Do weak institutions undermine global innovation production efficiency?

Abstract

This paper focuses on institutional influences on innovation efficiency across countries. Whereas various causes and effects of technological change have been examined, empirical investigations of the efficiency involved in innovation production are relatively few. Using data on a large sample of nations over 2018–2020 and considering corruption, regulatory quality, and state fragility as alternative institutional dimensions, our results show that greater corruption facilitates (“greases”) efficiency in the production of innovations. This is also the case with improvements in regulatory quality, while greater state fragility increases inefficiency. These findings for the overall sample are somewhat different for the OECD and non-OECD subsamples, although the greasing effect of corruption remains throughout. A robustness check with patent protection and government size as alternative institutional dimensions is also conducted.

1 Introduction

The importance of innovation for firms and economies has been recognized for quite some time. Economies are able to relax resource constraints, improve living standards, and ultimately boost economic growth by inventing new products and processes (Goel & Ram, 1994; Jones, 1995; Schumpeter, 2006). At a micro-level, firms and individuals are able to enter markets, increase market shares, leapfrog competition, and subsequently enhance profitability through inventions. Yet, for firms and policymakers, a relatively greater focus seems to be on the production of innovations, rather than on the efficiency associated with the production of innovations.¹ A part of this might be due to the one-off nature of many innovations—many inventors are pursuing one innovation and are not thinking of another “fallback” option where innovation resources may alternatively be employed. Still, efficiency in innovation production may be relevant for large firms simultaneously pursuing many innovations and, from society’s perspective, regarding inefficient use of resources. This research focuses on the efficiency of innovation production, rather than its production.

The internal incentives in a profit-maximizing firm are to ensure that the input-to-output relationship is efficient, both in the pursuit of output and innovation.² For the innovation production process to work efficiently, and for the innovations to diffuse efficiently, there is a need for external institutions and infrastructures to work effectively. Information on the latter is thin in the theoretical literature and policy literature/debates, and this paper seeks to make a contribution here.

The previous literature, both theoretical and empirical, has studied the causes (the impact of R&D, for instance (Kamien & Schwartz, 1982; Reinganum, 1989) and effects (on economic growth and productivity, for instance, see Jones, 1995) of innovation production. Relatively recently, some scholars have focused on the efficiency of innovation production (Barra & Zotti, 2018; Feng et al., 2021; Gao & Chou, 2015). Within this context, the primary focus has been on the transformation of the related inputs (or internal factors such as R&D and researchers) into innovation output (mainly measured via patents, see Mueller, 1966), although the role of institutional quality in dictating the efficient allocation of innovation resources has not been considered at length. Our research contributes to the innovation production efficiency stream of the literature by focusing on the influence of external factors (captured by different dimensions of institutional quality). Another contribution is the comparison of innovation efficiency in the OECD group of nations versus the non-OECD group.

At a broad level, institutions are relevant in many dimensions to the functioning and survival of nations. Institutions (and institutional structure) change rather slowly over time, and thus have latent effects or inertia (e.g., Acemoglu & Robinson, 2008), and they evolve over time via government actions and are nested in history, especially in nations that were former colonies. Their objectives are varied, ranging from maintaining law and order, regime stability, and smooth functioning of the markets. Institutions are crucial determinants of economic performance both at the micro- and macro-levels (see Knack & Keefer, 1995). Good institutions address the quantity and quality of social capital and ensure the internalization of research spillovers or externalities. Within this broad realm, one role of institutions has been on innovation generation (e.g., via patent policy). Better institutional quality can impact the timing and allocation of innovation inputs and, thereby, improve the production efficiency of innovations (see Fig. 1).

Fig. 1

Innovation production and the influence of external institutions

The focus of this paper, however, is more narrow, and we consider the influence of institutions on the efficiency of innovation production. In the context of technological change, institutions ensuring appropriation of rewards (through patents, for example), mitigating uncertainty, etc., would be important. On the other hand, weak institutions can create distortions that can undermine incentives for innovation and can even be counterproductive. With regard to technological change, institutions are tied to national/regional systems of innovation that crucially impact the inputs and outputs of the research process (Nelson, 1993). The importance of the topic studied, and the need for further research in the area have been highlighted in a recent review of the related literature (Zeng et al., 2021).

It is less clear whether policymakers around the world appreciate the importance of the efficiency of innovation production. Efficiency in the innovation process ties to the optimal use of resources devoted to the pursuit of innovation, which may be important both at the firm level and the broader country level. Do different dimensions of institutional quality similarly impact the efficiency of innovation production? The consideration of the impact of different institutions on the efficiency of innovation production, via their influence on the scale and speed of input allocations (Fig. 1), forms the main focus of this study.

Understanding this relationship requires that the following questions be considered:

• Which institutions significantly impact innovation production efficiency?

• Are the innovation production efficiency impacts of different institutions alike?

• Are the drivers of innovation production efficiency different in OECD nations from the rest of the world?

Alternately stated, we estimate an innovation production frontier and examine how the distance to the efficient frontier is affected by a change in different aspects of institutional quality (see La Porta et al. (1999) for background on institutional quality).

In framing institutions with regard to technological change, such as patent policy, policymakers have to consider the potential wedge between the social and private returns to inventive efforts (Griliches, 1992; Haruna & Goel, 2011). This is due to wasteful duplication in innovation races, the skewness of research or inventive effort towards certain technologies at the expense of others, and the lack of measurability of certain research efforts (i.e., when certain inventions are not patented or are unpatentable—e.g., Blanco et al., 2023). The benefit–cost accounting of appropriate research effort is complicated due to inherent uncertainty related to innovation success and payoffs, plus the inability to completely reap the rewards of most research efforts. Therefore, the potential for social returns to innovation to exceed private returns warrants that the efficiency in the production of inventions be evaluated so that resources, both public (through research subsidies and tax credits) and private, could be appropriately allocated. This issue has obvious technology policy implications: even nations faring well in innovation production might not be doing so efficiently.

The literature has mostly considered the causes and effects of inventive efforts (Kamien & Schwartz, 1982; Reinganum, 1989; Scotchmer, 2004). Initially, the focus was mostly on theoretical studies, but in recent years, with the greater availability of appropriate data and advances in computational and data transmission technologies, empirical studies examining various aspects and with varying data have emerged. In all this, the focus on the process of innovation production has been relatively less. What is the process of innovation production, and what does its efficiency depend upon? Broadly speaking, this aspect can be tied to the production of knowledge, which has received some research attention (see Buesa et al., 2006, 2010).

Beyond institutions, we also evaluate the impact of uncertainty on innovation efficiency. General, economy-wide, uncertainty is related to potential payoffs from inventive efforts, while budding inventors/firms also face specific uncertainties related to the timing and success of innovation (Kamien & Schwartz, 1982; Reinganum, 1989).

Our results show that greater corruption (an indicator of weak/poor institutional quality) facilitates efficiency in the production of innovations. This is also the case with improvements in regulatory quality, while greater state fragility increases inefficiency. These findings for the overall sample are somewhat different for the OECD and non-OECD subsamples. A robustness check with patent protection and government size as alternative institutional dimensions is also conducted and policy implications are discussed.

The results point to the role of institutions in impacting innovation efficiency. The findings have some useful input into the broader questions of how some nations are more innovative than others? (Fu & Yang, 2009); why some nations chose certain technological trajectories?; and why do they continue to be caught in a poverty trap? Based on our analysis, the answers to these questions seem to go beyond the usual inputs to innovation production. The format of the rest of the paper includes the literature and the model in the next section, followed by data and estimation, results, and conclusions.

2 Literature and model

2.1 Literature

Broadly speaking, this research ties three strands of the literature: (i) determinants of innovation; (ii) impacts of institutional quality; and (iii) production efficiency.

The literature on the determinants of innovation is based on the economic incentives to innovate. Beyond the costs, benefits, and the uncertainty of inventive efforts, innovation pursuit/success is determined by the market structure of the industry where the firms operate (Kamien & Schwartz, 1982), and, more broadly, the institutional setup of nations/jurisdictions where the firms operate. Most of the early studies were theoretical (for reviews, see Kamien & Schwartz, 1982; Reinganum, 1989), although, over time, data availability has enabled empirical investigations (Goel & Nelson, 2018).

Two large, related, aspects that have been hard to empirically capture are related to: (a) effectively quantifying the various innovation uncertainties (Goel & Nelson, 2021); and (b) effectively accounting for the spillovers from innovation (Audretsch & Feldman, 1996; Griliches, 1992).

Turning to the role of institutions in the inventive effort, it is hard to provide a definitive summary, since institutions are varied, both quantitatively and qualitatively. Still, we examine several significant institutions and their relative impacts on innovation and related efficiency.

Across the spectrum of government institutions, corruption is a widely present factor that undermines the strength or quality of institutions. For instance, corruption can undermine the strength of enforcement, compromising the functioning of bureaucracies. Yet, corruption can potentially have the opposite impact, enabling firms to overcome regulatory and bureaucratic barriers through bribes. The former is termed a “sanding effect”, while the latter is termed a “greasing effect”, (Dimant & Tosato, 2018). The relative strength of each effect would determine the overall impact of corruption, dictated by the underlying socio-economic-historical context of a nation. This is ultimately an empirical issue that we will address in the context of the impact of corruption on invention efficiency.³

As an indicator of weak institutional quality (see Knack & Keefer, 1995), corruption can also undermine the appropriability of innovation rewards. The appropriability of rewards is also accounted for directly in the analysis via (i) regulatory quality; and (ii) state fragility.⁴ While greater regulatory quality would enhance appropriability by ensuring a due process for inventors to reap the rewards of their actions (and allocate resources in the production of innovation accordingly), greater state fragility, on the other hand, would undermine incentives to allocate resources.

Along another related dimension, uncertainties of various kinds are associated with the innovation process. Some of these are pre-innovation, related to success in invention and sufficient incentive effort, while others are post-invention, related to diffusion and regulation (see Goel, 2003, 2007). The extant literature has considered the impact of uncertainty on research and innovation, both theoretically (Kamien & Schwartz, 1982; Reinganum, 1989), and empirically (Goel & Nelson, 2021; Goel & Ram, 2001). Given the aggregate nature of our data, we are able to empirically capture macroeconomic uncertainty, whereas firm- or inventor-specific uncertainties could be equally, if not more, significant. However, micro-level uncertainties are nearly impossible to measure empirically.

Whereas the literature has focused on production efficiencies, with and without the stochastic frontier models (see Kumbhakar & Lovell, 2000), the efficiency in the production of innovations has not been considered in much detail.3 In an interesting angle, Gao and Chou (2015) compare the innovation efficiency of multinational firms versus domestic firms. They find innovation efficiency in multinational firms to be lower. Further, Feng et al. (2021), using data from 2013 to 2017, compare the innovation efficiency in high-income and middle-income nations, while considering both the R&D and marketing stages of firms. They find that better quality institutions can help improve the innovation efficiency in all nations and that efficiency improvements were higher in high-income countries in the R&D stage, but higher in middle-income countries during the marketing stage.⁵ The stochastic frontier analysis has also been used to study the performance of university licensing (Coupet & Ba, 2022). Focusing on Eastern Europe, Kravtsova and Radosevic (2012) analyze a broader dimension by considering inefficiencies in the systems of innovation. Relatedly, the regional innovation system inefficiency in Italy has been studied by Barra and Zotti (2018). Furthermore, the contributions of international trade to innovation efficiency can also be important, and this has been studied by Klevenhusen et al. (2021) for OECD nations.

Related more directly to the methodology of the current study, a few recent studies have employed the stochastic frontier analysis to focus on innovation inefficiency (Barra & Zotti, 2018; Haschka & Herwartz, 2020; Huang, 2023). Unlike our work, these studies do not consider the external impacts of institutional quality on innovation efficiency, and they focus on specific world regions, rather than a cross-section of nations (see Fig. 1). For instance, Huang (2023) examines the impact of competition on innovation inefficiency among Chinese firms, Barra and Zotti (2018) study regional innovation inefficiency in Italy, and Haschka and Herwartz (2020) consider the innovation inefficiency in European high-tech industries.

A recent review of the literature on innovation efficiency can be found in Zeng et al. (2021), who also note the need for greater interdisciplinary research on the topic. This paper adds by focusing on the institutional quality-innovation efficiency nexus for a large sample of nations. The formal empirical model follows.

2.2 Model

Based on the above discussion and to tie to the theme of the paper, we pose our main hypothesis.

Hypothesis H1

Better institutions improve innovation production efficiency, ceteris paribus.

Since institutions are not one size fits all, we expect quantitative and qualitative differences across how specific institutions satisfy or refute hypothesis H1.

To answer the questions posed in the Introduction, we employ stochastic frontier analysis (SFA) to estimate an innovation production frontier and examine how the distance to the efficient frontier is affected by a change in different aspects of institutional quality. Given that institutional quality can be variously measured (see Knack & Keefer, 1995; La Porta et al., 1999), we consider different dimensions, including corruption, regulatory quality, government size, patent protection, and state fragility. Some qualitative differences across these institutional measures impacted our choice. For instance, corruption and government size are broad measures, with many causes and effects, while patent protection and regulatory quality are more specific. Furthermore, state fragility incorporates political dimensions more directly than some of the other measures.

To analyze the hypothesis of interest, we model the effect of institutions on innovation production efficiency by employing the following stochastic production frontier.

$$ Innovation_{i} = f\left( {x_{i} ;\beta } \right) \, + v_{i} - u_{i} $$

(1)

$$ v_{i} \sim {\text{N}}\left( {0, \, \sigma^{{2}} } \right) $$

$$ u_{i} \sim {\text{N}}^{ + } \left( {z_{{\text{i}}}^{\prime } \delta , \, \sigma_{u}^{{2}} } \right) $$

Our innovation production model is composed of 3 parts. First, a deterministic process by which inputs are transformed into innovation output (i.e., a production function). Second, a stochastic component captures randomness affecting the innovation production process. For any individual firm, this random error component creates distance between potential innovation output, determined by the production function and the level of inputs, and the observed output. Naturally, this distance between potential and observed outputs is purely random. Of paramount importance for our study, a third component is related to technical inefficiency, which adds to the distance between potential and observed innovation outputs in a systematic fashion. In aggregate, fully efficient firms make up a stochastic innovation production frontier, while technically inefficient firms are farther and below this frontier. Our interest is in understanding how external institutions and institutional infrastructures affect this distance between inefficient firms and the technically efficient innovation frontier.

Focusing on the first expression in (1), Innovation_i denotes the output variable for country i. Then, x_i is the input vector used to produce Innovation_i. The production technology f (x_i; β) is a deterministic process by which inputs are transformed into output, with β representing the vector of corresponding technology parameters attached to these inputs. Note that the “stochastic” part of the production frontier has two components. First, v_i represents pure randomness in the production of innovation for country i, and it is assumed to be normally distributed, as shown in the second expression in (1). The second component, u_i, represents inefficiency in innovation production for country i. Note from the last expression in (1) that inefficiency is assumed to be non-negative and can be explained by a vector of institutional covariates z_i, with the respective vector of parameters represented by δ.

The benefit of employing the model depicted in (1) is twofold. First, it allows us to estimate innovation production efficiency for each country. While the latter estimation is not our main objective, we deem it an important one given the scarce literature documenting cross-country innovation production efficiency. A second benefit relates to the model’s ability to associate variation in innovation production efficiency with variation in variables that characterize a country’s institutional environment. The latter allows us to directly test our hypothesis of interest. In what follows, we provide details about the variables employed in our analysis.

The output variable, Innovation_i, is proxied by the relative number of resident patent applications (patents-to-GDP ratio) in a nation. While patents are an imperfect measure of inventive activity (since they do not capture non-patented and non-patentable inventions), they remain the most readily available and comparable measure of research output (see Mueller, 1966).⁶

We consider the standard inputs in the (innovation) production process, namely labor, and capital. Labor is proxied by the full-time-equivalent (FTE) researchers per capita.⁷ The impact of R&D on innovation is intuitive and has been widely recognized in the literature (e.g., Pegkas et al. (2019)). One proxy for capital we consider is domestic credit as a percentage of GDP. This measure correlates to a higher level of domestic investment (Levine, 2005). To further account for these ancillary inputs, we also include population (net of researchers) and real GDP per capita. The size of the market (captured by population) addresses the potential market size post-innovation, plus it captures the strength of demand-pull innovation. Market size also relates to the scale effects. GDP includes capital consumption, which approximates the value of services rendered by capital stock, and thus, it proxies for the capital input.⁸

In addition, we include the inputs of E-participation and Inflation, both are assumed to influence the production of innovation directly. E-participation measures the extent of online services to facilitate the provision of information and public services by governments. We employ inflation variability to control for economic uncertainty (see Goel & Ram, 2001). Economic uncertainty can impact future payoffs from innovation and affect the allocation of re- sources devoted to the innovative effort. All these variables are represented by the x_i vector in (1).

We use several indicators of institutions, represented by z_i in (1), to account for their qualitative differences and possible different effects on innovation efficiency. First, we include a measure of regulatory quality, which reflects perceptions of the ability of the government to formulate and implement sound policies and regulations that permit and promote private-sector development. Second, we include corruption, which can potentially have a greasing or a sanding effect. On the one hand, greater corruption can impose transaction costs, undermine institutions (e.g., fairness of the courts, the award and scope of patents; Goel, 2002), and reduce innovation efficiency; on the other hand, bribes could enable firms to overcome regulatory barriers and can potentially have positive efficiency implications. We empirically assess which potential effect of corruption on innovation efficiency is the dominant one. Finally, we include a measure of state fragility, which captures vulnerabilities based on a state’s ability to cope with abnormal and unanticipated pressures.

Relatively speaking, of the three institutional aspects considered, state fragility, encompassing coups and government overthrows, might be more exogenous. State fragility can also be viewed as capturing political uncertainty, and, together with inflation variability, our results can show a comparison of political and economic uncertainties as they impact innovation production efficiencies.⁹ Whereas one might envision scenarios where greater uncertainties undermine the innovative effort, innovative activity might also increase in the face of uncertainty when innovation enables inventors to hedge against uncertainty (see Goel & Nelson, 2021). Furthermore, regulatory quality is largely fixed in the short term, while corruption levels and state fragility can change rather quickly (and unexpectedly).

3 Data and estimation

3.1 Data

We pool annual country-level data for 101 nations over the years 2018 to 2020.¹⁰ The span of our data is limited by the availability of some underlying variables across the large sample of nations considered. While the breadth of the coverage is welcome and enables us to compare the relative innovation production efficiencies in the OECD versus non-OECD samples, the short time span limits our ability to consider lagged effects, which may be important in the gestation period of certain innovations (Hall et al., 1986).

The primary source of data facilitating this analysis is the Global Innovation Index (GII).¹¹ This source provides comparable data across a large sample of nations on different dimensions of the inputs and outputs of the innovation process. Our output measure, the number of resident patent applications (patents-to-GDP ratio), and the main labor input of full-time-equivalent researchers per capita are from GII.

The data from GII are supplemented with data from other reputed international sources that are routinely used in the related literature. For instance, data on population, per capita GDP, and inflation come from the World Bank’s World Development Indicators (http://data.worldbank.org/). The E-participation index and domestic credit as a percentage of GDP are available within GII.¹²

Data on the first institutional variable we consider, i.e., regulatory quality, is available from the World Bank’s World Development Indicators and can also be taken from GII. Our corruption measure is the well-known corruption perceptions index (CPI) from the Transparency International (www.transparency.org/en). While this measure is imperfect (as it measures perceptions, not actual corruption incidence or its severity), it is known to provide a reasonable comparison of corruption prevalence across nations and over time since 2012. Finally, data on state fragility is produced by The Fund for Peace (https://fragilestatesindex.org/methodology/).

Table 1 reports summary statistics for all the variables employed. Starting with the output, the average number of patents (per bn GDP) is 4.82, and the sample has significant variation as suggested by the sample standard deviation (11.12) and range. At the bottom of the number of patents (per bn GDP) distribution, we have mostly African countries (e.g., Malawi has a minimum of 0.1 patents per bn GDP), while in the upper part of the distribution, we have South Korea (max of 84.5 patents per bn GDP), followed by China. Regarding inputs, the sample average values for population (net of researchers), GDP per capita, domestic credit as a percentage of GDP, and the 3-year moving standard deviation of inflation are about 65 million, $27,500, 73 percent, and 1.3, respectively. The E-participation variable has a sample average of about 72 and ranges from 6.8 to 100, where higher values suggest more online resources to facilitate the provision of information and public services by governments. On average, there are 2,118 full-time-equivalent researchers per million people, with notable differences across countries as suggested by the minimum and maximum values of around 12 and 8342. Focusing on the institutional variables that affect efficiency, the middle panel in Table 1 shows that the indexes for regulatory quality, corruption, and state fragility all have a significant level of variation in the sample. At the bottom of Table 1, we show summary statistics for an alternative set of inefficiency determinants, which are later employed in a robustness check. A brief discussion relevant to these alternative innovation inefficiency determinants is relegated to Sect. 4.4, where we report results from the robustness check exercise.

Table 1 Summary Statistics

In part motivated by the large extent of discrepancies in values for all the aforementioned variables, we separate the main sample into OECD and non-OECD members. Summary statistics for these two sub-samples are presented in Table 2. Relative to the OECD countries, non-OECD countries produce, on average, fewer patents (2.325 vs. 8.658). As expected, the average level of inputs in the production of our innovation measure also differs drastically by OECD membership status. The OECD countries are, on average, less populated, richer, and more engaged in e-participation. Also, OECD countries have, on average, 5 times more researchers, and less volatility as measured by average inflation. The average level of regulatory quality differs drastically by OECD membership, where richer countries have a higher average index (75 vs. 42) consistent with government policies and regulations that foster private-sector development. Average corruption levels are higher in non-OECD countries (i.e., a low corruption perception index is consistent with higher levels of perceived corruption). Finally, based on state fragility values, we report that OECD countries are, on average, better at coping with shocks whereas non-OECD countries are more vulnerable on average.¹³

Table 2 Summary Statistics by OECD Membership

3.2 Estimation

Estimation of the parameters in (1) is based on extensions to the econometric technique developed by Aigner et al. (1977). In particular, the maximum likelihood approach proposed by Aigner et al. (1977) can be employed to obtain estimates of the innovation production stochastic frontier in (1) but limited to a case in which the one-sided inefficiency component is assumed to have a mean of zero, i.e., u_i ∼ N ⁺(0, σ²).¹⁴ Subsequently, one can employ the technique developed by Jondrow et al. (1982) to obtain individual-specific estimates of inefficiency. The individual estimates of inefficiency are, in turn, used to estimate technical efficiency TE_i = exp{uˆ_i}, where uˆ_i is usually E[u_i|v_i − u_i]. One can summarize technical efficiency TE_i in the sample, where countries with values close to 1 are categorized as being highly efficient, and those with values closer to zero are inefficient in innovation production. These methods are well-known and have been widely employed in different literatures.¹⁵

The technique by Aigner et al. (1977) can be implemented as part of a two-step process, where, in the second step, one examines differences in innovation efficiency by levels of an institutional variable (e.g., comparing innovation efficiency between high and low corruption countries) or one simply regresses estimated efficiency on the institutional variables. The two-step process relies on the assumption that there is no correlation between levels of the institutional variables and no correlation between the inputs and institutional variables. In the presence of the latter types of correlation, a two-step process yields biased and inconsistent estimates. To address these caveats, a myriad of approaches that build on Aigner et al. (1977) have been introduced (see Kumbhakar & Lovell, 2000). To simultaneously estimate the innovation production frontier and model inefficiency as a function of external institutions (thereby determining the “distance” between efficiency and inefficiency), we employ the method developed by Huang and Liu (1994).¹⁶

Accordingly, we assume a Cobb–Douglas technology (i.e., f (·) in (1)), a normal distribution for the stochastic component (v_i), and a truncated normal distribution for the inefficiency component (u_i).¹⁷ Estimated parameters for model (1) are obtained via maximum likelihood. These estimates are employed not only to compute technical (innovation) efficiency, but to directly test our hypothesis of interest based on estimated parameters in δ. We turn next to the discussion of the results.

4 Results

4.1 Full sample

The main estimated results for the full sample of countries are reported in Table 3. Benchmark stochastic frontier estimates are reported in the first column.¹⁸ The estimated coefficients in the deterministic part of the frontier suggest that patent production is positively affected by population, e-participation, domestic credit, and the number of researchers. The latter estimates are statistically significant at the standard levels of confidence, except for e-participation. The population is related to market size and the potential pool of talent available as complementary inputs, domestic credit captures the functioning of financial markets, and e-participation is related to the information flows (or transaction costs more generally). The number of researchers obviously is the key input in innovation production. Keeping the underlying measurement units in mind, the magnitude of the positive impact of Researchers is greater than that of Population and Domestic Credit (Model 3.5).

Table 3 Innovation Production Cross-Section Frontier Estimates. Full Sample. Output Measure: Patents (in logs)

The estimated coefficients attached to GDP per capita and inflation variability (uncertainty) are negative but not statistically insignificant. Since innovations are forward-looking, the insignificance of the current GDP is not too surprising. The negative effect of uncertainty is in line with intuition (for a related theory, see Kamien & Schwartz, 1982) and some empirical evidence (e.g., Goel & Ram, 2001). The insignificance of the related coefficient could imply that inventor- and technology-specific uncertainties may be more relevant.¹⁹

For brevity and given our focus on inefficiency determinants, we opt for a brief discussion of the estimated coefficients in the deterministic part of the frontier for all specifications we consider (Models 3.2–3.5). We will just point out that estimated magnitudes, signs and significance remain, to a large extent, similar across specifications.

The estimates under the column labeled Model 3.2 include the variable Regulatory Quality as a determinant of innovative production inefficiency. Regulatory quality is a broad indicator of the quality of regulations, potentially impacting many aspects of firms’ operations. The estimated coefficient suggests that higher regulatory quality has a negative and significant impact on innovative inefficiency. This result is in line with intuition—better regulatory quality (related to efficiency, fairness, etc. in government dealings) would improve the efficiency of firms’ efforts to allocate resources and pursue innovation.

Model 3.3 estimates suggest that lower levels of corruption increase inefficiency in innovation; however, the estimated coefficient is not statistically significant. In other words, the greasing effect of corruption does not seem significant.

The estimate of interest under Model 3.4 is consistent with more inefficiency when state fragility increases. Institutions might not function very well in fragile states, and the potential uncertainty in fragile states might induce some firms to delay innovations (due to lower expected benefits or higher costs).

The specification reported under the Model 3.5 column is preferred because all three institutional determinants of inefficiency are added.²⁰ Since the institutional dimensions that we consider are intricately intertwined (with, for example, regulatory quality impacting corruption, and corruption being impacted by state fragility), consideration of the three aspects provides a useful indication of the overall impact on innovation efficiency. First, we report that the estimated signs of the effects of our institutional variables are consistent across specifications, so the previously discussed qualitative results hold. That is, under our preferred specification, regulatory quality and corruption decrease inefficiency in innovation production, while state fragility increases it. The greasing effect of corruption has some support in the literature, albeit in studies focusing on different aspects of technological change. Goel and Nelson (2018) found that corruption greased the introduction of process innovations to the market across countries, while, considering data across U.S. states and distinguishing between design and utility patents, Goel and Saunoris (2020) show that the greasing effect of corruption does not hold, and the influence of the timing of corruption effects is sensitive to the type of innovation considered. The current paper, on the other hand, considers the impact of corruption on the efficiency in the production of innovations. Second, we note that relative to Model 3.3, once all 3 institutional variables are included as inefficiency determinants, we gain precision since all estimated coefficients are statistically significant at the standard levels. Thus, we see that corruption reinforces the efficiency effects of regulatory quality, while the largely exogenous state fragility works in the opposite direction.

In Tables 4 and 5, we focus on reporting the main estimates for the subsamples of OECD and non-OECD countries, respectively. Institutional quality, cross-border trade, political stability, economic outlook, etc., are generally expected to be better in the OECD group, and it would be interesting to see whether these differences have significant bearings on the efficiency of producing innovations.

4.1.1 OECD sample

There are some noteworthy differences, regardless of model specification, in the estimated coefficients for the innovation production frontier across samples. For instance, the positive effect of researchers is notably larger for OECD countries, and the effect of economic uncertainty, captured by inflation, is negative for OECD countries but positive in the non-OECD sample. However, consistent with Table 3, the effect of economic uncertainty is statistically insignificant in both cases.

We now discuss the estimates of interest that are related to the innovation inefficiency determinants. Focusing on OECD countries, the estimated coefficient for regulatory quality (Model 4.2) is consistent with the reported estimate for the full sample—i.e., better regulatory quality lowers innovation inefficiency. We also note that the estimated coefficient under the preferred specification (Model 4.5) is consistent with a reduction in innovation inefficiency due to increases in regulatory quality for OECD countries.

The estimates under Models 4.3 and 4.4 suggest, respectively, that corruption and state fragility are not important determinants of innovation inefficiency, which stands in contrast with estimates under the preferred specification in Model 4.5. Relative to the full sample, the OECD sample estimates under Model 4.5 yield a similar qualitative conclusion in that higher regulatory quality and greater levels of corruption decrease innovation production inefficiency—the estimate attached to corruption is not statistically significant, which was not the case for the full sample. It could be the case that in the OECD group, with a relatively better economic outlook, the opportunity costs of engaging in corrupt acts (both for potential bribe-takers and bribe-givers) are relatively high. In contrast to the full sample, in the OECD countries, we find that greater state fragility decreases inefficiency in Model 4.4, but not in Model 4.5.²¹

4.1.2 Non-OECD sample

In this section, we consider the behavior of the non-OECD group of nations. This group is generally less economically developed, and the separate treatment provides an alternative check of the influence of economic prosperity.²²

For the sample of non-OECD countries (Table 5), we report that innovation inefficiency estimates under Models 5.2, 5.3, and 5.4 differ substantially from the estimates under the preferred specification in Model 5.5. The influence of population is less significant in the non-OECD group compared to the OECD group in Table 4. This may be due to the relatively fewer quantity and quality of complementary inputs that labor has to work with in developing nations (also see Goel, 1990). Furthermore, the relatively lower magnitudes (and relatively lower statistical significance) of the coefficients on Domestic Credit confirm the notion about capital markets being relatively less developed. Focusing on Model 5.5 specification, in non-OECD countries, the effects of regulatory quality and corruption on innovation inefficiency are qualitatively consistent with estimates for the sample of OECD countries.

Quantitatively, the effect of regulatory quality is smaller for non-OECD countries, while the effect of corruption on innovation inefficiency is statistically significant. The most striking difference across OECD membership is related to the estimated effect of state fragility on innovation inefficiency, where, for non-OECD countries, greater state fragility results in statistically significant increases in innovation inefficiency. As Table 2 shows, the average state fragility in the non-OECD group is substantially greater than that in the OECD group, and further increases in the fragility have a detrimental effect on innovation efficiency.²³ Fragile states would make the potential rewards from innovation more uncertain, while also causing distortions in the resources allocated towards complementary inputs to innovation.

Overall, the comparison across the two subsets of nations shows that the impacts of some institutions on innovation efficiency are somewhat sensitive to the underlying sample of nations considered.

Table 4 Innovation (Patents) Production Cross-Section Frontier Estimates. OECD Countries. Output Measure (Dependent variable): Patents (in logs)

Table 5 Innovation (Patents) Production Cross-Section Frontier Estimates. Non-OECD Countries, Output Measure (Dependent variable): Patents (in logs)

4.2 Robustness check: alternative inefficiency determinants

Given that institutions are multidimensional, many with qualitatively different features, we consider an alternative set of institutions to test the robustness of our findings. In light of the broad characteristics of innovation, government size and the degree of patent protection seem to be two key candidates for consideration. Government size relates to governance or institutional capacity, and often a fraction of the size of the government is allocated to directly funding research and innovation. However, when the size of the government contributes to bureaucratic delay, it can add to inefficiency in the innovation process.²⁴ Furthermore, the degree of patent protection is a direct indicator of regulation that is pertinent to innovation. Delays in the mechanisms to protect intellectual property (e.g., award of patents, court rulings in patent infringement cases) can add to inefficiencies in innovation production—especially since innovation is captured in the analysis by the number of patents. In contrast, the regulatory quality considered in the analysis above can be seen as a more general indicator of regulations (also encompassing dimensions of patent protection).²⁵

At the bottom of Table 1, we report summary statistics for a different set of determinants of innovation inefficiency. In particular, we now include the variables Government size and a patent protection index (Patent protection). Government size comes from the World Bank’s World Development Indicators, and it is measured by government consumption expenditure as a percent of GDP. The average for the full sample is 16.86 percent, with values ranging from 5.60 to 27.84 percent. In Table 2, we report that the average government size is larger for OECD countries relative to non-OECD countries (19.42 percent vs. 14.69 percent). The second determinant considered is the 1960 to 1990 average patent protection index, developed by Ginarte and Park (1997), and also reported in Park (2008).²⁶ As shown in Table 1, the average index value of patent protection for the full sample is 2, and the index values range from 0.66 (Jordan) to 4.14 (United States). As expected, average patent protection is larger for advanced economies, implying strengthened protection of intellectual property in these nations (see Table 2).

In Table 6, we report stochastic frontier estimates based on this alternative set of innovation inefficiency determinants. Focusing on the full sample, under column All, estimates for the deterministic part of the frontier (in the top panel of Table 6) are very similar to those previously reported estimates from Model 3.5 (see Table 3), except for the sign change in estimates attached to GDP per capita and e-participation.

Table 6 Innovation (Patents) Production Cross-Section Frontier Estimates. Alternative Inefficiency Determinants. Output Measure (Dependent variable): Patents (in logs)

The same observation applies when comparing estimates in Table 6 for OECD countries versus estimates for OECD countries from Model 4.5 (see Table 4). There are also some minor discrepancies when comparing estimates of the deterministic frontier for non-OECD countries versus estimates based on Model 5.5 (see Table 5). Namely, the sign attached to GDP per capita is now positive, which was not the case for Model 5.5.

Focusing on the bottom panel of Table 6, for the full sample, we report that the estimated coefficient attached to government size is negative but does not statistically significantly affect innovation production inefficiency. The latter result is observed for the subsample of OECD countries, while in the case of non-OECD countries, the reduction in innovation production inefficiency due to a larger government size is statistically significant. The patent protection estimated coefficient is also negative for the full sample and the OECD countries subsample, suggesting that patent protection reduces innovation production inefficiency; however, its effect is not statistically significant. For non-OECD countries, higher patent protection significantly reduces innovation production inefficiency. This finding can be seen as consistent with the notion that in the relatively less-prosperous non-OECD group, a part of the innovative activity might be of a duplicative or imitative nature. Firms in this subgroup might also be more likely to license complementary technologies from abroad (see Dohse et al., 2019). Therefore, an improvement in patent protection and/or government size (signifying better institutional/governance capacity) in these nations would reduce innovation production inefficiency.

4.3 Technical efficiency

In closing, it seems useful to consider the technical efficiency of our estimates. Technical efficiency in stochastic frontier enables one to evaluate the performance of the stochastic frontier models (see Jondrow et al., 1982 for details). These estimates are observation-specific measures that allow us to further characterize the efficiency variability in the sample (e.g., minimum and maximum).

Technical efficiency (TE_i) estimates, obtained by employing Jondrow et al. (1982) decomposition, are reported in Table 7 for the full sample and the subsamples of interest. The estimates reported were obtained employing the preferred specification (Models 3.5, 4.5, 5.5, respectively). The mean technical efficiency for the full sample is quite low at 0.171, and it has the minimum and maximum values of 0.006 and 0.990, respectively. The low mean technical efficiency in the full sample is likely due to the inclusion of largely heterogeneous countries. The more homogeneous sample of OECD countries yielded a mean technical efficiency of 0.908, with a minimum of 0.166 and a maximum of 0.995.

Table 7 Technical Efficiency in Innovation (Patents) Production

The mean technical efficiency in innovation production is notably smaller for non-OECD countries, with a value of 0.187. Also, the range in technical efficiency for non-OECD countries is larger (from 0.008 to 0.999). This finding is consistent with the generally lower quality of complementary inputs and government support/institutions in the relatively less prosperous non-OECD sub-group.

4.4 Discussion of results

With regard to the effect of corruption, which is an indicator of institutional quality, on innovation production efficiency, our findings support the greasing effects. The presence of corruption seems to enable firms to overcome some bottlenecks or regulatory lags and improve efficiency. These might be related to licensing, patenting, overcoming court (infringement) challenges, or to the procurement of essential inputs (see Goel, 2001). These findings can be seen as complementary to results in the literature that show corruption to facilitate the market introduction of innovations (Goel & Nelson, 2018). On the other hand, when a distinction is made between design and utility patents, the greasing effect of corruption on innovation production does not necessarily hold, at least not for the United States—see Goel and Saunoris (2020). From a broader policy perspective, these spillovers from corruption might make a case for tolerating some level of corrupt activity.

We further find that while strong patent protection is generally viewed as a crucial condition for the generation of innovation, the efficiency of innovation production generally is not impacted by patent protection (except for a select group of nations—Table 6). This contrasts with the innovation efficiency impacts of a broader index of regulation (i.e., regulatory quality) that is found across the board.

5 Conclusions

The predominant focus by policymakers and researchers on increasing the innovation capacity or innovativeness of different nations has left the consideration of the efficiency of innovation production behind. Yet, increasing competitiveness of the global markets and the relative scarcity of various resources necessitate that the efficiency of the innovation process also be considered on an equal footing.

Adding to the rather vast empirical literature on the determinants of technological change (see, for a recent example, Goel et al., 2023), this paper attempts to address the external or institutional drivers of innovation production efficiency. This aspect seems to have received relatively scant attention, especially with regard to the external influence of institutional factors on innovation efficiency (Fig. 1). Despite this, given the duplication of research effort in some innovation races and underinvestment in research in other cases due to imperfect appropriability, the focus on innovation production efficiency goes to the heart of the conservation and allocation of scarce research resources.

Using data on a large sample of nations over 2018–2020 and considering corruption, regulatory quality, and state fragility as alternative institutional dimensions, our results show that greater corruption facilitates (“greases”) efficiency in the production of innovations.²⁷ This is also the case with improvements in regulatory quality, while greater state fragility increases inefficiency. These findings for the overall sample are somewhat different for the OECD and non-OECD subsamples, although the greasing effect of corruption persists. This finding advises against framing blanket technology policies for all nations.

Another insight is that it is political uncertainty (captured by state fragility), rather than economic uncertainty (inflation), that significantly impacts innovation production efficiency. Uncertainty affects the expected payoffs from innovation, and can impact innovation production efficiency by altering the allocation and timing of inputs. The results with regard to the effects of uncertainty can be seen as complementary to studies focusing on the market introduction, rather than the production and related efficiency, of innovation. For instance, Goel and Nelson (2021) use firm-level survey information and show that both economic and political uncertainties enhance the market introduction of innovations—a finding consistent with the hedging story. In contrast, our results show that when it comes to the efficiency related to the production of innovation, only greater political uncertainty impacts firms, with inflation uncertainty having no significant impact. In contrast, when another dimension of uncertainty—economic policy uncertainty is considered—for instance, by William and Fengrong (2022), it is shown to discourage industry innovation.

From a policy perspective, this may not be great news—inflation management is relatively in greater control of policymakers than political uncertainty. Also, while the literature has shown that innovation uncertainty impacts innovation production (Goel & Ram, 2001), our results show that such uncertainty does not significantly affect the efficiency of innovation production. Thus, the present research points to the need for making a distinction between production and efficiency in the innovation process.

Turning to the questions posed in the Introduction, we are able to provide the following answers:

• Which institutions significantly impact innovation production efficiency?

Of the different institutional aspects considered (i.e., regulatory quality, corruption, state fragility, government size, patent protection), we find that all of them impact innovation production efficiency to some degree, depending upon the model/sample employed. The economically significant influences of the different institutions on innovation production efficiency seem to be not widely recognized in policymaking.

• Are the innovation production efficiency impacts of different institutions alike?

No, the innovation production efficiency impacts of different institutions are not alike. This is true both qualitatively and quantitatively.

• Are the drivers of innovation production efficiency different in OECD nations from the rest of the world?

Yes, we find differences in the impact of institutional influences on innovation production efficiency in the OECD group compared to the rest of the world. Specifically, corruption does not have a significant greasing effect for OECD nations’ innovation production efficiency, but it does significantly grease the innovation production process in other nations. Furthermore, government size and patent protection have a significant bearing on innovation production efficiency in the non-OECD group, but not in the OECD group (Table 6). Finally, in terms of relative magnitudes, the elasticity of patents with respect to researchers in the OECD group was substantially greater (= 0.9, Model 4.5) than that in the non-OECD group (= 0.6, Model 5.5), implying greater returns/productivity of researchers in the more developed OECD nations.

Besides adding insights into innovation production efficiency, the overall message from this work is that in regard to innovation production efficiency, not all institutions work in tandem. And the differences in the institutional influences across different subsets of nations advise against the formulation of blanket technology policies (see Nelson, 1993; also, Kravtsova & Radosevic, 2012). On a positive note, the role of effective patent protection comes through, although it is less clear whether the governmental efforts to appropriate the rewards to inventors via strengthened intellectual property protections also consider the related efficiency aspects. Finally, while this research provides some new insights into the role of institutions in fostering innovation production efficiency, consideration of the social efficiency of innovation, related to spillovers and duplicative research efforts (Haruna & Goel, 2011), remains a challenge for related empirical research.

Appendices

Appendix

Variable definitions from GII

Complementary to the information reported in Table 1, below we provide more details about the variables we employ from GII (Fig. 1).

Patents are the number of resident patent applications filed at a given national or regional patent office (per billion PPP$ GDP); GII (2020, p. 357).

“Domestic credit (as a percent of GDP) to private sector refers to financial resources provided to the private sector by financial corporations, such as through loans, purchases of nonequity securities, and trade credits and other accounts receivable that establish a claim for repayment. For some countries, these claims include credit to public enterprises.” GII (2020, p. 353).

E-participation is “... is derived as a supplementary index to the United Nations E-Government Survey. It extends the dimension of the Survey by focusing on the government’s use of online services in providing information to its citizens or “e-information sharing”, interacting with stakeholders or “e-consultation” and engaging in decision-making processes or “e-decisionmaking.”” GII (2020, p. 351). The index ranges from 0 to 1, with 1 denoting greater e-participation.

“Researchers in R&D are professionals engaged in the conception or creation of new knowledge, products, processes, methods, or systems and in the management of the projects concerned. Postgraduate PhD students (ISCED97 level 6) engaged in R&D are included.” GII (2020, p. 350).