Periodic Reporting for period 1 - BASEforHANK (A software tool for the Bayesian estimation of Heterogeneous Agent macroeconomic policy models)
Berichtszeitraum: 2024-09-01 bis 2026-02-28
Modern central banks, finance ministries, and research institutions rely on macroeconomic models to assess interest-rate decisions, fiscal packages, energy-price shocks, and other major policy interventions. However, the standard software used for this purpose was largely designed for models with one representative household or firm. This makes it difficult to analyse how policies affect different households, sectors, or countries differently, and therefore limits the ability to study inequality, redistribution, and social cohesion together with inflation, employment, and output.
BASEforHANK was designed to address this gap. Its overall objective was to turn frontier ERC research on heterogeneous-agent macroeconomic models into an open-source toolbox that can be used more broadly by policy institutions, researchers, and advanced students. In these models, households differ in income, wealth, debt, and exposure to shocks. This makes it possible to study not only whether a policy stabilises the economy as a whole, but also who benefits, who bears the costs, and how these distributional effects feed back into aggregate outcomes. The project is therefore firmly rooted in the social sciences, especially economics, because it translates research on inequality, household behaviour, and public policy into a practical analytical tool.
The project aimed to make this type of analysis easier, faster, and more transparent. Key objectives were to develop a Julia-based toolbox with a workflow familiar to users of standard macroeconomic software, improve the software architecture, reduce barriers to use, strengthen the numerical methods needed to solve large-scale models, and keep the toolbox openly available as a public resource rather than a proprietary product.
The expected impact is that policy institutions and researchers can move beyond “average-agent” analysis towards a more realistic understanding of how economic shocks and policy measures affect society. Better tools for modelling inequality and the business cycle can improve the design of monetary, fiscal, and crisis-response policies, especially where distributional effects matter strongly. Because the toolbox is open source, the project also lowers entry barriers, supports transparency and reproducibility, and creates scope for a wider community to contribute to future development.
BASEforHANK was designed to address this gap. Its overall objective was to turn frontier ERC research on heterogeneous-agent macroeconomic models into an open-source toolbox that can be used more broadly by policy institutions, researchers, and advanced students. In these models, households differ in income, wealth, debt, and exposure to shocks. This makes it possible to study not only whether a policy stabilises the economy as a whole, but also who benefits, who bears the costs, and how these distributional effects feed back into aggregate outcomes. The project is therefore firmly rooted in the social sciences, especially economics, because it translates research on inequality, household behaviour, and public policy into a practical analytical tool.
The project aimed to make this type of analysis easier, faster, and more transparent. Key objectives were to develop a Julia-based toolbox with a workflow familiar to users of standard macroeconomic software, improve the software architecture, reduce barriers to use, strengthen the numerical methods needed to solve large-scale models, and keep the toolbox openly available as a public resource rather than a proprietary product.
The expected impact is that policy institutions and researchers can move beyond “average-agent” analysis towards a more realistic understanding of how economic shocks and policy measures affect society. Better tools for modelling inequality and the business cycle can improve the design of monetary, fiscal, and crisis-response policies, especially where distributional effects matter strongly. Because the toolbox is open source, the project also lowers entry barriers, supports transparency and reproducibility, and creates scope for a wider community to contribute to future development.
The project focused on turning recent advances in heterogeneous-agent macroeconomics into a more usable and robust software environment for model development, solution, and estimation. Because such models capture differences across households, firms, sectors, or countries, they are much richer than standard macroeconomic models but also much more demanding computationally. The work therefore combined software development with methodological research on numerical solution techniques.
One major activity was to improve the parser that translates model descriptions into executable code. By the end of the project, the parser could handle a broader range of inputs and automatically generate repeated code lines for replicated equation blocks. This is particularly useful in larger applications with multiple industries, sectors, regions, or countries.
A second major activity was to reorganise the internal structure of the toolbox. We introduced a clearer separation between the general package infrastructure and the components that are specific to an individual economic model. This made the code base easier to maintain, extend, and reuse, and also improved reproducibility by distinguishing more clearly between model-specific elements and general numerical routines.
A third activity concerned model reduction methods, which are essential for making large heterogeneous-agent models computationally feasible. We explored more flexible first-stage reduction strategies inspired by recent methodological work. This produced an important scientific result: in our practical setting, the alternative approach turned out to be less powerful than initially expected, because it is better suited to frameworks that work directly with value functions than to frameworks that work mainly with their derivatives. This clarified the limits of a promising route and helped focus future work on methods that are more robust in practice.
Beyond the tasks originally foreseen, the project also implemented higher-order perturbation methods and advanced the treatment of non-linear distribution dynamics. A key outcome was the DEGM approach reflected in the Journal of Monetary Economics article "An endogenous gridpoint method for distributional dynamics". This goes beyond the previous state of the art for solving distributional dynamics in heterogeneous-agent settings.
By the end of the project, the main outcomes were an improved parser, a more modular software architecture, a clearer assessment of alternative reduction methods, and methodological advances for higher-order and non-linear dynamics, including DEGM.
One major activity was to improve the parser that translates model descriptions into executable code. By the end of the project, the parser could handle a broader range of inputs and automatically generate repeated code lines for replicated equation blocks. This is particularly useful in larger applications with multiple industries, sectors, regions, or countries.
A second major activity was to reorganise the internal structure of the toolbox. We introduced a clearer separation between the general package infrastructure and the components that are specific to an individual economic model. This made the code base easier to maintain, extend, and reuse, and also improved reproducibility by distinguishing more clearly between model-specific elements and general numerical routines.
A third activity concerned model reduction methods, which are essential for making large heterogeneous-agent models computationally feasible. We explored more flexible first-stage reduction strategies inspired by recent methodological work. This produced an important scientific result: in our practical setting, the alternative approach turned out to be less powerful than initially expected, because it is better suited to frameworks that work directly with value functions than to frameworks that work mainly with their derivatives. This clarified the limits of a promising route and helped focus future work on methods that are more robust in practice.
Beyond the tasks originally foreseen, the project also implemented higher-order perturbation methods and advanced the treatment of non-linear distribution dynamics. A key outcome was the DEGM approach reflected in the Journal of Monetary Economics article "An endogenous gridpoint method for distributional dynamics". This goes beyond the previous state of the art for solving distributional dynamics in heterogeneous-agent settings.
By the end of the project, the main outcomes were an improved parser, a more modular software architecture, a clearer assessment of alternative reduction methods, and methodological advances for higher-order and non-linear dynamics, including DEGM.
The project delivered several results that go beyond the previous state of the art in computational macroeconomics. Its central achievement was to strengthen an open-source software environment for building, solving, and estimating heterogeneous-agent macroeconomic models, that is, models that can capture how economic shocks and policy measures affect different households, sectors, or countries differently. Compared with standard tools designed mainly for representative-agent models, this is an important step forward because it allows policy analysis to combine aggregate outcomes with distributional effects.
Several project results are clearly beyond the previous state of the art. The software architecture was made more modular, improving transparency, extensibility, and reproducibility. The parser was substantially improved and can now automatically generate repeated code lines for replicated equation blocks, which is valuable for larger multi-sector, multi-country, or multi-region applications. The project also implemented higher-order perturbation methods and advanced the numerical treatment of non-linear distribution dynamics. In particular, it contributed the DEGM approach reflected in the Journal of Monetary Economics article "An endogenous gridpoint method for distributional dynamics", which expands the set of heterogeneous-agent problems that can be treated accurately and computationally feasibly.
The project also generated an important scientific result by clarifying the limits of one promising model-reduction route. In our practical setting, this approach turned out to be less powerful than initially expected, which helps focus future work on the strategies that are most robust and useful in practice.
The potential impact of these results is substantial. Better computational tools for heterogeneous-agent models can improve the evidence base for monetary, fiscal, and crisis-response policies, especially when distributional effects matter strongly. In the longer run, these results can help establish a shared open modelling infrastructure for this field.
To ensure further uptake and long-term success, several needs remain important: continued research on robust model-reduction methods and non-linear distributional dynamics; continued software development, documentation, and user support; demonstration through practical policy applications; training for users; and sustained financial support for maintenance and further development.
In overview, the main results of the project were an improved parser, a more modular software architecture, progress on higher-order perturbation methods, beyond-state-of-the-art advances in non-linear distribution dynamics including DEGM, and a clearer understanding of the strengths and limits of alternative model-reduction strategies.
Several project results are clearly beyond the previous state of the art. The software architecture was made more modular, improving transparency, extensibility, and reproducibility. The parser was substantially improved and can now automatically generate repeated code lines for replicated equation blocks, which is valuable for larger multi-sector, multi-country, or multi-region applications. The project also implemented higher-order perturbation methods and advanced the numerical treatment of non-linear distribution dynamics. In particular, it contributed the DEGM approach reflected in the Journal of Monetary Economics article "An endogenous gridpoint method for distributional dynamics", which expands the set of heterogeneous-agent problems that can be treated accurately and computationally feasibly.
The project also generated an important scientific result by clarifying the limits of one promising model-reduction route. In our practical setting, this approach turned out to be less powerful than initially expected, which helps focus future work on the strategies that are most robust and useful in practice.
The potential impact of these results is substantial. Better computational tools for heterogeneous-agent models can improve the evidence base for monetary, fiscal, and crisis-response policies, especially when distributional effects matter strongly. In the longer run, these results can help establish a shared open modelling infrastructure for this field.
To ensure further uptake and long-term success, several needs remain important: continued research on robust model-reduction methods and non-linear distributional dynamics; continued software development, documentation, and user support; demonstration through practical policy applications; training for users; and sustained financial support for maintenance and further development.
In overview, the main results of the project were an improved parser, a more modular software architecture, progress on higher-order perturbation methods, beyond-state-of-the-art advances in non-linear distribution dynamics including DEGM, and a clearer understanding of the strengths and limits of alternative model-reduction strategies.