Endogenous productivity of demand‐induced R&D: evidence from pharmaceuticals

• DOI: http://doi.org/10.1111/1756-2171.12289
• Author: Mark Pauly,Kyle Myers
• Date: 01 September 2019
• Published date: 01 September 2019

RAND Journal of Economics

Vol.50, No. 3, Fall 2019

pp. 591–614

Endogenous productivity of demand-induced

R&D: evidence from pharmaceuticals

Kyle Myers∗

and

Mark Pauly∗∗

We examine trends in the productivity of the pharmaceutical sector over the past three decades.

Motivated by Ricardo’s insight that productivity and rentsare endogenous to demand when inputs

are scarce, we examine the industry’s aggregate Research and Development (R&D) production

function. Using exogenous demand shocks to instrument investments,we find that demand growth

can explain a large portion of R&D growth. Returns to scale have been stable, whereas total

factor productivityhas declined significantly. Predicted rentsbased on our estimates and Ricardo’s

theory closely match the trends we observe.

1. Introduction

Considerable debate surrounds the growing research and development costs for new drugs.

These discussions often revolve around estimates of the average costs to bring a new drug to

market, which are estimated to have increased nearly six-fold in real terms from the 1980s to the

2000s (DiMasi et al., 1991; DiMasi, Grabowski, and Hansen, 2016).1Most discussions take this

trend as prima facie evidence of a growing friction in the R&D process, then conjecture as to

why it has occurred and how it may be corrected (e.g., Ruffolo, 2006; Pammolli, Magazzini, and

Riccaboni, 2011; Scannell et al., 2012).

Some blame regulatory burdens from purportedly stricter Food and Drug Administration

(FDA) requirements. Some blame an overrelianceon large-scale compound screening methods as

opposed to “rational” investigations based on theory and hypothesis testing. Others blame health

insurers for somehow pressuring drug firms to pursue products with lower success probabilities

∗Harvard University; kmyers@hbs.edu.

∗∗University of Pennsylvania; pauly@wharton.upenn.edu.

The Pharmaceutical Research and Manufacturers of America provided data that enabled our analyses, but no other

support, financial or otherwise. We are also thankful for the thoughtful comments and suggestions from Joe DiMasi,

Pierre Dubois, Tom Philipson, Neeraj Sood,and other attendees at the Bates White Life Science Symposium (2016) and

the Becker Friedman Health Sector and the Economy conference (2017). All errors are our own.

1See, for example, Hewitt, Campbell, and Cacciotti (2011). These R&D figures are frequently cited during debates

on pharmaceutical prices, although rarely following standard economic logic—ex post optimal prices are independent of

ex ante sunk costs.

C2019, The RAND Corporation. 591

592 / THE RAND JOURNAL OF ECONOMICS

but higher health improvements. Last, some argue that the stock of easy to develop ideas—the

“low-hanging fruit”—has shrunk and not been replaced. In response to these discussions, industry

leaders have called for the development of new R&D models, and for the government to bridge

the so-called “valley of death” in the drug discovery process (Butler, 2008; Collins, 2011).

More broadly, the declining pace of innovation and growing costs of R&D per output have

been documented across the US economy (Jones, 1995; Gordon, 2012; Bloom et al., 2018). Still,

it remains unclear to what extent these trends are driven by real economic frictions that might

necessitate policy intervention, or are instead the expected outcomes of rational firms making

investments in more costly, but increasingly demanded ideas.2

In this article, we clarify the nature of the productivity decline in pharmaceutical R&D,

and offer evidence that this decline is consistent with theoretical predictions about the way

productivity evolves when demand grows faster than the supply of ideas for new products. In

short, our findings are most in line with the “low-hanging fruit” hypothesis of scarcity.

We connect the classic idea of Schmookler (1966), expounded by Acemoglu and Linn

(2004)—the rate of innovation is directly related to demand growth—with Ricardo’s (1817)

point—demand and productivity will be inversely related wheninputs (here, profitable new drug

ideas) are rare. This connection guides our investigation of a simple aggregate R&D production

function,

N=αRβ,(1)

where the number of new products is a function of R&D investmentsper productivity parameters

α(TFP) and β(output elasticity). In our main analyses, we utilize the fact that firms’ optimal

investment leveldepends on the future size of the market, and identify the productivity parameters

using Acemoglu and Linn’s (2004, henceforth, AL04) exogenous demographics-driven measure

of demand to instrument.

The pharmaceutical sector provides both an inherentlyimportant and empirically ideal setting

to study R&D dynamics because we can (i) identify exogenous demand shocks, (ii) connect these

shocks to R&D investments and new products, and (iii) separately identify changes to TFP and

output elasticities. To do this, we utilize therapeutic-class-specific data on US consumer drug

expenditures, private US-based R&D investments, and approvals of New Molecular Entities

(NMEs, our proxy for new, highly valuable products).

In the sense that the exogenous demand measure is an instrumental variable, the reduced-

form evidence is clear; the elasticity of new drugs with respect to market size, the focal parameter

of AL04, is very stable over time. This is good evidence that any productivity decline was likely

not driven byallocative inefficiencies at firms. If managers had somehow gotten worse at directing

investments, we would have expected the relative rate of NMEs approved in larger versus smaller

markets to decline.

However, due to changes in industry reporting, we lack disaggregated private investments

post-2000, and cannot instrument investments throughout our sample (1985–2013). Toovercome

this limitation, we predict private R&D when unobserved under the assumption that firms are

equally responsive to demand shocks over time.3

Figure 1 shows the nature of the predictive R&D model, described in detail below; it plots

the sum of the predicted investments alongside actual investmentsto illustrate the role of expected

market size in stimulating private R&D. Notably,demand alone can explain anywhere from about

a third to virtually the entirety of the growth in total R&D since 1980.

2For instance, when discussing semiconductors and growingcosts of sustaining Moore’s Law,Bloom et al. (2018)

note that it may be the case that “Demand for better computer chips is growingso fast that it is worth suffering the declines

in idea Total FactorProductivity (TFP) there in order to achieve the gains associated with Moore’s Law.”

3This has some reasonable implications, for example, that managers didn’t get better or worse at forecasting

demand, but also implies a more debatable feature, that managers did not forecast any productivity shocks. We discuss

the implication of this assumption in Section 3.

C

The RAND Corporation 2019.