The innovator’s dilemma: the performance consequences of sequential or flexible exploration and exploitation patterns in turbulent environments

ABSTRACT We investigate the relationship between firm performance and patterns of exploration and exploitation dynamics under conditions of environmental turbulence. Adopting a contingency perspective, we develop arguments on how environmental turbulence moderates the effects of the simultaneous or sequential engagement in exploration and exploitation. Based on a longitudinal sample of 140 pharmaceutical firms, we find that the effectiveness of sequential or flexible simultaneous engagement in ambidexterity varies depending on the environmental contingency, namely level of technological change, that a company faces.


Introduction
The exploration-exploitation dynamics have been at the centre of the innovation research agenda since March (1991) published his seminal work. There is an emerging consensus in the literature that exploration and exploitation deliver complementary benefits since both are required to achieve and sustain competitive advantage. The exploration-exploitation research tradition maintains that the pursuit of both exploration and exploitation is driven by the need to achieve both short-and long-term benefits (Levinthal and March 1993). Although the pursuit of exploration and exploitation has recently received much research attention from different perspectives (e.g. Gibson and Birkinshaw 2004;He and Wong 2004;Stettner and Lavie 2014;Venkatraman, Lee, and Iyer 2007), we still know relatively little about the performance consequence of pursuing both exploration and exploitation under different environmental conditions. The issue is critical because what is beneficial in a stable environment may be detrimental in a more dynamic and turbulent context. The notion that the degree of turbulence faced by an organisation affects organisational outcomes is well established in the literature (Duncan 1972;Beckman, Haunschild, and Phillips 2004). More specifically, previous studies have put forth that the instability and dynamism of an environment represent key moderators that affect the impact of both exploration and exploitation on firm performance (Jansen, Van Den Bosh, and Volberda 2006;Kim and Rhee 2009;Jansen, Vera, and Crossan 2009). The above has been mainly focused on uncovering, under specific environmental conditions, the performance merits of either exploration or exploitation per se. We advance this research stream by explaining how the engagement in both exploration and exploitation in a turbulent environment influences performance. To address this issue, we use insights from the dynamic capabilities view (Teece, Pisano, and Shuen 1997) and the contingency perspective (Venkatraman and Prescott 1990;Volberda et al. 2012) to develop predictions about the contingencies of the relation between simultaneous and sequential engagement in exploration/exploitation and firm performance. In addressing this issue, we answer the call of scholars to bring the contingency perspective to exploration-exploitation research (Lavie, Stettner, and Tushman 2010). It is worth noting that the exploration-exploitation link to performance has been studied from a variety of perspectives, as pointed out by Lavie et al. (2010). In this paper, we focus on the organisational level and knowledge search of organisations operating in a technologically dynamic environment.
Our analyses have been conducted on a panel including 1495 firm-year observations for 140 firms. Our longitudinal analysis indicates that companies vary in the way they pursue exploration and exploitation. In particular, we find that firms differ by choosing to engage simultaneously in exploration and exploitation but changing the proportions of the two or to engage in exploration and exploitation sequentially. The results suggest that these choices impact firm performance differently depending on the level of turbulence in the environment. The present paper contributes to fine-tuning our understanding of how exploration/exploitation dynamics contribute to performance under conditions of technological change. Overall, the present research offers new insights to the academic debate on the simultaneous versus sequential pursuit of exploration/exploitation and to managerial practice.

Conceptual background
The present research starts from the consideration that the organisational learning orientation and performance are linked and influenced by the environmental context, more, in particular, the degree of turbulence (e.g. Hanvanich, Sivakumar, and Hult 2006). Expanding this line of reasoning, we argue that the performance consequences of pursuing exploration and exploitation can be better understood by considering specific environmental contingencies. Since March (1991) first proposed his theory, the role of the external environment has been identified as a key element in the pursuit of exploration and exploitation. Exploration aims to generate new knowledge; exploitation aims to find new applications for existing knowledge (Levinthal and March 1993). The returns of exploration are long term and uncertain, whereas those from exploitation are more short term, securing the resource base for investment in novel technologies (March 1991). The term ambidexterity is used to indicate the pursuit of both exploration and exploitation (Katila and Ahuja 2002;He and Wong 2004;Gupta, Smith, and Shalley 2006). The pursuit of ambidexterity has a positive impact on economic and innovative performance (Nosella 2014). The empirical evidence seems to confirm a positive association with business unit performance (Gibson and Birkinshaw 2004;Tushman 2004, 2013), sales growth rates (He and Wong 2004), survival of corporate venture units (Hill and Birkinshaw 2014), and new product introductions (Wang and Rafiq 2014).
When considering the pursuit of exploration and exploitation under conditions of environmental turbulence, we elaborate on how the engagement in exploration/exploitation is deployed. In particular, we argue that the firms may either engage in exploration or exploitation sequentially or might engage in both simultaneously but shifting the pronunciation of the two over time. The latter we refer to as flexibility in exploration/exploitation. Focusing on how exploration/exploitation is pursued is critical to understand its performance implications and the contingencies influencing its impact. Organisations need 'sufficient exploitation' and 'enough' exploration (Levinthal and March 1993, 105). A strict interpretation of ambidexterity suggests that there should be equal proportions of exploration and exploitation (e.g. He and Wong 2004). Levinthal and March (1993) point out, however, that the optimal mix is hard to find and it is not universal. The optimal proportion between exploration and exploitation remains an open question (Lavie, Kang, and Rosenkopf 2011).

Hypotheses
Adopting a contingency perspective, we develop arguments on how environmental turbulence moderates the effects of the simultaneous or sequential engagement in exploration and exploitation on company performance. We refer to environmental turbulence in a technology/knowledge interpretation. In this sense, the environment turbulence is characterised by a high pace of technological change, which makes a company's existing knowledge grow quickly obsolete, and by a high degree of uncertainty about future returns from investments on innovation (Bowman and Hurry 1993). Jansen, Van Den Bosh, and Volberda (2006) suggest that explorative innovation leads to higher performance under conditions of high environmental turbulence. The underlying logic is that, by expanding their knowledge base, companies may be better equipped and more prepared to face changing technological contexts. This seemingly generally accepted claim has been challenged by Posen and Levinthal (2012), who argue that in dynamic environments, the uncertainty associated with turbulence affects the payoff of not only existing, but also future new knowledge. In this sense, the returns of generating new knowledge through exploration may decay rapidly in situations in which external technological conditions continuously fluctuate. Therefore, managers are reluctant to pursue exploration sustaining 'the costs of change to capabilities that may become worthless if the environment reverts to its previous state' (Kogut and Kulatilaka 2001, 749). Pursuing exploration could be detrimental for performance in a turbulent environment as it risks to entail costs and fail to deliver a benefit in case the technological evolution moves in a different direction. While an increased level of exploration might be suggested by dominant logic, high levels of exploration come at a high cost, rewards are uncertain and hard to achieve. Therefore, under conditions of environmental turbulence, increasing levels of exploitative knowledge may be beneficial for performance (Posen and Levinthal 2012). Thus, in order to sustain performance under conditions of environmental turbulence, the firm may have to favour exploitation over exploration (or viceversa) and as a consequence, change the pronunciation of one of the two. Birkinshaw and Gupta (2013) mention that organisations pursuing both exploration and exploitation face a form of efficiency frontier challenge, having to decide where to sit on the frontier, how to reach it, and how to push it out. The optimal point may depend on specific circumstances. Consequently, we argue that changing circumstances may require the need to change that point in order to support performance. The literature has found that in uncertain environments, simultaneous exploration and exploitation is conducive to higher levels of innovation and survival rates and better financial results (O'Reilly and Tushman 2013). Taking this further, Burgelman and Grove (2007, 978) put forth the need for an adaptive system in which balance is achieved by varying the weights of exploration and exploitation over time 'but with none of the weights ever becoming zero'. The above suggests the need for having not only both exploration and exploitation present, but also changing their relative levels over time, favouring either exploration or exploitation.
The need to adjust the relative weights of exploration and exploitation may be particularly stringent when the firm is confronted with environmental turbulence which tends to make obsolete existing knowledge. Therefore, we hypothesise: Hypothesis 1. Environmental turbulence positively moderates the relationship between flexible ambidexterity and firm performance.
Pursuing both exploration and exploitation at the organisational level poses several challenges due to the fact that the two activities compete for scarce resources, entail conflicting practices and are based on different logics (Gupta, Smith, and Shalley 2006;March 1991;Sabidussi, Lokshin, and Duysters 2018). In order to pursue exploration and exploitation, organisations need to have in place a set of systems and processes to sustain both activities (Gibson and Birkinshaw 2004). Developing these coordination mechanisms is a complex process that is nevertheless necessary (Atuahene-Gima 2005). For this reason, firms may consider alternating phases of exploration and exploitation. Switching between exploration or exploitation implies, however, that resources devoted to exploitation need to be reconverted to exploration (or vice-versa), requiring a reorientation of the organisational activities. Companies are, therefore, required to overcome inertia, switching routines, practices, procedures, reward systems, and resource allocation guidelines (Simsek et al. 2009), which is costly and time consuming. This reorientation is especially critical when companies alternate sequentially between exploration and exploitation, switching completely from one activity to the other (Sabidussi et al. 2014). Swift (2016) confirms that transitioning from one activity to the other is perilous and keen to generate failures. In highly turbulent environments, the changes needed to reorient the organisation towards either exploration or exploitation may be made quickly redundant by other sudden technological changes. This, in turn, may be detrimental to performance. Therefore, we argue: Hypothesis 2. Environmental turbulence negatively moderates the relationship between sequential ambidexterity and firm performance.
Our model is illustrated in Figure 1.

Data and sample
We test our hypotheses on longitudinal data spanning the years 1990-2008 on publicly listed firms operating in the pharmaceutical & biotech industry. This industry is suitable for our study for several reasons. First, it is a dynamic high-tech sector where fundamental scientific advancements are applied to medical products introduced to the market. Second, it is a sector driven by innovation. Between 1990 and 2007, pharmaceutical R&D expenditures grew in the USA 5.2 times and in Europe 3.3 times (EFPIA 2008). Third, in line with previous research (e.g. Dunlap-Hinkler, Kotabe, and Mudambi 2010), we note that the pharmaceutical industry is a sector with a high propensity to engage in both explorative and exploitative innovation, which is the focus of our study.
In order to construct our sample, we identified companies included in the European Commission Industrial R&D Scoreboard report 2007. The R&D Scoreboard presents per sector firms that are global leaders in terms of R&D investment (European Commission 2007). In particular, we have relied on the list of the firms reported for the pharma & biotech sector. The focus on leading firms in this industry to test our hypotheses is appropriate for several reasons. First, the threshold for being actively involved in both explorative and exploitative innovation is considerably high, especially when compared to other less research-intense sectors. The investment required for a new drug development is  (IFPMA 2017). Second, R&D spending in this industry is highly correlated with other indicators of size such as sales or the number of employees. Additionally, the pharma industry is specifically known as a very mature and consolidated sector, with only the largest 10 companies representing approximately 40% of the industry and the leading 15 companies representing 50% of the industry. The consolidated structure of the industry makes the use of R&D Scoreboard an appropriate source to identify the industry firms.
For each sample firm, we collected firm-level information from several sources. Our primary source of financial data on the firms was Compustat, the North America as well as Global subsections. Since Compustat's coverage for European firms is less than complete, we complemented these data with information retrieved from Worldscope and the firms' annual reports. For the R&D data, we also drew on information given in the R&D Scoreboard. We used exchange rate information from the IMF's International Financial Statistics to express figures that were in domestic currencies in dollar terms. Information on patents was derived from the PATSTAT database. Since companies can also file patent applications under the names of their subsidiaries, we searched for patents applied for under the name of the parent firm or its subsidiaries for each firm and each year in our sample. We considered the patents of an acquired firm to be part of the patent stock of the acquiring company from the acquisition year onward. The data construction exercise provided us with an unbalanced panel dataset of firm performance for the years 1990-2008. We limited our data span to the 1990-2008 period to avoid any potential bias caused by the 2008 economic crisis. For the pharmaceutical industry, the '2008 Recession' resulted into a considerable drop of R&D expenditures, many R&D projects cut/put on hold, major restructuring plans and considerable layoffs (Mintz 2009;Zhang 2009;Cohen 2019). In addition to the direct effects of the crisis, the pharma industry suffered from the health policy changes implemented by countries (Leopold et al. 2014). These effects would have created a massive distortion in the relations that are at the core of our study. Due to missing values on some of the variables, not all firms presented in the Scoreboard could be used. In total, the panel included 1495 firm-year observations for 140 firms. The sample firms cover 93% of R&D expenditures and 87% of sales of all firms included in the R&D Scoreboard 2007 in the pharma & biotech sector.

Dependent variable
Following Chung and Pruitt (1994) we approximate Tobin's Q as a firm market to book value and taking the logarithm. It has 'the advantage of capturing short-term performance and long-term prospects' in a single measure (Uotila et al. 2009, 223). It is, therefore, suitable as a performance indicator for exploration and exploitation since they affect performance differently in the short and long term (March 1991).
Focal independent variables. Firm ambidexterity in exploration and exploitation has been investigated in a variety of contexts, including strategic alliances (e.g. Lavie and Rosenkopf 2006), organisational learning (e.g. Levinthal and March 1993), and product innovation (He and Wong 2004). We follow Lavie, Stettner, and Tushman (2010) in defining exploration as a deviation from the firm's current knowledge and exploitation as an expansion of its existing knowledge base. For the purposes of this study, a company's knowledge is approximated by patents (Ramani, El-Aroui, and Carrere 2008;Rosenkopf and Nerkar 2001). Yearly patent counts were used to distinguish between exploration and exploitation as follows: for each company in our sample, we created a technology profile based on patent activity in each patent class during its entire observable history and in five-year windows prior to a given year. The patent classes were determined at the two-digit level, which resulted in approximately 120 classes. We defined explorative patents to be applications in technological classes in which the firm had no patenting activity in the previous five years (e.g. Nooteboom et al. 2007). We defined exploitative patents to be all other applications in technological classes. In order to account for the fact that the same patent can be classified under a number of classes and subclasses (depending on that invention's components), a fractional count was used. Moreover, one invention can have multiple patents. All of these belong to a family that has a priority patent and a priority year. We used the very first patent of each family, but we took all the (unique) IPC classes mentioned in the whole family. We did deduplicate them afterwards to prevent a double count. Extant research suggests that technological knowledge depreciates sharply over time. We used a five-year moving window to calculate our variables in line with other studies based on patent data that have also used five-year moving windows (e.g. Katila and Ahuja 2002;Argote 1999).
The computation of the focal variables proceeds in the following steps. First, for each firm in our sample, we calculated standard deviation S i of exploration for the entire observation period, , where X i is the number of firm's i explorative patents per year and X is the mean value. We used the same approach to calculate the standard deviation of exploitation. These serve as our firm-specific benchmark. In the second step, we calculate the standard deviations of exploration and exploitation for a moving window of five years but eliminating the latest year of patents from the average. Elimination of patents for the most recent year may help isolate the effect of firms pursuing many different technologies due to uncertainty as opposed to a uniform move by majority of firms from one type of technology to another (e.g. analogue to digital) due to a common exogenous shock. The variable simultaneous flexible exploration exploitation takes the value 1 (else 0) if the firm engaged in exploration and exploitation in the previous five years, and the standard deviation for exploration or exploitation in the five-year moving window is larger than the corresponding value for the full observation period. In this way, we compare the adjustments in exploration or exploitation in the preceding five years to the normal adjustments (i.e. standard deviations for the entire history).
Sequential exploration and exploitation captures cycles of exploration and exploitation and takes the value 1 (else 0) if the company engaged only in exploration (exploitation) in the five-year moving window but engaged in exploitation (exploration) prior to that.

Moderating variable: environmental turbulence
We use patent classes that are relevant for pharmaceutical companies in our sample to construct a measure of technological change. Following Van de Vrande, Vanhaverbeke, and Duysters (2009), we selected the top 80% of the patent classes, based on the applications of the focal firms during the observation period, to determine the relevant technological fields. We used a threshold of 80% to reduce the noise in calculating this variable. Next, we calculated the number of applications worldwide for these patent classes for each year. We used all applications rather than only the patents of the focal firms so that this variable would be independent of the sample. To determine the similarities of the patent distributions for two consecutive years, we calculated the Pearson correlation coefficient, ρ. We then calculated technological turbulence as 1-ρ, so that higher values indicate higher levels of turbulence. We lagged this variable by one year.

Control variables
We control for simultaneous stable exploration exploitation. To construct this dummy, we followed steps one and two when constructing simultaneous flexible exploration exploitation, but in the third step, we apply the condition that the standard deviation in the five-year moving window is not larger than the overall standard deviation for both exploration and exploitation for the entire observation period.
Our analysis controlled for a number of firm-specific characteristics. Firm-level time-variant controls included the R&D stock/Total assets, where the consolidated R&D stock is calculated for each firm as a perpetual inventory of past R&D expenditures with a constant depreciation rate of 15% (Hall 1993), to account for intangible inputs, and the lagged patent-to-R&D ratio, the Patents/R&D variable, to capture the differences in the firms' prior patent activity. We controlled for firm age (Log Age). In all models, we include Firm size, calculated as the logarithm of the number of employees. We further controlled for firm profitability and debt (Mudambi and Swift 2014). Finally, the analysis controlled for general trends in patenting behaviour and economic fluctuations by including year dummies and for firm location by including country dummies.

Statistical method
The empirical model relates firm performance, as measured by Tobin's Q, to exploration and exploitation. Assuming that firm's assets are additive, we estimate the following linearised hedonic regression 1 : where V it is the firm's market value, A it is the book value of tangible assets, R&D it is firm's R&D stock, P it is a measure of firm's patent stock, EE_SIM_S it is simultaneous stable exploration exploitation, EE_SIM_F it is simultaneous flexible exploration exploitation and EE_SEQ it is sequential exploration exploitation. The remaining time-varying (control) variables such as firm size, environmental turbulence, firm age, profitability and leverage are collected in Z it . Time-invariant variables industry and country dummies are collected in vector W i . Coefficient b 1 reflects the relative shadow value R&D to tangible assets, b 2 measures the contribution to market value of additional patent per unit of R&D stock, b 3 − b 5 , the coefficients of interest, measure the contribution of ambidexterity variables to market value. To allow for unobserved firm-level heterogeneity in Tobin's Q across firms and for common macro-economic shocks, the error term 1 it in Equation (1) includes a firm-specific effect m i , a year-specific intercept l t , in addition to a measurement error u it : 1 it = l t + m i + u it for i = 1, . . . , N; t = 1, . . . , T i . In the regressions we account for unobserved heterogeneity and inter-cluster correlations by reporting for all models the two-way cluster-robust covariance estimates (Cameron, Gelbach, and Miller 2008). A core part of unobserved firm heterogeneity will be relatively fixed because a number of traits such as general managerial capabilities will change little over time. In the empirical models, we control for such influences by including the pre-sample mean of the dependent variable in a pseudo-fixed effects methods model due to Blundell, Griffith, and Van Reenen (1999). 2 Table 1 presents the descriptive statistics and bivariate correlations between the variables used in the estimation. The correlations are low to moderate in most cases, and we ascertained that the variance inflation factor scores are low (the average VIF is 3.24). An exception is the high positive correlation between firm size and firm age. The high negative correlations between the explorationexploitation strategy variables are by design.

Results
The firms in the sample are predominantly large and R&D intensive with total assets, R&D stocks and patent stocks equal, respectively $US 4,962 million, $US 1,781 million and 250 patents, on average. The Tobin's q value has a mean value of 3.43, above unity. Annual patent rates are high, with an average of 20 explorative and 30 exploitative patent applications per year.
The regression results are reported in Table 2. All models are estimated with robust two-way clustered errors and control for unobserved firm-level heterogeneity with a mean-scaling estimator (Blundell, Griffith, and Van Reenen 1999). Model 1 includes control variables only and serves as a reference point: R&D stock as a share of total assets, firm size, firm age, and patent intensity have the expected positive signs and are, with the exception of the latter two, statistically significant. Notes: All independent variables are lagged by one year.

TECHNOLOGY ANALYSIS & STRATEGIC MANAGEMENT
These coefficients remain consistent across the subsequent, more complete, models and are in line with prior evidence on the contributions of intangible assets to firm market value (e.g. Cockburn and Griliches 1988). In Model 2, when the focal variables are added to the model, we find that their coefficients are positive and significant. In Models 3-5, we include the interactions of the focal variables with environmental turbulence. The coefficient associated with Hypothesis 1 (Model 3) is significant and positive, thus confirming our prediction. The coefficient associated with Hypothesis 2 (Model 4) is significant and negative in line with our expectations. The complete model (Model 5) is consistent with the above-mentioned results.

Supplementary analysis
Given our main findings, the question remains of what drives the relative benefit of flexibility in explorationexploitation when the environment is turbulent. Faced with uncertainty, a firm All models are estimated on 140 firms and 1495 observations. The regressions contain 23 year dummies, 10 country dummies, and 1 patent dummy if zero. Robust (two-way clustered) standard errors in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001. could react by relatively larger adjustment in exploration or in exploitation or adjust both. In an attempt to understand this effect more, we carried out additional analysis presented in Model 6. In this model, we decomposed the flexible explorationexploitation into two dichotomous variables: adjustments driven by exploration and adjustments driven by exploitation. This additional analysis reveals that in times of environmental turbulence carrying out adjustments with respect to exploration, rather than exploitation, is especially beneficial for financial performance.
To test the robustness of our findings, we first considered alternative specifications of our model. Our results were robust to the use of a shorter, three-year window. We tested for endogeneity of the balance variables (one at a time) using a Sargan test, with firm age and its square as excluded instruments. We could not reject the null of all three balance variables being exogenous (Sargan statistic of 0.56, 0.61, and 0.98, respectively) at the conventional levels of significance. We also tested whether the strict exogeneity assumption of fixed-effects models held, using the test suggested by Wooldridge (2002, 285) by estimating models with one and two forward lags of the innovation variables. The null hypothesis that the forward lags of all the balancing modes are jointly zero could not be rejected at 5%, indicating that the exogeneity assumption cannot be rejected.

Discussion and conclusions
Companies face serious challenges posed by the need to both explore and exploit. In the face of this challenge, innovating firms vary in the way they pursue exploration and exploitation. The current study examined their merits for firm performance in conditions of technological change. We find that in times of increased turbulence especially flexible simultaneous explorationexploitation are rewarded.
Our study contributes to the ambidexterity literature by demonstrating that in the context of exploration and exploitation dynamics, environmental conditions matter for firm performance. Our study contributes also to the literature on environmental contingencies in the context of exploration and exploitation (e.g. Jansen, Van Den Bosh, and Volberda 2006). In particular, we highlight the role of technological change as a relevant form of external turbulence. In doing so, we answer the call of scholars to investigate the role of specific contextual factors in exploration -exploitation research (Lavie, Stettner, and Tushman 2010).
Our research shows that, in line with the dynamic capabilities approach to external technological change (Teece, Pisano, and Shuen 1997), adapting to external conjunctures represents a source of competitive advantage for companies. Our additional analysis reveals that a firm, which carries out adjustments with respect to exploration rather than exploitation benefits most under conditions of technological turbulence. This result adds support to the discussion on the signalling value of explorative activities (Gulati 1998;Powell, Koput, and Smith-Doerr 1996;Stuart 2000). Such signalling properties of exploration may be of particular value to investors as uncertainty increases (Gulati 1998).
Overall, our findings align with O'Reilly and Tushman (2008) argument that the dynamic capabilities view is an appropriate lens to study the pursuit of exploration and exploitation. This provides an important direction for future research.
Our findings have also policy implications. Monitoring the levels of technological turbulence may be relevant in order to fine-tune policy interventions that aim at stimulating innovative behaviours without hampering performance. Furthermore, government innovation policies and innovation subsidies under conditions of high technological turbulence should be geared towards both exploration and exploitation. This is in line with our findings for companies. Companies should therefore be enabled to engage in exploration and exploitation simultaneously and not be forced to focus on sequential ambidexterity due to financial constraints.
This study is not without limitations. First, we have considered specific dynamics in the pursuit of exploration and exploitation, flexible and sequential. Future research may further fine-tune these concepts. For instance, attention can be dedicated to differentiate whether the stable approach derives from not altering the levels of exploration and exploitation or altering them in equal proportion. Second, we also acknowledge the lack of subsidiary-level information since our investigation was limited to aggregated data. Additionally, an investigation into additional contingencies is needed to better understand patterns of exploration and exploitation. Future research could overcome the above limitations.
In conclusion, despite its limitations, we believe this paper paves the way for fruitful future research.

Notes
1. The analysis on all models was performed following the market-value approach (e.g. Griliches 1981;Lubatkin and Shrieves 1986), which draws on the hedonic price model and assumes that a firm is a bundle of assets and capabilities that are difficult to price separately on the market. Under the assumption that stock markets are efficient, the value that financial markets assign to these assets is equal to the discounted value of their corresponding future cash flows. Hence, it is possible to infer the value of bundles of intangible assets from their impact on the firm's market value. 2. Endogeneity of the exploration and exploitation decision over time may be of some concern, even when using lagged values of IV in the models. In a robustness checks we conducted endogeneity tests through the estimation of an IV version of the model. These results are qualitatively similar to the ones reported.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Notes on contributors
Anna Sabidussi is visiting associate professor at Tilburg University (Tilburg School of Economics and Management) and professor of international business at AVANS University of Applied Sciences.
Boris Lokshin is an associate professor at the Department of Organization and Strategy, Maastricht University. He holds PhD in economics from the University of Maastricht, The Netherlands and an MA in economics from Indiana University, USA.
Geert Duysters is a full professor of entrepreneurship and innovation at Tilburg University. He currently acts as Dean at the Tilburg School of Economics and Management (TISEM).