Optimal contracts for research agents

• Date: 01 March 2017
• Published date: 01 March 2017
• DOI: http://doi.org/10.1111/1756-2171.12169
• Author: Yaping Shan

RAND Journal of Economics

Vol.48, No. 1, Spring 2017

pp. 94–124

Optimal contracts for research agents

Yaping Shan∗

We study the agency problem between a firm and its research employees under several scenarios

characterized by differentResearch and Development (R&D) unit setups. In a multiagent dynamic

contracting setting, we describe the precise pattern of the optimal contract. We illustrate that the

optimal incentive regime is a function of how agents’ efforts interact with one another: relative

performance evaluation is used when their efforts are substitutes, whereas joint performance

evaluation is used when their efforts are complements. The optimal contract pattern provides a

theoretical justification for the compensation policies used by firms that rely on R&D.

1. Introduction

Over the last two decades, the industries of information and communication technologies

have emerged as the drivers of the US economy: between 1998 and 2012, they registered an

average annual growth rate of 9.3% when that of the Gross Domestic Product (GDP) was only

2.3%. A distinct feature of these so-called new-economy industries is the substantial investment

in R&D. In 2013, the total R&D spending in these industries was over $110 billion, which

accounted for about 10% of their revenues—the highest figures reported across all industries.

Clearly, the success of firms in these industries depends crucially on the performance of the

employees in their R&D units, and the compensation schemes for these researchers are the focal

point of decision making in these firms. This decision problem is similar in some respects to the

standard problem of providing incentives to workers; however, it also has some unique features.

As with its standard counterpart, a moral-hazard phenomenon is present in this specific agency

relationship. The outcome of research is uncertain; that is, the effort invested in research today

will not necessarily lead to a discovery tomorrow. However, the stochastic process governing the

outcomes is influenced by how much effort is put into research: higher levelsof effort increase the

chance of a discovery. Owing to task-complexity, the effort exerted by researchers is difficult to

monitor. Now, if the effortlevel is unobservable, then the imperfect monitoring of effort combined

∗The University of Adelaide; yaping.shan@adelaide.edu.au.

A special thanks to my advisor, Srihari Govindan, for his guidance and thoughtful support. I thank Yuzhe Zhang for

many insightful discussions. I sincerely thank the Editor, David Martimort, and two anonymous reviewers for their

constructive comments and suggestions. I would like to thank Martin Gervais, Ayca Kaya, Kyungmin Kim, Mandar

Oak, B. Ravikumar, Raymond Riezman, and seminar participants at the University of Iowa, University of Rochester,

the University of Adelaide, 2013 North American Summer Meeting of the Econometric Society in Los Angeles, 2013

Australasia Meeting of the Econometric Society in Sydney,2013 Asia Meeting of the Econometric Society in Singapore,

and the 68th European Meeting of the Econometric Society in Toulouse for their valuable advice and comments. Any

errors are my own.

94 C2017, The RAND Corporation.

SHAN /95

with the stochastic feature of innovation creates a moral-hazard problem. Further, because most

R&D projects last for long periods, the moral-hazard problem is dynamic in nature.

The agency problem that these R&D-intensive firms face with respect to their research

employees differs from the standard principal-agent problems in two aspects. First, unlike with

employees in traditional industries, it is difficult to measure research employees’ performance on

the basis of their day-to-day practice. However, most R&D projects progress through different

phases, with work in each phase depending on the outcomes of the previous phases. Thus, the

performance measure for research employees is usually linked to the completion of a sequence

of milestones. This multistage feature is particularly prominent in new-economy industries. For

example, since the first iPhone was launched in 2007, Apple has introduced ten new generations

of the iPhone into the market. Each generation is equipped with cutting-edge features, which

have, over the years, completely transformed the smartphone industry.

The second point of departure from standard agency problems is that R&D projects are nowa-

days typically undertaken by groups of researchers. Unlike when Edison invented the lightbulb

and Bell the telephone, R&D projects today are far too complex for technological breakthrough to

be realized by a single individual. Greater efficiencies can be achievedwhen multiple researchers

collaborate to overcomea key challenge in technological development. Hence, the most innovative

companies in the world, like Apple, Google, Facebook, Microsoft, and Amazon, have employed

innovation teams, which enable them to launch innovations faster. The widespread use of teams

in R&D projects suggests that a multilateral environment is the appropriate setting to approach

the agency problem between a firm and its in-house R&D unit.

In practice, these firms organize their research units in various ways. Most small start-up

firms focus on a single project owing to resource constraints, whereas tech giants, like Apple,

Google, and Facebook, usuallyadopt a parallel innovation strategy in which multiple teams work

on different research projects simultaneously. Such firms can be further categorized into two

groups. Facebook, Google, and many other firms encourage communication among research

teams. To enhance communication, they provide benefits and incentives, such as free food and

coffee and free on-site recreations, that motivate researchers to share their research experience

and exchange ideas. However, Apple organizes its research units such that multiple teams may

be assigned to the same area but work independently. Communication barriers are intentionally

created between teams so that researchers may not know who they are competing with or what

other teams are working on. Apple believes that secrecy drives internal competition and peer

pressure among employees enhances innovation efficiency. Given these different approaches to

organizing R&D units, how do firms arrive at the optimal compensation scheme under each of

these scenarios?

We approach this problemby constructing a theoretical agency model that captures both the

multistage nature of the innovative process and the multilateral feature of the incentive problem,

and design an optimal contract for each of the scenarios described above. Briefly, we construct

the model as follows. A principal hires two risk-averse agents to carry out a multistage R&D. At

any point in time, the agents can either choose to devote effort to work or shirk. Their actions

cannot be monitored by the principal, which creates a moral-hazard problem. The transition from

one stage to the next is modelled by a Poisson-type process, and the arrival rate of success is

jointly determined by the effort choice of both agents. Hence, the principal cannot consider each

agent separately. To overcome the moral-hazard problem, the principal offers each agent a long-

term contract that specifies a history-contingent payment scheme based on the information that

the principal can observe. In terms of public information, we consider two scenarios. In the first

scenario, the team-performance case, only the joint performance of the team can be observed. This

scenario reflects the situation in start-up firms where researchers work on a single project. In the

second scenario, the individual-performance case, the principal can identify the individual who

has completed the innovation. This scenario captures the situation found in largefir ms that pursue

parallel innovation, exploring severalnew ideas or approaches simultaneously. To model the two

approaches used by firms pursuing parallel innovation, we use different settings to mimic how

C

The RAND Corporation 2017.