| >Working Papers | |
|
| Working Paper 99-2 Research Note: KNOWLEDGE IN INTERNATIONAL CONSTELLATIONS January 29, 1999 VERSION 1.0-for discussion only. Please do not cite or copy without
permission. (781) 736-2264 John Hagedoorn Adam Jaffe Ben Gomes-Casseres is grateful to the Carnegie Bosch Institute for having funded part of his work on this project. Petia Topalova (MA candidate) and Ben Kriechel (PhD candidate) assisted in the research. © Benjamin
Gomes-Casseres, John Hagedoorn, and Adam Jaffe 1999 Alliances between firms from different nations are forging new units of economic power--groups of firms Gomes-Casseres calls "constellations." These constellations compete against other such groups and against traditional single firms. In such a world, the way firms manage the collaboration inside their constellation affects the competitive behavior and performance of the group as a whole. As a result, the performance of each firm comes to depend not only on its own capabilities and strategies, but also on those of its allies and on its relationships to these allies. One of the areas in which constellations may have an advantage over single firms is in the pooling and transfer of technological capabilities among member firms. Gomes-Casseres's previous case-based research suggests that member firms in a constellation cooperate in technology transfer and development more effectively than do unrelated firms. This paper is an attempt to test this finding on a broader sample of firms and using statistical methods.
Analytical Framework Capabilities. The rise of inter-firm collaboration has led to new empirical and analytical research on alliances. In our framework, single firms and constellations are alternative ways of controlling a set of capabilities. By capabilities, we mean the set of tangible and intangible assets that enable an organization to develop, make, and market goods and services. (This paper focuses on technological capabilities.) Control stands for the authority of a decisionmaker in using and deploying these capabilities. Simply put, the single firm has full control over all its capabilities; in a constellation, control over the set of capabilities of the group is shared among separate firms. At the same time, however, we expect that, compared to a collection of single firms, the alliances among members in a constellation facilitate the transfer and combination of capabilities of the member firms. Alliance. In this paper, an "alliance" is any governance structure to manage an incomplete contract between separate firms and in which each partner has limited control. Because the partners remain separate firms, there is no automatic convergence in their interests and actions. As a result, to deal with unforeseen contingencies the partners need to make decisions jointly. A contract is termed incomplete when, despite the fine print, it does not specify fully what each party must do under every conceivable circumstance. For many economists, the prevalence of incomplete contracts yields the basic rationale for existence of the firm. If such an incomplete contract is left to be managed by market principles, the parties--each acting in its own best interest--are likely to haggle over how to handle the "gaps" in the agreement. Integration is thus one way of governing incomplete contracts. But an alliance is also a way to manage the execution of an incomplete contract. Alliance agreements are typically open-ended and contain gaps typical of incomplete contracts. But, in contrast to full integration, alliances use some form of joint decision making to deal with unforeseen circumstances. This paper focuses on two activities typically thought to be subject to incomplete contracts: technology transfer and cooperation in the development of new technology. Because of the difficulties in monitoring inputs and outputs, in negotiating exchanges of value under conditions of uncertainty and asymmetric information, and in enforcing contracts, these activities are typically conducted better and at lower transaction cost within an integrated firm than between unrelated firms. We expect that when firms use alliances to transfer technology and to cooperate in technology development, these transaction costs would be lower than for unrelated firms, though they may still be higher than if the firms were fully integrated. As a result, we expect that more technology transfer and more cooperative technology development would take place among allied firms than among a comparable pair of non-allied firms. Patent Citations. To test this expected relationship, we use variables derived from cross-company citations in U.S. patents. According to U.S. patent regulations, every new patent granted to an inventor must cite the previous art upon which the new patent builds or that is closely related to the new patent. In other words, if an inventor in company 1 develops a new technology that is related to an earlier technology patented by an inventor in company 2, then the new patent for company 1 has to cite the older patent of company 2. The existence of a citation from company 2's patent to company 1's patent in itself does not imply a direct transfer of technology or joint development, nor does it require an alliance or contract. Two firms can be totally unrelated and have no communication with each other and still cite each other's patents. Yet, when firms are related to each other, and especially when they work jointly on new technologies or directly transfer technological knowledge between them, we expect that the citation pattern of their patents will reflect these cooperative activities. The reason for this expectation is that we interpret patents to be a reflection of an underlying technological capability, and citations among patents a reflection of relationships between specific technological capabilities. To summarize, we expect that allied firms will cooperate more on technology transfer and technology development than non-allied firms and that this bias will be reflected in a higher rate of cross-citations among allied firms than among a comparable pair of non-allied firms. Specific hypotheses are discussed below.
Data, Methods, and Variables For our statistical analysis, we combined data from two sources:
The two databases were combined by matching the firms. In other words, in our merged database, have information on all the alliances of each firm, as well as on all the citations of the firms and to the firms. Merging the data involved dropping some observations that could not be matched. The CATI sample included alliances among 733 different companies; but only 377 companies could be matched with a company from the patent database. The resulting merged sample of 377 companies contains 1,832 alliances, with sometimes more than one alliance between the same two partners. This sample contained the following self-explanatory variables used in the analysis:
We then constructed an additional series of variables to measure the citation patterns between companies. Two types of citation measures were constructed. "Citation frequency" measures the probability that any citation from company 1 is to company 2; "citation intensity" measures the probability that any patent of company 2 is cited by company 1. (These variables are asymmetric to the firm-pairs, and so were defined twice for each company pair: from 1 to 2 and from 2 to 1.) Specifically, these measures are:
Control variables. In addition to these independent variables that measure the effects in which we were interested, we constructed a series of controls variables. The first of these is a measure of the "similarity" between the technological capabilities of any two firm. The reason this is important is two-fold. First, we expect that two firms are more likely to cite each other's patents when their technological capabilities are similar, whether or not they are allied. Second, we expect that the degree of similarity of two firms may influence their propensity to form an alliance, though we could think of reasons why similarity and alliance propensity could be both positively or negatively related. At any rate, the degree of similarity needs to be a control variable in our analysis. The measure of similarity we used, developed earlier by Jaffe, calculated the extent of overlap between the number of patents of two firms when these patents are allocated to their "patent classes:" Similarity of patent portfolios of co. 1 and co. 2 
Sample of Non-allied Firms. Finally, and importantly, we needed a way to compare the citation pattern between allied firms to the pattern among non-allied firms. Because our sample was constructed by selecting allied firms from the CATI database and then matching them with the patent data, the 1,832 observations in the original sample contained only allied firms. A comparison sample of the same size was constructed by selecting 1,832 pairs of firms at random from the universe of all possible non-allied pairs. Since we have patent and citation data on all firms, regardless of whether they have an alliance or not, all the same variables defined above could be calculated for the sample of non-allied pairs. In order to calculate the variables involving the "year of alliance," we attached a random year to each of the non-allied pairs, making sure that the distribution of years in both samples was identical. In other words, we calculated for the non-allied firms the citation frequency "after" a certain year to compare with the citation frequency of allied firms "after" the alliance year; the same was done for citation intensity. Similarly, the "change" in citation patterns was measured for both samples with reference to either the alliance year or the randomly-chosen year. The fact that the distribution of years is identical in the two samples eliminates possible time-dependent biases introduced by our procedure. The final sample used in our analysis thus consisted of 3,664 observations,
of which one half were allied pairs from the CATI data and one-half
were randomly-chosen non-allied pairs. Descriptive statistics for
the variables in this final sample are below:
Hypotheses. The discussion above explained the effects that we sought to test, and the reasons for including certain control variables. In short, we expected the existence of an alliance as well as the number of alliances to have positive effects on all the four measures of citation probability. We expected the year of alliance to have no effect on the "after alliance" measures, but to be negatively related to the "change" in citation probability simply because when an alliance is formed in later years there are fewer years in the period after the alliance than before its formation. The similarity variable was expected to be positively related to citation probability "after" an alliance, but because we expected a comparable effect "before" an alliance, the "change" measure was expected to have no effect. We had no strong predictions about the size effects, though we felt that a large patent portfolio in absolute terms may give an ally greater scope for citing a partner's patents and so might be positively related to the citation probability measures. These hypotheses are summarized below:
Results We used linear least-squares regressions to test for these effects, and report standardized beta coefficients and significance levels in the tables below. (The standardized coefficient for any given variable shows the effect that one standard deviation change in that variable has on the dependent variable, again expressed in standard deviations. As such, the relative sizes of these coefficients can be taken to indicate the relative "importance" of each variable in accounting for the variance in the dependent variable, at least in this sample. The significance level of a variable shows the probability that the effect is different from zero, but does not indicate the size of the effect.) We are conscious of the fact that linear regressions may not be the optimal technique for our problem, in part due to the fact that our dependent variables are truncated at zero on the low end. We intend to address this issue in future versions of the paper. For now, we can say that the results presented below seem fairly robust and not sensitive to the various changes in specifications that we explored. The results of our tests are shown in the tables below:
Discussion The results for the most part are consistent with our hypotheses, though there remain some puzzles that we intend to address in future versions of the paper. The coefficients on "Allied" are positive and statistically significant in three of the four regressions, and usually are larger than the coefficients on other variables. In the regression (1) the coefficient on "Allied" is not statistically significant, but that on "No. of alliances" is, and again has the expected positive sign. The coefficient on the latter variable is not statistically significant in other regressions. Clearly, there is substantial multi-colinearity between these variables, as both are equal to zero in half the observations and equal to one in the bulk of the rest of the observations. Still, they measure the same underlying relationship that we sought to test and the results are consistent with our main argument that alliances facilitate the transfer and co-development of technology. Among the control variables, the most interesting results are relating to similarity and to size. "Similarity of patent portfolios" has positive coefficients in all regressions, and large, statistically significant effects in regressions (1) and (2). These results are consistent with the view that firms with similar portfolios have a greater probability of citing each other's patents, as we expected. In addition, however, the strong effect in (2) suggests that over time, the degree to which similar firms cited each other has increased; this appears to be true for both allied and non-allied firms. (This is consistent with the reports in other studies that over time, firms have increased the number of citations in their patents.) The two variables related to size had mixed results that we are still trying to understand. "Total patents of co. 2" has the strongest effects, notably in regressions (1) and (4). The results of regression (1) are to be expected; it suggests that when the company to which citations are made (company 2) has a large portfolio of patents, the citing firm (company 1) has a tendency to cite that company more than others. But the result in regression (4) is a bit more puzzling; it suggests that the probability that one company's patents are cited also increases with the size of that company's portfolio, all else being equal. Furthermore, this effect is stronger in regression (4) than in (3), suggesting that this probability has increased over time. One explanation for this pattern may be that as firms began to include more citations in their patents (see above), there were economies of scope in citing large firms, stemming perhaps from search costs in finding suitable citations among the patents of smaller firms. The last control variable, measuring the year of the alliance or the break-year in the non-allied pairs, also shows some unexpected results. The expected effect is in regression (4). This negative coefficient indicates that as we examine later break-years, the rate of change in citation intensity falls; this is consistent with the simple fact that there are fewer years after the break-year in these later years. But the effect in regression (2) was not expected. The positive coefficient there suggests that as we examine later break-years, the rate of change in citation frequency actually increases, regardless of the compression in time-frames. This is another indication that citation behavior was changing dramatically over time, though in ways unrelated to the alliance patterns. In further work, we intend to continue to unravel some of these puzzles and to include additional variables that may give insight into the patterns of technology exchange in alliances. For example, we intend to explore the effect of the type of alliance and also include consideration of the motivations for alliance formation. Conclusion This paper finds support for a specific but important effect of alliances. Much of the literature on alliances in high-technology industries speculates or illustrates with case studies that alliances may facilitate the transfer of technologies and the creation of new technologies through joint work. We set out to provide a statistical test of this idea and found results that are consistent with it. In future work, we hope to test the implications of this finding for the performance of constellations. If alliances facilitate the exchange of technology, as our results suggest, then it is reasonable to think that they may also lead to higher productivity in innovation and to higher overall performance. We will address this question in the next stage of this project. Endnotes
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
