Heterogeneity of optimal balance between exploration and exploitation:the moderating roles of firm technological capability and industry alliance network position

ABSTRACT

Although existing ambidexterity literature suggests that firms need to find the optimal balance between exploration and exploitation for superior performance, few studies have empirically examined the heterogeneity of this balance according to firm-specific conditions. Building upon the capability and social network literature, we contend that firms’ technological capability and network position within industry alliances determine the optimal balance between exploration and exploitation. Analysing 7-year panel data in the worldwide semiconductor industry from 1994 to 2000, we find support for the following hypotheses: 1) the proportion of exploration has an inverted U-shaped relationship with innovation performance; 2) as firm technological capability increases, the optimal point between exploration and exploitation moves towards the exploration side; 3) as network centrality within industry alliances increases, the optimal point moves towards the exploitation side. The results offer theoretical insights into the ambidexterity literature as well as managerial implications for firms making resource allocation decisions.

KEYWORDS:

• Ambidexterity
• exploration and exploitation
• technological capability
• industry alliance network

Acknowledgments

This research has been supported by the Institute of Management Research, Seoul National University.

Disclosure statement

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

Notes

¹ Following Helfat and Peteraf (2003), we define technological capability of a firm as the firm’s ability to perform a coordinated set of R&D tasks in specific technological fields, utilising organisational resources, for the purpose of achieving a particular end result. According to this definition, in order for the performance of an activity to qualify as a capability, the capability must have been routinised to work in a reliable manner. That is, technological capability is primarily developed by repetition of and accumulated experience with specific R&D tasks.

² In this theoretical development section, we do not distinguish between exploration-related and exploitation-related capabilities, but focus on aggregated technological competence developed by repetitive activity and accumulated experience of specific R&D tasks in firms. However, it is entirely possible that some capabilities could be more related to exploitation than exploration. For instance, routines established for consistency, stability, and control are related to exploitation (Stettner and Lavie 2014). According to our theory and definition of technological capability, even the accumulation of exploitation-related capabilities shifts the optimal balance point towards the exploration side, largely because continued exploitation is likely to lead to core-rigidity problems, as outlined by Leonard–Barton (1992) and Christensen (1997). That is, firms should shift their focus to exploration to improve performance in innovation.

³ Isolated nodes are not considered in this calculation and are assigned a value of zero.

⁴ Haans, Pieters, and He (2016) recommended that both interaction terms XZ and X²Z (X is the independent variable and Z is the moderator) should be included in the model even if only one moderation type (a shift in the turning point or a flattening/steepening of the curve) is hypothesised. However, they also stated that it is appropriate to exclude the interaction term X²Z only if a shift in the turning point is hypothesised and the coefficient for X²Z for the full specification is not statistically different from zero. We ran the regression including both XZ and X²Z, and confirmed that the coefficients for Proportion of Exploration² × Technological Capability ($\beta$ = 1.306, p = 0.544) and Proportion of Exploration² × Network Position ($\beta$ =0.564, p =0.107) are both non-significant. Thus, these two terms are excluded from our model.

⁵ The VIF test is performed using the ordinary least squares regression model.

⁶ One is added to the dependent variable before the log transformation.