# Research credits and the scale of French business innovation

Policy Brief · France · Innovation Policy

A research tax credit reduces the cost of eligible expenditure. Its policy return depends on whether that reduction causes additional research, changes its quality or location, and generates benefits beyond those retained by the recipient. The amount claimed is a measure of support, not an estimate of the innovation it causes.

France's crédit d'impôt recherche, or CIR, makes this distinction especially important because firms differ in size, financing, and research organization. A large group with an established research program and a small firm financing its first technical team can receive support through the same instrument while responding differently to it.

## Eligibility defines the tax base, not the additional research

The tax administration's [CIR eligibility guidance, updated October 15, 2026](https://www.impots.gouv.fr/professionnel/questions/puis-je-pretendre-au-credit-impot-recherche), identifies scientific and technical research categories and qualifying expenditures. The definition of qualifying R&D should be distinguished from the separate treatment of eligible innovation expenditure for smaller companies. A commercial improvement is not automatically qualifying research merely because the company describes it as innovative.

The relevant claim depends on documented activity, applicable costs, and the rules for the expenditure year. A published general rate cannot establish the credit available to a particular project without that assessment. Changes in eligible cost categories can also alter claims even when the underlying research effort is unchanged.

For evaluation, the tax base is an observed administrative outcome. Additional research is the difference between actual research and what would have occurred without the support. The two can differ substantially. A company may claim a credit for a project it would have undertaken in full, while another may expand its program because the net cost falls.

## The evidence differs by firm size and research history

France Stratégie's [2021 CNEPI evaluation](https://www.strategie-plan.gouv.fr/publications/evaluation-credit-dimpot-recherche-rapport-cnepi-2021) reports heterogeneous findings for the 2008 reform. Among firms already using the scheme, it identifies statistically significant effects for microenterprises and SMEs, while effects for the combined group of larger enterprises were not statistically significant. It also distinguishes descriptive gains among later entrants from a confirmed causal relationship.

That finding should retain its study population and reform period. It does not prove that every small company responds strongly, that larger firms never respond, or that the current scheme has identical effects. Statistical insignificance is not a demonstration that the true effect is exactly zero. The result does, however, make a uniform response assumption difficult to justify.

The Treasury's [September 2021 account of the reform evaluation](https://www.tresor.economie.gouv.fr/Articles/2021/09/20/evaluation-de-la-reforme-du-credit-impot-recherche-de-2008) likewise distinguishes the microeconomic evidence from model-based macroeconomic effects. A simulated increase in activity depends on assumptions about additional R&D, its returns, and the financing of support. It should not be presented as an observed GDP increment attributable directly to tax records.

## The financing mechanism needs to be specified

A credit can affect investment through expected project returns and through cash availability. Those channels differ when a firm has little current taxable profit or faces a delay before the benefit is realized. Administrative certainty and the timing of refunds can therefore matter alongside the statutory incentive.

Consider a hypothetical small research project costing one million euros. A tax benefit expected after completion reduces its eventual net cost, but the firm may still need to finance the full amount during development. If external funds are unavailable, a generous eventual credit may fail to resolve the immediate constraint. A larger firm using internal funds may face a different problem: the subsidy can improve the return on a project already planned.

The example does not establish a response coefficient for either group. It identifies why liquidity, profitability, and project timing should enter the evaluation. Firm size is a useful grouping variable but should not substitute for direct evidence of the financing constraint.

## More expenditure and more innovation are different outcomes

R&D expenditure can rise because researchers work more hours, more staff are employed, wages increase, or costs are reclassified. The real research input should be distinguished from the nominal tax base. When specialized labor supply is limited, some additional support can affect compensation rather than the number of projects conducted.

Innovation outcomes also arrive on different schedules. A technical experiment may fail commercially while producing useful knowledge. Patents provide information about some inventions but omit others and vary in economic value. Sales from new products, process improvements, productivity, and spillovers answer different questions and should not be compressed into one unsupported success rate.

A firm-level benefit is not the same as a social benefit. Knowledge can reach suppliers, workers, competitors, or other researchers, while support can also displace research elsewhere. The relevant appraisal considers those wider effects and the public financing cost. A rise in a recipient's turnover is insufficient to establish the net national return.

## Groups and research locations complicate the unit of analysis

A legal entity making a claim can belong to a group that moves expenditure or personnel between entities. Evaluation should identify whether the measured increase represents additional group research, a change in where it is reported, or relocation from another country. These changes have different implications for domestic employment and global knowledge creation.

The baseline should also distinguish firms entering the scheme from continuing claimants. Entry may coincide with growth, improved reporting, or a newly qualifying project. Comparing recipients with all nonrecipients without accounting for those differences can assign preexisting dynamism to the credit.

Credible evidence would follow research inputs and outcomes over time, use an appropriate comparison, and report uncertainty. Where a policy threshold supplies a useful comparison, the analysis must still examine sorting and behavior around it. A statistical design is persuasive only if its assumptions fit the actual rules and firms.

The French evidence supports a differentiated assessment. CIR claims establish the scale and allocation of support; evaluation establishes whether additional research and wider benefits follow. Policy conclusions should preserve the observed differences by firm size and research history and avoid treating every euro of eligible expenditure as an additional euro of innovation.

---

American Economic Senate