National Economic Review
National Institute of Economic and Industry Research
No. 66 September 2011
The National Economic Review is published four times each year under the auspices of the Institute’s Academic Board.
The Review contains articles on economic and social issues relevant to Australia. While the Institute endeavours to provide reliable forecasts and believes material published in the Review is accurate it will not be liable for any claim by any party acting on such information.
Editor: Kylie Moreland
© National Institute of Economic and Industry
This journal is subject to copyright. Apart from such purposes as study, research, criticism or review as provided by the Copyright Act no part may be reproduced without the consent in writing of the relevant Institute.
An overview of the national, state and regional modelling system
Peter Brain, Executive Director, NIEIR
Ian Manning, Deputy Executive Director, NIEIR
The present paper provides an overview of NIEIR’s national, state and regional modelling system. NIEIR’s forecasting methodology provides a strong and realistic basis for policy evaluation. An economic projection incorporating a policy change is compared with an otherwise similar ‘base case’ projection without the policy change.
Although using general equilibrium models is exceedingly fashionable in policy analysis, based as they are on a fundamental assumption that economies can usefully be divided into autonomous markets and analysed in terms of price-mediated balances of demand and supply in each market, NIEIR’s models are significantly closer to reality. They do not assume away mathematically inconvenient aspects of the economy and, hence, are less likely to deliver counter-productive advice.
The National Institute of Economic and Industry Research (NIEIR) originally entered the field of economic modelling as a forecaster. It maintains this role, preparing regular forecasts and checking them against actual forecast realisation, a process that results in learning from experience. However, NIEIR’s forecasting methodology provides a strong and realistic basis for policy evaluation. The concept is simple: an economic projection incorporating a policy change is compared with an otherwise similar ‘base case’ projection without the policy change.
After more than 25 years experience in economic forecasting and analysis, NIEIR has confirmed the value of dealing always in time sequences. This allows not only for the modelling of causation involving driver and driven variables but for the insertion of response lags and for the inclusion of lagged feedbacks. This time-driven structure of causation means that considerable complexity can be handled without major problems in ensuring analytical consistency.
A second benefit of experience is that NIEIR has developed a sense of relevance and used it to identify the drivers that have influenced the major forecast variables over the past six decades and more. These drivers have all been incorporated into the forecasting and analytical models in ways that reflect their perceived causative role. This is not to claim that a new wild card might not emerge (NIEIR continually scans the horizon in case one does) nor is it to claim that influences are constant in direction or strength, but it is to claim that the Institute has incorporated all historically-relevant drivers into its models and, furthermore, has endeavoured to ensure that their influence is determined by the data and not by assumption. Incorporation in the model is not the last word: historical behaviour is never completely replicated, especially the capricious historical behaviour of exchange rates and other variables strongly influenced by speculative financial markets. Again, although econometric relationships can provide evidence of the direction of causation, this evidence is never conclusive and the estimates of the strength of influence are not always stable. However, model specification emphasising lags and feedbacks provides a structure in which the complexities revealed by econometric analysis of historical experience can be formalised and brought into a logical relationship for forecasting purposes.
In the course of model development, NIEIR has learnt the benefit of a major simplifying device: the geographic layering of forecasting models. Some of the prices, flows and balance sheet values relevant to Australian forecasts are determined primarily on world markets, some are determined primarily at the all-Australia level, some at the level of large city-regions (which approximate to states in Australia) and some at the regional level. The Institute has thus evolved a tiered structure of models: the world is represented formally by the LINK models, to which NIEIR adds its own scenarios of world economic growth; the primary model is the National IMP model (from NIEIR’s IMP modelling suite), which is of particular importance in determining the values of variables influenced by imports, exports and the balance of payments and variables influenced by Commonwealth policy: broadly, the variables emphasised in the National Accounts.
It is also important as a means of ensuring that all-Australia markets add up; the state models include their own range of National Accounts variables and have their own city–region dynamics, but are individually constrained to national values for variables such as the exchange rate and inflation rate, and (subject to feedbacks) are constrained to national totals for a wide range of macroeconomic variables. Within these constraints there is scope for divergence from national trends, some brought about by differences in demography or by differences in industry mix, some by policy effects (particularly state government policies) and some by differences between states in the operation of markets, particularly such markets as housing; and the regional models again have their own dynamics, but are even more constrained by state and national values for variables, and state and national totals. When operated in ‘top down’ mode the regional models determine the local consequences of state and national forecasts. However, the modelling system can also be configured so that regional model results feed back to the state and national level.
All models are disaggregated by industry, with an emphasis on inter-industry relationships. Where a particular forecast or policy study emphasises a particular industry, the modelling of that industry is reviewed to ensure that the peculiarities of the industry are accurately represented. This can apply to industry modelling at national, state or regional levels.
The Institute originally developed two sets of models: annual models based on detailed annual data and projecting in 1-year increments, and stripped-down quarterly models. However, the modelling system has recently been rebuilt on a quarterly basis, this being the minimum time interval used in the National Accounts. Although this creates problems due to seasonality, it has major advantages in the treatment of causation.
The National (IMP) model
For operational and conceptual convenience, NIEIR’s integrated system of forecasting models is divided into modules. The most convenient point of entry to the system as a whole is the national model, because this model is most readily explained in relation to academic economics. It is also important to understand the national model because it determines many of the drivers that operate at the more detailed levels, and also guarantees the coherence of results at those levels.
The main data source at the macroeconomic level is the Australian Bureau of Statistics (ABS) System of National Accounts. The National Accounts comprise three main segments:
estimates of national income, expenditure and production;
financial or flow-of-funds accounts; and
the national balance sheet.
Although there is a tendency to regard the first of these as the most important, the other two provide information that is essential to forecasting growth in national income, expenditure and production. In particular, the national balance sheet includes important information on assets and liabilities.
The National Accounts are of fundamental importance for economic forecasting, for several reasons. They provide the following:
a guide for average or typical experience – if aggregate income is rising, individual incomes will also rise on average; consistency checks not only (by definition) within the National Accounts themselves, but checks useful in more detailed analysis, often expressed as column and row totals;
driver variables for more detailed analysis; a variable set within which a number of important dependant variables can be determined, particularly such variables as GDP, inflation, the exchange rate and the unemployment rate (these variables are also the subject of multiple feedbacks from the variables they drive: e.g. GDP drives energy use but any resulting changes in the efficiency of energy use feedback to GDP); and
a set of variables that is very attractive for econometric analysis, because data quality is high and virtually all the variables are the product of highly decentralised decision-making (the major variables affected by centralised decisions are government expenditure and taxation).
A disadvantage of the National Accounts is that they are published after a delay, and are subject to revision for many quarters after publication. This means that projections inevitably take off from a mixture of estimates of varying quality. NIEIR has tackled this problem and emphasises ‘lead indicators’ in its treatment of the latest published observations.
Forecasts of National Accounts variables provide invaluable background to forecasts at the industry and regional level and to policy-oriented analytical projections. NIEIR approaches the task of forecasting the National Accounts with the utmost seriousness. By longstanding practice, National Accounts forecasts have been ‘top down’; that is, the National Accounts variables, which are either aggregates or conceptually broad indices, are forecast in terms of other aggregates and indices, most of which also occur in the National Accounts or are easily related to the Accounts (e.g. national population). When forecasting in top-down mode, more detailed forecasts are largely driven by the national totals and, as a methodological principle, are reconciled to these totals, although not always completely: differences that can be highly significant at the industry and regional level are not always significant nationally, where they might lie within the acceptable range of forecast errors.
Although the top-down approach is standard, it is possible to move in the opposite direction, working from forecasts at the industry and regional level back to the national aggregates. and then down again to a further round of industry and regional detail.
National Accounts were first prepared after the Keynesian revolution and their basic structure continues to support Keynesian analysis. The familiar categories of aggregate demand are documented, including consumption, investment (fixed capital accumulation), government demand and net exports. Therefore, the National Accounts lend themselves to forecasting using the simplest of Keynesian models in which national income and GDP are determined by the sum total of consumption, investment and net export demand. As explained in university classes in elementary macroeconomics, this model is inherently dynamic. The consumption multiplier, which raises GDP following an exogenous increase in (say) investment demand, is usually explained as taking place in a series of steps, each step following one time period after its predecessor. This model is far too simple to yield useful forecasts, but it reveals two important points.
Demand is a very important underlying concept in economics. Marketed goods and services will not be produced unless they can be sold somewhere. Demand limits production.
Although the Keynesian multiplier can be explained as governing the transition from one steady state to another, it does not take much imagination to see it operating in conditions where exogenous shocks are occurring continuously. These do not prevent the multiplier from operating, but do prevent it from ever yielding a steady state.
Crude demand-dominated Keynesian models were common in the early days of National Accounting, in the 1950s and perhaps the 1960s. However, the
Institute’s forecasting model was never in this crude category. From the beginning it recognised the importance of Keynesian microeconomics and also the importance of explicit growth theory.
Keynesian macroeconomics is founded on Keynesian microeconomics, summarised as import parity pricing for trade-exposed goods and services and cost-plus pricing for all others. Where market structure indicates that monopoly or oligopoly pricing are present, these can be handled by varying the cost-plus mark-up.
The microeconomics of import-parity and cost-plus is not standard economics as taught in first year courses. Economic doctrine privileges pricing at the equilibrium of demand (which increases as price falls) and supply (which reduces as price falls). The fundamental reason for teaching this doctrine is its association with the normative defence of competitive markets. This apart, the equilibrium theory of price formation has been variously defended, for example on the grounds that it follows from the logic of optimisation in conditions of diffused economic power, or that it is approximated in at least some markets. The reasons that it is not assumed in NIEIR’s models are as follows.
The demand/supply concept is closely bound up with the concept of perfectly competitive markets. In practice, very few, if any, Australian product and service markets meet the onerous conditions required if competition is to be perfect. Instead, competition is generally restricted to a limited number of firms, each of which has incentives to adopt strategic pricing behaviour. In these circumstances, cost-plus subject to an import-parity maximum provides a reasonably accurate approximation to actual price formation.
A particular case where demand/supply pricing is inadequate is that of increasing returns to scale, which generate downward-sloping supply curves and indeterminate price. This is no small problem, because increasing returns to scale are endemic in manufacturing and possibly in other industries such as retailing. Once again, the import parity/cost plus theory yields determinate prices.
Even if competitive equilibrium provides a reasonably accurate account of price formation in some of the markets of an economy, the existence of cost-plus import-parity pricing in significant sectors is sufficient to generate Keynesian macroeconomic behaviour.
The fundamental reason for not using competitive equilibrium in forecasting models is that equilibrium is timeless and, therefore, unhelpful in a forecasting context.
Although the general import parity/cost plus approach remains, the disaggregation of the Institute’s modelling system by industry has made it possible to vary the approach to price formation by industry. For manufacturing industries, NIEIR’s developed models use the cost-plus approach with the mark-up a function of unit capital costs and export and import prices. Demand in relation to capacity is included as a short-period influence to allow for profit-taking during booms and price-cutting to generate cash flow during recessions.
Although NIEIR avoids the assumption that prices vary to bring markets into equilibrium, it respects the National Accounts identity: aggregate demand must equal aggregate supply. The difference is that this equality is generally unsatisfactory to the economic actors. It is a temporary accommodation rather than a lasting balance of forces.
Growth theory and inflation
In the 1950s, Keynesian macroeconomic theory was developed into a series of growth models (Harrod, Domar, Hicks and Robinson). These models recognised that investment (in the Keynesian sense of gross fixed capital formation) not only adds to current demand but also adds to the capital stock, resulting in increased productivity of labour and increased incomes for both workers and the owners of capital. Because the Institute forecasting model was designed to yield policy analysis over a time horizon of a couple of decades, it included these relationships, by contrast with a great many contemporary Keynesian forecasting models, which are limited to a time horizon of a few years.
This explicit treatment of the capital stock was of great importance in meeting the challenge of the 1970s: the failure of simple Keynesian models to predict the stagflation of those years. As Keynesian economics developed, it was quickly realised that demand was not the only determinant of GDP. There were also constraints on the supply side, and it was possible for ex-ante aggregate demand to exceed the aggregate supply capacity of the economy. Three relationships were posited
quick-working relationship by which demand which could not be satisfied due to limits to productive capacity spilled over into inflation,conventionally known as excess demand inflation;a quick-working relationship by which demand spilled over into imports and, less spectacularly, into reductions in exports (these relationships raised the whole question of the incorporation of international trade into economic analysis); and
a slow-working relationship by which excess demand for goods and services created additional demand for capacity-creating investment (this ‘accelerator’ relationship further increased capacity utilisation and initially worsened inflation, but to the extent that investment demand crowded out consumption it increased the capital stock, raised capacity and eventually dampened inflation).
Further investigation and experience transformed the accelerator into a relationship between investment and business retained surpluses: the greater the surplus, the greater the level of investment. Further investigation also transformed the account of the relationship between capacity and inflation. Current modelling allows for inflation resulting from the following:
cost-push (fundamentally a result of incompatible income claims);
imports (reflecting the net effect of inflation overseas and movements in the exchange rate); and
monetary sources (fundamentally a result of lack of control in the financial sector, public and private).
Capacity was gradually transformed from a near-engineering concept to one much more closely related to levels of activity above which inflation was likely to accelerate, and which itself depended on such variables as workforce skills.
Although long-run growth analysis is conveniently carried out in values adjusted for inflation, it is still important to include the inflation rate in forecasts, partly because it is a policy target (hence, a determinant of RBA behaviour and in some policy eras of Treasury behaviour as well) and partly because of its influence on economic behaviour, for example the behaviour of firms when assessing investment in fixed capital. NIEIR keeps in mind the various types of inflation, and makes an assessment of the strength of each mechanism. In the immediate wake of the global financial crisis the following assessments applied:
excess demand inflation was reasonably under control, but could break out if there was a reduction in the supply of imports;
cost-push inflation was initially defeated by the 1980s Accord and the probability of recurrence was further reduced by the Commonwealth’s moves to weaken the unions and transfer wage bargaining to the enterprise level;
a break-out of imported inflation will accompany any devaluation of the Australian dollar but was not a threat at current exchange rates, given the world outlook; and
monetary inflation was not a threat, given that the banks were more likely to be trimming their balance sheets than expanding them.
However, with the world economy in turmoil nothing should be taken for granted.
The overseas sector
There is a sense in which the overseas sector fits neatly and naturally into the Keynesian variables of the National Accounts. Exports add to demand and imports add to supply. Australian export earnings can be modelled as essentially demand-driven, industry by industry, from projections of world growth. Allowance can also be made for domestic supply constraints. Imports can likewise be modelled, industry by industry, by estimating domestic supply at the world-parity price and calculating imports as domestic demand less domestic supply and exports.
In this context, NIEIR has benefited as the Australian representative of the United Nations LINK project. Under this project, NIEIR annually prepares forecasts of Australian economic activity, including exports and imports by commodity. Along with its colleagues in other countries (most UN members participate), NIEIR submits its forecasts to the LINK secretariat, which reconciles the national forecasts using the requirement that one country’s exports are another country’s imports. The revised estimates are published and contribute to NIEIR’s forecasts.
Turning to the financial components of the balance of payments, earnings on Australian overseas investments can be calculated from the value of these investments and the rate of return, which is influenced by world growth and monetary conditions. Likewise, the earnings of overseas investors in Australia can be calculated from the value of their investments and the rate of return, as influenced (for equity investments) by the profitability of businesses in Australia and (for debt) by the Australian interest rate. As a consequence of Australian net indebtedness to the rest of the world, Australian interest rates are reliably above world rates, a requirement that limits the RBA’s capacity to influence interest rates.
It is agreed by all analysts that imports and net debt servicing have to be paid for and the ultimate source of foreign exchange with which to pay is export revenue. However, imports can also be paid for from capital inflow, known as a deficit on the balance of trade. Capital inflow results in net debt servicing costs and the addition of these to the balance of trade yields the balance of payments. The question for forecasters contemplating the typical Australian balance of payments deficit is how long it can be sustained by continued capital inflow and how far it will blow out. This involves forecasting both overseas willingness to lend to Australia and Australian willingness to borrow on the terms offered by overseas lenders.
Analysing Australian experience up to 1990, NIEIR employed the concept of the balance of payments constraint to growth. When the balance of payments deficit threatened to become excessive, three mechanisms came into play. First, the high interest rates required to attract overseas loans cut into Australian demand, reducing incomes and so reducing imports. Second, when alarmed over the deficit, the Reserve Bank imposed credit squeezes: quantitative controls over borrowing that acted to reduce incomes and imports. If these were not enough, the Treasury would tighten fiscal policy, further reducing incomes and imports. In the era of exchange and interest rate controls, up to the 1980s, the Commonwealth institutions alternated between periods when they used ‘high’ interest rates to support a ‘high’ exchange rate in the hope that low-priced imports would curtail inflation and periods when they used ‘low’ interest rates to support a ‘low’ exchange rate to encourage export and import-competing industries.
At deregulation the Reserve Bank forswore the quantitative regulation of the banks and the Treasury forswore active fiscal policy. The balance of payments constraint seemed to evaporate as the banks demonstrated a hitherto unsuspected capacity to absorb overseas loans, which they on-lent to the household sector. Successive national balance sheets chronicled an increase in bank liabilities to overseas and in household liabilities to the banks. The policy authorities regarded the resulting balance of payments deficit as benign: it was incurred between private parties and imports of low-cost consumers’ goods were welcome because they kept inflation down. The question for economic analysts is how long this pattern of household and bank debt accumulation can last. There was a severe wobble during the global financial crisis and the indications are for a return to balance of payments constrained growth, but when, and with how much of a bump, is one of the current conundrums of forecasting.
When deregulation was being pursued, one of its expected benefits was that market determination of the exchange rate would ensure appropriate pricing of imports and exports and so equilibriate the balance of payments. In the event, since 1990, the AUD/USD exchange rate has fluctuated between parity to AUD2 for each USD without any commensurate relationship to economic fundamentals. The exchange rate matters for forecasting – it affects the AUD values of all entries in the balance of payments and so finds its way into GDP – but has turned out to be very difficult to forecast. This would not have surprised Keynes, who had sufficient experience of financial markets to know their speculative jitteriness. In its forecasts, NIEIR takes into account commodity prices (which seem to influence the exchange rate far more than their significance for the economy) and interest rate differentials.
In neoliberal economics, finance for fixed capital investment is distributed by a cool and rational finance sector. By contrast, in the macroeconomics of Keynes’ General Theory, investment depends largely on animal spirits. NIEIR makes use of the Flow of Funds statistics, which show that there is very little net flow of funds from Australian financial intermediaries to businesses making major investments in fixed capital: funding is generally from internal sources backed up by direct access to international equity markets. In these circumstances, the Taylor rule is generally appropriate: fixed capital accumulation depends on industry retained surpluses with inflationary expectations taken into account through a downward adjustment when the inflation rate rises. This rule has the technical advantage of ease of econometric estimation at the industry level.
It has been argued in economics that forecasts of real capital accumulation should be forward-looking, emphasising the expectations of investors. For many years, NIEIR experimented with the data from surveys of investment expectations but found that realisation rates varied cyclically except for large-scale committed projects. NIEIR continues to use project lists to forecast expenditure for committed construction projects but otherwise argues that recent retained profits are as good a proxy as any for profit expectations. They accordingly exercise a strong influence on both the ability and willingness to invest.
Although based on the National Accounts and macroeconomic theory, NIEIR’s models have been extended from the world of Keynesian aggregates to include inter-industry accounting as pioneered by Leontief. As perceived by Leontief, the industries of any region take inputs and create outputs. The inputs comprise capital, labour and ‘materials’, the outputs of other industries in the region plus imports from other regions. The outputs of each industry are divided between inputs to other industries in the region, exports to other regions and consumption of final products by households in the region. This classification elaborates the Keynesian aggregates. For example, aggregate consumption is the total of industry outputs sold to consumers plus imports sold to consumers, while gross domestic product is the sum across all industries of the cost of capital and labour inputs. For this reason, it fits very neatly into NIEIR’s modelling system.
For reasons of data availability, this scheme is most readily actualised at the national level. Data are required on the values, by industry, of output, labour inputs, capital inputs, inputs from each other industry, inputs from imports, outputs sold to each other industry, outputs sold as exports, outputs sold to consumers and taxes paid less subsidies received. Price series are also required for all inputs and outputs. At the national level, all of these values are either directly estimated by the ABS or can be derived from ABS data. The input–output matrix is a central element. Unfortunately, it is not produced as frequently as the other data but after allowing for this it is possible to develop time series for all the variables required to describe activity in Australian industries as classified by the ABS: over 100 in the input–output table.
A crucial element in the analysis of this plethora of data is the functional form of the relationship between inputs and output. Because there are several inputs, the functional form must be able to deal with the choice of inputs. Assuming standard qualities for each input, this amounts to the rate of substitution of input for input when the ratio of input prices changes. Leontief responded to this problem by letting the data speak for itself. He specified a relationship in which outputs increased with inputs, but inputs could be either substitutes or complements: substitutes when purchases increased when relative price fell; and complements when purchases increased when the relative price of the complementary input fell. Apart from these limited priors, Leontief allowed the data to determine the parameters, including lagged changes. Applying this approach to Australian manufacturing industry data, NIEIR found that the response to an increase in demand is indeed dynamic, with inputs tending to be harder worked initially followed by an adjustment as capacity was increased. The effect of working inputs harder shows up as a short-term increase in productivity, or returns to scale, and the effect of increasing capacity is to remove at least some of these economies of scale. However, even after 5 years of adjustment, there were many industries in which increasing returns to scale persisted. Similarly, industries were identified in which at least some inputs were complementary: most commonly capital and inputs purchased from other industries (‘material’ broadly defined to include services). All of these estimates, including the dynamics, were well suited to incorporation into the model outlined above. Incorporation allowed the drivers of many of the macroeconomic variables (demand for labour, capital accumulation, value of output, imports and exports) to be calculated by aggregation from the industry level, subject to consistency conditions (e.g. total sales of consumption goods must equal total demand for consumption goods as determined by household incomes, wealth and the like).
The generalised Leontief cost function
Underlying the generalised Leontief production function is a cost function that has the desirable property that it can be regarded as a second-order approximation in prices to any arbitrary cost (and, hence, production) function.
The general form of the desired factor input function can be derived from the generalised Leontief production function as follows:
+ bj,n+1 fj(Q) + exogenous variables}
(j,k = 1, 2, …, n)
where fj(Q) and f(Q) are unspecified, monotonically increasing functions in output. For the equation to describe an underlying production technology, the regularity condition summarised above must be satisfied, for which bj,1 … bj,a+1 must be zero or positive for all j.
The generalised Leontief cost function is comprehensive, because it describes all the types of production technology that produce positive outputs. There will be a variable elasticity of substitution between factors j and k if at least one bj,k, j ≠ k, is non-zero. The function also allows for complementarity between factors, which exists if bj,k is negative: if bj,k is positive, the traditional substitutability assumption applies.
Although this approach to the data is straightforward, it requires a certain amount of unlearning by those who have previously encountered input–output tables only in the context of equilibrium theories. The first difference is the introduction of dynamic instead of instantaneous adjustment, the second the unrestricted form of the relationships (hence, for example, increasing returns to scale can occur), while the third, and more subtle, difference is that changes in outputs and inputs are not so strongly driven by prices. Instead, quantity adjustments can occur, in the same way as they occur in the macroeconomic model.
As an example, the energy industries are identified in the national model complete with input–output relationships with other industries and with their own internal relationships by which inputs are transformed into outputs. In the energy industries, inputs of capital are particularly important, reflecting not only the capital-intensive nature of the industries but the importance of technology in governing the transformation of fuels and other energy sources into useful energy. At the broad level, the various transformation technologies are represented by coefficients that reflect the productivity of capital embodied in succeeding vintages of the various technologies employed. The system thus describes the changing sensitivity of costs to the prices of fuel and capital. This summary account of the energy industries is incorporated into the model at the national level; however, much more detailed modelling of the demand for energy is required when preparing forecasts for particular energy industries.
The national model is complemented by state and local models, to which we now turn.
The state and regional models
Australia is geographically a large country, and the growth rates of economic variables generally diverge regionally. A first step in analysing these divergences is to move from the national to the state and territory level.
The ABS estimates and publishes most of the National Accounts data at the state and territory level, so providing the basis for constructing similar models to the NIEIR national model at the state/territory level. The main differences are as follows:
A number of drivers are determined at the national level and applied across the board to the states. These include the exchange rate, financial variables, such as the interest rate, and variables reflecting Commonwealth policy.
Capacity constraints are a little more flexible. For example, a state that is growing faster than the others will be better able to attract skilled labour, allowing its skilled labour supply to grow more rapidly than the national total.
Some of the statistical detail used in the estimation of the national model is not available at the state level, and has to be estimated. This is particularly true of the input–output table, which NIEIR estimates at the state level using a methodology similar to that used by the ABS for the national table, subject to national-level constraints.
At the state level, particular construction and investment projects increase in prominence in relation to the general flow of economic activity. Because they result from individual decisions, these are unsuited to econometric modelling. Instead, NIEIR maintains project lists and decides on the basis of such information as is available when listed projects are likely to be undertaken. Care is taken to avoid double counting.
When preparing general economic forecasts, which to a large degree are driven by world and all-Australia trends, NIEIR runs the national model and then estimates state impacts using the state models. This involves operating the state models in top-down mode, in which state estimates of macroeconomic variables are derived from the national estimates by applying ‘shift-share’ functions. These functions might be simple
(e.g. allocation by state in proportion to a single driver) or sophisticated (multiple drivers, feedbacks or lags). The simplifying assumption underlying this methodology is that national trends are experienced pro-rata by the states, without any interaction between the states. If interaction is expected, it is necessary to go to much more detailed modelling that covers the dynamics of each state and its effects on the other states and territories.
Although state models remain relevant to the assessment of state-level policies, their main role is to provide control totals for the regional modelling system. These align the regional models with National Accounts data, the state being the smallest jurisdiction for which these data are published.
Reflecting the structure of Australian governments, NIEIR generally defines local regions as local government areas (LGAs). However, other definitions are possible: virtually any geographic area can be defined as a region for modelling purposes.
As at the state level, reasonably accurate business-as-usual forecasts can be prepared with relatively little effort using top-down methods, applying shift-share functions to allocate national and state totals to LGAs. Similar methods are appropriate for most policy changes at the Commonwealth level, such as the effect on regional incomes of a change in tax rates. It is even appropriate to use top-down methods in the case of major investment projects that impact particular LGAs, one or a small number, provided there is no direct impact on their neighbours. In this case, the projects can be added to the LGAs concerned, to the state concerned and to the national total. The revised national forecast, excluding the effect on the directly affected LGAs, can then be allocated to the remaining LGAs by shift-share.
This approach is inadequate for the assessment of developments where regions interact with each other.
Interaction can only be described using ‘bottom-up’ modelling, in which each region is modelled in its own right as though it represents a country in a world model. This requires replication of the structure of the national model for each region. The national model then disappears from forecasting, national totals being calculated by summing the regional totals.
Replication of the national model at the regional level requires that each region should have its own household sector, its own industries and its own inter-industry relationships, all joined to other regions by explicit trade links (imports and exports) and explicit financial flows (including transfer incomes and commuter incomes and expenditures). The data requirements for this approach are very large, because for each region the following datasets are needed:
– aggregate household income and expenditure accounts showing income received, income outflows, savings and aggregate housing and consumption expenditures;
– basic aggregate household balance sheets, including property values and debt;
– input–output tables or inter-industry flows for industries operating within the region;
– foreign trade flows showing how each industry in a given region allocates exports to overseas markets and purchases imports from overseas;
– inter-regional trade flows showing how each industry in a given region sells goods and services to, and buys them from, industries in each other region in Australia;
– income flows showing how incomes flow between regions due to commuting, property incomes, government benefits and government-financed services; and
– expenditure flows showing how expenditures flow between regions due to taxes, superannuation contributions and out-of-region shopping.
The main problems in estimating models at the LGA level are due to data availability. Therefore, we name the major sources:
at 5-yearly intervals the Census provides detailed information on household demography, incomes, occupations, industry of employment and even basic information on asset ownership and indebtedness;
the Census also provides detailed information on the location of employment by industry and occupation. This is derived from the Census
‘journey to work’ question, and requires manipulation before it can be reconciled with Census data on employment by place of residence;
the Taxation Office provides detailed information on taxpayer characteristics by postcode, which NIEIR converts to LGA using a concordance produced by the ABS;
Centrelink likewise provides postcode data on the take-up of pensions and benefits;
at a variety of time intervals (generally getting longer) the ABS has conducted censuses of tourist accommodation, retail activity, manufacturing, mining and agriculture. For many years the ABS conducted a very basic census of businesses, known as the business register. This has been partially replaced by data from Dun and Bradstreet; sample surveys rarely yield valid data at LGA level, the partial exception being the labour force surveys of employment and unemployment produced by the Commonwealth Department of Education Employment and Workforce Relations. However, data from a variety of ABS surveys have been incorporated into NIEIR’s regional modelling; building approvals data are a source; the Real Estate and Stock Institute provides data on dwelling sales and values; and various other sources, mostly administrative data from state and local governments, are relevant from time to time.
Many of these data are costly and their use is limited by agreements to safeguard privacy and commercial confidentiality.
The model estimated for each LGA is structurally similar to those estimated at the national and state levels. However, the small size of LGAs results in a number of differences:
projects undertaken and decisions made by large employers (e.g. plant closure, plant upgrading) can completely dominate local economies and leave them open to idiosyncratic business and investment decisions. Where this is known to be the case, data on the particular project or employer decision is substituted for model-based forecasts; and LGAs, being small, are particularly open economies. Typically, a large proportion of total output is exported (to other LGAs, to overseas) while a large proportion of total supply is imported.
The lack of self-containment of LGAs also expresses itself in the flow of incomes from outside the LGA. These include commuter incomes (earnings of residents who work outside the LGA), private asset incomes and government pensions and benefits. LGA residents also contribute to taxation, and receive health, education and other publicly-financed services. These, in turn, generate employment, which is located at the discretion of governments. Although fixed capital capacity constraints apply strongly at the LGA level, labour can be imported readily, subject only to national constraints. However, labour imports may require incentive payments, particularly if the regional housing market is tight.
The household sector in the National Accounts includes households in their domestic activities plus family businesses and not-for-profit organisations. The incomes of the household sector are estimated from a combination of sources, including the Census, the Taxation Office and Centrelink, plus Institute calculations for the activities of unincorporated business. Consumption expenditures are estimated through microsimulation by matching the Household Expenditure Survey with the characteristics of local households. Consumption expenditures are initially classified by consumption item as in the Expenditure Survey, but these are translated into industry outputs (including imports from overseas by industry). Balance sheets are estimated by microsimulation from the balance sheet portion of the Household Expenditure Survey coupled with the limited Census data on assets and debts and data from other sources on dwelling prices and household debt.
Estimates of the quantum of agricultural output are available by LGA. For modelling purposes, past agricultural production is normalised to standard weather patterns, after which the quantum can be converted to a value by application of price indices (which themselves might require normalisation for weather). For other industries the value of output is estimated primarily from employment data by industry multiplied by regional labour productivity differentials based on postcode income tax data. The estimates for knowledge-based industries are further modified to take into account the productivity effects of regional industry clusters.
A separate input–output table is estimated for each LGA by matching industry input requirements to industry outputs in the same LGA, given the total outputs of each and the patterns revealed in the national input–output table, which, incidentally, restricts disaggregation to 106 industries.
The first step in the estimation of trade flows is the construction of household accounts for each region. On the income side, regional household income is known reasonably accurately from the Census, taxation and social security data. Microsimulation models are used in conjunction with information about house prices, rents, mortgages and survey data to estimate total financial assets, financial liabilities, savings and consumption expenditure of households resident in each region. Microsimulation modelling involves matching survey data at unit record level, principally the ABS Household Expenditure Survey, to Census and regional activity data (such as retail sales) to estimate household consumption expenditure by region. The estimates are highly disaggregated. Expenditures are constrained so that the sum of expenditures by commodity equals the total regional household expenditure estimate. This process ensures that income and socio-demographic factors are reflected in the estimates of regional expenditure patterns.
Households do not necessarily spend their incomes in their LGA of residence. Expenditures are accordingly allocated to local and nearby retailers by a gravity model. Similarly, households do not necessarily earn their incomes in their LGA of residence. The Census ‘journey to work’ question allows accurate allocation of work incomes received in each LGA to the LGAs in which they were generated.
The foundation for production estimates is the Census estimate of four-digit Australian and New Zealand Standard Industrial Classification employment in each LGA. Given the employment base, the value of production can be estimated by multiplying employment for each industry by regional productivity differentials based on postcode income tax data. Farm income is also checked from agricultural output data. The estimates are checked for consistency with state-wide industry output data and the National Accounts state-level estimates.
Aggregate demand in each region totals net consumption (after allowance for sales to residents of nearby regions balanced against cross-border purchases by residents of the region), government expenditure, tourist expenditure (estimated from employment structure), investment expenditure and industry demand. Investment expenditure by households is mainly on housing and is modelled with reference to household formation, the supply of dwellings and household balance sheets (which document the capacity to borrow). Business investment covers both construction and equipment and is modelled (as in the national model) on the basis of business cash flow. Industry demand comprises investment demand and the demand for business inputs, which are calculated from regional input–output relationships.
At the national level, the ABS publishes input–output tables that represent the flow of goods and services between industries. This information for the Australian economy as a whole can be adapted for regional use by taking four steps.
A national indirect allocation of imports table is prepared, showing the overseas import content of supply in each industry, the destination of supply (either inputs into various industries or final demands) and the mark-up between import costs and the prices charged to purchasers.
The information already described on industry output and consumption expenditure spent in the region is gathered. From the national input–output table, the region’s input requirements by industry are estimated given its industry outputs and consumption. How much of this input requirement is likely to be sourced locally is determined. This requires not only that local supply be available but that it be competitive with outside suppliers. The indicator used to assess likely local competitiveness is the import share in national supply as estimated in step 1.
For each industry, an increase in sales to other regions (exports) will yield increases in demand for the outputs of local industries, either directly as purchases of inputs or indirectly through the generation of household incomes which are spent locally. Dividing the resulting increase in local production by the increase in exports yields an estimate of the Type 1 multiplier: the increase in local output as a result of increased local sales, all other factors held constant. In further analysis there might be feedbacks: for example, increases in local wage rates that cause wage-sensitive local production to be curtailed, thus offsetting the initial stimulus. Whether or not this or other offsets occur depends strongly on local circumstances.
In addition to this basic analysis, regional input–output estimates can be strengthened by the addition of data, including details on employment, incomes and the extent to which profits generated in the region are retained within the region.
For each industry, data on overseas exports and imports is available by port and by state of origin/destination in both dollar value and tonnes. Given these constraints, a cost minimisation algorithm is used to allocate international exports and imports by port to the industries of each region. This assignment is iterated until a consistent balance is achieved across all regions and ports. Once international imports and exports by industry have been allocated by region, inter-regional exports and imports can be estimated as a residual. This is done using a gravity model. The gravity factors in the model are adjusted for variations in the substitutability of the items included in the output of each industry in a given location. The lower the substitutability, the greater will be the tendency of production in a given location to sell to the national market. Over time, increasing specialisation in production will tend to lower the degree of substitutability between plants in the same industry in different locations. The substitutability factors for each industry were estimated on the basis of differentials in net interstate imports by industry and by state.
A similar gravity model approach is taken with services, on the grounds that many services involve physical travel, which causes friction of distance, as with freight.
Although the basic unit of calculation is monetary values, trade flows in industries with physical outputs (as distinct from services) can be converted to tonnes using estimates of $/tonne. These can be compared to data on truck movements and reviewed if necessary.
Regional forecasting and analysis using the integrated regional model structure
Forecasts are prepared at the LGA level for the major economic indicators: population (including migration as a result of economic incentives), business value-added (or gross regional product, by industry), employment, income (by source) and consumption (by good or service purchased). The main factors driving the forecasts in each LGA include the following:
the dynamics internal to the LGA, including local demography, local holdings of wealth and debt, dwelling prices, local productive capacities of both capital and labour and local accumulation of capital and skills;
the effect of specific local changes, such as investment projects where the decision lies outside the forecasting model proper;
the local effect of changes in other LGAs through inter-LGA trade and income transfer mechanisms (these include the local effect of changes driven at the national, and occasionally state, levels, and these drivers are applied in each LGA-level model); and
in practice, and depending on judgement concerning the closeness of LGA relationships, changes in peripheral LGAs might be estimated by calculating a national total then cascading down from the national and state models through the effect of drivers determined in these models (e.g. interest rates and prices) and through the pro-rating process.
The regional modelling system is updated annually (with a special update after every Census). The update involves calculation of variables, such as regional value-added, which are not otherwise published. NIEIR groups Australia’s LGAs into 67 regions, and values for these regions along with short-term forecasts for each region are published annually in the ‘State of the Regions’ report for the Australian Local Government Association.
Modelling in practice
The NIEIR modelling system comprises a family of interacting, mutually-compatible econometric models adapted for both forecasting and analysis. Forecasting and analytical tasks are carried out using appropriate subsets of the family of models.
Because they are inherently dynamic, the models by their nature generate forecasts. These forecasts are driven partly by relationships internal to the models, but also by factors treated as exogenous, of which the most significant are world trade and finance. Exogenous variables become less and less reliable as they recede into the future, and so do endogenous relationships embedded in the models; therefore, beyond a decade or so, projections should not be regarded as forecasts, but rather as exploratory scenarios: business as usual perhaps, but not really business as expected.
Because the models cover all industries and all parts of the country, they can be used to prepare detailed forecasts for quite specific variables. A major area of forecasting expertise concerns energy demand, where the usual economic drivers have been supplemented by meteorological probability functions to predict peak electricity demand.
The models have also turned out to be powerful for policy analysis, using the simple methodology of dual projection: a business-as-usual or base case compared with a policy case. Welfare judgements can be made by a variety of variables, such as the effect on GDP, the effect on disposable income, the effect on sustainable consumption and the effect on the distribution of disposable income. Policies modelled can involve changes to prices, changes to technology, particular project investments variously financed, and changes to taxes, regulations and expenditures at all three levels of government. There is no difficulty in accommodating differences in timing.
Despite this general usefulness, when analysing particular proposals it has usually proved desirable to review and if necessary reconstruct the parts of the model(s) directly relevant to the proposal. This can include detailed attention to the local economy of an LGA, detailed review of the economics of an industry or perhaps detailed work on skills or finance. In these studies the general modelling provides background to the sector or region examined in detail. Lessons learnt at these detailed levels are fed back into the general modelling.
In the construction of forecasting models, the general methodology used by NIEIR has no serious competitors. However, in policy analysis it has been fashionable to resort to general equilibrium models, which claim to cover the whole ground of relationships relevant to economic policy assessment but in practice do so largely by assumption. General equilibrium models are exceedingly abstract, based as they are on a fundamental assumption that economies can usefully be divided into autonomous markets and analysed in terms of price-mediated balances of demand and supply in each market. NIEIR claims that its models are significantly closer to reality. They do not assume away mathematically inconvenient aspects of the economy and, hence, are less likely to deliver counter-productive advice.