Methodological approach:
To estimate the supply-side subsidies that EU member state governments provide for industrial policies, we use data provided in EU state aid transparency registers. The register data contains individual state aid award data, in compliance with the European transparency requirements. Member states are required to report most types of individual state aid above certain thresholds (for most legal bases, the threshold is at €500 000 or €100 000). By summing individual entries of the transparency register, total supply-side state aid can be approximated.
To filter for state aid relevant for industrial policies, we use the state aid decisions under which the individual state aid was approved. We can link them because the transparency register data includes the corresponding state aid decision number for each award. To classify state aid cases into industrial policy areas, we use a large language model to analyze the state aid decision text. For this purpose, we defined a number of industrial policy categories that capture most of the proclaimed EU industrial policy priorities: “Microelectronics / Semiconductors”, “Clean Tech” (further divided into various subcategories), “Biotech and Pharma”, “Defense/Military”, “Space”, “Software development”, and “Robotics, Internet of Things, 3D printing”. To improve the accuracy of classification, we also defined some other common state aid objectives (see table below), and provided a description for each category, refined in an iterative process and tailored to the specifics of the employed large language model (ChatGPT-4o and 4o-mini). Aid amounts are converted to 2021 prices.
Based on a small sample, the accuracy of the LLM in assigning the correct sector based on the state aid decision text was about 92% (i.e. in 92% of the random sample the LLM assigned the same sector category as a human evaluator. Note that this excludes state aid cases classified as COVID aid, for which accuracy is lower; however, this has no effect on the data, since COVID aid is excluded from the sectoral totals (see above)). About 1000 state aid decision texts were fully analyzed by the LLM. In some cases, the text was too short or contained too few information to infer the sector (especially relevant for amendments of existing state aid cases).
State aid disbursed to cushion the economic impact of Covid is considered not to constitute industrial policy in the context of this analysis. Hence, if the large language model classified the state aid case to be primarily to cushion the economic impact of the pandemic, the associated aid awards are not classified as industrial policy, even if the supported sector is among the sectors defined as being among the EU industrial policy priorities.
Data sources:
The state aid data, slightly more than 280 000 individual aid awards, was downloaded from the registries on 08.01.2025.
The state aid decision texts were last accessed on 08.01.2025, i.e. decision texts published since then where not analysed by the LLM.
The data can be provided upon request.
Comparison with NACE-codes:
For each state aid award, the transparency register already contains the NACE2 codes of the beneficiary company. However, the NACE-codes are inconsistently assigned by member states, and, given the coarse nature of some NACE categories, often not suitable to distinguish some industrial policy sectors from others (for instance, NACE category 26.11, “Manufacture of electronic components”, contains manufacturing of semi-conductors as well as manufacturing of some solar PV components). Hence, the NACE-codes were only used for some consistency checks.
Comparison with state aid scoreboard:
The EU state aid scoreboard contains aggregated data of state aid expenditure. As such, it is another suitable source to derive insights about industrial policy subsidies. The quality of the data is likely higher than in the transparency register, and subject to more rigorous checks. The scoreboard contains state aid expenditure, whereas the transparency register contains the granting of aid. This is one source of discrepancy between the two data sets, since the granting of aid can result in state expenditure in later years (e.g. if disbursements take place in several tranches stretched over multiple years). Another important source of discrepancy between the scoreboard and the transparency register consists in the amounts covered. Whereas the scoreboard includes all individual aid amounts, the transparency register only includes aid above certain thresholds (see above). Hence, the transparency register can typically only capture supply-side industrial policy subsidies, whereas the scoreboard also captures demand-side measures. However, the data provided in the scoreboard has a substantial time lag; the last edition, published in April 2024, contains the expenditure only up until 2022. Data for 2023 will only be available in Spring 2025. Moreover, the scoreboard aid objectives are relatively coarse, and do not allow to properly identify and delineate different industrial policy objectives. In the most disaggregated data set made available (“Disaggregated dataset for statistics (2000-2022)*), some state aid expenditure relevant for industrial policy is not included (e.g. IPCEIs) for confidentiality reasons.
Limitations:
Limitations of the raw transparency register data:
Limitations of the approach:
| Category (colored green if considered EU industrial policy priority) | Description |
|---|---|
| Microelectronics / Semiconductors | If the aid primarily and specifically supports the manufacturing of microelectronics or semiconductors, or research on them. Also assign this category if the aid supports hardware for quantum computing. |
| Biotech and Pharma | Includes CRISPR, new vaccines, medicines, manufacturing products needed for medicines, and other life sciences innovations crucial for health, agriculture, and industrial applications. |
| Clean Tech: Solar components | If the aid primarily and specifically supports the manufacturing of solar components, including PV (solar cells, solar modules, polysilicon, wafers) and thermal collectors, or research on them. Do NOT assign this category if it is about the generation of solar energy, e.g., support for building or operating solar parks. |
| Clean Tech: Hydrogen | If the aid primarily and specifically supports hydrogen production, electrolyzer manufacturers, hydrogen cars, or hydrogen research, or support for innovative electric fuels. DO NOT assign this category if it is about aid for biofuels, for instance from biomass or garbage. |
| Clean Tech: Decarbonizing energy-intensive manufacturing | If the aid primarily and specifically supports the shift to clean/renewable production in energy-intensive industries, such as decarbonizing steel, decarbonizing cement, etc. DO NOT assign this category if the aid is about compensation for indirect emission costs. |
| Clean Tech: Electric vehicles | If the aid primarily and specifically supports EV manufacturing or research on EVs. If it is about EV batteries, do NOT assign this category; instead, use the 'Clean Tech: Batteries' category. |
| Clean Tech: Wind energy components | If the aid primarily and specifically supports the manufacturing of wind energy components (such as wind turbines, nacelles, blades, etc.), or research in these areas. Do not assign this category if the aid supports the generation of wind energy directly, e.g., by supporting wind parks. |
| Clean Tech: Batteries and other energy storage | If the aid primarily and specifically supports the manufacturing of batteries and other types of energy storage, or research on these topics. |
| Clean Tech: Innovative nuclear power | If the aid primarily and specifically supports the manufacturing of components needed in the nuclear energy value chain, or related research. Assign this category if the aid is about nuclear fusion, small modular reactors, or research on nuclear power. DO NOT assign this category if it is about traditional nuclear power production. |
| Clean Tech: CCU / CCS | If the aid primarily and specifically supports carbon capture and use, and carbon capture and storage. |
| Clean Tech: Other clean manufacturing activities aimed at decarbonisation | This category should be used when the aid is specifically and primarily for a clean manufacturing activity that does not fit into the more specific clean tech categories. Do not assign this category if the aid is for general environmental cleanup, site remediation, or projects not directly related to clean technologies. |
| Space | If the aid primarily supports research and production processes required for value chains around space, for instance, R&D aid for rockets or satellites. |
| Software development | If the aid primarily and specifically supports software development. Also assign this category if the aid supports artificial intelligence development/applications. |
| Robotics, Internet of Things, 3D printing | If the aid primarily and specifically supports the advancement or production of robotics, the Internet of Things, 3D printing, or other highly advanced manufacturing. DO NOT assign this category if it is about a clean technology. |
| Rail | If the aid primarily and specifically supports the shift to railways, and the decarbonization of rail. If it is specifically about using hydrogen to decarbonize rail, classify instead as 'Clean Tech: Hydrogen.' If it is about supporting rail generally, but not about its electrification or sustainability, do not assign this category. |
| Clean energy production | If aid is given directly for the generation of clean energy, e.g., for producing electricity from solar or wind energy, or other clean sources. This includes support for building or operating facilities that generate clean energy. Do NOT assign this category if the aid is for reducing energy costs for energy-intensive industries. |
| Agriculture | If the aid exclusively and specifically supports agricultural or fishery activities. DO NOT assign this category if agriculture/fishery is just one of the possible beneficiaries, and companies active in other areas are also eligible for aid. |
| Defense/Military | If the aid primarily supports military capacities. |
| Combustion engine / Traditional automotive | If the aid primarily supports carmakers (or their suppliers) engaged in producing cars with combustion engines. If the aid concerns carmakers but it is not clearly stated that the aid is for zero-emission or low-emission cars, assign this category. |
| Telecommunications | If the aid primarily and specifically supports economic activities directly linked to telecommunications. |
| Energy cost support for households | If the aid primarily reduces energy costs for households, for instance (but not limited to) by capping gas and electricity prices, or by giving lump-sum payments. |
| Energy cost support for companies | If the aid primarily reduces energy costs for companies, for instance (but not limited to) by capping their gas and electricity prices, or by giving lump-sum payments. |
| Compensation for Indirect Emission Costs | If the aid compensates companies for high electricity prices due to the inclusion of greenhouse gas emission costs in the EU Emissions Trading System (ETS), commonly referred to as "indirect emission costs." |
| Fossil Energy Production | If the aid supports energy use from fossil fuels, such as coal, gas or oil. DO NOT assign this category if it is aid primarily given because of the Russian war in Ukraine and the resulting price spikes. |
| University research support | Aid for universities to conduct research. |
| Other: XXXX | If none of the above categories fit well, choose this 'Other: XXXX' category and describe very briefly the economic activity/support sector. |
| Error because multiple | If the document contains multiple, distinct new state aid cases or amendments that cover substantially different sectors and hence would fall into multiple categories. |
| Aid instrument | Discount factor applied to nominal amount |
|---|---|
|
85% |
|
90% |
|
No discount |