«John R. Graham, Co-Supervisor David T. Robinson, Co-Supervisor Manuel Adelino Alon P. Brav Manju Puri Aaron K. Chatterji Dissertation submitted in ...»
16 each year. I use the patent’s year of application instead of the year it is granted because that better captures the actual timing of innovation. I use the logarithm of one plus this variable, that is, lnp1 ` N ewP atentq (denoted as lnpN ewP atentq), to ﬁx the skewness problem for better empirical properties. I measure the quality of innovation, based on the average lifetime citations of all new patents produced by a ﬁrm in each year. Similar to the logarithm transformation performed on quantity, I use lnp1 ` P at.Qualityq (denoted as lnpP at.Qualityq).
The second layer of innovation data is citations ﬁrms make in their own patents.
By tracking the citations a ﬁrm makes, we can measure the technological areas in which the ﬁrm works and the speciﬁc underlying technologies. Moreover, examining the citation network among ﬁrms (including both established ﬁrms and startups) allows us to construct variables capturing the technological relation between CVCs and startups and to measure dynamic information ﬂows between ﬁrm pairs.
The third layer of innovation data concerns the micro-level information beyond patents—the inventors (engineers, scientists, etc.) who contributed to a ﬁrm’s patents and their mobility. As shown in Gonzalez-Uribe (2013), Bernstein, Giroud, and Townsend (2014), and Brav et al. (2016), inventor-level information can help us infer the motivation behind corporate activities from the perspective of labor adjustment.
Second, I construct a full set of patent transactions from the Google Patent database, and this panel of patent life cycles allows me to examine how information acquisition improves the eﬃciency in the market for technologies.
17 4 CVC Initiations: The Eﬀect of Innovation Deterioration Why do ﬁrms initiate CVC programs? Under the information acquisition view of Corporate Venture Capital, capacity-constrained ﬁrms trade oﬀ between acquiring information for new ideas and producing existing ideas (Nelson, 1982; Jovanovic and Rob, 1989). The allocation of capacity to information acquisition is determined by the quantity and quality of existing ideas available to the ﬁrm—the fewer (lower) the quantity (quality) of existing innovation ideas become, the more likely it is the ﬁrm that implement information acquisition strategies, such as CVC, in search of better innovation paths.
Figure 4.1 visualizes CVC parent ﬁrms’ innovation dynamics before initiating their CVC divisions.
Innovation performance, measured by patenting quantity (Panel (a)) and quality (Panel (b)), is tracked for ﬁve years from t ´ 4 to t (t is the year of CVC initiation). Firm-year measures are adjusted by the averages of all peer ﬁrms in the same 3-digit SIC industry in the same year to exclude the inﬂuence of industry-speciﬁc time trends.
Panel (a) tracks innovation quantity of CVC parent ﬁrms, measured by the
Figure 4.1: This ﬁgure tracks corporate innovation performance of CVC parents before the initiation of their CVC units.
lnpN ewP atentq is the logarithm of the number of new patents applied by a ﬁrm in each year. lnpP at.Qualityq is the logarithm of average citations of new patents. Each measure is adjusted by the mean of ﬁrms in the same year and industry (3-digit SIC level). The graph starts from four years before a ﬁrm launches its CVC unit (t ´ 4) and ends in the year of launching (t). 95% conﬁdence intervals are plotted in dotted lines.
19 logarithm of the number of new patent applications. Four years before initiating their CVC units, CVC parents were signiﬁcantly more innovative than their peers and on average doubled their peers’ patent production. This advantage shrinks continuously by about 25% until year t. In Panel (b), CVC parent ﬁrms’ innovation enjoys 15% higher average citations compared to their industry peers in t ´ 4, and this number decreases to well below 0 at the time of CVC initiation. In untabulated results, I ﬁnd that the performance deterioration pattern is robust to measures of product market performance, that is, ROA and sales growth. Overall, Figure 4.1 presents a clear pattern at the start of the CVC life cycle—that is, CVC initiations typically follow deteriorations in parent ﬁrms’ internal innovation, which is consistent with the information acquisition view of CVC.
Building on Figure 4.1, I ﬁrst conﬁrm the relation between innovation deterioration and CVC initiation using a simple empirical setting. I then explore an identiﬁcation strategy that controls for several endogeneity concerns and sharpens the role of the information acquisition motive by analyzing several alternative explanations of the pattern. Firm-level CVC initiation decisions are aggregated to an industrylevel pattern, which presents how the information acquisition function ﬁts into the technological evolution in each industry.
4.1 Baseline Results To statistically identify the eﬀect of innovation performance on CVC initiations, I
estimate the following speciﬁcation using a panel data of ﬁrm-year observations:
where IpCV Cqi,t is equal to one if ﬁrm i launches a CVC unit in year t, and zero otherwise.1 ∆τ Innovationi,t´1 is the change of innovation over the past τ years 1 Dummy variable IpCV Cqi,t, instead of the size of CVC investment each year, is more appropriate to capture the corporate decision on CVC investment for two reasons: (1) the decision to start a 20 ending in t ´ 1. I use a three-year (τ “ 3) innovation shock throughout the main analysis and report robustness checks using other horizons in the Appendix. Firmlevel controls Xi,t´1 include ROA, size (logarithm of total assets), leverage, and R&D ratio (R&D expenditures scaled by total assets). Industry-by-year ﬁxed eﬀects are included to absorb industry-speciﬁc time trends in CVC activities and innovation. A negative β indicates that the probability of starting a CVC increases with innovation deterioration.
4.1.1 Summary Statistics
Table 4.1 presents descriptive statistics based on whether a CVC division is initiated in the ﬁrm-year.
Only observations with valid ROA, size, leverage, R&D ratio, and at least $10 million in book assets are kept in the sample. Only “innovative ﬁrms,” deﬁned as those that ﬁled at least one patent application that was eventually granted by the USPTO, are included. Industries (3-digit SIC level) with no CVC activities during the sample period are removed.
Table 4.1 provides a benchmark to position CVC parent ﬁrms in the Compustat universe of publicly traded corporations.
First, CVC parents are typically large ﬁrms.
On average, a CVC parent has $10.1 billion in book assets in 2007 USD (median is $2.4 billion) just before launching its CVC unit, whereas non-CVC parent ﬁrms have less than $3 billion in book assets (median is $0.2 billion). Second, CVC parent ﬁrms are innovation intensive in terms of patenting quantity, echoing the size eﬀect. Third, corporate governance variables are comparable between the two subsamples. Overall, the basic characteristics are consistent with existing stylized facts that CVC parent ﬁrms tend to be larger corporations with more business resources (Dushnitsky and CVC unit is at the executive level, whereas the size of investment in subsequent years is plausibly determined by the CVC team; and (2) the data on investment size in VentureXpert have potential sample selection issues such as CVCs strategically hiding good deals they invested in (to avoid competition from other CVCs). I report the analysis on annual CVC investment size as an important result in Section 6.
21 Lenox, 2005a; Basu, Phelps, and Kotha, 2011).
Consistent with Figure 4.1, CVC parent ﬁrms on average experience more negative innovation shocks before starting their CVC divisions. CVC parents on average experience a -7% (-10%) change in patenting quantity (quality) three years before launching their CVC units, compared to the control ﬁrms, which experience a 12% (8%) shock. Similar to the deterioration in innovation, CVC parents appear to underperform in terms of ROA and market-to-book ratio before CVC initiations.
22 Table 4.1: This table summarizes ﬁrm characteristics at the ﬁrm-year level from 1980 to 2006. CVC observations (IpCV Cqi,t “ 1) are those when ﬁrm i launched a CVC division in year t (and those ﬁrms are categorized as non-CVC observations in other years). The CVC sample is deﬁned in Table 3.1. Observations are required to have valid ROA, size (logarithm of total assets), leverage, R&D ratio (R&D expenditures scaled by total assets), and with at least $10 million in book assets, and variables are winsorized at the 1% and 99% levels to remove inﬂuential outliers. A ﬁrm is included in the panel sample only after it ﬁled a patent application that was eventually granted by the USPTO. Industries (3-digit SIC) that did not involve any CVC activities during the sample period are removed. For each variable, mean, median, and standard deviation are reported. Variable deﬁnitions are provided in the Appendix.
Table 4.2 presents the estimation results of model (4.
1). Columns (1) and (2) focus on the eﬀect of changes in innovation quantity. In column (1), the model is estimated using Ordinary Least Squares (OLS). The coeﬃcient of -0.007 is negative and signiﬁcant, meaning that a more severe decline in innovation quantity in the past three years is associated with a higher probability of initiating CVC investment. This estimate translates a two-standard-deviation decrease (2σ-change) in ∆ lnpN ewP atentq into a 51.54% increase from the unconditional probability of launching CVC unites. Column (2) reports the model estimation from a Logit regression, and I report the marginal eﬀect evaluated at sample mean. Column (2) delivers an almost identical message as column (1).
Columns (3) and (4) study the eﬀect of deterioration in innovation quality and use OLS and Logit, respectively. In column (3), the coeﬃcient of -0.004 means that a two-standard-deviation decrease in ∆ lnpP at.Qualityq increases the probability of CVC initiation by 67.09%, and this is economically comparable to that in column (1). Column (4) delivers a consistent message.
It is worth stressing the importance of incorporating industry-by-year ﬁxed eﬀects in the estimation. Previous studies on technological evolution and restructuring waves highlight the possibility that certain industry-speciﬁc technology shocks could be driving innovation changes and organizational activities at the same time (Mitchell and Mulherin, 1996; Harford, 2005; Rhodes-Kropf, Robinson, and Viswanathan, 2005). However, after absorbing this variation using industry-by-year ﬁxed eﬀects, the results in Table 4.2 are identiﬁed using the cross-sectional variation within an industry-by-year cell. This issue is revisited and studied in Section 4.3.
Overall, Table 4.2 conﬁrms the pattern in Figure 4.1 that CVC initiations typically follow a deterioration in innovation, lending support to the information acquisition 24 Table 4.2: This table documents the relation between innovation deterioration and the initiation of Corporate Venture Capital. The analysis is performed using the following speciﬁcation: IpCV Cqi,t “ αindustryˆt ` β ˆ ∆Innovationi,t´1 ` γ ˆ Xi,t´1 ` εi,t, The panel sample is described in Table 4.1. IpCV Cqi,t is equal to one if ﬁrm i launches a Corporate Venture Capital unit in year t, and zero otherwise. ∆Innovationi,t´1 is the innovation change over the past three years (i.e., the innovation change from t ´ 4 to t ´ 1). Innovation is measured using innovation quantity (the natural logarithm of the number of new patents in each ﬁrm-year plus one), shown in columns (1) and (2) and innovation quality (the natural logarithm of average citations per new patent in each ﬁrm-year plus one), shown in columns (3) and (4). Firm-level controls Xi,t´1 include ROA, size (logarithm of total assets), leverage, and R&D ratio (R&D expenditures scaled by total assets). The model is estimated using Ordinary Least Squares (OLS) and Logit, respectively. Industry-by-year dummies are included in the model to absorb industry-speciﬁc time trends in CVC activities and innovation. T-statistics are shown in parentheses and standard errors are clustered by ﬁrm. *, **, *** denote statistical signiﬁcance at the 10%, 5%, and 1% levels, respectively. Economic signiﬁcance is calculated by changing two standard deviations of the ∆Innovation and is reported below the estimation results.
25 view of CVC. However, what if the results are due to some endogenous common factor that drives both innovation dynamics and CVC activities (for example, poor management)? Moreover, what alternative economic forces, other than informational motives, could drive deteriorating ﬁrms to launch CVC? The analyses that follow adapt the framework in Table 4.2 to discuss these issues.
4.2 Identiﬁcation Strategy
Potential endogeneity problems arise from unobservables that are hard to control for in model (4.1). For instance, agency problems (such as empire-building managers) could hinder innovation and lead simultaneously to the initiation of CVC as a pet project, biasing the estimation in favor of ﬁnding a negative relation between innovation and CVC investment. On the other hand, CEOs who are more risk tolerant could improve corporate innovation (Sunder, Sunder, and Zhang, 2014) as well as encourage interactions with entrepreneurs using CVC, biasing the estimation against ﬁnding the result.
4.2.1 Instrumental Variable and Empirical Strategy
To address endogeneity concerns and rule out competing interpretations, I construct a new instrumental variable by exploiting the inﬂuence of exogenous technological evolution on ﬁrm-speciﬁc innovation. The idea that technological evolution aﬀects corporate innovation is intuitive—a ﬁrm specializing in manufacturing 14-inch hard disk drives (HHD) was less likely to produce valuable innovation when 8-inch HHD technology emerged, and this happened repeatedly along the development path of HHDs (5.25-inch, 3.5-inch, 2.5-inch, Solid State Drives). Indeed, “new technologies come and go, taking generations of companies with them” (Igami, 2014).