The competition–innovation debate: is R&D cooperation the answer?

Abstract

Firms’ investment in research and development (R&D) depends on both product market competition and R&D cooperation. In this paper, we use a simple duopoly model of product innovation to show that firms should choose to enter into different cooperative R&D arrangements depending on competition. Equilibrium behavior implies different competition–innovation relationships conditional on the nature of the cooperative arrangements that firms join. Therefore, variation in the set of feasible modes of cooperation may explain why different competition–innovation relationships are observed in empirical studies based on field data. Experimental evidence confirms the presence of similar incentives for cooperating despite some deviations from predicted values of R&D.

Keywords:

• R&D cooperation
• investment
• innovation
• competition intensity
• experimental economics

JEL Classification:

• C92
• L13
• L41
• O31

Acknowledgements

We should like to thank two anonymous reviewers for constructive comments and suggestions. We should also like to thank seminar and conference participants at the University of Tromsø, Western Washington University, OFCE/DRIC, GREQAM, LAMETA, ESA meetings in Pasadena and Lyon, and the 3rd Nordic Conference on Behavioral and Experimental Economics in Copenhagen. This paper is part of the project ‘The Knowledge- Based Society’ financially supported by the Research Council of Norway (project 172603/V10). Remaining errors are our own.

Notes

Theoretical motives and empirical relevance for revealing technological information are surveyed by Lhuillery (2006).

Tingvall and Poldahl (2006) present results that show the nature and strength of the relationship between competition and innovation depends on the measure of competition (Herfindahl index versus price-cost margin) as well as the type of estimator used. In addition, they give a thorough review of the range of results in the empirical literature including six papers that find empirical evidence of a negative relationship in which more competition lowers R&D expenditures and four papers that find the opposite relationship. Finally, they cite four papers (including Aghion et al. 2005) that, like their own work, find evidence of a nonlinear relationship between product market competition and R&D.

R&D expenditures are an input to innovation. Increasing R&D expenditures may reduce innovative output in markets with non-exclusive property rights. Patent counts may also be a weak measure of innovative output since patenting is often done for defensive purposes and provides little benefits for new processes or products. See Gilbert (2006, 191) for a discussion.

De Bondt (1997) is a widely cited survey article. More recent reviews are provided by Sena (2004) and by Suetens (2005a).

See the survey of the small empirical literature in Suetens (2005a) and the discussion on data requirement problems. However, this may be too negative, reflecting the traditional skepticism among economists on the use of survey data, a view that seems to be slowly changing (see Bloom and van Reenen 2010). Several econometric studies based on innovation survey data suggest that cooperating firms spend more on R&D (Kaiser 2002; Tether 2002; Miotti and Sachwald 2003; Belderbos et al. 2004). There may be relevant studies from management and other adjacent fields. Jost and van der Velden (2006, 169), e.g. compare R&D cooperatives to mergers: ‘We note that mergers between firms is only an extreme form of an entire spectrum of horizontal cooperative arrangements \ldots we could also interpret a merger in the R&D contest in which insiders completely combine all previously separate firms as a Research Joint Venture.’ Hence, the literature on mergers and acquisitions may be relevant to the issues we discuss (Hagedoorn and Duysters 2002; Kleer 2012). A specific literature review on mergers and innovation is provided by Schulz (2007).

See Sørensen, Mattsson, and Sundbo (2010).

Most of the literature following the influential paper by d'Asprémont and Jacquemin (1988) is modeling deterministic R&D. Hauenschild (2003) provides a bridge by comparing stochastic and deterministic versions of some of the models. There is also a tournament literature surveyed by Reinganum (1989), studying the stochastic nature of R&D in oligopoly. See also Miyagiwa and Ohno (2002).

Setting α and δ either to 0 or 1, we span the entire parameter space which allows us to examine a wide variety of sharing arrangements. Intermediate values of α and δ would be an interesting extension, but would further complicate our experimental design.

Observe that convexity makes the relationship between innovation probability and R&D costs concave although the relationship between innovation probability and R&D by (1) is assumed linear for convenience. Therefore, we can look at the model as a reduced form reflecting an underlying structural form where nonlinearity depends not only on convexity of R&D costs, but also on possible nonlinearity between R&D and the innovation probability. This implies that the model is less restrictive than a literal interpretation suggests. We could, of course, explore the possible sensitivity of our results to the degree of nonlinearity by introducing a parameter in the R&D cost function that could be varied in the experiment. However, this would have added one or more additional treatments to an already complicated experimental design and therefore we have not pursued it here.

An alternative interpretation relates to the concept of absorptive capacity (Cohen and Levinthal 1989) – how quickly and to what degree firms can make use of new innovations. The common draw assumption we use in this paper is equivalent to assuming that firms are equal in their absorptive capacity.

This choice is not arbitrary. We wanted to keep the door open to the possibility of a comparison to the case where the thresholds for innovation success are drawn independently for each firm. It can be shown that in this independent draw case we get interior solutions for all sharing arrangements and competition levels if total knowledge is sufficiently small and 0.2 is just small enough.

To economize on space, we do not go through all the individual cases leading to the equilibrium values. A complete list of derivations is available from the authors upon request.

Beside the two pure Nash equilibria, there is a mixed strategy equilibrium. In the Symmetric/NO/soft case when 500 is played with probability 0.4. The mixed strategy equilibrium, however, is payoff dominated by the pure Nash strategy 900.

As for NO/Soft, the mixed equilibria are all payoff dominated by the aggressive pure Nash strategy.

Instructions are contained in the appendix.

Each sharing treatment had an equal number of periods with Soft, Moderate and Tough. To accomplish this, we generated a sequence of Soft/Moderate/Tough for each of the four blocks. For example, in Block 1, we began by labeling five index cards with Soft, five with Moderate and five with Tough. We then shuffled the deck and recorded the order for each of the fifteen periods in Block 1. We then reshuffled and repeated the process for Blocks 2, 3 and 4.

This is similar to some of the treatments in Sacco and Scmutzler (2011). It is in contrast with Suetens (2008) who explicitly examines endogeneous product market competition in an experimental setting.

The experiment was conducted using the z-Tree software package (Fischbacher 2007).

It is worth noting that we do speak to the possibility of subjects’ choices to cooperate (or not) by running an additional session under simplified conditions in the concluding discussion.

It should be noted that in making this prediction, we are assuming that the firms are able to take advantage of arbitrage opportunities, with the support of an appropriate side-payment. An alternative is to consider the default cooperative arrangement to be NO and requiring that both firms in an industry benefit in order for us to predict a move to an arrangement with more sharing.

Note that we made the decision to keep sharing exogenous in the experimental design in order to avoid introducing potential behavioral reactions that could make it harder to isolate the effect of sharing on investment. This means that our experimental data is not allowing a test of sharing choices, but rather an examination of what kinds of incentives to share that exist for our experimental subjects. We return to the possibility of endogenizing the decision to share at the end of the paper.

As we discussed when examining the overall results in Table 4, we conducted a series of one-sample signed-rank tests to compare the predicted values in Table 6 with the experimental data. We find that the experimental data differs from the Nash R&D prediction in 4 of the 12 cases.

Again, we use the a one-sample Wilcoxon signed-rank test to compare the predicted values of R&D to the experimental results and find deviations from the Nash prediction in 4 out of 12 cases.

If the proportion of ALL had been reduced from the very high 0.73 to a slightly lower 0.68, the regular U shape would have flipped to an inverted U if average investment under NO and ALL had remained unchanged.