1 Introduction

The role of a university is no longer limited to teaching and conducting research, but also considers the so-called “third mission”, meaning “a contribution to society” through the exchange of scientific and technological knowledge with industry and the government (Compagnucci & Spigarelli, 2020). Etzkowitz and Leydesdorff (1997) popularised the triple helix model, which involves collaboration between universities, the private sector (companies) and the public sector (government). An example of university business cooperation is the commercialisation of research results, and one important way to achieve this is by patenting and co-patenting with private companies. In the latter case, a patent has more than one holder, among which are both a university and a company. Yamaguchi et al., (2019) claim that it is generally difficult for universities and private enterprises to cooperate with each other. Foray and Lissoni (2010) list individual and institutional obstacles to such cooperation. First of all, individual scientists can be reluctant to invest their time in the transfer of technology and the commercialisation of academic inventions, since their research is evaluated mainly based on publications. System obstacles creating tension between universities and companies refer to the fact that higher education institutions (HEIs) and industry have different goals and a very different way to achieve them, and their economic benefits differ. On the other hand, there are some opportunities for exploiting potential transfers from academic research to industry. For example, HEIs provide human capital, e.g., trained engineers and researchers, to the stock of industrial employees, while industry can provide a university with needed material capital to conduct applied research.

The aim of our paper is to analyse the main determinants of universities' patenting and co-patenting with companies.

The main three research questions that we propose are as follows:

  1. 1.

    What are the main drivers of university patenting and co-patenting with companies?

  2. 2.

    Are determinants of patenting by university the same as co-patenting with companies

  3. 3.

    Are the motivations for patenting and co-patenting with companies different to start it than to continue?

To the best of our knowledge, co-patenting between a university and a company, when there are two holders of the patent, was not previously analysed which is the novelty of our study. Additionally, most of the previous studies were limited to one/two countries. Neves and Brito (2020) show that, out of 60 studies they analyse, most of them consider one country. The authors state that there were three papers with data from two countries, two papers with data from three countries and one paper with data from five countries. Conversely, we take into more than 400 universities from 17 countries over the period 2011–2018. We test several determinants of the commercial activity of universities, such as: student enrolment (size), age, public or private nature of the institution, students per academic staff, publications per academic staff (research orientation), non-academic staff per academic staff, funding structure (core and third-party budget) and prior patent activity. We estimate two-part models with zero-inflated negative binomial/zero-inflated beta regressions, which estimate separately the impact of the determinant of patenting (count data/proportion of joint company-university patents) and the probability of no patenting. The results indicate that the main determinants of universities patenting and co-patenting with companies are: size, age, research orientation and funding structure. As for patents per se, the determinants of starting patenting are the same as continuing patenting, while for joint company-university patents, most of the determinants differently affect starting joint patenting than increasing the proportion of co-patents with companies in all patents. For example bigger institution (characterised by higher number of students) are characterised by lower share of university-company patents but on the other hand they are more likely to start co-patenting. Next the share of third-party and core funds in the budget affects the probability of patenting with a company and the proportion of co-patents in a similar way. The higher the third-party budget and core budget, the lower the probability of no patenting, but for those universities that already patent, the proportion of patenting is lower. Our results can be deterministic not only to the management of an institution, but also to public policies, with the aim of strengthening university-company collaboration. We find evidence that there can be different drivers to initiate co-patenting with companies and for those HEIs that already are active in co-patenting. A wide range of policies has been implemented to encourage university patenting in hopes of spurring innovation to be effective, however, policies need to consider the motivation leading to starting patenting and/or co-patenting with companies and their continuing.

The structure of the paper is as follows. In the next section, we present the literature review focusing on the determinants of university of patenting as well in broaden sense taking into account academic entrepreneurship. Next we describe the data. It is followed by an empirical analysis. The last section is dedicated to a discussion and conclusions.

2 Literature review

Although the literature is mostly dedicated to the traditional role of a university (teaching and research), in recent years, university patenting has attracted more and more attention. Yamaguchi et al., (2019) provide a literature review of the studies dedicated to the analysis of factors influencing university patent creation. They divide previous studies into seven categories depending on the influencing factors. These are: research activity: the relationship between publications and patenting (Crespi et al., (2011), Grimm and Jaenicke (2015), Rizzo and Ramaciotti (2014), Wong & Singh (2010)); funding and its structure: public/private, internal/external (Coupe (2003), Azagra-Caro et al., (2006a) Lawson (2013), Azagra-Caro (2014)); institution support: law and regulations (Link and Hasselt (2019), Grimm and Jaenicke (2012), Hvide and Jones, (2018)); university type, e.g., private versus public, with and without medical schools (Duarte et al., (2020), Mathew et al., (2012)); researcher situation and individual motivations: age, seniority, position (Baldini et al., (2007), Neves and Brito (2020), Sellenthin (2009), Walter et al., (2018)); regional influence (Acosta et al., (2012), Rizzo and Ramaciotti (2014)); and multi factors (Duarte et al., (2020), Neves and Brito (2020), Rizzo and Ramaciotti (2014), Yamaguchi et al., (2019)).

The research activity refers to studies which examine the relationship between publishing and patenting by HEIs. The theme concerning publication versus patent activity is heavily studied. For an overview of the dedicated literature, see, e.g., Grimm and Jaenicke (2015), who conclude that although there are contradictory findings from previous studies, complementary effects between publishing and patenting are more often shown. For example, Wong & Singh (2010) find that the patenting output of the leading universities for the period 2003–2005 is significantly correlated with the quantity and quality of their scientific publications. Similarly, Grimm and Jaenicke (2015), for German scientists, show a positive relationship between patenting and publishing activities. Crespi et al., (2011), on a sample of UK academics, find that academic patenting is complementary to publishing, to a certain level of patenting output, after which there is evidence of a substitution effect. On the other hand, Rizzo and Ramaciotti (2014), examining the determinants of patenting of 59 Italian public universities from 2005 to 2009, find that there is a trade-off between patenting and publishing.

The next factor influencing patenting is funding and its structure. Generally, better funded academic research leads to more university patents (e.g., Coupe, 2003). However, the type of funding is also important. For example, Rizzo and Ramaciotti (2014) discovered that public funding is always positively related to patenting, while the impact of commercial revenues depends on university location: exerting a positive impact for central and northern Italian regions. Azagra-Caro et al., (2006a), analysing the laboratories belonging to Louis Pasteur university in Strasbourg, find that public funding positively impacts university-owned patents, while industry funds non-university-owned patents. On the other hand, Lawson (2013) suggests that researchers who obtain a large share of research contributions from industry have a higher propensity to patent. Similarly, Azagra-Caro (2014), conducting an analysis at the country level, analyses the patent activity of public research organisations (PROs) and universities, showing that the patenting activity of PROs is more responsive to R&D expenditure, while university patent ownership activity depends more on business funding. A country level analysis for 18 OECD countries for the period 1981–2016 is performed by Demir, (2019) who analyses whether the public, private and higher education sectors’ R&D investments impact the country’s innovation performance, measured by domestic patent applications. He finds that only the share of business R&D expenditures significantly raises the number of patents.

The next factor that can impact university patenting is institutional support (law and regulations). For example, Link and Hasselt (2019) evaluate the Bayh-Dole Act from 1980, which is United States legislation allowing universities the rights to innovative ideas that were federally funded. The authors show that the Act impacted the transfer of technology from universities to industry through patenting by providing an incentive for universities to invest in research infrastructure, especially technology transfer offices (TTO). Grimm and Jaenicke (2012) and Hvide and Jones (2018) evaluate the end of professor’s privilege, in East Germany and in Norway, respectively. Professor’s privilege is the right of academics to fully own the intellectual property of their research, inventions or patents (contrary to ownership by the university). Grimm and Jaenicke (2012) show that new public policy contributes to facilitate patent registrations. On the other hand, Hvide and Jones (2018) find a 50% decline in both entrepreneurship and patenting rates by university researchers after the reform.

University type can also have an impact on patenting activities. For example, Mathew et al., (2012) analyse North American universities with and without medical schools, while Duarte et al., (2020), analysing Brazilian HEIs, underline the division into public and private institutions.

Some of the studies analyse the motivations of individual academics to become involved in university patenting. Baldini et al., (2007), based on a survey of 208 Italian faculty members that were inventors of university-owned patents, find that earnings do not represent the main motivation, but rather respondents become involved in patenting to boost their prestige and look for new stimuli for their research. Neves and Brito (2020) conducted a Systematic Literature Review (SLR) of 66 studies dedicated to the drivers of academic entrepreneurship intentions (among them, patenting activities). They found that access to funding for the university and for research, is the most frequently described stimulus for entrepreneurial intention. Although there is an ambiguous effect of the age of an academic, most of the studies show a negative relation between age and academic entrepreneurship intentions. Academics engage in entrepreneurship activity in order to enhance their academic success, obtain needed infrastructure for their research and explore additional research topics. Walter et al., (2018) conducted an online survey at nine technical German universities and collected data from 1686 respondents, investigating their motivation for being involved in patenting. They propose three kinds of incentives to patenting: “gold”, “grace” and “glory”. “Gold” means that direct and indirect financial benefits account for roughly two-thirds of the total impact, “grace” is the freedom to pursue academic endeavours relatively undisturbed by commercialisation efforts, while “glory” refers to reputation and visibility. Respondents also underline the need for technology transfer achievements to be part of research evaluation.

When studying the region-university nexus, there can be a mutual relationship as far as university-owned patents are considered. Regional characteristics, e.g., GDP per capita, R&D expenditure, can affect university patenting. On the other hand, university patenting can be considered as regional R&D efforts (Azagra-Caro et al., (2006b), Azagra-Caro et al., (2007)). As far as regional factors affecting university patenting are concerned, the results are ambiguous. For example, Rizzo and Ramaciotti (2014) underline the differences in university patenting in the south and north of Italy, which can be the result of the wealth of the regions. On the other hand, Acosta et al., (2012) show that regional factors (level of development, industrial potential, and education expenditure on R&D), do not play a significant role in determining the quality (proxy by forward citation of a patent) of European university patents.

The literature review of previous studies indicates that the patenting activity of universities is determined by many factors. Neves and Brito (2020), based on their SLR, conclude that the drivers behind the intentions to patent are multiple, context-dependent, hierarchy-dependent and heterogeneous. Duarte et al., (2020) analyse 235 Brazilian universities for the years 2003–2011 using a production function with alternative output: publications measuring the basic research and the patent-applied research; they show that the main determinants of Brazilian scientific and technological production are the size of the university, its nature (whether public or private), the ratio of teaching staff and graduate students, and total investments in research and research support. Yamaguchi et al., (2019) provide multiple regression models for an analysis of the determinants of patents of 121 Japanese universities. They underline that university type and the number of graduate students has an influence on patent activity.

University patenting and co-patenting with companies can be considered from more broader view as the realisation of academic entrepreneurship. Besides the creation of spin-off companies, co-patenting is the main activity which links universities with companies. Changes a technology into a marketable products or services requires entrepreneurship capabilities of academicians and not just scientific achievements (Meoli et al., 2013). For example, Vismara (2014) presents that university-based firms (with a higher number of patents) have a lower propensity to pursue acquisition which may be due to their lower need for external technological resources. Additionally, Meoli et al., (2019) show that university patenting negatively correlated with the number of technology spin-offs. They underline that for the entrepreneur activities of university important is the composition of the governance body: a higher number of lay members (versus professors) in the governance body higher entrepreneur activities of the university is. Academics dedicated to teaching and research are focused less on entrepreneurship activities, while universities with a higher number of patents may create fewer surviving academic spin-offs since they are more highly risky businesses (Civera et al., 2020). Generally, the activity, characteristics and impact on the society of “entrepreneurial university” gained much attention in recent years (among others: Wright et al., 2004, Rothaermel et al., 2007, Audretsch et al., 2024, Marques et al., 2024, Abreu & Grinevich, 2024) however to the best of our knowledge these studies mainly focused on patenting activity and the creation of spin-offs, while co-patenting with companies has not been analysed in a systematic way so far.

3 Data

The individual characteristics of universities are derived from the RISIS European Tertiary Education Register(RISIS-ETER)Footnote 1 database, providing publicly accessible data on universities. First, the data is cleaned and the sample limited to general universities (as opposed to colleges and vocational schools). Furthermore, specialised institutions such as art, music, sport, police and theological are not taken into account. Finally, our sample is restricted to institutions with the number of students higher than 500 in order to avoid small outlier units. A similar cleaning procedure was introduced by Avenali et al., (2023) whose work is dedicated to a different topic, but the ETER database was also used. The data on publications come from the OrgReg database.Footnote 2 The data on university patents and co-patents with companies come from RISIS patent.Footnote 3 Then three datasets are merged: ETER, OrgReg and RISIS patent, restricting the analysis to the years 2011–2018, covered by three datasets.Footnote 4 As a result, we obtain information on more than 400 HEIs from 17 countries from 2011 to 2018; the sample of countries and years covered is driven by data availability. Table A1 in the Appendix shows the distribution of HEIs across different countries. Table 1 presents the descriptive statistics of the main variables used in the empirical study. On average, the biggest HEIs (regarding the number of students—first column) are in Italy and the Netherlands. The second column presents the number of students per academic staff, which can be treated as a proxy of the teaching load and is the highest in Italy and Czechia. The highest share of non-academic staff to total staff is reported in Latvia and the UK, where 50% of all staff are non-academics. Avenali et al., (2023) dedicated their study to the role and determinants of non-academic staff in a university. Non-academic staff can be crucial for university performance in providing help to academics; however, at the same time, they can provoke a bureaucratic burden (Martin, 2016). The fourth column shows the number of publications per academic staff member, which can be treated as a measure of scientific productivity. The highest value is achieved by Dutch HEIs, where, in the period analysed, one academic “produces” on average 2 publications per year. Wolszczak-Derlacz (2017) also finds that Dutch universities are leaders in publication production. The richest universities (with the highest revenues per academic staff) are from the Netherlands and the UK. At the other end, are Poland and Latvia, with half of those revenues per academic staff. The last two columns show the financial breakdown: the share of the core budgetFootnote 5 and the share of third-party funding.Footnote 6 The highest share of core funding is reported in Czechia (90%) and the lowest in the UK (27%). Looking at the share of third-party funding (last column), the highest is reported in Ireland and Lithuania.

Table 1 Key statistics on HEIs—mean values by country; time period 2011–2018
Full size table

Figure 1 shows the share of company-university patents across different countries.

Fig. 1
figure 1

Source Own elaboration based on RISIS Patent database

The share of joint company-university patents in total university patents across countries; all years pooled together.

Full size image

Overall, during the period 2011–2018, there are 27,937 patents, among which 2559 (9%) are joint company-university patents. Dispersion across countries can be seen. The maximum share is for the Netherlands, where 16% of all university patents are joint company-university patents, then France with 12.5% and Czechia 11.7%. On the other hand, only 1.2% of university patents in Lithuania and 2% in Poland are joint patents with a company.

4 Empirical analysis

The aim of our analysis is to evaluate the determinants of universities patenting and co-patenting with companies. In order to fulfil this and answer the research questions, we propose the following multivariable regression:

$$\begin{aligned} y_{ict} = & \alpha + \beta_{1} \ln {\text{Stud}}_{it} + \beta_{2} YearFound_{i} + \beta_{3} Private_{i} + \beta_{4} STEMStud\_acad_{it} + \beta_{5} Publ\_acad_{it} + \\ \beta_{6} Non\_acad_{it} + \beta_{7} {\text{Re}} v\_acad_{it} + \beta_{8} Third\_party_{it} + \beta_{9} Core\_budget_{it} + D_{c} + D_{t} + \varepsilon_{ict} \\ \end{aligned}$$
(1)

where i—stands for an individual university, c—country and t—time, yict is a dependent variable, either the number of university patents or the share of joint patents with a company to total patents. The independent variables include: the log of students—a proxy of the size of the institution (lnStudit); the age of the institution (YearFoundit), calculated as 2018—the year of foundation (the higher the year of foundation, the older the institution); the characteristic of the institution: private/public (Privatei); students from STEM fields (specifically students in fields: natural science, mathematics and statistics, information and communication technologies, engineering, manufacturing and construction) per academic staff (STEMStud_acadit), which can be a proxy of the teaching load; publications per academic staff (Publ_acadit), showing the research activity of academic staff; the share of non-academic staff in total staff (Non_acadit); financial information: revenue per academic staff (Rev_acadit); third-party funding in the total budget (Third_partyit); and core funding in the total budget (Core_budgetit). In all specifications, we add the country (Dc) and time (Dt) individual effect, which should control for country differences and time trends. The estimation method depends on the type of independent variable. When the dependent variable is patents (count data), we estimate the regression using zero-inflated negative binomial regression (see, e.g., Long, 1997), suitable for modelling count variables with excessive zeros and usually for over-dispersed count outcome variables (see histograms for patents, co-patents with companies and the proportion of co-patents with companies in total patents in Fig. A1 in the Appendix). When the dependent variable is proportion: the share of co-patents with companies in all patents, we use zero-inflated beta regression (Buis, 2010). Both types of estimations apply two-part models; a logistic regression model is estimated for the set of observations equal to 0, and a negative binomial count model or beta model for those between 0 and 1 in the case of proportion. Positive coefficients in lower parts (the probability of being zero) imply a higher chance of zeros (no patenting).

Table 2 shows the results for the regression with the dependent variable: number of university patents. We report seven specifications depending on the set of independent variables. Specifications (1)—(3) take into consideration all 17 countries, specification (4): 16 countries except France, specification (5): 15 countries except France and Spain, and specifications (6) and (7): 13 counties without France, Spain, Czechia and Finland, which is driven by data availability. The upper part shows the regression for count data and the logit model predicting whether there is a university patent or not (lower panel). In both the two models we use the same predictors. The number of students is positively correlated with the dependent variable: the bigger the institution, the higher the number of patents and the lower the probability of no patenting (negative coefficient in the case of the lower panel). Older institutions have a higher number of patents. Private institutions are characterised by a lower number of patents and a higher probability of no patenting., Universities with a higher number of students in STEM fields per academic staff have, on average, a higher number of patents and a lower probability of no patenting, but the coefficient is not statistically significant in all specifications (in lower panel). When we substitute the variable into students from all fields per academic staffFootnote 7 we have the opposite results: higher the number of students from all fields per academic staff the lower the patenting activity. This can be a sign that there is a trade-off between teaching and patenting and that the patent creation depends on the discipline: the number of patents in scientific disciplines is usually higher than the number of patents in humanities which can be measured by number of students from different disciplines Interestingly, we show that there is a positive relationship between the number of publications per academic staff and patenting: the higher the number of publications per academic staff, the higher the number of patents and the lower the probability that a university does not patent at all. In our study it seems that publications and patenting are complementary goods. There is a similar relationship between the share of non-academic staff in total staff, and patenting: the higher the share of non-academic staff in total staff, the higher the number of patents and the lower the probability of no patenting. Third-party and core funding is positively correlated with the number of patents. None of the financial variables are statistically significant in the case of the logit model (lower panel).

Table 2 Estimation results; dependent variable: number of university patents
Full size table

Since we are interested in the determinants of co-patenting with a company, we run the model (1), but this time, the dependent variable is the proportion of joint company-university patents in all university patents, and consequently, we use the two-part models with zero-inflated beta regression. The results are presented in Table 3: the upper part shows the result for proportion, and the lower part for the probability of being zero (logistic model). We find some interesting results, especially the differences in estimated effects between the proportion part and the logistic regression. For example, bigger universities (with a higher number of students) are more likely to co-patent (the negative coefficient in front of lnStud in the lower panel), but if they already patent, they patent less joint company-university patents (the negative coefficient in front of lnStud in the upper part). Similarly, older institutions are less likely not to co-patent, but if they patent, the share of co-patents is smaller (however, the coefficient is not statistically significant for all specifications of the upper panel). Private institutions patent less with a company than public ones. The next determinant, STEM students per academic staff, has different effects either on the proportion of joint company-university patents or the likelihood of patenting with companies. The higher the value of STEM students per academic staff, the higher the probability of patenting (lower panel), but for those universities that already patent with a company the proportion is lower. When we substitute the variable STEM students per academic staff by students from all disciplines per academic staff we obtain the reverse results. Again it can be a sign that indeed the characteristics of HEIs in respect to discipline matters for co-patenting with companies. As far as publications are considered. the higher the publications per academic staff, the lower the probability of no patenting, but the lower the proportion of joint company-university patents (the coefficient is not statistically significant in all specifications). We can say that universities which focus mainly on research (publications) are more likely to start patenting with companies, but at the same time their proportion of co-patents is lower. The share of non-academic staff in total staff seems not to be important for co-patenting. Revenue per academic staff is important for patenting with a company. The share of third-party and core funds in the budget affects the probability of patenting with a company and the proportion of co-patents in a different way. The higher the third-party budget and core budget, the lower the probability of no patenting (negative coefficients in the lower panel), but for those universities that already patent, the proportion of patenting is lower (negative coefficient in upper panel).

Table 3 Estimation results; dependent variable: the share of co-patents with companies in the total number of university patents
Full size table

Based on the results we can derive some interesting conclusions. First of all, the determinants of patenting can act differently for the number of patents and the share of co-patenting with companies. Specifically bigger institutions are characterized by a higher number of patents but a lower share of co-patents, older universities have more patents but younger co-patents. There is trade-off between teaching and patenting only in the case of share of co-patents with companies. Finally third and core funding is deterministic to patenting and co-patenting in different way: the higher the share of third and core funding higher the number of patents but the lower the share of co-patents with company. Additionally, in case of patents the determinants act in similar way for the number of patents and starting patenting while for co-patents the relationship is different. It can be specifically important for institution’s managers and stakeholders who want to enhance university commercialized output.

5 Extention and robustness

In order to check the sensitivity of our results we conducted several robustness checks. Firstly, we repeat the estimation for the panel dataset (each university is reported in all years) with the lagged number of patents/the share of co-patents as an additional independent variable (see Tables A2 and A3 in the Appendix). A similar approach was employed by Azagra-Caro (2014), who took into account prior patents in her econometric specification of country patenting. Sellenthin (2009), in his survey of university professors in Sweden and Germany, shows that the greater the researchers' years of experience in patenting, the more likely they are to apply for new patents. We find that those universities which patented/co-patent in the past are more likely to patent/co-patent in the future. The effect of the remaining variables is similar to our baseline regression, except for the coefficient in front of non-academic staff to total staff, which loses its statistical significance in the case of the count model (upper panel), but is still negative in the case of the logit model. This can mean that the share of non-academic staff is important to the decision of whether to patent or not, but when an institution already patented, it is not important for the number of patents. Similarly, some of the coefficients lose their statistical significance in the case of the regression of co-patents with companies. Specifically, now the size of the institution is not longer statistically significant, similarly whether the institution is private or public and the core budget. It can be the sign that for co-patenting the prior institution activity in more important than other determinants.

As the next sensitivity check, we also run the regression with all the time-varying variables lagged (see Tables A4 and A5 in the Appendix). It can solve the problem of potential endogeneity of the variables. The results are similar to the basic ones.

Since our sample is composed of 17 countries in all of our specifications we employ country-fixed effects which should control for different institutional settings. However, we run some robustness checks in this aspect. First, we repeat the estimation with an additional variable representing the share of the research and development expenditure in GDP (Tables A6 and A7 in the Appendix). It turned out that for most specifications, R&D is not statistically significant (being the sign that country-fixed effect already controlled for country-specific characteristics) and other variables are similar to the baseline model. Then in order to check if any country does not drive our results we estimated models limiting country by country. In the end, we have 17 regressions with 7 specifications for patenting and 17 with 7 specyfications for co-patenting—none of the countries drives the result.Footnote 8

6 Discussion and conclusion

This study examines the determinants of European universities patenting and co-patenting with companies. The analysis takes into account more than 400 universities from 17 countries over the period 2011–2018. We test several determinants of patenting activity and co-patenting with companies, such as: student enrolment (size), foundation year (age), public or private nature of the institution, STEM students per academic staff (teaching load), non-academic staff per academic staff, publications per academic staff, funding structure, and prior patent activity. We estimate two models, one for patenting (count data), the second for the proportion of joint company-university patents in all university patents. For those two regressions, we estimate a two-part model using zero-inflated negative binomial regression/zero-inflated beta regression, which estimate separately the impact of the determinant on the probability of patenting (count data)/the proportion of co-patenting with a company and the probability of no patenting (excess zeros). The results indicate that the bigger, older, public and richer institutions patent more. We find that there is no trade off between teaching load (STEM students per academic staff) and patenting, while the research orientation: publications per academic staff, positively affects patenting activities. Third-party and core funding is positively correlated with patenting. Finally, the higher the stock of patents of a university is, the higher its propensity to apply for patents will be.

As far as the proportion of joint patents with companies is concerned, there are some differences in relation to determinants of patenting. This time smaller and younger institutions are characterised by bigger share of co-patents. The teaching load (STEM students per academic staff) are negatively correlated with co-patenting. Similarly the higher the third party and core funding the lower co-patenting. Additionally we find that a different impact of the determinants of patenting or no patenting with companies and the proportion of joint patents (two part model). Specifically, the bigger, older, more research-orientated (higher publications per academic staff) institutions with a higher proportion of core and third-party budget are more likely to co-patent with companies, but if they already patent, they patent less joint company-university patents. Conversely, the higher the number of STEM students per academic staff, the lower the probability of no patenting, but for those universities that already patent with a company, the proportion is higher. Private institutions patent less with a company than public ones.

These are interesting results. As for patents per se, the determinants of starting patenting are the same as continuing patenting, while for joint company-university patents, most of the determinants differently affect starting joint-patenting than increasing the proportion of co-patents with companies in all patents. Our results are informative, to better understand the patenting activities performed by universities. To the best of our knowledge, this is the first analysis taking into account the determinants of universities co-patenting with companies which can be treated as one of the highest method of collaboration between HEIs and private industry, showing that collaboration between universities and companies is possible. It also shows which strategies could be performed by university management in order to enhance this collaboration. One of the main findings of this analysis is that a focus on research (higher publications per academic staff) is not unfavourable for patenting activity. In this sense, we contribute to the previous literature, which contains ambiguous results about the relationship between publishing and patenting (Grimm & Jaenicke, 2015). Next, we show that the funding structure is important for patenting, with third-party revenues especially determinant for patenting activity. Albeit the fact that this negatively impacts the proportion of joint company-university patents, it can be a sign that universities which already have a high proportion of joint company-university patents are no longer interested in obtaining higher funds from external sources.

Our results can be deterministic not only to the management of an institution, but also to public policies, with the aim of strengthening university-company collaboration. They should know that patenting per se can be characterised by different factors than patenting with companies. We showed that the determinants of patenting and co-patenting with companies are different.

The main limitation of our study is that it relies on the limited sample of the universities taken into account in our analysis. Additionally, there is a lack of information on the characteristics of the patents (fields and domains to which they are applied). Future research should further investigate the main International Patent Classification (IPC) classes of joint university-company patents, which can shed light on which fields the transfer of technology is offered by universities. Additionally, with the available data, we were not able to analyse the quality of patents, e.g., based on forward citation, we leave it for further research. Finally it would be interested to analyse the university patenting activity in the period of Covid-19. Bachmann et al., (2022) show that universities, as research institutions, played a significant role in the fight against the COVID-19 pandemic. They conclude that that during the pandemic universities demonstrated high R&D potential to quickly react to critical needs, offered open innovations, open licensing, showed collaborative abilities and effective use of their academic and student resources.