Riding the Innovation Wave: Global Insights Into Firm Performance

The Impact of National Innovation System on Firm Performance

While understanding firms’ responses to changes in NISs is essential as many countries are increasing their reliance on innovation to boost economic development and shape competitive advantages, this relationship is complex and subject to multifaceted, often contradictory interpretations. The first perspective suggests that NISs positively influence firm performance by fostering innovation and facilitating collaboration among economic actors via several channels. First, conventional theories on NIS models, such as Nonaka and Takeuchi’s (1995) Socialization, Externalization, Combination, and Internalization model and Etzkowitz and Leydesdorff’s (2000) Triple Helix model, assert that the promotion of NISs can facilitate knowledge spillovers, which directly benefit firms through improving their accessibility to knowledge generated by academia, research institutions and government agencies. Chen et al. (2018) highlight that the establishment of successful NISs entails continual interactions among various components, facilitating the diffusion of advanced knowledge and technologies across firms and industries. Furthermore, the collaborative networks formed from such interactions can promote resource sharing and joint problem-solving among involved parties, which leads to more cost-effective innovation and greater technological absorption (Karlsson & Warda, 2014). Another key mechanism fostering firm performance is the growth of dynamic capabilities resulting from innovation-enabling contexts that encourage firms to quickly adapt to market changes and leverage emerging opportunities. Dynamic capabilities, which are referred to as a firm’s ability to integrate, build, and reconfigure internal and external competencies, can enhance corporate resource utilization efficiency and assist them in shaping their innovation-based competitive advantages (Teece, 2018). Moreover, incentives provided by NISs (e.g., R&D subsidies, tax incentives, and innovation grants) can reduce the financial barriers to corporate innovation, thus driving firms to engage in the development of new products, processes, and technologies. This increasing utilization of governments’ supporting schemes for corporate innovation subsequently translates into more innovation outputs and improved corporate profitability (Cunningham et al., 2018). Looking forward, such government policies can foster a culture encouraging firms to continuously innovate to maintain their competitiveness, which is essential for sustaining long-term competitiveness in global markets.

Conversely, counterarguments cast doubt on the amplifying effect of NIS on firms’ ability to generate profits, at least in the short term. According to Edler and Fagerberg (2017), governments’ efforts to create more complete innovation systems may not always align with anticipated outcomes. Instead of focusing on businesses as the focal point in economic development strategies, governments come up with policies that prioritize macro-level outcomes. These policies can be beneficial for large firms that have available resources to pursue innovative activities; however, smaller firms that are under resource constraints or are tech-lagging may find it difficult to align their strategies with these priorities, resulting in diminished performance while trying to compete by means of innovation (Acs & Audretsch, 1988; Wellalage & Fernandez, 2019). From the viewpoint of the innovation cost paradox, despite that NISs are designed to encourage innovation, firms may falsely estimate the financial burden of adopting new technologies, conducting R&D and complying with stricter innovation-related regulations, causing innovation spending to exceed the short-term benefits (Terry, 2023; Wu & Liu, 2023). Furthermore, while firms might be more inclined to innovate following NIS improvements, their limited knowledge and experience disable them from effectively planning and managing innovation projects (Cruz-Gonzalez et al., 2015; Hoang, 2025a). This is especially true in emerging economies where firms often lack the knowledge, infrastructure, and networks needed to leverage the benefits of an upgraded NIS (Cirera & Maloney, 2017). Instead of improving performance, the increased complexity of the innovation system may lead to a widening gap between well-resourced firms and those struggling to keep pace, deepening inequalities and hindering overall productivity.

Even when firms are supplied with sufficient capital through government funding, firms might be subject to the crowding-out effect. Public R&D grants can lead firms to substitute external funding for their own internal resources originally allocated for innovation projects, thereby shifting (part of) the financial risk associated with innovation to public funders. This reliance on “free money” can result in inefficiencies, as inadequate government resource allocation and firms’ short-term budgeting misallocations diminish the effectiveness of financial support (Fang et al., 2023). In addition, even when firms take innovation seriously, excessive government interventions in NISs, such as direct funding of public research institutions or state-sponsored innovation projects, can reduce firms’ incentives to invest independently in innovation (Aghion & Jaravel, 2015). As a result, the lack of vigorous innovation strategies, coupled with firms’ over-reliance on government funding, could hamper firms’ long-term growth potential.

Despite differing opinions in the literature, the positive and negative views on the relationship between national innovation systems (NIS) and firm performance are not mutually exclusive. While some theories support a positive long-run relationship, they do not imply that the benefits materialize immediately once an NIS improves. Rather, these theories emphasize that by strengthening NISs, governments create a conducive environment for firms to learn, collaborate, and continuously innovate, thereby fostering sustainable economic growth (Freeman, 1995; Lundvall, 1992). Our framework is grounded in NIS theory and is complemented by the Schumpeterian model of innovation and competition (Aghion & Howitt, 1992; Schumpeter, 1942), highlighting NIS as a dual mechanism: it generates long-term gains through knowledge spillovers and capability accumulation, but it also triggers creative destruction, intensifying rivalry and eroding incumbent rents as new technologies diffuse. In such settings, firms face the innovation cost paradox – high upfront expenditures on R&D and capability building while immediate profits remain elusive (Peters & Taylor, 2017). This notion is reinforced by absorptive capacity theory, which argues that firms must invest in internal R&D and related knowledge assets to recognize, assimilate, and exploit external knowledge flows (Cohen & Levinthal, 1990). The returns to these investments are inherently lagged: only after sufficient absorptive capacity is built can firms fully capitalize on the spillovers created by an improved NIS. Similarly, the dynamic capabilities perspective (Teece et al., 1997) stresses that firms need to reconfigure and integrate resources to capture value from new technological opportunities, a process that entails significant adjustment costs and organizational learning before payoffs emerge.

Empirical evidence is consistent with this logic. Lee and Wu (2016) show that current-year R&D spending depresses contemporaneous firm performance, whereas accumulated R&D capital over the previous 3 years enhances it, illustrating the lag between innovation investment and financial returns. These findings support the idea that innovation, although value-creating in the long run, is initially associated with profit pressures and uncertainty. Investors, however, may anticipate these future returns, so forward-looking market valuations can remain stable or even increase despite short-term earnings reductions (Peters & Taylor, 2017). Taken together, these theoretical strands provide a coherent foundation for our hypothesis: improvements in a country’s NIS are expected to impose short-run cost pressures on firms’ accounting performance while laying the groundwork for long-run value creation that forward-looking investors may already price into firm valuations.

While ample studies have enriched the association between firms’ innovation and their performance, none of the prior studies address how firm performance reacts to changes in the country-level innovation systems. Understanding this link is crucial, as NISs shape the broader innovation environment in which firms operate and may impose costs and constraints that differ fundamentally from firm-level innovation choices. To seek an answer for this interesting yet imperative relationship, we propose Hypothesis 1 as follows:

Channel Hypothesis – Innovation Intensity

Corporate investments in technology and knowledge creation are expected to increase following heightened government efforts to foster innovative activities. Support from the government can be manifested in several forms and generally categorized into two groups of input support (e.g., financial aid, infrastructure, human capital) and output support (e.g., intellectual property rights protection). Such supports provide firms with access to necessary resources and facilities to initiate their innovative activities, ensure their legal ownership of innovation outputs, and discourage any commercialized ideas that violate the owners’ copyrights (Nguyen et al., 2024; Xu et al., 2021). These factors collectively increase firms’ likelihood of investing in innovation and producing measurable outputs, such as patents and new products (Castellacci & Natera, 2013).

At the same time, stronger national innovation systems often induce firms to intensify their innovation efforts, which can amplify short-run cost pressures and adversely impact firms’ short-term profitability. Innovation projects entail substantial upfront expenses and uncertain commercialization prospects, signifying that foreseeable expenditures precede measurable benefits, at least in the short run (Hall & Lerner, 2010). Therefore, firms that intensify R&D/innovation spending often experience significant financial strain (Lichter et al., 2025; McKinsey & Company, 2024). Elevated R&D intensity also increases earnings volatility and reduces free cash flows, leading equity markets to discount firm value in the near term (Chan et al., 2001; D. Li, 2011). This deterrent effect of the substantial upfront investments on firm performance, as suggested by Coad and Rao (2010), is not limited to small firms but also to those with high levels of slack resources, which aligns with the innovation cost paradox. Additionally, the pressure to innovate can lead firms to divert resources from core operations toward R&D, which in turn disrupts established routines and creates coordination frictions (Ahlawat et al., 2019). As a result, firms may experience short-term productivity slowdowns and less effective operations before any innovation benefits materialize (Hall & Lerner, 2010).

Together, these underlying reasons imply that innovation-intensive responses to improved NIS conditions impose immediate financial strain on firms, lowering both profitability and market valuation in the short run. Hypothesis 2 is, therefore, stated as follows:

Channel Hypothesis – Industry Competition

An improved NIS fosters innovation across firms, which is expected to increase the level of competition within industries. Bloom et al. (2019) suggest that innovation policies that promote widespread innovation capabilities intensify rivalries as firms compete for market share through differentiated products and technologies. Similarly, Granstrand and Holgersson (2020) highlight that strong innovation ecosystems amplify competition by reducing barriers to entry and enabling new firms to challenge established players in an industry. Specifically, greater diffusion of innovation capacity expands the pool of technologically capable firms, raising the number of rivals able to imitate or overtake competitors. This mechanism is consistent with Schumpeterian models, where innovation induces business-stealing effects that squeeze competitors’ profits and reshape the market positions (Aghion et al., 2015). As more innovative firms simultaneously join the competition, incumbent firms face mounting pressure to protect their relevance in the market, forcing them to tradeoff short-term profits for innovation opportunities.

The reduced firm performance can be explained via three major lenses. First, intense competition can erode profit margins as firms engage in price wars or invest heavily in marketing to differentiate their offerings (Coad et al., 2016; Porter, 1985). Second, firms may face higher costs related to rapid innovation cycles, such as frequent product upgrades and the need to attract and retain top talent (Dockner & Siyahhan, 2015). Accelerated technological change often requires continuous capability renewal, leading firms to invest heavily in advanced equipment, specialized human capital, and organizational restructuring. These recurring adaptation costs can significantly raise operating expenses and strain short-term financial performance (Jacobides et al., 2018). Third, the competition fueled by increasing innovation support can also be attributed to innovation herding – a phenomenon where firms within an industry engage in innovation activities to respond to similar behaviors by competitors (Xu et al., 2012). This innovation herding arises as firms seek to capitalize on the opportunities offered by an enhanced NIS to create their own innovation remarks. On the one hand, simultaneous engagement in innovation activities among various firms in an industry can accelerate technological advancements and market offerings. On the other hand, the convergence of innovation efforts increases pressure on firms to match the prevailing innovation tempo, often resulting in rushed or poorly integrated innovation investments and resource misallocation that potentially undermine firms’ short-run overall performance (Ahlawat et al., 2019; Yiu et al., 2020).

All in all, while an improved NIS can enhance competitiveness, the associated costs and pressures stemming from heightened competition may outweigh the benefits in the short term, negatively impacting firm performance. These rationales lead to our Hypothesis 3:

Global Innovation Index

This study examines the influence of national innovation on firm performance by employing the Global Innovation Index (GII) as a comprehensive metric for a country’s innovation capacity and output. Developed by the WIPO, the GII provides a multidimensional framework that evaluates the extent to which national policies, investments, and institutional frameworks facilitate innovation. Its extensive cross-country coverage and standardized methodology make it a robust and comparative indicator for analyzing the role of national innovation in shaping firm-level outcomes.

In the GII, seven innovation pillars are categorized into innovation inputs and innovation outputs. Innovation inputs reflect the enabling environment for innovation, including institutions, human capital and research, infrastructure, market sophistication, and business sophistication. Meanwhile, innovation outputs measure knowledge and technology production, as well as creative outputs, capturing tangible innovation-driven results such as patent filings, high-tech exports, and knowledge diffusion. Including those key input and output measures of NIS makes the GII particularly suitable for holistically analyzing the extent to which national innovation efforts translate into firm-level benefits. Furthermore, the employment of the GII as a country-level innovation measure enables cross-country comparisons, enabling this study to assess whether changing NIS conditions consistently impact firm-level outcomes, by that offering invaluable insights for policymakers aiming to enhance innovation-driven firm performance and national competitiveness through innovation-led growth

While the GII provides a relatively comprehensive measure of NIS, it has inherent limitations. It should be noted that some components of the GII rely on perception-based indicators and self-reported data, which may introduce subjectivity and measurement noise. Moreover, cross-country comparability may be affected by discrepancies in countries’ economic and institutional contexts, potentially leading to systematic biases favoring higher-income economies (Cirera & Maloney, 2017; World Intellectual Property Organization, 2023). This research seeks to address these shortcomings of GII by designing a research model that captures fundamental country-level indicators to enhance GII’s credibility as a proxy for NISs.

Research Model

To investigate the impact of NIS on firm performance, the fixed effect model is employed in this study. The use of the fixed effect model allows better controls for the variation at both the time series and cross-section that might otherwise bias the estimated effect. The research model is designed as follows:

F P i , t = β 0 + β 1 GI I c , t − 1 + β 2 SIZ E i , t − 1 + β 3 LEVERAG E i , t − 1 + β 4 CAS H i , t − 1 + β 5 REV _ GROWT H i , t − 1 + β 6 TANGIBILIT Y i , t − 1 + β 7 NW C i , t − 1 + β 8 GDP _ GROWT H c , t − 1 + β 9 INFLATIO N c , t − 1 + β 10 RULEOFLA W c , t − 1 + β 11 GIN I c , t − 1 + β 12 NATION _ RDEX P c , t − 1 + f firm + f country + f year + ε i , t

(1)

in which FP is measured by two alternative proxies, namely return on assets (ROA) and Tobin’s Q (TOBINSQ). ROA is computed by dividing a firm’s net income by its average total assets. Meanwhile, to calculate TOBINSQ, we use the following formula:

TOBINS Q i , t = Market value of equit y i , t + Total liabilitie s i , t Book value of equit y i , t + Total liabilitie s i , t

(2)

Although both ROA and Tobin’s Q are extensively utilized as proxies for firm performance in the literature, these two measures differ fundamentally. While ROA captures the accounting-based performance, Tobin’s Q represents market-based performance by reflecting market valuation relative to asset replacement costs. Thus, simultaneously employing ROA and TOBINSQ may not yield completely identical statistical results but can offer some interesting insights into how the accounting- and market-based performance vary with a changing NIS context. Regarding national innovation, GII represents the WIPO’s composite Global Innovation Index, which is re-evaluated on an annual basis to update the current capacity of innovation systems worldwide.

Two sets of control variables at the firm level and country level are employed to capture the differences among firms and countries. At the firm level, we follow H. H. Huang et al. (2018), Hu and Zhang (2021), and Gull et al. (2022) to account for firm size (SIZE), leverage (LEVERAGE), cash holdings (CASH), sales growth (REV_GROWTH), tangible assets (TANGIBILITY), and net working capital (NWC). At the country level, GDP growth (GDP_GROWTH), inflation (INFLATION), rule of law (RULEOFLAW), GINI index (GINI) and national spending for R&D (NATION_RDEXP) are included. To further account for the heterogeneity across firms, countries and periods, we include firm, country and year fixed effects in Model 1. The definition of our variables and associated sources are presented in Table A1.

Following the conventional approach in finance research, we winsorize all firm-level continuous variables at the top and bottom percentiles to relieve the outlier effect that can lead to estimation bias. Robust standard errors are clustered by country to alleviate the heteroskedasticity and autocorrelation misspecification.

Data

Data used in this study are compiled from several sources. Firm-level data are collected from Thomson Reuters’ Refinitiv. GII and its components are retrieved from WIPO’s annual Global Innovation Index reports. The macroeconomic and institutional variables are downloaded from the World Bank open database.

In accordance with the conventional approach in finance research, we exclude financial firms since the financial industry is typically highly regulated and is subject to distinct accounting and operation natures. Our sample is then gone through a filtering process to eliminate observations with insufficient data to estimate Equation 1. Eventually, the final sample consists of 77,819 observations of 16,228 firms from 2011 to 2019. These firms are scattered across 50 countries (see Table A2 for the complete list of countries), enabling us to observe the examined impact in an international context, thus adding significant understanding to the general knowledge revolving around NISs. Furthermore, the cross-country sample choice facilitates the generalization of the research findings, which is expected to help policymakers in formulating effective national innovation strategies.

We restrict our sample to 2019 since extending data beyond that point could introduce uncontrollable confounding factors, such as the health pandemic which triggered unprecedented disruptions in global supply chains, labor markets, and innovation activities, and international conflicts, led by the Russia–Ukraine war, that exacerbate economic instability. By focusing on the pre-2019 period, we aim to maintain the internal validity of our findings, ensuring that observed effects are attributable to changes in NISs rather than external crises.

Descriptive Statistics

The descriptive statistics of variables used in this study are presented in Panel A of Table 1. The statistics show that the firms averagely have a ROA as low as 0.3% and are highly overvalued (mean of TOBINSQ = 2.485). It seems that the market is optimistic about the firms’ prospects despite that firms are not profitable at the moment. On the flip side, this contradiction also presents a challenge for firms as the market may correct their valuation once future performance does not align with their expectation. Debt only accounts for a low proportion (≈11%) of firms’ total assets, and 12.7% of their assets are under the cash form. Meanwhile, more than one-fourth of the total assets are property, plant and equipment. The growth rate of revenue is quite noticeable at 17.4% per annum.

OPEN IN VIEWER

Table 1.

Descriptive Statistics.

Panel A. Summary Statistics.

Variable	Observations	Mean	Median	Standard Deviation	Minimum	Maximum
ROA	77,819	0.003	0.029	0.164	−1.174	0.303
TOBINSQ	77,819	2.485	1.409	4.417	0.284	40.801
GIIt−1	77,819	47.096	47.470	9.904	23.07	68.300
SIZEt−1	77,819	7.272	7.273	2.796	0.278	15.921
LEVERAGEt−1	77,819	0.109	0.047	0.148	0	0.797
CASHt−1	77,819	0.127	0.082	0.141	0	0.841
REV_GROWTHt−1	77,819	0.174	0.061	0.834	−1.117	8
TANGIBILITYt−1	77,819	0.278	0.233	0.226	0	0.911
NWCt−1	77,819	0.160	0.155	0.301	−1.551	0.888
GDP_GROWTHt−1	77,819	4.408	4.383	2.916	−10.079	24.370
INFLATIONt−1	77,819	2.391	2.000	2.536	−2.097	48.700
RULEOFLAWt−1	77,819	0.432	−0.010	0.885	−0.858	2.125
GINIt−1	77,819	0.358	0.358	0.048	0.240	0.539
NATION_RDEXPt−1	77,819	1.769	1.912	1.006	0.055	4.797

Note. Panel A, Table 1 reports summary statistics of all variables in this study. Variable definitions and data sources are shown in Table A1. All continuous variables are winsorized at 1st and 99th percentiles.

OPEN IN VIEWER

Panel B. Pearson Pairwise Correlation Coefficient Matrix.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)
(1) ROA	1.000
(2) TOBINSQ	−.092	1.000
(3) GIIt−1	−.111	−.019	1.000
(4) SIZEt−1	.235	−.116	−.214	1.000
(5) LEVERAGEt−1	−.082	−.005	−.027	.113	1.000
(6) CASHt−1	−.082	.138	.251	−.168	−.230	1.000
(7) REV_GROWTHt−1	−.031	.022	.018	−.038	−.003	.058	1.000
(8) TANGIBILITYt−1	.082	−.042	−.246	.232	.290	−.314	−.055	1.000
(9) NWCt−1	.164	−.015	.094	−.084	−.283	.457	.040	−.365	1.000
(10) GDP_GROWTHt−1	.105	.052	−.323	.241	−.176	.064	.031	.047	.082	1.000
(11) INFLATIONt−1	.047	.000	−.488	.145	.029	−.123	.020	.107	−.057	.129	1.000
(12) RULEOFLAWt−1	−.162	−.023	.698	−.355	.128	.032	.003	−.192	−.011	−.626	−.383	1.000
(13) GINIt−1	.108	.054	−.378	.226	−.035	.040	−.002	.088	.021	.337	.274	−.676	1.000
(14) NATIONAL_RDEXt−1	−.068	−.025	.775	.012	−.059	.214	−.001	−.204	.107	−.176	−.384	.466	−.246	1.000

Note. Panel B, Table 1 reports Pearson pairwise correlation coefficients of variables in Model 1. Variable definitions and data sources are shown in Table A1. Coefficients significant at 5% level are typed in bold. All continuous variables are winsorized at 1st and 99th percentiles.

Panel B, Table 1 reports the Pearson correlation coefficient matrix of variables in Model 1. It can be seen that almost all correlation coefficients are significant at the 5% level, stressing the necessity to include them in Model 1. Moreover, none of the coefficients has their values exceed 0.8, thus further indicating that our estimates are unlikely to be falsely driven by the multicollinearity concern.

Multivariate Regression

This section examines the direct effect of NIS on firm performance by estimating Model 1 using different model specifications and reporting the results in Table 2. First, we conduct bivariate regressions of firm performance measures on GII_t−1 (Columns 1 and 3). Next, Model 1 is estimated (Columns 2 and 4). The coefficients of GII_t−1 in all tests are significantly negative, indicating that an NIS upgrade is unbeneficial for firms in the short run, both in terms of accounting profits (ROA) and market valuation (TOBINSQ). These results validate our Hypothesis 1, which suggests a negative association between NIS development and firm performance. As NIS improves, firms may have better access to the facilities and financial resources needed for innovation projects and are more incentivized to engage in such activities. Increasing innovation spending represents a shift in operational direction that entails current financial burdens in exchange for potentially larger long-term gains. As a result, we witness a reduction in firm performance in the short run. Additional robustness tests, including estimating Model 1 in subsamples of non-high-tech industries and high-tech industries, and with Chinese and Indian firms dropped, are presented in Appendix A3.

OPEN IN VIEWER

Table 2.

Baseline Model Regressions.

	ROA		TOBINSQ
	(1)	(2)	(3)	(4)
Variable	Bivariate regression with FEs	Baseline model	Bivariate regression with FEs	Baseline model
GIIt−1	−0.002*** (−3.287)	−0.001** (−2.101)	−0.125*** (−3.920)	−0.069** (−2.563)
SIZEt−1		−0.009 (−1.475)		−1.142*** (−7.644)
LEVERAGEt−1		−0.039* (−1.910)		0.624** (2.178)
CASHt−1		0.052*** (2.947)		1.203*** (3.848)
REV_GROWTHt−1		0.008*** (4.616)		0.016 (1.134)
TANGIBILITYt−1		0.016 (1.447)		−0.747*** (−3.853)
NWCt−1		0.005 (0.433)		−0.625** (−2.644)
GDP_GROWTHt−1		0.000 (0.626)		−0.023 (−0.724)
INFLATIONt−1		0.000 (0.519)		0.000 (0.019)
RULEOFLAWt−1		−0.009 (−0.501)		−0.771** (−2.517)
GINIt−1		0.072 (0.612)		−7.970 (−1.121)
NATION_RDEXPt−1		−0.011 (−1.628)		−0.285 (−1.561)
Constant	0.111*** (3.377)	0.114 (1.506)	8.363*** (5.577)	17.915*** (4.221)
Firm, year and country FEs	Y	Y	Y	Y
Observations	77,819	77,819	77,819	77,819
Adjusted R−squared	.590	.593	.781	.792

Note. Table 2 reports the regression results of Model 1 with ROA and TOBINSQ as two firm performance proxies. Columns 1 to 2 report the results of the bivariate regression and Model 1 regression, using ROA as the dependent variable. Columns 3 to 4 replicate Columns 1 to 2, replacing ROA with TOBINSQ as the dependent variable. Variable definitions and data sources are shown in Table A1. Robust standard errors are clustered by firm. Robust t-statistics are in parentheses. All continuous variables are winsorized at 1st and 99th percentiles.

*, **, and *** indicate levels of significance at 10%, 5%, and 1%, respectively.

To get further insight into this association, we delve into the component analysis. We separately examine each of the seven pillars that constitute the GII, namely Institutions, Human capital and research, Infrastructure, Market sophistication, Business sophistication, Knowledge and technology outputs, and Creative outputs. The results shown in Table 3 imply that the impact of NIS is driven by the infrastructure and creative outputs pillars. In parallel, the coefficient of Market sophistication (MARKETSOPHIS_t−1) is significant in Panel A, suggesting that firms’ accounting performance is negatively influenced by this pillar, which comprises the key aspects of ease of getting credit, investment, trade, and market competition and scale. On the contrary, TOBINSQ is not affected by this dimension.

OPEN IN VIEWER

Table 3.

Component Analysis.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variable	Institutions	Human capital and research	Infrastructure	Market sophistication	Business sophistication	Knowledge and technology outputs	Creative outputs
PANEL A: ROA
INSTITUTIONSt−1	−0.000 (−0.461)
HUMANCAPITALRESt−1		0.000 (0.078)
INFRASTRUCTUREt−1			−0.001** (−2.489)
MARKETSOPHISt−1				−0.000 (−1.460)
BUSINESSSOPHISt−1					−0.000 (−0.817)
OUTPUT_KNOWTECHt−1						−0.000 (−1.083)
OUTPUT_CREATIVEt−1							−0.000* (−1.826)
Main effects	Y	Y	Y	Y	Y	Y	Y
Control variables	Y	Y	Y	Y	Y	Y	Y
Firm, year and country FEs	Y	Y	Y	Y	Y	Y	Y
Observations	77,819	77,819	77,819	77,819	77,819	77,819	77,819
Adjusted R-squared	.593	.593	.593	.593	.593	.593	.593
PANEL B: TOBINSQ
INSTITUTIONSt−1	−0.013 (−1.106)
HUMANCAPITALRESt−1		0.016 (1.136)
INFRASTRUCTUREt−1			−0.029* (−1.976)
MARKETSOPHISt−1				−0.040* (−1.733)
BUSINESSSOPHISt−1					−0.031 (−1.235)
OUTPUT_KNOWTECHt−1						0.002 (0.195)
OUTPUT_CREATIVEt−1							−0.028*** (−2.766)
Main effects	Y	Y	Y	Y	Y	Y	Y
Control variables	Y	Y	Y	Y	Y	Y	Y
Firm, year and country FEs	Y	Y	Y	Y	Y	Y	Y
Observations	77,819	77,819	77,819	77,819	77,819	77,819	77,819
Adjusted R-squared	.792	.792	.792	.792	.792	.792	.792

Note. Table 3 reports the regression results of Model 1, with ROA (Panel A) and TOBINSQ (Panel B) as two firm performance proxies, after replacing GII with each of its innovation pillars. Seven innovation pillars include Institutions, Human capital and research, Infrastructure, Market sophistication, Business sophistication, Knowledge and technology outputs, and Creative outputs. Variable definitions and data sources are shown in Table A1. Robust standard errors are clustered by firm. Robust t-statistics are in parentheses. All continuous variables are winsorized at 1st and 99th percentiles.

*, **, and *** indicate levels of significance at 10%, 5%, and 1%, respectively.

These results reveal that each GII pillar impacts firm performance differently. The findings that infrastructure is a primary driver of firm performance align with the infrastructure-driven competition effect. Specifically, improvements in infrastructure lower search, logistics, and entry costs economy-wide, leading to compressed markups and intensified rivalry – the mechanisms in settings where connectivity reduces price dispersion and strengthens competitive pressure (Brown & Goolsbee, 2002; Brynjolfsson & Smith, 2000). Market-linking infrastructure also raises the threats of new entrants, thus slashing average profitability in the short run as competition intensifies (Donaldson, 2018; Porter, 1985).

Likewise, an environment abundant in creative outputs implies dense product variety and fierce competition. Such markets are characterized by high revenue volatility and squeezed average profits even as a few winners capture large rents (Caves, 2003). Creative outputs might also bid up skilled employees’ wages, resulting in raising businesses’ input costs across the economy (J. Li et al., 2020). Near-term firm performance, hence, is deemed to decline. Moreover, our finding that market sophistication is negatively associated with Tobin’s Q but unrelated to accounting performance supports the efficient market hypothesis, suggesting that increased competition – due to anticipated heightened innovative activities – factors into valuations faster than reported profit, which might not recognize a profitability reduction in the immediate accounting period (Huang, Lin et al., 2023).

The non-significance of the other pillars (institutions, human capital and research, business sophistication, and knowledge and technology outputs) does not necessarily imply irrelevance. Research on NIS emphasizes complementarities and capabilities, which means these inputs contribute to absorptive capacity and the efficiency of knowledge use, but effects typically materialize over longer horizons and through interaction with downstream enablers (Fagerberg & Srholec, 2008). We therefore cautiously interpret these insignificant coefficients as slower, capability-dependent transmission rather than an absence of impacts.

Endogeneity Diagnoses

Alternative Measures of Firm Performance and National Innovation System

To inspect whether our finding is sensitive to measurement choice, we make use of different measures of firm performance and GII in our Model 1. Since firms’ profitability among a diverse range of industries might not be comparable, we adjust ROA by industry; that is, we compute the annual industry-average firm performance, then take the difference between a firm’s ROA and its industry average. The use of industry-adjusted measures allows us to focus on examining firm performance in relation to other companies within the same industry, thereby enabling an assessment of the impact of the GII from an industry-specific perspective. As for NIS, we transform the 1-year lag of GII to derive the natural logarithm of GII_t−1 as an alternative measure. Furthermore, since GII is subject to bias due to inappropriate weight generation and indicator aggregation (Alqararah, 2023), employing another proxy for NIS is necessary. We make use of the OECD Research and Development Statistics Data, which offer data on Gross domestic expenditure on R&D, R&D personnel, and Government budgets on R&D. We construct an alternative OECD-based innovation index using the data reduction technique Principal Component Analysis (PCA) method. Applying PCA helps retain the combined informational content of the innovation indicators and provides a more neutral representation of the present state of the national innovation system (Shabir et al., 2023). In addition, PCA serves to lessen potential measurement noise and reduce multicollinearity among these closely related variables (Chao et al., 2017). Only the first factor extracted from the PCA is employed, as the factor itself accounts for 94.88% of the variation of the newly created NIS index. The higher the value of the factor, the greater the innovation intensity of the country in a given year.

As reported in Table 4, we find the results using alternative proxies for firm performance and NIS consistent with our main finding in Section 4.1. That is, all the coefficients of NIS variables are statistically negative at 10% significance level for ROA and at (at least) 5% significance level for TOBINSQ. These pieces of evidence bolster our confidence in the robustness of the findings.

OPEN IN VIEWER

Table 4.

Alternative Measures.

	(1)	(2)	(3)
Variable	Industry adjusted firm performance	Logarithm of GII	OECD-based innovation index
PANEL A: ROA
GIIt−1	−0.001* (−1.827)
LN_GIIt−1		−0.044* (−1.704)
OECDINNOt−1			−0.044* (−1.815)
Main effects	Y	Y	Y
Control variables	Y	Y	Y
Firm, year and country FEs	Y	Y	Y
Observations	77,819	77,819	16,706
Adjusted R-squared	.587	.593	.666
PANEL B: TOBINSQ
GIIt−1	−0.062** (−2.292)
LN_GIIt−1		−2.639** (−2.024)
OECDINNOt−1			−1.854*** (−3.519)
Main effects	Y	Y	Y
Control variables	Y	Y	Y
Firm, year and country FEs	Y	Y	Y
Observations	77,819	77,819	16,706
Adjusted R-squared	.787	.792	.606

Note. Table 4 reports the regression results of Model 1, with ROA (Panel A) and TOBINSQ (Panel B) as two firm performance proxies, after replacing PF with the industry-adjusted firm performance (Column 1), GII_t−1 with the natural logarithm of GII_t−1 (Column 2) and OECD-based innovation index (Column 3). Variable definitions and data sources are shown in Table A1. Robust standard errors are clustered by firm. Robust t-statistics are in parentheses. All continuous variables are winsorized at 1st and 99th percentiles.

*, **, and *** indicate levels of significance at 10%, 5%, and 1%, respectively.

Entropy Balancing for Continuous Treatment

Selection bias can arise from our international research setting given that our sample might skew toward larger listed firms whose financial statements are reported more consistently. The exclusion of firms with incomplete data might lead to the overrepresentation of successful firms rather than capturing the overall performance of all listed firms. To examine whether our research results are biased, we use Tübbicke’s (2022) entropy balancing for continuous treatments (EBCT) method, which can be utilized to eradicate the correlations between the treatment (and its higher moments) with all covariates. Entropy balancing is considered “doubly robust” as compared with the conventional propensity score matching (Zhao & Percival, 2017). The summary statistics on balancing quality, presented at the foot of each panel of Table 5, indicate that the entropy balancing is successfully done since the R-squared of the regression of firm performance (ROA and TOBINSQ) on all the control variables equals zero.

OPEN IN VIEWER

Table 5.

Endogeneity Diagnoses.

	(1)	(2)	(3)
VARIABLES	EBCT	IV/2SLS First stage	IV/2SLS Second stage
PANEL A: ROA
GIIt−1	−0.001* (−1.822)		−0.002*** (−4.496)
ARTICLESt−1		0.047*** (183.974)
Control variables	Y	Y	Y
Firm, year and country FEs	Y	N	N
Firm FE, and country and year dummies	N	Y	Y
Observations	77,819	77,819	77,819
Adjusted/Uncentered R−squared	.606		.014
EBCT summary statistics on balancing quality
	R-squared	F-statistic	p-Value
Pre-balancing	.132	1,147.134	.000
Post-balancing	.000	0.000	1.000
Post 2SLS’s specification tests
F-test of excluded instruments		33,846.46
Underidentification test			1.1 × 104 ***
Weak identification test			3.4 × 104 ***
Hansen J-statistic			0.000
PANEL B: TOBINSQ
GIIt−1	−0.080*** (−3.417)		−0.067*** (−6.606)
ARTICLESt−1		0.047*** (183.974)
Control variables	Y	Y	Y
Firm, year and country FEs	Y	N	N
Firm FE, and country and year dummies	Y	Y	Y
Observations	77,819	77,819	77,819
Adjusted/Uncentered R-squared	.850		0.078
EBCT summary statistics on balancing quality
	R-squared	F-statistic	p-Value
Pre-balancing	.048	382.391	.000
Post-balancing	.000	0.000	1.000
Post 2SLS’s specification tests
F-test of excluded instruments		33,846.46
Underidentification test			1.1 × 104 ***
Weak identification test			3.4 × 104 ***
Hansen J-statistic			0.000

Note. Table 5 reports the regression results following the application of EBCT (Column 1) and IV/2SLS methods (Columns 2 and 3). Variable definitions and data sources are shown in Table A1. Robust standard errors are clustered by firm. Robust t-statistics are in parentheses. All continuous variables are winsorized at 1st and 99th percentiles.

* and *** indicate levels of significance at 10% and 1%, respectively.

Afterward, Model 1 is estimated with the weight generated from the EBCT procedure. The GII’s coefficients are negative and significant at 10% level, thus relieving our concern about reverse causality and lending support to our main finding on the inverse relationship between NIS and firm performance (see Table 5).

Market Development and the Forward-Looking Effect of National Innovation Systems

Another question arising from this finding: Is this impact uniform across the country? Given that each country has a different build of its NIS, assuming the uniformity of firm performance sensitivity to NIS development is illogical. For instance, a country that is struggling to transition to a knowledge-based economy might not have its innovation ecosystem as mature as the first-world countries. We dissect this potential discrepancy by segregating our sample into two country groups based on the MSCI market classification. The first group consists of the markets that are classified as developed markets, while the other groups comprise markets that are assigned to either emerging, frontier, or standalone categories. As can be seen from Panel A of Table 6, the negative effect of GII on ROA only exists in developing countries. We attribute this effect to the fact that firms in less developed markets are less attentive to innovation due to their lack of resources and inadequate regulatory frameworks supporting corporate innovation activities. This negligence also indicates an enormous potential for innovation-driven economic development once more inclusive NISs and generous innovation incentives are introduced. Figure 1 compares the sum of patent applications in developing and developed markets during the period of study, showing that developing markets dominate in this regard. Firms’ profitability, therefore, declines to a greater extent in developing markets. Conversely, with well-built NISs and innovation activities being an integral part of operations, firms in developed markets do not need to adjust their financing decisions as much, thus having milder performance reactions to NIS advancement.

OPEN IN VIEWER

Table 6.

Market Development and the Forward Effect of National Innovation.

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	Developed markets	Developing markets	Year t+1	Year t+2	Year t+3	Year t+4
PANEL A: ROA
GIIt−1	−0.001 (−0.349)	−0.001* (−1.745)	−0.001*** (−4.786)	−0.001** (−2.278)	−0.001 (−0.897)	0.001 (0.804)
Main effects	Y	Y	Y	Y	Y	Y
Control variables	Y	Y	Y	Y	Y	Y
Firm, year and country FEs	Y	Y	Y	Y	Y	Y
Observations	11,127	66,692	63,809	49,300	33,318	23,250
Adjusted R-squared	.710	.509	.601	.605	.616	.631
PANEL B: TOBINSQ
GIIt−1	−0.105** (−3.227)	−0.059** (−2.320)	−0.102*** (−4.232)	−0.068*** (−3.094)	−0.032 (−1.256)	0.051** (2.248)
Main effects	Y	Y	Y	Y	Y	Y
Control variables	Y	Y	Y	Y	Y	Y
Firm, year and country FEs	Y	Y	Y	Y	Y	Y
Observations	11,127	66,692	63,276	48,663	32,727	22,803
Adjusted R-squared	.636	.813	.807	.802	.802	.799
PANEL C. Mean-difference test between R&D expenditure between developed and developing markets
	Obs	Mean	Std. dev.	95% confidence interval
(1) Developing markets	32,429	0.090	0.471	0.085	0.095
(2) Developed markets	6,639	0.271	0.971	0.247	0.294
(2)–(1)		0.181***

Note. Table 6 reports the regression results of Model 1, with ROA (Panel A) and TOBINSQ (Panel B) as two firm performance proxies, with the subsample of developed markets (Column 1), the subsample of developing markets (Column 2), and PF replaced by their forward-looking measures from year t+1 to t+4 (Columns 3 to 6). Panel C reports the mean-difference test of firms’ R&D expenditure between developing markets and developed markets. Variable definitions and data sources are shown in Table A1. Robust standard errors are clustered by firm. Robust t-statistics are in parentheses. All continuous variables are winsorized at 1st and 99th percentiles.

*, **, and *** indicate levels of significance at 10%, 5%, and 1%, respectively.

The significance is 1%, which has been displayed next to the number as ***.

OPEN IN VIEWER

Figure 1.

Patent application comparison by years.

Differently, the adverse impact of GII on TOBINSQ is consistent regardless of the level of market development. If reduced performance is a logical reason for lowered Tobin’s Q in developing markets, the negative impact of GII on TOBINSQ in developed markets can be viewed from the cost perspective. As shown in the foot of Table 6, developed markets’ firms spend significantly more than their counterparts in developing markets on R&D, indicating that market participants might have their eyes on immediate costs that cannot be converted into profits immediately. Market devaluation might reflect investors’ concerns about firms’ weaker short-run profitability and return uncertainty since innovation requires persistent investments to build up and maintain competitive advantages (Gu, 2005). Explaining the heterogeneous effect of NISs on firms’ market-based performance, in developed markets with mature NISs, the rapid spread of knowledge and technologies erodes the advantages of early adopters of such advancement. If firms cannot leverage NIS upgrades to earn abnormal profits, market participants may expect lower profits in the future due to the business-stealing effect, leading to a decline in stock valuations (Bloom et al., 2013; Kogan et al., 2017). On the contrary, financial constraints and weaker absorptive capacity, alongside less effective NISs, discourage developing markets’ firms from deeply integrating innovation activities in corporate strategies. As a result, once deciding to innovate, those firms must incur substantial costs associated with initiation, implementation and adjustment frictions, which subsequently depress accounting performance. Meanwhile, valuations co-move more closely with near-term earnings in thinner, noisier information environments (Hail & Leuz, 2006). Consequently, the negative effect on Q in developed markets reflects anticipated rent erosion, whereas in developing markets it primarily tracks realized profitability pressures.

It is important to note that the aforementioned results only capture the short-term effect of NIS, while how long this impact persists has not been inspected. To investigate this, we replace FP with its forward measures in Model 1, then regress the model. The results presented in columns 3 to 6 in Table 6 unveil that the impact of GII on ROA remains in the subsequent 2 years (years t+1 and t+2) and then disappears. In terms of TOBINSQ, the impact even reverses in year t+4, hinting that although the bottom line has not been improved, the market starts to appreciate the innovation endeavors made by firms and willingly factor them into the firm valuation. These results demonstrate that NIS upgrades only slow down firms’ profitability in the short run while potentially casting a desirable effect in the longer horizon, which corroborates the long-term beneficial impact of national innovation on firm performance.

Channel Analysis

Innovation Intensity Channel

If building a comprehensive NIS can unlock the innovation potential of a country by providing a business environment conducive to innovation at multiple levels, the degree to which organizations and individuals commit to promoting innovation plays a vital role in boosting economic development. Israel is a salient example of a contemporary innovation powerhouse that is typically referred to as a “startup nation.” To realize its current success, Israel fuels its innovation through enormous R&D expenditure – considerably one of the largest innovation spenders worldwide (Lemarchand et al., 2016). Furthermore, Israel continuously promotes the entrepreneurial spirit, achieving the highest rate of startups per capita globally with high-tech industries driving 20% of Israel’s economic output (Scheer, 2024). These facts highlight the vital role of fostering innovation in boosting national economic development.

To examine whether the impact of GII transmits via the innovation intensity channel, we employ the natural logarithm of the number of patent applications per annum in each country as a proxy for innovation output. As reported in Table 7, we uncover a weaker firm performance in countries with more patent applications as the coefficients of GII_t−1 × PATENTAPPS_t−1 are significantly negative (Columns 1 in Panels A and B, Table 7). These results are in line with our discussion earlier that in response to the government’s efforts to foster NIS, firms are more inclined to engage in innovative activities by leveraging the supporting regulatory framework and infrastructure, resulting in more patent applications. Consequently, both ROA and Tobin’s Q drop since investing in innovation is costly but does not yield immediate profits or guarantee future cash inflows.

OPEN IN VIEWER

Table 7.

Explanatory Channels.

	Innovation intensity channel	Market competition channel
	(1)	(2)	(3)
VARIABLES	Country-level patent applications	HHI	Boone’s competition
PANEL A: ROA
GIIt−1	0.001 (1.028)	−0.002** (−2.426)	−0.001 (−1.424)
GIIt−1 × PATENTAPPSt−1	−0.000* (−1.779)
GIIt−1 × HHIt−1		−0.002 (−0.857)
GIIt−1 × BOONEt−1			−0.002 (−0.608)
Main effects	Y	Y	Y
Control variables	Y	Y	Y
Firm, year and country FEs	Y	Y	Y
Observations	77,371	55,153	77,819
Adjusted R-squared	.594	.604	.593
PANEL B: TOBINSQ
GIIt−1	0.091*** (2.872)	−0.080*** (−4.280)	−0.060** (−2.660)
GIIt−1 × PATENTAPPSt−1	−0.031*** (−3.868)
GIIt−1 × HHIt−1		0.009 (0.188)
GIIt−1 × BOONEt−1			−0.063 (−1.188)
Main effects	Y	Y	Y
Control variables	Y	Y	Y
Firm, year and country FEs	Y	Y	Y
Observations	77,371	55,153	77,819
Adjusted R-squared	.792	.773	.792

Note. Table 7 reports the cross-sectional tests with PATENTAPPS as the proxy for country-level innovation intensity, and HHI and BOONE as the proxy for market competition. ROA (Panel A) and TOBINSQ (Panel B) as two firm performance proxies. Variable definitions and data sources are shown in Table A1. Robust standard errors are clustered by firm. Robust t-statistics are in parentheses. All continuous variables are winsorized at 1st and 99th percentiles.

*, **, and *** indicate levels of significance at 10%, 5%, and 1%, respectively.

Market Competition

Firm performance might also decline due to an increase in competition within an industry. To capture within-industry competition, we employ two common proxies: the Herfindahl–Hirschman Index (HHI) and the Boone indicator (Boone, 2008; Fosu, 2013). Following Fosu (2013), we measure HHI by summing the squared market shares of each firm in a given industry in a year.

HH I j , t = ∑ i = 1 N , j ( Sale s i , j , t ∑ i = 1 Nj Sale s i , j , t ) 2

where HHI_j,t is the HHI of industry j and year t, and Sales_i,j,t is the revenue of firm i of industry j and year t.

Meanwhile, the Boone indicator is built upon the idea that in a highly competitive industry, the punishment for firms being inefficient is more profound (Boone, 2008; Boone et al., 2007). Hence, in such industry, a rise in marginal cost causes a greater reduction in variable profits. The Boone indicator is estimated by the following regression:

VRO A i , t = α 0 + α 1 lnM C i , j + ϑ i , t

in which VROA_i,t is the variable profits of firm i in year t; lnMC_i,j represents the marginal cost, computed by taking the natural logarithm of a firm’s marginal cost of firm i in industry j year t. In this equation, α 1 is the parameter capturing the level of competition in the industry. While α 1 is expected to carry a negative sign, Fosu (2013) suggests the use of the absolute value of α 1 as the higher value of | α 1 |, the more competitive the industry is. We denote the second proxy for market competition as BOONE = | α 1 |.

Using HHI and BOONE as proxies for industry competition, we find that both HII and BOONE do not play moderating roles in the relationship between GII and PF. Hence, market competition, though is supported by the literature as a potential channel, does not contribute to the explanation of the NIS – firm performance relationship.

Overall, although two feasible channels are hypothesized and empirically tested to explore the relationship between NIS and firm performance, only the innovation intensity channel shows its explanatory power. NIS and its components operate as economy-wide knowledge public assets; thus, although industry rivalry might accelerate the acquisition of new knowledge and technology, the impact of NIS on firms depends more on absorptive capacity and the ability to capture value through complementary assets (Castellacci & Natera, 2013; Cohen & Levinthal, 1990). Empirical evidence from our analysis documents that the primary NIS pathway to firm performance is through fostering innovation outputs – particularly patent applications, whereas the industry competitiveness channel is muted. The insignificant industry competitiveness channel indicates that innovation-driven competition may operate through more dynamic and less observable channels. NISs often stimulate technology-based rivalry/collaboration as well as knowledge spillovers, and are therefore more likely to shape firms’ competitive positions through capability accumulation than immediate price or market share adjustments. As a result, the competitive pressures induced by stronger NISs may not be adequately reflected in traditional industry competition measures, such as market structure and price pressure as specified in Boone (2008) and Fosu (2013). This suggests that innovation-led competition is likely to be multidimensional, involving both rivalry and cooperation, and may require alternative metrics that better capture knowledge-based and network-oriented competitive dynamics. Exploring such measures represents a promising direction for future research. We, therefore, cautiously interpret that the impact of NISs is channeled via innovation intensity while remaining neutral about the market competition channel.