In this article we will discuss about:- 1. Introduction to the Future Development of Agricultural Biotechnology 2. Casual Analysis of Mergers 3. Methods and Techniques to Estimate Consolidation.
Introduction to the Future Development of Agricultural Biotechnology:
Over the past decade the structure of the plant-breeding and agricultural biotechnology industries has been radically transformed. Through dozens of mergers, acquisitions and strategic alliances, there has been a rapid and ‘dramatic concentration of control over value-generating assets. At the time of many of these acquisitions and mergers, the recorded valuations were surprising. In August of 1996, the announced purchase of Plant Genetics Systems (PGS) for (US)$730 million was made at the time PGS’s market capitalization was $30 million.
According to AgrEvo, $700 million of the purchase price was assigned to the valuation of the patent-protected trait technologies owned by PGS. The acquisition of Holden’s Foundation Seeds by Monsanto may have been even more surprising. Here, a privately owned company, Holden’s, with gross revenues of only $40 million, was acquired for a purchase price of $1.1 billion. A principal regulatory issue in this merger was the potential effect that might arise for germplasm access by Monsanto’s competing trait developers.
Holden’s germplasm is widely disbursed throughout the industry and at least one of its elite lines is present in most commercial maize pedigrees. In the case of Monsanto’s acquisition of DeKalb Genetics, Monsanto paid not only a control premium of 122% for the 60% of DeKalb that they did not already own, but also indemnified DeKalb against any disapproving regulatory action.
Moreover, DuPont acquired 80% of Pioneer for $7.7 billion that it did not already own. In this instance, the control premium was only 14%, while the initial premium paid for 20% of Pioneer (purchase price of $1.7 billion) was significantly higher. More generally, the pattern of acquisition in agricultural biotechnology is over the period of 1984-2000.
As assets have been reshuffled, and in many instances newly created, much controversy has arisen. The content of the controversy has ranged from regulatory concerns about the exercise of market power, academic researchers’ concerns about freedoms to operate, competitors’ concerns about litigation threats, consumer concerns about genetically altered foods, and environmental concerns about insect resistance build-up. Our specific concern is with the role that intellectual property has played, if any, in the consolidation and restructuring of the US agricultural biotechnology industry.
The validity and scope of intellectual property in agriculture has a long and controversial history. During much of this century, conflicts and disputes have continuously arisen between farmers’ and breeders’ rights. Up until the 1930s, farmers’ rights generally prevailed. Prior to the development of hybrid maize varieties, farmers had direct access to any germplasm that was developed.
Attempts to secure premiums or differentiable pricing for new germplasm were generally unsuccessful. Annual crops had short reproductive cycles and their seeds could easily be saved by farmers. Until the introduction of hybrid maize, these saved seeds bred true and maintained their productivity.
In the case of maize, however, the introduction in the 1930s of hybrid varieties meant that saved seed was no longer a viable option for farmers. As a result, the biological science of the time allowed private breeders to protect their discoveries and innovations through trade secrecy. Initially, private breeders used public ‘inbred lines’ to develop the parents of their proprietary hybrid varieties. As a result, hybrid maize was the first significant example of intellectual property in the agricultural industry.
The first legislation in the USA to protect the investment of plant breeders came through the Plant Patent Act of 1930. The act protects asexually reproduced varieties, i.e. those that are reproduced by cutting, layering, budding or grafting. The legislation specifies that in order to qualify as intellectual property, the variety must be ‘distinct’ and ‘new’. These requirements are much weaker than those that apply to utility patents. Both kinds of patents are administered by the US Patent and Trademark Office.
It was not until 1970 that protection was provided to sexually reproduced varieties through the Plant Variety Protection (PVP) Act. This act is administered by the US Department of Agriculture (USDA), which offers certificates to breeders on the basis of distinctiveness, uniformity and stability. Infringement cover’s selling, importing or sexually multiplying a protected variety. There are, of course, several exemptions allowed under this act. In particular, there is a research exemption, which allows anyone to use a protected seed variety to breed a new variety.
Moreover, another exemption allows farmers to save seed for reproductive purposes as well as to sell seed to other farmers whose primary occupation is growing crops for consumption or feed. The PVP Act was amended in 1994 to eliminate controversial provision allowing the sale of ‘saved seed’ to others for reproductive purposes.
Under this amendment, a farmer can use saved seed only for his own re-plantings, and can sell his purchased seed only for purposes other than reproduction. The subsequent court decision, Asgrow Seed Company vs. Winterboer, established that agents who sell any amount of seed for reproductive purposes now violate the rights of certificate holders.
With the introduction of modern biotechnology, utility patents have provided stronger intellectual property protection for plant-related innovations. A utility patent is a property right granted by the US government to inventors to exclude others from imitating, manufacturing, using or selling the invention over a specified period of time.
In exchange for this exclusive right, the public receives a detailed description of the invention, so that others can use it after the patent-specified term has expired. To obtain a utility patent, the subject matter must meet certain criteria – it must be novel, non-obvious, useful and amenable to the descriptive requirements of the law.
Until the landmark Supreme Court ruling in the matter of Diamond vs. Chakrabarty, plant-related inventions based in genes or cells from nature or applied to living organisms were viewed as natural phenomena and were thus deemed un-patentable. In this case, however, the court held that ‘anything under the sun that is made by man’ is patentable subject matter. Specifically the court found that ‘the patentee has produced a new bacterium with markedly different characteristics from any found in nature and one having the potential for significant utility.
His discovery is not nature’s handiwork, but his own; accordingly it is patentable subject matter under the section 101.’ This decision broadened the narrow reach of utility patent laws to encompass living organisms. Accordingly, utility patents are now granted in the USA for genetically engineered organisms, for processes that transform cells and express proteins, and for the genes themselves.
With the current widespread issuance of utility patents, agricultural biotechnology suffers from the same problems as biotechnology intellectual property in general – (i) many layers of patented technology are necessary for production, and those layers may be owned by different firms; and (ii) new technologies embodied by biotechnology patents are frequently ill-defined, which leads to uncertainty over patent scope and validity.
The new patent structure is one possible causal source for the merger wave that has swept through the industry. Indeed, it may be that uncertainty in patent rights causes a breakdown in contracting which provides incentives for consolidation. However, to date, there is no hard evidence to support this hypothesis. Weak or uncertain property rights may lead firms to develop alternative organizational forms with which to manage their intellectual property.
Certainly consolidation is one such alternative and may in part be explained by patent right uncertainty. In this context, one obvious question that arises is whether the evolution of patent ownership mirrors the changing structure of the industry. There is a through between 1995 and 1996 reflected by the Herfindahl-Hirschman index of the patents in our sample.
To our knowledge, no empirical study has investigated the role of intellectual property holdings in merger decisions. Using patent data enables us to directly address the intellectual property issues that the industry raises in regard to consolidation, and affords us two methodological benefits. By using patent data, we are able to examine both public and private firms. Most merger studies focus exclusively on public firms because of the dearth of economic data on private firms.
Since patent data are available for all patenting firms, they enable us to use a wide variety of informative measures without relying on information available for only public firms. Moreover, by building on previous work of Marco we are able to investigate the consequences of intellectual property uncertainty in the consolidation of agricultural biotechnology.
Following a brief literature survey on causal analysis of mergers and other types of restructuring, the methodology is presented. To address the limitations of conventional methodologies discovered by the prior literature, we combine duration and logit models. By doing so, we hope to capture the strength of each approach.
We use duration analysis to investigate the timing and factors influencing the merger decision. However, once that decision is made, we use a conditional logit model to estimate the probability of ‘matching’. The conditional logit model is appropriate for the matching decision since – by conditioning on the acquisition decision – we have reduced the decision to a static one.
Casual Analysis of Mergers:
A variety of tools exist to examine the causes and consequences of mergers and other types of restructuring. Hall and Sinay investigate the consequences of mergers by comparing merged firms to non-merged firms. Hall examines the R&D behaviour of firms under different types of restructuring. Sinay investigates the effects of mergers on hospital costs by examining pre- and post-merger cost function estimates.
Qualitative choice models have also been used to examine the determinants of mergers. Hall presents an excellent description of the econometric issues that arise in applying qualitative choice models to the market for corporate control. The article discusses the problems of a market where the buyers and sellers are ex ante indistinguishable, and the problems involved in defining the choice set.
In the merger market, the set of choices is equal to the number of possible participants in the market. In some cases, this calls for some simplification in order feasibly to analyse a given sample. Since Hall uses a large inter-industry sample, sampling is advanced in order to reduce the choice set for each firm.
Some authors have analysed industries in isolation. For instance, Bacon et al. use a logit analysis to predict whether firms belong to the merged or non-merged groups in the rural electricity market. Tremblay and Tremblay estimate the probability that beer manufacturers will be involved in mergers.
The benefits of focusing on a particular industry are that the choice set becomes feasible, and also that the results do not suffer from potentially contaminating differences among industries. Of course, to apply the results then to other industries becomes problematic.
With regard to merger analysis, qualitative choice methods suffer from the problem that they are inherently static. When the analysed period of time is small, this may not be a problem. However, when the range of the study is large, static analysis does not seem appropriate.
In the Tremblay and Tremblay study this problem is dealt with by estimating the probability of merger year by year. However, because of the dynamic nature of merger decisions, and the fact that they occur at different times within a sample, duration analysis has also been used.
Two examples of duration models applied to corporate structure are Ravenscraft and Scherer and Van de Gucht and Moore. Van de Gucht and Moore use a duration model to estimate the factors that influence the survival of LBOs. Many LBOs revert to public forms of ownership, while some remain LBOs. Since these events happen over a length of time, and since some observations are truncated, duration analysis is appropriate.
Ravenscraft and Scherer investigate the probability that firms will sell off divisions (the flip side of the merger market).
They point out three benefits of using duration analysis – (i) when events occur at different times; (ii) when the probability of events may be changing over time; and (iii) when observations are censored. The intuition is that duration analysis uses valuable information about the timing of events that logit analysis is not able to capture. However, they also point out two problems with the approach – (i) it requires specification of a particular hazard function (at least for parametric approaches); and (ii) it is difficult to deal with time- varying covariates. Regarding (ii), they argue that while theoretically time-varying covariates can be incorporated, ‘(i) n practice, this step is plagued with computational complexities and collinearity’.
Methods and Techniques to Estimate Consolidation:
Three models are estimated that identify the factors that increase the likelihood of consolidation. First, we estimate a duration model measuring the rate at which firms pursue acquisitions in agricultural biotechnology. Then, we estimate a second duration model – this one on the rate of being acquired. Last, conditional on a firm deciding to pursue an acquisition, we use a fixed effects model to investigate the probability that a given acquirer will match with a potential target.
i. Likelihood of Making an Acquisition:
The first duration study examines the probability (the hazard rate) of making an acquisition in the agricultural biotechnology industry. The factual support for this study is event data. In particular, we identify a sample of agricultural biotechnology firms and track control changes in these firms in the post-1994 period. Beginning from this sample, we obtained merger dates and other information from searching Lexis’ Mergers and Acquisitions file. The sample was augmented with agricultural mergers going back to 1984, up to April 2000.
In total we have merger histories for 111 firms. Note that we include agricultural mergers only. So, while Dow is involved in agricultural chemicals, if it purchased an electronics firm, that firm and that merger were not included in the sample. We obtained data on the patent portfolios of these firms from Micro-patent.
In the merger context, the duration – or ‘spell’ – refers to the length of time without making an acquisition. We tracked the merger history for all potential acquirers, where history refers to the number of previous acquisitions by the firm in the sample. With regard to tracking mergers, we needed to make several assumptions about who was buying whom.
These assumptions are laid out below:
a. The sample consists only of patent holders. Since we are interested in the consolidation in agricultural biotechnology, we do not examine mergers of non-patent holders.
b. Parent firms are always the acquirers. That is, if a subsidiary makes an acquisition, we code that as an acquisition by the parent.
c. Parents are assumed to have a patent portfolio consisting of the current patents of all their subsidiaries.
d. Companies formed by the merger of equals are considered to be new entities, e.g. Novartis formed by the merger of Ciba Geigy and Sandoz. This makes a difference only in coding the merger history of the firm. However, a name change is not considered to be a new entity, e.g. ELM becomes Savia. So, Savia retains the merger history of ELM.
e. The beginning of a firm’s spell is assumed to be the month in which is applies for its first patent, or 1 January 1984, whichever is later. When a firm makes an acquisition, its spell has ended, and the following month it begins a new spell with its history augmented by one.
f. A firm remains in the sample until the earliest of – (i) the date it is acquired; (ii) 10 years after the issuance of its last patent; or (iii) the end of the sample period.
g. Measuring time-varying covariates in duration models necessitates some simplification. In our data, time-varying explanatory variables for a given firm are measured at the end of the spell (at the event date, t1). So, the probability that a firm will make an acquisition at any time t < t1 is a function of the firm’s characteristics at time tr For shorter spells this simplification is not troublesome, since patent portfolios change slowly over time.
Patent data were obtained from Micro-patent by searching on company name. The data consist of 94,976 US patents issued by the 111 firms in the sample between the years of 1975 and 1998. For each firm, and for each event date (the date of a merger between any two firms in the sample).
All of the explanatory variables listed above are calculated using firm’s ‘live’ patent portfolio as of time t. In our analysis, a patent is alive from the application date until 17 years after the issue date. Because we use the application date, the portfolio includes patents that are ‘in the pipeline’, i.e. those that have been applied for, but not yet issued. This is a reasonable measure since firms will base their decisions on in-process technology as well as developed technology.
Once a firm acquires a target, the target’s portfolio is absorbed by the parent. Since HHIs needed to be calculated at all event dates, it was necessary to calculate each firm’s market share at all event dates. We tracked the portfolios of each firm over each event date, accounting for all consolidations of portfolios through mergers, and using only live patents.
Calculation of Patent Enforceability:
Firms in agricultural biotechnology claim that one of the reasons that they engage in mergers is because of the difficulty in enforcing their property rights and the difficulty in producing where other firms are enforcing theirs. A patent is only enforceable if a court finds it both valid and infringed.
Therefore, we interpret the predicted probability of validity and infringement (conditional upon being litigated) as ‘enforceability’. A variable prvi.ag is constructed reflecting the average enforceability for a firm’s agricultural patents.
The probability that a patent will be found valid if it is brought to court is estimated by:
Pr(V = 1) = f(Xß),
An agricultural patents are defined as those assigned to international patent classes A01, C07H, C07K, C12M, C12N or C12Q.
Where X is a matrix of patent characteristics, including:
a. Age1 – The age of the patent at the time of litigation;
b. Age2 – The age of the patent at the time of adjudication;
c. Dummy variables for the year in which the patent was issued;
d. Forlif – average annual forward citations to the patent;
e. Selfor – the proportion of forward citations that are self- citations;
f. Numback – The number of backward citations;
g. Numicd – the number of unique four-digit international patent classes to which the patent has been assigned;
h. Dummies for the technology field of the patent (agriculture, medicine, chemicals, electronics, mechanical);
i. Patdelay – the delay of the patent between application and issuance.
For validity, we estimate a probit model and obtain the parameter estimates found in Table 12.2. Using these parameters, we predict the probability that each patent in our sample would be found valid if litigated. We repeat this analysis for the probability of infringement to obtain the parameter estimates found in Table 12.3.
To create our enforceability measure, we construct an interaction term equal to the product of the predicted probability that a patent is valid and the predicted probability that it would be found infringed. Presuming independence, prvi.ag = Pr(patent is valid and infringed) = Pr(valid)-Pr(infringed).
Note that we could have used the litigation data to directly estimate a logit model of the joint probability of validity and infringement findings. However, some cases do not rule on both matters. So we are able to increase the sample size by estimating them separately.
Summary:
The merger data yield 133 observations – 48 acquisitions, 31 observations are censored on the right because they are acquired and 54 observations are censored because the firms do not acquire anyone before they exit the sample. The means of the variables for the acquisition analysis are given in Table 12.4.
Note that the maximum history is seven. This firm is Monsanto, who acquires seven firms in the sample before it is acquired. After its seventh acquisition, its history is seven, at which point it exits the sample by being acquired. Also, the maximum duration is 16 years, which reflects firms who are in the sample for the entire sample period, but which never acquire. All the Japanese firms belong in this category.
Estimation and Accuracy:
We specify a reduced form model for the probability of acquisition. Our model is motivated by the assumption that a firm will choose to make an acquisition in the next small interval of time when the value of doing so exceeds the reservation value (the value of no acquisition). Of course, the value to any particular firm of an acquisition is dependent upon the choice set of possible targets. However, we are not (at this point) interested in which target will be chosen, but only whether a firm chooses to make an acquisition at all.
Since the choice set is (almost) the same for all firms, the only distinguishing characteristics are the characteristics of the potential acquirer. The choice set is almost the same, because for any firm j, the set does not include the firm j.
Or alternatively, the set includes j, but acquiring oneself is equivalent to making no acquisition. Because of this, the probability that a given firm j will make an acquisition relative to other firms is dependent only on its own characteristics. If this is an industry typified by a highly attractive acquisition set, then this will show up in the intercept term.
Accordingly, we model the probability of a firm making an acquisition at time t as a function only of the firm’s characteristics and the characteristics of the market (the HHI):
λ(t) = exβ+ε
where X is a matrix of firm and market characteristics,
λ(t) = f(t)/1 – F(t)
is the hazard function, and f(t) and F(t) are the usual density and cumulative probability functions.
Our specification assumes a constant hazard – λ(t) = λ, so that the hazard function does not vary with time. That is, there is no duration dependence; the length of time a firm has gone without a merger does not, ceteris paribus, affect the likelihood of merger in the next interval of time. The hazard rate is constant in t if 1 – F(t) is distributed according to the exponential distribution.
Estimation involves maximum likelihood estimation where the censored observations are incorporated much like the Tobit model:
The results of the estimation are given in Table 12.5.
Degrees of freedom – 133 total; 126 residual.
-2 x log-likelihood – 241.
% = 0.0440 at the means of the independent variables.
SE, standard error.
The signs of the coefficients can be interpreted in the usual way; since λ = eXß, In λ = Xß. So, a positive coefficient indicates a positive relationship with In?, meaning a positive relationship with?. The hazard rate is 4.4%, evaluated at the means of the independent variables. This rate means that – on average – the probability that a firm will make an acquisition in the following year is 4.4%.
From Table 12.5, we see that there are several factors that appear to be important in determining the rate at which firms acquire. The number of previous mergers is related to a higher rate of acquisition. That is, it appears that firms that merge more also merge increasingly frequently. This can be viewed as a firm-specific ‘taste’ for mergers.
To see the quantitative effects of changes in history, it is instructive to graph the hazard rate as a function of history, with the other variables assumed to be equal to their means. We do this for each independent variable in Fig. 12.4. Note that variables that are not significant, like pct.ag, will not have a large effect on the hazard rate.
The inclusion of the HHI of agricultural patents is intended to control for the acquisition behaviour of other firms in the industry. If acquisitions are reactions to competitors’ mergers, than we should see a positive influence between hhi.ag and the hazard rate. However, the opposite appears to be true. This can partly be explained by a reduction in the set of available targets as concentration grows.
Other factors that appear to be important are size (which increases the likelihood, as measured by the share of industry patents owned by the firm), recent patenting behaviour (which also increases the likelihood), as well as the derived explanatory variable prvi.ag. Agricultural intensity itself does not affect the likelihood of acquisition.
A large value for prvi.ag indicates a ‘high-quality’ patent portfolio, or one that is characterized by less uncertainty about both validity and scope. Note that a low value for prvi.ag does not indicate that the patented technologies are not valuable-only that they are difficult to enforce. As the results suggest, stronger intellectual property rights at the firm level are associated with higher rates of acquisition. We will return to this hypothesis when we examine the matching model.
ii. Likelihood of Being Acquired:
The parallel analysis for the acquirer duration is the target duration. This model estimates the probability that a firm will be acquired, conditional upon its characteristics and industry characteristics. We use only the target’s characteristics – as opposed to the acquirer’s – for the same reason elaborated in the acquirer duration analysis.
The data for the target duration analysis (the rate of being acquired) is very similar to the acquirer duration analysis above, with one additional restriction. We assume that independent firms are the only candidates for acquisition. That is, a firm can only be acquired once. Once it is a subsidiary, it is ‘off the market’.
Clearly there is a market for the acquisition of assets, including wholly owned subsidiaries. We do not include these assets in the sample. Had we, it would have involved tracking the patenting behaviour of all subsidiaries. Unfortunately, different firms handle post-merger patenting differently.
While some maintain independent patenting by the subsidiary, some absorb the R&D activities of the new subsidiary into those of the parent, making the entities inseparable. Since we cannot observe the difference from available information, we exclude sales of subsidiaries from the analysis.
Our data include 90 observations, of which 31 involve firms that are acquired, and 59 are truncated (or censored), because they are not acquired by the end of the sample period. Summaries of the variables used are in Table 12.6.
All independent variables are calculated as described in the acquirer duration analysis. We again estimate an exponential model for the target analysis. The results are summarized in Table 12.7.
Degrees of freedom – 90 total; 83 residual.
– 2 x log-likelihood – 117.
X = 0.0321 at the means of the independent variables.
SE, standard error.
The target duration results are similar to those of the acquirer duration model, with one exception – the size of the firm’s patent portfolio was marginally important for making an acquisition. Here we find the converse. This result is to be expected since acquirers in this industry tend to be larger, more diversified firms.
Somewhat surprisingly, intensity of agricultural patenting continues, to be unimportant, and, since we are examining the agricultural biotechnology sector, we would expect that target firms will be ag-patent intensive.
However, the patent classes are only rough guidelines for the actual technological uses of patents. Figure 12.5 shows the response of the hazard rate to changes in the independent variables. Again, insignificant independent variables show up as very flat curves.
Not surprisingly, recent patenters are also attractive targets as measured by the positive coefficient for pct.yng. Also, the industry concentration (hhi.ag) has a similar coefficient as in the acquirer analysis, which is to be expected because it is an environmental variable, and not firm specific. That is, if concentration leads to fewer acquisitions, then it will lead to both fewer acquirers and fewer targets, on average.
Patent enforceability appears to enter in the same direction (a positive effect on the hazard rate), and magnitude (a coefficient of 18.1 in the target analysis vs. 20.6 in the acquirer analysis). This result is interesting. We found that high expectations of enforceability led to a higher likelihood of making an acquisition. In the target analysis, we find that high values for enforceability also increase the likelihood of being acquired.
The interpretation is that firms with more enforceable patents are more attractive targets, and more aggressive suitors. Whether firms with enforceable patents align themselves one with another is a question that cannot be answered by examining acquirers and targets independently. Thus, we turn to a model of acquirer and target matching.
iii. Matching:
We are interested in addressing the question of who acquires whom. To do this we examine the acquirers who made acquisitions, and obtain information on their contemporaneous choice set. In this fashion, we can ascertain which characteristics of the targets made the realized target the best choice for the acquirer. Thus, we are conditioning on the acquirer having made a decision to acquire at date t.
Our methodology is to use conditional – or fixed effects – logit. To do so requires the development of more explanatory variables, and to arrange the data in a particular way. The data consist of acquirer-target pairs for each acquirer at the date of an actual acquisition. One acquirer-target pair will be the actual consummated deal. The other acquirer-target pairs will consist of the actual acquirer matched with all possible targets at date t. A possible target is any independent firm as of the date of acquisition.
Our sample contains 31 acquisitions of independent firms. These events can be described by acquirer at date t – At. An acquirer may enter more than once, so that At = At + 1, but acquirer-date combinations are unique. Each acquirer-date combination contains one observation for each potential target. In data, there are – on average – 62 available targets, yielding 1921 observations. At any given date there may be more or fewer available targets, due to entry and exit. An event equals 1 if the acquirer actually purchased the target, 0 otherwise.
The explanatory variables for the matching data are similar to those of the duration models. We use the target’s values for share.pat, pct.ag, pct.yng. Furthermore, we create two new variables following Podolny. Specifically, let BA be the set of patents that are cited by the acquirer’s patent portfolio (where the time subscript is omitted). Similarly, let BT be the set of patents that are cited by a potential target’s portfolio.
Then the overlap between the acquirer and target is OAT = (number of patents in BA n BT)/(number of patents in BA). Similarly the overlap between the target and acquirer is is OTA = (number of patents in BA n BT)/ (number of patents in BT). Table 12.8 summarizes the overlap variables.
The overlap variables are intended to measure the similarity in the research programmes of a pair of firms. To the extent that backward citations define a technology space, then the overlap variables measure whether the firms’ research programmes lie in the same space.
Essentially, this measure allows us to observe whether the firms lie in the same space or not. Note that the measures are not symmetric; if BT C BA, then OTA = 1, and OAT < 1. The overlap variables provide a basis for inferring whether overlapping property rights, complementarities or uncertain property rights are at the heart of the matches between firms.
The matching data contain information for 23 acquirers making 31 acquisitions. For each acquisition there were on average 62 potential targets, so that each acquisition accounts for approximately 62 observations, for a total of 1921 observations. Our sample is narrow enough that we can feasibly include all possible targets in the industry.
For the available data, we estimate the matching model using a conditional, or fixed-effects, logistic regression following Greene. Let At be the acquirer at time t, and let Tt= {Ttr…Tm} be the set of potential targets at time t. yti is an event variable describing whether Tti was acquired by At (yti = 1) or not (yti = 0). Importantly, the acquirer is restricted to making one and only one acquisition per period. So, we want to measure the probability that yti = 1, conditional on SNtyti = 1.
For At to find it worthwhile to acquire Tti, it must be that the value (V) of making the acquisition is greater than that for the other potential targets. Let the value of acquiring Ti at date t be Vti. If At chooses target Ttj, then it must be that Vt > Vti for all Tti e Tt. If eti is distributed with the Weibull distribution, we can write the relevant probability as –
Since at enters all the terms, it drops out of the probability. That is, acquirer specific effects do not alter the probability that a particular target is chosen, conditional on the fact that the acquirer has already chosen to make an acquisition. Also, note that joint characteristics (which involve characteristics of At, but which vary with the target) remain. Of course, these characteristics include the overlap variables.
Estimation is done via maximum likelihood. The results are reported in Table 12.9, where ‘_t’ represents values of a variable for the target.
Number of obs. = 1921.
LR chi2(6) = 22.47.
Prob. Chi2 = 0.0010.
Log likelihood = – 116.57998.
Pseudo R2 = 0.0879.
SE, standard error.
The independent variables, with the exception of over_at and over_ta, are the same as those of the target in the duration analysis. We see that the only variables with any independent explanatory power are pctag_t and over_at. The signs of the coefficients can be interpreted in the usual way – a positive coefficient increases the probability of a match.
The agricultural intensity appears to have positive effect on the probability of a match. However, the enforceability variable does not have any explanatory power. So, while enforceability helps to explain the likelihood of being acquired, it does not help to explain the likelihood of being acquired at a specific point in time. The same is true of the percentage of young patents. These variables tend to make the targets more attractive, but more attractive for all acquirers, not just one in particular.
The overlap variables are more revealing. The positive coefficient on over_ta shows that the greater the overlap in the backward citations (the technology space), the greater likelihood that the firms will match. The value of over_ta will generally be larger than over at, since acquirers tend to have larger portfolios; hence the effect over_ta is larger than the effect of over_at. The result is not unexpected.
However, it can be interpreted in two ways:
First, a high value for over_ta may indicate complementarities in intellectual property.
Second, a high value may also indicate overlapping property rights, which lead to mutually blocking technologies.
Without additional information, it is impossible to determine which story explains matching. However, bringing together results from the three studies, inferences about the determinants of consolidation in agricultural biotechnology can be drawn.
Conclusion:
Our results show, among other things, that the average enforceability of a firm’s patent portfolio is positively related to the probability of acquiring and being acquired. Put another way, the estimated ability of a firm to enforce its property is important in its consolidation decision. At the same time, the amount of overlap in technology space helps to explain who will merge with whom. So, firms with enforceable intellectual property boundaries tend to be involved in mergers, and those with overlapping property rights tend to match up with one another.
While the causal connection with the overlapping variable suggests complementarities, the additional impact of our enforceability measure is consistent with industry anecdotal evidence which suggests that many of the mergers were rooted in conflicts over mutually blocking patents. In fact, a handful of mergers, e.g. Calgene/Monsanto and DeKalb/Monsanto, were completed in the midst of patent infringement suits.