The duration of research joint ventures: theory and evidence from the Eureka program

The duration of research joint ventures:theory and evidence from the Eureka program

K. Miyagiwa (Emory and Kobe) and A. Sissoko (LCU)

Introduction - 1RJV = partners (A) coordinate research efforts and (B) share innovation

Incentives for RJVsAvoid duplications (Katz 1986)Internalize technical spillovers (dAsprement and Jacquemin 1988, Kamien et al. 1992, Miyagiwa and Ohno 2002)Introduction - 2Instability of RJVs Lack of monitoring of R&D effort (free-rider problem)

Solutions to monitoring problems1. random termination 2. green-porter3. deadlines (Miyagiwa 2011)

Introduction - 3Theory: Pre-commitment to the dissolution of RJV at a pre-set date (duration)Optimal duration is positively related to innovation valuesIntroduction - 4Time consistency problemSolution for RJVsPrivate research grants have time limitsHelp from government regulationsRJVs are required to ask for permission from government to be exempted from antitrust lawsU.S. DOC Advanced Technology Program (ATP)Europe EUREKAFlow of the presentationTheory Model of optimal RJV durationsProperties of optimal RJV durations

EmpiricalData from EurekaMain estimation resultsRobustness checks

Part 1: TheoryInfinite horizon, discrete time t = 1, 2 m firms try to find a new product or technologyGoing it alone:v : expected value of R&D per firm (v 0).

RJV parametersRJV => share innovation, independent R&D effort

= value of innovation per partner

k = R&D cost (fixed)

q = (conditional) probability of failure per partner per time

qm = (conditional) joint probability of failure for RJV

RJV without monitoringRJV with an infinite durationNo monitoring and no punishing shirkingV = value of RJV per firm when everyone exerts effort

V = - k + (1 qm) + qmV

V = [- k + (1 qm)]/(1 qm )

Assumption 1: V > v (RJV is worthwhile)Unstable RJVShirking saves k but lowers (joint) probability of innovation, yielding to a shirker the payoffWd = (1 qm-1) + qm-1V

Assumption 2: V Wd < 0.V Wd = - k + qm-1(1 q)( V) < 0.

A one-period RJVAgree to dissolve RJV between t = 1 and t = 2

Equilibrium payoffR(1) = = - k + (1 qm) + qmv

Shirking yieldsRd(1)= (1 qm-1) + qm-1v

R(1) - Rd(1)= - k + qm-1(1 q)( v)

Prop 1:Given assumption 1 (V > v) and assumption 2 (V Wd < 0), there are ranges of parameters in which R(1) - Rd(1) 0.

Compare:R(1) - Rd(1)= - k + qm-1(1 q)( v) 0

V Wd = - k + qm-1(1 q)( V) < 0

Extending durationIf prop 1 holds, consider a two-period RJVR(2) = - k + (1 qm) + qmR(1).

An n-period RJVR(n) = - k + (1 qm) + qmR(n-1)

Properties of R(n) R(n) is increasing in n.As n goes to infinity, R(n) goes to VOptimal durationProp 2: If prop 1 holds, there is an optimal duration n*

Shirking (at date 1) yieldsRd(n)= (1 qm-1) + qm-1R(n-1)As n goes to infinity, Rd(n) goes to Wd

R(1) - Rd(1) > 0 As n goes to infinity, R(n) Rd(n) goes to V - W d < 0,

Properties of optimal duration (n*)Prop 3: An increase in tends to raise n*.

Proof: In R(n) appears with positive probability so an increase in raises R(n) Rd(n)= - k + qm-1(1 q)( R(n-1)).

Properties - 2An increase in the number of partners (m) has two effects: reduces (value per member)raises probability of successThe effect on R(n) and hence on n* are ambiguous.

Let the data determine the effect.Part 2: EmpiricalEuropean Eureka program (1985 )Promotes pan-European RJVs with subsidies and no-interest loansPartners are sought from separate countriesMonitoring problem exists as R&D conducted in different countriesRJVs required to pre-commit to durationsTime inconsistency problem is resolved.

Ideal for testing the theory

Data detailswww.eurekanetwork.orginitiation yeardurationcoststypes of industriesnames, addresses, and nationalities of all partners. identities and nationalities of RJV initiators. 1,716 Eureka RJVs started and completed (1985-2004)8,520 partners: 4,700 firms and 1,937 other partners (research centers or universities) from the EU-15Data summary

MethodologyEmpirically examine the factors determining the durations of the Eureka projects

Normality test fails

Duration or survival modelsProportional hazards models death as an eventHazard decomposes into a baseline hazard h0 and idiosyncratic characteristics of RJVshj(t)= h0(t) exp(xj x). Proportional hazard modelsCox model no restriction on functional form

Prior info specific functional form - Weibullh0(t) = ptp-1 exp(0)p determines the shape of a baseline hazardBaseline hazard increasing if and only if p > 1p = 1 : exponential hazard model

Strategy hereUse Weibull basic model (some ancillary evidence)Use other models for robustness

Hazard ratioHazard ratio = effect of a unit change in the explanatory variableHazard ratio < 1 => explanatory variable has a negative impact on RJV death (increases duration)Hazard ratio = 0 => explanatory variable has no impact

Explanatory variablesNo data on innovation valuesRJV cost per partner per month (in million euros) = main proxy of innovation values expected hazard ratio < 1 Number of partners - ?Initiator dummy firm initiated shorter durationsMulti-sector dummy multi-sector longer durationsInitiation year dummiesMain industry dummies

Table 2: Weibull

Robustness testingWeibull PH model assumes that all Eureka RJVs have a common baseline hazard, which is Weibull.

Model 6: questions the Weibull distribution assumptionCox (non-parametric) modelTable 3

Robustness checkModel 7: common hazard assumption stratified Weibull

Stratum 1: small (2 4 partners), 64 % of the samplesStratum 2: medium sized (5 - 8 partners),27.3 %Stratum 3 large RJVs (9 - 196): 8.7 per centResults: large RJV shape para. p significant at a 5% levelNo significant difference between the small and the med-sizeClose resemblance to V

Stratified Weibullh0(t)= exp(-13.030) (2.974)t j1.974 (small)h0(t)= exp(-14.029) (2.974)t j1.974 (medium-sized)h0(t)= exp(-13.030) (2.542) tj 1.544 (large)Large versus small and med-sized

Robustness checksModel 8: Hidden heterogeneity between data-wise identical RJVsfrailty Weibull test baseline hazard - Zh0(t); Z random

Model 9: make sure that time is not affecting the rsults exponential prop. Hazard model

ConclusionsTheoryRJV partners can overcome monitoring problems by committing to dissolve the RJVs at a fixed dateGovernment oversight of RJVs help the renegotiation problemOptimal duration depends positively on innovation valuesAmbiguous effect from the number of partners

Conclusions - 2Empirical evidenceEureka program ideal for testingProportional hazards modelsRJVs cost per partner has a positive effect on durationNumber of partners has a positive effect on durationFirm-initiated RJVs have shorter durationsMulti-sector RJVs have longer durations