1. Incentives and the Innovative Environment Suzanne Scotchmer ESNIE, Cargèse, May 2007 references: Chs 2, 8 Innovation and Incentives (MIT Press, 2005)

2. Innovation: the problem of “procuring knowledge” Framing Questions: Is “procuring knowledge” different than “procuring infrastructure?” Intellectual property is a procurement mechanism. Why do we use it in one realm, but not the other? (Is that true?) Is technological progress (growth) simply a matter of spending resources? How does ‘imagination’ fit into incentive theory?

3. R&D: How much? By whom? Percentages of GDP are similar, but sources of funds are different: E.U.: about 44% of R&D spending is public U.S. 26% of R&D spending is public (ratio of public/private was 2:1 in 1950; now 1:3) Brazil, Chile, Costa Rica and Mexico: substantially over half

4. The Plan of this Lecture Frame the problem: What is the incentive system trying to accomplish? Compare some important mechanisms: Intellectual Property Prizes Contests Grants (contracts)

5. The inescapable question: What is primitive? How do opportunities for innovation arise? 1.In most economic models of R&D and growth, the primitive is a production function for knowledge: Models of R&D races Most models of procurement Endogenous growth theory 2.The model in my own work: There is no such thing as a production function for knowledge. Instead: (1) People have ideas for investment (exogenously). (2) With investment, ideas may become innovations This is where incentives enter.

6. Ideas …. may be scarce The idea defines the need as well as the solution. ….. or substitutes Examples: longitude, HIV vaccine, X prize v For example, a product idea is (v,c) v = consumers surplus at p=mc  v c = cost of developing the idea into an innovation

7. Substitute ideas: the problem of aggregating information An idea is “best” if (v 1 /r-c 1 ) > (v 2 /r-c 2 ) (Not necessarily the lower-cost idea.) c 2 v 2 /r c 1 v 1 /r The incentive problem is about aggregating information, not about choosing the right number of firms in a race, or the tradeoff between deadweight loss and innovation.

8. Intellectual Property In its favor: A weak efficiency test (sort of). In its favor (maybe): Users pay Not in its favor: Deadweight loss Payment is ex post - who funds the investment? Information is not aggregated (the best ideas or firms are not chosen)

9. Prizes Prizes are complicated objects: Two types: blue-sky versus targeted Two styles: first-past-post versus best-in-class In their favor: Can give the same incentive as a patent, but without deadweight loss (set the prize equal to the patent value). It is always possible to make the value verifiable. Cremer and McLean 1988, Markets signals. Not in their favor (maybe): Users do not pay the costs.

10. Contests (best-in-class prizes) Examples (1) Simple Commitment to Pay (Nobel Prizes) (2) (Vickrey Auction) (3) Prototype Contest In their favor: Nothing needs verification Not in their favor: (1) duplication. (2) Need to make contingent contracts in advance

11. The problem of choosing the best idea Neither patents nor prizes aggregate information. What happens if we auction the right to develop? c 1 v 1 /r c 2 v 2 /r Vickrey auction works, but only if value can be observed. Report (v i /r – c i ), Pay v 1 /r – (v 2 /r – c 2 ) Prototype contest: Leads to duplicated cost and possibly not the best

12. Two Modern Grant Systems Peer-reviewed grants Not monitored; future grants depend on past success. Grant is given ex ante, not withheld ex post on the basis of failure. Government Subsidies. Government funds research, then gives IP, often to firms in return for matching funds. Prevalent in biomedicine. What is the role of IP in the subsidy system?

13. Grants NSF: 99% research budget given as grants (NSF also supports education) NIH: about 80% given out as grants Would you grant-supported R&D to be different than in-house R&D? Why? What is the point of grant support? How does NSF/NIH know whom to support? What constraints on abuse are there?

14. A simple Model of the Grant Process Suppose an “idea” worth funding costs c. Suppose it is impossible to punish a grantee for cheating (not delivering), e.g., by getting the money back. The only punishment is to kick a researcher out of the grant system. Suppose that each researcher has an idiosyncratic “idea rate,”  ideas per year Who does the NSF want to support? Should it depend on the “idea rate”  ? What is the objective? Let  be the size of a grant for a given idea. How large should  be? E.g.,  =c? Does the NSF/NIH have to “waste money” in order to keep researchers from cheating?

15. Keeping Grantees Honest: the grant size  Value of ideas in period t: The value of staying in the system (future grants): Invest if the value of cheating, c, is less than value of future grants (selects the high-fertility researchers)

16. Questions about grants If researchers have “fertile minds” (high  ), what does that do to the premium  -c? How much surplus does a high-cost researcher make? What if you are a low-fertility (low-  )? Will the government have to “over pay?” Homework: If the government could give different  for researchers with different , would high fertility researchers get larger or smaller grants per idea? Would they get more or less money per unit time?

17. vT+svT+s vTvT s s+m v m v Public Subsidies, Private Matching, IP Space of ideas c Matching m entitles firm to IP on subsidized innovations If the cost is less than m, the surplus m-c>0 goes to general fund. (Subsidy is irrelevant.) If the cost is more than (m+s), the firm pays the difference c-(m+s), and receives IP

18. Suppose agents i=1,2 have irreducibly different beliefs q 1 =.4, q 2 =.6, endowments = 10 each. w 1, w 2 = WTP in case of success Cost/benefit test (maybe): w 1 q 1 +w 2 q 2 >c What if w 1 =w 2 =0, c=1? Invention has intrinsic value. But investment can still make a Pareto improvement. Give19 units to agent 2 in case of success and to agent 1 in case of failure. Expected utility of each is greater than the value of the endowment: (0.6)(19)>10. Efficient investments: A paradox (Could this explain the “dot bomb”?) from Innovation and Incentives, Scotchmer 2005.

19. Example: Inefficient, self-reinforcing patent race Suppose costs are known. Suppose it is unprofitable for even a single firm to invest if signals are (L,L). Suppose each firm will invest, even if its own signals is L, if the probability is at least half that the other firm is H. Equilibrium strategies, leading to multiple outcomes: Invest if and only if the other firm invests. If the signals are (L,L), there are two equilibria, with and without investment. At signals (L,L), the outcome with investment is not profit-maximizing.

20. Aggregating Correlated Information Kremer (1998) aggregates correlated info ex post. Information often needs to be aggregated ex ante. Interpret v 1,v 2 as signals of a common value v Suppose v 1,v 2  {L,H}, c 1,c 2  {,h} The “cost” of innovation is min{c 1,c 2  The “value” is E(v | v 1,v 2 ). Information aggregation: In general it might be optimal to ask firm 1 to invest on the basis of information possessed by firm 2! Firm’s investment strategy depends on 1)firm’s own signal 2)firm’s observation of the other firm’s investment.