Europe’s innovation gap with the US and China lies in breakthrough rather than incremental innovation. This column argues that part of the explanation is the labour market. Breakthroughs come from risky experimentation and experiments often fail and cost workers their jobs. When job loss is costly, workers demand a wage premium to join risky young firms, and that premium acts as a tax on experimentation. Combining a new model, Danish matched employer–employee data and a randomised survey experiment, the authors show that more fluid labour markets lower this premium, raise experimentation, and lift aggregate productivity.

Europe is falling behind at the technological frontier. In Philippe Aghion’s words, Europe “remains stuck in mid-tech, incremental innovation, while the US and now China have moved into breakthrough, high-tech innovation” (Aghion and Johnson 2026). The Draghi report (Draghi 2024) and Fuest et al. (2024) describe a ‘middle-technology trap’, and European young firms are less dynamic than their US counterparts (Adilbish et al. 2025). Columns here on VoxEU have examined barriers to entry and the Single Market (Di Mauro et al. 2026), innovation funding (Fuest et al. 2024), and size-contingent labour regulation that chills incremental innovation (Aghion et al. 2018). The worker’s side – the cost of losing a job – has received far less attention.

It should not. Young firms that scale up account for a disproportionate share of job creation and productivity growth, but many exit quickly (Haltiwanger et al. 2013, 2016). Entrepreneurship is experimentation: most risky ventures fail and a few become very large (Kerr et al. 2014, Sterk et al. 2021). As Aghion puts it, “I don’t think you can have breakthrough innovation without accepting, and even encouraging, failure” (Aghion and Johnson 2026). Every experiment – in high tech, retail, or logistics – must recruit workers who bear the risk that the firm will close. Yet European labour markets remain far less fluid than the US (Schoefer 2025). About a quarter of the US unemployed find a job each month, while in France or Italy unemployment spells are long. Asked why they do not move to comparable, better-paid jobs, German workers most often answer that their current job is safer (Clymo et al. 2026). European labour market institutions, Schoefer argues, may have been designed for large manufacturers and incremental innovation – the very model that now puts Europe at a disadvantage.

In our new paper (Faccini et al. 2026), we ask a simple question: does the cost of losing a job shape how much experimentation an economy undertakes? We bring together theory, administrative data, and a randomised experiment, and the three lines of evidence point in the same direction.

A compensating differential for experimentation

We build a directed-search model of firm dynamics in which entrants choose between a safe business model and a risky, experimental one. Risky entrants face a high probability of failure but a chance of drawing very high permanent productivity; safe entrants do not. Workers understand that a job at an experimental firm is more likely to end, and that a lost job means a spell of unemployment. To attract workers, risky firms must therefore pay a compensating wage premium. The premium does not require workers to be risk averse; it reflects the expected cost of job loss. The worse the outside option, the larger the premium, and the more expensive experimentation becomes.

We calibrate the model to Danish data and use it as a laboratory. As a proxy for institutions that make the labour market more fluid, we raise the quarterly job-finding rate from 0.44 to 0.56. Because job loss becomes less costly, the wage that experimental entrants must offer falls relative to safe entrants (Figure 1, panel a). The share of entrants choosing the risky technology rises by 6.4 percentage points. Wages at young and mature firms rise by 1.0% and 1.6%, respectively, so workers share in the gains.

Aggregate total factor productivity (TFP) rises by about 1% in the new steady state (Figure 1, panel b). The gain comes mainly from a better productivity distribution, as more experiments produce more high-productivity survivors, and from improved allocation of workers towards more productive firms. Both outweigh a smaller mass of firms. This is a permanent increase in the level of productivity, not a change in its growth rate, but it is not trivial: discounted at 4%, the cumulative output gain over ten years amounts to about 8% of one year’s GDP.

Figure 1 Higher job-finding rates lower the experimentation premium and raise TFP

Notes: Steady-state comparison in the calibrated model when the quarterly job-finding rate rises from 0.44 to 0.56. Panel (a) shows wage changes by firm type; panel (b) decomposes the change in aggregate TFP. Source: Faccini et al. (2026), Tables 2 and 3.

Evidence from Danish register data

Experimentation is not directly observable, but its risk is concentrated among young firms and resolved as firms mature. We therefore use firm age as a proxy and test the model’s central prediction: where job finding is easier, the wage that young firms pay relative to mature firms should be lower. We use Danish matched employer–employee data covering 2008–2022 and exploit variation in job-finding rates across 29 commuting zones and over time.

The data confirm the prediction (Figure 2). A one-standard-deviation increase in the local job-finding rate lowers the young-firm wage by between 0.6% and 1.1% relative to mature firms. Our preferred estimate, –0.62%, compares the same worker across firms and the same firm across local labour market conditions, controlling for year effects and time-varying worker characteristics. The result survives every robustness check we have run, from restricting the sample to manufacturing or high-tech industries to alternative definitions of young firms and establishment-level estimates.

Figure 2 Young firms pay a smaller wage premium where job finding is easier

Notes: Coefficient on the interaction of a young-firm indicator (under three years old) with the standardised commuting-zone job-finding rate. Dependent variable: log real hourly wage. 95% confidence intervals. Workers aged 30–52, 2008–2022. Source: Faccini et al. (2026), Table 4 and Appendix B.

We do not give these estimates a causal interpretation. Job-finding rates are not randomly assigned across commuting zones, and local conditions may reflect unobserved features of labour demand.

A randomised survey experiment

To address causality directly, we designed a pre-registered randomised survey experiment, fielded in early 2026 through Danmarks Nationalbank’s Survey of Consumers’ Expectations to 2,556 respondents in permanent employment. Each respondent chooses between two otherwise identical jobs: one at a safer firm with a 4% annual closure probability, and one at a riskier young firm with a 20% closure probability and greater growth potential. We elicit the wage premium that makes the respondent indifferent between the two. Respondents are randomly told that, if they lose their job, finding a new one would take either half or twice as long as they themselves expect.

On average, respondents require a premium of 17.6% to join the riskier firm. Doubling expected unemployment duration raises the required premium by 2.3 percentage points. Because the duration shock is randomly assigned, this effect is causal. And doubling is not an extreme scenario: job-finding rates in France or Italy are about half of Denmark’s (Schoefer 2025). High-skilled respondents, the workers most relevant for innovative firms, expect similar unemployment durations and require a similar premium, so the channel is not confined to low-skilled workers.

Theory, data, and experiment line up

The three approaches are quantitatively consistent (Figure 3). When we administer the same experiment to workers in the model, doubling unemployment duration raises the risky–safe premium by 2.85 percentage points, inside the 95% confidence interval of the survey estimate. Separately, the model predicts that a one-percentage-point increase in the relative wage of young firms reduces their share of employment by 1.3 percentage points; the register data give 0.9.

Figure 3 The model’s predictions match the experimental and register evidence

Notes: Panel (a) compares the model’s response of the risky–safe wage premium to a doubling of unemployment duration with the survey estimate, with 95% confidence interval. Panel (b) compares the model’s response of the young-firm share to a one-percentage-point higher young-firm relative wage with the register estimate. Source: Faccini et al. (2026), Section 5.2 and Tables 6 and 7.

Implications for Europe

Our results identify a labour market channel behind Europe’s innovation gap. The channel runs through risk, not technology, so it taxes experimentation in every sector. And it falls precisely on the ventures Europe says it wants more of: unlike size-contingent regulation, it bites on the radical margin, because only risky entrants pay the premium. Where losing a job means a long spell of unemployment, workers are reluctant to join firms that might fail, experimental firms face higher labour costs, and fewer of them are created. Workers stay in safe jobs because they may not find another. The cost is not only to the workers themselves but to aggregate productivity.

This channel supports the call for labour market institutions that “protect workers, not jobs” (Adilbish et al. 2025). Our evidence comes from Denmark, whose flexicurity model of easy hiring and firing, generous income support and active help in finding a new job Aghion holds up as an example for Europe (Aghion and Johnson 2026). Even there, workers demand a sizeable premium to join a firm that might fail; where job finding is slower, the tax on experimentation should be heavier still. Policies that raise the speed at which workers find new jobs, such as effective active labour market policies, better matching institutions and lower barriers to job mobility, are also innovation policy, because they lower the cost of experimentation (see also Hombert et al. 2020, Van Doornik et al. 2022).

In one respect, our estimates are likely to be conservative. The model abstracts from knowledge spillovers across firms and from technology ladders in which today’s experiments raise the productivity of tomorrow’s. With such feedback loops, a labour market that makes it cheaper to experiment would raise not only the level of productivity but potentially its growth rate.

Source : VOXeu