This chapter provides a review of the Federal Planning Bureau’s analytical models. It includes a technical assessment of the appropriateness and comprehensiveness of the Bureau’s economic models and workflows.
OECD Review of the Belgian Federal Planning Bureau
![](/adobe/dynamicmedia/deliver/dm-aid--bc6b3eac-0e5e-4c97-8d7b-7b53e9103d7a/19.png?quality=80&preferwebp=true)
2. Review of the Federal Planning Bureau’s analytical models
Abstract
2.1. Introduction
The objective of the review of analytical models is to help the FPB increase the quality of its work related to the ex ante assessment of reforms envisaged in the future by the federal and regional governments in areas such as pensions, taxation, the labour market, energy, and investment, particularly those linked to Country Specific Recommendations from the Council of the European Union.
The technical assessment looks at the appropriateness of the Bureau’s economic models across seven academic and practical criteria that an independent fiscal institution should consider when developing tools to deliver its mandate: (1) Theory; (2) Accuracy; (3) Communication; (4) Transparency; (5) Proportionality; (6) Sustainability; and (7) Precedent.
2.2. The OECD’s model assessment framework for IFIs
The review team assessed the models of the FPB according to the assessment framework for IFIs developed by the OECD’s Directorate for Public Governance. That framework answers the question: Are the institution’s models comprehensive and appropriate for delivering its mandate?
To answer the question, the review team identified the needs of the Bureau to deliver its mandate and the constraints it faces in fulfilling these. An assessment was then made regarding whether the suite of models that the Bureau has developed meets those needs given its constraints. Finally, each of the Bureau’s current models are reviewed individually and in depth to determine their individual appropriateness according to seven academic and practical considerations summarised in Table 2.1.
Table 2.1. Model assessment criteria
Theory |
Does peer-reviewed literature support (or not provide a strong argument against) this model for the analysis, given the context and available data? |
Accuracy |
Is this model likely to give the most accurate results (or avoid the most systematic bias) if applied to this problem? |
Communication |
Can the model’s outputs provide a coherent and intuitive narrative to stakeholders? |
Transparency |
Can the model’s methodology and assumptions be provided to the Bureau’s stakeholders in a manner that will satisfy its requirements for transparency and accountability? Are they doing so now? |
Proportionality |
Is the level of effort proportional to the overall importance of the model in terms of the mandate and in the context of the Bureau’s resources? |
Sustainability |
Does the model require a level of resources and expertise that is appropriate to expect from the IFI’s analysts and hiring pool to avoid analytical disruptions from staff turnover? |
Precedent |
Is the approach used widely at other IFIs and public finance institutions? |
Some criteria are complementary, while others conflict. For example, a structural model grounded in economic theory may score highly in its ability to provide an intuitive narrative to stakeholders but may have higher forecast errors than a simple univariate time series model that relies only on its own history. Analysts at IFIs must consider these trade-offs and strike a balance when choosing models. For this reason, it is not possible to offer a total score or pronouncement on whether a model is the best tool for the analysis.
Instead, the assessment criteria is used to form an opinion on whether the chosen tool is appropriate or inappropriate for delivering the Bureau’s mandate in the country’s context — that is, whether any models or analytical decisions are not currently suited to their purpose, fail to advance the Bureau’s mandate, or do not adhere to the OECD Principles. If a model is assessed as appropriate but yet there are recommendations to bring it in-line with best practices, a qualified opinion may be issued (Table 2.2).
Table 2.2. Overall opinion of the review team
Appropriate |
The model meets industry standards (according to benchmarks from the OECD Working Party of Parliamentary Budget Officials and Independent Fiscal Institutions) and no further action is recommended. |
Appropriate, qualified opinion |
The model is appropriate for delivering the IFI’s mandate and meets industry standards, but there are areas in which the IFI should invest to improve it. |
Adverse opinion |
The model is not suited to the task and should be changed as soon as possible. |
This review is not a line-by-line code audit, nor does it undertake out-of-sample validation of alternative specifications. Instead, it seeks to identify any analytical gaps or areas where the Bureau should invest in broadening or deepening models.
External model reviews are an important element of an IFI’s accountability mechanisms and help reassure stakeholders of the quality of the IFI’s work. However, macro-fiscal forecasting is above all a human process that relies on considerable judgment — no two analysts with the same model will produce the same results. A periodic external assessment cannot take the place of an IFI’s other legislated channels of accountability. For the Bureau, this is formal scrutiny by the Federal Parliament and its ongoing dialogue with academics, peer institutions and the public.
2.3. Identifying the modelling needs and constraints of the Bureau
IFIs fall across a spectrum of roles and responsibilities. The assessment framework must be adapted for the needs of an IFI’s institutional arrangements — that is, the functions defined by its primary and secondary governing legislation, memorandums with government agencies and the discretionary operating guidelines it sets for itself. The framework must also consider the constraints of the Bureau — its resources and the economic and fiscal data available to it, which will drive model selection.
2.3.1. The Bureau’s modelling needs
The Bureau’s main responsibilities and modelling needs are laid out in an exceptionally complex array of laws. New laws in Belgium that have a requirement for monitoring or technical expertise tend to name the Federal Planning Bureau as the institution responsible for it either directly, or by serving as the technical secretariat for another responsible body. Table 2.3 lists some of the Bureau’s main responsibilities named in legislation. The Bureau also has responsibilities arising from established practices and agreements of various formality with government departments.
Table 2.3. The main responsibilities of the Federal Planning Bureau named in legislation
Law |
Responsibility |
---|---|
Law of 21 December 1994 on social and miscellaneous provisions giving |
Preparing the economic forecasts for the federal budget, drawing up the five-yearly input-output tables and other satellite accounts for Belgium, along with the overarching mandate to assess and forecast socio-economic and environmental developments with a view to improving their rationality, efficiency, and transparency. |
Law of 10 April 2014 that implemented Directive 2011/85/EU requiring multi-annal budgets |
Serving as the arm’s length independent body responsible for producing the macroeconomic forecasts underlying the budget to comply with the enhanced budget co-ordination and surveillance framework in Regulation (EU) No 473/2013 of the “Two Pack” of reforms. |
Law of 5 May 1997 relating to the coordination of the federal sustainable development policy |
Preparing an ex ante and ex post monitoring report consisting of indicators and scenarios to support the newly created federal plan for sustainable development to achieve domestic, international, and European commitments, and placing a representative of the Bureau on the Interdepartmental Commission for Sustainable Development as an observer. |
The Law of 5 September 2001 guaranteeing a continuous reduction of public debt and creating an Ageing Fund |
Serving as the secretariat of the Study Committee on Ageing, of which the Vice-Chair and one member are from the Federal Planning Bureau. |
The Royal Decree of 14 November 2003 implementing the law of 28 April 2003 |
Supplying the mortality tables and demographic studies for certain calculations related to annuity payments. |
The Programme Law of 23 December 2009 |
Developing and maintaining a database of transport indicators and satellite statistical accounts for the Ministry of Mobility and Transport and to carry out transport simulations with impact analysis and policy analyses on request and in consultation with the Ministry of Mobility and Transport. |
The Law of 8 January 2012 amending the Act of 29 April 1999 on the organisation of the electricity market |
Collaborating with the Directorate General for Energy to draw up a report on the monitoring of the security of the supply of energy every two years. Note – this law is currently under review. |
The Law of 25 November 2018 establishing the National Productivity Council |
Nominating two of its members to the National Productivity Council and contributing to the meetings and reports on the topics of diagnosing and analysing developments in productivity and competitiveness, associated challenges, and the consequences of policy options on productivity and competitiveness. |
The law of 21 May 2015 establishing a National Pensions Committee, a Centre of Expertise, and an Academic Council |
Serving as the secretariat of a support committee for the Centre of Expertise and appointing a representative to sit on the committee. The Centre of Expertise on pensions is responsible for grouping all the knowledge on pensions from various administrations, public establishments, and public interest organisation. |
The Law of 22 May 2014 |
Costing the election manifestos of political parties, amended 2018 to restrict requests to a minimum 3 and maximum 5 priorities and to political parties represented in the House of Representatives. The analysis is to include the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. |
The Law of 22 May 2014 amending the Civil Code |
Supplying the mortality tables for calculating the value of the rights of surviving spouses to enjoy a property. |
The Law of 15 May 2014 implementing the Compact for Competitiveness, Employment and Recovery |
Calculating increases in wage limits and other social security contribution parameters for various stakeholder decisions and negotiations. |
The Law of 14 March 2014 amending the FPB’s 1994 legislation |
Calculating a set of additional indicators for measuring quality of life, human development, social progress, and the sustainability of our economy. |
2.3.2. Identifying the constraints of the Bureau
As illustrated previously in Figure 1.7 in Chapter 1, the Bureau has one of the largest teams of analytical staff among European IFIs. However, it also has among the most tasks and the most diverse responsibilities among IFIs.
The Bureau is generally able to attract staff with the required expertise and the pool of high-calibre analysts in Belgium is large, including a large pool of Belgian PhD economists and PhD economists from other European countries with the right to work in Belgium. That said, there are some constraints that limit the Bureau’s ability to find the expertise for its modelling needs:
There are language constraints, with the Bureau working primarily in French and Dutch.
Competition for analysts in Brussels is fierce among the many domestic and international institutions, governments and think tanks.
The tendency for the Bureau’s analysts to remain in the same position on the same model for many years has lowered its attractiveness to graduates and early-career staff.
The Bureau has enviable access to data, both through legislation for its participation in the National Statistics Institute and membership on numerous official commissions, councils, and committees, as well as through well-established informal peer-to-peer relationships.
However, the Bureau does face some challenges relating to data access:
GDPR compliance mean that some of the Bureau’s access to data is being increasingly questioned, delayed, or even withdrawn.
The Bureau’s data sharing arrangements with regions — which retain full autonomy over agreeing to supply data in many areas — is a perennial sticking point.
2.4. Assessing the comprehensiveness of the model suite and effectiveness of its analytical workflow
2.4.1. Comprehensiveness of the FPB’s model suite
The Bureau’s responsibilities are summarised under five areas and mapped to its current model suite in Table 2.4. For providing the macroeconomic forecast for the budget, the Bureau takes its quarterly MODTRIM macroeconometric model and splines it with its medium-run annual forecasting and policy model HERMES. The Bureau uses monthly monitoring, expert judgment and smoothing to arrive at the most recent quarters for which national accounts data is not yet available and current quarters. Most peer IFIs rely either on the same quarterly macroeconometric model for the short term and medium term or extend it from a nowcasting model for the current and subsequent two quarters. The Bureau’s use of an annual model for outer years of the medium-term forecast is somewhat unique.
For financial and economic policy costing and analysis of the taxes and transfer system, the Bureau relies primarily on HERMES and microsimulation models. The Bureau has quickly caught up to the CPB Netherlands Bureau for Economic Policy Analysis in pursuing a form of election costing that focuses on economic impact assessments — the effects of policies on employment, productivity, growth, inflation, and other macro aggregates. This reflects the Bureau’s history and its staff skillsets.
For sectoral modelling for government ministries, the Bureau has its PLANET model to support the Federal Public Service Mobility and Transport and CRYSTAL SUPER GRID to support its work on energy modelling for various stakeholders such as the climate team within the Ministry of Health. For the Bureau’s work monitoring the Recovery and Resilience Plan, it uses the Belgian-adapted QUEST III R&D model for the long term, the framework most common among EU analysts and promoted by the European Commission, and the HERMES model for the short-medium term. The structural studies team is also working on a new tool for structural reform analysis, the DynEMIte DSGE model.
For long-run fiscal sustainability analysis, the Bureau has developed MALTESE, a tool for projecting social expenditures over the long-run. The Bureau does not regularly publish summary statistics of sustainability such as fiscal gap calculations.1
For its statistical compilation, the Bureau relies on its partners in the statistics framework—particularly those that comprise the National Accounts Institute—to procure the data it needs, which is then processed (mainly) in its Python-LArray platform. Stakeholders are satisfied with the largely mechanical compilation and dissemination of statistics that the Bureau offers as a service provider.
Overall, the Bureau has a broad and diverse range of tools at its disposal to cover an area of vast swath of policy analysis both broadly and in-depth. While there is a risk that in defining an institution’s scope too broadly it loses focus, the Bureau’s stakeholders universally praised the advantages of having all the Bureau’s diverse modelling expertise under one roof. The review team noted some areas of the overall workflow that set the Bureau’s model suite apart from its peers:
The Bureau’s models are generally more sophisticated than those of peers. This partially reflects the age of the Bureau and the experience of its analysts, many of whom have been there since the modern Bureau’s formative years. However, sophistication is not always better. Models are a tool to help think through a problem and tell a story. When the development and deployment of sophisticated models becomes the goal unto itself, it may distract from timely and responsive analysis that may not be as elaborate but could better fulfil the Bureau’s purpose of informing stakeholders when it matters most during the policy process. Although stakeholders are not looking for quick and dirty analysis when they approach the Bureau with a question, there is a balance to be found between providing a timely answer and providing an answer based on sophisticated modelling.
The Bureau has considerably more resources devoted to microsimulation than its peers. This is largely because they can: they have far better access to administration data than peers and Belgium has a wealth of interesting public microdata sets. However, the availability of microsimulation options can be a crutch and come at the expense of analytical options that have a weaker footing in administrative micro-data but may sometimes be more informative.
Table 2.4. Models and methodologies corresponding to mandated responsibilities
Macroeconomic forecasts and medium-term economic and fiscal outlook |
m = missing u = underserved ✓ = comprehensive |
||||
---|---|---|---|---|---|
Task 1: Macroeconomic forecasting |
MODTRIM, HERMES, HERMREG |
u |
|||
Task 2: Fiscal forecasting |
HERMES, PROMES |
✓ |
|||
Task 3: Estimating potential GDP and the business cycle (including long-run potential GDP projections) |
Output gap module (EC method), HERMES, SBS3 (long-run potential GDP) |
u |
|||
Task 4: Forming a view of the cyclical and structural budget balance |
HERMES |
✓ |
|||
Financial and economic policy costing and analysis of taxes and transfer system |
|||||
Task 5: Costing the first-round financial impact of polices1 |
Ad hoc models, EXPEDITION, PROMES |
✓ |
|||
Task 6: Costing the second- round budgetary and economic impact of policies |
HERMES, HERMREG, |
✓ |
|||
Task 7: Preparing policy simulation and scenario analysis |
HERMES, HERMREG, EXPEDITION (distributional impact of policies) |
✓ |
|||
Sectoral modelling for government ministries |
|||||
Task 8: Modelling energy prices and energy distribution |
CRYSTAL SUPER GRID |
✓ |
|||
Task 9: Forecasting and analysis of emissions and climate change |
PLANET |
u |
|||
Task 10: Modelling transportation and freight demand |
PLANET |
u |
|||
Task 11: Monitoring the Recovery and Resilience Plan |
QUEST III R&D, DynEMIte |
u |
|||
Long-run fiscal sustainability analysis |
|||||
Task 12: Population projections (including mortality tables) |
DEMO |
✓ |
|||
Task 13: Projecting the long-term trajectory of public debt and assessing long-term fiscal sustainability |
MALTESE |
✓ |
|||
Task 14: Modelling health care spending |
Module (without specific name) included in MALTESE |
✓ |
|||
Task 15: Modelling the long-term sustainability of national pensions and social spending |
MALTESE, MIDAS |
✓ |
|||
Statistical compilation and dissemination |
|||||
Task 16: Compiling input-output tables and environmental economic accounts |
Statistical compilation tool (without specific name) |
✓ |
|||
Task 17: Compiling indexes to measure quality of life |
Statistical compilation tool (without specific name) |
✓ |
1. Note – this is not the core business of the Bureau and is instead usually done by government ministries. The exception is in relation to election platform costing.
Potential gaps in the model suite
The review team indicated several areas in Table 2.4 as underserved according to their modelling needs or in comparison to the practices of other peer IFIs with similar mandates. Not all underserved areas can be addressed (for example, if missing data is irresolvable).
Task 1: Macroeconomic forecasting
Nowcasting. Many peer IFIs have adopted nowcasting models that use dynamic factor analysis or principal component analysis to statistically assess high-frequency data (monthly, daily, or continuous) to arrive at the recent past (to fill the lag in national accounts publication), the current quarter, and the next two quarters. For example, the Independent Authority for Fiscal Responsibility in Spain (AIReF) uses its MIPRed dynamic factor model at the monthly frequency to determine the concurrent two quarters.
Task 3: Estimating potential GDP and the business cycle (including long-run potential GDP projections)
Contribution to the business cycle debate. The Bureau’s analysis of the business cycle—particularly its estimates of potential output and the output gap—does not receive the same attention or carry as high a modelling priority as many of its peers. This is partly because the structural budget balance in the context of the EU Stability Programme has not yet become as heated a national debate in Belgium as elsewhere. However, it cannot be taken for granted that it will not be an issue in the future and the Bureau would be well-advised to get ahead of it.
Other institutions have made valuable contributions to the business cycle debate using two approaches that would complement the Bureau’s existing model work: (1) Preparing simple, intuitive visual heat maps that assess specific industries and regions and how they are performing relative to their trend, for example in Finland, Latvia, Estonia and others from the Baltic-Nordic network, along with Ireland; (2) Preparing several alternative projections of actual and potential GDP using different model types or specifications, either as a sense check of the primary forecasting model, or to be averaged for their published outlook in a suite modelling approach, as in the case of Ireland. The Bureau accomplishes some sense checks on the model results; however, it could be more systematic and provide greater discussion surrounding the different results and how they have been reconciled.
Task 5: Costing the financial impact of polices
Ad hoc financial models and satellite structural tax and transfer models. While the Bureau does some financial cost assessments off model on an ad hoc basis, it primarily views policy costing through a macroeconomic lens—that is, working out the many ways that a policy could potentially affect the macroeconomy, such as output, inflation, wages, unemployment, household disposable income, and purchasing power. That perspective is guided by the Bureau’s traditional role in economic research. However, the new costing mandate requires the Bureau to be much more focused on the financial and accounting elements of new policies. In focusing on the macroeconomic implications of policies, they may miss important financial details important in getting the budgetary impact of measures correct and making them useful for decisions makers. These include aspects such as administration costs, accruals and cash considerations for financial statements, take-up or noncompliance considerations, and base erosion and planning or evasion, which should all be incorporated in a cost estimate or presented as supplementary analysis.
As Belgium looks to improve its public finances over the medium term, the provision of rich financial information on policies will be useful for stakeholders to undertake effectiveness evaluations, impact assessments, and ex post audits. Other institutions provide simple back-of-the-envelope arithmetic explanations underlying cost estimates such as the number of taxpayers affected, or the number of benefits recipients and average payment, so that stakeholders understand the moving parts underlying the results and can approximately replicate them.
Behavioural adjustments. The Bureau’s approach to costing means the analysis is presented without adjusting for likely behavioural responses, giving rise to systemic bias in the Bureau’s results. Other institutions invest more in complementing the initial results of microsimulation models with top-down spreadsheet financial models that adjust the results for behavioural assumptions derived from the academic literature or empirical assessments of similar policy changes in their own country’s past or in other jurisdictions.
Tools for evaluating the impact of taxes on income from wealth. The Bureau’s tools for assessing the impact of personal income tax measures commonly exclude any income from capital sources and the surrounding costs and economic implications.
Task 9: Forecasting and analysis of emissions and climate change
Environment and climate modelling. The Bureau’s models for monitoring efforts to reduce emissions and act on climate change have fallen behind its leading-edge approaches to evaluating macroeconomic issues and are undeveloped compared to some peer institutions like the Danish Economic Councils, the CPB Netherlands Bureau for Economic Policy Analysis, and the Parliamentary Budget Officer of Canada. They have begun on a work programme to address these gaps and are developing an environmental CGE model.
Task 10: Modelling transportation and freight demand
Freight transportation modelling. The Bureau has not found a sophisticated solution to model freight transportation, as required to fulfil its transport modelling mandate. This is largely owing to data gaps.
Task 11: Monitoring the Recovery and Resilience Plan
Structural reform assessments. The Bureau’s solutions for assessing the Recovery and Resilience Plan remain under development. This is an issue common across institutions in the European Union and elsewhere, where there are no easy fixes. The Bureau’s work programme for developing the new DynEMIte tool may make progress toward this goal.
2.4.2. The FPB’s analytical workflows
The FPB has a diverse range of modelling responsibilities and uses a wide array of software packages for managing it. It is a mixture of proprietary packages like Stata, SAS, Gams, Matlab and Excel; open source languages (and their usual libraries) such as Python and R; and in-house developed software packages or libraries like IODE, LIAM2 or LArray, which are further described below.
For econometric models, the econometrics platform IODE2 is the backbone of the Bureau’s workflow to co-ordinate its data resources, inputs, and outputs across models, teams, and projects. It is a powerful software package for statistical analysis and model solving.
IODE was developed in-house by the IT Unit of the FPB. It assists analysts by streamlining activities such as (1) Automating data retrieval from databases, (2) importing and exporting series between the office’s open-source and licensed software packages like Python and the LArray library, Stata, R, Excel, (3) documenting databases, (4) writing and estimating equations, (5) facilitation scenario simulations, and (6) generating graphs and tables, among other helpful functions like scripting.
While it has many benefits and is fast and efficient in keyboard navigation and processing, such an in-house software solution is unique among IFIs. It is largely a carry-over from an earlier computing workflow — it was developed as the replacement for the Bureau’s mainframe computer econometric software in the 1980s.
The software is written in C and C++ and requires dedicated specialists to maintain, refine and add functionality. New techniques must be translated into IODE rather than simply being applied as imported libraries from R or Python that outside researchers often publish alongside their results (although such files can be passed back and forth to IODE).
The look and feel of the IODE GUI divides users internally, with some having an affinity and others wanting a more modern solution. Some outside stakeholders also see it as outdated.
Overall, IODE is observed to play a crucial role in the Bureau’s workflows. Nonetheless, the Bureau’s ongoing commitment to IODE should be reviewed with an eye to converting it over the long-term to a more modern software solution with greater penetration in the field of economics, along with the gradual conversion of models specific to it, such as HERMES. Doing so will have several benefits:
The ability to quickly incorporate leading-edge techniques from outside academic working papers and other researchers, that are increasingly published open-source in Python and R.
The ability to leverage the tools coming online to assist code drafting, such as AI “co-pilot” programmes that autocomplete code based on code comments which is improving the productivity of researchers by leaps and bounds.
The ability to participate in, and benefit from, larger modelling communities providing support for choices like Python and R.
LIAM2 is an open-source software package developed in Python to help economists develop microsimulation models. The MIDAS model is developed in LIAM2. The toolbox is made as generic as possible so that it can be used to develop almost any microsimulation model as long as it uses cross-sectional ageing, i.e. all individuals are simulated at the same time for one period, then for the next period, etc. The goal of the software is to let modellers concentrate on what is strictly specific to their model without having to worry about the technical details. It was made available for free to outside researchers to build a community to reduce the development costs of microsimulation modelling.
LArray (which stands for Labelled Array) is an open-source Python library and GUI for analysing multi-dimensional matrices and creating models with them. It is used for many models of the Bureau (demographic projections, MALTESE). The most important feature is to access data via meaningful labels to make models more readable and easier to maintain, but it also helps modelers automate large parts of their workflow from importing data in various formats and cleaning it to generating data reports, charts or even dashboards.
All in all, the Bureau is considered to be further ahead than many of its peers in adopting collaborative open-source software in several areas of its model suite.
2.5. Assessing the appropriateness of the Bureau’s individual models
Through discussions with the Bureau the review team has identified 12 models in the Bureau’s primary toolset that are currently in use and appropriate to review individually, along with three that are in development for the future.
Models in use:
HERMES |
Forecasts the short- to medium-term macroeconomic outlook for gross domestic product and its components, prices and incomes, employment and unemployment, energy consumption and greenhouse gases, as well as the public finances at an annual frequency. It also calculates the macroeconomic impacts of policy measures and their effects on the budget balance and public debt. |
HERMREG |
Same functions as HERMES but for Belgium’s regions. Two model versions: one for forecasting (top-down version), one for impact analysis (bottom-up version). Macroeconomic aggregates at the national level are constrained to HERMES in the top-down version. |
QUEST III R&D |
Simulate the long-term macroeconomic effects of structural reform measures, with special coverage of measures supporting research and development, market functioning and public investment, as well as tax shifts between labour income, capital income and consumption. |
EXPEDITION |
Calculates the direct impact of measures on the distribution of household disposable income, presented by income decile or household characteristics. |
TYPECAST |
Calculates the effect of measures on the financial incentive to work (the change in a worker’s decision to move between unemployment and employment) arising from changes to household disposable income. |
HINT |
Assesses the redistributive effects of measures that affect the prices of goods and services, presented by income quartile or household composition. |
MIDAS |
Simulates the life spans of individuals for the years between 2012 and 20702070 to assess the adequacy of pensions, replacement ratio, inequality, and poverty risk indicators of the elderly. |
MODTRIM |
Forecasts the macroeconomic outlook in the short run using national accounts data in the quarterly frequency (versus HERMES at an annual frequency). |
PROMES |
Used to compute detailed health care expenditure in the medium term. First operational version in 2019. |
MALTESE |
Estimates long-term implications of budget measures, especially those related to public pensions. The model has been used to compute Belgian pension projections published in the European “Ageing Report”, is used annually to compute the total social expenditure projection in the Report produced by the Study Committee on Ageing and for the budgetary impact of social benefit policy measures. |
PLANET |
Calculates the effects of changes in economic activity and policy reforms on demand for personal and freight transportation, including congestion, the environment and welfare. |
CRYSTAL SUPER GRID |
Assess the impact of different assumptions and policy reforms on the prices and distribution of the electrical system, particularly the long-run supply security, sustainability, and affordability. |
Models in development:
LASER |
Assesses the expected impact on labour supply of measures influencing household disposable income. |
DYnaMITe |
DSGE model based on QUEST III R&D, incorporating multiple industries and knowledge spillovers so that propagation of (productivity and other structural reform) shocks through the production and innovation networks can be modelled. |
Environmental CGE model |
Standard multi-sector recursive dynamic model, covering Belgium and its regions. Currently, special attention is paid to the interaction between energy inputs and heterogeneous labour demand, to the modelling of labour markets and to empirical underpinning. Linkage with microsimulation models is explored. |
For in-depth assessments of each model, interviews were held with the relevant modelling team along with a review of work papers, specifications of equations, and in some cases the model code to scrutinise the suitability according to the framework described above. The remainder of this sections provides the results of the individual model assessments.
2.5.1. HERMES
HERMES (Harmonised Econometric Research for Modelling Economic Systems) was the outcome of a 1981 project proposed by the Commission of the European Communities (Directorate-General for Science, Research and Development) to create an econometric model that could simulate alternative assumptions about the world environment and economic and energy policies. It was motivated by the contemporary energy crisis and a realisation that energy distribution, prices, and policy would spread across borders (Donni, Valette and Zagame, 1993[1]).
The Federal Planning Bureau was a key member of the HERMES Club of 12 institutions in 12 countries (Belgium, Denmark, Spain, France, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Germany, and the United Kingdom) that was intended to work closely through co-ordination at the European level, although the goal was never fully realised and the “H” in HERMES is a misnomer (Donni, Valette and Zagame, 1993[1]).
The first version of the model in Belgium was finalised in the late eighties after four years of development, with the first results published on the topic of the impact of economic activity on the environment in 1989 and the use of taxes to reduce CO2 emissions in 1990.
Since then, most other HERMES Club members have stopped supporting development of their domestic models, with a few exceptions such as Ireland. The FPB has carried the torch and Belgium’s model has been regularly updated and significant improvements have been made over the years.
Table 2.5. Overview and evaluation of HERMES
Name |
HERMES (Harmonised Econometric Research for Modelling Economic Systems) |
Type |
Large-scale macroeconometric model |
Description |
Developed in-house as a conventional macroeconometric model consisting of a new Keynesian final demand-driven system of equations estimated econometrically for consumption, investment, exports, imports, inventories. Private consumption consists of several stylised products, the labour market has several categories of workers, and there is a detailed public finance block reflecting the complexity of the Belgian institutional environment. Demand is linked to supply via an industrial exchange matrix and utilisation rates and corporate behavioural functions are based on installed production capacities, disaggregated into industries for sectoral analysis. The model treats energy as a special production factor introduced through an energy sub-model determining the energy demand requirements for different products for private consumption. It is developed and maintained by the General Directorate (ADDG). |
Mandate justification |
Law of 21 December 1994 on social and miscellaneous provisions giving the Bureau the responsibility of preparing the economic forecasts for the federal budget, as well as the following analytical responsibilities: Assessing the medium-term impact of legislation Calculating the macroeconomic impacts of policies Calculating the financial impact of polices on the budget balance and the public debt, of the measures. Calculating the impact of economic policy measures Calculating the impact of external shocks Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, including the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. Law of 10 April 2014 that implemented Directive 2011/85/EU on requirements for multi-annal budgetary frameworks of Member States and pre-emptively complied with the enhanced budget co-ordination and surveillance framework in Regulation (EU) No 473/2013 of the “Two Pack” of reforms requiring that a Member State’s budget be based on macroeconomic forecasts produced or endorsed by an independent body. |
Outputs |
Short- to medium-term expenditure-side forecasts of gross domestic product and its components, prices and incomes, inflation, (un)employment, energy and greenhouse gases and detailed public finance tables typically for five or six years (for example, the 2019 election platform costings show HERMES results for 2020 to 2024). |
Working paper |
How the HERMES model works - Description using variants (Bassilière, Dobbelaere and Vanhorebeek, 2018[2]) Description and use of HERMES: Document drafted in the framework of preparing for the 2019 costing of electoral programmes (Bassilière et al., 2018[3]) A new version of the HERMES model – HERMES III (Bassilière et al., 2013[4]) |
Major reports |
The FPB uses HERMES to forecast and publish its economic outlook for Belgium over a six-year horizon once a year, typically in June. Since 2014, the Bureau has also produced a preliminary version of the outlook in February or March for the preparation of the Stability Programme and the National Reform Programme, which must be submitted to the European Commission in April. Macroeconomic and fiscal effects of the draft National Recovery and Resilience Plan - Report to the Secretary of State for Recovery and Strategic Investments (Federal Planning Bureau, 2021[5]) HERMES was used for the short run effects, while QUEST was used for the long run effects. |
Key judgments |
Technological progress is exogenous. Government measures have no direct influence on total factor productivity. Labour supply is exogenous and does not react to economic fluctuations or to economic policy measures (any increase in the demand for labour can be satisfied by the existing supply). The model does not distinguish between different types of households and so distributional effects of policy measures cannot be assessed. Cross-border purchases are not influenced by excise duties or VAT rates. Foreign investment does not respond to changes affecting the relative attractiveness of Belgium. Assumptions relating to oil prices, exchange rates and interest rates for the first two years are based on quotations on the futures markets. From the third year, the Bureau imposes assumptions for medium-term equilibrium targets. |
Software |
The Bureau ’s econometric platform IODE (Intégrateur d'outils de développement économétrique). |
Theory and context |
Good. Macroeconometric modelling is suited to the Bureau’s twin goals of capturing dynamics with enough structure to trace effects of policies and shocks. Large-scale macroeconometric models are having a resurgence in the literature following a period of falling out of favour in preference to DSGE models. |
Accuracy |
Good. Strong theoretical underpinnings combined with empirically estimated equations pinned on medium-run equilibrium conditions (closing of the output gap, use of levels and dynamics via error correction models, all likely to improve upon naïve statistical forecast benchmarks over the medium run). |
Communication |
Good. Macroeconometric modelling produces coherent, intuitive narratives in-line with economic theory. Coefficients and directions are meaningful. |
Transparency |
Good. Working papers provided with equations, complete with parameters and statistical test tables. Detailed data with sources. Could be replicated externally by an experienced external analyst. Judgment plays a significant role in specification and introduces some opacity. IODE is provided for public download; however, the HERMES code is not proactively shared. |
Proportionality |
Good. Team of six analysts, along with two analysts on HERMREG. Commensurate with the considerable weight of medium-term forecasting in the Bureau’s mandate and its connections across modelling areas. |
Sustainability |
Fair. Once developed, experienced analysts with a degree in economics or a numerate field could support and run the model. Maintenance and development are likely to require a PhD economist or analyst with an MSc-level background and considerable experience. Implementation in IODE requires a learning curve and a modelling background specific to the Bureau. |
Precedent |
Good. HERMES was once developed for several countries including Spain, Ireland, Dutch, French, German, Italian as an initiative of the Commission of the European Communities (Directorate-General for Science, Research and Development). Other countries have shifted away or stopped maintaining their models; however, some institutions continue to refine and develop their domestic HERMES models, for example Ireland’s Economic and Social Research Institute (ESRI). |
Opinion |
Appropriate. HERMES is a well-developed and well-supported staple of the macro-fiscal forecast and policy assessment framework in Belgium. The Bureau should review the theoretical basis for using futures markets quotations as short-run forecasts for oil prices, exchange rates and interest rates—a practice which is common but has a poor theoretical justification and poor forecasting performance. The current economic environment and energy crisis have created a climate akin to that during the initial HERMES initiative and the European Union has grown even more open and co-ordinated with more harmonised national accounts data. The Bureau should consider leading a movement to reboot the HERMES project with the FPB at the centre, convert to open source software like Python hosted on GITHUB, and work with other Belgian and international institutions to co-operate on the development and co-ordination of HERMES projects. To do so it could look for partners and stakeholders among the EU institutions or member states for resources. |
2.5.2. HERMREG
HERMREG is the regional companion of the national model HERMES for estimating regional economic output and its components. A first top-down version (HERMREG 1) was developed to produce regional economic projections in 2006 as a collaborative and jointly-owned owned initiative of the FPB and its three regional counterparts: SV (Statistiek Vlaanderen), IWEPS (Institut Wallon de l’Évaluation, de la Prospective et de la Statistique) and IBSA (Institut Bruxellois de Statistique et d’Analyse). In 2015, the Bureau developed a new bottom-up structure for policy impact assessments by constructing a block of regional equations for sectoral production factors and regionalised demand (private consumption, investments by delivery sector, trade regionalised external demand) relying on interregional input-output tables.
A multi-phase model development programme has since deepened the bottom-up version by improving the household income block, integrating the public finances block of the HERMES model, and improving the link between public finances and public consumption. For 2022 to 2026, the four partners have decided to fund a sixth phase of enhancements extending and improving the short- and medium- term projections of the top-down model, improving impact assessments using the bottom-up model, and enhancing the backend database. The sixth phase will also involve broadening and deepening the regionalising of the energy component of HERMES.
Table 2.6. Overview and evaluation of HERMREG
Name |
HERMREG (HERMES regional) |
Type |
Large-scale macroeconometric model |
Description |
The regional version of the national model consisting of both a top-down and bottom-up approach to estimating regional output and its components. The bottom-up version contains 16 000 equations, of which 500 are econometrically, to cover 14 areas of economic activity in the three regions. The main components of demand—household consumption, investment, international exports, and imports—are error-correction equations combining short-term dynamics and long-term equilibrium conditions. Public expenditure is initiated both federally and regionally, and consumed locally. Data comes from interregional input-output tables compiled and co-ordinated by the FPB and the regional accounts from the National Bank. The HERMREG projection process (using the top-down model) is initiated with the HERMES results in March and work runs until the middle of July. HERMREG is developed and maintained by the General Directorate (ADDG). |
Mandate justification |
Law of 21 December 1994 on social and miscellaneous provisions giving the Bureau the responsibility of preparing the economic forecasts for the federal budget, as well as the following analytical responsibilities: Assessing the medium-term impact of legislation Calculating the macroeconomic impacts of policies Calculating the financial impact of polices on the budget balance and the public debt, of the measures. Calculating the impact of economic policy measures Calculating the impact of external shocks Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, including the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. |
Outputs |
Somewhat less detail than the HERMES output, but for the Brussels capital region, Flanders, and Wallonia. For impact assessments tables of the percentage change in (1) nominal hourly labour costs (market activities), domestic employment, domestic employment (persons), real household disposable income, personal consumption, investment, exports, imports, GDP. |
Working paper |
The HERMREG bottom-up model (Baudewyns and Lutgen, 2022[6]). How the HERMREG bottom-up model works: Description using variants (Baudewyns and Lutgen, 2022[7]). HERMREG: A regionalisation model for Belgium (Hoorelbeke et al., 2007[8]). |
Major reports |
The HERMREG regional economic outlook is published annually in July. Labour cost reduction measures: what effect on employment and public finances in the Brussels Region? (Baudewyns, Dewatripont and Michiels, 2020[9]) Past policy impact assessments include an increase in public investments, , a general reduction in employers' social security contributions, a reduction in social security contributions targeted at low wages, a reduction in personal income tax, an increase in family allowances, a reduction in personal social security contributions, target group reductions in Flanders and the Brussels Region, and activation of unemployment benefits in Wallonia and Brussels. HERMREG was also used for an assessment of both the Recovery and Resilience Plan and Wallonia's economic recovery plan. |
Key judgments |
The active population in the labour market is fixed. Production equals demand. Constrained to aggregate HERMES projection (top-down version). |
Software |
The Bureau’s econometric platform (Intégrateur d'outils de développement économétrique), C programming language. |
Theory and context |
Good. Structural models with dynamics in the form of ECM are suited to the twin goals of tracing the effects of policies and shocks while capturing the time series statistical behaviour to improve forecasting power. Large-scale macroeconometric models well-supported in the literature, fell out of favour, but are having a resurgence as the limitations of DSGE models are better understood. |
Accuracy |
Good. New version of the bottom-up model rigorously tested in fall of 2021. Strong theoretical underpinnings combined with empirically estimated equations pinned on medium-run equilibrium conditions (, use of levels and dynamics via error correction models, all likely to improve upon naïve statistical forecast benchmarks over the medium run. |
Communication |
Good. Macroeconometric modelling produces coherent, intuitive narratives in-line with economic theory. Coefficients and directions are meaningful. |
Transparency |
Good. Working papers (bottom-up model) provided with equations, complete with parameters and statistical test tables. Detailed data and sourced. Could be replicated by an experienced external analyst. Judgment plays a significant role in tuning and combining the models and introduces some opacity. IODE is provided for public download; however, the HERMREG code shared with partner institutions. |
Proportionality |
Good. Team of 2 analysts and the support of the three regional institutions. Suited to the weight of regional analysis in the Bureau’s mandate and commitments to other organisations. |
Sustainability |
Fair. Once developed, experienced analysts with a degree in economics or a numerate field could support and run the models. Maintenance and development are likely to require a PhD economist or analyst with an MSc-level background and equivalent experience specific to the Belgian context and to the niche IODE and C programming environment. Some knowledge sharing procedures and documentation but still some exposure to business continuity problems if the two analysts were to leave. |
Precedent |
Good. As with HERMES, there is a common modelling heritage among institutions within the HERMES Club. Other countries have looked to HERMREG for inspiration in their research, including France, Estonia, Chile. |
Opinion |
Appropriate. The HERMREG model is well-supported financially and analytically by the Bureau and by partner institutions. It is already a leading-edge tool that will be refined and expanded with the model development programme scheduled over 2022 to 2026 and guided by considerable expertise. Regionalising the energy component of HERMES will be an important step in supporting policymakers in facing the current crisis. Sustainability of the model could be improved by converting to a more widely used software environment in the economics community. |
2.5.3. QUEST III R&D
QUEST III R&D is a Dynamic Stochastic General Equilibrium (DSGE) model that the FPB uses to calculate the long-run steady state impact of some structural reforms that affect the productivity of labour and capital. For example, it was used during the 2019 election costing period to assess proposals to invest in research and development, improve market functioning and increase public investment. The QUEST III R&D model was also used to calculate the medium- to long-term impact of Belgium’s draft National Recovery and Resilience Plan (only the investment part of the plan).
The model is the Belgian module of the QUEST III R&D model developed by the European Commission (DG ECFIN), that has been calibrated to Belgium’s national accounts data. Researchers in other EU member countries have similarly received their country-specific module, have updated the calibration and have applied it to a wide range of country-specific policy cases.
The Bureau has begun to look at extending the model to handle more sophisticated modelling of public investments and public support for private-sector investment, among other areas. It is also developing its own in-house DSGE model, which is based on the structure of QUEST III R&D but incorporates multiple industries, intermediate consumption (with input-output linkages between industries) and labour-augmenting semi-endogenous technological growth.
Table 2.7. Overview and evaluation of QUEST III R&D
Name |
QUEST III R&D |
Type |
Smets-Wouters Dynamic Stochastic General Equilibrium (DSGE) model. |
Description |
Belgium-specific DSGE model adapted from a model developed by the European Commission (DG ECFIN) to simulate the medium- to long-term impact of structural reforms (changes to a country’s structural settings, for example in the labour market, product market regulation, that influence the efficiency of the supply side of the economy). There are three regions (Belgium, the euro area, and the rest of the world). The Bureau used it to simulate, e.g. the impact of deregulation of professional services through markup reductions, leading to changes in prices and productivity both in the sector itself and in downstream sectors. Data is from the most recent National Accounts and the Bureau plans to update the model annually. It is maintained by the FPB’s Sectoral Directorate (SDDS). |
Mandate justification |
Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, including the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. During the 2019 election period it was used to simulate the long-term impact of reforms in the field of market functioning, research, and development (R&D), administrative burden and public investment. The Law of 25 November 2018 establishing the National Productivity Council which prescribes that the bureau nominates two of its members to the National Productivity Council and must contribute to the meetings and reports on the topics of diagnosing and analysing developments in productivity and competitiveness, associated challenges, and the consequences of policy options on productivity and competitiveness. Monitoring the draft National Recovery and Resilience Plan. |
Outputs |
The long-term percentage change in a list of macroeconomic variables for the “structural” equilibrium year of 2040. Macroeconomic variables include GDP, personal consumption, government consumption, private investment excluding R&D, private investment in R&D, public investment, exports, imports, the GDP deflator, real wage cost (private sector), labour productivity, employment rate, and lump sum taxes or public debt. |
Working paper |
Description of the QUEST III R&D model (Biatour et al., 2018[10]) |
Major reports |
Economic impact of professional services reform in Belgium – A DSGE simulation (Kegels and Verwerft, 2018[11]) Public investment in Belgium – Current state and economic impact (Biatour et al., 2017[12]) |
Key judgments |
Only one “final” sector. One total factor productivity variable. The knowledge spillover coefficients are calibrated on the basis of patent citations. Rotemberg (1982[13]) quadratic adjustment costs. |
Software |
Dynare pre-processor and library for Matlab and Octave. |
Theory and context |
Good. In the lineage of Smets and Wouters (2003[14]), a seminary and well-scrutinised class of DSGE models that have become industry standard and are a common practical solution for structural reform questions. Industry-specific return to additional investment in R&D in the style of “Innovation networks” from Liu and Ma (2022[15]). |
Accuracy |
Fair. DSGE models are a device to simulate the propagation of shocks through a highly stylised and simplified theoretical and empirically validated framework, but are not intended to be used for forecasting. |
Communication |
Fair. DSGEs are grounded in economic theory and should be able to tell a coherent and consistent economic story of the relationships between variables of interest. However, the complexities of the model in practice make it somewhat of a black box. |
Transparency |
Good. Although initially in the open-source code Octave, the model was transitioned to Matlab for computational speed and the more well-developed tools for publishing and exchanging results with other model teams and colleagues. Nonetheless, the Matlab version is shareable and publishable, and Matlab is accessible with reasonable license fees. It would also be trivial for an outside researcher to convert to Octave. Dynare is freely available. The QUEST framework is available through the community. The Bureau has published several working papers and descriptions with equations. |
Proportionality |
Good. Two researchers committed full time. Three knowledgeable in the area and able to assist. Would like more expertise for specific issues, especially from outside consultants. Structural reform analysis will play an increasingly important role in the Bureau’s work monitoring the Recovery and Resilience Plan and election platform costings. |
Sustainability |
Fair. Maintaining and developing DSGE models generally requires a PhD economist in the field. Once developed, experienced analysts with a degree in economics or a numerate field could run the models but would need support of colleagues or external consultants. Most central finance ministries with a DSGE model retain external academic economists for developing new capabilities. The QUEST modelling community is relatively active, and expertise could be found in Brussels institutions. Little risk to business continuity. |
Precedent |
Good. Smets-Wouters DSGE models are used commonly in central banks. Used across EU governments. QUEST III R&D and its forebearers are used commonly in think tanks and other fiscal institutions. |
Opinion |
Appropriate. DSGE models are what they are: a theory-based solution to think through problems that are difficult to estimate with econometric models. It is appropriate for this use. A number of limitations of QUEST III R&D (e.g. some lack of detail on the supply side) could be revisited as the new DSGE model (DynEMite) is developed in-house. While the new DSGE model will be a welcome tool, it is sensible to also maintain QUEST III R&D for EU policy analysis commitments and to benefit from the relatively large QUEST community. |
2.5.4. EXPEDITION and TYPECAST
EXPEDITION is a static microsimulation model developed in-house for analysis of policy measures related to personal income taxation, social security, and social assistance. Its key output is the impact of measures on disposable income in nominal terms by different household categories. EXPEDITION was developed over the 18 months leading up to the 2019 election platform costing exercise and is based on the EUROMOD platform, through using administrative data in place of EU-SILC (Statistics on Income and Living Conditions), which is the default data source for EUROMOD.
The model covers six policy areas: (1) pensions; (2) allowances payable by the National Employment Office; (3) compensation for sickness and disability; (4) personal income tax, (5) personal social security contributions and deductions from allowances; and (6) social assistance allowances and family allowances.
EXPEDITION is able to assess the effects of policy changes on the full set of households represented in the administrative microdata. By contrast, the TYPECAST module uses EXPEDITION’s analysis of disposable income to simulate the impact of a measure on a selection of specific standard household types that are useful for illustrating policy effects.
The Bureau has considered freezing development of EXPEDITION to switch to BELMOD, a similar project developed with other partners; however, negotiations surrounding the development and maintenance of BELMOD are ongoing and its future is currently uncertain.
Table 2.8. Overview and evaluation of EXPEDITION and TYPECAST
Name |
EXPEDITION (EX-ante simulation of Policy reforms and an Evaluation of their Distributional ImpacT on Income and Other welfare Notions) TYPECAST (Type case simulaTing) |
Type |
Static microsimulation model |
Description |
EXPEDITION was developed in-house to compute the direct impact of tax and social policies on nominal household disposable income. Simulations are based on a representative sample of Belgium’s entire population and allows for assessing distributional effects of policies. The model uses administrative microdata on individuals and their households from the Labour Market and Social Protection Datawarehouse of the Crossroads Bank for Social Security (a collaboration between dozens of institutions involved in social security). It adds data from tax records (IPCAL) and CENSUS (highest level of education). Households are linked together with a register of exact identifiers. Some variables like education are inferred contextually. TYPECAST extends the output of EXPEDITION to calculate the intrinsic and extrinsic impact of a measure on a selection of standard household types (composition of family and place of residence, status in the event of non-work, status in the event of full-time work). The model is developed and maintained by the General Directorate (ADDG). It must be run on-site but the Bureau is investigating a remote access option. |
Mandate justification |
Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, including the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. |
Outputs |
EXPEDITION: Change in taxes and allowances by household member. Baseline and counterfactual for the direct effects of a measure on the distribution of household disposable income, presented by income decile or household characteristics. Change in household disposable income in euros in nominal absolute terms and percentage change by income decile, socioeconomic position (employee, civil servant, pensioner, independent, social assistance recipient, etc.), and household composition. Number and percentage of gaining and losing households in terms of disposable income by decile, socioeconomic position, and household composition. TYPECAST: Baseline and counterfactual for the direct effects on the financial incentive to work, that is, the change in disposable income in the event of a transition from inactivity to full-time employment. |
Working paper |
Description and use of the EXPEDITION model (De Vil et al., 2018[16]) |
Major reports |
EXPEDITION and TYPECAST have been used to compute the impact of measures taken during the pandemic to support disposable income in COVID-19 Crisis: A simulation of the impact of the loss of wages for temporary unemployment in the case of force majeure and the loss of income in the case of bridging rights (Thuy, Van Camp and Vandelannoote, 2020[17]). In addition, EXPEDITION has also been used to analyse the budgetary and distributional effects of the regional child benefit reforms (Nevejan, 2021[18]). Results of the model were presented for each party’s policy proposals, where relevant, on the DC2019 election policy costing portal: https://www.dc2019.be/. For example, proposals to increase the portion of income exempt from tax, change the tax break on company cars, and reform unemployment benefits. |
Key judgments |
Does not capture property income of households so an incomplete picture of the distribution of disposable income. Socio-demographic characteristics of individuals are assumed to be constant. Costing of measures does not consider behavioural responses in allocation decisions of economic agents. No automated adjustment of EXPEDITION for the macroeconomic outlook. Tax year and benefit year align to calendar years, rather than legal years and implementation schedule. |
Software |
EXPEDITION takes its framework from EUROMOD in C+, SAS is used for data manipulation, Stata is used for data processing. TYPECAST is written in Stata. |
Theory and context |
Fair. Microsimulation models transcribe the tax and benefits laws into computational identities linked to survey and administration data and therefore have a direct correspondence to key concepts for assessing the fiscal impact and household impact of changes in tax rates and thresholds and social benefits amounts and qualification criteria. They allow flexible aggregation and cross-tabulation after simulation, unlike macro approaches that determine aggregation in advance. However, not modelling behavioural responses either in the model or with satellite models to refine cost estimates does not take advantage of the literature. TYPECAST is a helpful solution to bring in potential behavioural responses and some of the literature on labour supply effects of changes in taxes and benefits, albeit for stylised examples of specific cases rather than general fiscal cost estimates. |
Accuracy |
Fair. Microsimulation models on administration data are the gold standard for determining the non-behavioural costs of policies. The exclusion of behavioural effects will bias results but can be adjusted in satellite models for costing measures, which is being done to an extent with TYPECAST as supplementary information, albeit not incorporated as an aggregate costing. Cost estimates in years other than the benchmark year are conditioned on the accuracy of the forecast of economic cost drivers and growth factors taken from auxiliary forecast models, and the degree of model detail in microsimulation generally reduces prediction power through the accumulation of errors and biases of individual variables that are amplified (rather than smoothed as in macro approaches). Administrative data should be an improvement in accuracy and granularity over SILC. Satellite models that adjust the results for tax or benefit year effects could improve accuracy (particularly important for the first year of a programme that is introduced as a partial year, or if indexation factors are applied mid-year). The benchmark administration and survey year is considerably out of date and likely to significantly affect the accuracy of results, especially during a period of rapid economic fluctuations and change. |
Communication |
Fair. Communicating the results from microsimulation models is easy and intuitive as the model is a rote translation of the tax code and benefits legislation. The results of TYPECAST are somewhat more of a challenge to illustrate intuitively and the Bureau has struggled with presenting results in a comprehendible way to non-economists. |
Transparency |
Good. A detailed working paper on EXPEDITION has been published. The underlying equations are mechanical identities, and aside from weights to scale results to the population level and economic growth factors to shift the results between years, little estimation and no judgment is directly applied. TYPECAST described only briefly in broader documents but enough information is given on assumptions that a sophisticated analyst outside of the Bureau could approximately reproduce the results. |
Proportionality |
Good. At peak demand during the costing period, six analysts, with four working on child benefits, pensions, PIT, and social assistance; one for data; and one for labour supply. Appropriate during the costing period. Some streamlining could be done by automating the links to the macroeconomic outlook. |
Sustainability |
Good. Microsimulation models require considerable resources to develop and maintain; however, once the framework is built (or borrowed) it requires some expertise to operate but is within the expected tool kit of someone with an undergraduate economics degree. TYPECAST requires some additional expertise in labour market economics and more professional judgment from experience. The Bureau has the documentation in place and enough analysts familiar with the model to ensure business continuity. |
Precedent |
Good. EUROMOD, BELMOD, UK, others in the EU, Canada SPSD/M |
Opinion |
Appropriate, qualified. EXPEDITION and TYPECAST meet standard practices and are appropriate for delivering the FPB’s mandate. In the opinion of the reviewers the Bureau should continue to develop EXPEDITION rather than switching to BELMOD. There are nonetheless several areas in which the IFI should invest to improve its usefulness: (1) Work with data providers to refine the co-ordination process, (2) Improve the model’s mechanical link to HINT to leverage the greater detail that HINT will produce in the next election, which will take place in the context of a sustained period of significant price volatility, (3) Work with STATBEL to compare and contrast its analysis on disposable income, (4) Review construction of the weights and matching to uprate and reweight tax benefit years to bring them forward to at least 2017 or more recent years, (5) Support the main model with satellite models to adjust results for implementation date and policy year, (6) Explore how to present the results of TYPECAST in a more comprehensive and comprehendible way for a broader audience, (7) Explore a link with the LASER labour supply model, (8) Refine the model so it has the capacity to simulate policies in more detail for wealth taxes, personal income taxes, and regional policies. |
2.5.5. HINT
HINT was developed for the 2019 election costing exercise to calculate the impact of policy measures on consumer prices faced by different household types (incomes and family composition). The results of the model complement the results of EXPEDITION to provide a more complete picture of the welfare effects of policies that alter the prices of specific goods or services. For example, a subsidy for public transportation may raise the real disposable income of households in lower income brackets more than wealthier households.
The model traces the effects of price changes both on the standard CPI consumer price basket faced by different households and on Belgium’s “health index” indicator that excludes alcoholic beverages, tobacco, and motor fuels. The latter is used for the indexation of housing rents and certain salaries, social benefits, and pensions.
Table 2.9. Overview and evaluation of HINT
Name |
HINT (Household type INflation Tool) |
Type |
Static accounting model |
Description |
Developed in-house to calculate the impact of price shocks on different household categories (income classes in quartiles and family compositions—single/couple, one child, two children, etc.). Built and maintained by the FPB’s General Directorate (ADDG). Data for consumption patterns comes from the Household Budget Survey (weights of products and services per household type) which is linked to the National Consumer Price Index for 244 product groups corresponding to the 5-digit COICOP aggregation level (United Nations Classification of individual consumption by purpose), which is enriched with a breakdown of the expenditure item 'Restaurants and cafes” into 15 sub-items at the 6-digit level to exclude alcohol for calculating the impact on the health index. To go from the 292 groups of products in the Household Budget Survey to the 244 groups of the CPI, several headings are aggregated, while others are removed (drugs, prostitution, and owner-occupied housing). |
Mandate justification |
Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, including the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups for social security, and of the impact on the environment and transportation. |
Outputs |
The direct redistributive effects of price change measures, presented as the percentage point change in the CPI basket or health index by household income quartile and family type. Results are provided separately from the EXPEDITION model’s assessment of redistributive effects on disposable income. |
Working paper |
No standalone working paper but it is discussed in the appendix of Description and use of the EXPEDITION model (De Vil et al., 2018[16]). |
Major reports |
Results of the model were presented for each party’s policy proposals, where relevant, on the DC2019 election policy costing portal: https://www.dc2019.be/. For example, proposals to reduce the VAT on lenses in eyeglasses and proposals to reduce the VAT on repair services to encourage reduction of waste. |
Key judgments |
The model assumes no changes in consumption behaviour in response to price changes. It is not linked mechanically to other microsimulation models. Only the price changes of goods and services in the basket of the National CPI and health subindex can be modelled. It assumes that every household pays the same price, which does not hold for important policy areas like the social tariff for gas and electricity. |
Software |
IODE (Intégrateur d'outils de développement économétrique) |
Theory and context |
Fair. The tool is appropriate as a practical accounting replica of the specific price subindices faced by different households in the Household Budget Survey. It fulfils its requirements of allowing the Bureau to change prices and compute the resulting mechanical household-level impact. However, the assumption of no changes in consumption patterns in response to price changes does not take advantage of the literature on price elasticities of demand for items in the consumption bundle. |
Accuracy |
Fair. The model is not intended as a forecasting tool; however, it is presented as a prediction of the changes in prices faced by households in response to a policy change. For prediction it is likely to be biased—for example, a policy that raises the costs of a good or service would always have a larger predicted impact than would materialise because households will reduce consumption or substitute for cheaper alternatives. For tracing mechanical changes in consumption pattern identities, it is accurate. |
Communication |
Good. Accounting models are a rote translation of the underlying relationships (in this case, survey results mapped to the equations underlying price indexes) so the results can be communicated coherently and intuitively. |
Transparency |
Fair. The underlying model equations are mechanical identities that are described in broad terms in an appendix to the EXPEDITION working paper and could in theory by recreated by an expert familiar with the topic. There is no role for hidden judgment. IODE is publicly available for download; however, the code for HINT is not. The software is in C and C++, which does not require licenses and has a large community but is not widely practiced among economists and policy researchers and is not as collaboration-ready as some of the Bureau’s model suite in Python. |
Proportionality |
Good. No analysts are currently working on HINT, so it is not pulling resources from other projects. The model will be revisited before the next round of election platform costing. The mechanical identities are what they are and would not benefit from additional resource investments; however, the Bureau should invest in satellite models to estimate behavioural responses to price changes that can be introduced to complement and refine the results of HINT. |
Sustainability |
Good. Accounting models can require some moderate resources to develop; however, once constructed they are relatively easy to maintain and update. In theory, operating and explaining the model is well within the tool kit of someone with an undergraduate economics degree. In practice, this is done by an economist with experienced in the field of prices and inflation, who is also responsible for further development of the model. |
Precedent |
Good. Canada’s SPSD/M commodity tax module, the Indirect Tax Tool (ITT) plug-in for EUROMOD. |
Opinion |
Appropriate. The model is appropriate for delivering the FPB’s mandate and meets industry standards. The Bureau could explore additional links to environmental taxes, more direct interaction with EXPEDITION, and options to refine income quartiles, which are not immediately comparable to other models which present results as quintiles and deciles. The Bureau should explore satellite models that could estimate behavioural responses, including on own-price elasticities of demand and cross-price elasticities that could be used to complement or refine the results of HINT. |
2.5.6. MIDAS
MIDAS is a microsimulation and projection model that the Bureau has used since 2009 to study the risk of poverty and inequality among the elderly and the long-term effects of social and economic policies on pension adequacy.
The model starts with a cross-section of the Belgian population from administration data and the national census then guesses the life path of each individual as they choose a level of education, form a family, pursue a career, and save for retirement. It then takes the projections, works out the implications for pensions, and combines the results with simulations for social benefits to form a set of indicators for income inequality and the risk of poverty in each year of the outlook.
It is distinct from EXPEDITION in that it is a longitudinal and dynamic model. Moreover, the core model focuses on pensions and it uses the LIAM2 modeling apparatus. The model is aligned as much as possible with the financial sustainability projections of MALTESE and the composition of households is aligned with the Bureau’s LIPRO lifestyle projections that models the position of individuals within households.
Table 2.10. Overview and evaluation of MIDAS
Name |
MIDAS (Microsimulation for the Development of Adequacy and Sustainability) |
Type |
Dynamic microsimulation model, with behavioural projections coming primarily from empirically estimated logistic equations. |
Description |
MIDAS is a collection of microsimulation models to study the lifecycle paths of family and career decisions to determine the adequacy of retirement planning, particularly the adequacy of public pensions. It consists of several modules that can be grouped into five blocks (1) the demographic characteristics, (2 labour market position and incomes (gross and net), (3) social security and pensions, (4) taxation, and (5) output. It is developed and maintained by the General Directorate (ADDG). The demographic modules use dynamic cross‐sectional ageing to simulate the life spans of individuals for the years between 2012 and 2070. The demographic variables, events and developments include birth, mortality, education level, leaving the parental home, partnership and marriage, having children, and divorce or separation. The modelling of family formation has been revised and expanded to align with annual publications of the Bureau’s projection of the position in the household (LIPRO position). The model also projects birth rates and migration (including the different socio-economic profile of immigrants). The labour market modules simulate eleven career stages, choices, or events, including: (1) student or pupil (including out-of-school children), (2) civil servant, (3) self-employed, (4) employee in the public sector, (5) employee in the private sector, (6) unemployed with single payment, (7) invalid (formerly self-employed), (8) invalid (formerly employee), (9) unemployed with company allowance, (10) pensioner, and (11) other inactive. The number of individuals in situations (2) to (9) is aligned (by age category and sex) to projections by MALTESE. The income modules simulate the wages and salaries of employees and civil servants, labour incomes of the self-employed, old-age and survivor's pensions and the Income Guarantee for the Elderly (IGO), unemployment benefits and career break benefits, disability benefits, the Child benefit and birth premium, the Living wage (social assistance) and the Income Replacement Allowance, and the Allowance for assistance to the elderly (THAB) and its regional equivalents. Foreign pensions are also projected, albeit to a limited extent. MIDAS is at the annual frequency, with each transition assumed to take place at the beginning of a simulation year. The basic data is for the year 2011 coming mainly from social security institutions aggregated by the Datawarehouse Labour Market and Social Protection, which is managed by the Crossroads Bank for Social Security. The data has been supplemented with tax variables from the IPCAL (database of the Ministry of Finance), and some data from the administrative Census of 2011 (for example, education and housing status). The entire sample includes 601 683 persons, stratified by region. After selecting useful cases, the sample includes 553 722 individuals. The Bureau compares the results with EU-SILC. |
Mandate justification |
Law of 5 September 2001 guaranteeing a continuous reduction of public debt and creating an Ageing Fund which named the Federal Planning Bureau as the secretariat of the Study Committee on Ageing to oversee the fund, of which one member is from the Federal Planning Bureau. The Law also created an Ageing Fund to manage additional spending between 2010 and 2030 as a result of population ageing. Law of 21 May 2015 establishing a National Pensions Committee, a Centre of Expertise, and an Academic Council assisted by a support committee of which the Federal Planning Bureau serves as the secretariat and also appoints a representative to sit on the committee. The Centre of Expertise on pensions is responsible for grouping all the knowledge on pensions from various administrations, public establishments, and public interest organisations. Law of 21 December 1994 on social and miscellaneous provisions, Art. 127. §1. The Federal Planning Bureau is responsible for analysing and forecasting socio-economic development, the factors which determine this development and for evaluating the consequences of economic and social policy choices with a view to improving their rationality, efficiency, and transparency. |
Outputs |
Percentage of pensioners and elderly at risk of poverty, by gender and under alternative poverty definitions. Inequality in equivalised income of pensioners (Gini coefficient and S80/S20 inter‑quintile ratio). Although the working-age population and children are also modelled, only projections for pensioners and the elderly are published. |
Working paper |
A working paper is available at the FPB’s website: MIDAS 2.0: Revision of a dynamic microsimulation model (Dekkers, De smet and Van den Bosch, 2023[19]). The long-term adequacy of the Belgian public pension system: An analysis based on the MIDAS model (Dekkers, Desmet and De Vil, 2010[20]). |
Major reports |
The model has been used in studies contributing to the Pension Adequacy Report of the EU, in the annual report of the Study Group on Ageing and for producing input for supplementary table 29 “Accrued-to-date pension entitlements in social insurance”, a part of the European System of Accounts 2010, published by Eurostat. 2021 Pension adequacy report (Social Protection Committee (SPC) and the European Commission (DG EMPL), 2021[21]). 2021 Annual report of the Study Committee on Ageing (High Council of Finance, 2022[22]). |
Key judgments |
Each transition is assumed to take place at the beginning of a simulation year; therefore, every situation or position applies for the full year and incomes are received throughout the year. The distributions of the population (by sex and age group) over labour market positions and family types, to which the MIDAS projections are aligned, do not record how labour market positions are distributed among people in different family types (e.g. the proportion of people in work among single people). For this MIDAS aims at maintaining the distributions observed in the start data. Each person in MIDAS has only one labour market position in each simulation year. For example, combinations of employed and part-time unemployment or part-time disability are not simulated, except for career breaks. |
Software |
LIAM2 (a Python library and framework developed in-house). |
Theory and context |
Good. Microsimulation models are the gold standard for assessing the impact of policy on social benefit delivery as they have a direct correspondence to the life course and behaviour of individual economic agents. The use of logistic functions in dynamic projections is well-established in the literature and the Bureau’s framework has led to the emergence of a community of researchers building similar models and implicitly reviewing the seminal research of the FPB. Microsimulation models allow flexible aggregation and cross-tabulation after simulation, unlike macro approaches that determine aggregation in advance. |
Accuracy |
Good. The model is for scenario analysis to aid decision-making and frame the debate and not pure projection. However, it should be expected to give accurate conditional projections of life cycle decisions given the empirically derived logistic regressions and mechanical demographic identities. No ex ante reason to expect results to be biased or have particularly large variance among alternatives, although the degree of model detail in microsimulation can reduce prediction power through the accumulation of errors and biases of individual variables that are amplified (rather than smoothed as in macro approaches). Out-of-sample projections for pensions generally compare favourably to data from Datawarehouse and EU-SILC (used to monitor poverty and social inclusion as part of the European Semester), albeit with some discrepancies in percentage of recipients and a smaller average proportion below the EU-SILC poverty threshold. Recent refinements of emigration and immigration have improved projections versus earlier modelling. MIDAS has also undergone a major overhaul in recent years to improve its validity, such as its treatment of the Income Guarantee for the Elderly (IGO). The MIDAS load module has been improved over time by validating it against data from the Datawarehouse and IPCAL database. |
Communication |
Fair. Microsimulation models capture the tax code and benefits system in a rote manner and the effect of policy simulations on real households can typically be communicated easily and intuitively. However, the complex and time-intensive dynamic framework of MIDAS means some modules are treated by analysts as a black box and interpreting the results is not always easy. |
Transparency |
Good. The Bureau participates in the modelling community discussing its code and providing the LIAM2 software it developed. There is no significant judgment applied that is not discussed in publications. The Bureau should continue to finalise the working paper that discusses developments since Dekkers et al. (2010[20]) and publish it. |
Proportionality |
Good. The Bureau currently commits two to three analysts part time to the model. This is appropriate given its legislated requirements to support the Study Committee on Ageing and the Centre of Expertise on pensions, considering that the Bureau receives additional funding specifically to do so. That said, MIDAS is a significant investment of time for its main output of a line plot in publications showing the share of pensioners at risk of poverty over the next sixty years, which could potentially be accomplished by simply reweighting the starting dataset based on the age, family and labour variables projected by MALTESE. However, the Bureau expects important research questions to emerge in the future such as the implications of immigrants among pensioners which could not be modelled by ad hoc adjustments to MALTESE. Further, accurate and useful simulations of future pensions require simulating the careers of future pensioners. It is therefore not enough to reproduce the correct number of people in work, unemployed, etc. for each year; the transitions between labour market situations should also be simulated. This is only possible with a dynamic microsimulation model such as MIDAS. |
Sustainability |
Fair. Microsimulation models in general require considerable resources to develop and maintain, but once the framework is in place, it can be run by someone with the tool kit of an undergraduate economics degree. Model development is within the expected tool kit of a graduate-level economist or an undergrad with practical programming experience. The Bureau has sufficient people trained on the model to deal with a disruption. |
Precedent |
Good. There is a growing community of similar modelers in other countries. For example, work in preparation for the European Commission’s 2024 Pension Adequacy Report, in which projections of poverty and inequality are made for Belgium and a number of other countries, involves the use of models similar to MIDAS. |
Opinion |
Appropriate. The model is appropriate for delivering the IFI’s mandate and meets industry standards. The Bureau could consider capturing second pillar pensions (fully discretionary group insurance schemes funded by employers), which are not currently modelled. |
2.5.7. MODTRIM
MODTRIM is the Bureau’s quarterly national accounts forecasting model for the short- to medium-term. It was built in 2003 but has undergone several reviews and major overhauls. It is a structural macroeconometric model that uses behavioural equations to forecast the demand components of expenditure-based GDP. Error-correction models underpin most relationships in aggregate demand, allowing for long-run equilibrium conditions with short-run correction paths to reach them following shocks. Short-run MODTRIM forecasts are combined with medium-term HERMREG modelling to feed into the Bureau’s forecasts for the Economic Budget.
Table 2.11. Overview and evaluation of MODTRIM
Name |
MODTRIM (Modèle trimestriel) |
Type |
Quarterly structural macroeconometric model |
Description |
Large-scale macroeconometric model for projecting the quarterly national accounts. Short-run neo-Keynesian and long-run neo-classical assumptions. Most components of aggregate demand are specified as error-correction equations. |
Mandate justification |
Law of 21 December 1994 on social and miscellaneous provisions giving the Bureau the responsibility of preparing the economic forecasts for the federal budget. Law of 10 April 2014 that implemented Directive 2011/85/EU on requirements for multi-annal budgetary frameworks of Member States and pre-emptively complied with the enhanced budget co-ordination and surveillance framework in Regulation (EU) No 473/2013 of the “Two Pack” of reforms requiring that a Member State’s budget be based on macroeconomic forecasts produced or endorsed by an independent body. |
Outputs |
Detailed expenditure-side forecasts of GDP in current and constant prices by components, typically for eight quarters. Prices (CPI, health index, export and import prices, terms of trade, GDP deflator), employment and wages (hourly wage cost, real hourly wage cost, unit labour cost, employment, value added, hourly labour productivity), income (real disposable income households including and excluding property income), household savings rate. |
Working paper |
A new version of MODTRIM II (De Ketelbutter et al., 2014[23]). MODTRIM II: a quarterly model for the Belgium economy (Hertveldt and Lebrun, 2003[24]). |
Major reports |
Annual forecasts for the budget, such as Economic Budget – Economic forecasts 2022-23 for September 2022 (Bureau, 2022[25]). |
Key judgments |
Net property income and equity holdings excluded from household consumption, after shown to have no explanatory power. Business investment previously derived from Cobb-Douglas production function but dropped from latest version of model in favour of a behavioural equation in levels. |
Software |
IODE (Intégrateur d'outils de développement économétrique) |
Theory and context |
Good. Structural macroeconometric models with error correction equations are the gold standard of macroeconometric models required to balance forecasting and policy analysis. Suited to twin goals of capturing data and dynamics with enough structure to trace effects of policies and shocks. Large-scale macroeconometric models well-supported in the literature, fell out of favour, but are having a resurgence as the limitations of DSGE models are better understood. |
Accuracy |
Good. Because of its theoretical underpinnings and reliance on medium-run equilibrium conditions and use of levels and dynamics via error correction models, these models are likely to improve upon naïve forecasts for the medium run. However, empirical comparisons of model classes for the short run tend to show that purely statistical time series models outperform structural models. For the Bureau’s role it must undertake risk scenarios and sensitivity analysis, as required by the new European Directive on budgetary framework, and so a structural model rather than a statistical time series forecasting model is required. |
Communication |
Good. Structural macroeconometric modelling can produce coherent, intuitive narratives in-line with economic theory compared to purely statistical macroeconomic forecasting models. Coefficients and directions are meaningful. |
Transparency |
Good. Working papers with equations provided, complete with parameters and statistical test tables. Detailed data and sourced. Could be replicated by an experienced external analyst. That said, judgment plays a significant role in tuning and combining the models and introduces some obscurity. |
Proportionality |
Good. Four analysts spend a small part of their time with MODTRIM, which is approximately in line with the weight of short-run forecasting in the Bureau’s mandate. |
Sustainability |
Fair. Once developed, experienced analysts with a degree in economics or a numerate field could support and run the models. Maintenance and development are likely to require a PhD economist or an analyst with an MSc-level background and substantial experience. |
Precedent |
Good. The Bureau’s model is in-line with the large-scale IS/LM and supply-side structural econometric models used at other IFIs and institutions such as the Parliamentary Budget Office of Canada and Portuguese Public Finance Council. |
Opinion |
Appropriate. Appropriate unqualified. The tool meets the standard practices of peers and is appropriate for delivering the Bureau’s mandate. No further action is recommended. |
2.5.8. PROMES
PROMES is a microsimulation model developed at the request of the National Institute for Health and Disability Insurance. The model was also used in 2019 for the Bureau’s election platform costing mandate to compute the medium-term budgetary impact of changes to health care policy in detail. It can estimate the budget implications of measures such as a percentage reduction in user fees, an increase in coverage for sickness insurance, an increase in dentistry fees, or other measures targeted at health expenditures.
The model consists of 25 modules corresponding to major expenditure groups, for example consultations and visits, dentistry, or physiotherapy. For each expenditure group, a behavioural model was estimated to explain the use of care according to individual characteristics, living environment, and previous use of care to arrive at projected healthcare volumes and expenditures. It was developed with the assistance of the National Institute for Health and Disability Insurance (NIHDI) and relies on the Permanent Sample (EPS), a longitudinal administrative database on the use of health care with more than 300 000 respondents. The results obtained for the sample are extrapolated to the future population using reweighting factors.
Expenditure projections from PROMES also contribute to the economic outlook in HERMES and the acute and long-term care in MALTESE (for the medium-term part).
Table 2.12. Overview and evaluation of PROMES
Name |
PROMES (PROjecting Medical Spending) |
Type |
Partly dynamic microsimulation model based on logistic regressions of the probability of using care. |
Description |
PROMES computes detailed health care expenditure projections over the medium term for 25 expenditure groups (12 modules with subgroups). Projections use two-step modelling of the volume, or number of care units, by estimating the probability of accessing care with logistic regression and the average volume as a function of demographic, socio-economic characteristics, indicators of morbidity, previous consumption, and environmental factors. The model is calibrated on technical estimates from the National Institute for Health and Disability Insurance. Input data is longitudinal and comes from the Permanent Sample (EPS), an anonymous, randomised, and representative sample of 1/40th of the Belgian population from the database of the Inter-Mutualistic Agency, which consists of members of mutual insurance companies. To project health states, a chronic disease and disability indicator is constructed from data on insured persons and morbidity indicators have been constructed from data on the consumption of prescription drugs: an indicator of general health based on the number of different drugs taken, a series of indicators for chronic diseases (cardiovascular conditions, COPD/asthma, rheumatoid arthritis, diabetes, epilepsy, Parkinson's, Alzheimer's, psychosis and conditions of the thyroid gland) and other indications of chronic conditions. PROMES also includes an influenza epidemic variable constructed using data from the Scientific Institute of Public Health that can be set to assume above normal outbreaks of flu. |
Mandate justification |
The Bureau also has responsibilities toward the Belgian Health Care Knowledge Centre, such as calculating sustainability indicators, and toward the National Institute for Health and Disability Insurance. Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, including the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. |
Outputs |
Budget impact of measures, deviation from baseline, in thousands of euros, typically for a 4-to-6-year horizon. Expenditure by categories, such as dental practitioner fees, medical imaging, clinical biology, nurse practitioner fees etc. typically for a 4-to-6-year horizon. |
Working paper |
Description and use of the PROMES model (Geerts, Van den Bosch and Willemé, 2018[26]) |
Major reports |
The report by the Belgian Health Care Knowledge Centre, Assessing the sustainability of the Belgian health system using projections (Lefèvre and Gerkens, 2021[27]), contains the Bureau’s projections for medical care consumption up to 2025. Results of the model contributed to a report commissioned by the Minister of Health aimed at preparing a multiannual budget for health care. Results of the model were used to calculate the outcomes presented for each party’s policy proposals, where relevant, on the DC2019 election policy costing portal: https://www.dc2019.be/. For example, “Cheaper medicines through public tendering” (Proposal 103 of party “Groen”). |
Key judgments |
Data drawn from the EPS does not provide information on several important individual characteristics that may influence demand for care, such as household income, level of training, lifestyle (diet, consumption of alcohol and tobacco, physical activity, etc.), background and working conditions. Some of these variables are available in principle if the database is linked to other databases, but this has not yet been undertaken for reasons of privacy protection. PROMES only models the consumption and expenditure on care that fall under the health and disability insurance nomenclature (AMI). Some aspects of long-term care, which have been devolved to the Communities and Regions for a longer time (such as care for the disabled), are not included in the AMI nomenclature. Specific actions taken in these areas cannot be assessed with the PROMES model. Because the model is based on historical data it cannot directly evaluate new initiatives such as extending insurance to reimburse psychotherapy. The expected effects of such measures on expenditure are assessed through external estimates and then added to the projection results. |
Software |
SAS and Stata (for weights), with Python for summaries and data visualisation. No server license for SAS, three workstations. Looking at remote desktop licensing. NIHDI uses SAS, so the Bureau must follow for data aggregation. Rewriting weights in Python. GUI Tools for NIHDI are in conceptual phase. Needs more programming, which is planned for next year. |
Theory and context |
Good. Microsimulation models based on administrative data are the gold standard for health care expenditure modelling. The projections leverage the latest literature on the life-course perspective of health and social welfare analysis and empirical techniques of logistic regression for two-step modelling of number of care units and the probability of accessing them. They allow flexible aggregation and cross-tabulation after simulation, unlike macro approaches that determine aggregation in advance. |
Accuracy |
Good. Dynamic elements used for projection are empirically estimated. The first results were available in 2019, after which COVID-19 complicates ex post assessments of accuracy. Initial results of some out of sample assessments show some sectors better than others. In theory should be reasonably good forecasts given exogenous growth variables. The model is not only for forecasting, but also for scenario analysis to aid decision-making and frame the debate. The degree of model detail in microsimulation generally reduces prediction power through the accumulation of errors and biases of individual variables that are amplified (rather than smoothed as in macro approaches). However, it should be expected to give accurate conditional projections given the empirically derived logistic regressions. Economic growth factors and certain long-run spending factors taken exogenously will reflect the underlying accuracy of HERMES. |
Communication |
Good. For the most part easy to explain with intuitive paths and mechanisms. The dynamic framework complicates the story-telling ability but the decision framework for each category is straightforward to explain. |
Transparency |
Good. The model uses a mix of low-fee licensed software and open source (SAS and Stata and Python). The code could be made accessible and shared. A working paper has been published. Some room for judgment obscures underlying equation specifications. |
Proportionality |
Good. Model uses a little over two full-time equivalents (2. 5). Health care is not a core sectoral activity of the FPB but is critical fo2r thinking about long-run fiscal sustainability. DFR |
Sustainability |
Good. There is good documentation and transition plans for business continuity. In theory, the model can be run by undergraduate economists and developed by master’s level economists with experience in the area. However, given the small size of the team this is not feasible in practice. Significant project capacity development in the form of new modules require a PhD economist or someone with considerable experience in the field of healthcare. |
Precedent |
Good. A similar approach has been used to project LTC expenditure in Flanders by Steunpunt Welzijn, Volksgezondheid en Gezin. France has expressed an interest in the FPB’s modelling. |
Opinion |
Appropriate. The model is appropriate, and no further changes are required. An industry-leading model that the OECD will be recommending as an example for other independent fiscal institutions and central ministries. The COVID-19 challenges and data complications will require an investment to overcome. |
2.5.9. MALTESE
The MALTESE model was developed to support the Bureau’s mandate added in 2001 to serve as the secretariate for the Study Committee on Ageing (SCA). Results have appeared in the Committee’s reports since 2002. The FPB also uses the model to represent Belgium at the EU’s Working Group on Ageing Populations and Sustainability (AWG) established in 1999 by the Economic Policy Committee of the Economic and Financial Affairs Council (ECOFIN). The model has also informed several high-profile policy impact studies for pension reforms, such as those in 2015. The model is also used to answer pension questions posed to the Bureau as part of the Knowledge Centre of Pensions (established in 2015).
The model consists of a set of modules for translating demographic projections into budgetary developments for social protection spending, particularly public pensions, over a horizon of 50 years (currently until 2070). Results are published by branch of social protection. Total revenues and indicators of public finances (balances, debt) are also modelled.
Table 2.13. Overview and evaluation of MALTESE
Name |
MALTESE (Model for Analysis of Long-term Evolution of Social Expenditure) |
Type |
Macroscopic accounting model |
Description |
MALTESE consists of a central model and several specific peripheral models (MOSES, PENSION, PUBLIC, HORBLOK) for estimating the long-term budgetary implications of ageing, especially on public pensions. MOSES determines the average pension in the scheme for self-employed workers, PENSION determines the evolution of the average pension in the scheme for salaried workers, PUBLIC determines the average pension in the scheme for civil servants, and HORBLOK carries out a projection of the number of pensioners by scheme and by category of pension within each scheme. For the AWG projections, it uses demographic data from Eurostat’s population projection. For the SCA projections, it is based on the national demographic projection made by Statbel and the FPB. The projection horizon is fifty years, currently to 2070. Additional data comes from the national accounts and various social protection institutions. The starting point for the detailed public finances estimates is the HERMES medium-run forecast. The long-term MALTESE socio-economic and wages projections feed into the MIDAS simulation framework through alignment tables. It is developed and maintained by the General Directorate (ADDG). |
Mandate justification |
Law of 5 September 2001 guaranteeing a continuous reduction of public debt and creating an Ageing Fund, which named the Federal Planning Bureau as the technical and administrative secretariat of the Study Committee on Ageing to oversee the Ageing Fund to manage additional spending between 2010 and 2030 because of population ageing (the fund was cancelled in 2016 but the Study Committee remains). The Federal Planning Bureau also appoints a member of the Study Committee. The Study Committee produces a yearly report on the budgetary and social implications of ageing. MALTESE is used for the budgetary component. Law of 21 May 2015 establishing a National Pensions Committee, a Centre of Expertise, and an Academic Council assisted by a support committee of which the Federal Planning Bureau serves as the secretariat and also appoints a representative to sit on the committee. The Centre of Expertise on pensions is responsible for grouping all the knowledge on pensions from various administrations, public establishments, and public interest organisation. Since 2001 the FPB also represents Belgium at the EU’s Working Group on Ageing Populations and Sustainability (AWG) established in 1999 by the Economic Policy Committee of ECOFIN), where MALTESE is used for the pension projection. The AWG is “responsible for producing common budgetary projections” on age-related public expenditure items and each member state must project its long-term pension expenditure under common assumptions. The Bureau must complete a detailed pension questionnaire about the results and the results undergo peer review of the pension projection results by a Member State and the European Commission. |
Outputs |
Benchmark and alternative scenarios (namely on productivity, employment rate or demographic parameters) for the additional cost of ageing on pensions and other social expenditure, expressed as the change in gross social expenditure in percentage points of GDP. Impact analyses of reforms. In the annual report of the Study Committee on Ageing, the following results are available: social protection expenditure by branch (pensions by scheme, health care, long-term care, incapacity to work, unemployment, family allowances, other social expenditure) in % of GDP; macroeconomic projections; socio-economic projections; population projections. For the Ageing Report of the European Commission, the results are focused on pensions: gross public pension spending by scheme as a % of GDP (with a breakdown for old-age and early pensions, earnings related, non-contributory pension, disability pensions, survivor pensions, other pensions); new pensioners; replacement rate at retirement; benefit ratio. |
Working paper |
High-level description in Tools and methods used at the Federal Planning Bureau (Federal Planning Bureau, 2006[28]) |
Major reports |
The projections of public pensions are published in the EU Ageing Report and in the Fiscal Sustainability Report of the European Commission that assesses the medium-term and long-term fiscal situation of Member States. The projections of all social protection expenditure are used in an annual report produced by the Study Committee on Ageing (Comité d’étude sur le vieillissement, or CEV) on the budgetary and social implications of ageing. The first report was in 2002. Normally one long-term projection published in July for the Study Committee on Ageing where the medium-term projection is taken from the Economic Outlook (HERMES model) published in June for consistency. On a more irregular but therefore no less intensive basis, the model is used to answer policy questions, and in particular pension reforms, posed to the Bureau as a member of the Knowledge Centre of Pensions. The model was used in an influential impact assessment of pension reforms. The last one being the current proposition of reform 2022-2023 with behavioural-financial incentives to work longer and postpone retirement and the impact on the labour force. The MALTESE model was not used within the framework of DC2019 given the evaluation horizon retained by law (a short and medium-term calculation) and the team was required for social benefits analysis with HERMES and EXPEDITION. For this purpose, the underlying (pension) models of MALTESE were used for the medium-term calculation. Outlook 2019-2070 (High Council of FInance, 2020[29]) Annual report of the Study Committee on Ageing (High Council of Finance, 2022[22]) |
Key judgments |
For the AWG-projections, assumptions are based on common demographic, scope of pension, and macroeconomic assumptions discussed in the EU’s Ageing Working Group and approved at the EPC level. These are different from those used in the national projection of the Study Committee on Ageing. Simulations of pension measures in the context of the Knowledge Centre of Pensions are usually simulated according to the SCA-scenario. Assumes no policy change. Legislation is included: for the AWG only if it has already been voted and passed; for the SCA promulgated measures are taken into account. Does not include second and third pillar pensions (private voluntary individual pensions schemes) for which there is a data gap. All social expenditure are automatically adjusted to the consumer price index, unless otherwise stipulated by legislation. Social benefits are also adjusted to living standards in real term, for example the pensions of retired public servants are partly indexed to the real wage of working civil servants. |
Software |
Iode for the core model of MALTESE and Python LArray, a library developed by the Bureau, for (almost) all peripheral models. |
Theory and context |
Good. Well-grounded in established methodologies that mimic the underlying accounting identities. Peer-reviewed among Member States and the European Commission. Appropriate for the EU context. |
Accuracy |
Good. Long-term projections are a thought exercise to identify whether immediate policy action is required and are therefore not intended or expected to be a most-likely scenario. As conditional projections, the framework is likely to give accurate results given the exogenous demographic projections and assumptions. In the context of the AWG, the results of the pension projection were peer reviewed in detail by the European Commission (DG ECFIN) and by the Romanian delegates to the AWG. They also receive ongoing scrutiny by other Member States. |
Communication |
Good. It is easy to communicate the moving parts and assumptions of accounting models. The Bureau could explore summary statistic that could improve communication (for example, a “pensions gap” that discounts the future stream of unfunded liabilities and gives the immediate and permanent adjustment to contributions as a share of GDP required to bring the accounts to a stable steady-state funding equilibrium). |
Transparency |
Fair. No public working paper. However, the Python LArray initiative is a fruitful transparency and open-source initiative for the policy community and appreciated by other institutions internationally. |
Proportionality |
Good. Two analysts part time to the core model MALTESE and 3 FTE for the peripheral models strikes the right balance within the Bureau’s overall responsibilities. However, the team also is responsible for social analysis with HERMES and EXPEDITION, leading to potential conflicts in time commitments. |
Sustainability |
Good. As constructed, the model could in theory be both maintained and expanded by a relatively junior analyst with an undergraduate degree with a background in Python programming. However, in practice the small size of the team means that all tasks (including developing and extending the model) are performed by master’s degree economists with large experience in the field of social expenditures. There is no sufficient duplication of knowledge to ensure business continuity. The Bureau should improve documentation. |
Precedent |
Good. Uses a common EU framework and assumptions (for the AWG projections). Demographics, cohort model, and supply side long-run potential GDP are all standard assumptions used by many other fiscal councils and PBOs for similar analysis. |
Opinion |
Appropriate. The tool is appropriate, and no further action is recommended. |
2.5.10. PLANET
PLANET is a model developed in-house to make long-term projections of the demand for passenger and freight transport in Belgium and to carry out transport-related policy analysis. The model derives transport demand by mode and period (peak and off-peak hours) from the evolution of demographic, economic and price variables (fuel prices, transport fees, and time costs). It also considers externalities such as pollutions and congestion.
The latest PLANET version used for the 2019 election costing period including teleworking in the commuting module and distinctions between morning and evening peak hours, outward and return trips, and private and company cars.
Table 2.14. Overview and evaluation of PLANET
Name |
PLANET (PLANning Economy and Transport) |
Type |
A framework including representative agent models, nested trees, (un)constrained gravity models and discrete choice modelling. |
Description |
A collection of tools developed in-house to assess the relationship between the economy and transportation demand to produce: (1) medium- and long-term projections of transport demand in Belgium, both for passenger and freight transport; (2) simulations of the effects of transport policy measures; and (3) cost-benefit analyses of transport policy measures. Produces the endogenous change in aggregate transport demand, mode choices, and time spent in response to changes in pecuniary and time costs. It uses the economic outlook from HERMES for the first five years and MALTESE projections afterward. Data comes from a variety of sources, including administrative databases and surveys. For measures assessed during the 2019 election costing period, the Belgian car fleet model CASMO (CAr Stock MOdel) was used as an input for PLANET. CASMO has undergone a recent overhaul to estimate the discrete choice model econometrically based on actual purchase transactions over several years and a distinction between the type of owner (private persons versus legal persons) by region allowing analysis of region-specific tax regimes. |
Mandate justification |
The Programme Law of 23 December 2009 requiring the Federal Planning Bureau to develop and maintain a database of transport indicators and satellite statistical accounts for the FPS Mobility and Transport and to carry out transport simulations with impact analysis and policy analyses on request and in consultation with the FPS Mobility and Transport Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, including the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. |
Outputs |
Transportation demand and consumption for twenty years (currently to 2040). The medium and long-term impact of transportation policy proposals on traffic, emissions, and welfare. |
Working paper |
The PLANET Model: Methodological Report, PLANET 4.0 (Daubresse and Laine, 2020[30]). Telework and transport demand: an evaluation in the PLANET model (Daubresse and Laine, 2020[31]). Description and use of the PLANET model (Daubresse et al., 2018[32]). The PLANET model methodological report: Modelling of short sea shipping and bus-tram-metro (Gusbin et al., 2010[33]). |
Major reports |
The Bureau is required to publish a Transport Outlook every three years as part of an agreement with the Federal Public Service Mobility and Transport, with whom it discusses and agrees upon which alternative scenarios to assess. In 2021, a technical report was published that studied the potential of a range of measures to encourage carpooling, including penalties for driving alone. |
Key judgments |
Own-price and cross-price elasticities from the literature. The only endogenous variable is the speed of the road. Always work on a constant policy basis. In CASMO, the Berry-Levinsohn-Pakes method for estimating discrete choice models based on aggregate market data was used. |
Software |
GAMS, some ad hoc in R, plan to migrate GAMS components to Python/LArray. |
Theory and context |
Good. Elasticities are taken from peer-reviewed literature. Structural relationships are firmly grounded in theory. Estimation techniques are well-founded. The overall model packages and individual specifications are sensible for the Bureau’s unique mandate for transportation analysis to support government decision-making. |
Accuracy |
Fair. The team has performed ex post assessments that provide satisfactory results for the prediction. However, much of the data is ten to twenty years out of date. Accuracy could be improved with new survey programmes. |
Communication |
Fair. The various transport modelling modules generally tell an intuitive and convincing story; in some cases however, sometimes opaque inner workings or complexities emerge that are more of a black box. The team has occasional difficulties tracking the exact causal channels and explaining the results to stakeholders. |
Transparency |
Fair. The Bureau has published several detailed working papers. GAMS is a moderately priced licensed software with a relatively small community and is not commonly part of the software suite in other institutions. Some ad hoc components of the model suite are in open source. The Bureau has a plan to migrate the complete package to open-source software. The team is open to publishing the model and considering it as they migrate to open-source software. |
Proportionality |
Good. Four full-time staff, two of which receive funding under an agreement with the FPS Mobility and Transport. The size of the team is appropriate for the Bureau’s unique mandate for transportation analysis and its duties to the government. |
Sustainability |
Good. Large team with distributed expertise and routines in place for knowledge transfer. Option to use commercial transportation modelling software would ensure development and support but would forego some of the adaptability and coherence of the current model suite and introduce further black box elements. The best practice is to continue to use the Bureau’s in-house solution. Although niche, the model is accessible for non-PhD analysts from a range of generalist backgrounds and the Bureau has shown that they can find talent and onboard them quickly. |
Precedent |
Good. Switzerland has a similar model that is developed and maintained under a commercial software license with an external supplier. Government departments in other countries use a variety of patch-work solutions to model and assess transportation policy that are not unlike the Bureau’s approach. The lack of close examples in other IFIs is an artifact of the Bureau’s unique mandate, not that they’re pursuing an inappropriate approach that is not widely prevalent. |
Opinion |
Appropriate. The tool is appropriate. The Bureau is a leader in the field and can be an inspiration to other governments and institutions looking to wade into similar analysis. The Bureau should persist in its plan to migrate the model to Python and make the code publicly available. The Bureau should continue to invest in a solution to redesign the freight transport module and adopt an approach more in line with physical flows and less constrained by a theoretical allocation of economic flows. This would allow improved modelling of mode choices and geographical influences of freight transport demand. |
2.5.11. CRYSTAL SUPER GRID
CRYSTAL SUPER GRID is a “unit commitment” and “economic dispatch” model linking up to thirty-three European countries to assess the impact of different assumptions on prices and distribution within the electricity sector. Unit commitment determines the start-up and shut-down schedule of energy production units. Economic dispatch determines the power output of each energy production unit according to its cost and operational constraints, as well as the limits of the transmission network. The model determines both unit commitment and economic dispatch with optimisation routines that match supply and demand, while enforcing operational constraints (e.g. production limits, ramping constraints), by minimising total system production costs.
The Bureau has used the model since 2015. It is maintained and developed by Artelys, an external commercial software provider that specialises in energy modelling. In addition to the optimisation solvers, Artelys maintains an extensive library of physical and financial assets (thermal power plants, renewable energy sources, power lines, etc.) that is used in the scenario’s construction.
Table 2.15. Overview and evaluation of CRYSTAL SUPER GRID
Name |
CRYSTAL SUPER GRID |
Type |
Optimal unit commitment and economic dispatch model. |
Description |
An optimisation tool that minimises total system production costs while aligning demand with supply, and enforcing operational constraints, at any point in time for up to thirty-three European countries. Developed by Artelys, an external commercial software supplier. It contains an extensive library of both physical and financial assets (thermal power plants, renewable energy sources, power lines, etc.). Results cover imports and exports between zones (countries or regions), marginal costs of electricity generation, as well as the CO2 emissions of the national and European electricity sector. Analysis can be performed at an hourly resolution (or indeed any frequency). |
Mandate justification |
Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, including the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. The Law of 8 January 2012 amending the Act of 29 April 1999 on the organisation of the electricity market requiring the Federal Planning Bureau to collaborate with the Directorate General for Energy of the FPS Economy, in concertation with the Commission for Electricity and Gas Regulation (CREG), to draw up a regular report on the monitoring of the security of the supply of energy. The Climate Responsibility Mechanism Act of 6 January 2014 requiring the Federal Planning Bureau to monitor the methodologies and compliance with the trajectories and the realisation of the objectives with regard to the reduction of the Belgium's emissions under European legislation and the United Nations Framework Convention on Climate Change and its protocols. Various agreements and commitments to collaborate in the work of Elia (Belgium’s transmission system operator), DG Energy of the EC, and FPS Economy. |
Outputs |
The long-term effects of energy policies on its security of supply, sustainability, and affordability. Imports and exports between countries or regions, marginal costs of electricity generation, CO2 emissions of the national and European electricity sector, load losses, productions by generation type/unit, production of renewables, estimated future price effects. |
Working paper |
Description and use of CRYSTAL SUPER GRID (Devogelaer, 2018[34]) |
Major reports |
The Bureau provides feedback to Elia for the Federal Development Plan and the Adequacy & Flexibility Study and collaborates with the DG Energy of the FPS Economy and the CREG in the Etude Prospective. The Bureau supplies both analysis and text for Energy Outlook/Studies and the long-term energy outlook published every three years. Analysts may also use the model to contribute to the National Energy and Climate Plan. |
Key judgments |
Energy demand is exogenous. Therefore, prices determined on the wholesale market will not influence energy demand. |
Software |
Proprietary software and web interface from Artelys, an external provider. Written in Java. Bureau has a license for the cloud-based instance of the model. |
Theory and context |
Good. The optimisation routines come from a long history within the engineering field and have been well-scrutinised. |
Accuracy |
Good. The model is used for scenarios and sensitivity analysis, not prediction. Energy industry and systems planners use similar software and optimisation routines to determine energy investment and supply so likely to closely approximate real-world decisions and allocations. |
Communication |
Good. Can trace the effects of policy simply and intuitively and tell obvious stories—for example, increasing the price of one source of energy with a tax would instigate a reallocation to other generation technologies. |
Transparency |
Fair. The optimisation routines are largely a black box from proprietary software; however, the model largely consists of a set of mathematical solvers without a great deal of judgment applied in either producing results or choosing model specifications. The Bureau has drafted and published a working paper describing how it engages with the model with as much detail as possible. |
Proportionality |
Good. Two analysts, with one backup in another workstream. Appropriate for the weight of energy analysis within the Bureau’s mandate and business agreements. |
Sustainability |
Good. The team has shown it can handle business disruptions as it has recently had turnover of most of the team. Analysts report that they can be trained up to speed in a day. Artelys also provides training and support under a maintenance contract. That maintenance contract provides for the delivery of some new functionalities. Model developments or updates are not handled by the FPB’s analysts themselves—if there is a need for new functions or specifications an additional contract has to be signed with Artelys. Somewhat expensive and the Bureau is exposed to price increases that could disrupt operations. |
Precedent |
Good. Similar models and optimisation routines used by the energy industry in its commercial decision making. METIS used by government departments and energy institutions in other countries (same company, a few differences, core model is the same). |
Opinion |
Appropriate. The model is appropriate, and no changes are recommended. The Bureau is ultimately paying for a set of solvers that it has assessed would not be cost-effective to reproduce in-house. The review confirms this assessment. |
2.5.12. Models under construction
The assessment also included three models that are not yet in full production (LASER and DynEMIte) or are very early in their development stage. While it is not possible to issue an opinion on the appropriateness of the models at this stage, the models were nonetheless partially assessed, with preliminary feedback provided below.
LASER
LASER is a static structural discrete-choice model that estimates the change in an individual’s labour supply in response to a policy that affects household disposable income. The policy’s affect on household disposable income is taken from the EXPEDITION model.
The model has been in an ongoing state of development since 2017 to contribute to the Bureau’s election platform costing mandate. It is currently undergoing a refinement of its parameters and elasticities in response to the lessons learned during the 2019 election and to allow a link with the HERMES model.
Table 2.16. Partial overview and evaluation of LASER (the model is still in development)
Name |
LASER (LAbour Supply model to Evaluate policy Reform) |
Type |
Cost of Working Model, a static structural discrete choice model of labour supply that uses multinomial logit with mass points on household consumption. |
Description |
The model uses the output of EXPEDITION to assess the expected impact of a measure on labour supply (both the extensive and intensive margin) after households consider the impact on their disposable income. It is estimated using the same set of administrative data as EXPEDITION—that is, microdata on individuals and their households from the Labour Market and Social Protection Datawarehouse of the Crossroads Bank for Social Security (a collaboration between dozens of institutions involved in social security), tax records from the IPCAL administrative database, and census data for the highest level of education attained. The model is still under construction within the General Directorate (ADDG). |
Mandate justification |
The Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, amended 2018 to restrict requests to a minimum 3 and maximum 5 priorities. The analysis is to include the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. |
Outputs |
Change in unemployment, change in labour force participation, change in hours worked, plus the same outputs as EXPEDITION, but corrected for labour supply responses |
Working paper |
Not available. |
Major reports |
Results of the model will be featured election policy costing portal for the next election. |
Key judgments |
Do not observe wage equation. Needs a wage floor. |
Software |
Stata |
Theory and context |
Good. The methodology to estimate a COW model is appropriate and its application is less complex than other estimation methods such as those used to estimate a Random Utility Random Opportunity Model (RURO). |
Accuracy |
N/A. The predictive power of the COW model will need further investigation. |
Communication |
Good. The model is theory-based and the responses of workers to changes in incentives should be an intuitive story to communicate to stakeholders. |
Transparency |
N/A. When the model is put into use it should be accompanied by a working paper. The Bureau’s plan to transition the model development to Python will score highly on transparency. |
Proportionality |
Good. The Bureau currently has one analyst developing the model part time. This is appropriate, as the labour supply analysis will be only a minor component of the election platform costings and is not strictly required for delivering the Bureau’s other legislated mandates, although it will eventually be a useful information point for stakeholders in all relevant areas of the Bureau’s policy analysis. |
Sustainability |
Good. Once developed, the module could in theory be manageable by a junior analyst with an undergraduate economics toolkit and has a short learning curve. Maintenance and development require a skilled economist. There are no obvious risks to business continuity with employee turnover. |
Precedent |
Good. Similar labour-response models are used by researchers in other think tanks, at universities, and in fiscal institutions that supply similar analysis, like the CPB Netherlands Bureau for Economic Policy Analysis. |
Opinion |
N/A. The model development plan is appropriate for delivering the Bureau’s mandate and the resulting tool, when finished, should meet industry standards. As the Bureau continues to improve LASER, it should study the possibility of (1) expanding the target population with more types of working status, such as unemployed and self-employed and people with disabilities, (2) linking administrative data to data from the labour force survey, and (3) using LASER’s elasticities in the new environmental CGE model. |
DynEMIte
Following the 2019 election platform costing exercise, the Bureau began developing a new in-house DSGE model it has named DynEMIte. The model is largely based on the structure of QUEST III R&D, but incorporates multiple industries, intermediate consumption (and hence, input-output linkages between industries) and labour-augmenting semi-endogenous technological growth. It is currently in the calibration and estimation phase. The team is having problems with the convergence of the model under certain parameter values .
Table 2.17. Partial overview and evaluation of DynEMIte (the model is still in development)
Name |
DynEMIte |
Type |
Dynamic stochastic general equilibrium model |
Description |
Based on the structure of QUEST III R&D but incorporating multiple industries, intermediate consumption (and hence, input-output linkages between industries) and labour-augmenting semi-endogenous technological growth. |
Mandate justification |
Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, including the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. The Law of 25 November 2018 establishing the National Productivity Council which prescribes that the bureau nominates two of its members to the National Productivity Council and must contribute to the meetings and reports on the topics of diagnosing and analysing developments in productivity and competitiveness, associated challenges, and the consequences of policy options on productivity and competitiveness. |
Outputs |
Equilibrium percentage change of typical macroeconomic variables. |
Working paper |
Not available. |
Major reports |
N/A. Model still in development. |
Key judgments |
N/A. Model still in development. |
Software |
Dynare and Matlab. |
Theory and context |
Good. As with QUEST III R&D, theory-derived DSGE models are generally considered to be the only practical solution for structural reform questions. |
Accuracy |
N/A. Model still in development and will need to undergo testing. DSGE models are a device to simulate the propagation of shocks through a highly stylised and simplified theoretical and empirically validated framework, but are not intended to be used for forecasting. |
Communication |
Fair. DSGEs are grounded in economic theory and should be able to tell a coherent and consistent economic story of the relationships between variables of interest. However, the complexities of the model in practice make it somewhat of a black box. |
Transparency |
N/A. Model still in development. |
Proportionality |
Fair. Two researchers committed part time. The Bureau should reflect on the resources it devotes to DSGE research considering the growing literature (such as (Korinek, 2015[35]) and (Stiglitz, 2017[36])) on its practical limitations and the expertise and attention it demands of the Bureau’s analysts. |
Sustainability |
Poor. Maintaining and developing DSGE models generally requires a PhD economist in the field. Once developed, experienced analysts with a degree in economics or a numerate field could run the models but would need support of colleagues or external consultants. Most central finance ministries with a DSGE model retain external academic economists for developing new capabilities. If the Bureau departs too much from the stock QUEST III R&D model, there could be a risk to business continuity with fewer potential candidates in the event of staff turnover. |
Precedent |
Good. DSGE models are commonly found in central banks and across EU governments. QUEST III R&D and its forebearers are used commonly in think tanks and other fiscal institutions. |
Opinion |
N/A. While the new DSGE model will be a welcome tool to answering questions, it may be necessary to also maintain QUEST III R&D for EU policy analysis commitments and to benefit from the relatively large QUEST community. Despite clear synergies between QUEST and DynEMIte allowing to increase the efficiency of resources used, two DSGE models may be out of step with the practices of the Bureau’s peers and could be a misplacement of resource priority. |
New environmental CGE model (unnamed)
The Bureau is developing a new computable general equilibrium (CGE) model to focus on climate and energy policy. It is in the very early stages of conceptual development and is supported by the Belgium Research Action Through Interdisciplinary Networks (BRAIN-BE). It is intended to be a standard multi-sector recursive dynamic model covering Belgium and its regions with an emphasis on energy inputs and heterogeneous labour demand. It is also to be linked to microsimulation data for distributional analysis.
Table 2.18. Partial overview and evaluation of the new environmental CGE model (the model is still in development)
Name |
Unnamed |
Type |
Dynamic (recursive) Computable General Equilibrium Model |
Description |
Standard multi-sector recursive dynamic model covering Belgium and its regions with an emphasis on energy inputs and heterogeneous labour demand. It is also to be linked to microsimulation data for distributional analysis. |
Mandate justification |
Law of 21 December 1994 on social and miscellaneous provisions giving the bureau the responsibility of calculating a set of additional indicators for measuring quality of life, human development, social progress and the sustainability of our economy. The Law of 22 May 2014 requiring the Federal Planning Bureau to cost the election manifestos of political parties, amended 2018 to restrict requests to a minimum 3 and maximum 5 priorities. The analysis is to include the short- and medium-term consequences for public finances, the purchasing power and employment of various income groups, for social security, and of the impact on the environment and transportation. |
Outputs |
N/A. Model still in preliminary stage of development. |
Working paper |
Not available. Model still in preliminary stage of development. |
Major reports |
N/A. Model still in preliminary stage of development. |
Key judgments |
Intended to develop alternative to standard nested CES production functions. |
Software |
GAMS initially and eventually an open-source alternative. |
Theory and context |
Fair. There is a wide body of literature on CGE models applied to environmental and climate studies; however, the results of the literature are mixed on the usefulness and appropriateness of CGE models in policy applications. |
Accuracy |
N/A. Will need to be assessed. However, CGE models are geared toward theory and conceptual scenario analysis, rather than real-word predictive power, and are not generally considered to be adept at forecasting or capturing empirical relationships accurately. |
Communication |
Fair. CGE models are theory-based and should allow for intuitive narratives; however, their complexity and moving parts means they become black boxes in practice. |
Transparency |
N/A. Model still in preliminary stage of development. |
Proportionality |
Poor. The Bureau’s responsibilities surrounding environmental and climate modelling are considerable and growing. Currently one full-time analyst with some part-time support does not reflect the weight of climate analysis in the Bureau’s mandate and it has fallen behind in preparation for the policy conversation. The Bureau should invest rapidly to catch up and devote sufficient staff resources to the project. |
Sustainability |
Fair. CGE models require specialised skills and have a steep learning curve. However, the Bureau is capable of recruiting and there is a sufficient stock of PhD economists in Belgium and Brussels. |
Precedent |
Good. The most common model applied to climate analysis and emissions trading schemes. Precedents in eQuest, GEM-E3. HMRC CGE model in the United Kingdom, Finland, Denmark. |
Opinion |
N/A. Model still in preliminary stage of development. CGE models have limitations but are generally considered the only tool capable of the time of climate and emissions trading analysis for which the Bureau intends to build capacity. |
2.6. Recommendations
During interviews with analysts and stakeholders it was clear that the Bureau has many strengths. For example:
The Bureau has renowned expertise internally and productive relationships with external experts in both the academic and practitioner communities.
It has unparalleled data access and relationships with government ministries compared to other OECD fiscal institutions stemming from its engagement in Belgium’s statistical framework and its role in directly supporting government departments, committees, and other stakeholders. More generally, the administrative and survey data available in Belgium facilitates economic and fiscal models would make researchers in other countries envious.
Analysts at the Bureau have leveraged their expertise, data, and relationships to develop a remarkable suite of sophisticated models and have cemented the institution as a leader in the modelling community.
Even institutions at the leading edge of policy analysis can learn from outside views and fresh perspectives. In that spirit, the review team identified several areas where the Bureau could benefit from reviewing its practices and receiving support from its peers in the OECD Working Party of Parliamentary Budget Officials and Independent Fiscal Institutions.
2.6.1. Summary of overall recommendations for model coverage and workflow
1. The institution’s strength — its collection of sophisticated models — may become a weakness. The strict commitment to rigid supermodel frameworks means long lead times to fulfil requests in new areas of analysis. Where other fiscal institutions in some cases prefer to fulfil novel requests in a quick manner, often within 48 hours, the Bureau generally prefers several months or even years of lead time to analysis to the same sophistication it has grown accustomed. In all cases, there should be a reflection on the potential gain in terms of precision and accuracy from additional months of development.
Further, sophisticated modelling can miss simple but valuable analysis. For example, IFIs commonly introduce behavioural adjustments from the academic literature to make simple calculations that may not have the depth of a microsimulation model run on administrative data but may ultimately provide reasonable results.
The Bureau also overestimates the sophistication of stakeholders and their ability to understand the models and commit their purpose to memory. For example, on the election platform website and publications, there are references to model names, which – when seen out of context - may be confusing for non-technical stakeholders.
Finally, with such gold-standard hammers, all analysis tends to become nails—policies that may not be worth an exhaustive assessment of every facet of economic consequences receive full treatment, taking analytical effort away from areas where it may be better prioritised.
Recommendation: As a sense test and communication device for cost estimates, the Bureau should make a habit of accompanying the results of its showpiece models with high-level “back-of-the-envelope” calculations of the financial costs with simple behavioural responses taken from the literature to make their results more transparent and intuitive for outside stakeholders to replicate. It could also provide more information and detail on the financial costs of policies as part of its election platform costings. For example, rather than only showing the ultimate impact on the budget deficits, it could show each offsetting line of higher revenues or expenses, the different affected budget line categories, and any breakdowns of more granular costs and revenues underlying the ultimate budget impact.
2. The Bureau has undertaken many more open-sourced projects than stakeholders and peers may be aware of.
Recommendation: The Bureau should promote its existing open-source tools and models and continue to proactively make its work available to the public. It should advertise a Bureau-branded GitHub repository and participate in or host conferences, code camps in collaboration with universities and think tanks, and do more modelling community outreach. These efforts can pay dividends in forming relationships that result in other people contributing to model refinements that the Bureau can then leverage. The Bureau should continue to transition its models to open-source collaborative working approaches with modern software with widespread use in the economic community.
3. The Bureau once was a leader in the multi-country HERMES Club initiated by EU institutions to jointly address challenges of the 1980s energy and inflation crisis. With the European Union facing similar conditions again, it may be worth revisiting that project. The world has changed, with better modelling software options, better harmonisation of national accounts, and a more connected EU with greater institutional support. Further, large-scale macroeconometric models are regaining favour in the academic and practitioner community.
Recommendation: The Bureau should explore the appetite among institutions in other countries for restarting the HERMES project. If there is sufficient demand, it could look for partners in other domestic institutions or international organisation for funding to explore converting HERMES to open-sourced software hosted on GitHub for a collaborative modelling initiative across European Member States.
4. The Bureau does not devote as much attention to the measurement of the business cycle and cyclical budget as other jurisdictions. This is largely because the business cycle in Belgium is subject to more limited variation, making the issue of measuring the output gap less important. However, Belgium is qualified by the European Commission as a “high risk country” in terms of public debt and the issue may come under the spotlight if Belgium was to be under the Excessive Deficit Procedure.
Recommendation: To get ahead of potential future contention, the Bureau could explore alternative tools to assess the output gap from different perspectives, such as the heat maps used by peers, or the suite of model averaging used by the Irish Fiscal Advisory Council.
5. The Bureau does not currently have statistical forecasting tools as a sense check for its structural models. For example, dynamic factor analysis for nowcasting, or vector autoregression models (VARs) for short-run forecasts up to eight quarters.
Recommendation: The Bureau should explore ready-made or out of the box dynamic factor models and simple reduced-form VARs to serve as a sense check for its structural models and for adding nowcasting capacity to its model suite.
2.6.2. Summary of key recommendations for individual models
HERMES. The Bureau should review the theoretical basis for using futures markets quotations as short-run forecasts for oil prices, exchange rates and interest rates—a practice that is common but has a poor theoretical justification and poor forecasting track record.
EXPEDITION and TYPECAST. The Bureau should continue to develop EXPEDITION and invest in the following areas to improve its usefulness for policy analysis: (1) Work with data providers to refine the co-ordination process, (2) Improve the model’s link to HINT to leverage the greater detail that HINT will produce in the next election, which will take place in the context of a sustained period of significant price volatility, (3) Work with STATBEL to compare and contrast its analysis on disposable income, (4) Review construction of the weights and matching to uprate and reweight tax benefit years to bring them forward to more recent years, (5) Support the main model with satellite models to adjust results for implementation date and policy year, (6) Explore how to present the results of TYPECAST in a more comprehensive and comprehendible way for a broader audience, (7) Explore a link with the LASER labour supply model, (8) Refine the model so it has the capacity to simulate policies in more detail for wealth taxes, personal income taxes, and regional policies.
HINT. The Bureau should explore additional links to environmental taxes, more direct interaction with EXPEDITION, and options to refine income quartiles, which are not immediately comparable to other models which present results as quintiles and deciles. The Bureau should explore satellite models that could impose assumptions on own-price elasticities of demand and cross-price elasticities that could be used to complement or refine the results of HINT.
PLANET. The Bureau should continue to invest in a solution to redesign the freight transport module and adopt an approach more in line with physical flows and less constrained by a theoretical allocation of economic flows. This would allow improved modelling of mode choices and geographical influences of freight transport demand. They should leverage their close connection with the network of Belgium statistical agencies to fill any data gaps to address the gap in freight transport modelling capacity.
LASER. As the Bureau continues to improve LASER, it should study the possibility of (1) expanding the target population with more types of workers, such as unemployed and self-employed, (2) linking administrative data to data from the labour force survey, and (3) using LASER’s elasticities in the new environmental CGE model.
References
[4] Bassilière, D. et al. (2013), A new version of the HERMES model - HERMES III.
[3] Bassilière, D. et al. (2018), Description and use of HERMES: : Document drafted in the framework of preparing for the 2019 costing of electoral programs, https://www.plan.be/uploaded/documents/201901110952260.WP-1-DC2019_HERMES_11843_F.pdf.
[2] Bassilière, D., L. Dobbelaere and F. Vanhorebeek (2018), How the HERMES model works - Description using variants, https://www.plan.be/publications/publication-1822-fr-le_fonctionnement_du_modele_hermes_description_a_l_aide_de_variantes.
[9] Baudewyns, D., A. Dewatripont and P. Michiels (2020), Labor cost reduction measures: what is the effect on employment and public finances in the Brussels Region?, https://www.plan.be/publications/article-2013-fr-les_mesures_de_reduction_du_cout_du_travail_quel_effet_sur_l_emploi_et_les_finances_publiques_en_region_bruxelloise.
[7] Baudewyns, D. and V. Lutgen (2022), How the HERMREG bottom-up model works: A description using variants, https://www.plan.be/uploaded/documents/202202030749560.WP_2202_12562_F.pdf.
[6] Baudewyns, D. and V. Lutgen (2022), The HERMREG bottom-up model: A multiregional model of the Belgian economy, https://www.plan.be/uploaded/documents/202202030739580.WP_2201_12561_F.pdf.
[10] Biatour, B. et al. (2018), Description of the QUEST III R&D model, https://www.plan.be/publications/publication-1848-fr-description_du_modele_quest_iii_r_d.
[12] Biatour, B. et al. (2017), Public Investment in Belgium - Current State and Economic Impact, https://www.plan.be/publications/publication-1650-fr-public_investment_in_belgium_current_state_economic_impact.
[25] Bureau, F. (2022), Economic Budget – Economic forecasts 2022-23 for September 2022, Federal Planning Bureau, https://www.plan.be/publications/publication-2283-fr-budget_economique_previsions_economiques_2022_2023_de_septembre_2022.
[1] Communities, C. (ed.) (1993), HERMES, Elsevier, https://doi.org/10.1016/C2009-0-10171-X.
[32] Daubresse, C. et al. (2018), Description and use of the PLANET model, https://www.plan.be/uploaded/documents/201901111505430.WP-6-DC2019_PLANET_11848_F.pdf.
[31] Daubresse, C. and B. Laine (2020), Telework and transport demand: an evaluation in the PLANET model, https://www.plan.be/publications/publication-2059-fr-teletravail_et_demande_de_transport_une_evaluation_dans_le_modele_planet.
[30] Daubresse, C. and B. Laine (2020), The PLANET Model: Methodological Report, PLANET 4.0, https://www.plan.be/publications/publication-1967-fr-the_planet_model_methodological_report_planet_4_0.
[23] De Ketelbutter, B. et al. (2014), A new version of MODTRIM II, https://www.plan.be/uploaded/documents/201407111401360.WP_1405_10748.pdf.
[16] De Vil, G. et al. (2018), Description and use of the EXPEDITION model, Federal Planning Bureau, https://www.plan.be/publications/publication-1849-fr-description_et_utilisation_du_modele_expedition.
[19] Dekkers, G., R. De smet and K. Van den Bosch (2023), MIDAS 2.0: Revision of a dynamic microsimulation model, Federal Planning Bureau, https://www.plan.be/uploaded/documents/202301240701580.WP_2302_12752.pdf.
[20] Dekkers, G., R. Desmet and G. De Vil (2010), The long-term adequacy of the Belgian public pension system: An analysis based on the MIDAS model, https://www.plan.be/publications/publication-946-fr-the_long_term_adequacy_of_the_belgian_public_pension_system_an_analysis_based_on_the_midas_model.
[34] Devogelaer, D. (2018), Description and use of SUPER CRYSTAL GRID, https://www.plan.be/uploaded/documents/201901111032530.WP-5-DC2019_CrystalSG_11847_F.pdf.
[5] Federal Planning Bureau (2021), Macroeconomic and fiscal effects of the draft National Recovery and Resilience Plan: Report to the Secretary of State for Recovery and Strategic Investments, https://www.plan.be/publications/publication-2106-en-macroeconomic_fiscal_effects_of_the_draft_national_recovery_resilience_plan_report_to_the_secretary_of_state_for_recovery.
[28] Federal Planning Bureau (2006), Tools and methods used at the Federal Planning Bureau, Federal Planning Bureau, https://www.plan.be/uploaded/documents/200610230931210.wp0607_en.pdf.
[26] Geerts, J., K. Van den Bosch and P. Willemé (2018), Description and use of the PROMES model, https://www.plan.be/publications/publication-1850-fr-description_et_utilisation_du_modele_promes.
[33] Gusbin, D. et al. (2010), The PLANET model methodological report: Modelling of short sea shipping and bus-tram-metro, https://www.plan.be/uploaded/documents/201007060844530.wp201016.pdf.
[24] Hertveldt, B. and I. Lebrun (2003), MODTRIM II: a quarterly model for the Belgium economy.
[29] High Council of FInance (2020), Outlook 2019-2070: a sharp increase in social spending until 2040 but a decrease in the risk of poverty of pensioners, https://www.plan.be/press/communique-2019-fr-perspectives_2019_2070_une_augmentation_prononcee_.
[22] High Council of Finance (2022), Annual report of the Study Committee on Ageing, https://www.plan.be/publications/publication-2263-en-study_committee_on_ageing_yearly_report_2022.
[8] Hoorelbeke, D. et al. (2007), HERMREG: A regionalisation model for Belgium, https://ecomod.net/sites/default/files/document-conference/ecomod2007/93.pdf.
[11] Kegels, C. and D. Verwerft (2018), Economic impact of professional services reform in Belgium - A DSGE simulation, https://www.plan.be/publications/publication-1798-fr-economic_impact_of_professional_services_reform_in_belgium_a_dsge_simulation.
[35] Korinek, A. (2015), Thoughts on DSGE Macroeconomics: Matching the Moment, But Missing the Point?, https://www.ineteconomics.org/uploads/downloads/Korinek-DSGE-Macro-Essay.pdf.
[27] Lefèvre, M. and S. Gerkens (2021), Assessing the sustainability of the Belgian health system using projections, https://kce.fgov.be/en/evaluatie-van-de-duurzaamheid-van-het-belgische-gezondheidssysteem-op-basis-van-projecties-1.
[15] Liu, E. and S. Ma (2022), Innovation Networks and R&D Allocation.
[18] Nevejan, H. (2021), Regional Child Benefit Reforms - An Impact Analysis Using the EXPEDITION Microsimulation Model, https://www.plan.be/uploaded/documents/202105180544440.WP_2104_12405.pdf.
[13] Rotemburg, J. (1982), “Monopolistic Price Adjustment and Aggregate Output”, Review of Economic Studies, pp. 517-531.
[14] Smets, F. and R. Wouters (2003), “An estimated dynamic stochastic general equilibrium model of the euro area”, Journal of the European Economic Association, Vol. 1, pp. 1123-1175.
[21] Social Protection Committee (SPC) and the European Commission (DG EMPL) (2021), 2021 Pension adequacy report, https://op.europa.eu/en/publication-detail/-/publication/4ee6cadd-cd83-11eb-ac72-01aa75ed71a1.
[36] Stiglitz, J. (2017), Where Modern Macroeconomics Went Wrong, NBER Working Paper 23795, https://www.nber.org/papers/w23795.
[17] Thuy, Y., G. Van Camp and D. Vandelannoote (2020), COVID-19 Crisis: A simulation of the impact of the loss of wages for temporary unemployment in the case of force majeure and the loss of income in the case of bridging rights, pp. 149-174, https://socialsecurity.belgium.be/sites/default/files/content/docs/nl/publicaties/btsz/2020/btsz-2020-1-covid-19-crisis-simulatie-impact-van-het-loonverlies-en-het-inkomensverlies.pdf.
Notes
← 1. Although the Bureau does compute these statistics and publishes them occasionally. See for instance - https://www.plan.be/uploaded/documents/201902280925280.PP_117_11866_F.pdf
← 2. Initially "Intégrateur d'outils de développement économétrique”