Evidence from Pharmaceuticals Mark Pauly Wharton and NBER Kyle Myers Wharton and NBER Are RampD Productivity Woes Warranted RampD Productivity typically refers to Average cost per new thing of assumed similar marginal socialprivate value ID: 1047060
Download Presentation The PPT/PDF document "Endogenous Productivity of Demand-Induce..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. Endogenous Productivity of Demand-Induced R&D: Evidence from Pharmaceuticals Mark Pauly, Wharton and NBERKyle Myers, Wharton and NBER
2. Are R&D Productivity Woes Warranted?R&D Productivity typically refers to:Average cost per new thing of assumed similar marginal (social/private) valueIn pharma.Avg. $ per NME: $230M(1980s), $540M(1990s) $1,300M(200s); $2,012M, DiMasi et al.Economy-widePrevailing thought: stable growth due to inputs amidst productivityJones 1995; Gordon 2012; Bloom, Jones, van Reenan, Webb 2017“We find that research productivity for the aggregate U.S. economy has declined by a factor of 41 since the 1930s, an average decrease of more than 5% per year” Bloom et al 2017
3. Who cares?Are private gains from new (and/or improved) products worth incurring extra costs? They must be if firms and buyers were rational and competitive.Should we be concerned? Depends…Are extra costs due to addressable frictions?Regulation (too much or too little)Intra-firm (e.g. short-sidedness, cognitive biases)Or are R&D inputs scarce and should we think like Ricardo (and Schmookler)?“Demand for better computer chips is growing so fast that it is worth suffering the declines in research productivity there in order to achieve the gains associated with Moore’s Law” – Bloom et al 2017
4. This paperRQsHow did production function in pharma change and is a decline in measured productivity explainable? Make policy-relevant distinctions about productivity parametersDid the NIH role in basic research have anything to do with what happened?Is the data consistent with Ricardo’s theory?Is demand growth spurring investments into scarce (i.e. less productive) “ideasCircumstantial evidence—is the path of rents to infra-marginal drug investments consistent with the observed path of productivity changes?
5. This paperGeneral approachExamine time-varying industry “production function” for new drugs:Instrument industry spending with therapeutic class-time () demographic and income growth related demand shocks (Acemoglu & Linn 2004)Identify productivity shifts (i.e.
6. Main findings from 1985-2013 200% decline in , no significant change in Nowadays larger markets still getting more new drugs, but everyone is getting lessNIH spending has had a shrinking impact on the flow of productive new ideasPattern of pharma productivity changes is in line with Ricardian story
7. Going back to Ricardo RecentFirms anticipate future market sizes and invest in R&D accordingly (Schmookler)Acemoglu & Linn (2004): (Demand->New Drug) causation in pharma.RicardoWhen markets growing while some inputs are constrained (i.e. land, new drug ideas), firms are incentivized to less productive inputs to increase profits/rentsWill likely continue to be important in:Healthcare – expenditure share increasing in wealth (i.e. Hall & Jones 2007)IT – returns to automation appear to be increasing (positive externalities)
8. Our theory ideasImplied modelExogenous shocks (NIH support; demographic-driven demand), determine the distribution of new drug ideas and their expected profitabilityDemand growth increases the ROI for all drug ideasIdeas with costs too high for >$0 returns in period t become profitable in t+1 if demand grows (due to new revenues)Importantly, unless demand shocks are accompanied by equal growth in the supply of ideas, the newly profitable ideas have lower expected or average (and marginal) productivity (higher costs/NME) because less productive investments have now become profitable.To the extent that promising ideas are scarce, increasing the output of new drugs in response to demand has to lower the productivity of R&D Unless there is an equivalent offset in the number or quality of ideas
9. Intro pt 2, pharma specificWhat are “ideas” and why might they be scarce? Jones (2005): ideas as production instructions; scarce ideas = lower productivityChanges in the distribution of ideas will manifest as changes in parametersRemain agnostic as to how scarce ideas might arise w.r.t. (classic TFP) and/or (classic output elasticity)Practically speaking, what drives differences in the ideas? Cross-section: costs of trials relative to exclusivity time (Budish, Roin, Williams 2015); relative NIH spending on disease-specific basic researchOver time: “low hanging fruit” gone” & not replenished, specialization (Jones 2009)
10. Data, R&D OutputsDrug classification schemeNeed mapping of investments to outcomesUse Anatomical Therapeutic Chemical (ATC) System 16 major categories; treated independentlyTradeoff: control for unobservables (with F.E.) versus allow for spilloversThis paper: priority is unobservables, ignore spillovers: simple specification w/ few interactions.Outcome: New Molecular Entities, 1985-2013FDA approvals receiving NME status
11. Data, R&D InputsPhRMA“Industry-wide” therapeutic-class-specific R&D spending1970-20008 class categories – mapped to 16 major ATC categoriesLag construction is ad hocNo disease specific post-2000 private R&D data; it is the most interesting period! Will work around…Look at NIH disease class supportMarket DemandFollow Acemoglu & Linn (2004)CPS – age-time income (I)MEPS – age-class spending avg. (S)Key assumption: panel income variation uncorrelated with firm’s investment decisions
12. EmpiricsSimplest Reduced Form: Conditional Poisson for class j year y and time-period t (4-year bins)Need within t variation to identify variation in productivity over timeFirst stage:Log-linear with class F.E.
13. EmpiricsI.V. with Conditional PoissonBlundell, Griffith, Windmeijer (2002)Can only estimate pre-2000Work around lack of post-2000 dataAssume industry response to demand shocks is stableImplication: identified productivity shocks were unexpected (otherwise response to demand would have changed)Predict private R&D post-2000 using observed demand shocksEstimate as 2-stage predictor substitution (2SPS), given bias potentialTest for signs of bias using reduced formNote: focus is on panel variation in parameters, not absolute accuracy
14. Has Market Responsiveness Changed over Time?Time-varying estimates From 1980-2013, a 1% increase in steady-state demand caused a 3% increase in same-year NME approvalsAcemoglu Linn (2004) ~3.5
15.
16. How did Demand Influence Investments Pre-2000?Stable and time-varying estimatesFrom 1980-2000, a 1% increase in steady-state demand caused a ~1.2% increase in private R&D seven years priorRobust to alternative lags
17.
18. How Productive was Demand-Induced R&D, Pre-2000?Output elasticity ~2-2.5Insignificant decline 1980-2000GMM vs 2SPSUpward bias in 2SPS ~30%Cross-section bias stable over timeInterpret 2SPS estimates accordingly
19.
20. How much of R&D Growth was Demand-Driven?Since 1980 roughly 50% of growth in PhRMA R&D can be attributed to demand growthLikely underestimateDon’t allow for spillovers (evidence suggests are positive)Demand measure assumes income elasticity of health spend is 1 (Hall & Jones 2007 argue > 1)What about other 50%?Can’t address in this paperNote that “demand-induced R&D” is likely most policy-relevant
21.
22. How Productive was Demand-Induced R&D, overall?All of productivity decline is within “TFP” ()Agrees with reduced-form evidence that NME-Demand elasticity is unchanged ( stable)
23.
24. What happened with NIH support and productivity?
25. Robustness & Supply-Side ConcernsRobustnessVary regression weights, R&D lags and drug-class categorization schemesOutput elasticity always stable, TFP declines range 100-200%Are the demand shocks uncorrelated with supply-side?Namely, NIH investments Test for correlation between private R&D and NIH spending: little (despite what you may have heard).Using wide range of lag structures, no significant correlations ariseEmpirically: supports exclusion restrictionPolicy: maybe lack of correlation is one cause of idea scarcity…
26. Back to Ricardo, and making some guesses: how much of growth in infra marginal rents is due to marginal productivity declines (which were driven by demand growth)?Demand-driven productivity declines will lead to proportional increases in the rents accrued to firms: corr(Demand , TFP) < 0corr(Demand , Rents) > 0Rent growth is larger when productivity declines and/or demand growth are largerWithout any nefarious intentions…Estimates from analysesData-driven ProxyUse revenue per market size per drug on patent (isolate price effects)
27. Suggestive Result: how much of growth in rents is due to productivity declines (which were driven by demand growth)?Virtually all of pharma’s “rents” could plausibly be explained by Ricardo’s theoryIn other words, all of growth in revenues due only to pricing could plausibly be due to productivity declines
28.
29. Summary and ContributionsNo point in handwringing over productivity decline or the “valley of death”Connection between productivity declines and higher prices/rents is unavoidable and is the fault of consumersWe may not need NCATSBut some attention should be paid to the flow of ideas from NIH (type and quantity)