Technological innovation and Acquisitions

Further, although experiencing less decline in innovation output than bidders that cancel the deals, bidders that complete the deals still trail non-bidding firms in technological innova

Technological innovation and Acquisitions

Xinlei Zhao1Department of FinanceKent State UniversityKent, OH 44242330.672.1213xzhao@kent.eduSeptember 2007

Abstract:

I examine firms’ make-or-buy decision in the context of mergers and acquisitions via an

investigation of the relation between technological innovation and acquisition activities

I find that firms engaging in acquisition activities are less innovative and have

experienced declines in technological innovation during the years prior to the bid Among

the bidders, the relatively more innovative ones are less likely to complete a deal,

suggesting that these bidders may feel less pressure Further, although experiencing less

decline in innovation output than bidders that cancel the deals, bidders that complete the

deals still trail non-bidding firms in technological innovation during the three years after

the acquisitions This finding suggests that buying-innovation can only partially alleviate

the internal innovation deficiency problem

Keywords: patents, citations, innovation, Coase Theorem, make-or-buy, R&D

JEL Classification: G34

1 I thank Kai Li for help with the M&A data and Bronwyn Hall for providing the NBER Patent Data I thank Raj Aggarwal,Jarrad Harford, Kai Li, John Thornton, and seminar participants at the Kent State University for their helpful comments All errors are mine

1 Introduction

This paper examines firms’ make-or-buy decision in the context of mergers and acquisitions Thisquestion is one of the fundamental questions in modern economics and it goes back as far as Coase(1937) Coase argues that the make-or-buy decision depends on the environments in which firms performrelative to the market Obtaining goods or services via the market entails a number of transaction costs,including searching and information costs, bargaining costs, and contract enforcement costs, etc Because

of the additional costs in the external market, the Coase Theorem argues that firms are formed to producegoods or services more cheaply internally This theorem also implies that firms resort to the externalmarket only when they are not able to internally produce the goods or service as efficiently in the market.This argument has been reintroduced into economics by Williamson (1975); it has been more thoroughlydeveloped by Klein, Crawford, and Alchian (1978), Grossman and Hart (1986) and others However,empirically proving this theory is difficult Some studies have tried to shed light on this theory byexamine how firm boundaries or asset ownership affect its behavior (for example, Baker and Hubbard(1993) and Mullainathan and Scharfstein (2001)) Another stream of literature directly examines thistheory in the context of mergers and acquisitions (for example, Higgins and Rodriguez (2006))

This paper belongs to the second stream of literature If the Coase Theorem sufficiently describes

a firm’s make-or-buy decision, firms with successful internal innovation are more likely to rely oninternal growth and less likely to participate in the acquisition market Consequently, firms would bemore likely to engage in acquisition activities following a lack of success in their internal innovationefforts Higgins and Rodriguez (2006) find some evidence in support of this buying-innovation prediction

in the pharmaceutical industry However, their work is limited to one industry and it is not clear whetherthe findings can be generalized to other industries and to different types of deals

My study contributes to this literature in four ways First, I examine firms’ make-or-buy decisionsacross industries and over time In contrast, the prior studies have been limited to specific industries, such

as the trucking industry (Baker and Hubbard (1993)), the vinyl chloride monomer (VCM) industry(Mullainathan and Scharfstein (2001)), or the pharmaceutical industry (Higgins and Rodriguez (2006))

Second, I examine the role innovation plays in the likelihood of deal completion, which has notbeen investigated before The Coase Theorem implies that, among the firms involved in acquisitionactivities, the less innovative firms should be more likely to complete a deal.

Third, an important question that has yet to be thoroughly investigated in the existing literature iswhether buying-innovation is detrimental to internal innovation and whether it can fully make up for thelack of success in internal innovation I try to answer this question by examining changes in innovationoutput following a completed acquisition and a failed acquisition

Finally, this paper uses citation data as a measure of innovation output quality Prior studiesusually draw conclusions based on R&D expenditure (a measure of innovation input), or patent counts (ameasure of innovation output quantity) However, neither of these measures is able to sufficientlydescribe the true technological innovation level of a firm

This study uses a sample of 1,053 acquisitions during the period of 1984-1997 I combine thissample with the NBER patent data set created by Hall, Jaffe, and Trajtenberg (2001) I find that,controlling for other firm characteristics, the number of patents of the bidders is comparable to that of thenon-bidding firms However, bidders have significantly fewer citation counts to their patents, and theyhave also lagged behind in terms of citation growth in the past three years This finding suggests thatbidders are as innovation-oriented, but the quality of their innovation is deficient, so they attempt toacquire quality and innovation skills externally The likelihood of deal completion is negatively associatedwith the citation counts prior to the bid, suggesting that bidders with relatively higher-impact innovations

in the past are more confident with their own research productivity and feel less pressure to complete thedeal

Further, acquisitions do not seem to stifle technological innovation since bidders that completethe deals experience less decline in innovation output than those that cancel the deals On the other hand,bidders completing the deals still lag behind non-bidding firms during the three years after dealcompletion This finding suggests that buying-innovation can only partially alleviate the internalinnovation deficiency problem In addition, I find that less innovative bidders experience more increases

in innovation after the acquisitions are completed

These results hold among both diversifying and non-diversifying deals, among bidders from moreinnovative industries and during the sub-periods of 1980s and 1990s Therefore, the findings indeedconfirm that the make-or-buy decision is determined by whether a firm performs better than the market ornot The findings here also suggest that what motivates the make-or-buy decision is the innovation outputquality, not innovation input (R&D) or innovation output quantity (patent counts), further highlighting thevalue of citation data On the other hand, although acquisitions do not stifle technological innovation, theycan only partially make up for the internal deficiency In other words, my findings suggest that firmsshould rely primarily on internal innovation success rather than on acquisitions in the external market.

My study also contributes to the mergers and acquisitions literature Theory has suggested manypossible answers to the question of why acquisitions occur: efficiency-related, market power, disciplining,agency costs, and diversification However, data have not strongly supported any one of these views asconsistently explaining a significant portion of acquisition activities over time My paper sheds insight onthe motivations of acquisitions by examining the quantity and quality of technological innovation on thebidder side, and my evidence suggests that buying-innovation is one reason why firms merge

The plan of the paper is as follows I review the literature on the role of technological innovation

in acquisition and develop my hypotheses in the next section Section 3 describes the sample andvariables, and provides descriptive statistics Section 4 examines how technological innovation affects afirm’s takeover decisions, and Section 5 studies post-acquisition innovation Section 6 discusses someadditional investigations, and Section 7 concludes

2 Role of technological innovation in Acquisitions and Hypothesis Development

Acquisition activities and technological innovation each has a large impact on economic activity,and technological innovation has often been mentioned as one of the justifications for mergers andacquisitions (for example, Chesbrough (2003)) Yet, there is surprisingly little empirical study on theinteraction between the two Typical studies of the efficacy of acquisitions focus on operating synergiesand performance post-acquisition, or announcement return and long-term returns.2 My current paper tries

2 A large amount of literature argues that many acquisitions destroy value for the acquirer (see, for example,

Loughran and Vijh (1997), Rau and Vermaelen (1998), and Moeller, Schlingemann, and Stulz (2005)) Bradley, Desai, and Kim (1988) shows that the increase in the gains to the target shareholders has come at the expense of the stockholders of acquiring firms Nonetheless, there is a growing body of empirical literature documenting that

to fill the void in the literature by investigating the interaction between technological innovation andacquisitions

Earlier studies investigating the role technological innovation plays in acquisition activitiesusually relies on R&D expenditure However, R&D expenditure is an input to innovation, not the output

On the other hand, patents are innovation output and have long been recognized as a very rich andpotentially fruitful source of data for the study of innovation and technological change (see Griliches(1990) for an excellent review of the literature on patents) In particular, patent data include citations toprevious patents and to the scientific literature, allowing researchers to create indicators of the

“importance” of individual patents and cope with the enormous heterogeneity in the “value” of patents.3Past research has shown that citations as a whole do provide useful information about the generation offuture technological impact of a given invention (Trajtenberg (1990), and Hall, Jaffe, and Trajtenberg(2005)).4

acquisitions are efficient means for assets to be reallocated within the economy (Andrade and Stafford (2004)) Large sample evidence by Healy, Palepu, and Ruback (1992) on post-acquisition operating performance, as well as Jensen and Ruback (1983), Jarrell, Brickley, and Netter (1988), and Andrade, Mitchell, and Stafford (2001) for a review of the literature on announcement returns and long-term profitability, suggests that acquisitions on average increase value, and lead to improved profitability in subsequent years

3 The main advantages are: each patent contains highly detailed information on the innovation itself; there are a very large number of patents; the patent system has been around for more than one hundred years; patents are supplied entirely on a voluntary basis; patent citations perform the legal function of delimiting the patent right by identifying previous patents whose technological scope is explicitly placed outside the bounds of the citing patent The major limitations to the use of patent data are: not all inventions meet the patentability criteria set by the United States Patent and Trademark Office (USPTO) (the invention has to be novel, non-trivial, and has to have commercial application); the inventor has to make a strategic decision to patent, as opposed to rely on secrecy or other means of appropriability Thus, not all inventions are patented

4 There is a small but growing finance literature that employs the patent data to examine various interesting

questions Lerner and Wulf (2006) examine the impact of the compensation incentives of the heads of corporate research and development on R&D output, and show more long-term incentives (e.g., stock options and restricted stock) are associated with more heavily cited patents These incentives also appear to be associated with more patentfilings and patents of greater originality Seru (2006) uses patent-based metrics to examine the impact that the conglomerate form may have on the scale and novelty of corporate R&D activity He concludes that conglomerates

do stifle innovation Atanassov, Nanda, and Seru (2006) hypothesize that established firms with innovative projects and technologies will make relatively greater use of arm’s length financing (such as public debt and equity); whereasless innovative firms will tend to use relationship based borrowing (such as bank borrowing) Using a large panel of

US companies from 1974-2000, they find that consistent with their predictions, firms that rely more on arm’s length financing receive a larger number of patents and these patents are more significant in terms of influencing

subsequent patents

Prior Literature

Hall (1990) uses data on R&D to examine the impact of acquisitions on industrial research anddevelopment She finds evidence of declines in R&D intensity as measured by the ratio of R&D to salesafter acquisition, but the declines are statistically insignificant Hall (1999) extends her earlier work bystratifying potential bidders based on their propensity to acquire She finds that firms with a highacquiring propensity that actually make an acquisition have a significantly higher increase in their R&D.Hall concludes that the overall finding of no impact obscures some real heterogeneity across firms

Hitt et al (1991) examine the effect of acquisitions on R&D inputs (firm R&D/sales adjusted forindustry average) and outputs (patents/sales), based on a sample of 191 firms from 29 industries from theperiod 1970 - 1986 They examine the seven-year window around the merger completion They find thatacquisitions have a significant and negative effect on R&D intensity, and diversifying acquisitionsnegatively affect patent intensity However, this study needs to be updated, since technologicalinnovation really took off in the past two decades and more sophisticated statistical methods are nowavailable Further, this study draws conclusions based on R&D and patent, measures that are not able tofully capture innovation quality

Hagedoorn and Duysters (2000) focus on a single, high-tech industry to examine the effect ofacquisitions on the technological performance of the combined firms, as measured by the number ofpatents They conclude that acquisitions can contribute to increases in innovative activities if there isboth the organizational and strategic fit of the companies involved in these mergers They measure “fit”using the SIC codes, the patent classification codes, R&D intensity, and firm size

Higgins and Rodriguez (2006) examine the performance of 160 pharmaceutical acquisitions from

1994 to 2001 They find that firms experiencing declines in internal productivity are more likely toengage in an outsourcing-type acquisition in an effort to replenish their research pipelines Theydocument post-acquisition improvement in three performance measures: positive announcement periodreturns, significantly positive changes in both the research pipeline and sales (the year of acquisitionversus the year after) Besides the fact that this study is limited to one specific industry, it is also unclearwhether the documented improvement can be attributed to better innovation success post acquisitions.For example, the improved research pipeline can result from the additional drug candidates acquired from

the target, which does not necessarily indicate an improvement in innovation output Therefore, theabove study still leaves unaddressed the question whether acquisitions (or a lack of acquisition) affecttechnological innovation.

My Hypotheses

This study examines the interaction between technological innovation and acquisition acrossfirms and over time If the Coase Theorem provides a good description of a firm’s make-or-buy decision,then bidders facing a decline in the success of internal innovation efforts would attempt to buy continuedinnovation externally I call this the buying-innovation hypothesis This hypothesis has the followingpredictions

1) Less innovative firms are more likely to engage in acquisition activities

2) Among the bidders, the relatively less innovative firms are more likely to complete a dealbecause they feel more pressure to do so, while more innovative firms are less likely to complete a deal

The buying-innovation hypothesis does not have clear predictions on the post-acquisitioninnovation levels, which can be affected by many other factors One factor is the strategic fit, whichplays an ambiguous role in the Coase Theorem Post-acquisition innovation could increase if there areeconomies of scale in R&D activities On the other hand, acquisition activities can be affected by agencycosts such as the integration problem and post–acquisition innovation could also show a significantdecline as a result.5

Therefore, whether technological innovation increases or decreases following a successfulacquisition is an interesting empirical question An answer to this question is important to the make-or-buy decision because it can shed light on whether the ‘buy’ decision adds value or not, and whether the

‘buy’ decision can fully make up for internal innovation deficiency

5 A very good example of the latter case is IBM’s acquisition of Rolm, a leading maker of telephone switchingequipment, in 1984 The purpose of the acquisition was to create a technology powerhouse However, because of theenormous differences in business models, the acquisition was not successful and technical experts gradually left.IBM eventually sold the unit to Simens

3 Sample Formation, Variable Construction, and Sample Overview

My empirical design is to relate the quantity and quality of technological innovation to takeoverdecisions and to explore the impact of takeover from a technological perspective

Sample Formation

To form the sample of acquisitions, I begin with all announced (both completed and cancelled)

US acquisitions with announcement dates between January 1, 1984 and December 31, 1997 as identifiedfrom the Mergers and Acquisitions database of the Securities Data Company (SDC).6 I identify all dealswhere the bidder is a public firm and the form of deal was coded as a merger, an acquisition of majorityinterest, or an acquisition of assets I require that the transaction value be no less than 1 million.7 Further, Ionly retain an acquisition if the bidder owns less than 50 percent of the target prior to the bid and isseeking to own greater than 50 percent of the target For completed deals, I require that the bidder ownsmore than 90 percent of the target after the deal completion These filters yield 10,457 deals

To clearly delineate the effect of each acquisition on innovation and reduce the risk ofcontamination, I only include isolated acquisitions, i.e., those acquisitions that do not overlap when asample firm makes multiple acquisitions More specifically, I only keep the first bid by the same bidderwithin a three-year window Note that this three-year time frame is chosen because I also computechanges in patent/citation counts over the same three-year window both before and after the acquisition.This filter yields 4,269 deals

I then match the bidder sample with Compustat and CRSP data, and the NBER patent data TheNBER patent data set was created by Hall, Jaffe, and Trajtenberg (2001).8 This dataset provides amongother items, annual information on patent assignee names, on the number of patents, on the number of

6 The sample period is chosen because the information in SDC is less reliable before 1984 The patent data end in

2002 For completed deals, I impose the additional filter based on the acquisition effective date to be no later than the end of 1997 due to the fact that the average lag between application and grant is about two years, and I would like to have up to three years of citation data post acquisition for the sample firms This helps mitigate the post-1989bias to some extent

7 Given that the success of a knowledge-based firm is not its size, and a small firm may hold many patents or have the potential for many patents So I do not impose any relative size filter that may eliminate many interesting and innovation-enhancing acquisitions

8There are 400 3-digit main patent classes by USPTO, and Hall et al further refine them into 36 2-digit

technological sub-categories, and 6 main categories: categories: Chemical (excluding Drugs);

Computers and Communications (C&C); Drugs and Medical (D&M); Electrical and Electronics (E&E); Mechanical; and Others

citations received by each patent, on the technology class of the patent, and on the years that the patentapplication was filed and was granted Hall, Jaffe, and Trajtenberg (2001) match the assignees of thepatents in the NBER dataset, by name, to manufacturing firms from Compustat, as of 1989 – theNBER/Compustat population is the base for my analysis The fact that the matching occurs for firms thatexisted on or before 1989 might introduce a new listing bias, since firms that went public after 1989 (forboth bidding and non-bidding firms) are not included in the study and older firms dominate the latter half

of the sample (I address this problem in a sub- sample analysis in Section 6.) Using these cusip numbers,

I merge the financial data in Compustat and SDC M&A data with the NBER patent dataset I only keepnon-utility manufacturing firms (SIC codes 2000- 4899 and 5000-5999) and I require a firm to have datathree years before the announcement date and three years after the completion date or the announcementdate I also require that failed acquisitions are not due to regulation reasons These requirements reducethe sample size to 1,349 deals

The number of bidding firms in the population is quite small relative to the total number of firms

To overcome the possible existence of non-linearity and ensure that I am comparing bidders to their mostcomparable non-bidding competitors, I rely on the matching method in this paper I require that thecontrol group of non-bidding firms 1) have no bid within a seven-year period, 2) are from the same Fama-French 48 industry-year as the bidding firm, and 3) have total sales within 25% deviation of the biddingfirm This requirement reduces the final sample to 1,053 deals, including 988 completed deals and 65failed deals There are 7,798 non-bidding firm-year observations in the study.9 Note both bidders andnon-bidders are restricted to firms established before 1989 because of the NBER patent data

Innovation Variable Construction

I focus on the following two measures of technological innovation success: patent counts andcitation counts.10 Both measures are based on the application year, as it is closer to the time of the actualinnovation than the grant year (Griliches, Pakes, and Hall (1987)) The first measure is a simple patentcount for each firm year and measures a firm’s innovation intensity Because only patents that have been

9 I allow for more than one matching firm for each bidder as long as they are from the same industry-year and have total sales within 25% deviation of the bidding firm Results do not change if I pick only one matching firm for eachbidder and if I change the size cutoff point from 25% to 30% or 15%

10 I have checked with the U.S patent office and confirmed that if a firm applied for a patent and then was acquired before the patent was granted, the patent was still granted to the old firm instead of the combined firm Therefore, mergers and acquisitions do not contaminate the identity of the patent assignee

granted are reported in the dataset, the patent data are truncated; i.e., there is a declining number ofpatents towards the end of the sample period I correct for this truncation bias following Hall et al.(2001) Further, there is an increasing trend in the number of patents granted over the past a few decades,and the number of patents from different years will not be directly comparable To adjust for the timetrend, I deflate the patent counts by the average number of patents of that year, using 1990 as the baseyear This measure captures the quantity of a firm’s innovation output

The second metric, the number of citations to patents, measures the quality of a firm’s innovationoutput This measure is motivated by the recognition that patent counts do not distinguish breakthroughinnovations from less significant ones Past research has shown that the distribution of the impact ofpatents is extremely skewed, with most of the value concentrated in a small number of patents (Griliches,Pakes, and Hall (1987)) Trajtenberg (1990), and Hall et al (2005) among others have shown that patentcitations captures the true value of innovations This measure is calculated as the ratio of (1) the totalnumber of citations (as recorded in the dataset) received by a firm’s patents applied in year t to (2) thetotal number of patents applied in year t For example, a firm may applied for five patents in year 1988,and future patents citing these five patents could come from any year after 1988 I count all futurecitations as citations of year 1988

However, the raw citation data suffer from a serious truncation bias, since citations to patents tend

to arrive over time, and the citation lag can take decades Therefore, the number of citations to morerecent patents is not comparable to the number of citations to older patents Further, different industriesmight have different practices in citing patents I correct for these biases using the fixed-effect methodrecommended by Hall et al (2001) This method divides the citations to year t patents by the averagecitations to patents applied in the same technology group and the same year Therefore, the mean citationcounts are normalized to be 1 across all firm-years in the NBER citation database The NBER patent dataend in 2002, so the data after 2000 are more subject to the data truncation problem To further alleviatethis problem, I stop the acquisition sample in 1997, so that patent data after 2000 are not used in the study

Sample Overview

Table 1, Panel A presents the temporal distribution of my sample of announced acquisition dealsover the 1984-1997 period It is clear that acquisitions tend to be highly cyclical, as the total number of

acquisitions closely follows the business cycle expansion over most of the sample period The largenumber of deals in the first year of the sample period, 1984, is due to my requirement that only the firstdeal by the same bidder over a three-year window is kept in the sample Panel B presents the industrybreakdown based on the Fama and French (1997) classification I find a high concentration of bidders inelectronic equipment, machinery, computers, construction materials industries, and wholsale, suggestingthat a large proportion of the bidding firms are not from the high innovation industries

Panels A and B of Table 2 present bidding firm and deal characteristics All firm characteristicsare as of the year prior to the announcement year, and all dollar values are in 1990 dollars The medianmarket value of assets is $430 million My measure of firm size, sales, has a median of $416 millions Iuse as the measure of R&D intensity as the ratio of R&D expenses (data 46) to total sales (data 12).Median R&D expenditure is 1.24 percent The bidders have relatively low market leverage and areprofitable, with a median market leverage ratio of 13 percent and median ROA at 11 percent.11 Thebidders experience median annual stock return of 18.7 percent In the sample, 94 percent of theannounced deals are eventually completed Almost half of the deals use only cash as the method ofpayment, and less than 20 percent of the deals are pure stock swaps Approximately 3 percent of thetargets receive competing bids within one year There are more diversifying deals than within-industrydeals The mean relative deal size, defined as the ratio of the transaction value to the market value of thebidder, is 20 percent, while the median relative size ratio is 8 percent These statistics are quitecomparable to those in the existing literature

Panel C of Table 2 reports the three measures of technological innovation - (1) R&D expenditure,(2) the number of patents and (3) the number of future citations generated by these patents - during theyear before the announcement date for both the bidders and non-bidders The distributions of all threemeasures are quite skewed, both among the bidding and non-bidding firms Second, bidders seem to beless innovation-oriented, as R&D expenditure and both the numbers of patents and citations are

11 Market leverage is defined as the ratio of total book value of debt (data34+data9) to the market value of assets, which is defined as the market value of equity plus the book value of asset minus the book value of equity, and the book value of equity is defined as stockholders’ equity (data 216) or common equity (data 60) + preferred stock par value (data 130) or total asset (data 6) – total liabilities (data 181), plus balance sheet deferred taxes and investment tax credit (if available, data 35) and post-retirement benefit liabilities (if available, data 330), minus the book value

of preferred stocks (estimated in the order of the redemption (data 56), liquidation (data 10), or par value (data 130), depending on availability) ROA is defined as the ratio of data13 (operating income before depreciation) to data12

significantly lower among the bidders than among non-bidders This panel presents the first piece ofevidence suggesting that bidding firms might intend to acquire innovation from the market.

Table 3 presents averages of my measures of technological innovation, patent counts and citationcounts per firm year, as well as R&D expenditure, broken down by industry The top five industries forpatents are aircraft, chemicals, computers, automobiles, and electronic equipment The list is slightlydifferent for citations: medical equipment, chemicals, automobiles, computers, and aircraft Forcomparison, I also present the average R&D expenditure normalized by sales across different industries.The top five industries for R&D expenditure are pharmaceuticals, medical equipment, computers,measuring and control equipment, and electronic equipment It is clear that the ranking of industries byinnovation input, R&D, is quite different from those based on measures of innovation output, patents andcitations to patents This evidence suggests that there is a difference between measuring R&D input andoutput In this paper, I focus on innovation output in my examination of the interaction betweeninnovation and acquisitions

4 Innovation and Acquisition Decisions

I investigate two questions in this section First, I examine the role innovation plays in a firm’sdecision to make a bid Second, I study whether innovation is a factor in a firm’s likelihood of completing

a deal In all models, I include industry and year fixed effects and use heteroskedasticity-adjustedstandard errors that are robust to clustering at the firm level In all regression models, I include both thelevel of past innovation measures and the change in the innovation measures to capture innovationdynamics Further, I use patent counts (citations counts) from year t 3 instead of patent counts(citations counts) from year t 1 to allow for the uncertainty about the innovation quality, as it may taketime for the impact of the innovation to be realized, particularly since the application-grant lag is around 2years (Hall, Jaffe, and Trajtenberg (2001)

4.1 Propensity to Engage in an Acquisition

I base the empirical model on previous work such as Higgins and Rodriguez (2006), and add mymeasures of innovation as explanatory variables,

&

/

)1()

Pr(

1 10

1

9

1 8 1 7

1 6

1 5 1 4

3 3

1 2

3 1

0

it it it

it it

it it-

it

it it

it t

industry it

e Leverage turn

ROA D

R B

M Sales

Citations in

Change

Citations Patents

in Change Patents

f f

year t-3 to t-1 by one citation count reduces the propensity to bid by 3.5% Therefore, Table 4 indicates

that bidders have significantly lower quality innovations than the non-bidding competitors, consistentwith the first prediction of the buying-innovation hypothesis

On the other hand, although there is a significant difference in R&D expenditure and numbers ofpast patents between bidding and non-bidding firms (Table 2), after controlling for other firmcharacteristics, I do not find a significant difference in R&D expenditure, the number of patents and pastchanges in patent counts between bidding and non-bidding firms These results imply a distinctive roleplayed by innovation output quality versus innovation input or innovation output quantity

However, the results in Table 4 may hide possible heterogeneity among bidding firms Inparticular, technological innovation may not play a role for bidders engaging in diversifying deals.Further, technological innovation should be a concern primarily only for highly innovative industries Ifthe results in Table 4 are driven by firms from less innovative industries, they may not provide muchinsight to the role that technological innovation plays in acquisition activities

To address these concerns, I conduct the same analysis across various sub-samples, and presentthe probit results in Table 5 To save space, I only report coefficient estimates of the patent and citationmeasures Panels A and B report results for diversifying deals and non-diversifying deals, respectively

Panel C presents results among the 15 most innovative industries, defined by number of patents as perTable 3.12 The evidence that bidders have lower innovation quality and have experienced a decline ininnovation quality is quite robust across the different sub-samples

It is interesting that, in both Tables 4 and 5, neither lagged patent counts nor past changes inpatent counts play any role in a firm’s decision to bid Also, although not reported in Table 5, there is nosignificant difference in R&D expenditure between bidding and non-bidding firms These findingsfurther imply that it is the quality of technological innovation, rather than the input or innovation outputquantity, that plays the major role in firms’ make-or-buy decision

Pr(

15

14 13

12 11

1 10

1 9

1 8 1 7

1 6

1 5 1 4

3 3

1 2

3 1

0

it it

it it

it it

it

it it

it it

it

it-it it

it t

industry it

e Competing

Stock All Cash

All Size

lative Diversify

Leverage

turn ROA

D R B

M Sales

Citations in

Change

Citations Patents

in Change Patents

f f

This table shows that bidders with more citations in year t  3 are less likely to complete the

deal The marginal effect of model (3) suggests that an increase of one citation count in year t-3 increases

the likelihood of deal incompletion by 1.9% This magnitude is economically significant given theunconditional deal incompletion rate at about 6% This finding is consistent with the second prediction ofthe buying-innovation hypothesis However, bidders completing and cancelling the deals do not differ

statistically in terms of citation changes from year t-3 to year t-1 Since bidding firms cancelling the deals

have more citation counts in year t-3, with a similar magnitude of citation count deterioration, these firmsstill have relatively more internal innovative power than bidders completing the deals

The observation that patent counts and the change in patent counts turn out to be statistically significant is consistent with the finding from Tables 4 and 5, again suggesting that the quantity of

non-12 Results do not change when I use top 10 or top 20 innovative industries

innovation is not of primary importance Results from sub-samples, i.e., diversifying and non-diversifyingdeals, and bidders from more innovative industries, are very similar to those presented here and are notreported because of space limitations

Table 5 also shows that large firms with high prior year stock returns are more likely to completethe deal, and R&D expenditure is not related to the likelihood of deal completion On the other hand,diversifying deals, deals with smaller relative size differences, and deals involved either pure cash or purestock swaps are less likely to complete Not surprisingly, deals with competing bids are less likely tocomplete

Results from this section add further support to the buying-innovation hypothesis They implythat less innovative firms are more likely to complete a deal possibly because they feel more pressure; ormore innovative firms are more careful and would not complete if the deal is not right The findings hereprovide additional evidence that it is the innovation output quality, rather than innovation input or outputquantity, that is of primary concern in firms’ make-or-buy decision

However, it shall be interesting to examine 1) whether acquisitions stifle technologicalinnovations among bidders that complete the deals, and 2) whether the ‘buy’ decision can fully make upfor internal innovation deficiency We answer these questions in sections 5.1 and 5.2, respectively, via anexamination of the post-event innovation levels

5 Post-Event Innovation

5.1 Do acquisitions stifle technological innovations?

I illustrate in Figures 1.1 and 1.2 the number of patents and citations three years before and threeyears after the acquisition (acquisition announcement) for bidders that complete (cancel) the deals Figure1.1 shows that the average patent counts of bidders completing the deals remains stable during the sevenyears By contrast, bidders canceling the deals experience an almost monotonic drop in the averagenumber of patents during the same period The gap between these two groups of bidders is 2.4 three yearsbefore the bid, which grows to 6.3 in the year of the bid Following the bid, the gap in the number ofpatents between the two groups widens to 8.5 by the third year

Figure 1.2 tells a similar story Both groups of bidders experience a decline in citation counts;however, the drop is more severe for bidders canceling the deals These bidders actually have slightlymore citations per patents than the deal-completing bidders three years before the bid, consistent with thefindings in Table 6 Nevertheless the difference reverses to 0.10 fewer citations for the bidders notcompleting the deals in the year of the bid, a gap which remains for at least three years following (thecompletion of) the bid

Bidders not completing the deals experience a decline of 5.6 in patent counts and a decline of0.35 in citation counts during this seven-year period, both of which are statistically significant Bycontrast, bidders completing the deals do not experience a drop in patent counts and a reduction of a muchsmaller magnitude (0.09) in terms of citation counts The difference in patent count changes between thetwo groups of bidders is not statistically significant, but the difference in citation count changes isstatistically significant at the 5% level

I also conduct the same analysis among diversifying deals, non-diversifying deals, and biddersfrom more innovative industries All findings, not reported here because of space limitations, suggest thatbidders that do not complete the deals experience a reduction in patent counts, although the decline isonly marginally significant Further, these same bidders face an economically and statistically significantdrop in citation counts Although the bidders completing the deals also have a decline in citation counts,the magnitude is significantly smaller than that among bidders canceling the deals These results suggestthat acquisitions do not stifle technological innovations, since bidders completing the deals experienceless decline in technological innovations than bidders cancelling the deals

5.2 Can the ‘buy’ decision fully make up for internal innovation deficiency?

The decline in citations experienced by bidders completing the deals in Section 5.1 is difficult tointerpret, since there is also an overall decline in citations among the non-bidding firms in the sample.Because I normalize the average citation counts to be equal to 1 for each technology category of each year(see appendix), the decline only suggests that the technological innovation by firms in the sample (firms

in existence before 1989) becomes less important over time compared to new firms - firms establishedafter 1989 and thus are not included in the sample Further, the bidding and non-bidding firms differ interms of patent counts and citation counts even before the deal year Therefore, directly comparing the