The Scottish Government

« Previous | Contents | Next »

Listen

1. Introduction to the CGE modelling approach

Policy actions to reduce the generation of greenhouse gas ( GHG) emissions may have impact that permeate throughout an economy, leading to a series of adjustments in the production and consumption of different goods and services. These adjustments cannot be adequately captured within a partial equilibrium framework but may be explored through the use of Computable General Equilibrium ( CGE) models of the macro-economy. CGE models are widely used in the investigation of energy and climate policy, partly as a consequence of the increasing availability of modelling frameworks and associated benchmark data. In the case of Scotland, an energy-economy-environment CGE modelling framework has been in development at the Fraser of Allander Institute ( FAI), Department of Economics, University of Strathclyde over the last decade. CGE model development for Scotland at the FAI, in the form of the AMOS suite of models ¹, has been greatly enhanced by the availability of comprehensive region-specific economic accounts in input-output ( IO) format for Scotland, the required core database for CGE modelling. However, policy applications of the environmental variant, AMOSENVI, have been limited, due to the absence of an appropriate environmental augmentation to the Scottish IO accounts to provide a region-specific linked environmental component to the model database. The purpose of the current project is to provide illustrative findings from AMOSENVI as an experimental energy-economy-environment CGE model for Scotland. This serves to demonstrate the type of analytical capability that is potentially available given the appropriate commitment to developing the data and modelling infrastructure by the policy and research communities.

This section of the report begins by clarifying the strengths and weaknesses of multi-sector CGE models for investigating the economy-wide impacts of policies to reduce GHG emissions, summarising the theoretical underpinnings and basic methodology of the CGE modelling approach, before outlining the specific AMOSENVI model employed here.

1.1 Basic characteristics, including strengths and weaknesses, of the CGE approach

Computable General Equilibrium ( CGE) modelling involves numerically simulating the general equilibrium structure of an economy, where a general equilibrium is characterised by a set of price and output levels across all sectors of the economy such that market demand equals supply in all markets simultaneously. The technique is an important tool in evaluating the economy-wide impact of exogenous shocks, and has proved to be appropriate for economic policy appraisal. CGE modelling has been employed to examine a whole range of policy and other non-policy disturbances in a variety of research areas, including questions relating to regional trade agreements (see Lloyd and Maclaren (2004) for a review), public finance (Shoven and Whalley, 1984), tax reform (Jorgenson, 1997) and the distributive impacts on different household groups of policy change ( e.g. Bourguignon et al, 1991). Furthermore, it has become the most widely use approach for system-wide analysis of energy-economy-environment issues at both national (Beausejour et al, (1995), Bergman (1990), Bohringer and Loschel (2006), Conrad and Schroder (1991), Goulder (1998), and Lee and Roland-Holst (1997), and Conrad (1999) provides a review) and regional levels ( e.g. Despotakis and Fisher (1988) and Li and Rose (1995)). The key characteristics motivating the use of CGE models to analyse economy-environment problems are their multi-sector nature (sharing the desirable characteristics of input-output models, where sectors with differing resource use and/or pollution generation characteristics can be distinguished) and the simultaneous modelling of prices and quantities, supply and demand (characteristics not shared by input-output). A fuller literature review and overview of the application of CGE modelling techniques to environmental issues/problems is given in Technical Appendix 1 below.

CGE analysis is grounded in economic theory, but can deal with circumstances that are too complex for analytical solutions. As such, CGE analysis can be considered a numerical aid to analytical thought. For example, in the first set of simulations in this project, CGE analysis can simulate the various substitution, income, output and composition effects that may follow from energy efficiency improvements and give rise to what are referred to as 'rebound effects'.

CGE models are generally parameterised to reflect the structural and behavioural characteristics of a particular economy. As a result, they can estimate not only the direction, but also the order of magnitude of effects that may result from a particular exogenous disturbance, such as any one of the scenarios simulated in the present project. A feature of CGE models is that they tend to have a very well developed supply side, allowing investigation of supply-side policies, such as energy efficiency improvements, where other models ( e.g. input-output) are inappropriate. CGE models also make it easier to evaluate the net impacts of a given policy or other disturbance, where the counter-factual is simply the model run without any changes. Since all changes in output, employment and energy use are measured relative to this baseline, the marginal effects of the disturbance simulated are clear. Evaluating the same policy using time series or cross-sectional statistical data would require the counter-factual to be identified by appropriate statistical control, which may be harder, and risks confusing the drivers of changes in activity.

However, CGE models do have a number of well-established weaknesses. For example, most represent production behaviour through the use of 'well-behaved' but relatively restrictive functional forms, with limited facility for testing their appropriateness. Parameter values for these functions may be assigned through calibration to a base year, but this may not be representative an equilibrium in the economy. Alternatively, they may be taken from empirical studies, but these may relate to different countries and/or time periods from that to which the CGE model is applied. While sensitivity tests are feasible, they are not always conducted in practice.

CGE models generally also assume that firms minimise costs, that consumers maximise utility and often that the source and direction of technical change is exogenous. Each is partly inconsistent with empirical evidence. Markets may also be assumed to be competitive and factor inputs may be assumed to be mobile, although neither is a necessary feature of CGE models. While the results of CGE models may sometimes be driven by assumptions that are not readily apparent, a CGE model should not be regarded as a 'black box'. Transparency may be considerably improved by providing information on key features and assumptions, and by the modeller explaining the results with reference to economic theory.

1.2 The AMOSENVI model

Over the last decade the regional modelling team based at the Fraser of Allander Institute ( FAI), Department of Economics, University of Strathclyde, has developed an energy-economy-environment computable general equilibrium modelling framework of the Scottish economy ( AMOSENVI). This framework has mainly been (and continues to be) developed with the support of the ESRC and EPSRC.

AMOSENVI is a 25-sector CGE model of the Scottish economy. The current database is a social accounting matrix ( SAM) that incorporates a 25-sector aggregation of the 1999 Scottish IO tables. The motivation for continuing to use the 1999 IO tables is that these include a limited experimental (not publicly available) disaggregation of the Scottish electricity sector (carried out by the Scottish Government IO team, with input/development by the FAI regional modelling team), which is clearly important for environmental analyses. The 25 sector aggregation of the 1999 IO tables is also augmented with sectoral physical energy use and pollution data, constructed by the FAI modelling team as part of a project carried out in consultation with the Scottish Environmental Accounts Working Group in 2003 (see Turner, 2003).

However, as noted in Section 7 (Conclusions and Recommendations), the current database is limited and would benefit from more environmentally-focussed development of the Scottish Government's Input-Output economic accounting framework. For example, the current (experimental) disaggregation of the Scottish electricity supply sector is limited to generation type and could be improved particularly by separately identifying the different stages of the electricity supply process ( i.e. generation, distribution and supply to consumers). In terms of physical energy-use pollution generation, the current environmental input-output component of the model database is heavily reliant on adapting UK national energy and pollution intensities to the Scottish case, and would benefit from the development of more region-specific environmental input-output data for Scotland.

Details of the AMOSENVI modelling framework can be found in Hanley et al (2008). Key details are that it has three transactor groups, namely households, firms and government; 25 commodities and activities (five of which are energy commodities/supply - coal, gas, oil and electricity from renewable and non-renewable sources); and two exogenous transactors, the rest of the UK ( RUK) and the rest of the world ( ROW). We regard AMOSENVI as a modelling framework as there is a high degree of flexibility in terms of choice of key parameter values and functional forms, assumptions about how the labour market functions and the nature of macroeconomic constraints (government budget constraints etc). Greenhouse gas emissions are modelled as linked to energy-use in the case of CO2 emissions generated through fuel combustion in each production sector and final consumption sector, and otherwise to sectoral outputs/final demand expenditures (see below).

The model can be run for three conceptual time periods, namely the short, medium and long- run, which allows examination of the sectoral and macroeconomic impacts of any disturbance under alternative assumptions about factor supply.

• In the short run, the population and capital stock is fixed.
• In the medium run, population can adjust through migration.
• In the long run, capital stocks can vary through net investment

The model can also be run in period-by-period (year-by-year) mode in order to examine the path of adjustment over time as the labour supply adjusts in response to changes in real wages and investment responds to changes in profitability. This allows us to examine the extent of adjustment, and impacts on key economic and environmental indicators over different timeframes and towards a long-run equilibrium. Simulation results are reported so that the impacts of any disturbance can be examined in isolation and relative to a 'no change' baseline. That is, we do not attempt to forecast the future performance of the Scottish economy; rather we focus on examining the impacts of a given policy scenario in isolation under different assumptions about supply and demand conditions in the economy at the time the shock is introduced. Disturbances/policy scenarios may be introduced as transitory or permanent shocks and they can be introduced gradually or as step-changes.

Modelling pollution generation in AMOSENVI

The simplest way to model pollution as a result of economic activity is through fixed coefficients linking pollution outputs to each sector's output level. This approach was one of the earliest steps in general equilibrium economy-environment modelling, developed in Leontief's (1970) environmental IO framework. Nonetheless, it remains common in both IO and more general CGE modelling e.g. Ferguson et al (2005). However, as explained in more detail in Section 4 of Technical Appendix 1, below, the major limitation of relating emissions to sectoral outputs only is that there is no scope for changes in emissions due to technical substitution within sectors. That is to say, if pollution coefficients are output-based and/or only pure Leontief technology is modelled, then the only way to reduce emissions within any sector is to reduce that sector's output. In discussing this issue, Beghin et al (1995) identify three underlying components of changes in emissions levels over time. The first component is composition: the change in pollution induced by a change in the commodity composition of aggregate production (more or less dirty/clean goods). Secondly, technology relates to evolving cleaner technologies (which usually result in a change in the input mix or input substitution). Finally, scale: the increase/decrease in pollution attributable to an increase in aggregate economic activity

The present AMOSENVI model captures input substitution by relating emissions of CO2 to different types of energy use through input-pollution coefficients. In the absence of appropriate economic-environmental input-output accounts for Scotland, these coefficients are determined using data on the CO2 emissions intensity of different types of fuel use in the UK economy (see Turner, 2003 and Hanley et al, 2008). The application of fuel-use emissions factor data is fairly straightforward in the case of CO2 emissions, as these are primarily dependent on fuel properties rather than combustion conditions and/or technology. Modelling input-pollution relationships becomes more complex when it comes to non-CO2 emissions. This is because non-CO2 emissions tend to be dependent not only on fuel type, but also combustion conditions and technology, meaning that appropriate emissions factors are likely to be more difficult to identify and numerous for models with a high level of sectoral detail. Thus, at this time we do not attempt to extend the input-pollution approach to any other pollutants. In the environmental CGE literature, models that adopt an input-pollution approach have indeed tended to focus solely or primarily on CO2 emissions (see Turner, 2002, for a review).

We also include an output-pollution component for the generation of CO ₂ emissions (see Hanley et al, 2008). This is following the argument put by Beauséjour et al (1994, 1995) that there is a role for modelling both input-pollution relationships, and output-pollution relationships where emissions not only result from input use but also from processes that are inherently polluting. Beauséjour et al (1994, 1995) identify processes such as non-ferrous smelting, which generates SO _X, and pulp and paper production, which generates CO2. Here, in the case of CO2 emissions, we identify industrial process emissions relating to the production of mineral products and metal in the 'Mfr metal and non-metal goods' sector. We also apply output-pollution coefficients to capture CO2 emissions that occur during extraction activities in the 'Oil and gas extraction' sector and flaring in the 'Refining and distribution of oil' sector. While these are obviously related to energy supply, they are not easily related to energy input use through the application of emissions factors.

Due to a lack of Scottish-specific data on sectoral pollution data ( i.e. Scottish environmental input-output accounts), the input- and output-pollution coefficients in AMOSENVI are currently mainly based on UK direct emissions intensities for each SIC-classified production sector and for household final consumption (see Ferguson et al, 2005 and Hanley et al, 2008), adjusted to reflect the composition of Scottish output at the aggregate and sectoral levels. A more detailed account is given in Turner (2003), but basically we have taken the following steps to derive the input- and import-pollution coefficients. First we used UK data on physical fuel intensities for the broad (directly polluting) fuel types - oil, gas and coal - to estimate total Scottish fuel uses. These are then distributed across the production and final consumption sectors identified in the model according to the distribution of local and imported purchases of these fuels implied by the Scottish IO tables and the experimental data on commodity imports to estimate sectoral fuel uses. UK data on the level of emissions (tonnes) per unit of each fuel type (tonnes of oil equivalent) are then used to derive estimates of direct CO2 emissions resulting from each production and final consumption sector's use of local and imported coal, gas and oil. Finally, we divide each sector's estimated emissions from each type of fuel use by the IO and (experimental) import-by-commodity data on fuel purchases to derive the input- and import-pollution coefficients for the model (tonnes of CO2 per £1million expenditure on each local and imported fuel respectively).

However, it is important to introduce Scottish-specific data on sectoral emissions where it is possible to do so (Turner, 2006). As explained in Turner (2003), even though region-specific estimates of CO2 and other GHG emissions have been made for 1999 by Salway et al (2001), there are problems in mapping emissions reported for IPCC classified activities to the SIC classification used in economic IO accounting. However, it is possible to map for some activities, most notably electricity production and supply. Moreover, we were able generate the output-pollution coefficients for non-fuel-combustion emissions of CO2 in the 'Mfr metal and non-metal goods', 'Oil and gas extraction' and 'Refining and distribution of oil' sectors, using the estimates of CO2 emissions in 1999 from the relevant sources reported by Salway et al (2001). These are simply divided by the base year outputs for each of these sectors.

In general though, given the limitations of appropriate energy-economy-environment data currently available for Scotland, and the many uncertainties involved in modelling the types of policy that are of interest here, it is important to note that results should be regarded as indicative of the scale and direction of impacts on the Scottish economy of a given policy scenario.

Reporting results from the AMOSENVI model

All simulation results are reported in terms of the percentage change relative to the (no change) base case scenario represented by the 1999 model database. For each simulation we report a range of key economic variables - including GDP, employment, unemployment, exports and imports, wages, household consumption and CPI. We also report results for energy consumption, separately identifying electricity and non-electricity energy types, and for CO2 emissions, and three composite indicator variables: GDP per unit of energy consumed ( GDP divided by total, economy-wide, electricity and non-electricity consumption, respectively, in physical units) and CO2 intensity of production (total CO2 emissions generated from production and consumption divided by GDP). For improved environmental productivity (sustainability), the value of the first two (energy) indicators should rise, while the value of the third should decrease.

Table 1.1 Sectoral breakdown of the 1999 AMOSENVI model

We also report results for the individual sectors identified in the model (see Table 1.1), giving particular attention both to the main sectors particularly affected by different shocks, and to particular sectors of interest, i.e. the Key Sector identified in the current Government Economic Strategy. These are: Creative Industries (which we have we mapped to our Sectors 17 and 18, 'Communications, Finance and Business' and 'R&D' respectively, through the input-output classifications used in the model database); the energy supply sectors (Sectors 21-25); Financial and Business Services (also mapping to our Sector 17); Food and Drink (covering our Sector 1, 'Agriculture' and the two fishing sectors, Sectors 1 and 3, as well as Sector 7, 'Mfr Food, Drink and Tobacco'). We also focus on 'Distribution' (Sector 15), as the one with the highest share of output serving tourist expenditure (largely because of the inclusion of hotels etc), and 'Education' and 'Public and Other Services' (Sectors 19 and 20) sectors, where more than 50% of output goes to meet public sector demand. In the case of the energy efficiency simulations, we also identify some groups of sectors that are likely to be of particular interest, which we label 'Agriculture and Primary' (Sectors 1-6), 'Manufacturing' (Sectors 7-12), 'Energy Use' (Sectors 1-20), 'Energy Supply' (Sectors 21-25).

« Previous | Contents | Next »