Estimating dynamic R&D choice: an analysis of costs and long‐run benefits

• DOI: http://doi.org/10.1111/1756-2171.12181
• Author: Mark J. Roberts,Bettina Peters,Helmut Fryges,Van Anh Vuong
• Published date: 01 May 2017
• Date: 01 May 2017

RAND Journal of Economics

Vol.48, No. 2, Summer 2017

pp. 409–437

Estimating dynamic R&D choice: an analysis

of costs and long-run benefits

Bettina Peters∗

Mark J. Roberts∗∗

Van Anh Vuong∗∗∗

and

Helmut Fryges∗∗∗∗

This article estimates a dynamic structural model of discrete Research and Development (R&D)

investment and quantifies its cost and long-run benefit for German manufacturing firms. The

model incorporates linkages between R&D choice, product and process innovations, and future

productivity and profits. The long-run payoff to R&D is the proportional difference in expected

firm value generated by the investment. It increases firm value by 6.7% for the median firm in

high-tech industries but only 2.8% in low-tech industries. Simulations show that reductions in

maintenance costs of innovation significantly raise investment rates and productivity, whereas

reductions in startup costs have little effect.

1. Introduction

Firm investment in R&D is a key mechanism generating improvements in firm perfor mance

over time. Estimating the ex post return to the firm’s R&D investment has been a major focus of

empirical studies for decades, with most of the empirical literature built around the knowledge

∗Centre for European Economic Research (ZEW); b.peters@zew.de.

∗∗Pennsylvania State University and NBER; mroberts@psu.edu.

∗∗∗University of Cologne; vananh.vuong@ewi.uni-koeln.de.

∗∗∗∗University of Tasmania; helmut.fryges@utas.edu.au.

We are grateful to Uli Doraszelski, Jordi Jaumandreu, Ken Judd, Florin Maican, Jacques Mairesse, Matilda Orth,

Joris Pinkse, Marc Rysman, Spiro Stefanou, Jim Tybout, Daniel Xu, Hongsong Zhang, and three referees for helpful

comments and discussions. Wethank particpants at the ZEW Conference on Innovation and Patenting (Mannheim, 2013),

IIOC (Boston, 2013), EEA (Gothenburg, 2013), German Economic Association (D¨

usseldorf, 2013), EARIE (Evora,

2013), CDM Workshop (London, 2013), Conference on Innovation and Entrepreneurship (Copenhagen, 2014), MaCCI

Annual Conference (Mannheim, 2014), Berlin IO Day (2016), as well as seminars at universities of Cologne (2013),

Brussels (2013), Valencia (2013), Turin (2013), Barcelona (2013), Cornell (2013), Leuven (2013), Tinbergen Institutue

Amersterdam (2013), Heidelberg (2014), W¨

urzburg (2014), Gothenberg (2014), Duke (2014), Oregon (2014), ETH

(2016), Zurich (2016), Copenhagen Business School (2016). We thank the Centre for European Economic Research

(ZEW) for providing data access and research support.

C2017, The RAND Corporation. 409

410 / THE RAND JOURNAL OF ECONOMICS

production function developed by Griliches (1979). In this framework, firm investment in R&D

creates a stock of knowledge that enters into the firm’s production function as an additional input

along with physical capital, labor, and materials. The marginal product of this knowledge input

provides a measure of the return to the firm’s investment in R&D and has been the primary focus

of the empirical innovation literature.

The goal of this article is to estimate the expected payoff to R&D investment at the firm

level. Unlike most of the empirical literature that relies on the knowledge production function,

we focus on the firm’s R&D investment decision. The discrete decision to invest in R&D contains

information on the costs of innovation and the expected long-run payoffto the firm from engaging

in R&D investment. We develop a dynamic structural model of the firm’s choice to invest in

R&D, estimate the model using microdata on German manufacturing firms, and summarize the

implicit expected long-run payoff to R&D that rationalizes the firm’s observed R&D investment

decision.

Our model of the firm’s dynamic R&D choice captures five important features of the R&D

investment process. The first is the impact of R&D on the probability that the firm realizes

a product or process innovation. The second is the effect of these realized innovations on the

firm’s revenue productivity and short-run profitability. Third, these effects can be long lived,

affecting the incentives of the firm to invest in the future and impacting the long-run value of

the firm. Fourth, there is uncertainty surrounding the effect of R&D on innovation and the effect

of innovation on productivity. Fifth, the cost of generating innovations is likely to differ between

firms based on their size and whether they are spending to maintain ongoing R&D activities

or establishing new R&D programs. Incorporating these features into the model, the structural

parameters characterize the linkages between R&D, innovation, and productivity as well as the

costs of producing innovations.

Weuse the model to estimate the long-r un payoff to R&D for a sample of German manufac-

turing firms across a range of high-tech and low-tech industries. The data source is the Mannheim

Innovation Panel (MIP) collected by the Centre for European Economic Research (ZEW). This

is the German contribution to the Community Innovation Survey (CIS) that is collected for most

countries within the Organisation for Economic Co-operation and Development (OECD). The

key features of the MIP survey that we utilize are questions on product and process innovations

realized by the firm, R&D input measures, production expenditure, capital stocks, and firm sales.

The structural estimates can be briefly summarized. First, firms that invest in R&D have a

substantially higher probability of realizing product or process innovations; but R&D investment

is neither necessary nor sufficient for firm innovation. The group of high-tech manufactur-

ing industries has a higher probability of innovation, given R&D, than the group of low-tech

industries. Second, product as well as process innovations lead to increases in future firm pro-

ductivity; but product innovations are more important for the high-tech industries, whereas

process innovations are more important for the low-tech industries. Third, firm productivity is

highly persistent over time, which implies that innovations that raise productivity have long-

run effects on firm performance. Fourth, the cost of generating innovations is significantly

smaller for firms that are maintaining ongoing R&D investment rather than beginning to in-

vest in R&D. This means that firm R&D history is an important determinant of current R&D

behavior.

Using the structural parameters, we estimate the expected payoff to firm R&D as the pro-

portional difference in the expected future value of a firm when it invests in R&D versus when

it does not. This expected payoff varies with the productivity, capital stock, age, and industry of

the firm and can be constructed for all firms, not just firms that choose to invest. We find that the

expected payoff varies substantially across industries and across firms within each industr y. For

the five high-tech industries, a firm with the median productivity, capital stock, and age has an

expected payoff equal to 6.7% of firm value. In the seven low-tech industries, the corresponding

payoffis 2.8%. Our results show that the difference between the high-tech and low-tech industries

arises from differences in the magnitude of the effect of innovation on the firm’s productivity

C

The RAND Corporation 2017.