happen decades into the future. This does not seem possible given the state of the science of economics. What are your beliefs on the usefulness of the dynamic model in the future? What are some possible directions for improving the model in the future?

Ann Harding
Dynamic modeling is very difficult and it is very hard to build the models. We are currently four years into the five-year APPSIM project. There were not large amounts of longitudinal data available that could be used to estimate the probability of things happening to individuals, and this could lead to errors in the estimates. Also, it is not entirely clear how well government officials will be able to change the parameters and interpret the results.

It is possible that only the Treasury Department will end up using this model as it tends to have very sophisticated users of these types of models. Other ministries can struggle to retain a staff that can properly use and interpret the models. NATSEM trains many individuals in the use of the models who go on to work for the government.

It seems that even if politicians cannot get any sort of guide to the likely distributional outcomes of policy changes, they still have to make decisions on policies. Those decisions can be made on the basis of information on what the impact may be, whether or not that information has caveats, or the decision can be made without such information. Either way, the decision will be made.

These dynamic microsimulation models have been used successfully by European governments. For example, Norway was having problems with its pension system and the results from its dynamic microsimulation model were used in public debates. Norway's pension system was made more affordable and sustainable through use of microsimulation models. This modeling is complicated, but the alternatives to using these models are not desirable.

Q: Are there any movements based on these models in Australia to stimulate complicated numerical debates between political parties?

Ann Harding
When I first started doing this kind of modeling, I envisioned a national academic center that would create these sorts of models for all kinds of organizations involved in public policy. The media, for example, can commission NATSEM to change tax schedules and see what the impact will be. I assume that these models have improved the quality of the policy debate as the detailed results are available to policymakers across the spectrum. Lobbying groups could make very good use of these services at very low costs.

These models decrease the likelihood of politicians bringing up absurd or unwise policy options. For example, STINMOD includes a flat tax option which was used in parliament to quickly swat down the proposal to institute a flat tax by some politicians.

Q: A good simulation should be based on good data. Income data is very difficult to get, especially in Japan. The Japanese household survey queried 50,000 households. In this survey, many respondents were reluctant to let even the government know their income due to privacy concerns. For this reason, it is very difficult to collect income distribution data in Japan, thus affecting the ability to ascertain income and pension expenditures as well. How can correct income data be collected?

Ann Harding
Good data is certainly at the heart of a good model. The problems with sample survey data in Australia have recently come to light. While privacy concerns are not as much of an issue in Australia, high-income persons often do not fill out surveys correctly. NATSEM spends a lot of time validating sample survey data against other data sources like social security administrative data, tax administrative data, etc. Adjustments are made to improve the quality of the microdata but it is not always possible to do this perfectly. Results with a 5-6% range of accuracy are accepted for certain revenue and expenditures models, but this cannot be guaranteed for all areas since at times the data is simply not good enough.

Q: What is the major cause of the fertility rate recovery in Australia?

Ann Harding
The baby bonus policy was the central factor. The Treasurer called upon every family to have three children, supported by the generous family support of the Australian government. Demographers dispute this and attribute it to a catch-up effect. They hold that women who delayed having children are now having more children while they are still able to.

Q: In what other fields can microsimulation modeling be applied?

Ann Harding
NATSEM has specialized in the household sector, though the UK and other countries have built corporate tax microsimulation models. Some firms have done industry-specific applications of microsimulation modeling. Any application where separate microdata or separate micro-actors whose behavior is desired for simulation exist is suitable for microsimulation.