Determinants of Patent Applications across European Regions: Evidence from Panel Estimators, K-Means Clustering and Predictive Validation

In the modern economy, innovation is often treated as a measurable commodity, something that can be counted, ranked, and compared across borders. For decades, European policymakers have relied on a specific metric to gauge how well their regions are doing: the number of patent applications filed. A patent is a legal document that protects an invention, and the logic has been straightforward. If a region produces many patents, it is assumed to be a hub of creativity and technological progress. This assumption drives the allocation of billions of euros in public funds, shaping strategies intended to help Europe compete with the United States and China. The underlying belief is that these numbers reflect a region's ability to turn scientific ideas into commercial reality, and that by tracking them year after year, officials can see if their investments are working.

However, this reliance on a single count raises a fundamental question: what exactly does a patent application tell us about a region? Does it capture the true effort of companies to innovate, the strength of the links between universities and industry, or simply the legal habits of local businesses? A new study challenges the way these numbers are read, suggesting that the standard approach may be missing the most important parts of the story. By looking at data from 245 European regions over eight years, the researchers found that the picture is far more complex than a simple league table suggests. They discovered that the ability to innovate is deeply rooted in a region's long-term character, changing very little from year to year, and that different regions protect their ideas in completely different ways. Some rely on patents, while others rely on trademarks, and treating them all as if they were the same leads to a misunderstanding of where Europe's strengths and weaknesses actually lie.

The researchers began by examining a massive dataset from the Regional Innovation Scoreboard, which tracks 245 regions across 31 countries from 2016 to 2023. They focused on three main drivers that theory suggests should lead to more patents: the amount of money companies spend on research, the frequency of formal collaborations between universities and businesses, and the tendency of firms to register trademarks. Trademarks are the legal protections for brand names and logos, distinct from patents which protect inventions. The team wanted to see how these factors influenced the number of patent applications filed under international treaties. They used advanced statistical methods to separate the differences between regions from the changes that happened within a single region over time.

What they found first was a surprising lack of movement. While the data covered eight years, the vast majority of the differences in patenting activity came from the fact that some regions were simply more innovative than others, not from any region changing its performance significantly over time. In fact, about 95 percent of the variation in the data was due to the differences between regions, while only about 2 to 5 percent came from changes within a region over the eight-year period. This means that a region's position on the innovation ladder is remarkably stable. The small movements that do occur are often so slight that they look like noise rather than genuine progress. This finding suggests that judging the success of a policy program over a typical five-year cycle is like trying to measure the growth of a tree by looking at it for a few days; the changes are too small to see, and the real story is in the long-term structure.

When the researchers looked at what actually drives patent numbers, they found that corporate research spending was the strongest predictor. Regions where companies invest more in research tend to file more patents. Collaboration between universities and industry also played a role, though a weaker one. Interestingly, the study found that regions with high trademark activity also tended to have more patents. This suggests that protecting a brand and protecting an invention are not competing strategies; rather, they are signs of a single, robust capability to protect intellectual property. Companies that are good at one are likely good at the other.

However, the story changes dramatically when the researchers stopped treating all 245 regions as a single group and instead sorted them into four distinct profiles based on their actual behavior. The first group consisted of leading regions, which were high in every category. The second group was made up of peripheral regions, which were low in everything. The third group, which was the largest and included many regions in Southern and Central Europe, was the "intermediate" group. Here, the researchers found something startling: the strong link between research spending and patenting that held true for the whole continent simply disappeared. In these intermediate regions, spending more on research did not lead to more patents. The relationship that policymakers often use to justify funding was effectively invisible in the very places where it was needed most.

The fourth group was the most revealing. It consisted of 27 regions from 17 different countries, including some officially classified as innovation leaders. These regions were unique because they filed far more trademarks than patents. They were not failing to innovate; they were innovating in design, services, and branding, and protecting those ideas through trademarks rather than patents. Because the standard scoreboard ranks regions by patent counts alone, these dynamic, brand-focused economies were being misclassified as laggards. The study showed that a one-dimensional ranking system cannot capture the reality of these regions. If a policy is designed to boost patents, it will miss the mark entirely for these territories, which are already successful but in a different way.

To ensure these findings were not just an artifact of the math, the researchers tested their models against data they had not used before. They used computer algorithms to see if the patterns held up when the models tried to predict the performance of regions they had never seen. The results confirmed that the relationship between research and patents was real, but only when looking at the differences between regions, not the changes within them. They also found that complex computer models that seemed to work perfectly at first were actually just memorizing the specific characteristics of each region, rather than learning a general rule. Once the models were forced to treat each region as a unique case without relying on past data from that same region, their accuracy dropped significantly. This proved that the apparent success of these advanced models was an illusion created by the way the data was structured.

The implications of these findings are profound for how Europe manages its innovation policy. First, it suggests that the timeline for evaluating success needs to be much longer. Since regional innovation systems are so stable, short-term fluctuations in patent numbers tell us very little about whether a policy is working. Second, it shows that a "one-size-fits-all" approach is flawed. The strategies that work for leading regions or for lagging regions do not work for the large middle group, where the connection between research and output is broken. For these intermediate regions, simply throwing more money at research is unlikely to yield results; instead, the focus needs to shift to improving the institutions and connections that turn research into protected products. Finally, the study argues that Europe needs to stop measuring success solely by patent counts. By ignoring the regions that innovate through brands and design, the current system is misdiagnosing a significant part of the continent's economy. To truly compete with global rivals, Europe needs an evidence base that recognizes the diverse ways its regions create and protect value.