Do Corporate Mergers Bring about New Combinations of Knowledge?

Transcription

1 . Do Corporate Mergers Brng about New Combnatons of Knowledge? - Emprcal Evdence from Patent Data - Atsush INUZUKA The Unversty of Tokyo Research Center for Advanced Scence and Technology, Komaba, Meguro-ku, Tokyo JAPAN Orgnal Source: Internatonal Journal of Knowledge Management Studes Vol.3, Nos.1/2, 2009, pp Abstract. The aftereffects of corporate mergers, partcularly whether corporate mergers facltate new combnatons of knowledge, were nvestgated usng patent applcaton data for two corporate mergers n the Japanese chemcal ndustry. The results show that new combnatons are more lkely to occur after mergers and that the choce of a new combnatons strategy can dffer between companes. Ths strategc choce has an mpact on the aftereffects of corporate mergers, and the dfference n strategc choce mght depend on whether power and ntatve resde n the same premerger frm. Keywords. corporate merger, new combnaton, patent applcaton, nventor, after-effect, effcency, effectveness, collaboraton, technology map, selecton and concentraton. Reference to ths paper should be made as follows: Inuzuka, A. "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp Bographcal Notes: Atsush Inuzuka receved a Ph.D. n Knowledge Scence from Japan Advanced Insttute of Scence and Technology. He s currently a project assocate professor at Research Center for Advanced Scence and Technology, The Unversty of Tokyo. He has over fve years of ndustry experence n product development. Hs research nterests nclude knowledge management, organzatonal behavor, and organzatonal psychology.

2 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp INTRODUCTION Recent lterature on the role of frm-specfc resources has ndcated that knowledge s the most strategcally sgnfcant resource of a frm. A partcular vew of a frm, namely, the knowledge-based vew, emphaszes the capacty of a frm to ntegrate tact knowledge (Nonaka, 1994; Grant, 1996; Conner and Prahalad, 1996). As ths resource s usually dffcult to mtate and s socally complex, heterogeneous knowledge bases comprse the major determnants of sustaned compettve advantage and performance. However, as Grant (1996) and other researchers have shown, the knowledge-based vew s not yet a theory of a frm at some respects. If knowledge and ts management are of such great mportance, knowledge strateges should play a crtcal role n a frm s strategc plannng. Of the numerous strategc choces that a frm can make, one of the most drastc ones s whether to merge wth another frm. However, a corporate merger can be lkened to a double-edged sword. If t s carred out successfully, a frm can gan access to new resources and, by the redeployment of resources, ncrease ts revenue and reduce costs, whch wll result n ts performance mprovng sgnfcantly. However, as a worst case, a corporate merger smply creates turmol and results n poor performance. In fact, many emprcal studes have shown that there s no guarantee that the aftereffects of a corporate merger wll be postve (Lubatkn, 1983; Mueller, 1997). From a knowledge-based perspectve, mergers can be vewed as combnatons that broaden a frm s knowledge base. A broader knowledge base may provde many opportuntes for recombnng the exstng elements of knowledge nto new syntheses. In other words, the emergence of new combnatons of knowledge that lead to nnovaton s one of the mportant motves for sufferng the stress of a corporate merger. However, emprcal fndngs suggested that the mpact of mergers depends on the characterstcs of the knowledge of the merger and merged frms (Ahuja and Katla, 2001; Mowery, Oxley, and Slverman, 1996, 2002; Sampson, 2004). If the merger s successful, the frm should be able to generate new knowledge that could not have been generated n ts absence. If t s unsuccessful, the frm s knowledge base could be runed. Although knowledge s seen as the most mportant asset of a frm, untl now, hardly any research from the knowledge-based vew has been conducted on how the aftereffects of corporate mergers have been successfully acheved, especally at the mcro level. It s possble that the lack of emprcal work documentng the aftereffects of mergers s partly due to the dffculty of measurng the technologcal and other capabltes of frms (Mowery, Oxley, and Slverman, 1996). Owng to the dffculty nvolved n measurng frms technologcal capablty, ths paper focuses on the usefulness of patent applcaton data. Patents have three partcular advantages: they are systematcally compled, they have detaled nformaton, and they are 1

3 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp avalable contnuously across tme (Almeda, 1996) 1. Consderng these advantages, many scholars have obtaned useful results about frms nnovatve actvtes by usng patent data (Pavtt, 1985). Patents, because of ther nature, represent technologcal advances and nclude speculatons about how a partcular technology may be appled. Therefore, parents enable a detaled understandng of how and to what extent each technology was used before and after a corporate merger. Therefore, I nvestgated the aftereffects of corporate mergers, usng patent data, and put forth two questons: (1) Do corporate mergers facltate new combnatons of knowledge? (2) If they do, how and why does a frm choose a new combnaton strategy? To answer these questons, I created a new measure for evaluatng new combnatons of knowledge and tested three hypotheses. The remander of ths paper s organzed as follows. In the next secton, I brefly survey the academc contrbutons wth respect to ths ssue and present the three hypotheses. The followng secton descrbes the analytcal procedure and the varables used for testng the hypotheses. The results are presented and dscussed n the next two sectons. The last secton summarzes the key ponts, descrbes several lmtatons, and provdes drectons for future research. 1 It should be noted that patents are not perfect ndcators of frms nnovatve actvty. However, they may serve as an nterestng montorng devce to dentfy the man lnes and trends, and even under specfc condtons, enables us to analyze R&D processes n greater detal (Engelsman and Raan, 1992). Three potental lmtatons of patents were ponted out by Patel and Pavtt (1997): (1) patents do not measure the extent of the frm s external technologcal lnkages, (2) patents measure only codfed knowledge, and (3) patentng does not fully measure competences n software technology. 2

4 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp BACKGROUND New Combnatons Evdenced by Patent Data Perhaps the most sgnfcant contrbuton of Schumpeter was to defne nnovaton as new combnaton (Schumpeter, 1934). Innovaton s consdered to be accomplshed by creatvely combnng two (or more) prevously unrelated peces of knowledge to form a qualtatvely new pece of knowledge. For a new dea to come nto exstence, dfferent peces of knowledge have to come together. Ths mples personal nteracton between ndvduals because most knowledge s held wthn them. In ths respect, a corporate merger provdes many new opportuntes for dfferent peces of knowledge to come together. It s expected to reduce the psychologcal and/or economc cost of nteracton between the members of the two formerly separate frms. Addtonally, as each mergng frm has a specfc technologcal portfolo, the merger should facltate dversfcaton and the fllng of gaps between the portfolos. Ths mples that the possblty of new combnatons occurrng n a frm should be hgher after the merger. H1: Corporate mergers facltate new combnatons of knowledge. To test ths hypothess emprcally, a relable measure for new combnatons of knowledge s requred. If a narrow defnton of knowledge s permtted, patent statstcs provde a proxy for knowledge and are a good resource for nvestgatng ths problem 2. The patent systems of major countres use the Internatonal Patent Classfcaton (IPC) code, whch had ts orgns n the Councl of Europe s 1954 European Conventon on the Internatonal Classfcaton of Patents for Inventon. Under ths scheme, each patent s assgned one or more IPC codes correspondng to the patent s contents. Snce the codes correspond to technology areas, the relatedness between felds of technology can be estmated by analyzng the co-occurrences of the codes assgned to each patent (Jaffe, 1986; 1989; Engelsman and van Raan, 1992; Bresch, Lsson, and Malerba; 2003, 2004). For example, f two codes co-occur frequently, the combnaton of the two correspondng technology felds s probably not new (.e., the combnaton s central n the frm s total portfolo). Smlarly, f they co-occur nfrequently, ther combnaton s thought to be new (.e., the combnaton s margnal n the frm s total portfolo). We can thus estmate the extent of newness of a combnaton of knowledge on the bass of the frequency of the IPC code pars assgned to 2 There s a dscusson that patent statstcs show dfferent aspects of ndustral nnovaton (Pavtt, 1982). 3

5 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp patents. New Combnatons as a Strategc Choce A frm s strategc choces strongly affect the behavors of the nventors t employs; therefore, focusng on the nventors s a reasonable way to assess the effects of a strategc choce. Jaffe, Trejtenberg, and Henderson (1993), for example, analyzed patent ctaton data by the U. S. patent offce and found that ctatons were more lkely to occur among patents nvented by persons from the same regon. Such a localzaton effect of knowledge was also confrmed by many scholars (Almeda, 1996; Almeda and Kogut, 1999; Stolpe, 2002; Song, Almeda, and Wu, 2003; Thompson and Fox-Keanm, 2004). Narn and Bretzman (1995) and Ernst, Lepten, and Vtt (2000) used patent data to verfy Lotka (1926) s law,.e., the number of hghly productve scentsts s a relatvely small fracton of all scentsts. In a frm, there s a relatvely small number of hgher-performance nventors and a large number of lower-performance nventors. The hgher-performance nventors are assumed to be the key person n the frm s research actvtes. Besdes developng new nventons, they also nteract wth each other to fnd promsng combnatons of the knowledge that they are creatng. Snce ther role ncludes developng a large number of nventons as well as those that are unque (pursung unque combnatons of knowledge), we are led to the followng hypothess. H2: Inventors pursung new combnatons of knowledge are characterzed as hgher performers. However, the pursut of new combnatons nvolves greater rsks. Frst of all, a frm confronts hgher uncertanty because the probablty of the occurrence of new combnatons s relatvely low. In other words, new combnatons generally do not occur n the core area but n the frnge areas of the frm. Snce new combnatons are unexpected, the probablty of commercal success would be lower. Due to ths low probablty, a frm wll encourage ts nventors to develop many nventons regardless of the probablty of ther success. Ths can be desgnated as a he who shoots often, hts at last polcy that compensates for lower effectveness. Ths dscusson leads to the followng hypothess. H3: A frm pursung new combnatons of knowledge as a strategc choce after a merger can be characterzed as havng hgher effcency but lower effectveness. 4

6 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp ANALYTICAL PROCEDURE Case Selecton The three hypotheses posted above were tested usng data on patents fled at the Japanese Patent Offce (JPO). Although Japanese patent data may be based, they have some advantages that permt vald analyses. Frst and foremost, the Japanese two-step procedure for grantng patents enables a reasonable assessment of an nventor s performance. Addtonally, owng to the detaled and ntegrated system, they are very helpful for a long-term analyss (both characterstcs are fully descrbed n later). Owng to these reasons, the Japanese patentng system provdes extensve and reasonable data for the followng analyss. The cases for the testng of the hypotheses were systematcally selected. Frst, the total number of patent applcatons fled durng the fve years pror to the merger was extracted of all horzontal M&As (a type of merger of two companes that are n drect competton and share almost the same product lnes and markets) from 1988 to 2002, summarzed n the Databook of M&As n Japanese Frms by RECOF Co. Ltd., n the Tokyo Stock Exchange Market. The dentfcaton and patent data used hereafter was taken from the PATOLIS database. The database contans extensve nformaton about patent applcatons fled n Japan, e.g., (a) the name and address of the applyng frm and nventors, (b) the date on whch the applcaton was fled at the JPO and nformaton on ts after status, and (c) the IPC/FI code(s) assgned by the patent examner. After dentfyng the number of patent applcatons of each corporate merger, the cases were selected as follows. Consderng ths paper s purpose, corporate mergers that are not n technologcal areas, for example, retal or the bankng sector are not consdered approprate for the case. Also, n order to obtan accurate results, the case should have many patents fled by the merger and merged frms, and ther proporton should preferably be close to equal. Therefore, I set three crtera for case selecton: (1) the total number of patent applcatons fled by both frms durng the fve years pror to the merger should be more than 5000, (2) the rato of the number of applcatons fled between the two frms should not exceed 1:2, and (3) the name of the nventor s frm was mostly confrmed on the patent documents (descrbed below). Two cases met all the three crtera: Mtsubsh Chemcals and Mtsu Chemcals (Table 1). Both cases were n the same chemcal ndustry. Fortunately, ths enables a comparatve analyss based on almost the same condtons. Although both cases occurred n large Japanese busness groups, the merger and merged frms dd not have a strong busness relatonshp pror to the merger. In fact, they were rvals wth a contentous relatonshp. 5

7 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp Table 1. Cases for analyss frm name (after merger) merger-sde frm merged-sde frm date of merger Mtsubsh Chemcals Mtsubsh Kase Kogyo Mtsubsh Yuka 1994/10/1 Mtsu Chemcals Mtsu Sekyu Kagaku Kogyo Mtsu Toatsu Kagaku 1997/10/1 number of patents fled durng the fve years pror to the merger 7654 (1989/10/1-1994/9/30) 6072 (1992/10/1-1997/9/30) number of patents fled durng the fve years after the merger 6650 (1994/10/1-1999/9/30) 4981 (1997/10/1-2002/9/30) Identfyng Inventors Under the patent flng system n Japan, applcatons are publshed n a patent publcaton bulletn (a patent journal) wthout a search report 18 months after the date of flng the applcaton. The patent journal contans nformaton about nventors, as shown n Fgure 1. From the example, we can determne that nventor 3 Taro Suzuk worked for frm X and that Hanako Tanaka worked for frm Y on the date of flng the applcaton. (71) applcant(s) X-company Shnjyuku-ku, Tokyo. (72) nventor(s) Taro Suzuk In X-company, Shnjyuku-ku, Tokyo. Hanako Tanaka In Y-company, Shbuya-ku, Tokyo. Fgure 1. Example of patent nformaton n Japanese patent journal Usng ths nformaton, I dentfed whch of the frms (merger frm or merged frm) each nventor worked at before the merger. Hereafter, I wll use A to represent the merger frm and B to represent the merged frm (.e., Mtsubsh Kase Kogyo and Mtsubsh Yuka are referred to as frm A and frm B, respectvely). I examned all the patent documents fled by both the frms durng the fve years before the merger and lsted the nventors who could be lnked to ether of the frms. Two lsts were created. One was created for the merger frm (lst A) and the other, for the merged frm (lst B). I excluded the nventors who had provded nformaton other than the frm s name n the address feld (the cases where the addresses of nventor s home, afflate frm, jont research frms, unversty, etc. were mentoned). 3 In Japanese patent law, nventors are defned as the person(s) who practcally contrbuted to the nventon, excludng just managers, assstants, and sponsors. 6

8 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp However, the nventors whom I could lnk to any one frm through one or more patents and not by other patents, were regarded as havng worked for that frm and ther names were added to the lst 4. Thereafter, usng the same data, I checked whether each person on lst A and each person on lst B had fled at least one patent applcaton durng the fve years after the merger and elmnated those who had not fled any patent applcaton. The lsts, then, contaned only those nventors who had fled one or more patent applcatons durng the fve years both before and after the merger. The followng analyss uses the patent applcatons fled by the nventors on the two lsts. Varables of Inventors I ntroduced sx varables to desgnate nventors characterstcs. Hereafter, S s the set of patent applcatons fled durng the target perod (fve years before or after the merger). Further, P, belongng to set S, s the set of patent applcatons on whch nventor p s named. (1) New combnatons ndex for each nventor (new combnatons) Snce the newness of combnatons corresponds to a lower probablty of co-occurrence between the technology felds, the degree of nnovatveness s consdered to be the nverse of the number of combnaton occurrences. The assumpton made s that the frequency wth whch two classfcaton codes are jontly assgned to the same patent document corresponds to the strength of the knowledge relatonshp (Bresch, Lsson, and Malerba, 2003). As descrbed above, IPC codes, whch are assgned to patent documents by patent examners, are useful for estmatng ths frequency. However, snce the codes are revewed perodcally (modfed slghtly every fve years), they are not sutable for long-term studes. Therefore, I used the FI code, whch s the Japanese orgnal code havng ts orgn n the IPC, and havng almost the same classfcatons as the IPC, as an alternatve. The man dfference between IPC and FI code s that the latter s updated frequently by the JPO n order to be consstent wth other patent codes to enable long-term searches. Therefore, the FI s more sutable for tracng frms technologcal areas over the long term. The FI has the same herarchcal structure as the IPC,.e., class, subclass, man group, 4 Ths procedure has certan lmtatons. Frst, f there were nventors wth the same famly and personal names n both frms A and B, I could not strctly dentfy whch frm the nventors belonged to. Ths problem was almost neglgble because there was only one nstance for each case (the nventor was elmnated from both the lsts). However, t s possble that one nventor had been lsted twce because the nventor mght have changed hs or her famly name due to marrage durng the target perod. 7

9 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp and subgroup. I used the hgher level, whch s represented by an alphabetcal secton symbol followed by a two-dgt class symbol number (e.g., C07 corresponds to organc chemstry ). Usng ths classfcaton, I frst estmated the degree of newness by the frequency of co-occurrence between the FI class codes for the patent applcatons fled durng the target perod. The formula used for the calculaton s derved from Shannon s nformaton theory, accordng to whch, the less lkely the event (.e., co-occurrence) s to occur, the greater the value. The degree of newness between FI classes j and k and the one of FI class j are estmated by the followng equaton. f newness j, k fcount ln 1, S f newness j, k fcount fcount 1 j ln S 1 f S S j 1, where fcount represents the number of unque FI classes for patent. f(j,k), and f(j) takes bnary values: 1 means co-occurrence between FI class j and k ( fcount 1) and occurrence of FI class j ( fcount 1) for patent-.: 0 means no (co-)occurrence. After calculatng the newness between the FI classes, I assgned a new combnaton ndex to each nventor. It s a weghted average that accounts for exam rato (descrbed below). The ndex s larger for nventors wth patent applcatons n technologcal felds wth a low probablty of (co-)occurrence. In other words, the less the probablty of (co-)occurrence n technologcal areas n whch the nventor generates patents, the larger s the ndex for hm or her. new _ combnatons j k, jk newness( j, k) p P f j, k n fcount fcount 1 newness( l) f l P 1 n where n represents the number of unque nventors for patent-. l P n, 8

10 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp (2) Inventors performance (patents, partal patents, exam rato) Inventors performance s measured n terms of effcency and effectveness. Effcency s related to the number of patent applcatons for the nventor. Credt s assgned n two ways: An nventor gets full credt (a patent ) for an applcaton and gets partal credt (a partal patent ) for an applcaton f more than one nventor s named; for example, f three nventors are named, each s credted for one-thrd of the patent (Narn and Bretzman, 1995). Effectveness s related to the patent examnaton rato for the nventor ( exam rato ). In Japan, a request for examnaton must be made to the JPO (the request must be made before three years have passed snce the applcaton flng date 5 ). Such a request s not made for all applcatons for several reasons. It may be decded that the patent s not mportant after the applcaton s fled, or that dsclosng the patent s contents s suffcent for buddng the novelty of related patents to be granted. Ths judgment process by a frm mples that the rato of patents for whch a request s made can be used as a measure of an nventor s performance on the bass of qualty. Hence, I consdered ths rato for each nventor as a proxy of hs or her effectveness and calculated t n terms of a weghted average of the number of co-nventors. For example, suppose an nventor has one patent applcaton for whch a request has been made along wth one other person and another patent applcaton for whch a request has not been made along wth two other people. In ths case, the exam rato for the nventor s calculated as a weghted average: (1/2 + 0/3) / (1/2 + 1/3) = 0.6, not smply 0.5. Although some mght argue that the rato of patent regstratons s a better ndcator of an nventor s performance, because regsterng of a patent takes many years after the applcaton s fled (e.g., Jaffe, et al., 1993; Hall, 2002), most of the patents n the two cases mentoned here stll have ther examnatons pendng. Smlarly, many researchers have ponted out that a patent s strength s best measured by the number of patent ctatons to t (e.g., Narn, Carpenter, and Woolf, 1984; Albert et al., 1991), however, t also takes consderable amount of tme to measure the number of targeted patent ctatons. Therefore, I used only exam rato as the metrc for an nventor s effectveness. 5 The perod was changed from seven to three years as of 2001/10/1. Snce I am nterested n applcatons fled from 1992/10/1 to 2002/9/30 n the Mtsu Chemcals case, some of them mght not have had a request for regstraton by the frm. 9

11 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp patents p exam rato p P partal patent s 1 p P P d P n 1 n 1 n (3) Average number of co-nventors per patent for each nventor (co-nventors) I calculated the average number of co-nventors (ncludng the targeted nventor) for each patent by nventor p, usng the followng equaton. co _ nventors p P P n 1 (4) Weghted average number of FI classes per patent for each nventor (weghted FIs) Smlar to Lerner s (1994) work, I calculated the weghted average number of dfferent FI classes whch s weghted by the number of co-nventors per patent usng the followng equatons: weghted FIs p P fcount P 1 n n 10

12 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp RESULTS Aftereffects of a Corporate Merger The average of (partal) patents and the exam rato of an nventor durng the fve years before and the fve years after a merger, and the dfference between them wth t-statstcs 6 are summarzed n Table 2. The table also shows the ndex of new combnatons and weghted FIs. The latter two are hgher for after the merger than before the merger n both cases, mplyng that a corporate merger tends to broaden the perspectve of research actvty. In both cases, the merger appears to have played a postve role n promotng the encounterng of new knowledge wthn the company, whch can lead to nnovaton. Ths supports H1. It mples that mergers ncrease the probablty of encounterng people wthn the frm who have knowledge of dfferent technologes or dfferent ways of thnkng. Consequently, nventors are more lkely to fnd new ways of coordnatng or combnng technologes. In terms of nventor performances, the number of (partal) patents declnes sgnfcantly after a merger (Mtsu Chemcals case s not sgnfcant but at the level p <.1) probably due to the reassgnment of technology areas for researchers after the merger as a part of the restructurng. In contrast, nventors effectveness (exam rato) mproves sgnfcantly for Mtsubsh Chemcals as shown by the hgher exam rato, whle t declnes slghtly for Mtsu Chemcals. Table 2. Characterstcs of nventors frm name Mtsubsh Chemcals (N=1522) Mtsu Chemcals (N=1306) perod patents fled patents partal patents exam rato new combnatons co-nventors weghted FIs 5 years before merger years after merger dfference (after-before) *** * *** *** *** 5 years before merger years after merger dfference (after-before) * *** * *** Note : A pared sample t-test s used ( * p <.05, *** p <.001). Effect of New Combnatons as a Strategc Choce I used three regresson models to estmate whether the dfference n the nventor s 6 Owng to the large sample, the average of pared dfference for the exam rato s also regarded as t dstrbuted. 11

13 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp effectveness (exam rato) between the two frms could be confrmed after controllng the effects of other varables and determne the extent to whch these characterstcs affect the nventors performance. The frst model defnes three characterstcs as ndependent varables: new combnatons, co-nventors, and weghted FIs. The second and thrd regresson models have one or two dummy varable(s), respectvely for nventor classfcaton. The classfcaton of nventors was carred out by the followng procedure. Usng lst A and lst B, whch I had prepared, I checked each patent applcaton fled durng the fve years after the merger to see whether t named nventors from frm-a and frm-b. Then, nventors who shared at least one patent applcaton wth someone from the other frm were classfed as collaborators; the remanng nventors, as non-collaborators. On the bass of ths classfcaton, the dummy varable n model 2 was set to 1 for non-collaborators and to 0 for collaborators. In Model 3, dummy A was set to 1 for non-collaborators from the merger frm, and dummy B was set to 1 for non-collaborators from the merged frm, and both were set to 0 for collaborators. I used the ordnary least squares (OLS) regresson for partal patents as the dependent varable (I took the natural log of the number of partal patents because the dstrbuton of partal patents was hghly skewed) and used the probt regresson for exam rato as the dependent varable. The results are summarzed n Tables 3 and 4. Table 3. Results of regresson analyss (Mtsubsh Chemcals) dependent varable ln partal patents ln partal patents exam rato exam rato (5 years before merger) (5 years after merger) (5 years before merger) (5 years after merger) ndependent varable new combnatons model *** model *** model ** model *** model *** model *** model model model * model *** model *** model *** co-nvetors *** *** *** *** *** *** weghted FIs *** ** ** *** *** *** ** ** ** ** ** *** dummy A+B *** *** dummy A *** *** dummy B *** *** * ntercept *** *** *** *** *** *** ** ** ** R F / χ *** 59.1 *** 50.2 *** *** *** *** Note : N=1522. Numercal varable represents standarzed coeffcent estmated by OLS regresson analyss / probt analyss ( * p <.05, ** p <.01, *** p <.001). Table 4. Results of regresson analyss (Mtsu Chemcals) dependent varable ln partal patents ln partal patents exam rato exam rato (5 years before merger) (5 years after merger) (5 years before merger) (5 years after merger) ndependent varable new combnatons model *** model *** model *** model *** model *** model *** model *** model *** model *** model * model ** model * co-nvetors *** *** *** *** *** *** *** *** *** weghted FIs *** *** *** *** ** ** *** *** *** * ** * dummy A+B *** *** ** dummy A *** *** *** dummy B *** *** ntercept *** *** *** *** *** *** * ** R F / χ *** 46.1 *** 40.5 *** 34.0 *** *** 82.5 *** Note : N=1306. Numercal varable represents standarzed coeffcent estmated by OLS regresson analyss / probt analyss ( * p <.05, ** p <.01, *** p <.001). 12

14 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp The results ndcate that new combnatons and co-nventors had a sgnfcantly negatve effect on nventors effcency (.e., partal patents), whle weghted FIs had a sgnfcantly postve effect both before and after the merger. Ths mples that nventors who work relatvely ndependently and cover many but related technologcal areas are lkely to be more effcent. All the results show that new combnatons negatvely affect nventors effcency, whch contradcts H2. Furthermore, the negatve sgns for the dummy varables mean that non-collaborators are less effcent than collaborators. In other words, collaborators are more effcent than non-collaborators, whch mply that collaborators must play key roles n developng nnovaton. Moreover, the ncrease n the absolute values of the dummy varables after the merger means that collaborators play an even more mportant role after the merger. The results for effectveness (.e., exam rato) dffer somewhat between the two cases. For example, the sgn for co-nventor s negatve for Mtsubsh Chemcals, whle t s postve for Mtsu Chemcals. Moreover, for Mtsu Chemcals, the coeffcents for co-nventors are sgnfcantly greater after the merger. Ths suggests that Mtsu Chemcals emphaszed more on collaboratve work among researchers after the merger. More mportantly, the values of dummy varables (non-collaborators) for Mtsu Chemcals were sgnfcantly greater after the merger, whch s n contrast to for the case of Mtsubsh Chemcals. Ths ndcates that technologcal collaboraton between the merger and merged frms was less successful after the merger for Mtsu Chemcals than for Mtsubsh Chemcals. The sgns of the new combnatons are all negatve, whch means that pursung new combnatons does not result n effectve nventons. Ths, together wth the results descrbed above, leads to the concluson that H2 s not supported n terms of both effcency and effectveness. In addton, careful analyses of the results confrm the dfferences between the two cases. In the Mtsubsh Chemcals case, the values for new combnatons are lower after the merger, whle n the Mtsu Chemcals case they are hgher. Ths ndcates that Mtsubsh Chemcals followed a selecton and concentraton of technology strategy after the merger n terms of effectveness. In contrast, Mtsu Chemcals emphaszed new combnatons rather than selecton and concentraton of technology. Ths strategc contrast affects the roles of collaborators as well as non-collaborators. From Table 3, the effectveness of Mtsubsh Chemcals collaborators and non-collaborators dd not change much after the merger. In contrast, as shown n Table 4, the collaborators of Mtsu Chemcals worked less effectvely than the non-collaborators after the merger. These cases mply that f a frm makes new combnatons ts strategc choce, effectve nventons mght not be expected for collaboratve work. Also, the results shown on the left sde of Table 4 mply that the aftereffect for collaborators n terms of effcency s hgher after the merger (the values for non-collaborators sgnfcantly declne after the merger), showng that collaborators are lkely to produce more nventons than non-collaborators after a merger. In short, Mtsu Chemcals apparently adapted a he who shoots often, hts at last polcy for collaborators to compensate for ther lower effectveness, whch supports H3. 13

15 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp DISCUSSION Thus far, I have dscussed the aftereffects of corporate mergers, usng patent applcaton data. The fndngs show that Mtsubsh Chemcals and Mtsu Chemcals made dfferent strategc choces for flng patent applcatons. In ths secton, I consder how ther technology areas affected ther strategc choces because ths choce s strctly rooted n the technology area n whch a frm stands. Although new combnatons after a merger demand a certan level of technologcal change, n general, such changes cannot be easly realzed. In fact, there s consderable evdence that a frm s nnovatve actvtes are cumulatve n the sense that ts technologcal specalzaton tends to reman stable over lengthy perods of tme (Cantwell and Anderson, 1996; Malerba, Orsengo, and Peretto, 1997; Malerba and Orsengo, 1999; Cefs and Orsengo, 2001; Cefs, 2003; Bresch, Lsson, and Malerba, 2003). Ths s because there s nerta that renforces specalzaton and techncal nterdependence (Dos, 1982; Patel and Pavtt, 1997); thus a frm s technology area generally changes gradually over tme. Ths means that people must take strong ntatve for new combnatons to occur. I used a technology map to dentfy who took the ntatve that led to new combnatons. I created the map, usng correspondence analyss, a method of factorng categorcal varables and dsplayng them n a feature space. Frst, I determned the total number of patent applcatons 7 n each FI class for each flng year, usng the formula gven below (I counted the number of partal patents for each FI class for each flng year). Subsequently, correspondence analyss was conducted wth FI class as a row category and the classfcaton of nventor (lst A, lst B, or both) for each flng year for both cases as a column category. Fgure 2 shows a scatter plot of ponts for each FI class. The dstance between FI classes s shorter when they have relatvely the same rato of patent applcatons among column categores. It must be emphaszed that the axes do not represent specfc parameters; the relatve postonng tself s essental. contrbute number of patents j, t f( j) n fcount P t P(t) :Set of patent applcatons fled by target nventors n perod t 7 Snce the purpose s to analyze the propertes of patent data as a proxy of the frm s nnovatve actvtes, patent applcatons were chosen nstead of regstered patents. 14

16 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp B27 E03 E02 D21 E21 C40 H05 E01 F A43 F21 A44 B26 B07 B67 B63 C04 A62 A01 D03 B41 C09 B28 C07 A45 F17 A24 G02 F23 B42 F25 B24 B81 C01 G12 A42 F28 B68 B82 A61 C11C23 E04F02 F27 B32 A63G09 B05 E06 B01 C03G11 B60 B29 B30 C05 C08 B66 C25 D06 G06 C02 F26 B22 B02 0H01 H04 F01 G A41 B43B65 B25 A23 D02 G05 G03 B03 G21 B08 B04 B09 F16 G01 C12 C22 A47 F04 D01 F03 B44 C10 G08 C13 D04 B23 H02 B62 C30 A21 C21 F15 A46 B21 G04 B31 D07 B61 H03 B06 B Fgure 2. Technology space Next, I plotted the nventors technologcal centrod postons n the same technology space (Fgures 3 and 4). The patent applcatons fled before the merger by nventors for fle A (merger frm) and fle B (merged frm) are labeled A and B, respectvely. The applcatons fled after the merger by nventors for both fle A and fle B are labeled C. The numbers n parentheses ndcate the flng perod,.e., ( 1) mples flng n the year pror to the merger, and (1) mples flng n the year after the merger. I added ellpses to cover all the flng perods for the nventors classfcaton. 15

17 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp Mtsubsh Chemcals (after merger) C(3) C(4) C(2) C(1) A(-1) 1.2 C(5) A(-2) A(-3) A(-4) B(-3) B(-2) B(-5) A(-5) Mtusbsh B(-4) Kase Kogyo B(-1) Mtsubsh Yuka (merger-sde) (merged-sde) -1.2 Fgure 3. Technologcal poston (Mtsubsh Chemcals) B(-1) B(-3) 1.2 Mtsu Toatsu Kagaku (merged-sde) B(-2) B(-4) B(-5) C(1) C(3) C(5) C(4) Mtsu Chemcals C(2) (after merger) A(-5) A(-4) A(-3) A(-2) A(-1) Mtsu Sekyu Kagaku Kogyo (merger-sde) -1.2 Fgure 4. Technologcal poston (Mtsu Chemcals) 16

18 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp Usng these fgures, we can estmate the frm that took the ntatve n terms of collaboraton after the merger. For example, Fgure 3 shows that patent applcatons from the merger frm (A), are located n the area plotted n the fourth quadrants, and those from the merged frm (B) are located n the thrd quadrant. The applcatons fled by nventors from both frms after the merger (C) are located n the frst and fourth quadrant and closer to those from the merger frm (A) than those from the merged frm (B). Ths mples that the ntatve after the merger was taken manly by the nventors from the merger frm n the Mtsubsh Chemcals case. In contrast, n the Mtsu Chemcals case (Fgure 4), the postons of the applcatons for the merged frm (B) and for both frms after the merger (C) partly overlap, ndcatng that the ntatve was taken manly by the nventors from the merged frm. Supportng ths, the famous Japanese busness newspaper, Nkke, reported the followng on the day of the merger, producng a cash flow by the stable petro-chemcal busness of Mtsu Sekyu Kagaku Kogyo (the merger frm), and nvestng t n Mtsu Toatsu Kagaku (the merged frm) s fne-chemcal busness that has hgh growth potental the newly formed Mtsu Chemcals develops such a scenaro. Ths fndng provdes one explanaton for makng a dfferent strategc choce for new combnatons. Usually, the merger frm has much strong power than the merged frm. Therefore, the results for the Mtsubsh Chemcals case mply that a merger frm can adopt a top-down approach to mplement a selecton and concentraton strategy because the merger frm s more powerful and takes strong ntatve. On the other hand, the results for the Mtsu Chemcals case ndcate that f ntatve s lackng n the merger frm, the resultng maldstrbuton of power and ntatve can make the adopton of a top-down approach dffcult. Ths forces the frm to choose a new combnatons strategy nstead of a selecton and concentraton one. These cases mply that f the selecton and concentraton strategy s to be appled, both power and ntatve should resde n the same premerger frm. 17

19 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp CONCLUSION I began ths paper by posng two questons: (1) Are new combnatons of knowledge facltated by corporate mergers? (2) How and why does a company make a strategc choce for new combnatons of knowledge? My analyss of the two cases ndcated that new combnatons were more lkely to occur after mergers. However, the focus on the new combnatons strategy revealed a dfferent story. Mtsubsh Chemcals concentrated on ther technology area for flng patent applcatons after the merger. In contrast, Mtsu Chemcals encouraged ts nventors, especally the collaborators, to produce many nventons regardless of ther effectveness. Further analyses revealed that the choce of a new combnaton strategy depends on whether or not power and ntatve resde n the same premerger frm. Moreover, strategc choce seems to have a sgnfcant mpact on the aftereffects of a corporate merger. These evdences have some manageral mplcatons. Frst of all, although a corporate merger facltates new combnatons of knowledge, pursung new combnatons n nnovatve actvtes does not necessarly result n hgher performance. Ths mples that f a frm consders new combnatons as the strategc purpose for a merger, the frm must endure low performance for research actvtes. To compensate for the lower performance, a frm mght need some schemes that wll encourage research actvtes, especally n terms of the number of research actvtes. The second pont s that the possblty of choosng an approprate strategy for a frm depends on the overall portfolo of the frm after the merger. If maldstrbuton of power and ntatve exsts, a frm mght face the dffculty of followng the selecton and concentraton strategy rather than the new combnatons strategy. Contrary to ths, f maldstrbuton does not exst, the selecton and concentraton strategy mght be better because the emprcal results exhbt sgnfcant mprovement n effectveness after the merger. From a human resource management perspectve, two mplcatons can be drawn. The frst s the fact that there s a dfference n the condtons wth regard to the creaton of nnovaton before and after a merger. Therefore, a frm should create a dstnct organzatonal envronment for research actvtes before and after the merger. Second, the classfcaton of collaborators and non-collaborators mples that effectve nnovaton mght occur n the frnge area of a frm s techncal portfolo as well as of human resources (although the effcency of nventons s manly led by collaborators, t s not always acheved effectvely by them). Effectve nventons can be developed by unexpected nventors of a frm, mplyng that such people should not be treated casually. Whle these fndngs are nsghtful, there are several sgnfcant lmtatons of ths study that should be noted. Perhaps the greatest one s the lmted number of cases and the lmted geographcal area covered by the study (n Japan). The small number of corporate mergers wth a suffcent number of fled patent applcatons n Japan lmted the analyss to 18

20 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp two cases. The lmted condtons n Japan as well as n the chemcal ndustry make the generalzaton of the results too narrow. To develop the theoretcal underpnnngs of how to develop nnovatons, more emprcal works examnng the dfference between frms, ndustres, or countres s requred. Thus, so that the cases can be compared on an equal bass, some weghts that reflect the techncal and/or commercal mportance of patents need to be devsed because each merger s dfferent n ts sze, type, purpose, etc. Ths study also has lmtatons that have resulted oeng to the analytcal method ths paper has taken. For example, ths paper measured an nventor s performance by hs or her patentng actvty, whereas the qualty of the resultng patents s not explctly taken nto account (I assumed that each patent has an equal weght). In addton, snce patent data s not a perfect ndcator of research actvtes, I neglected many other aspects that are mportant for generatng nnovaton. Ths study s just an ntal step. The followng questons could be nvestgated n future studes. (a) Who are the real contrbutors n a corporate merger? Ths paper classfed nventors as collaborators or non-collaborators, and assumed that collaborators took an ntatve for the collaborated work. However, the real contrbutors could exst n both the categores. More meanngful classfcaton should be devsed for detaled analyses. (b) What knd of organzatonal envronments can a frm offer to acheve the purpose of a merger? Snce ths paper focused only on the varables of nventors actvtes, the varables at the organzatonal (team, dvson, etc.) level were neglected. Includng these varables provdes a clearer lst of condton that a frm should prepare n order to merge successfully. (c) How are nternal research actvtes related to a frm s outcome? There s a mssng lnk between each nventor s performances and hs/her real contrbutons to a frm. An endeavor for brngng together many measures relatng to an nventor s actvtes n a frm wll provde us wth a more detaled understandng of real contrbutons of patentng actvtes. Despte several lmtatons, I beleve that ths study mght be the frst emprcal test of the effects of corporate mergers at the mcro level wth the purpose of dentfyng the condtons that may affect the success of corporate mergers. As the analyses revealed, the aftereffects of corporate mergers dd not occur n a vod. It should be stressed that they depended largely on the frms strategc choces. 19

21 Atsush Inuzuka, "Do corporate mergers brng about new combnatons of knowledge?: Emprcal evdence from patent data," Internatonal Journal of Knowledge Management Studes, Vol.3, Nos.1/2, 2009, pp Acknowledgement Ths research was partally supported by the Mnstry of Educaton, Scence, Sports and Culture, Grant-n-Ad for Scentfc Research (B), References [1] Ahuja, G., and Katla, R. (2001), Technologcal acqustons and the nnovaton performance of acqurng frms: A longtudnal study, Strategc Management Journal, 22(3): [2] Albert, M. B., Avery, D., Narn, F., and McAllster, P. (1991), "Drect valdaton of ctaton counts as ndcators of ndustrally mportant patents," Research Polcy, 20: [3] Almeda, P. (1996), "Knowledge sourcng by foregn multnatonals: Patent ctaton analyss n the U.S. semconductor ndustry," Strategc Management Journal, 17(Wnter Specal Issue): [4] Almeda, P., and Kogut, B. (1999), "Localzaton of knowledge and the moblty of engneers n regonal networks," Management Scence, 45(7): [5] Bresch, S., Lsson, F., and Malerba, F. (2003), "Knowledge-relatedness n frm technologcal dversfcaton," Research Polcy, 32: [6] Bresch, S., Lsson, F., and Malerba, F. (2004), "The emprcal assessment of frms' technologcal coherence: Data and methodology," In Cantwell, J, Gambardella, A., and Granstrand, O.(eds.), The Economcs and Management of Technologcal Dversfcaton, 69-97, Toutledge. [7] Cantwell, J., and Anderson, B. (1996), "A statstcal analyss of corporate technologcal leadershp hstorcally," Economcs of Innovaton and New Technologes, 4: [8] Cefs, E.(2003), "Is there persstence n nnovatve actvtes?," Internatonal Journal of Industral Organzaton, 21: [9] Cefs, E., and Orsengo, L.(2001), "The persstence of nnovatve actvtes: A cross-countres and cross-sectors comparatve analyss," Research Polcy, 30: [10] Conner, K., and Prahalad, C. K. (1996), "A resource-based theory of the frm: Knowledge versus opportunsm, Organzaton Scence, 7(5): [11] Dos, G.(1982), Technologcal paradgms and technologcal trajectores, Research Polcy, 11: [12] Engelsman, E. C., van Raan, A. F. J.(1992), "A patent-based cartography of technology," Research Polcy, 23: [13] Ernst, H., Lepten, C., and Vtt, J.(2000), "Inventors are not alke: The dstrbuton of patentng output among ndustral R&D personnel," IEEE Transactons on Engneerng Management, 47(2): [14] Grant, R.M.(1996). "Toward a knowledge-based theory of the frm," Strategc Management Journal, Vol.17(Wnter Specal Issue): [15] Hall, B. H., Jaffe, A. B., and Trajtenberg, M.(2002), The NBER patent-ctatons data fle: Lessons, nsghts, and methodologcal tools, In Jaffe, A. B., and Trajtenberg, M.(eds.), Patents, Ctatons, and Innovatons: A Wndow on the Knowledge Economy, , Cambrdge, MA: MIT Press. [16] Jaffe, A. B. (1986), "Technologcal opportunty and spllovers of R&D: Evdence from frms' patents, profts, and market value," Amercan Economc Revew, 76(5): [17] Jaffe, A. B. (1989), "Characterzng the technologcal poston of frms, wth applcaton to quantfyng technologcal opportunty and research spllovers," Research Polcy, 18: