NC-1034 meeting on The Future of Agricultural Research: Funding, Funding Mechanisms, and Public-Private Collaborations March 15, 2012 Identifying and Rewarding Success with Proportional Prizes Will Masters Friedman School of Nutrition, Tufts University http://nutrition.tufts.edu http://sites.tufts.edu/willmasters Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application • Diagnosis: Ag R&D is constrained by asymmetric information – Funders cannot observe impact directly; they see only impact claims – Innovators have access to more data, but little incentive to reveal it – This is Akerlof’s market for lemons • Remedy: An incentive to reveal hidden information – A type of quality certification, to elicit outcome data for third-party audit – A type of contest, to attract participants and reveal relative performance • Today: Design and performance of proportional prize contests – Typology and motivation for the new design – Performance in laboratory experiments • A “real effort” experiment, with endogenous entry (J. of Public Economics 2010) • A “chosen effort” experiment, with equilibrium benchmarks (submitted March 2012) – Specification of a proportional prize contest for agricultural R&D Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application • Why not just intellectual property rights (IPRs)? – Well suited for proprietary, excludable innovations, with value capture …but not for non-excludable, public services • Why not just grants & contracts? – Well suited for both private and public services, of predictable value …but not for services where the preferred vendor is unknown • Why not conventional contests? – Well suited for discrete breakthroughs, with one or few winners …but ag involves many sequential, location-specific, cumulative successes • The proposed new contest design would: – Specify how impact is to be measured, then audit and reward results – Offer artificial market-like incentive, proportional to measured success – Mimic stock markets, other real-life competition with market share Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application A typology of innovation incentives Investor: Private forprofit (to avoid need for value capture) Instrument: Direct grants & contracts Ex-post payments and prizes (to avoid need for project selection and supervision) Public or philanthropic Many private labs, or… Novartis, BP to UC Berkeley; Chocolate makers to STCP for cocoa in West Africa Many government labs, or… grants and contracts to public and private institutions, universities and other agencies Eli Lilly and others on Innocentive (since 2001); Procter & Gamble etc. on NineSigma (since 2000) X Prizes for space flight etc. (1996- ), AMC for new pneumococcal vaccine (launched June 2009) Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Philanthropic prizes have a long history (shown here: 1700-1930) Net present value of prizes paid French Academy of Sciences Montyon prizes for medical challenges (2006 US dollars, not to scale) $51,118,231 Deutsch Prize for flight between the Aero-Club de France and Eiffel Tower British Longitude prize for determining longitude at sea $12,600,000 The Daily Mail prize for flight $5,997,097 across the English Channel $3,364,544 French government prize for food preservation techniques $1,045,208 Hearst prize for crossing continental US in 30 days French government prize for large scale hydraulic turbine French government prize for producing alkali soda $421,370 $644,203 Milan Committee prize for flight across Alps $618,956 $582,689 The Daily Mail prize for transatlantic flight $515,770 $289,655 Chicago Times-Herald prize for motors for self-propelling road carriage $123,833 Scientific American prize for first plane in US to fly 1 km $56,502 Wolfskehl prize for proof of Fermat’s Last Theorem 1700 1750 1800 1850 Orteig prize for solo flight NY to Paris 1900 $31,690 1930 Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Philanthropic prizes have grown quickly (shown here: 1930-2009) Net present value of prizes paid (2006 US dollars, not to scale) Advance market Commitment for pneumococcal disease vaccine up to $1.5 billion Soviet Incentive Awards For Innovative Research Bigelow Space Prize for crew transport into orbit $ 50,000,000 Super Efficient Refrigerator Program for highly efficient CFC free refrigerator $165,755,396 $37,682,243 Virgin Earth Challenge for removal of greenhouse gases $ 25,000,000 European Information and Communication Technology Prize $ 10,917,192 Ansari X Prize for private manned space flight $ 10,717,703 Archon X Prize for sequencing the human genome $ 10,000,000 $7,000,000 Millennium Math Prizes for seven unsolved problems DARPA Grand Challenge for robotics in vehicles $6,660,406 $6,000,000 Lemelson-MIT Prize for invention of a patented product useful to society $4,300,000 Methuselah Mouse Prize for demonstration of slowing of ageing process on mouse NASA Centennial Challenges for Improvements in space exploration $2,000,000 $1,882,290 Schweighofer Prize for Europe’s forest industry competitiveness $1,600,000 International Computer Go Championship $1,210,084 Budweiser Challenge for first non-stop balloon flight around the globe $1,210,084 Rockefeller Foundation Prize for Rapid STD Diagnostic Test $1,210,084 Grainger Challenges for development of economical filtration devices for the removal of arsenic from well water in developing countries $588,092 Kremer Prize for Human Powered Flight Across the English Channel Kremer Prize for Human Powered Flight (Figure 8) $290,153 Feynman Prizes for nano tech robot technology $250,000 $59,240 1940 CATS Prize for inexpensive commercial launch of payload into space $654,545 $250,000 Electronic Frontier Foundation Cooperative Com$50,000-250,000 puting Challenge for new large prime numbers Beal’s $128,489 Fredkin Prize for Chess Computer Program Conjecture Prize $100,000 Loebner Prize for Computer that can pass the Turing Test $100,000 Polytechnische Gesellscaft Prize for Human Powered Flight 1930 Goldcorp Challenge for best gold prospecting methods or estimates 1950 1960 1970 1990 2000 Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application A typology of contest designs Target is pre-specified Success is ordinal (yes/no, or rank order) Success is cardinal (increments can be measured) Target is to be discovered Traditional prizes (e.g. X Prizes) Achievement awards (e.g. Nobel Prizes, etc.) AMC for medicines, COD for schooling (fixed price per unit) Proportional prizes (fixed sum divided in proportion to impact) Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Experiment #1 Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Subjects solved arithmetic problems as quickly and accurately as possible, choosing how they want to be paid. Table 1. Contest results under piece-rate (PR), winner-take-all (WTA) and proportional-prize (PP) payments, with endogenous entry Start with piece rate to see skill Then offer contests, either traditional WTA or proportional Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Offering proportional contests not only increased entry and total performance, but also reduces inequality Winner-Take-All Contests Lost Did not enter Won Proportional Prize Contests Distribution includes entrants and non-entrants Results shown are for 207 contests involving 69 subjects Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Experiment #2 • A “chosen effort” contest between two symmetric players, so can solve for equilibrium in: • winner-take-all contests won by the best performer, • winner-take-all lotteries where odds of success are proportional to performance, and • proportional-prize contests with rewards shared in proportion to performance. • Performance depends on both effort and random noise to reflect imperfect information: • outcome (𝑦𝑖 ) depends on both effort and noise: 𝑦𝑖 (𝑒𝑖 |𝜀𝑖)=𝑒𝑖𝜀𝑖 • noise (𝜀) is uniformly distributed on the interval [1−𝑎,1+𝑎], 𝑎∈[0,1] • success (𝑝𝑖 ) is relative to other contestants: 𝑝𝑖 (𝑒𝑖,𝑒j|𝜀𝑖,𝜀 j)=𝑦𝑖𝑟/(𝑦𝑖𝑟+𝑦 j 𝑟) • payoff (𝜋𝑖 ) depends on the value of prize (v) and cost of effort: 𝐸(𝜋𝑖)=𝑝𝑖𝑣−𝑐(𝑒𝑖) • The three forms of competition are special cases of the success function • Traditional WTA contest if r=∞ • “Tullock” WTA lottery if r=1 and pi is probability of winning a lump-sum prize • Proportional prize contest if r=1 and pi is share of the prize that is won • With uniform noise and quadratic costs [𝑐(𝑒)=𝑒2/𝑏], we can • solve for pure strategy equilibria, and compare to laboratory behavior Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Proportional contests elicit more realistic behavior, less optimism bias Table 1: Experimental Parameters and Equlibrium Predictions Treatment Value of the Prize, 𝑣 Noise Parameter, 𝑎 Equilibrium Effort, 𝑒 ∗ Expected Payoff, 𝐸 𝜋 ∗ DET-L 100 0.5 70.7 0.0 DET-H 100 1 50.0 25.0 PROB-L 100 0.5 34.6 38.0 PROB-H 100 1 31.1 40.3 PP-L 100 0.5 34.6 38.0 PP-H 100 1 31.1 40.3 Table 2: Observed Average Efforts and Payoffs (144 subjects, 2880 rounds) Treatment DET-L DET-H Equilibrium Average Median St. Dev. 70.7 62.4 65.0 20.9 50.0 51.2 50.0 17.4 Equilibrium Average Median St. Dev. 0.0 6.7 0 47.1 25.0 20.8 0 49.0 PROB-L PROB-H Effort 34.6 31.1 51.3 46.1 51.0 47.0 20.0 17.2 Payoff 38.0 40.3 19.7 25.8 0 0 49.7 49.5 PP-L PP-H 34.6 45.2 45.0 15.6 31.1 42.4 41.3 17.8 38.0 27.1 27.6 16.5 40.3 28.9 28.4 27.2 Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application How proportional prizes would work in African agriculture • Donors offer a given sum (e.g. $1 m./year), to be divided among all successful new technologies • Innovators assemble data on their technologies – controlled experiments for output/input change – adoption surveys for extent of use – input and output prices • Secretariat audits the data and computes awards • Donors disburse payments to the winning portfolio of techniques, in proportion to each one’s impact • Investors, innovators and adopters use prize information to scale up spread of winning techniques Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Implementing Proportional Prizes: Data requirements Data needed to compute each year’s economic gain from technology adoption Price D S S’ S” Variables and data sources J (output gain) P K (cost reduction) ΔQ Field data Yield change × adoption rate J Input change per unit I I Economic parameters Supply elasticity (=1 to omit) K Δ Q Demand elasticity (=0 to omit) (input change) Q Market data P,Q National ag . stats. Q’ Quantity Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Implementing Proportional Prizes: Data requirements Data needed to impute each year’s adoption rate Fraction of surveyed domain Other survey (if any) First survey Projection (max. 3 yrs.) Linear interpolations First release Application date Year Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Implementing Proportional Prizes: Data requirements Calculation of NPV over past and future years Discounted Value (US$) “Statute of limitations” (max. 5 yrs.?) First release Projection period (max. 3 yrs.?) Year NPV at application date, given fixed discount rate Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Hypothetical results of a West African contest Example results using case study data Example technology 1. Cotton in Senegal Measured Social Gains (NPV in US$) Measured Social Gains (Pct. of total) Reward Payment (US$) 14,109,528 39.2% 392,087 2. Cotton in Chad 6,676,421 18.6% 185,530 3. Rice in Sierra Leone 6,564,255 18.2% 182,413 4. Rice in Guinea Bissau 4,399,644 12.2% 122,261 5. “Zai” in Burkina Faso 2,695,489 7.5% 74,904 6. Cowpea storage in Benin 1,308,558 3.6% 36,363 231,810 0.6% 6,442 $35.99 m. 100% $1 m. 7. Fish processing in Senegal Total Note: With payment of $1 m. for measured gains of about $36 m., the implied royalty rate is approximately 1/36 = 2.78% of measured gains. Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Opportunity for a single-country trial in Ethiopia New technology adoption is stalled: Share of cropped area under new seeds for major cereal grains, 1996-2008 Source: Ethiopian Central Statistical Agency data, reprinted from D.J. Spielman, D. Kelemework and D. Alemu (forthcoming), “Seed, Fertilizer, and Agricultural Extension in Ethiopia.” Draft chapter for P. Dorosh, S. Rashid, and E.Z. Gabre-Madhin, eds., Food Policy in Ethiopia. Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application Opportunity for a single-country trial in Ethiopia Adoption is especially slow for seeds: Number and proportion of farm holders applying new inputs, by education Proportion of farms using new inputs: No. of farms Fert. Impr. Seed Pesticide Irrigation 12,916,120 44% 12% 24% 8% Illiterate 8,239,615 41% 10% 22% 8% Informally educated 1,016,284 48% 13% 23% 12% Some formal education 3,660,222 51% 16% 30% 8% All farm holders Of whom: Source: Author's calculations, from CSA (2010), “Agricultural Sample Survey 2009-2010 (2002 E.C), Meher Season.” Version 1.0, 21 July 2010. Addis Ababa: Central Statistical Authority of Ethiopia. Available online at http://www.csa.gov.et/index.php?&id=59. Identifying and Rewarding Success with Proportional Prizes Motivation| Experimental Results | Application In conclusion… • Diagnosis: Ag R&D is constrained by asymmetric information – Funders cannot observe impact directly; they see only impact claims – Innovators have access to more data, but no incentive to reveal it – This is Akerlof’s market for lemons • Remedy: An incentive to reveal hidden information – A type of quality certification, to elicit outcome data for third-party audit – A type of contest, to attract participants and reveal relative performance • Today: Design and performance of proportional prize contests – Typology and motivation for the new design – Performance in laboratory experiments • A “real effort” experiment, with endogenous entry (J. of Public Economics 2010) • A “chosen effort” experiment, with equilibrium benchmarks (submitted March 2012) – Specification of a proportional prize contest for agricultural R&D Well-designed prize contests offer very powerful incentives • By “well-designed prizes”, we mean: – An achievable target, an impartial judge, credible commitment to pay • Such prizes elicit a high degree of effort: – Typically, entrants collectively invest much more than the prize payout – Sometimes, individual entrants invest more than the prize • e.g. the Ansari X Prize for civilian space travel offered to pay $10 million • the winners, Paul Allen and Burt Rutan, invested about $25 million • Why do prizes attract so much investment? – contest provides a credible signal of success – so winners can sell their product more easily • the X Prize winners licensed designs to Richard Branson for $15 million • and eventually sold the company to Northrop Grumman for $??? million • total public + private investment in prize-winning technologies ~ $1 billion …but traditional prize contests have serious limitations! • Traditional prize contests are winner-take-all (or rank-order) – this is inevitable when only one (or a few) winners are needed, but... • Where multiple successes could coexist, imposing winner-take-all payoffs introduces inefficiencies – strong entrants discourage others • potentially promising candidates will not enter – pre-specified target misses other goals • more (or less) ambitious goals are not pursued – focusing on few winners misses other successes • characteristics of every successful entrant might be informative • New incentives can overcome these limitations with more market-like mechanisms, that have many winners New pull mechanisms allow for many winners • From health and education, two examples: – pilot Advance Market Commitment for pneumococcal disease vaccine • launched 12 June 2009, with up to $1.5 billion, initially $7 per dose – proposed “cash-on-delivery” (COD) payments for school completion • would offer $200 per additional student who completes end-of-school exams • What new incentive would work for agriculture? – what is the desired outcome? • unlike health, we have no silver bullets like vaccines • unlike schooling, we have no milestones like graduation • instead, we have on-going adoption of diverse innovations in local niches – what is the underlying market failure? • for AMC and COD, the main problem is making commitments • for agriculture, the main problem is learning what works, where – Innovations are location-specific; investors cannot observe success directly What new incentives could best reward new agricultural technologies? • New techniques from elsewhere did not work well in Africa – local adaptation has been needed to fit diverse niches – new technologies developed in Africa are now spreading • Asymmetric information limits scale-up of successes – local innovators can see only their own results – donors and investors try to overcome the information gap with project selection, monitoring & evaluation, partnerships, impact assessments… – but outcome data are rarely independently audited or publically shared • The value created by ag. technologies is highly measureable – gains shown in controlled experiments and farm surveys – data are location-specific, could be subject to on-side audits • So donors could pay for value creation, per dollar of impact – a fixed sum, divided among winners in proportion to measured gains – like a prize contest, but all successes win a proportional payment