Kati La Acquisitions Academic Research

Strategic Management JournalStrat. Mgmt. J.,22: 197220 (2001)

TECHNOLOGICAL ACQUISITIONS AND THEINNOVATION PERFORMANCE OF ACQUIRINGFIRMS: A LONGITUDINAL STUDYGAUTAM AHUJA1* and RIITTA KATILA21College of Business Administration, University of Texas at Austin, Austin,Texas, U.S.A.2Robert H. Smith School of Business, University of Maryland, College Park, Mary-land, U.S.A.

This paper examines the impact of acquisitions on the subsequent innovation performance ofacquiring firms in the chemicals industry. We distinguish between technological acquisitions,acquisitions in which technology is a component of the acquired firms assets, and nontechnologi-cal acquisitions: acquisitions that do not involve a technological component. We develop aframework relating acquisitions to firm innovation performance and develop a set of measuresfor quantifying the technological inputs a firm obtains through acquisitions. We find that withintechnological acquisitions absolute size of the acquired knowledge base enhances innovationperformance, while relative size of the acquired knowledge base reduces innovation output.The relatedness of acquired and acquiring knowledge bases has a nonlinear impact oninnovation output. Nontechnological acquisitions do not have a significant effect on subsequentinnovation output.Copyright 2001 John Wiley & Sons, Ltd.

In this paper we examine the impact of acqui-sitions on the subsequent innovation performanceof acquiring firms. Studying the impact of acqui-sitions on postacquisition innovation performanceis important from at least three perspectives. First,this evaluation is important from the perspectiveof organizational learning and innovation, andhelps us understand how organizations absorb anduse external knowledge. Firm-level theories oftechnical change suggest that a firms inno-vativeness is an outcome of increases in itsknowledge base (Griliches, 1984, 1990; Pakesand Griliches, 1984; Henderson and Cockburn,1996). While a firms knowledge base can growthrough a series of knowledge-enhancing invest-ments by the company over time, firms can alsogrow their knowledge through acquiring or graft-ing of external knowledge bases (Cohen andLevinthal, 1989; Huber, 1991). Interestingly,

Key words: innovation; acquisitions; knowledge*Correspondence to: Gautam Ahuja, College of BusinessAdministration, University of Texas at Austin, Austin, TX78712-1174, U.S.A.

Copyright 2001 John Wiley & Sons, Ltd. Received 27 August 1998Final revision received 12 October 2000

while the relationship between firms investmentsin knowledge and their innovation output hasbeen studied extensively (Hall, Griliches andHausman, 1986; Griliches, 1990), relatively littleresearch has focused on the role of acquisitionsin growing the firms knowledge base (Granstrandand Sjolander, 1990; Huber, 1991; Gerpott, 1995).This lacuna is all the more surprising given find-ings which indicate that obtaining technologicalknow-how and developing technical capabilitiesare increasingly important motives for acqui-sitions (Link, 1988; Granstrandet al., 1992;Chakrabarti, Hauschildt, and Suverkrup, 1994;Wysocki, 1997a, 1997b).

Scholars studying the market for corporate con-trol have also examined the relationship betweenacquisitions and innovation output (Hittet al.,1991, 1996; Hoskisson, Hitt, and Ireland, 1994).However, in contrast to the findings of the inno-vation literature, studies in the corporate controltradition have generally found that acquisitionshave a negative impact on the postacquisitioninnovation output of acquiring firms. Agencyproblems, reduction in managerial commitment to

198 G. Ahuja and R. Katila

innovation, and the absorption of managerialenergy in the acquisition integration process atthe expense of routine management have beenposited as possible explanations for these results(Hitt et al., 1991, 1996). This conclusion, how-ever, is puzzling since acquisitions continue tobe a popular strategy for corporate growth. Inrecent years more dollars have been invested inacquisition activity than in any other equivalenttime period in history (Curry, 1997). Evaluatingthe postacquisition innovation output of acquiringfirms provides one indicator, albeit an indirectone, of the returns to corporate investments inacquisition activity.

A third reason for studying the impact of acqui-sitions on the postacquisition performance ofacquiring firms comes from the growing literatureon the resource-based view of the firm. Accordingto this perspective acquisitions are an importantpart of the business process of redeployingresources into more productive uses (Anand andSingh, 1997; Capron, Dussauge, and Mitchell,1998; Capron, Mitchell, and Swaminathan, 1998).Through acquisitions firm-specific assets housedwithin one organization are merged with assetsin another organization to improve the productiv-ity of the combined assets (Haspeslagh and Jemi-son, 1991; Anand and Singh, 1997). Evaluatingthe postacquisition performance of firms providesevidence on the efficiency of this asset-matchingand combining process.

In this paper we draw upon theories of techno-logical innovation, learning, and the resource-based view to develop a theoretical model andpredictions relating acquisition characteristics tothe innovation performance of acquiring firms(Cohen and Levinthal, 1989; Grant, 1996). Inno-vation performance can be measured in terms ofinnovative inputs such as R&D expenditures, orinnovation outputs such as patenting frequency(Griliches, 1984; Henderson and Cockburn,1996). Acquisitions can affect both innovativeinputs and innovative outputs (Hittet al., 1991).For instance, a firms R&D expenditures candecrease after it conducts an acquisition as thefirm eliminates certain streams of research or asmanagers become more risk averse (Hittet al.,1991). Yet, even while research efforts decrease,the productivity of those efforts can increase asthe two hitherto separate research teams combinetheir skills and knowledge. In this research wefocus on the impact of acquisitions on innovation

Copyright 2001 John Wiley & Sons, Ltd. Strat. Mgmt. J.,22: 197220 (2001)

outputs as measured by the patenting frequencyof the acquiring firm. Accordingly, we adopt aninnovation production function framework andmodel patenting frequency as the output of aproduction function (Griliches, 1984). Thehypotheses of this study are statements aboutthe relationship between this patented knowledgeoutput and the firms stocks of owned andacquired knowledge. In estimating the impact ofacquisitions on the innovative output of a firmwe statistically control for the levels of innovativeinputs such as R&D expenditures, but leave theirsubstantive examination to future work.

HYPOTHESES

In this research we focus on two contingenciesthat may be critical to clarifying the relationshipbetween acquisitions and the postacquisition inno-vation performance of acquiring firms. First, wedraw attention to the fact that technologicalreasons do not motivate all acquisitions. Forexample, acquisitions can be motivated by thedesire to obtain access to distribution channels,to gain entry into new markets, or to obtainfinancial synergies or market power (Lubatkin,1983; Balakrishnan, 1988; Chatterjee, 1991; Has-peslagh and Jemison, 1991; Capron, Dussauge,and Mitchell, 1998). Such acquisitions may pro-vide no technological inputs to the acquiring firmand therefore cannot be expected to improve itsinnovation output. Second, we note that evenamong technologically motivated acquisitions theimpact of acquisitions may depend on the charac-teristics of the relationship between the knowl-edge of the acquired and acquiring firms(Lubatkin, 1983; Singh and Montgomery, 1987;Lane and Lubatkin, 1998).

In summary, we argue that the impact of acqui-sitions on the acquiring firms innovation outputcan be understood in the context of the techno-logical inputs provided by the acquisition. Weargue that acquisitions that provide no technologi-cal inputs cannot be expected to have a positiveimpact on firm innovation output (Hypothesis1). Second, within acquisitions that do providetechnological inputs, we predict that the impactof an acquisition on the postacquisition innovationoutput of the acquiring firms is likely to varypositively with the absolute size of the acquiredfirms knowledge base, negatively with the rela-

Technological Acquisitions and Innovation 199

tive size of the acquired and acquiring knowledgebases, and curvilinearly with the relatedness ofthe acquired and acquiring knowledge bases(Hypotheses 2, 3 and 4).

Technological vs. nontechnologicalacquisitions

Acquisitions can affect the acquirers subsequentinnovation output through two possible mecha-nisms. First, an acquisition of another firm canbe viewed as an absorption of the acquired firmsknowledge base into the acquiring firms knowl-edge base. Such a union can potentially expandthe acquirers knowledge base and increase itsinnovation output by providing economies ofscale and scope in research and by enhancing theacquirers potential for inventive recombination(Henderson and Cockburn, 1996; Fleming, 1999).Since nontechnological acquisitions add less tothe knowledge base of the acquirer they are lesslikely to lead to such innovation output-enhancingeffects.1 However, an acquisition can also disruptthe established routines of the acquiring firm andthose of its newly acquired component, andthereby reduce productivity (Jemison and Sitkin,1986; Haspeslagh and Jemison, 1991). Priorresearch suggests that acquisition integrationentails far-reaching disruption, and involves sig-nificant managerial attention and transactionscosts (Pritchett, 1985; Haspeslagh and Jemison,1991; Hitt et al., 1991, 1996; Hoskissonet al.,1994). However, whether this disruption relatedto nontechnological acquisitions will produce anegative impact on innovative productivity is notclear, a priori. On one hand it is possible thatas management focuses more on acquisitions andtheir integration, decision making on routine tech-nological matters can be delayed, activities suchas product championing can suffer, and a crisismentality on the management of the acquisitioncan lead to only residual energies being suppliedto day-to-day operations even in the technologicalcore of the company (Thompson, 1967; Pritchett,1985; Hitt et al., 1996). Alternately, it is pos-sible that since such acquisitions do notinvolve a technological component by definition,they may not affect the technological subsystem

1 As suggested by one of the reviewers, in some casesnontechnological acquisitions can increase innovation output,for instance by providing contacts with new customers.


(Thompson, 1967) and innovation routines of thefirm, and therefore have no impact on innovationoutput at all. The final impact of an acquisitiondepends on the degree to which these effectscome into play.

Accordingly, we present the following basehypothesis:

Hypothesis 1: Nontechnological acquisitionswill affect the postacquisition innovation outputof acquiring firms either negatively or nonsig-nificantly.

Technological acquisitions and the absolutesize, relative size, and relatedness ofknowledge bases

Technological acquisitions are acquisitions thatprovide technological inputs to the acquiring firm.Thus, they potentially expand the acquirersknowledge base and provide scale, scope, andrecombination benefits (Henderson and Cockburn,1996; Fleming, 1999). However, technologicalacquisitions can also entail a disruption in organi-zational routines. Further, this disruption is mostlikely in the set of routines that are closest tothe innovation arena, the technological subsystemof the firm. Thus, technological acquisitions canalso have a negative impact on the innovationoutput of the acquiring firm. On balance,assessing whether technological acquisitions willhave a positive or negative impact on postacqui-sition innovation output is likely to depend uponthe quantity and nature of knowledge elementsthat they bring to the acquiring firm. To evaluatewhether the scale, scope, and inventive recombi-nation benefits of acquisitions outweigh theirnegative effects on organizational routines wecompare the knowledge bases of acquired andacquiring firms along three key characteristicsthat have also been prominently used in prioracquisition research: absolute size, relative size,and relatedness (Lubatkin, 1983). In the followingsections we examine and develop distinct hypoth-eses for each of these characteristics.

Absolute size of acquired knowledge base

The absolute size of an acquired knowledge basecan affect the post acquisition innovation outputof the acquiring firm through at least two mecha-nisms, both indicating a positive effect of larger


knowledge bases. First, the integration of the twohitherto separate knowledge bases may enableenhanced economies of scale and scope(Henderson and Cockburn, 1996). For instance, along tradition of research in technology suggeststhat new innovative outputs are often the resultof recombining existing elements of knowledgeinto new syntheses (Schumpeter, 1934; Hendersonand Clark, 1990; Kogut and Zander, 1992; Tush-man and Rosenkopf, 1992; Utterback, 1994;Fleming, 1999). From this combinatorial perspec-tive, the number of direct combinations that afirm can create from its own knowledge elementsincreases with the size of the acquired knowledgebase. While a firm with five units of knowledgecan generate 10 combinations using two elementsat a time, acquiring another firm with three unitsof knowledge increases the number of combi-nations that become available to 28. Similarly,the merger of two knowledge bases can alsoprovide scale or scope economies by reducingduplication in research efforts or by providing alarger research base to defray costs.

Second, acquiring a larger knowledge base mayenhance a firms absorptive capacity. Priorresearch indicates that as a firm expands its inter-nal knowledge base and technological capability,it also enhances its ability to absorb and utilizeexternal knowledge (Cohen and Levinthal, 1989;Cohen and Levinthal, 1990). Thus, when a firmacquires a knowledge base it obtains access notonly to the acquired firms internally createdknowledge but also to a larger external domainof knowledge that is understood and used bythe acquired firm. Thus, acquisitions increase thenumber of elements of both internal and externalknowledge that are available to the acquiring firmand hence increase its potential for inventiverecombination. Hence, we hypothesize:

Hypothesis 2: The greater the absolute size ofthe acquired knowledge base, the greater thesubsequent innovation output of the acquiringfirm.

Relative size of acquired knowledge base

The arguments above focused on the increasedscale, scope, and recombination benefits possiblethrough the acquisition of a knowledge base. Yet,several steps must be completed before newlyacquired knowledge can improve the acquirers


performance. The acquirer needs to recognize thevalue and content of the acquired knowledge,assimilate it, and apply it (Cohen and Levinthal,1990). The degree to which these tasks can besuccessfully accomplished is likely to vary withthe relative size of the acquiring and acquiredknowledge bases. The larger therelative size ofthe knowledge base to be integrated, the moredifficult these stages are likely to be, and themore negative the impact on postacquisition inno-vation output.

Knowledge is primarily transferred throughinteractions between the acquired and acquiringunits, and entails both teaching and learning onboth sides (Haspeslagh and Jemison, 1991). Inte-gration teams, meetings within and between thetwo R&D units, and extensive face-to-face com-munication are integral parts of the process bywhich the merging units learn about each otherstechnology and processes (Gerpott, 1995). Sinceevery communication needs both a sender and areceiver, the relative size of the knowledge basesin the merger becomes relevant. If the mergedknowledge bases are relatively equal in size, mostof the knowledge resources of the combined firmwill be devoted to the task of integrating the twoknowledge bases. As the two approximately equalgroups educate each other, fewer resources willbe available for conducting the actual businessof innovation. On the other hand, if the twoknowledge bases are relatively dissimilar in size,the absorption and assimilation activity willoccupy only a part of the larger groups resourceseven if it entails the preoccupation of the smallerof the two groups.

A second dimension of the relative size effectis the disruption of existing organizational rou-tines (Haspeslagh and Jemison, 1991; Singh andZollo, 1997). For successful assimilation andapplication of the newly acquired knowledge,many changes have to be introduced into thefunctioning of the organization. Pathways of com-munication, routing of work and authority, andformal and informal organizational structures allhave to be adapted to incorporate the acquiredunits knowledge (Gerpott, 1995). If the acquiredfirms knowledge base is small relative to theacquirer the modifications required are likely tobe minor, and therefore not be very disruptive.However, if the acquired firms knowledge baseis large relative to the acquiring firm, fairly majorchanges would have to be made in the acquiring


firm, leading to a significant disruption of existingprocesses. Accordingly, we predict:

Hypothesis 3: The greater the relative sizeof the acquired knowledge base, the less thesubsequent innovation output of the acquiringfirm.

Relatedness of acquired and acquiringknowledge bases

The third critical dimension in the unification oftwo knowledge bases is their relatedness(Lubatkin, 1983; Singh and Montgomery, 1987;Lane and Lubatkin, 1998). While the previousarguments concerned themagnitude of theacquired and acquiring knowledge bases therelatedness argument concerns thecontent ofthese knowledge bases. We predict thatrelatedness between the acquired and acquiringknowledge bases is likely to have a nonmonotonicinfluence on the subsequent innovation perfor-mance of acquiring firms. Innovation output willincrease with increasing relatedness, but beyondsome optimum innovation output will decreasewith increasing relatedness.

The absorptive capacity argument suggests thatthe ability to use new information to solve prob-lems is enhanced when the new knowledge isrelated to what is already known (Cohen andLevinthal, 1990). Elements of similar knowledgefacilitate the integration of the acquired andacquiring knowledge bases (Kogut and Zander,1992; Grant, 1996). Common skills, shared lan-guages, and similar cognitive structures enabletechnical communication and learning (Cohen andLevinthal, 1989; Lane and Lubatkin, 1998).Further, the recipes for conducting research, orthe innovation routines of the acquired andacquiring firms, are also likely to be different ifthe firms come from distant realms of technology(Kogut and Zander, 1992; Spender, 1989). Insuch circumstances the integration of knowledgebases can be resource consuming, or evencounterproductive as routines inappropriate toeither or both knowledge bases can be adopted(Haspeslagh and Jemison, 1991; Singh andZollo, 1997).

On the other hand, an acquired knowledge basethat is too similar to the acquiring knowledgebase may also contribute little to subsequent inno-vation performance. From an absorptive capacity


perspective acquired knowledge can help improveperformance through two effects. First, acquiredknowledge can provide a cross-fertilization effectas old problems can be addressed through newapproaches, or by a combination of old and newapproaches (Cohen and Levinthal, 1990). Second,new acquired knowledge can serve as the basisfor absorbing additional stimuli and informationfrom the external environment. If an acquisitionbrings in knowledge that is too closely related tothe existing knowledge base of the acquiring firm,both these benefits might be limited.

Knowledge bases with moderate degrees ofrelatedness provide the benefits of enhancing thevariety of possible combinations that the firm canuse, while maintaining the elements of com-monality that facilitate interaction between theacquired and acquiring knowledge bases. Basedon the above arguments we suggest that acqui-sitions that are characterized by a moderatedegree of overlap in the knowledge bases arelikely to provide the most significant positiveimpact on the acquiring firms subsequent inno-vation output. Accordingly, we predict:

Hypothesis 4: The relatedness of the acquiredknowledge base will be curvilinearly (invertedU) related to the subsequent innovation outputof the acquiring firm.

DATA AND METHODS

Organizational knowledge base

In the previous section we argued that an acqui-sition can be viewed as the union of two knowl-edge bases. Measuring an organizational knowl-edge base is then the key operational issue intesting the hypotheses. Specifically, we need toidentify empirical measures that capture the abso-lute size, relative size, and relatedness of knowl-edge bases. In the arguments below we suggestthat an organizations patent portfolio provides ameans for capturing these dimensions and map-ping an organizations knowledge.

A patent, by definition, represents a unique andnovel element of knowledge. A set of patentsthen represents a collection of discrete, distinctunits of knowledge. Identifying a set of patentsthat a firm has demonstrated familiarity with, ormastery of, can be a basis for identifying therevealed knowledge base of a firm, the distinct


elements of knowledge with which the firm hasrevealed a relationship (Kim and Kogut, 1996).

The patents owned by a firm represent theknowledge that the firm is acknowledged as hav-ing created (Jaffe, Trajtenberg and Henderson,1993). Such patents are naturally elements of thefirms knowledge base. However, the firms pa-tents also build on prior patents, the knowledgecreated by the same firm in the past or by otherfirms which have preceded it in that line ofinquiry. These prior patents are cited in the pa-tents as recognition of their contribution to theknowledge embodied in the focal patent. By cre-ating a patent that builds on these prior patents,the firm provides evidence that the knowledgecontained in those past patents is a part of thefirms knowledge set. Thus, the patents cited bythe firm should also be included in its knowledgebase. The Appendix describes in more detail thelogic underlying this approach to measuringorganizational knowledge bases.

Using patent data to measure organizationalknowledge bases corresponds closely to the con-ceptual abstraction of a firms knowledge base asa set of knowledge elements (Grant, 1996). Thenumber of cited and obtained patents provides ameasure of the size of the knowledge base. Theindividual patents in the firms knowledge baseprovide the basis for comparing the firms knowl-edge base with other knowledge bases. Sinceeach patent number uniquely identifies a distinctcomponent of knowledge, the higher the numberof patents that are common across two knowledgebases, the higher the relatedness between thoseknowledge bases (see also Stuart and Podolny,1996; Mowery, Oxley and Silverman, 1998).

Methods

A panel data design was used to test the hypoth-eses. We selected a sample of firms from theglobal chemicals industry independent of theiracquisition behavior, and traced the acquisitionbehavior of these firms over a 12-year period,from 1980 to 1991 (more details on the sampleare provided later). Several of the sample firmswere very active in the acquisitions arena, whileothers conducted few or no acquisitions at all.We attempted to collect data on every acquisitionconducted by these firms. The acquisition charac-teristics of these firms were modeled as time-varying influences on the subsequent innovation


performance of these firms. A distributed laganalysis was used to identify the effects of anacquisition on innovation performance in the 4years succeeding the acquisition.

This approach addresses several methodologicalproblems that arise in evaluating the impact ofacquisitions on the postacquisition performanceof acquiring firms. First, examining a singleindustry over a common period controls forindustry and period effects that have been citedas a problem with prior acquisition research(Fowler and Schmidt, 1988). Second, this researchdesign naturally includes both firms that are activein acquisitions and those that are inactive. In theabsence of the latter group of firms it is difficultto refute the argument that good or bad perfor-mance by acquiring firms was not shared bysimilar nonacquiring firms (Fowler and Schmidt,1988). Third, this approach resolves the problemof handling firms with multiple acquisitions. Toreduce the problem of confounding caused byfirms making several acquisitions over the studyperiod, some researchers have omitted such firmsfrom analyses (Fowler and Schmidt, 1988). How-ever, omitting such firms may lead to biasedfindings. With our research design such firms canbe retained in the analysis by including acqui-sition and firm characteristics as time-varyingcovariates in a panel regression. Finally, our paneldata set includes distributed lag effects. The dis-tributed lag technique enables us to use multiple-period lagged values of the independent variablesas additional regressors in the estimated equation(Judgeet al., 1988). By using this approach, wecan, in principle, trace the effects of the acqui-sition on the performance of the acquiring firmfor several periods after the acquisition.

Model specification and econometric issues

We now describe our econometric approach, fol-lowed by a discussion of the industry setting andthe variables used to test the hypotheses. Thedependent variable of the study, innovation out-put, as measured by patent counts, is a countvariable and takes only non-negative integervalues. The linear regression models assumptionsof homoskedastic, normally distributed errors arethus violated. A Poisson regression approach isappropriate for such data (Hausman, Hall, andGriliches, 1984; Henderson and Cockburn, 1996).Accordingly, we specified the following Poisson


regression model:

Pit = exp (Xit1 + Ait11 + Ait22+Ait33 + Ait44) (1)

where Pit is the number of patents obtained byfirm i in year t, Xit21 is a vector of controlvariables affectingPit, and Ait-year is the laggedvector of the acquisition variables for yearst 21 to t 2 4.

Intuitively, this specification implies that thenumber of patents obtained by any firm in anyyear is randomly distributed following a Poissonprocess, where the covariate vectorsXit21 andAit21, Ait22, Ait23, and Ait24 determine the meanof this process. Changes in the value of individualcovariates thus influence patenting frequency byaffecting the mean of the Poisson distributionfrom which observations are drawn, in a manneridentical to an ordinary regression.

Since the impact of an acquisition is likely tobe felt over a number of years, rather thanentirely in any one year, we used a distributedlag model (Judgeet al., 1988). To capture thelag effects, the one-period, two-period, three-period, and four-period lagged values of all acqui-sition related variables were included as covari-ates in the above model. This distributed lagmodel tests the impact of acquisitions for up to4 years after the year the acquisition was orig-inally made. Thus, illustratively speaking, in thesemodels the acquiring firms innovation perfor-mance in 1986 is potentially influenced by acqui-sitions made in 1982, 1983, 1984, and 1985. Insensitivity tests (results available from theauthors), instead of using four lags we also esti-mated the models using three and five lags andfound results substantively identical to thosereported here.

The use of distributed lags provides two bene-fits. First, it enables us to examine the timepattern of the impact of acquisitions on firminnovation output. For instance, if an acquisitioncontributes to improved performance for the first2 years but thereafter leads to no further improve-ment in innovation output, the 1- and 2-yearlagged acquisition variables will be positive andsignificant, while the 3- and 4-year lagged acqui-sition variables will be nonsignificant. Second,since the total impact of an acquisition is likelyto be distributed over several periods followingthe acquisition and may be statistically incon-


sequential in any one period, the regression coef-ficients on the distributed lags can be summed toobtain the total impact of an acquisition acrosstime (Gujarati, 1988: 507). By using the variancesand covariances of the individual lag coefficientsfrom the regression output the variance of thesummed coefficient can be calculated. Thesummed coefficient can then be used for hypoth-esis testing. For instance, the hypothesis that thetotal impact of acquisitions summed across allyears is zero can be tested by computing thefollowing t-statistic and checking whether it isstatistically significant at the desired confidencelevel (Greene, 1993; Gujarati, 1988).

t = (t1 + t2 + t3 + t4)

/ Variance (t1 + t2 + t3 + t4)

where the t2is are the regression coefficientsfor the ith period lagged acquisition variable(Judgeet al., 1988). The variance for the summedcoefficients can be computed using the followingrelation (Gujarati, 1988: 507):

Variance (t1 + t2 + t3 + t4)

= [Var(t1) + Var(t2)

+ Var(t3) + Var(t4) (2)

+ 2Cov(t1t2) + 2Cov(t1t3)

+ 2Cov(t1t-4) + 2Cov(t2t3)

+ 2Cov(t2t4) + 2Cov(t3t4)]

The conditioning vectorX in Equation 1 helpsus to control for alternative explanations. Forinstance, since an acquisition represents theabsorption of one firm by another, a simple expla-nation for an increased postacquisition innovationoutput would simply be increased innovativeinputsthe one postacquisition firm reflects thecombined efforts of the two preacquisition firms.By including innovative inputs as time-varyingcovariates in the conditioning vectorX, we candirectly control for such effects. Other possibledeterminants of innovation outputs such as firmsize, diversification, nationality, and time are alsocontrolled for through this vector.

This specification does not account for unob-served heterogeneity, or the possibility that obser-vationally equivalent firms may differ on unmeas-ured characteristics. For instance, firms may enter


the sample with inherently different innovationgenerating capabilities. Such unobserved hetero-geneity, if present and not controlled for, cancause estimation problems. First, it can lead tooverdispersion in the data. For the Poisson distri-bution the variance is restricted to equal themean. If this restriction is false and the data areoverdispersed in that the variance exceeds themean, the computed standard errors in a Poissonregression are understated. Second, unobservedfirm effects can lead to serial correlation amongthe residuals of observations from the same firm.Under overdispersion or serial correlation, com-puted regression coefficients remain consistentlyestimated; however, the standard errors are inac-curate. Thus, hypothesis testing and inference canbe invalidated. To address the possibility of unob-served heterogeneity we used the Presample PanelPoisson approach (Blundell, Griffith, and VanReenen, 1995).

In the Presample approach, unobserved hetero-geneity is modeled as an additional covariate inthe basic Poisson model. The values of the depen-dent variable in the periods immediately preced-ing the study period are used to construct aninstrumental variable. This instrumental variableserves as a fixed-effect for the firms in thepanel and helps to partial out unobservable differ-ences across firms. Thus, presample informationon the firms provides the basis for controllingfor unobserved heterogeneity. In the context ofthe current research the Presample variable canbe interpreted as a measure of the unobserveddifferences in knowledge stocks between the sam-ple firms.

Although the presample approach can help inthe reduction of overdispersion and serial corre-lation, a more complete treatment of these poten-tial problems would be to use an estimationapproach that accounts for any remaining over-dispersion and serial correlation even after theinclusion of a presample variable. The Gen-eralized Estimating Equations (GEE) method-ology provides a direct approach to modelinglongitudinal Poisson data with serial correlationin a regression context (Liang and Zeger, 1986).The GEE estimation procedure involves twostages. In the first stage of the procedure betacoefficients are estimated with the assumption ofindependence across observations. This processyields consistent estimates of the beta parametersand the residuals from this regression provide an


estimate of the working correlations betweenthe errors of different observations. This workingcorrelation matrix is then used as an input in asecond regression. The beta coefficients and stan-dard errors from this second regression provideconsistent estimates of the underlying parameterswhile accounting for the observed correlationbetween observations. To ensure that all residualcorrelation was accounted for, we used the GEEprocedure to estimate all models. Even for theGEE approach instead of using the model-basedstandard errors we used the more conservative(i.e., larger) empirical standard errors. Thisensures that other potential misspecifications ofthe variance function, such as any residual over-dispersion, are also accounted for.

Sample and data

We tested the hypotheses on a longitudinal dataset comprising the acquisition and patentingactivities of 72 leading firms from the globalchemicals industry. Focusing on the largest firmsof the industry was necessary to ensure the avail-ability and reliability of data. Obtaining infor-mation on the key variables for smaller firms isextremely difficult. This focus on large firms isalso consistent with prior research on acquisitions(Hitt et al., 1991, 1996). We identified the 100leading players in the chemicals industry fromlists of the largest chemicals firms that are pub-lished annually by trade journals such asChemi-cal WeekandC&E News. To avoid survivor bias,the selected sample was drawn from the lists atthe beginning of the study period. In these pub-lished lists subsidiaries were often listed sepa-rately from parent firms. After combining subsidi-aries with parent firms, a sample of 82independent firms was identified for inclusion inthe sample. However, for 10 firms data could notbe reliably obtained and they were subsequentlydropped from the analysis. The remaining firmsinclude the key firms in the industry over thestudy period and comprise of 30 European, 26American, and 16 Japanese firms. The panel isunbalanced as some of the firms were acquiredby other firms or restructured in a fashion thatmade comparison difficult beyond a particularyear. Even though the sample was focused onthe largest firms in the chemicals industry theinclusion of 72 firms provides significant depthto the sample and ensures that there is consider-


able variety within the sample on the key vari-ables of this study. For instance, the number ofemployees for firms in the sample varies from aminimum of 2300 to a maximum of more than181,000. Similarly, the number of patentsobtained yearly varies from 0 to 760. Financialfigures and personnel data on these firms wereobtained from Compustat, Worldscope, JapanCompany Handbooks, Daiwa Institute ResearchGuides, and trade publications and companyannual reports. For all firms, financial data wereconverted to constant (1985) U.S. dollars toensure standardization within the sample. A fulllist of the sample firms is available from theauthors.

The chemicals industry is an appropriate settingfor several reasons. First, technology-based acqui-sitions have been a significant feature of thisindustry (Chemical Week, 1983; Gibson, 1985).Second, patents are generally regarded to beeffective, and used widely and consistently in thechemicals industry (Levinet al., 1987; Arundeland Kabla, 1998). For the firms in the samplewe obtained yearly patent counts from 1975 to1992 and acquisition and firm attribute data forthe years 1980 to 1991. The need to use laggedrelationships between patent counts and the othervariables and to construct a control for unob-served heterogeneity reduced the final panel forregression analysis to 9 years. We describe thedata collection and coding procedures for thethree sets of variables in some detail below.

We used U.S. patent data for all firms, includ-ing the foreign firms in the sample. This wasnecessary to maintain consistency, reliability, andcomparability, as patenting systems across nationsdiffer in their application of standards, system ofgranting patents, and value of protection granted.The United States represents one of the largestmarkets for chemicals, and firms desirous of com-mercializing their inventions typically patent inthe United States if they patent anywhere. Studiesby Dosi, Pavitt and Suete (1990) and Basberg(1983) show empirically that U.S. patent dataprovide a good measure of foreign firms inno-vativeness. Prior research using patent data oninternational samples has also followed this strat-egy of using U.S. patent data for internationalfirms (e.g., Stuart and Podolny, 1996; Patel andPavitt, 1997).

Data on acquisitions were obtained throughdetailed archival research on the chemicals sec-


tor. Three main types of data sources were usedto identify acquisition activity and to collectdata on acquisition transactions: (1) commercialdata bases, including general business newsmedia such as the Dow Jones News RetrievalText Index, and chemicals sector-specific databases such as Metadex, (2) general businessprint media such as the Frost and Sullivan Predi-casts Index (United States, International, andEurope),Mergers and Acquisitions JournalandMoodys Manuals, and industry-specific publi-cations such asChemical Weekand PlasticsTechnology, and, (3) government publicationsand consultant reports for the chemicals indus-try. We were able to identify 1287 acquisitionannouncements for the sample firms over theperiod of 1980 to 1991.

To identify the technological acquisitionswithin this larger sample two approaches wereused. First, we obtained detailed news storiesassociated with each acquisition announcement.We were able to obtain news stories providingdetails for 516 of the 1287 acquisitions identifiedfor the sample firms. For the remaining acqui-sitions no further details, or inadequate details,were available. Second, we searched the U.S.Patents Data Base to determine if the acquiredfirms had obtained any patents in the 5 yearspreceding the acquisition. 165 of the acquiredfirms had obtained patents during the 5 yearspreceding their acquisition. For all subsequentanalyses we retained only those acquisitions forwhich we were able to find either corroboratingnews stories or patenting activity (534 acqui-sitions in total). Acquisitions for which we wereable to obtain no further information were omit-ted. The paucity of information on these acqui-sitions suggests that these acquisitions were verysmall or unimportant. Therefore, our analysis isbased on the 534 acquisitions for which we haverelatively complete information.

We used two criteria to distinguish technologi-cal acquisitions from all other acquisitions. First,we examined the news stories to establish if theacquiring firm reported technology as a motivat-ing factor for the acquisition or if technologywas a part of the transferred assets. We classifiedthe acquisition as technological if either of theseconditions was met. Second, we classified theacquisition as technological if the acquired firmhad any patenting activity in the 5 years preced-ing the acquisition. Of the 534 acquisitions on


which we had information, 283 met at least oneof the two above criteria and were classifiedas technological acquisitions. The remaining 251acquisitions were classified as nontechnological.

This classification scheme reflects a fairlyinclusive definition of technological acquisitions.Firms need not have patented to be classified astechnology acquisitions. Further, acquired firmsobtaining even a single patent in the 5 yearsprior to the acquisition are classified as techno-logical acquisitions.

This classification scheme provides two bene-fits. First, it uses two indicators to identify tech-nological acquisitions and therefore enables morecomplete identification of technological acqui-sitions than either indicator by itself would. Forinstance, our patent-based measures cannot becomputed if only a part of a firm is acquired,rather than the complete entity, or if the tech-nology has not been patented at all. In suchcircumstances the news story could provide anindication of whether the transaction entailedtechnology or not. Similarly, if the acquisitionwas motivated by multiple factors of which tech-nology was not necessarily the one mentioned inthe news story, the patents measure provides anindicator of whether technology was involved.Thus, using the input from the news stories sup-plements our patent-based measures.

Second, using this scheme of classificationmakes our statistical tests more conservative.With this scheme we are likely to capture anyacquisition that includes a technological compo-nent. Further, we are more likely to err in thedirection of defining acquisitions as technologicaleven when they have a relatively small tech-nology component. If such misclassificationsoccur, we reduce our likelihood of findinga positive impact of technological acquisitionson the innovation performance of acquiringfirms.

Finally, we need to select a time period orwindow for measuring a firms knowledge base.At one extreme only the current years patentscould be considered relevant. At another extremeany patent obtained by the firm in the past couldbe included in computing its current knowledgebase. Prior research suggests that knowledge capi-tal depreciates sharply, losing significant valuewithin 5 years (Griliches, 1979). Although thedepreciation rate for knowledge capital is likelyto vary across industries, a boundary of 5 or 6


years seems reasonable and has been used byother researchers (Podolny and Stuart, 1995;Henderson and Cockburn, 1996). In this researchwe use the patents obtained by a firm in thepreceding 5 years to compute our knowledgebase measures. We also computed some of themeasures using patents for the preceding 6 years.These measures were highly correlated with the5-year measures, suggesting that the construct isnot unduly influenced by changes in the timeperiod used to compute it.

Variable definitions and operationalization

Dependent variable

Patents Innovation output, the dependent vari-able, is measured through the patenting frequencyof firms, that is, the number of successful patentapplications by a firm in a given year. Patentshave both significant strengths and weaknesses asmeasures of innovation output. First, patents aredirectly related to inventiveness: they are grantedonly for nonobvious improvements or solutionswith discernible utility (Walker, 1995). Second,they represent an externally validated measure oftechnological novelty (Griliches, 1990). Third,they confer property rights upon the assignee andtherefore haveeconomicsignificance (Kamien andSchwartz, 1982: 49; Scherer and Ross, 1990:621).

Patents also correlate well with other measuresof innovative output. Empirical studies find thatpatents are closely related to measures such asnew products (Comanor and Scherer, 1969), inno-vation and invention counts (Achilladelis,Schwarzkopf and Cines, 1987), and sales growth(Scherer, 1965). Expert ratings of corporate tech-nological strength are also highly correlated withthe number of patents held by corporations(Narin, Noma, and Perry, 1987). Further, surveysof patent holders indicate that the rate of utili-zation of patents is reasonably high, with esti-mates indicating that between 41 percent and 55percent of all patents granted are put to commer-cial use for at least a limited time (Griliches,1990). Similarly, about 50 percent of all patentsgranted are still being renewed and a renewal feeis being paid 10 years after the patents hadoriginally been applied for (Griliches, 1990;Schankerman and Pakes, 1986). Given a nonneg-


ligible renewal fee, this indicates a significantusefulness for the majority of patents for a sig-nificant time period.

However, the use of patents as a measureof innovative output also has limitations. Someinventions are not patentable, others are not pa-tented, and the inventions that are patented differgreatly in economic value (Cohen and Levin,1989; Griliches, 1990; Trajtenberg, 1990).Research, and the logic of appropriability, indi-cate that the degree to which the first two ofthese factors are a problem varies significantlyacross industries (Cohen and Levin 1989; Levinet al., 1987). Limiting the study to a singleindustrial sector or a few closely related sectorsminimizes such problems as the factors that affectpatenting propensity are likely to be stable withinsuch a context (Basberg, 1987; Cohen and Levin,1989; Griliches, 1990).

We measure Patentsit, as the number ofsuc-cessfulpatent applications, or granted patents, forthe acquiring firmi in year t. The granted patentcarries the date of the original application. Weuse this date to assign a granted patent to theparticular year when it was originally appliedfor. This procedure permits consistency in thetreatment of all patents and controls for differ-ences in delays that may occur in granting patentsafter the application is filed (Trajtenberg, 1990).Note that the patent count for the dependentvariable is based on the patents of the acquiringfirm obtained 14 yearsafter the acquisition inall cases. These patents are therefore differentfrom the patents on which the independent vari-ables are based. The independent variablesdescribed below are based on patents obtainedby the acquired and acquiring firms, respectively,in the 5 yearsbefore the acquisition.

Independent variables

Number of nontechnological acquisitionsAsnoted earlier, acquisitions were coded as techno-logical acquisitions if either of the following twocriteria are met: first, if the acquiring firmreported technology as a motivating factor for theacquisition or if technology was reported as apart of the transferred assets of the acquired firm;second, if the acquired firm had any patentingactivity in the 5 years preceding the acquisition.Acquisitions that did not meet either of the abovecriteria were coded as nontechnological acqui-


sitions. We used four lagged versions of this andall other acquisition related variables listed below.

Absolute size of acquired knowledge baseToobtain this variable we used the following pro-cedure: for each acquiring firm for each year, weprepared a list of the patents its acquisitions hadobtained in the preceding 5 years. These patentswere then combined with the patents cited bythem. Thereafter, all duplicates were removed toensure that a patent number appeared only onceon this list. The acquired knowledge base wasthen computed as the number of patents (i.e.,knowledge elements) on this list.

Relative size of acquired knowledge baseThisvariable was obtained by dividingAbsolutesize of acquired knowledge baseby the size ofthe acquiring firms knowledge base. The size ofthe acquiring firms knowledge base wascomputed using the same procedure as the sizeof the acquired firms knowledge base (seeabove). In a few cases the acquired knowledgebase was larger than the acquiring knowledgebase. In these cases we used the larger numberas the denominator. Since our theoretical mech-anism is concerned with the relativeproportionof the merged firms resources that are likely tobe occupied with integrative rather than inventiveactivity, a number greater than 1 is not meaning-ful.

Relatedness of acquired knowledge baseTomeasure the relatedness of the acquired knowl-edge base, the following procedure was followed.First, the list of patent numbers that appeared inboth the acquired firms knowledge base and inthe acquiring firms knowledge base was pre-pared. Then, the number of elements on thislist was divided by Absolute size of acquiredknowledge base.

Number of technological acquisitions where pa-tents unavailable This is the number of acqui-sitions for which news stories indicated that tech-nology was a component of the transferred assetsbut where no patents could be identified with theacquisition. This would occur if either theacquired unit had obtained no patents or if theacquired unit was a part of a larger parent firmand patents were not assigned separately to theacquired unit and therefore could not be separated


from the parent firms patents.

Controls

We included several control variables in the mod-els. These control variables include yearly R&Dexpenditures (R&D), firm size as measured bynatural log of number of employees(Logemployees), firm diversification as measuredby entropy (Diversification), and a measure ofnational cultural distance between the acquiredand acquiring firms (Foreign acquisitions). Thefollowing formula was used to calculate theDiversification measure: Entropy = SPj 3ln(1/Pj), wherePj is defined as the percentage offirm sales in segmentj and ln(1/Pj) is the weightfor each segmentj (Palepu, 1985).Foreign acqui-sitions was computed as

Foreign acquisitions= O4i=1

{( IijIiu)2/Vi} /4 (3)

where Ii stands for the index of theith culturaldimension, j and u are subscripts indicating thecountries j and u, and Vi is the variance of theindex of theith dimension (Hofstede, 1980). Theindex thus indicates the cultural distance betweenthe acquirers country (j) and the acquired firmscountry (u) (Kogut and Singh, 1988). We antici-pate that foreign acquisitions might be more dif-ficult to integrate than domestic acquisitions.Finally, in all models we included the firm hetero-geneity control variablePresample patents(thesum of patents obtained by a firm in the 3 yearsprior to the firms entry into the sample) anddummy variables for acquirer nationality and cal-endar year.

RESULTS

Table 1 provides descriptive statistics and corre-lations. The table indicates the diversity of firmsincluded in the sample. Even though the sampleinvolves the prominent players in the industry,there is considerable variance on all the keyvariables such asPatents, R&D, Logemployees,and the acquisition variables. The variablesreflecting the hypothesized effects are not veryhighly correlated among themselves or with thecontrol variables. However, the correlationsbetween some of the control variables are high,


notably the correlation betweenR&D andLogem-ployees (0.72), and the correlation of the firmfixed-effect variable Presample patentswithR&D (0.89) andLogemployees(0.72). Robustnesstests (reported at the end of this section), how-ever, indicate that the results on the hypothesizedeffects were strong and unaffected by these highcorrelations between some of the control vari-ables.

Table 2 provides results for all models usingGEE Poisson estimators (reported with empiricalstandard errors). The variables reflecting thehypothesized effects were entered into theregression individually and likelihood ratio testsare reported for all models. Model 1 in Table2 presents the base model with firm- and acqui-sition-related control variables. Models 26include theNumber of nontechnological acqui-sitions, Absolute size of acquired knowledgebase, Relative size of acquired knowledge base,Relatedness of acquired knowledge baseandRelatedness of acquired knowledge base2 vari-ables entered successively. We use the fullmodel to discuss the results of the hypothesistests. The summed coefficients and theassociated standard errors for the lagged acqui-sition variables for these models are given atthe bottom of the table.

In Hypothesis 1 we predicted a negative ornonsignificant relationship between the number ofnontechnological acquisitions conducted in a yearand the subsequent innovation output of theacquiring firm. The coefficients ofNumber ofnon-technological acquisitionsare nonsignificantfor all four periods individually. The summedcoefficient, which represents the total effect ofacquisitions across the 4 years, is not statisticallysignificantly different from zero. Thus, we do notfind an appreciable impact of acquisitions withouta technological component on the innovation out-put of the acquirer for the four periods followingthe acquisition.

The coefficients of the fourAbsolute size ofacquired knowledge basevariables in Model 6 inTable 2 are positive and significant, supportingHypothesis 2. The summed coefficient reflectingthe total impact of the acquisition is also positiveand significant. However, as the summed coef-ficient indicates (0.0004), the absolute magnitudeof this effect is low. A one-unit increase in theknowledge base of the acquired firm leads to a0.04 percent increase in the acquirers innovation


Tab

le1.

Mea

ns,

stan

dard

devi

atio

ns,

min

imum

and

max

imum

valu

esan

dbi

varia

teco

rrel

atio

nsfo

ral

lva

riabl

es

Var

iabl

eM

ean

S.D

.M

in.

Max

.1

23

45

67

89

1011

1213

1415

1617

1819

2021

2223

2425

2627

2829

3031

3233

34

1P

atie

nts t

93.9

013

7.65

0.00

760.

002

No.

ofno

ntec

hnol

ogic

al0.

350.

790.

006.

000.

12ac

quis

ition

s t1

3N

o.of

nont

echn

olog

ical

0.34

0.78

0.00

6.00

0.12

0.26

acqu

isiti

ons t

2

4N

o.of

nont

echn

olog

ical

0.29

0.68

0.00

5.00

0.11

0.29

0.23

acqu

isiti

ons t

3

5N

o.of

nont

echn

olog

ical

0.25

0.69

0.00

7.00

0.12

0.21

0.29

0.30

acqu

isiti

ons t

4

6A

bsol

ute

size

ofac

quire

d26.

0827

5.94

0.00

6227

.00

0.13

0.070

.03

0.01

0.0

2kn

owle

dge

base t

1

7A

bsol

ute

size

ofac

quire

d34.

5933

2.94

0.00

6227

.00

0.150

.03

0.05

0.0

30.

000.

02kn

owle

dge

base t

2

8A

bsol

ute

size

ofac

quire

d34.

8333

8.80

0.00

6227

.00

0.180

.02

0.03

0.01

0.0

30.

030.

01kn

owle

dge

base t

3

9A

bsol

ute

size

ofac

quire

d33.

2033

7.48

0.00

6227

.00

0.19

0.000

.02

0.03

0.02

0.51

0.03

0.01

know

ledg

eba

se t4

10R

elat

ive

size

ofac

quire

d0.

020.

080.

000.

900

.04

0.12

0.00

0.00

0.05

0.44

0.0

10.

000.

16kn

owle

dge

base t

1

11R

elat

ive

size

ofac

quire

d0.

020.

090.

000.

900

.01

0.01

0.12

0.0

10.

010.

000.

560

.01

0.00

0.0

2kn

owle

dge

base t

2

12R

elat

ive

size

ofac

quire

d0.

020.

090.

000.

900.

010.

040.

000.

090.

000.

040.

000.

550

.01

0.05

0.0

2kn

owle

dge

base t

3

13R

elat

ive

size

ofac

quire

d0.

020.

080.

000.

900.

020.

020.

030.

030.

050.

370.

030.

000.

560.

120.

050.

00kn

owle

dge

base t

4

14R

elat

edne

ssof

acqu

ired

0.01

0.06

0.00

1.00

0.03

0.02

0.02

0.0

30.

030.

040.

000

.01

0.01

0.09

0.0

10.

020.

27kn

owle

dge

base t

1

15R

elat

edne

ssof

acqu

ired

0.01

0.06

0.00

1.00

0.04

0.000

.01

0.02

0.0

30.

010.

030.

000

.01

0.00

0.08

0.00

0.01

0.0

2kn

owle

dge

base t

2

16R

elat

edne

ssof

acqu

ired

0.01

0.06

0.00

1.00

0.05

0.03

0.000

.01

0.03

0.07

0.0

10.

030.

000.

070.

000.

070.

000.0

20.

02kn

owle

dge

base t

3

17R

elat

edne

ssof

acqu

ired

0.01

0.06

0.00

1.00

0.06

0.02

0.03

0.000

.01

0.09

0.06

0.0

10.

030.

020.

060.

000.

070.

040.0

10.

01kn

owle

dge

base t

4

18R

elat

edne

ssof

acqu

ired

0.00

0.05

0.00

1.00

0.0

10.

030.

020

.03

0.03

0.01

0.0

10.

000.

010.

040.

010

.01

0.36

0.92

0.0

10.

010.

00kn

owle

dge

base t

12

19R

elat

edne

ssof

acqu

ired

0.00

0.05

0.00

1.00

0.00

0.02

0.0

30.

020

.03

0.00

0.00

0.0

10.

000.

010.

030.

010.

000

.01

0.92

0.0

10.

010

.01

know

ledg

eba

se t22



Tab

le1.

Con

tinue

d

Var

iabl

eM

ean

S.D

.M

in.

Max

.1

23

45

67

89

1011

1213

1415

1617

1819

2021

2223

2425

2627

2829

3031

3233

34

20R

elat

edne

ssof

acqu

ired

0.00

0.05

0.00

1.00

0.00

0.000

.02

0.03

0.03

0.02

0.00

0.00

0.01

0.03

0.01

0.03

0.01

0.0

10.

010.

920

.01

0.01

0.0

1kn

owle

dge

base t

32

21R

elat

edne

ssof

acqu

ired

0.00

0.05

0.00

1.00

0.01

0.01

0.00

0.0

20.

030.

050.

020.

000.

000.

010.

030.

010.

040.

020.0

10.

010.

930.

000

.01

0.01

know

ledg

eba

se t42

22N

o.of

tech

nolo

gica

l0.

170.

470.

003.

000.

190.

220.

160.

170.

200.

030.0

30.

020.

020.

060

.04

0.00

0.0

20.

010.

010.

030.

000.

010

.02

0.00

0.0

2ac

quis

ition

sw

here

pate

nts

unav

aila

ble t1

23N

o.of

tech

nolo

gica

l0.

160.

460.

003.

000.

210.

100.

230.

180.

170.0

10.

010

.03

0.02

0.00

0.05

0.0

30.

010

.03

0.01

0.02

0.04

0.03

0.0

10.

010.

000.

28ac

quis

ition

sw

here

pate

nts

unav

aila

ble t2

24N

o.of

tech

nolo

gica

l0.

140.

430.

003.

000.

220.

150.

110.

230.

220.

010.0

10.

010

.03

0.00

0.0

20.

020

.02

0.05

0.0

30.

020.

000.

030.

020

.01

0.02

0.19

0.31

acqu

isiti

ons

whe

repa

tent

sun

avai

labl

e t325

No.

ofte

chno

logi

cal

0.13

0.41

0.00

3.00

0.19

0.15

0.15

0.08

0.200

.02

0.00

0.0

20.

020

.02

0.00

0.0

20.

010.

020.

060

.03

0.00

0.0

10.

030

.02

0.01

0.14

0.21

0.31

acqu

isiti

ons

whe

repa

tent

sun

avai

labl

e t426

For

eign

acqu

isiti

ons t1

0.20

0.86

0.00

14.6

60.

020.

230.

060.

070.

040.

150

.02

0.00

0.0

10.

330

.03

0.11

0.01

0.12

0.02

0.17

0.0

10.

080

.02

0.09

0.0

10.

390.

140.

120.

0427

For

eign

acqu

isiti

ons t2

0.19

0.85

0.00

14.6

60.

030.

060.

240.

050.

080.

000.

120

.01

0.00

0.01

0.29

0.03

0.12

0.01

0.13

0.02

0.19

0.00

0.08

0.01

0.09

0.10

0.40

0.16

0.14

0.15

28F

orei

gnac

quis

ition

s t30.

150.

600.

006.

860.

040.

240.

100.

150.

100.

080.

000.

010

.01

0.11

0.00

0.15

0.02

0.0

10.

020.

190

.01

0.02

0.01

0.12

0.0

20.

180.

170.

410.

220.

420.

2329

For

eign

acqu

isiti

ons t4

0.15

0.69

0.00

8.97

0.03

0.10

0.21

0.07

0.11

0.01

0.06

0.0

10.

010.

010.

080.

010.

160

.01

0.00

0.01

0.17

0.01

0.0

10.

000.

100.

050.

150.

140.

430.

150.

370.

2530

R&

Dt

119

.32

30.2

40.

2316

1.25

0.89

0.18

0.19

0.16

0.16

0.08

0.09

0.10

0.11

0.0

40.

020

.01

0.01

0.03

0.05

0.07

0.07

0.00

0.00

0.01

0.02

0.24

0.30

0.30

0.26

0.07

0.11

0.15

0.12

31Lo

gem

ploy

ees t1

2.76

1.14

0.83

5.20

0.70

0.31

0.31

0.28

0.28

0.13

0.15

0.15

0.15

0.00

0.03

0.05

0.04

0.05

0.05

0.06

0.05

0.01

0.01

0.01

0.00

0.29

0.28

0.28

0.24

0.13

0.14

0.1

80.

170.

7232

Div

ersi

ficat

ion/

Ent

ropy t

10.

530.

410.

001.

600.

040.

100.

080.

070.

060.

100.

120.

110.

100.

050.

090.

100.

100.

010.

010.

000.

010.

020.

020.

010.

020

.03

0.07

0.0

80.

120.

010.

000

.06

0.08

0.0

50.

1433

U.S

.A.

0.35

0.48

0.00

1.00

0.10

0.04

0.0

40.

000.

020.

040.

070.

090.

090.0

10.

030.

040.

080.

080.

070.

070.

070.

070.

070.

070.

070.

050.

040.

070.

080

.13

0.12

0.1

20.

110

.02

0.08

0.13

34Ja

pan

0.24

0.43

0.00

1.00

0.1

90.

240

.23

0.22

0.2

00.

040

.05

0.05

0.0

50.

080

.09

0.09

0.0

90.

070

.07

0.07

0.0

60.

050

.04

0.04

0.0

40.

170

.17

0.15

0.1

60.

050

.07

0.07

0.0

70.

210

.53

0.10

0.4

235

Pre

sam

ple

pate

nts

t1

2.28

3.26

0.01

15.8

00.

940.

110.

120.

090.

100.

070.

080.

110.

110

.07

0.05

0.0

30.

030.

050.

050.

050.

050.

010.

010.

010.

010.

200.

210.

210.

180.

000.

020.

050.

040.

890.

720.

020.

100

.20



Table 2. GEE presample Poisson regression with distributed lag analysis predicting patentst variable

Variable 1 2 3 4 5 6

Intercept 1.906*** 1.916*** 1.994*** 2.055*** 2.041*** 2.048***[0.333] [0.334] [0.296] [0.293] [0.295] [0.292]

No. of 0.004 0.009 0.009 0.008 0.004nontechnological [0.017] [0.017] [0.016] [0.016] [0.017]acquisitionst1No. of 0.002 0.0003 0.004 0.005 0.004nontechnological [0.018] [0.017] [0.016] [0.016] [0.016]acquisitionst2No. of 0.001 0.006 0.002 0.003 0.002nontechnological [0.023] [0.023] [0.023] [0.023] [0.024]acquisitionst3No. of 0.010 0.012 0.008 0.010 0.008nontechnological [0.014] [0.014] [0.013] [0.013] [0.012]acquisitionst4Absolute size of 0.0001** 0.0001*** 0.0001*** 0.0001***acquired knowledge [0.00004] [0.00002] [0.00002] [0.00003]baset1Absolute size of 0.0001*** 0.0001*** 0.0001*** 0.0001***acquired knowledge [0.00003] [0.00002] [0.00002] [0.00002]baset2Absolute size of 0.0001*** 0.0001*** 0.0001*** 0.0002***acquired knowledge [0.00002] [0.00002] [0.00002] [0.00002]baset3Absolute size of 0.0000 0.0001** 0.0001** 0.0000*acquired knowledge [0.0000] [0.00004] [0.00004] [0.0001]baset4Relative size of 0.393** 0.387** 0.348*acquired knowledge [0.165] [0.167] [0.156]baset1Relative size of 0.447*** 0.442*** 0.455***acquired knowledge [0.144] [0.142] [0.144]baset2Relative size of 0.388*** 0.405*** 0.488***acquired knowledge [0.125] [0.125] [0.121]baset3Relative size of 0.479** 0.430* 0.369*acquired knowledge [0.196] [0.198] [0.209]baset4Relatedness of 0.207 0.221acquired knowledge [0.176] [0.558]baset1Relatedness of 0.002 0.821*acquired knowledge [0.112] [0.384]baset2Relatedness of 0.158 0.847*acquired knowledge [0.111] [0.371]baset3Relatedness of 0.151 0.836**acquired knowledge [0.121] [0.335]baset4Relatedness of 0.820acquired knowledge [0.911]base2t1



Table 2. Continued


Relatedness of 1.580**acquired knowledge [0.572]base2t2Relatedness of 1.301*acquired knowledge [0.572]base2t3Relatedness of 1.251**acquired knowledge [0.501]base2t4No. of technological 0.023 0.034 0.016 0.018 0.009acquisitions where [0.026] [0.024] [0.023] [0.023] [0.024]patentsunavailablet1No. of technological 0.011 0.025 0.016 0.016 0.014acquisitions where [0.024] [0.022] [0.021] [0.021] [0.022]patentsunavailablet2No. of technological 0.039 0.053* 0.043 0.047 0.045acquisitions where [0.030] [0.027] [0.026] [0.025] [0.025]patentsunavailablet3No. of technological 0.063* 0.068* 0.066* 0.073** 0.077**acquisitions where [0.020] [0.027] [0.026] [0.026] [0.027]patentsunavailablet4Foreign 0.034** 0.031 0.026 0.035 0.038* 0.040*acquisitionst1 [0.012] [0.016] [0.017] [0.019] [0.017] [0.017]Foreign 0.020* 0.019 0.014 0.023 0.022 0.024acquisitionst2 [0.010] [0.013] [0.012] [0.015] [0.015] [0.015]Foreign 0.020 0.042* 0.051** 0.042* 0.046** 0.043*acquisitionst3 [0.013] [0.017] [0.016] [0.017] [0.016] [0.017]Foreign 0.009 0.008 0.010 0.004 0.008 0.011acquisitionst4 [0.013] [0.015] [0.015] [0.017] [0.017] [0.017]R&Dt1 0.0004 0.001 0.001 0.001 0.001 0.001

[0.001] [0.002] [0.002] [0.002] [0.001] [0.001]Logemployeest1 0.554*** 0.553*** 0.534*** 0.517*** 0.522*** 0.516***

[0.105] [0.109] [0.099] [0.097] [0.097] [0.096]Diversification/ 0.176* 0.187* 0.217* 0.180* 0.183* 0.174*Entropyt1 [0.076] [0.083] [0.085] [0.080] [0.080] [0.076]US firm 0.553*** 0.549*** 0.526*** 0.509*** 0.507*** 0.514***

[0.128] [0.130] [0.120] [0.118] [0.117] [0.114]Japanese firm 0.997*** 1.006*** 0.991*** 0.942*** 0.950*** 0.951***

[0.233] [0.229] [0.219] [0.217] [0.217] [0.216]Presample patentst1 0.135*** 0.137*** 0.145*** 0.142*** 0.143*** 0.142***

[0.022] [0.024] [0.024] [0.023] [0.023] [0.022]

output. If we consider a hypothetical acquisitionin which the acquired firm has a knowledgebase with 50 elements (50 elements correspondsapproximately to a firm with 10 owned patentsplus 40 cited patents) then the above coefficientsuggests that this acquisition should lead to a 2percent increase in innovation output (0.00043


50 = 0.02) for the acquiring firm after the acqui-sition, other things being equal.

In Hypothesis 3, a negative relationship wasproposed between the relative size of an acqui-sition and the subsequent innovation performanceof the acquiring firm. This prediction was alsoborne out. Specifically, the individual and


Table 2. Continued


N 598 598 598 598 598 598Pearson chi sq./d.f. 10570/579 10491/571 9923/567 9454/563 9448/559 9280/5552 log-likelihood vis 158*** 1136*** 938*** 12*** 336***a vis the precedingmodel

Summed coefficients (Model 6)No. of nontechnological 0.001acquisitions [0.054]

Absolute size of 0.0004**acquired knowledge [0.000]baseRelative size of 1.660**acquired knowledge [0.661]baseRelatedness of 2.725**acquired knowledge [1.332]baseRelatedness of 4.951**acquired knowledge [2.020]base2

No. of technological 0.145*acquisitions where [0.081]patents unavailableForeign acquisitions 0.011

[0.049]

*p , 0.05; **p , 0.01; ***p , 0.001 (one-tailed tests for hypothesized variables, two-tailed tests for controls)The table gives parameter estimates; standard errors are in brackets. Year dummies are included, but not shown.

summed coefficients forRelative size of acquiredknowledge baseare negative, indicating thatacquiring firms that are large relative to theacquirer leads to a decline in postacquisition inno-vation output for the acquirer. We also find sup-port for Hypothesis 4. As shown in Model 6 ofTable 2, the coefficients forRelatedness ofacquired knowledge baseare positive, and thatfor its squared term,Relatedness of acquiredknowledge base2, are negative. These findingssupport the argument that the relatedness ofacquisitions has a curvilinear impact on the sub-sequent innovation output of acquiring firms. Thesummed coefficients at the bottom of Table 2support the same results.

Among the control variables, the coefficientson Number of technological acquisitions wherepatents unavailablefor the 1-year, 2-year, and 3-year lags are positive but not significant. How-ever, the 4-year lag is positive and significant.The summed coefficient is positive and signifi-cant. Thus, technological acquisitions for which


we were unable to identify patents also improvepostacquisition innovation output.

Foreign acquisitions, which represents the cul-tural distance between acquired and acquiringfirms, has a nonsignificant impact on innovationoutput. Although the 1-year and 3-year laggedvariables are marginally significant but inopposite directions (positive and negative,respectively), the summed coefficient is positivebut not statistically significant (Table 2). Althoughthese results are not conclusive it appears that,on balance, foreign acquisitions neither help norhurt innovation output. This finding is consistentwith recent research on international acquisitions,which finds that acquisitions in which the acquirerand the acquired are from different countries donot result in greater postacquisition conflict, and,in fact, often lead to superior postacquisitionperformance (Weber, Shenkar, and Raveh, 1996;Very et al., 1997). Nevertheless, further researchis required to understand this issue more com-pletely.


Logemployees, the control measuring the sizeof the acquirer, was found to be positivelyassociated with patenting frequency.Diversifi-cation is negatively associated with patenting fre-quency. Prior results on the impact of diversifi-cation on innovative activity have been mixed;studies find that diversification has both a positiveand a negative impact on innovation (Cohen andLevin, 1989). Diversification can either encourageinnovation by providing a stimulus of multipleknowledge bases within a single firm and byleading to cross-fertilization of ideas, but it canalso imply a loss of focus in a given technologicalarea as research efforts are spread in multipledirections. The results of this research supportthe latter interpretation.

The Presample patentsvariable and several ofthe year dummies (not reported) were also sig-nificant in all the models. This indicates that itwas important to control for unobserved firmeffects as well as period effects in these data.The Year dummies for 198386 were negativeand significant, while all the other year dummieswere nonsignificant, relative to the omitted cate-gory1991. These results indicate that patentinghad significantly increased for this set of firmsin the later years (198791), relative to the earlieryears (198386). The nation dummy variablesreflecting acquirer nationality, Japanese and U.S.,were also positive and significant, indicating thatJapanese and U.S. firms were likely to obtainmore patents than European firms.

Sensitivity analyses

We also ran several sensitivity tests to check therobustness of the results. We reconstructed allthe knowledge base measures after separating thepatents obtained by a firm (Own Patents) fromthe patents cited by the firm (Cited Patents), andran two distinct sets of models. In the first setwe used knowledge-based measures for acquiredand acquiring firms based on only the Own Pat-ents of the firm (i.e., we excluded Cited Patents).In the second set we computed all knowledge-based measures for acquired and acquiring firmsbased on only the Cited Patents of the firms (i.e.,we excluded Own Patents). The results of theseanalyses are presented in Table 3. In the interestof brevity, we only report the summed coef-ficients and associatedt-tests for the models (thefull regression table is available from the authors).


Table 3. GEE presample Poisson regression predictingpatentst. Own-cited patents measures. Summary of thesummed coefficient results for all lagged variables

Variable Own Citedpatents patents

1 2

No. of nontechnological 0.007 0.0006acquisitions [0.057] [0.054]Absolute size of acquired 0.002***knowledge base (Own [0.001]patents)Relative size of acquired 1.632*knowledge base (Own [0.708]patents)Relatedness of acquired 2.411knowledge base (Own [3.562]patents)Relatedness of acquired 3.568knowledge base2 (Own [7.533]patents)Absolute size of acquired 0.0004***knowledge base (Cited [0.000]patents)Relative size of acquired 1.653***knowledge base (Cited [0.465]patents)Relatedness of acquired 2.752*knowledge base (Cited [1.266]patents)Relatedness of acquired 4.759**knowledge base2 (Cited [1.865]patents)No. of technological 0.196* 0.141acquisitions where patents [0.083] [0.081]unavailableForeign acquisitions 0.003 0.011

[0.003] [0.048]

Reported coefficient estimates are the sum of the four laggedcoefficients (bt1, bt2, bt3, bt4) for each variable.*p , 0.05; **p , 0.01; ***p , 0.001 (one-tailed tests forhypothesized variables, two-tailed tests for controls)

The summed coefficients in Table 3 indicate thatwith knowledge base measures based only onOwn Patents (Model 1), the hypotheses (2 and3) on absolute sizeof acquired firm andrelativesize of acquired firm were strongly supported.However, the coefficients for therelatednesshypothesis (4) carried the right signs (+ forrelatedness and2 for its square term), but theywere statistically nonsignificant. In the analysisbased on Cited Patents only (Model 2) all hypoth-eses (2, 3, and 4) were strongly supported. Thus,using only patents, or only citations to measureknowledge bases, provides rather similar resultsto using both patents and citations for the absolute


size and relative size effects. However, in termsof measuring relatedness, the patents-only meas-ures do not appear to capture relatedness as wellas the citations-based measures. A comparison ofthe results in Table 3 (based on Own Patents orCited Patents taken individually) with the resultsin Table 2 (based on combining Own and Citedpatents to provide a single measure of knowledgeelements), however, suggests that the relatednessmeasure based on both own and cited patentscollectively is more predictive than the measuresbased on either Own Patents or Cited Patentsonly.

In the results discussed above we used thePresample Patents variable as a measure of unob-served differences in the knowledge bases offirms. The patent production function literatureprovides several alternate approaches to constructindices, reflecting differences in firms knowledgestocks (Griliches, 1984; Hallet al., 1988; Hender-son and Cockburn, 1996). We constructed severalsuch indices using different assumptions and inputdata. First, we used capitalized historical R&Dexpenditures to construct a knowledge stock index(Hall et al., 1988).2 Second, we constructed anindex based on the moving average sum of R&D expenditures for the previousn years. Weused a depreciation rate of 0.20 (consistent withHenderson and Cockburn, 1996, and other studiesin this tradition) for the first approach andn =5 years for the second approach, and estimatedmodels using these indices in lieu of the R&Dvariable. We also computed the knowledge stockindex using cumulative lagged depreciated patentstocks rather than cumulated lagged depreciatedR&D, and used this as a control variable in ourregression models in place of the Presample Pa-tents measure. In additional estimation we usedthe one period lagged values of the dependentvariable in lieu of the presample variable. Theresults for all these approaches (available fromthe authors) are very similar to the earlierreported results.

As noted earlier, the correlations between the

2 Specifically, for theith firm andtth period, knowledge stock(Kit) was computed using the formulaKit = 1 2 (Kit21) + Rit,where stands for depreciation rate for knowledge capital,and Rit represents the current-period knowledge flow addition.The initial stock of capital was computed by dividing thefirst observed years flow by + g, where g is the firmshistoric rate of growth of real expenditures on R&D(Henderson and Cockburn, 1996).


three control variables,R&D, LogemployeesandPresample patentswere high. Although this highcollinearity does not bias coefficient estimates, itcan affect the stability of the estimated coef-ficients. Consequently, omitting even a few obser-vations can change the sign or the significanceof the affected variables (Greene, 1993). Toensure that our results were robust, we drew 50random samples of 90 percent of the observationsand estimated the full model for each of thesesamples. The results of this sensitivity analysisindicated that the results of our hypothesis testingwere robust.

DISCUSSION

The results of the study indicate support forour theoretical predictions. Specifically, separatingnontechnological acquisitions from technologicalacquisitions and distinguishing between techno-logical acquisitions on the basis of the absolutesize, relative size, and relatedness of the knowl-edge bases of the acquired firm helps in predictingpostacquisition innovation output. We do not findany statistically significant impact of nontechno-logical acquisitions on subsequent innovation out-put. Within technological acquisitions, absolutesize of the acquired knowledge base has a posi-tive impact on innovation output, while relativesize of the acquired knowledge base reduces inno-vation output. The relatedness of acquired andacquiring knowledge bases has a nonlinear impacton innovation output, with acquisition of firmswith high levels of relatedness and unrelatednessboth proving inferior to acquiring firms with mod-erate levels of relatedness. These findings haveimplications for theory, research, and practice.We discuss these below.

Implications for theory and research

Recent research has highlighted the role of acqui-sitions as a mechanism for the redeployment ofresources that are subject to market failure(Anand and Singh, 1997; Capron, Dussauge, andMitchell, 1998). In this study we investigated theeffectiveness of such redeployment in the case ofone set of resources that are very susceptible tomarket failure, knowledge-based or technologicalresources. Our results provide both good andbad news.


The positive outcome of this evaluation is that,when the characteristics of acquisitions areaccounted for, acquisitions improve the techno-logical performance of the acquiring firm.Although prior research concluded that acqui-sitions reduce innovative outputs, this study sug-gests that under the appropriate circumstances,even after controlling for innovative inputs suchas R&D, acquisitions can introduce a positiveshock onto innovation output. This finding is alsoconsistent with the sheer volume of acquisitionactivity in the high-technology sector that sug-gests that managers also view acquisitions as amechanism for accessing technology.

The results of this study are also importantfrom a broader economic perspective. The rapidgrowth of technical knowledge in the past fewdecades has meant that building and maintainingexpertise in multiple technologies is difficult foreven the largest corporations (Granstrand and Sjo-lander, 1990; Arora and Gambardella, 1994). Yet,bringing together different streams of knowledgeis becoming an important precondition to success-ful innovation in many industries (Grant, 1996;Powell, Koput, and Smith-Doerr, 1996). Thisneed for increased differentiation in the develop-ment of knowledge and increased integration inits application has given new importance to tech-nology markets (Demsetz, 1991; Grant, 1996).Our results indicate that the process of obtainingtechnological assets from external sources andmatching them with internally developed assetsto enhance their productivity can work, at leastinsofar as the frequency of innovation outputis concerned.

The above arguments highlighted the brighteraspects of the results. Another perspective onthese results, however, draws attention to theirless positive facets. Prior research has identifiedthree kinds of deficiencies in the context oforganizational learning: hubris, or underestimatingthe likelihood of failure or the difficulty of thetask at hand; overexploration, or venturing intodomains of completely new knowledge; and,overexploitation, or focusing only on the immedi-ate neighborhood of the well known and under-stood and overlooking more distant options(Levinthal and March, 1993). Our data and resultsindicated evidence of all three kinds of errors. Tothe extent that relative size serves as a measure ofthe relative difficulty of integration, it appearsthat underestimating the magnitude of the inte-


gration task is not uncommon. Further, the vari-ation on relatedness, and the curvilinear relation-ship that we identified between acquisitionrelatedness and innovation output, suggests thatin selecting acquisitions managers err in bothdirections, acquiring both businesses that are onlydistantly related or unrelated to the existing busi-ness, as well as those that are too closely relatedto the current business. The problems of acquiringunrelated businesses are well documented and ourresults further reinforce those lessons (Singh andMontgomery, 1987). Additionally, our findingshelp to highlight errors in the opposite direction.Managers do make mistakes in picking acqui-sitions that are too closely related to their extantdomains, and these mistakes are penalized withpoorer performance. Thus, we find some empiri-cal evidence of competency traps (Levinthal andMarch, 1993).

Implications for measurement

Our findings also have implications for the litera-ture on acquisitions in general. Managerial hubrisand agency problems (Roll, 1986; Hayward andHambrick, 1997), imitation effects (Haunschild,1993), inappropriate application of learning(Haleblian and Finkelstein, 1999), and underesti-mating the process impediments to postacquisitionintegration (Jemison and Sitkin, 1986; Hittet al.,1996; Singh and Zollo, 1997) are all valid andcomplementary explanations for why acquisitionsconsistently fail to help the acquiring firm. Thefindings of this paper suggest an additional expla-nation, one that is perhaps more sympathetic tomanagerial motivations and decisions. Evaluatingall acquisitions on the same performance metric,for instance financial performance, may not beappropriate. Acquisitions motivated by differentobjectives may differ in their timing and modeof impact on firm performance.

The recent literature has provided several excit-ing ways of using patent data to measure theconstruct of knowledge (Mowery, Oxley, and Sil-verman, 1998; Jaffeet al., 1993; Stuart andPodolny, 1996). We build on these studies bypresenting an additional set of measures andapplying them to the context of acquisitions. Thepatent-based approach to knowledge measurementused by this study, and similar approaches usedin prior studies (e.g., Stuart and Podolny, 1996),have several strengths. A primary strength of


such approaches is that by using information onindividual elements of knowledge theseapproaches make poss

Date post:	17-Nov-2015
Category:	Documents
Upload:	mailitin
View:	216 times
Download:	0 times

Kati La Acquisitions Academic Research

Documents