Independent report

Trade modelling review expert panel: report

Published 31 January 2022

This report was written by Professor Tony Venables, chair of the Modelling Review Expert Panel. It was informed by the views of the rest of the panel.

The report does not represent the views of the department, but provides independent advice which the department will use to guide the development of its analytical tools. The department will respond in due course.

Summary and recommendations

The Department for International Trade (DIT) has rapidly developed its trade analysis capacity and capability over the last 5 years. The objectives of this analytical work are to inform the development of trade policy and the choices between policy alternatives. Analysis is also presented to the public in the form of scoping and impact assessments of the free trade agreements (FTAs) the DIT is negotiating.

The Trade Policy Modelling Review has explored how the DIT meets its objectives of informing policy and the public, and how its capacity and the tool-kits it employs should best be developed in coming years. The DIT’s current analysis is in line with international best practice and is well-regarded. The recommendations of the review panel build on this base, while increasing capability to focus on issues that will be particularly important for the performance of the UK economy in coming years.

Our recommendations involve:

1. Development of the DIT’s core modelling approach (computable general equilibrium) to provide a robust platform in which extensions – to dynamics, regional impacts, and environmental effects – can be implemented in a straightforward way.
2. Complementing this core approach with the ability to ‘zoom-in’ on sectoral detail, particularly in areas in which the UK has comparative advantage and in sectors related to new technologies.
3. Filling evidence gaps and broadening the evidence base used to understand and to model the full range of economic mechanisms through which trade affects the economy.
4. The presentation and communication of results in as transparent and clear form as is possible.

1. Computable general equilibrium (CGE)

The DIT should maintain and develop its CGE modelling to ensure comprehensiveness of coverage and the internal consistency of results obtained. This should continue to be based on the frontier techniques used by the international community of trade policy modelers which supports scientific excellence, access to data, and model representation of some 140 country and region economies.

We recommend a focus on ensuring that the model is robust and explainable, which may involve some simplification of the core model. Building on a robust core model, we recommend extensions to capture regional impacts of trade, the dynamics of adjustment, and environmental impacts. These additions will require working with the international modelling community and with other UK government departments to develop data sources and modelling techniques that enable the wider impacts of trade policy to be studied, and that incorporate relevant policy instruments.

2. Complementary Modelling

The DIT should further develop its capacity to model effects that go beyond those captured in the core CGE model. This is best done with ‘bespoke’ modelling based on detailed understanding of particular sectors.

Emphasis should be on:

(a) sectors that are most likely to be affected by changes in trading relationships

(b) sectors where the supply response to new opportunities is complex, such as in tradable services and new technology sectors

(c) the impact of agreements on value chain participation

This will require further development of ‘partial equilibrium’ (PE) modelling, drawing on a variety of data sources and analytical methods. Such models may in some cases be formally incorporated as ‘plug-ins’ to CGE, or may be stand alone, providing a complementary view of possible outcomes. Some of these developments will be best done in collaboration with other government departments.

3.Evidence

The DIT should further develop the evidence base it uses in modelling.

There are multiple sources and types of evidence, including:

• econometric estimates of key economic relationships (elasticities) and trade costs (including non-tariff barriers)
• case studies of historical changes in trade relationships
• ‘bottom-up’ information from sources familiar with the detail of particular sectors or policies

CGE modelling relies almost exclusively on the first of these, and capacity should be developed to draw more heavily on the second and third, particularly in the context of sectoral analysis. Data gaps need to be filled in several areas, particularly service sectors and estimates of non-tariff border frictions.

4. Presentation of results

DIT should build trust in its results by presenting them in a manner that enables ‘reality checking’ and that provides insight into their derivation.

This will generally require providing a range of results for different sectors, time-scales, and scenarios, making clear the assumptions and mechanisms that generate differences between them. It requires presentation of the results in a manner that makes clear the key factors which drive them.

Background and remit

In August 2020 the Secretary of State for International Trade commissioned a review of DIT’s modelling capability by independent external experts who would advise on the future development of this capability.

The Modelling Review Expert Panel (MREP) was established with the remit to:

• advise on the development of modelling approaches and strategies to support the trade policies of the department, based on a broad base of academic expertise
• highlight methodologies to assess the breadth of gains that are expected to be achieved from FTAs and the degree to which they can be captured in trade modelling frameworks
• advise on cutting edge studies and research and how to apply it so that DIT’s modelling reflects the forefront of research excellence
• provide advice on the development of specific modelling tools, econometric work and the development of existing models
• offer a forum to exchange ideas on current and future research trends pertinent to international trade and economic modelling
• offer advice to the DIT Chief Economist and Director of Analysis on potential research agenda

The terms of reference and the membership of the MREP are given in Annexes 1 and 2.

During the review the MREP has considered a wide range of modelling-related topics including;

• trade model specifications
• sectoral responses, including service sectors
• alternative modelling approaches
• regional modelling
• labour market modelling
• dynamic modelling
• sector-specific (partial equilibrium) modelling
• sources and uses of evidence
• trade and the environment

Annex 2 presents details of issues raised in these discussions.

The deliberations by the MREP, including the formation of its recommendations, have also benefitted from contributions from a wide range of academic experts (Annex 3.)

Report and recommendations of the Modelling Review Expert Panel

Modelling is a part of the DIT’s analytical tool-kit for assessing the likely impact of changes in trading relationships. This tool-kit and the impact assessments they generate serve 2 main functions. The first is to provide rigorous economic analysis to inform the design of trade policy and the choices that have to be made between alternatives. The second is to inform the public and other decision-takers about the likely effect of policy changes.

Providing these assessments is a complex task. The range of alternatives that are potentially considered at the various stages of trade policy formulation – from initial scoping of possible agreements to their final negotiation – is vast. In addition, the economic issues are complex as changes in trading relationships have multiple effects through different channels, expanding some activities while contracting others, and benefiting some businesses and workers while requiring adjustment and redeployment by others.

There are many criteria that the tool-kit, including modelling, should satisfy.

These include the capability to:

• analyse the range of policy measures that may be implemented, recognising that such measures contain fine detail, generally specific to the economic sector, mode of commerce and partner country affected, and often couched in non-quantitative terms. Change in the world economy means that policy measures may need to be evaluated under a number of different scenarios
• capture a full range of economic mechanisms through which trade and investment impact the national and regional economies. Trade agreements may bring direct cost savings and unlock comparative advantage, enabling beneficial structural change to occur. There may also be productivity effects arising from scale economies created by new market opportunities, by firms’ adjustment to international competition, by foreign direct investment (FDI), and by participation in global value chains. Some of these mechanisms are likely to be particularly important in high-technology sectors, service sectors, in digital trade, and for trade in intellectual property and data
• provide quantitative estimates of the changes that are expected to occur, and place values on these changes. There will be direct impacts on the prices and volumes of imports and exports and on trade and investment opportunities, leading to further change as the economy responds. Estimates of these effects are needed both in aggregate summary form, and also for a wide range of indicators that are salient for policy makers and the public. These include impacts on particular sectors of the economy, on particular regions and socio-economic groups, on the timescale over which adjustment occurs, and on the environment.

These criteria should be met in a manner that is:

• rigorous, ensuring internal consistency and avoiding pitfalls of double-counting some effects or omitting others. Economy-wide resource and budget constraints have to be imposed, requiring recognition of ‘displacement’ effects and the fact that expansion of one activity will draw resources out of others
• evidence based, drawing on as wide a range of relevant sources and experience as possible, and being open to ex post evaluation of results
• transparent, based on mechanisms that are understood and yielding quantitative results that can be subject to ‘reality checks’ by experts and a wider audience
• proportionate and timely, prioritising issues where analytical and modelling input is most valuable

The DIT’s Trade Modelling Unit is based in the Macro Analysis and Modelling Team under the direction of the Chief Economist. The core modelling is based largely on CGE techniques which are generally used to produce a central estimate of the changes in trade, output and economic welfare that are likely to be generated by trade policy. The techniques used are best-practice amongst the class of trade models developed and currently employed by researchers and by many countries and international organisations. This is supplemented by partial equilibrium (PE) and ‘off-model’ analyses undertaken in the Modelling Unit and in other analytical teams in DIT as well as other government departments. Results from each of these parts feed into policy decisions where, according to evidence from within the DIT, modelling results are influential and highly valued.

The existing capacity has a strong base which meets a number of the criteria outlined above.

Our principle recommendations are intended to:

1. develop the core CGE model to provide a robust platform in which recommended extensions – including dynamics, regional impacts, and environmental effects – can be implemented in a straightforward way
2. ensure that a wide range of economic mechanisms are captured, particularly those relevant to sectors in which the UK has comparative advantage, and sectors related to new technologies
3. develop capacity to draw on a wide range of evidence to inform and provide essential inputs to the modelling process
4. present results for a range of outcomes in as transparent of a manner as possible.

We discuss each of these issues in the remainder of this report. Full details of the panel’s discussions and technical details on the topics discussed are given in the annexes. Annex 5 also provide additional comments by one panel member.

1: General equilibrium modelling

CGE modelling scores highly on several of the criteria outlined above. It has explicit modelling of trade barriers, both in the UK and in its multiple trading partners. It is based on standard models of economic behaviour and hence on the adjustment of firms and households to change. It provides quantitative estimates of changes in economic variables, of the value of these changes, and thus of economic costs and benefits. It is rigorous, ensuring that results are internally consistent. Importantly, it imposes the discipline that if some sectors expand following a policy change, others may need to contract to release required labour and other resources. The mechanisms through which these costs and benefits arise are explicit in the model, are based on familiar principles of supply and demand, and can therefore be communicated through a relatively simple narrative.

However, CGE models score less well on other criteria, and are subject to several types of criticism.

First, while capturing many sectors of the economy and many trading partners, they impose a highly stylised structure across sectors. This fails to capture the detail of some of the sectors that may be most impacted by trade policy, in particular service and knowledge intensive sectors, which are particularly important to the UK economy. Missing this detail, they also miss important mechanisms through which trade affects economic performance. For example, changes in FDI, in aspects of service trade, or in the productivity of domestic firms are, in most cases, not well captured.

Second, the approach as currently employed by the DIT does not address all the issues of concern to policy makers who might, for example, be concerned with the time-paths of employment, the regional distribution of effects, or environmental impacts of a trade agreement. The presentation of results, often as a single aggregate point estimate, does not provide a credible narrative about the effect of policy for the broader public, or facilitate understanding of the factors that drive the results.

Third, the CGE approach is not sufficiently grounded in empirical evidence. Inputs to the model do not make sufficient use of empirical observations from studies of historical changes in trading relationships, or of macro-econometric modelling results. Outputs from the model – the predictions made about the effects of trade policy – are rarely validated by comparison with actual outcomes, and where this has been done, the models typically do not perform particularly well.

Nevertheless, our recommendation that the DIT should maintain and develop its CGE capacity is based on the rigorous, economy-wide and multi-country framework it provides, and its potential for future development as the core platform on and around which a fuller picture of the impact of trade can be built. Alternative core modelling approaches were considered by the panel, and are discussed in more detail in Annex 4.1. The remainder of this section addresses the core model and its extensions to include dynamics, regional and environmental effects. Sections 2 and 3 of the report cover the development of complementary modelling approaches to capture granular detail, and wider sources of evidence respectively.

1.1 The core CGE model

The core CGE model has 2 main elements. One is the specification of supply and demand in each sector, and the other the elasticities that summarise key economic relationships.

Three different supply and demand specifications are commonly employed in CGE models, each of them highly stylised.

They are:

1. perfect competition with nationally differentiated product varieties (‘Armington’)
2. monopolistic competition with firm level product differentiation (‘Krugman’)
3. monopolistic competition with firm level product differentiation, plus heterogeneity in the levels of productivity attained by firms within each sector (‘Melitz’). They successively capture somewhat richer effects, while imposing greater data requirements and becoming less transparent

A modelling choice arises as to whether to capture a relatively rich range of effects and sector heterogeneity by either (a) using the more complex specifications in a number of sectors within the core model, or (b) by ‘bespoke’ PE modelling of a few selected sectors, set in a relatively ‘simple’ core CGE specification. The panel’s recommendation is that there should be increased emphasis on (b), using approaches outlined in section 2 below.

In view of this, the panel advise that DIT may benefit from a simpler, robust, and more understandable core CGE model. Priority should be given to improving the underlying parameters and data, and providing a robust platform for extensions to dynamic, regional, and other issues.

Specifically:

• prioritise improvements in the robustness and application of a relatively simple core specification (such as Armington), including:
  • improved estimates of parameters that capture key economic relationships
  • improved estimates of other inputs that feed into the model, such as measures of non-tariff barriers
  • improved understanding of model sensitivity to uncertain estimates of key variables such as non-tariff barriers
• develop methods to improve the capacity of the core model to act as a platform for further experiments, including:
  • development of ways of running at different levels of sectoral aggregation
  • development of ways of integrating PE data in CGE via sectoral ‘plug-ins,’ ensuring consistency where possible
  • enabling the model to incorporate dynamic and regional extensions

1.2 Dynamics

DIT’s current CGE model, in keeping with many CGE models, employs a ‘comparative static’ framework, meaning that results are presented as long-term changes brought about by a change in policy, and expressed relative to a ‘base-line’ description of the economy.

In practice, policy implementation is likely to be staged through time, and different economic variables adjust at different speeds. For example, exchange rates move in expectation of changes in policy, while asset depreciation, investment in capital stocks and changes in management practices are slow processes. Redeployment of workers across sectors may be slow and far from complete, with critical implications for affected workers. Better understanding of the dynamic path of future change, both in the short- and medium-run, is an important demand of policy-makers and the public.

There are different technical approaches to dynamic economic modelling ranging from a fully specified multi-period model with forward looking decision-takers, to a model in which the ‘endpoint’ is essentially that which comes out of the comparative static modelling (as discussed in Annex 4.3.1), while the dynamics specify the short- and medium-run adjustment path between the baseline and this long-run outcome.

The former approach encounters complex computing issues as well as the risk that the results obtained are hard to understand and may not be credible. We therefore recommend the latter approach, focusing on the short- and medium-run adjustment process. This will enable attention to be focused on the frictions that slow down adjustment, particularly in the labour market, and will facilitate identification of possible complementary policy measures.

Recognising the importance of introducing a time element to its modelling, we recommend that the DIT develops a dynamic extension of its core CGE model, including:

• develop short- and long-run dynamic capabilities of the core CGE model, using these features to provide better descriptions of the time path of change, including the phasing in of agreements and the effect of FTAs being signed sequentially
• introduce short- and medium-run analysis that captures labour market impacts, labour market rigidities and adjustment processes, developing these capabilities to analyse the impacts of policy choices on different segments of the labour market
• implement these extensions as a series of scenarios under different adjustment rules, not involving full forward-looking expectations and inter-temporal optimisation

1.3 Labour markets

Adjustment to changes in trading arrangements involves the reallocation of resources, including labour and capital, across activities in the economy. This, in particular the reallocation of labour across sectors and skill categories, is politically salient and crucial for identifying the distribution of gains and losses across individuals. Labour market adjustment is often slow as there may be mismatches in the skill requirements and geographical locations of expanding and contracting activities.

A standard assumption employed in CGE models is that labour markets adjust to long-run equilibrium full employment, implying no change in unemployment. A richer modelling of labour market adjustment requires that short- to medium-run labour market frictions are included in a dynamic CGE model. Such modelling extensions are available in the academic literature and have been applied elsewhere. Developing modelling capacity to apply this to the UK will involve working with other government departments and drawing on results from macro-econometric modelling techniques, where relevant.

We recommend that DIT:

• develops its capacity to analyse labour market adjustment processes triggered by changes in trade. This will involve both building the evidence base and incorporating adjustment frictions in modelling approaches
• apply this capacity to inform both dynamic and regional extensions of the core model. Further developments may include analysis of the impact of trade on different segments of the labour market, such as gender, skills, and education

1.4 Regional impacts

Changes in trade and technology can have impacts that are highly localized and very persistent, as demonstrated by the economic history of UK cities and regions. This is particularly true for labour markets where adjustment to change is principally a local, rather than a national, issue. Modelling provides various tools to explore these effects.

One possibility is to develop a full model of inter-regional trade within the UK, embedded in an inter-national CGE model. Such an approach is not recommended. It poses data requirements that are insurmountable in the near term and, as DIT focuses on national policy, it is excessively complex for the issues at hand.

The alternative is a ‘top-down’ approach that maps the sectoral changes coming from the CGE and complementary modelling to a sub-national level – regional, or perhaps even finer – to capture local labour market impacts. Currently, DIT utilises a simple mapping approach, apportioning CGE results with respect to a sector’s relevance in each region. This initial mapping is straightforward but needs to be accompanied by 3 further steps.

First, significant inter-regional linkages through the value chain need to be captured; in some cases, these are associated with specialized clusters in which agglomeration economies are important. Second, local linkages (‘regional multipliers’) through spending effects, principally on non-traded goods, need to be accurately modelled. And third, while spatial impacts may be driven by changes in sectoral employment, their economic effects often show up in housing markets, selective inter-regional migration, and the occupational and skill mix of the population. These are all areas where further development of the DIT’s analytical and modelling capacity is needed.

To best analyse regional impacts without overcomplicating the core model DIT is recommended to continue using a ‘top-down’ approach but to develop this significantly to include the further 3 steps mentioned in the previous paragraph.

We recommend that this approach is developed as follows:

• core CGE model results should be mapped across regions of the UK according to their sectoral impact, patterns of UK regional specialisation, and key sectoral input-output linkages
• within-region impacts should be captured by a ‘local multipliers’ approach, incorporating spending multipliers on local goods and other features of local economies
• regional analysis should be combined with improved modelling of labour market adjustment, enabling local employment impacts of trade policy to be captured

1.5 Environment

The environmental impacts of trade and trade policy are intrinsically important and bear on key government objectives. Environmental impacts of trade policy are, and will increasingly be, subject to intense public scrutiny. At the same time new trade policy instruments – notably carbon border adjustment mechanisms and other carbon controls – will likely come into play internationally.

CGE models are well-suited to capture the economy-wide adjustments that will be associated with climate policy, and to explore the economic consequences of changing climate, technology, and of adjustment to net zero. Capturing these effects will require substantial modelling inputs, both in terms of data and further extensions to the core CGE model.

The panel recommends that:

• DIT works with others in the international modelling community to develop extensions of the core CGE model that capture environment impacts and environmental policy instruments
• DIT should consider working with other government departments to develop modelling expertise that spans domestic and international aspects of climate and climate policy

2. Complementary modelling and sectoral approaches

Complementary modelling is needed for 2 main reasons.

First, much of the action from trade policy takes place within relatively few sectors of the economy and often involves detailed industry-specific changes for example in non-tariff measures. This requires a level of detailed industry-level modelling that goes beyond the representation of a CGE model.

Second, some of the channels through which trade policies affect the economy (for example changing the intensity of competition, affecting productivity, or encouraging FDI) are not well incorporated in CGE. The panel’s view is that it is appropriate to capture these effects by detailed modelling of sectors in which they are most likely to be important.

There is a wide range of issues and sectors for which such modelling – in some cases involving innovative approaches –is appropriate.

These include:

• the service sector: non-tariff barriers, standards, market access, mode of supply, FDI.
• innovative and high-technology sectors: the role of market size and export opportunities in promoting R&D and innovative activities
• digital trade, including e-commerce and trade in data
• market power and competition: the role of trade as a pro-competitive discipline
• productivity improvement: firm heterogeneity and selection
• FDI: the impact of trade policy on both home-market serving and export oriented FDI: productivity differentials between domestic and foreign firms and potential productivity spillovers
• value chains: disruption or facilitation of supplier networks and impacts on the location of productive activity
• agriculture: complex tariff structures, product standards, subsidy regimes

Capturing the key features of these sectors requires detailed empirical case studies, knowledge of the research literature on the impact of trade in the sectors, and ‘bottom-up’ information from industry experts and actors in the affected sectors. While these approaches are often termed partial equilibrium, they may involve several closely linked sectors (for example in a supply chain).

In all cases such models must be rigorous, transparent and evidenced. Objective criteria should be used in the application of such models to avoid a selection bias towards their use in favoured sectors. Proportionality is needed in choosing where complementary modelling is best used, bearing in mind that direct impacts of most FTAs will be very small in many industries. The panel recognises that the feasibility of modelling these is dependent on data and technical constraints.

The relationship between complementary models and CGE is important. In some cases it will be appropriate to present results from each of the approaches alongside each other, with commentary on why they differ (for example the mechanisms that are captured in one approach, but not the other). In others, consistency can be achieved either by having partial equilibrium modules that plug-in to the larger model, or by results from the partial equilibrium used as ‘extraneous’ inputs to the CGE. For example, a partial equilibrium study may indicate a change in levels of investment or productivity in affected sectors; these changes can be taken as input to the CGE model.

DIT is recommended to develop its modelling capabilities as follows:

• identify priority features that can be explored in a granular manner in complementary modelling. Potential candidates are: (a) tradable service sectors, (b) increasing returns to scale, particularly in innovative sectors, (c) FDI, (d) productivity and firm selection
• develop datasets to allow for highly disaggregated analysis and expand sector coverage to other areas of the economy such as services
• develop a family of models that are able to capture these features, identify data requirements, and move to trial calibration and simulation in relevant sectors
• explore the possibility of utilising partial equilibrium models and data nested within a CGE model

3. Developing and using a wider evidence base

Multiple sources and types of evidence are needed to inform trade policy impact assessment. Good descriptions of the baseline economy and of the proposed policy change are necessary. To capture the response of the economy to change, evidence includes econometric estimates of key economic relationships (‘elasticities’); case studies of previous changes in trade relationships, these being either statistical or qualitative; and ‘bottom-up’ information from sources familiar with the details of particular sectors or policies.

A priority is to ensure that the description of the policy change itself is accurate. This requires knowing and understanding the baseline value of economic variables, including the policy variables themselves. In the case of many non-tariff barriers this is not straightforward, requiring further research to assess the quantitative significance of initial barriers and the impact of regulatory and other changes.

The ‘base-line’ description of the economy reflects best available current data, but may not be the best description of the expected future state of the national and world economy at the time that the full impact of the policy comes through. Capacity to input alternative base-lines and phased implementation of policy should be developed. Alternative base-lines should, where possible, be drawn from publicly available sources (such as the IMF or OECD). The DIT itself should not be drawn into forecasting future states of the world, but should project its understanding of future economic developments and perform policy experiments in this context. Care needs to be taken in the presentation of results that follow, making an explicit distinction between changes in the base-line and those attributable to policy measures.

The DIT should continue to develop its work on estimates of key elasticities, updating estimates and assessing their appropriateness for the UK. This would improve the ability of the core CGE model to reflect the operation of the UK economy. The development of dynamic modelling will make it important to distinguish elasticities with respect to short-, medium-, and long-term time horizons. We recommend that work in this area is checked for consistency with relationships established in macro-econometric modelling.

There are major gaps in UK data coverage of the service sector that need to be addressed. In particular, the core CGE and PE databases used by DIT should be expanded to include further detail on and disaggregation of service sector trade. The development of an evidence base to analyse this area of the UK economy should be considered a priority given the economic importance of this sector.

There is an extensive research literature on the effect of historical trade policy changes on aspects of economic performance in countries around the world. Generalising from such case studies is problematic, but we take the view that information in these studies is a valuable input to model design and to appraisal of prospective policies.

We recommend that the DIT establishes dialogue with researchers who are expert in this area for the purposes of:

(a) identifying mechanisms that have been quantitatively important – or unimportant – in previous changes in trading relationships

(b) developing methodologies to utilise ex-post estimates of FTA impacts in the suite of PE and CGE models used by the DIT

A further source of sectoral level information is engagement with stakeholders in affected sectors. Extensive dialogue with stakeholders and sector experts (including those in other government departments) is encouraged to better understand key mechanisms. This will be required in areas where information from standard statistical sources is scarce, for example in the service sector, digital trade, and in the regional structure of intra-UK supply chains.

Ex-post model validation is an extremely demanding task as trade policy changes often take many years – or decades – to work through, during which time the economy will have been subject to numerous shocks. DIT should facilitate work by academic researchers to undertake such validation exercises.

In summary, we recommend that the DIT continues to:

• work to quantify the impact of existing trade frictions, in particular non-tariff measures
• upgrade its estimates of key model relationships
• improve data coverage of service sectors, developing approaches to analyse these sectors in granular detail
• establish greater engagement with researchers in order to learn from the extensive literature on the economic consequences of previous changes in trading relationships
• engage with industry and commerce across the UK to identify key mechanisms in affected sectors and to ‘reality-check’ model results
• promote ex-post evaluation of the accuracy of results obtained from DIT trade models

4. Presentation

The results from trade policy modelling are presented to different audiences; to government for informing policy choices; to business for planning future change; and to the wider public for informed debate. Each of these audiences have different requirements for speed of results, for sectoral detail, and for the time-scale of projections. They all require that results are credible, and 2 aspects of communication strategy are important to reinforce this.

The first is that there needs to be an understanding of why the results are what they are. This requires some understanding of the key variables (their order of magnitude) and of the mechanisms that underpin results. Essentially, results need to be presented with sufficient information and in such a manner that they can be subject to a ‘back-of-the-envelope’ reality test. For example, in sectors that are expected to experience an increase in production, the DIT should explain this in terms of the key factors driving the increase (that is lower trade costs, increase in market access, decrease in input costs due to lower trade cost on key inputs, and the like). Similarly for decreases, the change can be explained in terms of increased import competition from lower cost countries, or other relevant factors.

The second is that a range of results needs to be presented. This may be across sectors and time-scales, or perhaps more importantly across different scenarios. It is evident to all audiences that there are optimistic and pessimistic scenarios for any trade policy or FTA negotiation. Understanding is built by laying these out and making clear the circumstances and assumptions under which one outcome is more likely than another (this involving more than just presenting a confidence interval).

While popular presentation of results is outside the control of the modelling unit, the following recommendations might promote more informed use of results:

• provide the key estimates of inputs and parameters that drive model mechanisms, highlighting how these drive results and what their limitations are
• present ranges of likely scenarios rather than point estimates, these varying with respect to future states of the world as well as across estimates of parameters and inputs
• where possible present dynamic as well as static CGE scenarios, providing results that demonstrate the transition process to the final long run result
• support results with illustrative model calculations, clearly demonstrating how results are arrived at, and how they depend on the assumptions made

5. Concluding remarks

Our recommendations are intended to build on the sound modelling base that the DIT has already established and the capabilities that we observed during the course of the review. The lines of work we recommend should reinforce DIT’s core modelling capacity and guide its future development to better capture features of sectors that are key to UK trade performance, now and in the future, and to enhance all stakeholders’ understanding of the effects of trade policy. We recognise that implementing these recommendations will involve a considerable investment by the DIT, taking both time and additional resource.

In addressing these recommendations, the DIT will need to carefully consider the institutional setup and supporting technology in which it develops and implements its modelling developments. The robust economic underpinning of the modelling tool-kit must be accompanied by an organisational setup that facilitates the sustainability of the work (for example when staff change) facilitates reactiveness to policy work, and maximises transparency and accountability.

Annex 1: expert panel terms of reference: DIT modelling review

Objectives

The MREP was established to advise DIT on the development of its approaches to modelling trade policies, in order to best meet the department’s mandate and objectives.

The specific remit of MREP was to:

• advise on the development of modelling approaches and strategies to support the trade policies of the Department, based on a broad base of academic expertise
• highlight methodologies to assess the breadth of gains that are expected to be achieved from FTAs and the degree to which they can be captured in trade modelling frameworks
• advise the Modelling Unit on cutting edge studies and research and how to apply it so that DIT’s modelling reflects the forefront of research excellence
• provide a sounding board and technical advice on the development of modelling tools, econometric work and the development of existing models and ideas that the DIT modelling unit may have for further development
• offer a forum to exchange on current and future research trends on international trade and economic modelling
• offer advice to the DIT Chief Economist and Director of Analysis on potential research agenda

It should be noted that the MREP served solely as an external advisory panel and have had no authority or oversight over DIT research or policymaking. Nor have the MREP been asked to actively model a live agreement or agree to any modelling results as these remain the sole ownership of DIT.

Outline of the DIT Modelling Review Expert Panel (MREP)

Members of the MREP were individually invited by the DIT Chief Economist in consultation with Tony Venables, who led the MREP.

The selected panel formed a core membership and attended every meeting throughout the review. Other academics were invited to join the MREP’s deliberations where issues relating to their area of expertise were discussed.

The MREP and guest academics reflected a spectrum of experts in trade who are at the forefront of their research fields, mostly International Trade Economists including but not limited to Trade Modellers.

The MREP met monthly, with each meeting focusing on a specific topic in trade modelling. Meetings normally included:

• a discussion topic led by a MREP panellist
• a discussion topic led by DIT Trade Modelling Unit

Between the virtual meetings, Members of MREP occasionally met or corresponded on an individual basis with the DIT Chief Economist or the members of the Modelling Unit to discuss specific or outstanding issues.

Annex 2: membership of panel and overview of panel meetings

Members of the panel:

• Professor Tony Venables, University of Manchester – Chair of the Panel
• Dr Eddy Bekkers, World Trade Organization
• Associate Professor Swati Dhingra, LSE
• Dr Peter Dixon, Victoria University, Melbourne
• Dr Graham Gudgin, Cambridge University
• Dr Christine McDaniel, Mercatus Centre, George Mason University
• Professor Michael Plummer, SAIS Europe, Johns Hopkins University

Other experts were invited to join Panel discussions where issues relating to their area of expertise were discussed. The DIT thanks the panel members and the expert guests for their contributions.

These guests include:

• Dr Stefan Boeters - Labour market modelling
• Dr Jean Chateau - Environmental modelling
• Dr Erwin Corong - Dynamic modelling
• Dr Maros Ivanic - Sectoral modelling
• Dr Pham Van Ha - Dynamic modelling
• Dr Frank Van Tongeren - Productivity and COVID-19
• Dr Fan Zhai - Modelling specifications

Topics covered in meetings of the panel are listed below:

Meeting 1: trade theory – the specification of supply and demand

This meeting focused on the following topics:

• the objectives of trade modelling in evaluating policies
• the specification of supply and demand; which specification for example ‘Melitz’, ‘Armington’, ‘Krugman’ should DIT utilise in its models
• parameter and data requirements for different specifications

This meeting benefitted from a presentation by expert guest Dr Fan Zhai, who presented on the Melitz specification model he has developed, and the importance of utilising firm level data to accurately estimate parameters for such models.

Meeting 2: services trade – how to model trade in services

This meeting focused on the following key topics:

• modelling the different modes of services
• estimating NTBs for services trade and the role of gravity modelling
• modelling the structure of services sectors
• the importance of industry engagement

This meeting featured presentations from DIT and an open discussion on these topics.

Meeting 3: regional modelling – how to model impacts in different UK regions

This meeting focused on the following key topics:

• the merits and methods of regional modelling
• DIT’s current approach to regional modelling
• new features and developments to improve DIT’s approach

This meeting featured presentations from MREP member Professor Peter Dixon, focusing on his previous work on the US and Australian government’s regional models, illustrating the uses and benefits of different approaches.

Meeting 4: labour markets – how can labour market impacts be captured?

This meeting focused on the following key topics:

• short vs long-run modelling of labour market impacts
• capturing labour market frictions in trade models
• validation of model results

This meeting featured presentations from expert guest Dr Stefan Boeters, His presentation focused on approaches to capturing labour market impacts, highlighting how to combine modelling and off-model empirical evidence to successfully capture labour market impacts.

Meeting 5: dynamic modelling – should DIT incorporate dynamic analysis? What are the best approaches and uses of dynamic modelling?

This meeting focused on the following key topics:

• dynamic modelling and policymaking
• modelling dynamic baselines
• approaches to dynamics – recursive or forward-looking?

This meeting featured presentations from expert guests Dr Erwin Corong and Dr Pham Van Ha. Dr Erwin Corong presented on the uses of dynamic CGE models for policy questions, and the recursive dynamic approach whilst Dr Pham Van Ha highlighted the alternative, forward-looking dynamic approach, and its usefulness.

Meeting 6: alternative modelling approaches – are there useful alternatives to CGE models?

This meeting focused on the following topics:

• limitations of CGE models
• macro-econometric models and how well they model trade
• new quantitative trade (NQT) models
• micro data and ex-post analysis to inform CGE

This meeting featured presentations from several MREP members: Dr Graham Gudgin, Associate Professor Swati Dhingra and Dr Eddy Bekkers. They presented on the limitations of CGE models and macro-econometric models, NQT models, and micro data and ex-post analysis to inform CGE respectively.

Meeting 7: sector specific modelling – how can DIT better analyse specific sectors?

This meeting focused on the following topics:

• DIT’s current use of PE modelling, and how this informs sectoral analysis
• developments to DIT’s current PE models
• publication of sectoral results and supporting modelling approaches with sectoral case studies
• linking sectoral models and nesting PE with CGE models

This meeting featured a presentation from Dr Maros Ivanic, who focused on the challenges in capturing sectoral nuance in CGE and how multiple modelling approaches can be used together.

Meeting 8: modelling trade and the environment – how should DIT analyse the environmental impacts of trade policies?

This meeting focused on the following topics:

• the role of trade in environmental policy
• current modelling of trade and the environment
• limitations of modelling the environment, when should models be used and when is off-model analysis more beneficial?

This meeting featured a presentation from expert guest Dr Jean Chateau whose presentation focused on recent work and best practises in environmental modelling and the importance of trade in environmental issues.

Meeting 9: trade, productivity and COVID-19 – what economic impacts and events can and should be captured in trade models?

This meeting focused on the following topics:

• capturing the link between trade and productivity in trade models
• capturing wider economic events and structural changes such as off/onshoring and value chains
• capturing exogenous shocks for example. COVID-19 in trade models

This meeting featured a presentation from expert guest Dr Frank Van Tongeren who highlighted his recent work from the OECD in modelling reshoring and shocks such as COVID-19, and the lessons learned from these exercises.

Annex 3: wider academic consultation

Following completion of an initial draft report, the panel chair Professor Venables hosted a wider consultation, sharing the draft version with a wide list of more than 40 academics. Live consultation sessions were hosted, whilst written feedback was also welcomed. These helped to provide a check on the panel’s conclusions and accepted wider feedback beyond the core panel.

In attendance of the live discussion sessions were:

• Professor Denis Novy
• Professor Joseph Francois
• Dr Linda Yueh
• Dr Erwin Corong
• Dr Frank Van Tongeren
• Dr Robert Koopman
• Professor Alan Winters
• Dr Dan Ciuriak
• Professor Ngaire Woods
• Hector Pollitt

Outside of these sessions, written feedback was also received from:

• Professor Patrick Minford
• Professor Richard Baldwin
• Professor Bernard Hoekman
• Professor Dominique Van Der Mensbrugghe
• Professor Thomas Rutherford
• Professor Christoph Böhringer

Overall, consultees generally agreed with the main points and direction of the initial draft, but desired further development on some points, which have been reflected in the recommendations of this final report. Consultees also shared their expertise and technical details on avenues for DIT to address the recommendations, which will be crucial to DIT’s response to this report.

Annex 4: modelling developments, issues and approaches

This annex contains a detailed record of the modelling topics covered in the review. This highlights some of the evidence considered throughout the review and provides greater detail on the potential models and model features considered by the panel.

4. Computable equilibrium modelling

4.1 General equilibrium modelling: what are the alternatives?

The panel considered what alternatives there might be to the core CGE framework that DIT currently employs for its main modelling analysis. While CGE models have widely recognised limitations, as discussed in the main paper, they are the most widely used models for trade and environmental policy analysis by governments and institutions worldwide.

CGE models offer a compromise between the micro-economic foundations driving agents’ behaviour in the economy and the wider macroeconomic implications. Current CGE models offer a strong level of disaggregation in terms of sectors and regions. Currently, 65 economic sectors and 141 regions/countries are present in the core GTAP database typically employed in these models. Aside from their analytical strengths, CGE models benefit from the support of a wide community of analysts in governments, institutions, and academia. This allows for a collaborative, supportive environment upon which frequent model developments can be made.

Despite being considered the state-of-the-art approach by many, the panel explored whether there were alternative, more useful tools to capture the economy wide impacts of trade policies. Two main alternative frameworks were identified and discussed by the panel and were also raised in consultation sessions: Macro-econometric models and structural gravity (SG) models. While offering their relative advantages, most panelists took the view that CGE models are the best choice of core model for DIT’s purposes.

The following describes the overall merits and drawbacks of the respective approaches.

Macro-econometric models

While macro-econometric models are the basis of generally short-run projections of the macro-economy, they are seldom used to analyse trade policies. Typically, these models are designed with respect to policies or economic changes that principally have macro-economic impacts, such as fiscal policy, exchange rate movements and financial system shocks. As a result, these models are most developed in their representation of the financial sector, exchange rates and interest rates rather than trade policy changes (Perali and Scandizzo, 2016). Macro-econometric models are often highly aggregated to the point of featuring just a handful of sectors. There are some newer models that go beyond this, providing a greater level of disaggregation and therefore potential usefulness to DIT. Amongst these models, a promising candidate for DIT was identified as the Cambridge Econometrics E3ME model. During the aforementioned consultation process, Tony Venables, the chair of the MREP panel discussed this model with Cambridge Econometrics’ Chief Economist, Hector Pollitt. This model features a moderate disaggregation, featuring 61 regions and 43 sectors in each. However, for international trade analysis, these sectors are aggregated further to just 19 sectors, significantly fewer than the GTAP (used for CGE) database’s 65, although this could likely be disaggregated with some data development. While the latest version now incorporates bilateral trading relationships, the model largely provides analysis on the impact of macroeconomic policies/changes on trading relationships, rather than simulating the individual trade policies in which DIT is principally interested (for example FTAs). Macro-econometric models are commonly used by central banks and financial institutions due to their strengths, but are rarely used by trade policy institutions due to their limited applications in this area.

Considering these limitations, macro-econometric models are not the strongest contender for DIT’s central FTA modelling tool. However, these models can be useful as a complement to provide analysis on major macroeconomic changes that CGE models handle poorly (for example the impact of recessions). These models’ main strength is their econometric modelling of the time-series behavior of economic variables, but are limited for DIT’s purposes in the policy questions they can answer. Macro-econometric modelling techniques have a role in DIT’s modelling analysis, particularly if concerned with short-term changes and expectations dynamics, but should not be used as DIT’s central, core model when trying to understand the range of likely long-run economic effects from an FTA.

Structural gravity models

While econometric/gravity model estimates often serve as inputs to CGE or other models, the use of these models can be extended to provide a simulation of a change in trade, allowing for econometric general equilibrium analysis, often known as NQT models or SG models.

SG models are a type of general equilibrium model but with a much simpler structure than a CGE model (Yu et al., 2017). SG models, unlike CGE, follow economic data rigidly – opting to exclude the incorporation of any parameters that cannot be theoretically posited and easily empirically validated. Instead, SG models estimate parameters within model, using internal model data – like gravity models.

SG models put more emphasis on transparency and simplicity and less on detail, using a medium-sized model that aims to be rich enough to capture the most important effects (Costinot and Rodriguez-Clare, 2014). While these models are appealing in that they are simpler and more data driven, they ignore many economy-wide mechanisms such as the importance of savings and investment. Savings, and therefore investment, are excluded from SG on account of unknown parameter rates and elasticities. In addition, SG models can be used to calculate sectoral and aggregate welfare impacts but not impacts on employment, wages and other economic variables that are often of interest to policy makers. And, while they do not rely on as many parameters as CGE models, the parameters generated within SG models can be harder to explain as the model does not provide the mechanism or explanation for its effects. This can mean that the mechanisms driving results are more difficult to understand despite an overall simpler model. In contrast, the equations driving a given result in CGE models can be sense-checked against reality through evaluation of the model equations.

Ultimately, members of the panel experienced in this class of models summarised that analytically, either a CGE or SG model could be used to provide similar analysis. However, CGE models have the overall advantage for policy practitioners such as DIT due to their ‘consortium approach’, where there is a wider community internationally that can be drawn upon to constantly develop, improve and expand versions of the model. If DIT were to embark on development of an SG model, it would be an independent, resource-heavy, custom-built approach, and DIT could not make use of the existing data, model developments and research from the wider trade modelling community. The main valuable addition that SG models could make for DIT would be as a sense-check to more-complex CGE models, and a parallel tool upon which mechanisms and results can be compared. However, it is unlikely that SG models would be more useful for DIT than CGE as a core model.

While their limitations are recognised, CGE models are recommended as the central modelling tool for DIT’s trade analysis as they provide the analytical basis for reconciling the microeconomic and macroeconomic impacts of trade policy while offering a wider analytical community for developments, improvements, and new methodologies. CGE analysis should, however, be supplemented with other relevant modelling approaches and evidence that better represent specific issues. As economics and trade modelling is a constantly developing field, DIT is encouraged to engage with the literature on the latest and most cutting-edge approaches in development, experimenting with these and adopting them where available.

4.2 Core CGE specification

CGE models come with varying specifications that capture different economic mechanisms and are based on different theories of international trade. Armington models, based on perfect competition, are the simplest and most empirically researched models of international trade. It is recommended in the main report that DIT pursue a model with a simple, empirically grounded Armington specification before building upon this model with relevant extensions to capture further impacts.

Armington CGE model specifications are based on the Armington (1969) theory of trade, which set out the premise that consumers not only distinguish between different types of goods (for example machinery, clothing, chemicals) but also differentiate between goods according to their country of origin. Consumers therefore consume goods from different sources and consider them to be imperfect substitutes (for example British confectionary versus American confectionary, German cars vs Japanese cars). These models feature perfect competition, so do not recognise the difference between different firms operating in the same sector of a given country. Consumers only differentiate individual goods by their country of origin.

The Krugman (1980) model of trade builds upon Armington’s work, introducing firm set-up costs, monopolistic competition, and product differentiation at the firm level. However, firms in a particular sector/country are assumed to be symmetric, all having the same productivity level.

Melitz (2003) extends the Krugman model, introducing fixed costs to international trade. However, the major departure from prior models is that the Melitz model allows for variation in the productivity of firms. In Armington and Krugman models, all firms in a market sell on all trade links, while in Melitz only the most highly productive firms can sell on trade links due to the fixed costs that must be overcome. The introduction of heterogeneity in firm productivity has important implications when evaluating the welfare effects of trade, with trade leading to the selection of more productive firms while less productive firms contract, leading to an overall rise in productivity in the economy.

There have been increasing attempts to implement the Melitz system into CGE models, although these are often less empirically founded due to requiring greater assumptions and difficult to estimate parameters. In practice, the features represented in newer specifications such as Melitz are more likely to apply to some sectors than others. For example, some modellers argue that many services sectors are best represented by Melitz’s theory as they are likely to be differentiated between firms and potentially require fixed costs to enter international markets. As a result, the best compromise between empiricism and realistic representation of the economy would be to selectively apply Melitz specifications to specific sectors where the theory is more likely to occur, while retaining Armington characteristics in others. The use of these features should be driven by policy imperatives, qualitative sectoral engagements, and the latest research, with analysts using this knowledge to select which scenarios and sectors each economic theory should be applied to.

The use of complex specifications, without any improvement in the underlying data driving the model would not improve the accuracy of results and would instead generate more confusion as results become more difficult to interpret. More complex specifications, such as Melitz can also be computationally challenging and are less widely used and developed. These arguments suggest that the impact of trade on firm selection – enabling more productive firms to expand at the expense of less productive ones, may be better evaluated in partial equilibrium frameworks.

In view of the panel’s recommendations concerning complementary models (partial equilibrium models that look at selected sectors in detail), the panel took the view that a priority for DIT should be to focus on the simple Armington approach in its core model, and prioritise getting good data to apply this model in a robust and well-grounded manner. Improved estimates of non-tariff barriers, better data in sectors with weak coverage, and better estimates of elasticities will substantially improve the performance of the model.

4.3 Modelling extensions

4.3.1 Dynamics

CGE methodology first produces a baseline simulation of the economy, matching a set of baseline data, and then is shocked by introducing a change in policy or environment. This is the method of comparative statics, and is currently used by the DIT. It does not present results for the way in which the economy transitions through time to its final steady-state result – dynamic models aim to address this.

Dynamic models build upon the static framework and introduce a specific time element, tracking the transition between the initial and final steady-state equilibrium. While static CGE models demonstrate what the eventual long-run projection of a shock is, dynamic CGE models are of interest because they answer the question of not only what the projection of a shock is, but also how the economy would evolve were it not for the shock (Dixon, Rimmer, 2010).

Dynamic CGE models can provide insights far beyond static models when analysing trade policy questions such as the impact of an FTA. By modelling how the economy evolves over time, dynamic models can offer a picture of the adjustment processes as the economy progresses to a new equilibrium, allowing for a deeper understanding of the impact of a shock and results that can be easier to interpret and explain. Some of the various additions that dynamic CGE models can offer beyond static models in the context of trade analysis are described below.

Short-run impacts

Most CGE models are not useful for short-run analysis because they are comparative static models that assume full employment both before and after the policy change (Morley et al., 2011, Radulescu and Stimmelmayr, 2010). Adding dynamics enables CGE models to analyse short-term behaviour of the economy, and thereby address issues concerning the effect of adjustment costs, such as labour market frictions, on short- and medium-run employment and unemployment. There is significant literature employing the short-term results from a dynamic model to inform their analysis. For example, O’Ryan et al. (2011) use a dynamic model to examine a Chilean FTA with the EU and USA and find that significantly positive short-term effects are diminished in the long run through lower investment (due to less government revenue) and a gradual reduction in net-exports. The dynamic path ultimately allows for an improved understanding of both the short and long-term impacts of a shock.

The use of dynamic CGE models could also extend to analysis of other policy questions that involve shocks on trade that will have both short term and long-term impacts, such as the impact of COVID-19 or a recession. Indeed, Dixon et al. (2010) utilise a dynamic CGE model to examine the impacts of a possible H1N1 epidemic, using a quarterly dynamic model to capture the short-run impacts of a pandemic on areas such as unemployment, GDP, investment, and the like. More recently, Dixon and Rimmer (2021) have adopted a similar, updated methodology to analyse the impact of COVID-19 and government stimulus packages on the US economy. Within a static framework, short-run analysis like this would not be possible as the results would only show the final long-run equilibrium.

It should be noted that typically, the fundamental mechanics and structure of static and dynamic CGE models are largely the same (unless extensions for example labour market modelling are implemented), meaning that starkly different long-run results should not be expected compared to the static framework. Instead, analysts should expect improved short and long-term insights along with greater transparency, tractability, and contextualisation of results. Some of these additions are described below in detail.

Analysing sequential FTAs and staging of FTAs

Dynamic models permit analysis of questions surrounding the optimal sequencing of FTAs. While there have been few published studies on the sequencing of FTAs in a modelling framework, Moon and Lee (2010) use a dynamic GTAP CGE model to examine potential sequences of Korea signing FTAs with different partner countries and find that there can be an optimal sequence that maximises economic gains. They describe how the dynamic CGE model is suitable for this analysis, with the model being able to capture trade diversion/creation between countries/regions when liberalisation events happen sequentially. As a result, their modelling captures the changing incentives for countries considering signing an FTA once other FTAs have been signed, an aspect that could be important for DIT’s analysis. Moon and Lee (2010) compare their results with existing non-CGE economic/political game theory in this area and find that their results empirically support the theory. Their use of dynamic modelling demonstrates its potential usefulness in providing unique trade policy insights surrounding the sequencing of FTAs.

Similarly, introducing a dynamic framework could allow for the accurate modelling of the phasing out of tariffs from trade agreements. Aguiar et al. (2019) suggest that valuable results can be derived regarding a prospective trade agreement once the staging of other FTAs are added to dynamic models. McKibbin et al. (2004) agree that staging matters, finding that more gradual staging led to greater gains from a Korea-Japan FTA in their dynamic model. Considering this, access to credible analysis that indicates the short-term effects of altering the staging of prospective FTA’s will support the UK’s understanding of how industry behaviour will change over time – which may become a powerful tool for examining welfare implications of staging proposals.

Setting accurate baselines and capturing structural economic change

The effects of trade policy will interact with other changes – including other changes in government policy – that occur through time. This means that the ‘baseline’ of the model should incorporate expected changes in policy and other aspects of the world economy that are likely to occur. Accounting for time-based changes in policies improves modelled baseline forecasts by creating a counterfactual from which to accurately compare the economic effects of a future trade policy change (Aguiar, et al., 2019). In adopting such an approach, Aguiar recommends the inclusion of time-varying government policies as much as possible into baseline settings as this can significantly alter the true change in key figures such as GDP growth and employment rates.

One of the most unique and useful aspects of dynamic CGE studies are that they often attempt to incorporate some structural changes of the economy into their baseline, providing analysis in the context of a future world economy. Chateau et al. (2020) explain that incorporating the latest information on the projected structural change of the economy can lead to more accurate baselines, arguing that a realistic baseline cannot be produced without considering supply side structural change. Britz and Roson (2018) argue that long-run structural change is especially relevant for analysis, explaining that while no one can predict future technology and events, modellers should still incorporate important slow economic adjustment processes that are currently active. Among other impacts, these can include productivity changes in sectors over time, debt accumulating over time, and aggregate saving rates linked to population and income dynamics.

There are various examples of structural change being modelled in dynamic CGE modelling exercises, with this area being particularly important for the analysis of agreements with developing countries. In one example, Bekkers et al (2021) implement structural change to project the Chinese economy and its trading relationships in the future, including factors such as a falling savings rate, increased share of skilled workers and rising productivity in high-end manufacturing. Dixon and Rimmer (1999) highlight the importance of incorporating such changes, showing that doing so can lead dynamic models to disagree with static models on the outlook of a policy. Through projecting that tourism would have a greater share in Australia’s economy in the future and thus emphasising this sector’s importance in the future Australian economy, their MONASH model presented a more pessimistic outlook on a tax policy than static models. Dynamic modelling therefore offers a way of incorporating structural changes to the national and world economy in the baseline, allowing for more insightful results that consider the future economic environment in which the effects of trade policy are playing out.

Model transparency and validation

Another aspect of dynamic CGE models is that they offer a plausible way of validating CGE model results and improve model transparency. Validation exercises are not very common in the CGE space, partly because most experiments are static. Static CGE modelling typically aims to demonstrate the marginal economic impact of an isolated shock. As a result, it is difficult to validate results against actual outcomes as real world shocks happen among a variety of others, making it hard to disentangle the true impact of the shock (Dixon and Rimmer, 2010). Given that dynamic models show how the economy will evolve over each period, the results and baseline can be compared with real outcomes as they occur, creating an avenue for results validation.

Dixon and Rimmer (2010) demonstrate how CGE models can be ex-post validated through baseline forecasting with a dynamic model. Through ex-post analysis of US economy from 1992 to 1998 they found that dynamic CGE forecasts were much more accurate than a trend forecast and were made extremely accurate when implementing more accurate exogenous movements (retrospective forecasts that would not have been available at the time.)

While dynamic models are more complex in their nature, their results can be easier to interpret and explain as they show how the economy evolves over time to reach a final equilibrium, offering tractability and transparency not seen in static models. Dynamic modelling frameworks therefore offer results that can be easier to interpret and the opportunity to validate and critique a CGE model through ex-post analysis.

Approaches to modelling dynamics

There are 2 main approaches to dynamics in CGE modelling. Recursive dynamic models are the more common form of dynamic CGE model (Brocker and Korzhenevych, 2013). A common assumption in recursive models is that they are myopic, meaning agents do not see beyond the current period. The main alternative approach to these are forward-looking models, where agents can respond to a policy event before its actual implementation. Forward-looking models are much less common due to analytical and computational difficulties (Brocker and Korzhenevych, 2013).

Recursive dynamic models

Most dynamic extensions are recursive, essentially performing a series of static simulations and linking these together with an endogenous and exogenous variable updating procedure (Cockburn et al., 2004). While it is possible to implement rational expectations in recursive models (Dixon et al, 2005), it is very computationally time consuming and not truly viable for widescale use. As a result, these models typically feature static or adaptive expectations, where expectations adjust each period as economic changes occur (Van Ha et al, 2016). Ultimately, while recursive models are more common, better developed, and computational than forward-looking models, they do not contain rational forward looking behaviour so are unable to capture, for example, announcement effects (that is announcement not of a future policy causing reaction today).

Most of the dynamic models used by institutions around the world are recursive such as the LINKAGE model (World Bank), GTAP-Dyn (widely used by researchers) and the G-RDEM model (US Bureau of environment). Many of these are inspired by or based upon the MONASH model (Australian government).

Forward-looking dynamic models

In contrast to recursive models, forward-looking models link current decisions to future economic values (Melinkov et al. 2020), allowing agents to fully respond to current and future policy shocks. Van Ha et al. (2016) suggest that this is particularly important for trade policies, as agreements are typically phased in over time. Forward-looking models must solve all time periods at the same time (Robinson, 1991) leading to significant computational restraints. As a result, Lemelin (2014) argues that while forward looking models can support full intertemporal optimisation, recursive dynamic models are more practical in most situations due to them being much more computable.

Despite this limitation, recent literature has looked extensively at making forward-looking models more computable, finding techniques that massively reduce the computational intensity and make these models much more viable and scalable. Indeed, Van Ha et al (2016) present a new, less computationally intensive method of solving forward-looking models through parallel processing that is easily scalable. Recursive models cannot make use of this parallel processing innovation as they must solve sequentially, one period to another. Similarly, Melnikov et al. (2020) present an alternative method which reduced computational time by 97% in their forward-looking model, from 10 hours to only 20 minutes. Therefore, forward-looking models may be more achievable than before with recent literature offering solutions to the computational intensity.

Lecca et al. (2012) question whether the importance of forward-looking expectations may be overstated, demonstrating using their own model that there is little difference in the adjustment paths and zero difference in the long-run equilibria between a recursive and forward-looking version of an identical model. Thus, they argue that it does not matter which is used for looking at the long-run, although it is still an issue if interested in short-run impacts of policies and the dynamic path.

Data and requirements

As recursive dynamic models are solved as a sequence of time-based comparative statics models, much of the data and parameters requirements are the same as conventional static CGE models. The distinct difference in data requirements, however, centres around the presence of exogenous dynamic variables that need to be accounted for to produce accurate modelling baselines. These capture the ‘base case’ of how modellers expect the world to develop without the shock, accounting for aspects such as population growth, productivity changes, GDP growth over time, and the like. Though dynamic models can be solved without the need for additional data; this would require informed estimations as to the evolution of endowments, population, and the labour force. Instead, modellers should aim to use existing data and forecasts from international organisations (for example OECD, IMF and World Bank) or government departments to estimate movements in exogenous variables such as projections on GDP growth for different countries, demographic change and educational attainment (Aguiar, et al., 2019).

4.3.2 Labour markets

In any CGE modelling simulation, the selection of ‘closure’ rules is paramount to scenario design. These rules dictate which variables are exogenous and endogenous, and thus which variables the model is computing rather than taking as given in its simulations. The selection of these rules can be controversial as it impacts the mechanics and assumptions of the models, and thus the meaning and interpretation of results. A common closure rule in CGE modelling exercises is the assumption of no change to long-run unemployment, meaning that the model solves for a final solution in which unemployment is unaffected. This is a common, well-studied and accepted economic assumption but can be controversial as permanent changes to employment have historically occurred, and disagreement amongst economists is frequent.

Some economists challenge the assumption of zero impact on long run unemployment. They argue that it is unrealistic, citing real world evidence in which long-term structural unemployment occurred following shocks. While it is possible for DIT to model changes to long-run unemployment in this way, such an approach would require significant and difficult to estimate empirical evidence to be performed accurately. There would be little benefit in applying such an approach as standard, but DIT could experiment with long run unemployment analysis if it arises as an area of particular concern to policymakers.

Dynamic labour market analysis

By adopting a dynamic CGE model, DIT would open viable avenues for analysing short-term labour market impacts, including variations in employment. This can first be incorporated in a simple manner with modifications to closure rules, allowing for unemployment changes in each period. Additionally, frictions derived from empirical evidence could be introduced to the model to realistically represent the costs of moving between jobs. This model would estimate short-term changes in employment and unemployment while considering the cost of short-run frictions and maintaining the assumption of long-run full employment. By assuming the labour market returns to equilibrium in the long run, unemployment is only affected in the short-run and short-term labour market impacts would translate into long-term wage changes.

Dixon and Rimmer (2018) describe an example of how a labour market module could be set up. This helps to provide analysis on the structural adjustment problems that can arise from international trade. Their example identifies 10 broad occupations in the US economy. Here, where sectors are damaged by an FTA, employment is impacted, and workers in that industry (of a given occupation) will seek employment in other sectors and can displace workers from other sectors/regions.

Dismissal rates, willingness of workers to move occupation and region, and skill closeness between occupations are all estimated assumptions which will drive the extent of short-term impacts. Dixon and Rimmer (2018) describe the key ingredients of labour market modules as:

• the division of the workforce into categories, at the start of year t reflecting activities in year t-1. (for example in a previous study analysing impacts on legal and illegal workers, labour was divided by: ‘Legal worker, employed in occupation X in year t-1,’ ‘illegal workers, employed in occupation X in year t-1,’ ‘legal worker, long term unemployed in year t-1’, ‘legal worker, new entrant’)
• the determination of labour supply from each category to each activity. Each ‘category’ of labour has a different optimisation problem, and the model specifies which activities people in each category prefer to perform. These category-specific optimisation problems capture a variety of ideas from labour economics: people in long-run unemployment become discouraged and search less actively than do employed and short-run unemployed people; people in occupation X offer strongly to continue in occupation X; and people in occupation X cannot make an effective labour supply to occupation Y if the qualifications required for these 2 occupations are incompatible
• the determination of demand for labour in employment activities. Demand for labour in each occupation is specified for each industry via hiring choices and then aggregated across industries
• the specification of wage adjustment processes reflecting demand and supply. These equations recognize that when a shock affects either the demand for or supply of workers in occupation X, it takes time for wages to adjust to their market clearing level
• the determination of everyone’s activity: who gets the jobs and what happens to those who don’t? This part of labour-market modules specifies vacancies in each occupation taking account of demand for workers in that occupation and desires of incumbents to continue in their occupation. The modules then describe competition to fill vacancies in occupation X between new entrants, unemployed workers, and workers from other occupations.

Dixon and Rimmer (2018) present a visual summary of these labour market dynamics below:

Overall, the way in which this process is defined for example categories, activities, labour demands, and the like. can be adjusted based upon the needs of the given economy or policy in question.

Labour market segmentation

As mentioned prior, when implementing a labour market response to a core dynamic CGE model, DIT will need to segment the labour market in an informative and relevant way, to provide effective labour market analysis which informs policy objectives (for example skill, occupation, gender and the like). There are various ways in which this could be approached, with the aim of decomposing where in the labour market impacts are likely to be felt. The following describes some approaches that have been advocated within trade modelling literature and applied in models.

4.3.3 Regional modelling

Employment impacts are inherently linked to the impacts on regions, as different sectors are concentrated in different areas of the UK and communities have interests in local industries. Regional analysis provides an important element of DIT’s scoping and impact assessments, highlighting the likely economic impacts on different regions of the UK. There are 2 conventional modelling approaches to achieving this. A ‘bottom up’ approach involves explicitly modelling each region in a country, essentially splitting up a country’s base input data into various regional components. The alternative is a ‘top down’ approach where a country is modelled as a single entity in the core model, but results are then applied to a further model containing supplementary regional economic data which dissects regional impacts from the core results.

The ‘bottom-up’ approach is not recommended for DIT’s purposes. It involves immense data requirements, with explicit trade data between different regions of the UK being required. This data is sparsely available in the UK due to the lack of regional input-output tables for UK regions. Estimating intra-regional trade data for the UK would require significant resources and would suffer from inaccuracies due to crude assumptions and estimations. There are also important computational restraints, as an intra-regional trade may require a trade off with other potential modelling extensions as computing capabilities may struggle to handle multiple extensions along with a model explicitly modelling trade between areas of the UK as well as FTAs with other countries. The best compromise between data, computational restraints and capturing important regional economic effects is therefore to develop a bespoke ‘top down’ approach drawing upon the regional economic data available.

DIT’s current ‘top down’ approach takes core CGE model results and apportions these based on location quotients, which are coefficients of sector concentration in different regions. This is a sound approach to gauging regional welfare effects as a first step when data are limited but can be developed significantly to capture regional economic interlinkages. DIT is recommended to continue using a ‘top-down’ approach, but to develop this significantly to capture regional economic linkages such as local multipliers and agglomeration economies. Top-down regional models provide robust regional analysis without posing insurmountable data challenges or computational difficulties (Ghaith et al, 2021.) This could be a bespoke regional model which uses local multipliers and other regional data to analyse core results and estimate economic relationships between regions of the UK. These are common best practise in regional economic analysis.

There are various examples of top-down regional models, with these being regularly deployed in regional studies. Dixon and Rimmer (2007) demonstrate an example of how core CGE results can be regionalised in a top-down framework. This application inputs results into a system of equations, which are based on coefficients representing regional shares of various aspects of the simulation for example. regional share of employment, regional sectoral concentration, regional share of consumption, and the like. They describe the top-down approach’s main limitation being the assumption that the shares discussed do not change. In other words, if interested in analysing how the regional composition might change, a top-down approach is not as useful. Dixon and Rimmer (2007) conclude that the top-down approach is most appropriate when concerned with national policies, for example tariff liberalisations, which are within DIT’s remit.

4.4 Environmental modelling

As the effects of climate change are becoming more understood, the critical importance of environmental preservation and the protection of global ecosystems is increasingly recognised. DIT will need to assess the environmental implications of trade policy and evaluate the role of trade policy in the light of the government’s objective of net zero carbon emissions.

Many FTAs now have specific environmental chapters, especially those signed by the OECD nations (WTO, 2010). Commitments to enforce minimum environmental regulations as well as clear litigation mechanisms within the treaty’s text has created a clear need for investment into environmental analysis from trade agreements. New trade policy instruments – such as carbon border adjustment mechanisms – also need to be studied.

CGE models of the type used in trade policy modelling have become increasingly popular in the realm of environmental modelling. Dynamic CGE models have been widely accepted as the preferred approach to trade/environmental modelling as environmental policies and changes occur over time and in an uncoordinated fashion across countries. Similarly, the cumulative size of emissions is of greater concern than the long-run annual emissions after a trade shock. Furthermore, modellers have begun using dynamic CGE models to endogenize consumer preferences towards carbon-friendly products, as the prevalence of CO2 in the atmosphere intensifies. These crude attempts at capturing the demand size responses are crucial to getting a truer sense of the feedback effects on the economy.

Modern trade theory on the effects of FTAs on the environment was developed by Grossman and Kreuger (1991) who presented 3 dominant mechanisms for which trade shocks could induce change to the surroundings:

• scale effect – trade leads to reductions in prices for both parties and therefore an income effect which expands economic activity. This expansion will unambiguously increase emissions and pollution to the environment
• composition effect – trade encourages national specialisation towards their comparative advantage sectors. As a result of this, the mix and share of economic activity by sector will adjust from an FTA which may expand highly emitting industries and shrink cleaner ones (or vice versa). The direction is therefore ambiguous
• technique effect – trade liberalisation accelerates technological transfer to the developing world. If modern technology and manufacturing processes are more green than current techniques, then FTAs could disseminate a lower carbon-intensive approach to energy demanding sectors (lowering their carbon dioxide per output ratio)

Environmental CGE extensions

There are various CGE models capable of environmental analysis. Nong (2020) describe some key features present in environmental models:

• emissions accounting: data on carbon dioxide and greenhouse gas emissions is implemented into the model, with emissions being linked to production processes. These are released by industries, households, the government, and the like
• carbon taxation: Using emissions accounting, a carbon price variable can be implemented to apply to emissions from particular sources
• emission permits and emission trading: Regional blocks of emission traders are created. These regional blocks might have emission quotas, within which regions might exceed quotas, but an overall bloc of regions will stick to its respective quota
• energy disaggregation: Models often go further, implementing a disaggregation of the energy sector and linking emissions to different fuel sources for example coal, oil, gas, petroleum

Data

Access to highly disaggregate, long-time series, data remains the largest hurdle to environmental models (US EPA, 2017). The UK, unlike most countries, does have an extensive suite of disaggregated data on emissions from institutions such as the National Atmospheric Emissions Inventory (BEIS), UK AIR (Defra), and the ONS. Harnessing these datasets within DIT’s new framework for FTA analysis will be an important data requirement for capturing environmental impacts.

2. Complementary approaches

Sectoral modelling

Aggregation issues

As a model of the world economy, CGE models often fail to account for many of the sector-specific economic details that are present in each industry. CGE studies typically apply the same broad economic structure (albeit with different parameter values) to a wide range of economic sectors, and require consistent data for many nations, sectors, and economic variables, which is sparsely available. As a result, CGE models are weak at drawing out sectoral detail and analysing tariff-line level policies, where contentious and late-stage trade negotiations often occur (Jafari et al. 2021).

Narayanan et al. (2010) propose 3 main reasons as to why highly disaggregate modelling is needed for credible trade policy analysis:

1. There exist significant variations in tariff lines across similar commodities, within a sector, creating serious aggregation biases.
2. Aggregation of sectors may result in ‘false competition’ that is 2 countries that do not compete in a disaggregated market may appear as competitors once aggregated.
3. Many policies aimed at specific products are not identified among highly aggregated sectors.

Aggregation issues can be particularly troublesome in analysis where there are sector-specific features such as zero initial trade flows between countries analysed or tariff rate quotas and other non-tariff instruments.

Partial equilibrium modelling

The most common models used for highly disaggregate trade analysis are partial equilibrium (PE) models. These models sacrifice many of CGE’s economy-wide linkages to focus on disaggregated sectors (Narayanan et al., 2010). PE models are a general description of a wide range of models that do not include general equilibrium effects; these can be econometric-based PE models, PE models nested within CGE frameworks or independent models with their own equations, parameters, and unique specifications designed for different purposes. Importantly, these models must be designed to capture features of sectors that are absent from the stylised representation of CGE models, including features such as FDI, increasing returns to scale, firm heterogeneity, and value chains.

International organisations often present PE results alongside CGE results or even link these models together - acting as a complement that provides more detailed sectoral insights than CGE modelling can. In previous impact assessments, organisations such as the US International Trade Commission (USITC) or European Commission have used PE modelling alongside CGE to look at specific sectors of economic interest or to identify granular sectors with significant trade opportunities. For example, the EU-Japan sustainability impact analysis used PE modelling to focus on granular food sectors that CGE could not solve for. Delzeit et al. (2020) support the use of PE, highlighting PE models as useful tools that can be used to assess important sector-specific economic issues more accurately such as land and energy usage.

Nested PE models

Naranayan et al. (2010) argue that a PE model nested within a CGE framework is superior to either model alone, as it borrows the strengths of both models and allows users to analyse more granular sectors in a way that is consistent with the CGE framework. They propose nested PE as an effective way to follow the natural evolution of FTAs, from early-stage negotiations where fine details of an agreement are not clear and the whole economy impact is important, to later stages where analysis on tariff line level detail can be crucial. By nesting, users obtain aggregate, whole economy analysis while also obtaining specific and granular sectoral results that are consistent with the wider CGE model.

Ultimately, there are 2 main avenues in which DIT should develop its PE capabilities. On one hand, standalone partial equilibrium models and datasets should be developed to handle greater disaggregation, coverage of more economic sectors and features that capture sectoral intricacies. DIT should develop its current partial equilibrium models in this way, exploring more disaggregated data, expanded coverage and potential model features that would allow a greater understanding of specific sectors of economic interest. On the other hand, DIT should experiment with using this partial equilibrium data in a linked/nested fashion in its CGE framework, exploring the use of nested PE simulations on select sectors that might be of significant relevance to a given policy scenario.

Annex 5: Additional submission by panel member Professor Graham Gudgin

This report contains a welcome diversity of approaches to estimating the economic impact of international trade agreements. In particular a number of welcome suggestions are made in the direction of greater simplicity and transparency of analysis including a greater focus on empirical studies of the key sectors in any trade negotiation. These are proposed as adjuncts to run alongside the GTAP general equilibrium modelling which is proposed to continue as the core of the DIT’s approach. However, the fact that the core strategy is based on a single rather theoretical model is in my view an important weakness. The accuracy of the core model is unproven and it is unclear how complementary empirical estimates will be conducted or how much salience they will receive in providing guidance to ministers. What for instance happens if empirical estimates contradict the results of the core modelling.

The complex and theoretical nature of the core model (the GTAP general equilibrium model) is unlikely to be sufficiently appreciated by ministers and their non-economist advisors. There is little in the report to describe the nature of the GTAP model in terms that non-specialists can appreciate. It seems unlikely that ministers and their advisors are likely to understand that this is a theoretical system based on often unrealistic assumptions. Further detail is needed to give ministers and others at least a feel for what is involved.

The fact that general equilibrium models such as GTAP are built on the idea of representative agents in which the representative consumer maximises his/her ‘utility’ and firms maximise profits with prices determined in order to achieve these maximisations, should be explained. The idea that wages and prices clear in competitive markets to optimise welfare describes an ideal than a reality. This is most obvious in the labour market where the basic model assumes that wages adjust to achieve full employment in some unspecified long-term with no identified path from to that long-term ‘equilibrium’ outcome. Much of the Panel’s deliberations were focussed on ameliorating these shortcomings, but inevitably the adjustments to the model are likely to be in technical and sometimes arbitrary.

The important initial stage of the model consists of estimates on the impact on trade of changes in tariff and non-tariff barriers. An accurate or at least meaningful specification of the links between tariff and non-tariff barriers (so-called elasticities) on the one hand, and trade volumes on the other, are critical in obtaining helpful results. However, the report says too little in my view about how these key parameters might be obtained. At the very least the elasticities for key sectors should be checked with the main firms trading in those sectors.

Although the GTAP model is very widely used in various forms across the world, including by the United States International Trade Commission (USITC) there have been rather few attempts to assess their predictive accuracy. The report itself refers to this in stating that ‘outputs from the model – the predictions made about the effects of trade policy – are rarely validated by comparison with actual outcomes, and where this has been done the models do not perform particularly well’ (page 5). Despite this rather serious admission the report but does not follow-up the implications. Ackerman and Gallagher (2008) for instance conclude that ‘CGE models fall short of offering a useful, comprehensive framework for thinking about and measuring important effects of trade. Despite all its complexity, the theoretical apparatus ironically enforces arbitrary, undesired simplifications from the esoterica of Armington elasticities and the rigidities of static analysis, to the central flaw of ignoring employment effects by design’. A number of studies have evaluated the accuracy of general equilibrium models in predicting NAFTA impacts on changes in bilateral industry-level trade flows. Kehoe et al (2017) conclude that these ‘show that the models did poorly. The models did so poorly, in fact, that the predictions of the models were often negatively correlated with the actual changes observed post-NAFTA’. Unless these concerns about the accuracy of GE models can be successfully addressed it calls into question DIT’s intention to continue with GTAP as a core model.

It is fair to conclude, I feel, that since the DIT economists are currently organised to run a GTAP model, and this group has continued to recruit the skilled personnel to do so, there was never a realistic possibility that any other conclusion would be reached in this report than to continue with GTAP. It is true that highly experienced GTAP modellers on the panel supported this approach, and in some cases had made efforts to validate the GTAP approach, these validations do not (in my view) provide sufficient confidence in the accuracy of the approach. For instance, the claim of Dixon and Rimmer^{[footnote 1]} that ‘past movements [in US trade] are a very imperfect guide to future movements. Encouragingly, we have found that USAGE [model] output forecasts at the 500-commodity level comfortably outperform [extrapolation] trends” is followed by the admission that: “On the other hand average errors for these for these forecasts seem alarmingly high’.

Although there is mention in the report of alternative modelling approaches these were never seriously followed up and there was no recommendation to run alternatives such as the Cambridge Econometrics E3ME model alongside GTAP. This type of model is much more dependent on actual economic data than is GTAP and the database consists of historical time-series data parameters are validated against historical relationships. The model already contains features such as annual forecasts and regional and environmental impacts that the report proposes to add onto its basic GTAP capacity. The model is more realistic; agents are not assumed to have perfect knowledge and are not assumed to optimize their decisions. Full employment is not assumed and involuntary unemployment is an important model outcome. The model was rejected due to its lower level of detail on trade compared with GTAP but little effort was made to determine how much effort would be needed to extend the model in this direction.

In case the above views be thought unusual it is worth pointing out similar views were expressed earlier this year by of Jason E Kearns chair of the USITC. Kearns added an ‘additional view’ at the end of the USITC’s 5-year review of the impact of US FTAs ion the US economy^{[footnote 2]}. Although the analysis of the Commission he heads relied on GTAP modelling he stated that ‘the trade community needs to fundamentally rethink how to assess the impact of trade agreements on U.S. workers, businesses, and farmers. We need to ask more and different questions, question our assumptions, and dig deeper into the substantive terms of our trade agreements’. He adds that the USITC model ‘relies on unrealistic assumptions about the economy, such as the assumptions of full employment (which implies that all workers seeking jobs are employed and that trade agreements cannot cause unemployment) and costless switching (that workers have the ability to freely move across industries and occupations). This contrasts with some recent academic studies that highlight how significant the transition costs are for US workers and how long the transition can be after a wave of import competition’. ‘As a result of the unrealistic assumptions in the economy-wide modeling, this report improbably concludes that all demographic subgroups (‘labor types’) analysed gained from trade agreements’.

Furthermore he states that, ‘our modeling approach has been adapted from a time when removing or reducing tariffs and nontariff barriers was the focus of trade agreements. But trade agreements today do not simply reduce or eliminate (‘liberalise’) trade barriers and expand trade, and not every domestic rule or regulation should be viewed as an ‘unnecessary obstacle to trade. Finally, too often, economic analysis of provisions in our trade agreements focuses too narrowly on the quantifiable expansion in trade, as if trade expansion is an end in itself, rather than a means to achieve the broader objectives articulated in our trade agreements, such as higher standards of living’.

In conclusion, the absence of convincing research to verify the accuracy of GTAP modelling runs the risk that apparently scientifically-based advice given to ministers may not be reliable or accurate. This is why the more comprehensible and simpler research conducted without complex theoretical models is so important. If I were a minister wishing to understand what impact new trade agreements might have on the UK economy I would wish to see estimates of the direct impacts on exports and imports of changes in tariff and non-tariff barriers and to understand how the key elasticities used to obtain these estimates were calculated. Indirect, second-round effects could then be added to appreciate how the immediate impacts might spread to other parts of the economy. Finally, estimates could be presented for the wider macro-economic impacts of the direct changes including the redeployment of workers displaced by new agreements. Unless the last of these 3 stages comes from a model with realistic assumptions, this stage of conclusions should, in my view, be given limited credence.

References

Aguiar, A., Corong, E. & Mensbrugghe, D. v. d., 2019. The GTAP Recursive Dynamic (GTAP -RD) Model Version 1.0, s.l.: GTAP Purdue University.

Annabi, N., Cockburn, J. and Decaluwé, B., 2004, June. A sequential dynamic CGE model for poverty analysis. In Preliminary Draft for the Advanced MPIA Training Workshop in Dakar, Senegal.

Armington, P., 1969. A Theory of Demand for Products Distinguished by Place of Production. Palgrave Macillan on behalf od International Monetary Fund.

Bekkers, E., Koopman, R.B. and Rêgo, C.L., 2021. Structural change in the Chinese economy and changing trade relations with the world. China Economic Review, 65, p.101573.

Britz, W. and Roson, R., 2018. G-RDEM: A GTAP-based recursive dynamic CGE model for long-term baseline generation and analysis. University Ca’Foscari of Venice, Dept. of Economics Research Paper Series No, 11. Bröcker, J. and Korzhenevych, A., 2013. Forward looking dynamics in spatial CGE modelling. Economic Modelling, 31, pp.389-400.

Chateau, J., Corong, E., Lanzi, E., Carrico, C., Fouré, J. and Laborde, D., 2020. Characterizing supply-side drivers of structural change in the construction of economic baseline projections. Journal of Global Economic Analysis, 5(1), pp.109-161.

Costinot, A. and Rodríguez-Clare, A., 2014. Trade theory with numbers: Quantifying the consequences of globalization. In Handbook of international economics (Vol. 4, pp. 197-261). Elsevier.

Delzeit, R., Beach, R., Bibas, R., Britz, W., Chateau, J., Freund, F., Lefevre, J., Schuenemann, F., Sulser, T., Valin, H. and van Ruijven, B., 2020. Linking global CGE models with sectoral models to generate baseline scenarios: Approaches, challenges, and opportunities. Journal of Global Economic Analysis, 5(1), pp.162-195. 

Dixon, P. & Rimmer, M., 2010. Validating a Detailed, Dynamic CGE Model of the USA*. Economic Record, 86, pp.22-34.

Dixon, P. and Rimmer, M.T., 2018. Creating a labor-market module for USAGE-TERM: illustrative application, theory and data. Centre of Policy Studies (CoPS), Victoria University.

Dixon, P.B. and Rimmer, M.T., 1999. Changes in indirect taxes in Australia: A dynamic general equilibrium analysis. Australian Economic Review, 32(4), pp.327-348.

Dixon, P.B. and Rimmer, M.T., 2021. Coping with seasonality in a quarterly CGE model: COVID‐19 and US agriculture. Australian Journal of Agricultural and Resource Economics.

Dixon, P.B., Lee, B., Muehlenbeck, T., Rimmer, M.T., Rose, A. and Verikios, G., 2010. Effects on the US of an H1N1 epidemic: analysis with a quarterly CGE model. Journal of Homeland Security and Emergency Management, 7(1).

Dixon, P.B., Pearson, K.R., Picton, M.R. and Rimmer, M.T., 2005. Rational expectations for large CGE models: a practical algorithm and a policy application. Economic Modelling, 22(6), pp.1001-1019.

Dixon, P.B., Rimmer, M.T. and Tsigas, M.E., 2007. Regionalising results from a detailed CGE model: Macro, industry and state effects in the US of removing major tariffs and quotas. Papers in Regional Science, 86(1), pp.31-55.

Ghaith, Z., Kulshreshtha, S., Natcher, D. and Cameron, B., 2021. Regional Computable General Equilibrium Models: A Review. Journal of Policy Modeling.

Grossman, G.M. and Krueger, A.B., 1991. Environmental impacts of a North American free trade agreement.

Jafari, Y., Himics, M., Britz, W. and Beckman, J., 2021. It is all in the details: A bilateral approach for modelling trade agreements at the tariff line. Canadian Journal of Agricultural Economics/Revue canadienne d’agroeconomie, 69(3), pp.415-442.

Krugman, P., 1980. Scale Economies, Product Differentiation, and the Pattern of Trade. American Economic Association.

Lecca, P., McGregor, P.G. and Swales, J.K., 2013. Forward-looking and myopic regional Computable General Equilibrium models: How significant is the distinction? Economic modelling, 31, pp.160-176.

Lee, C.S. and Moon, D., 2010. Impacts of sequential free trade agreements in East Asia: A CGE and political economy analysis. Global Economic Review, 39(4), pp.365-381.

McDougall, R. and Golub, A., 2007. GTAP-E: A revised energy-environmental version of the GTAP model. GTAP Research Memorandum, 15.

McKibbin, W.J., Lee, J.W. and Cheong, I., 2004. A dynamic analysis of the Korea–Japan free trade area: simulations with the G‐cubed Asia–Pacific model. International Economic Journal, 18(1), pp.3-32.

Melitz, M. J., 2003. The Impact of Trade on Intra-Industry Reallocation and Aggregate Industry Productivity. The Econometrics Society, pp. Vol. 71, No. 6.

Melnikov, N.B., Gruzdev, A.P., Dalton, M.G., Weitzel, M. and O’Neill, B.C., 2021. Parallel extended path method for solving perfect foresight models. Computational Economics, 58(2), pp.517-534.

Morley, S.A., Piñeiro, V. and Robinson, S., 2011. A dynamic computable general equilibrium model with working capital for Honduras. Internat. Food Policy Research Inst.

Narayanan, B.G., Hertel, T.W. and Horridge, J.M., 2010. Disaggregated data and trade policy analysis: The value of linking partial and general equilibrium models. Economic Modelling, 27(3), pp.755-766.

O’RYAN, R., De Miguel, C.J., Miller, S. and Pereira, M., 2011. The Socioeconomic and environmental effects of free trade agreements: a dynamic CGE analysis for Chile. Environment and Development Economics, 16(3), pp.305-327.

Peters, J.C., 2016. GTAP-E-Power: an electricity-detailed economy-wide model. Journal of Global Economic Analysis, 1(2), pp.156-187.

US EPA, 2017. SAB Advice on the Use of Economy-Wide Models in Evaluating the Social Costs, Benefits, and Economic Impacts of Air Regulations, Washington D.C: Office of the Administrator EPA Science Advisory Board.

Van Ha, P. & Kompas, T., 2016. “Solving intertemporal CGE models in parallel using a singly bordered block diagonal ordering technique,” Economic Modelling, Elsevier, vol. 52(PA), pages 3-12.

Yu, W., Elleby, C., Lind, K.M.H. and Thomsen, M.N., 2017. Modeling the potential impacts of 2 BREXIT scenarios on the Danish agricultural sectors. Department of Food and Resource Economics, University of Copenhagen.

1. Forecasting with a CGE Model. Does it work? (PDF, 4,942KB) Monash Centre for Policy Studies G-197, 2009 ↩

2. https://www.usitc.gov/publications/332/pub5199.pdf ↩