Complementarities and spillovers in mergers: an empirical investigation using patent data

Abstract

We investigate the merger behavior of firms in the plant biotechnology sector using firm-level patent data for public and private firms in the 1980s and 1990s. Conditional logit estimation is used to estimate the probability that the firms will match in mergers and spinoffs. We calculate several patent portfolio-based measures of complementarity and spillovers between firms, and find that both are important to defining a good match of acquirer and target. However, complementarities provide the more robust explanation. The mergers and spinoffs observed in plant biotechnology may have been designed to overcome the anti-commons problem of mutually blocking technology, an extreme form of complementarity. Our results highlight the need to integrate patent and competition policy.

Keywords:

• mergers
• acquisitions
• patents
• plant biotechnology
• agricultural biotechnology
• complementarities
• spillovers

JEL Classification :

• L22
• L65
• O34

Acknowledgements

The authors would like to thank the Economic Research Service of the US Department of Agriculture for financial support.

Notes

Some recent papers address this shortfall (Hussinger 2005; Ornaghi 2009).

See Ali-Yrkkö, Hyytinen, and Pajarinen (2005) for another study investigating private firms using patent data.

2002 Monsanto Annual Report.

The HHI for patents is the sum of the squares of the firms’ market share of sample patents, so that the index can vary between 10,000 (all patents owned by a single firm) and close to zero (patents are diffusely held by many firms). Brennan et al. (2005) present a similar computation in terms of pre- and post-merger HHI for the US Department of Agriculture field trials instead of for patents.

See Ornaghi (2009) for a similar matching model in the pharmaceutical industry. Ornaghi controls for match-specific covariates, but does not account for target or acquirer-specific effects.

While the principal-agent terminology is useful for exposition, we do not employ a principal-agent model per se.

We consider only intra-industry mergers in this context.

It is presumed that the principal makes a unilateral choice. It is possible that for the agent there may be a more suitable principal (e.g. where the complementarities are greater). However, had this been the case, the agent should already have merged with the more suitable principal. Hence, at time t, all potential agents should be willing to merge with the principal, assuming that they are able to bargain for a share of the surplus. Another caveat is our implicit assumption that the most profitable acquisition actually makes the principal better off. This assumption is subsumed in our condition that the principal has already decided to make an acquisition, implying that it must be made better off by doing so.

Our data are monthly, so restricting a firm to one merger per month is not onerous. See Blonigen and Taylor (2000) for an analysis of annual data. They employ a Poisson count model to account for multiple mergers within a year.

Japanese firms are not included in the sample because they do not engage in mergers during the sample period. In general, they are unlikely to engage in mergers for reasons that are particular to their corporate structure. Our results are robust to whether or not they are included in the sample.

The sample of mergers is intended to be as complete as possible; however, it is dependent on the initial sample of plant biotech firms as defined by Graff, Rausser, and Small (2003). Because manual name searches were used within the Lexis–Nexis database, it is possible that some mergers were missed. Merging partners identified in Lexis–Nexis were added to augment the sample.   Our sample of plant biotech firms accounted for 70% of the firms identified as the top plant biotech companies by the US Department of Agriculture's Economic Research Service (ERS). See http://www.ers.usda.gov/data/AgBiotechIP, Table 11, ‘Top 100 patent holders, US and non-US, companies only (excluding subsidiaries)’.   The ERS list actually includes some universities and government agencies, with the top two patentees being the US Department of Agriculture and the University of California at Berkeley. After excluding Japanese firms and non-firms, our sample identifies 70% of the residual entities. Additionally, we identify some firms not found on the ERS list. However, the overlap is significant.

Company name searches are known to introduce some error. First, unknown subsidiaries will not be observed. Second, typographical errors at the patent office will make some patents unobservable. Our data are a subset of the true universe of patents owned by our sample firms. However, every effort was made to make the subset as close as possible to the whole.

Hall (1988) examines a sample of over 2000 mergers of publicly traded companies. It is not feasible to include all possible targets (all publicly traded firms) in a sample of that size, so she relies on sampling. A total of 2000 mergers by 2000 acquirers would imply at least 4,000,000 observations. While this may in itself be feasible, calculating the dependent variables for our study would require calculating the overlap variables for each observation – a task involving hundreds of millions of backward patent citations.

Where applicable, variables are measured in percentages rather than proportions. These scales aid in the interpretation of the parameter estimates.

The median merger in the 1970s, 1980s, and 1990s involves an acquirer that is nine to ten times the size of the target, in terms of market capitalization (Andrade, Mitchell, Stafford 2001).