You are on page 1of 20

Prepared for Information and Organisation: Essays in Honour of Donald M. Lamberton, S. McDonald and J. Nightingale (eds).

, Elsevier: Amersterdam, 1998 (forthcoming).

A Strategic Theory of In-House Research and Development*
by

JOSHUA S. GANS Melbourne Business School University of Melbourne
First Draft: 12 November, 1996 This Version: 11 June, 1997

This paper considers the question of whether research and development is optimally undertaken within the firm responsible for employing productive ideas or by independent researchers contracted with it. It is demonstrated that, in addition to the incentive considerations identified in the property rights literature, firms have an additional strategic motive to undertake in-house research and development. In contrast to the previous literature, when the possibility for multiple research teams are explicitly taken into account, an inhouse team serves to provide the firm with an advantage in negotiations with independent researchers. Therefore, it is possible that firms could inefficiently conduct their own research. Journal of Economic Literature Classification Numbers: Bargaining Theory (C78), Vertical Integration (L22) and R&D (O32). Keywords. bargaining, renegotiation, research and development, vertical integration.

*

Discussions and comments from seminar participants at the Australian National University and especially Catherine de Fontenay and Scott Stern are greatly appreciated. Of course, all responsibility for views expressed lies with the author. All correspondence to J.Gans@mbs.unimelb.edu.au. The latest version of this paper is available at: http://www.mbs.unimelb.edu.au/home/jgans.

2 A long standing issue in the economics of technology is why private firms conduct their own research and development. This was viewed as a puzzle since once knowledge is produced it is difficult to exclude others from using such knowledge (particularly one’s competitors). Thus, it was argued that the fruits of knowledge production were more difficult to appropriate than other forms of production (see Nelson, 1958; Cohen and Levinthal, 1989). Even when innovations are patentable, there are spillovers associated with knowledge production that mean that the social value of innovations are less than their private value (Arrow, 1962; Dasgupta and Stiglitz, 1980; Romer, 1990; Bresnahan and Trajtenberg, 1994). However, it was recognised by some researchers that having an in-house research capability was important if one was to capture spillovers from others. Lamberton (1980) argued that information use required an understand of the language of a particular technology. Hence, firms who manufactured products would have to have some ability to research simply to learn that code associated with a given innovation. A similar line of reasoning was present in Cohen and Levinthal’s (1989) conception of “absorptive capacity.” In situations in which spillovers were not considered as important, attention was been drawn to the difficulties associated with writing contracts for innovative activity. Aghion and Tirole (1994) argued that it is often impossible to contract on the outcomes of innovative activity ex ante. Hence, if firms conduct their own research, scientists face limited incentives to direct effort towards producing innovations that firms value most. Instead, if they were to sell such innovations to the firm, they would face greater incentives as their returns from innovative activity would be tied directly to the economic value of any

3 resulting innovations.1 In the absence of wealth effects, the observed portion of in-house research and development reflects an efficient assignment of property rights of the fruits of innovation (Grossman and Hart, 1986). However, Aghion and Tirole (1994) argue that scientists are often wealth constrained and hence, cannot develop innovations to their full economic value. This may be because of the difficulties smaller firms face in financing innnovative activity (Arrow, 1982). Hence, the initial assignment of property rights might be impaired. Contracting difficulties are even greater when property rights are weak or absent. Anton and Yao (1994) investigate situations where innovations, once generated, are not patentable. This could be because there are too closely related to other inventions or that they are of a non-patentable concepts for new products or management systems. Thus, scientists face a problem of expropriation with potential developers. In negotiations, they must reveal their idea and are, therefore, at the risk of that idea simply being appropriated by the other party. Anticipating this, they choose not to engage in innovative activity. Anton and Yao (1994) show how competition can resolve this dilemma. In negotiations, if expropriation occurs, the scientist can credibly threaten to give the idea to competing firms. Hence, the potential developer opts to pay the scientist not to reveal the idea to others. This paper also focuses on contracting difficulties associated with innovative activity. Here, like Aghion and Tirole (1994), I assume that individual scientists or small research firms face wealth constraints in developing any innovations they might generate. Thus, they must form a relationship with a larger firm if the innovation is to have any value. But I also adopt the view that the original innovators are also essential to ensure the innovation was subsequent value. In complex technical environments, the knowledge of innovators is not simply codifiable and, as such, their human capital would be required to properly develop the innovation. Regardless of the patentability or otherwise of the innovation, an acquiring firm will need to employ the scientists responsible for the

1

Chesbrough and Teece (1996) also argue that independent researchers have more higher powered incentives than their in-house counterparts. However, the latter may have an advantage when it comes to developing systemic innovations that require more radical product redesigns.

4 innovation. Thus, scientists are not at risk of expropriation of the kind discussed by Anton and Yao (1994). In this environment, we seek to understand the conditions under which a larger firm might choose to employ scientists before innovative activity has taken place -- that is, conduct in-house research and development -- versus acquiring their research units and employing them after innovative activity has taken place and it has been successful -independent research and development. A critical driving force in this model will be the lack of an ability to write binding employment contracts in this environment. The

inalienability of human capital means that scientists must be employed. However, those scientists cannot be bound to continue working for the firm until an innovation is fully developed. Given the uncertain nature of this activity, it is difficult to know when such activity is actually complete and hence, it is not possible verify whether contractual conditions have been met. Similarly, the firm can choose to end a scientist’s employment, so employment contracts are cannot be binding in that regard. With non-binding

employment, we must, therefore, take care in ensuring outcomes are stable with respect to renegotiation (Stole and Zwiebel, 1996). This paper offers a strategic explanation of the degree of in-house research and development. This strategic explanation extends the model of de Fontenay (1997) to the particular characteristics of innovation. She shows that firms might choose to partially integrate with suppliers even when there are efficiency losses in so doing. This is because that integration improves their ex post bargaining position with independent suppliers. The approach of this paper is to build upon de Fontenay’s insight to understand how the integration process occurs in a market environment prior to any productive activity taking place. Below it will be shown that for there to be a benefit to the firm from in-house research it must affect a scientist’s ability to leave the firm after innovative activity has taken place. While an independent scientist is free to enter into any contract with the firm so long as they have a successful innovation, the in-house scientist after innovative activity has

5 taken place cannot renegotiate their contract. Since wage payments have been made, the firm also cannot renegotiate the contract. The reason for these binds becoming effective ex post when they were not available ex ante can be rationalised as follows: we view firms as using scientists to generate a research capability (as in Nelson and Winter, 1982 among others). This capability rests with the employees of the firm. If this capability is generated in-house, no single scientist can remove that capability from the firm. If this capability is generated independently, then it requires the full employment of the scientist in order to be realised.

I

The Basic Set-Up

Assume there there is a single firm in an industry and a pool of 2 potential scientists or scientific teams.2 Each scientist, if they undertake innovative activity, has a given probability p > 0 of generating an innovation that is of value to the firm. Suppose that x ≤ 2 such innovations are realised. Then the value to the firm of those innovations is R(x). The only assumptions we make about R(.) is that it is non-decreasing for all x and that R(0) = 0. Thus, the firm potentially values all innovations. Moreover, the firm is the only agent that values innovations. One might view the firm as a monopolist in the relevant product market. As such, scientists cannot bring innovations to market on their own. Finally, both scientists and the firm are assumed to be risk neutral. The basic timeline is as follows: Stage 1 (Integration): From the 2 scientists, the firm bargains with a subset, I, to become in-house scientists, while the remainder, n = 2 - I, become independent researchers. Stage 2 (Innovation): Research activity is conducted. Initially, it is supposed that the productivity of scientists does not depend on whether they are in-house or independent. This assumption is relaxed in a later section. With probability, p, a scientist produces a
2

The model below can be readily extended to the case of N scientists without any change in the conclusions offered. The general case is omitted here for ease of exposition.

6 useful idea. Failure yields no benefit.3 Both the firm and scientists know what innovations have been realised, thus, avoiding problems associated with information asymmetry. Stage 3 (Acquisition): Having observed the set of successful innovations, the firm bargains with independent scientists over payment for useful ideas. Given our

assumptions, the expected number of these is a = np and the expected number of in-house ideas is b = Ip. Thus, realised revenue is R(a+b). Note that this bargaining is simply over the fee. Recall that the scientist or firm can always reopen negotiations with the firm prior to the value from an innovation being realised. We will, therefore, examine solutions to this bargaining game that are immune to such renegotiation. This will rule out benefits to sequential contracting at this stage (see McAfee and Schwartz, 1994). The incentives for the firm to integrate into research in stage 1 depend on the expected outcomes in later stages. We will show that, despite the fact that there is no productive or motivational benefit to integration, firms have an incentive to integrate to improve their bargaining position over innovation payments. This is the effect notice by de Fontenay (1997). However, the costs of integration are endogenous in that scientists take into account their forgone license fees in stage 3 in bargaining over their wage within the firm at stage 1. Thus, it is possible that, despite the stage 3 strategic benefits to integration, the firm might find it too costly to do so.

II

Stage 3 Benfits to Integration

To solve the model, we work backwards, considering stage 3 outcomes first. Here we demonstrate the strategic benefits to integration that are realised in stage 3. These

mirror the effects identified by de Fontenay (1997) in her application of the Stole and Zwiebel (1996) wage bargaining model to integration. These bargaining issues only arise when complete integration has not already occured. The consequence of complete

3

One might suppose that one can still yield the benefits of “absorptive capacity,” even in the event of failure (see Lamberton, 1980; Cohen and Levinthal, 1989).

7 integration is that firms do not have to bargain over subsequent payments and simply realise R(.). Ex ante, these expected revenues are π ( I = 2) = p 2 R(2) + 2 p(1 − p) R(1) . If one scientist is in-house and the other is not, we assume that, provided the independent scientist has generated an innovation, that scientist and the firm engage in Nash bargaining over the acquisition price and subsequent wage for the scientist, with equal ex post bargaining power. If the in-house scientist has not produced an innovation, the amount, t, paid to the independent scientist is t = 1 R(1). If the in-house scientist has 2 produced an innovation, t = 1 ( R(2) − R(1)) . Thus, the expected revenues from I = 1 are: 2
3 π ( I = 1) = p 2 1 ( R(2) + R(1)) + p(1 − p) 2 R(1) , 2

while

the

expected

payment

is:

T ( I = 1) = E[t I = 1] = 1 p(1 − p) R(1) + p 2 1 ( R(2) − R(1)) . 2 2 Bargaining when there is no integration is more complex. If only a single

independent scientist has a successful innovation, then simple Nash bargaining takes place and the payment is t = 1 R(1). When both scientists negotiate over successful innovations, 2 the bargaining environment is as is formulated by Stole and Zwiebel (1996). To describe this, note that when negotiations between the firm and a single scientist break down, the firm is left with the outcomes of negotiations with the remaining scientist. In that case, the ˜ payment negotiated is t (1) = 1 R(1) and the firm realises 2
1 2

R(1). Given the non-binding

nature of employment, this becomes the firm’s outside option when negotiating with both scientists. As such, splitting the surplus in this case implies that: ˜ ˜ ˜ R(2) − 2t (2) − ( R(1) − t (1)) = t (2) . Rearranging and substituting in ˜ t (1) , we have ˜ t (2) =
1 3

( R(2) − 1 R(1)) 2

and

˜ π (2) = 1 ( R(2) + R(1)) . Notice that this payment exceeds the payment agreed upon in the 3 single scientist case if and only if R(2) > 2R(1), that is, if there are increasing returns to innovations. This might be the case if the two innovations are complements in production

8 or demand creation.4 In contrast, if there are decreasing returns to innovations, an

independent scientist becomes worse off if the firm has another scientist in-house. The expected payment is: ˜ ˜ ˜ T ( I = 0) = E t I = 0 = p 2 t (2) + p(1 − p)t (1) = p 2 1 ( R(2) − 1 R(1)) + p(1 − p) 1 R(1) . 3 2 2 Thus, expected profits from I = 0 are:

[

]

π ( I = 0) = p 2 1 ( R(2) + R(1)) + p(1 − p) R(1) . 3
Observe that greater integration raises expected revenues in stage 3. As de Fontenay

(1996) notes, the expected threat point of the firm in negotiations with scientists is greater the more scientists there are in-house. Under decreasing returns to innovation an

independent scientist does not expect to capture as much of the surplus from innovation when the firm has another scientist who has potentially produced an innovation of value. This analysis has assumed that scientists provide more than a commodifiable innovation to the firm. They acquire a capability that requires their continued cooperation in order to generate economic value. In contrast, one could imagine stronger property rights regimes in which the innovation generated is commodifiable. In this case, when a deal is struck with one scientist, there need be no concern about subsequent renegotiation with that scientist. While this does not change outcomes for the complete or partial integration cases, it does potentially alter the bargaining solution for the non-integration case. It is not difficult to show, however, at the expense of some notational complexity, that the strategic benefits of in-house research and development continue to hold. For

instance, suppose that the firm engages in sequential bargaining where one scientist is chosen at random and one-on-one bargaining takes place with the expectation that an agreement will result in a different outcome with the remaining scientist than a breakdown.5 Therefore, the results here are likely to continue to hold for stronger property rights regimes.

4

For example, they might be complementary product and process innovations (Athey and Schmutzler, 1995). 2 5 For this protocol, it is easy to show that π ( I = 0 ) = p 1 ( R( 2 ) − 2 R(1)) + pR(1) . 4

9

III

Stage 1 Costs of Integration

The above analysis shows that the firm might have an incentive to have in-house scientists despite there being no productive advantages to so doing. However, that analysis completely neglects the cost hiring such scientists. This, as Aghion and Tirole (1994) show, is a critical issue in predicting the level of integration actually chosen. When looking at bargaining between a single independent scientist and the firm, they demonstrate that when the firm has all of the ex ante bargaining power, the allocation of property rights ex ante is efficient in that the scientist will own the innovation if their effort is most critical for the research being successful and the firm will own it otherwise. In contrast, when the scientist has all of the ex ante bargaining power, a socially inefficient allocation of property rights could occur. However, it does not seem a priori reasonable to suppose that the relative bargaining power between the parties will necessarily change between the ex ante and ex post bargaining stages. One can think of reasons such as a “fundamental

transformation” to bilateral monopoly (as in Williamson, 1985) but Aghion and Tirole (1994) do not address this. Therefore, it seems reasonable to begin with the situation in which ex ante and ex post relative bargaining power between scientists and the firm are equal. In the case of the model discussed thusfar, this means that the firm and scientists have equal bargaining power. Therefore, I will consider the stage one bargaining game using the same assumption.6 Continuing on from the previous section, I will assume, however, that there are no productivity differences between in-house and independent scientists. This assumption will be relaxed in the next section. The wages received by in-house scientists are the outcome of bargaining between the scientists and the firm in stage 1, prior to any innovative activity taking place or inventions being realised. They are, therefore, related to expected stage 3 outcomes. To see this, suppose that the firm wished to employ both scientists in-house. As the firm has

6

Stole and Zweibel (1996) demonstrate that their bargaining solution can be extended to an arbitrary assumption of relative bargaining power between the parties. Thus, the results here could readily by extended to this case.

10 to negotiate with both scientists individually over their wage, once again the bargaining formulation at this stage is akin to that of Stole and Zwiebel (1996). Here, however, if negotiations break down, scientists have an outside option of obtaining their expected payment in stage 3. This amount is T(I=1), as a scientist leaving negotiations can expect the other scientist to become integrated. However, the cases of decreasing and increasing returns to innovation are different in this regard. Therefore, we will deal with each in turn. Suppose that there are decreasing returns to scale (i.e., 2R(1) ≥ R(2)). In general, equal division of the surplus in stage 1 implies that: ˜ ˜ ˜ π ( I = 2) − 2 w(2) − (π ( I = 1) − w(1)) = w(2) − T ( I = 1) ˜ ⇒ w( 2 ) = =
1 3 1 3

˜ ( p ( R(2) − R(1)) + p(1 − p) R(1) + w(1))
2

˜ (π ( I = 2) − π ( I = 1) + T ( I = 1) + w(1))

.

Using Nash bargaining with a single scientist, it is easy to see that ˜ w(1) = 1 (π ( I = 1) + T ( I = 0) − π ( I = 0)) 2 =
1 2

(p

2 1 2

R(2) + p(1 − p) R(1))

˜ ˜ What is critical in this determination is that w(2) and w(1) satisfy Stole and Zwiebel’s ˜ ˜ (1996) feasibility condition. This requires that w(2) ≥ T ( I = 1) and w(1) ≥ T ( I = 0) . These conditions will be satisfied if the surplus in efficient bargaining covers each party’s outside options. If this were not the case, bargaining would optimally breakdown, with each party receiving their outside options. Under decreasing returns, feasibility is satisfied. It is easy to see that

˜ w(1) ≥ T ( I = 0) as the generated surplus, π (I=1), is greater than the sum of outside options, π(I=0) + T(I=0). The firm’s strategic benefit from hiring the scientist ex ante ˜ ˜ exceeds the payment the scientist expects to receive. Substituting w(1) into w(2) , we have, ˜ w( 2 ) =
1 3

(p (

2 5 4

3 R(2) − R(1)) + 2 p(1 − p) R(1)) .

Once again, it is easy to check that this is greater than T(I=1) and, hence, is feasible. Therefore, the expected profits, v(I=2), of the firm are:

11 ˜ v( I = 2 ) = π ( I = 2 ) − 2 w ( 2 ) = pR(1) + p 2 1 ( R(2) − 2 R(1)) 6 Under increasing returns to innovation, however, the feasibility conditions are not satisfied. This is because π ( I = 1) < π ( I = 0) + T ( I = 0) . The firm’s strategic benefit is outweighed by the scientist’s expected payment. Thus, negotiations with a single scientist will result in that scientist not being hired. Therefore, the firm receives π(I=0) in the event of a breakdown when it is negotiating to hire both scientists. In this case, equal division of the surplus implies: ˜ ˜ π ( I = 2 ) − 2 w( 2 ) − π ( I = 0 ) = w( 2 ) − T ( I = 0 ) ˜ ⇒ w(2) = 1 (π ( I = 2) − π ( I = 0) + T ( I = 0)) 3 However, as π ( I = 2) − π ( I = 0) = 2T ( I = 0) , the feasibility condition is just satisfied in ˜ this case and w(2) = T ( I = 0) . Hence, under increasing returns: ˜ v( I = 2 ) = π ( I = 2 ) − 2 w ( 2 ) = π ( I = 2 ) − 2 T ( I = 0 ) = π ( I = 0 ) . This is the same as the expected payoff in stage 3 when no scientists are hired. Now we turn to determine the returns to the integration of a single scientist. In this case, if negotiations with one scientist breaks down, the firm can go to the other scientist. The wage in that instance would be equal to
1 2

(π ( I = 1) − π ( I = 0) + T ( I = 0)) . Therefore,

when there is a replacement scientist, equal division of the surplus implies that: ˜ ˜ π ( I = 1) − w(1) − 1 (π ( I = 1) + π ( I = 0) − T ( I = 0)) = w(1) − T ( I = 0) 2 ˜ ⇒ w(1) = 1 (π ( I = 1) − π ( I = 0) + 3T ( I = 0)) = p 2 4
1 24

(7 R(2) − 2 R(1)) + 1 p(1 − p) R(1) 2

This wage is only greater than T(I=0) if 2R(1) > R(2). Once again, under increasing returns, the firm chooses not to hire a single scientist in stage 1. However, under

decreasing returns, we can derive the expected profits from having a single in-house scientist,
3 v( I = 1) = 1 π ( I = 0) + 4 (π ( I = 1) − T ( I = 0)) 4

= pR(1) + p 2

5 24

( R(2) − 2 R(1))

Finally, note that the expected profits from having no in-house scientists are:

12 v( I = 0) = π ( I = 0) = pR(1) + p 2 1 ( R(2) − 2 R(1)) . 3 Having derived these payoffs a simple comparison leads to the following proposition. Proposition 1. If there are increasing returns to innovation (i.e., R(2) > 2 R(1) ), the firm chooses 0 or 2 in-house scientists. If there are decreasing returns to innovation (i..e, R(2) < 2 R(1) ), the firm chooses 2 in-house scientists. If innovations are separable in firm revenues (i.e., R(2) = 2 R(1) ), the firm’s expected profits are invariant to the number of in-house scientists. Thus, the desire of the firm to have in-house scientists is contingent on how innovations determine firm revenue. While there is a stage 3 benefit to having in-house scientists always, the amount paid to independent scientists in that stage is much greater if there are increasing returns to innovation. This is because, if two innovations are generated, the marginal contribution of each individual scientist to firm revenue is very high. Thus, an independent scientist can extract higher payments in this stage if they have an innovation to sell. This effect creates a non-convexity in the benefits to having in-house scientists. So, when there are increasing returns, the firm chooses either to have no in-house research or to have all research in-house. Hiring a single independent scientist is never optimal for the firm. When there are decreasing returns, these effects are reversed. Scientists extract lower stage 3 rents from innovation, and while this reduces the bargaining benefits to having an in-house workforce, it also reduces scientist’s position in wage bargining in stage 1. Consider the position of a scientist bargaining with a firm who wishes to integrate both scientists. If bargaining breaks down this scientist will obtain its expected payment when they are the only independent scientist. But where are the strategic benefits for the firm here? If it hires the scientist it cannot improve its stage 3 bargaining position with other independent scientists as there are none of these. However, if bargaining breaks down it will be forced to renegotiate its contract with the remaining in-house scientist. That scientist will be able to extract part of the firm’s strategic benefit to in-house research. Thus, under renegotiation, in-house wages will rise. As such, the firm’s strategic

advantage in hiring the last independent scientist is the improvement in its bargaining position with other in-house scientists. In the many scientist case, both of these effects will

13 be present. That is, hiring an in-house scientist gives the firm both a stage 3 and stage 1 bargaining advantage. Both of these effects mean that scientists are better off being inhouse than independent and this reduces their wage claims, lowering the costs of integration and making complete integration optimal.

IV

Productivity Differences Independent Scientists

Between

In-House

and

Thusfar, our analysis has assumed that the productivities of in-house and independent scientists do not differ. However, there is much research that suggests that there are such differences (e.g., Aghion and Tirole, 1994). We capture these differences by assuming the probability of a successful innovation differs among in-house versus independent researchers. We see this as arising from motivational and resource related effects rather than instances of absorptive capacity. Those effects are independent of the success or otherwise of in-house research, whereas absorptive capacity means that even failed innovative activity has value (Cohen and Levinthal, 1989). Suppose that the probability of success of an independent scientist is p and that of an in-house scientist, q. At this point, we do not specify which is greater. However, this will become of relevance as the analysis proceeds. Under this change in assumptions, it is easy to show that the stage 3 expected profits and stage 3 payments are as follows:

π ( I = 2) = q 2 R(2) + 2 q(1 − q ) R(1) π ( I = 1) = pq 1 ( R(2) + R(1)) + (q(1 − p) + p(1 − q ) 1 ) R(1) 2 2
T ( I = 1) = 1 p(1 − q ) R(1) + pq 1 ( R(2) − R(1)) 2 2

π ( I = 0) = p 2 1 ( R(2) + R(1)) + p(1 − p) R(1) 3
T ( I = 0) = p 2 1 ( R(2) − 1 R(1)) + p(1 − p) 1 R(1) 3 2 2

14 However, before describing the stage 1 expected payoff we need to specify the feasibility conditions for bargaining at this stage. ˜ w(2) ≥ T ( I = 1) , 8 p( 1 q − 2 p) R(1) 3 ≤ (≥) (i) 2 if ( R(2) − 2 R(1))( p − q ) ≤ (≥)0 p−q R(2) − 2 R(1) (ii) − 1 ( p + q ) ≤ (≥) 2 R(1) if ( R(2) − 2 R(1))( p − q ) ≤ (≥)0 R(2) − 2 R(1) ˜ That is, for (i) w(1) ≥ T ( I = 0) 7 and (ii)

Note that as q becomes equal to p, these conditions collapse to the decreasing returns to innovation case. Note that if p > q, then the left hand side of (i) is less than the left hand side of (ii). Thus, if p > q, under decreasing returns, (ii) implies (i) and under increasing returns, (i) implies (ii). If q > p, then the left hand side of (i) is greater than the left hand side of (ii). Thus, if q < p, under decreasing returns, (ii) implies (i) and under increasing returns, (i) implies (ii). If these feasibility conditions hold, then using the same methods as above we can calculate: v( I = 0) = p 2 1 ( R(2) − 2 R(1)) + pR(1) 3
3 v( I = 1) = p( 8 q − p 1 )( R(2) − 2 R(1)) + 6 1 4

( p + 3q) R(1)

v( I = 2) = 1 q(2 q − p)( R(2) − 2 R(1)) + qR(1) 6 If both conditions (i) and (ii) do not hold then, the firm does not choose to have in-house research as stage 1 hiring does not generate a positive bargaining surplus. If only condition (i) holds, then I = 2 is never chosen. If only condition (ii) holds, then while I = 1 is never chosen and the payoff from choosing I = 2 is: v( I = 2 ) = π ( I = 2 ) − 2 T ( I = 0 ) > π ( I = 0 ) = v( I = 0 ) Thus, the firm prefers complete integration.

7

Note that this condition is exactly the same regardless of whether bargaining is occuring for I = 1 or I = 2. That is, the condition is π ( I = 1) − π ( I = 0 ) ≥ T ( I = 0 ) . 8 This condition is π ( I = 2 ) − π ( I = 0 ) ≥ 2 T ( I = 0 ) .

15 While a complete characterisation of the optimal choices is cumbersome, the result of interest is whether integration will occur in conditions when non-integration involves a higher rate of innovation. For instance, suppose that p > q. In this case, non-integration is socially optimal. However, as the following proposition occurs, this is not necessarily the equilibrium outcome. Proposition 2. Suppose that p > q. If R(2) ≥ 2R(1), then non-integration (i.e., I = 0) is always chosen. If R(2) < 2R(1), then if (i) holds, non-integration (i.e., I = 0) is never chosen, being dominated by partial integration (i.e., I = 1). If (ii) holds, there exist parameter values such that either complete integration (i.e., I = 2) or partial integration is chosen. If neither (i) nor (ii) hold, non-integration is chosen. The proof is in the appendix. Basically, so long as there is sufficient decreasing returns to innovation or the probability of success of independent and in-house scientists are sufficiently high, some form of integration is chosen by the firm at stage 1. Under these conditions, the stage 3 strategic benefits to integration are relatively high and hence, justify the stage 1 costs of integration. Now turning to the case in which integration is socially optimal, that is, q > p, we can show that non-integration is never chosen. Proposition 3. Suppose that q > p. If R(2) ≥ 2R(1), then complete integration is always chosen. If R(2) < 2R(1), then non-integration (i.e., I = 0) is never chosen, being dominated by partial integration (i.e., I = 1). There exist parameter values such that either complete integration (i.e., I = 2) or partial integration is chosen. Once again the proof is in the appendix. This is a similar conclusion to Proposition 2 except that integration is favoured under different conditions. Partial integration can

actually be optimal if the marginal revenue of the second innovation is very small and the differential in productivity is very high. For instance, p = 1/20; q = 1; R(1) = 1 and R(2) = 1.2 yields a value of v(I=1) ≈ 0.75 > 0.74 = v(I=2). Because in-house innovation is relatively more productive, then the marginal return to having more than one in-house scientist is relatively low, especially given the low value of a second innovation. The result in examples such as this hinge upon the low wages that need to be paid to in-house scientists. These low wages occur because the expected stage 3 rents are low and an individual scientist is replaceable in stage 1 negotiations. Thus, despite the large

16 productivity differential, the firm chooses to leave one scientist independent so as to lower the wage of the in-house scientist. In summary, while it is the case that the relative productivity of in-house versus independent scientists alters the equilibrium likelihood of integration, the firm choice only accords with the social optimum in the case of increasing returns to innovation. Under this assumption, the strategic benefits to integration are completely absorbed in stage 1 hiring costs so no distortion occurs. Under decreasing returns, there is a net strategic benefit to integration. This means that integration is observed when it is not socially optimal, but also that partial integration can result under conditions when complete integration is socially optimal.

V

Concluding Remarks and Future Directions

This paper has demonstrated two things. First, firms have a strategic incentive to undertake their own internal research and development as this improves their expected bargaining position with independent scientists producing either complementary or substitutable innovations. This incentive is still present in equilibrium when the complete ex ante costs of integration are taken into account. Second, when there are productivity differences between in-house and independent scientists, while for the case of complementary innovations, the decentralised equilibrium is socially efficient, this need not be true when innovations are imperfect substitutes. Hence, the strategic incentives of firms to conduct their own research and development and their own anticipation of ex post bargaining can lead to allocations of the property rights over innovations that reduce the rate of innovation overall. Nonetheless, these conclusions have been made using a somewhat simply view of the innovation process. In reality, research activity is a dynamic process where resources are expended over time with the hope of accelerating the rate of innovation. In Gans and Stern (1997) an explicitly dynamic process of innovation is considered, based on the

17 traditional patent race literature in industrial organisation (see Reinganum, 1989). In that paper, incumbent firms have a strategic incentive to develop their own research capabilities in order to influence potential bargaining with independent research teams. However, the mechanism of that strategic advantage is somewhat different from the one presented above. Rather than being concerned about the ability of in-house versus independent researchers to renegotiate their employment contracts, Gans and Stern (1997) demonstrate the, by conducting their own research, a firm is able to become relatively more patient during negotiations. This is because in that bargaining model negotiations take time and hence, a firm with its own in-house research team has an ability to continue research during negotiations. Therefore, they could potentially create an innovation and pre-empt further negotiations leaving independent scientist “out in the cold.” This possibility means that any bargaining struck between the two parties must compensate the firm for its option value of continued research. Hence, a firm with its own in-house scientists can appropriate a greater share of the rents from independent innovation. All this suggests quite strongly that looking at the strategic implications of in-house research and development represents a fruitful complement to existing theories of in-house research.

18

Appendix
Proof of Proposition 2 If R(2) ≥ 2R(1), (ii) and hence, (i) does not hold. Hence, non-integration is always optimal. If R(2) < 2R(1), then if (ii) holds, (i) holds as well. Suppose that (ii) holds. Comparing I = 2 and I = 1. v( I = 2) > v( I = 1) ⇒
4 3

q 2 + p 2 2 − 13 qp R(1) 3 6 < p−q R(2) − 2 R(1)

It is easy to demonstrate that the left hand side of this inequality is always greater than the left hand sides of (i) and (ii). Thus, there exists parameter values such that I = 2 is preferred to I = 1 and vice versa. Now, consider the choice between I = 0 and I = 1. If (i) holds, v( I = 1) > v( I = 0) ⇒
3 p( 8 q − 1 p) R(1) 2 < 3 R(2) − 2 R(1) 4 ( p − q)

This condition is equivalent to (i). Thus, if (i) holds, partial integration is preferred to nonintegration. Finally, consider the choice between I = 0 and I = 2. If (ii) does not hold, these are equivalent. If (ii) holds v( I = 2 ) > v( I = 0 ) ⇒
1 3

(q

2

− p2 ) − 1 p 6 p−q

<

R(1) R(2) − 2 R(1)

If (ii) holds then this condition holds. Thus, non-integration is never optimal if (ii) holds. Proof of Proposition 3 If R(2) ≥ 2R(1), then (ii) always holds. Notice too that if 3q > 4p, then (i) holds. Thus, there are conditions under which some form of integration is optimal. When (i) holds, comparing I = 2 and I = 1, v( I = 2) > v( I = 1) ⇒
4 3

q 2 + p 2 2 − 13 qp R(1) 3 6 < p−q R(2) − 2 R(1)

The left hand side of this inequality is always less than the left hand side of (i). Therefore, if (i) holds, I = 2 is preferred to I = 1 always. Comparing I = 1 to I = 0,
3 p( 8 q − 1 p) R(1) 2 v( I = 1) > v( I = 0) ⇒ 3 < R(2) − 2 R(1) 4 ( p − q)

19

Once again this is equivalent to (i). Thus, if (i) holds, I = 1 is preferred to I = 0. However, if (i) does not hold then, I = 2 is preferred to I = 0. Thus, I = 2 is always optimal. If R(2) < 2R(1), then (ii) implies (i). Notice that under this condition, (i) and (ii) always hold. Comparing I = 2 and I = 1, v( I = 2) > v( I = 1) ⇒
4 3

q 2 + p 2 2 − 13 qp R(1) 3 6 > p−q R(2) − 2 R(1)

As the left hand side of this inequality is always less than the left hand side of (i) and (ii), there exist parameter values under which I = 2 is preferred to I = 1 and vice versa. Comparing I = 1 to I = 0, v( I = 1) > v( I = 0) ⇒
3 p( 8 q − 1 p) R(1) 2 > 3 R(2) − 2 R(1) 4 ( p − q)

Once again, this condition is equivalent to (i). Thus, I = 1 is always preferred to I = 0.

20

References
Aghion, P. and J. Tirole (1994), “The Management of Innovation,” Quarterly Journal of Economics, 109 (4), pp.1185-1210. Anton, J.J. and D.A. Yao (1994), “Expropriation and Inventions: Appropriable Rents in the Absence of Property Rights,” American Economic Review, 84 (1), pp.190209. Arrow, K.J. (1962), “Economic Welfare and the Allocation of Resources for Invention,” in The Rate and Direction of Inventive Activity, Princeton University Press: Princeton, pp.609-625. Arrow, K.J. (1982), “Innovation in Large and Small Firms,” in Ronen (ed.) Entrepreneurship. Bresnahan, T. and M. Trajtenberg (1995), “General Purpose Technologies -- Engines of Growth?” Journal of Econometrics, 65 (1), pp.83-108. Chesbrough, H.W. and D.J. Teece (1996), “When is Virtual Virtuous?” Harvard Business Review, Jan-Feb., pp.65-73. Cohen, W.M. and D.A. Levinthal (1989), “Innovation and Learning: The Two Faces of R&D,” Economic Journal, 99 (397), pp.569-596. Dasgupta, P. and J.E. Stiglitz (1980), “Market Structure and Innovative Activity,” Economic Journal, de Fontenay, C. (1997), “A Model of Vertical Integration and Bargaining with Many Suppliers,” mimeo., Stanford. Gans, J.S. and S. Stern (1997), “Incumbency and R&D Incentives: Licensing the Gale of Creative Destruction,” mimeo., MIT and Melbourne. Grossman, S. and O. Hart (1986), “The Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration,” Journal of Political Economy, 94, pp.691-719. Lamberton, D.M. (1980), “Economic Analysis of the UNISIST Experience,” in First International Conference on Technology for Development, Canberra. McAfee, R.P. and M. Schwartz (1993), “Opportunism in Multilateral Vertical Contracting: Nondiscrimination, Exclusivity, and Uniformity,” American Economic Review, 84(1), pp.210-230. Nelson, R.R. (1958), “The Simple Economics of Basic Scientific Research,” American Economic Review Nelson, R.R. and S. Winter (1982), An Evolutionary Theory of Technological Change, Romer, P. (1990), “Endogenous Technological Change,” Journal of Political Economy, Stern, S. (1996), “Incentives and Knowledge in Organizational and Technological Change: The Case of Drug Discovery in the 1980s,” mimeo., MIT. Stole, L.. and J. Zwiebel (1996), “Organizational Design and Technology Choice under Intrafirm Bargaining,” American Economic Review, 86 (1), pp.195-222.