RDP 2014-08: The Effect of the Mining Boom on the Australian Economy Appendix A: Description of AUS-M

Full documentation of AUS-M has not yet been published, but details on individual equations and sectors underlying the results in this paper are available on request to Outlook Economics. The macroeconomic properties of AUS-M are very similar to those of the Treasury Macroeconomic (TRYM) model, from which AUS-M was developed. Documentation of the TRYM model is available in Taplin et al (1993) and at http://archive.treasury.gov.au/contentitem.asp?NavId=016&ContentID=238. TRYM was shown to have macroeconomic responses that were similar to the Pagan-Dungey 10-equation VAR model of the Australian economy and the small RBA model (Stone et al 2005). Given that the properties of TRYM have been well documented and scrutinised, it is probably useful to focus on the key points of difference between the two models.

A major focus in the development of AUS-M has been closer integration of GDP(E) and GDP(P), or what input-output analysis refers to as demand and supply. The components of GDP(E) are mapped into industry output and imports using input-output tables.[5] A framework similar to the Almost Ideal Demand System (Deaton and Muellbauer 1980) is used to model industry and import volumes, where each individual supply component depends on: (1) the input-output weighted demand term; (2) relative prices; (3) the change in total demand; and (4) a stochastic time trend to capture effects from changing technology and tastes. Cross price elasticities are jointly estimated. Following Deaton and Muelbauer, the dependent variables are the weights in total supply, cross price elasticities are symmetric, and income and substitution effects on weights sum to zero.

Other key elements of AUS-M in addition to those in TRYM include:

  • Household consumption is disaggregated. Like the demand system for imports and industry outputs, this is based on an Almost Ideal Demand System style framework where each component is estimated on the basis of relative prices, the change in total consumption (to pick up cyclical and income effects), and stochastic trends (to capture changing tastes and technology over time).
  • Expenditure deflators are estimated in a system which maps production deflators to the demand-side expenditure items. Each demand-side deflator has an input-output weighted production equivalent. Indirect taxes are also allocated using the input-output coefficients leading to implicit tax variables for each demand component. Expenditure deflators respond directly to supply prices, indirect taxes and the GST.
  • Production functions for each industry are estimated indirectly by jointly estimating equations for employment, investment and output prices. (The long run of each equation depends on similar first-order conditions and hence the equations contain common parameters.) Stochastic trends capture changes in underlying labour and capital productivity. Each industry has a derived measure of potential output and its capacity utilisation can be compared with measures from the business surveys.
  • Trade equations – for import and export prices and export volumes – are disaggregated.
  • Equations for inventories are disaggregated.
  • The labour market framework includes unfilled job vacancies and a detailed cohort model that forecasts the duration structure of unemployment. Employment equations are estimated on an industry basis, and employment demand for each industry is adjusted for unfilled vacancies and average hours worked in each industry.
  • The dwelling sector includes an equation for rental vacancies, reflecting imbalance between demand and short-term supply. Lower vacancies lead to higher rents, a higher dwelling Q ratio, and a higher level of investment. Higher rents also lead to a substitution away from the consumption of dwelling services.
  • Explicit equations have been developed for non-dwelling construction investment in finance and insurance, property and business services, and consumer services to capture some of the movement in commercial property investment.
  • Government budget identities are more detailed, reflecting the greater detail available in the model to establish various tax bases and that the model has a complete representation on the income side adding up to GDP(I).
  • Extensive use is made of systems estimation. In one sense the model can be viewed as a kind of Sudoku, utilising constraints across a number of dimensions to reconcile estimates and fill in a complete picture of how the economy works. Any information on expenditure components has implications for production and income components, and vice versa, both in volumes and in values. The model forecasts converge on a steady state growth path that essentially stems from a small scale CGE growth model.


This table is known as the Primary Input Content of Final Demand. If I is the identity matrix; A the matrix of industry to industry interactions; and D the matrix of industry to final demand coefficients, then the table is given by [IA]–1D (i.e. the Leontief inverse times the final demand component of the supply-use tables). As the ABS national accounts produce chain-linked, constant price, time series estimates of industry value added we are mainly interested in the allocation of value added to final demands (and vice versa), thus avoiding much of the detail of a standard CGE model. To build an econometric time series model consistent with the historical data we need to focus on the available time series data. There are no available time series data for the thousands of industry to industry interactions. Hence, we solve out for the industry to industry interactions and model them as a reduced form by forming a demand system for individual industry output, which depends on relative prices and components of demand. That is, rather than impose a set of industry to industry substitution effects by calibrating a detailed set of equations based on imputed data series, we let those effects be determined by the responses to relative prices evident in the historical time series data. [5]