Representation of Technological Change: Some Economic Approaches

1. Representation of Technological Change: Some Economic Approaches Jim Turnure Principal, Energy Group ICF Consulting ORNL Conference on Estimating the Benefits of Government-Sponsored Energy R&D Arlington, VA March 4, 2002

2. Outline I. Overview II. NREL Research Task: Process and Methodology • Purpose of Study • Scope and Limitations • Research Methodology III. Summary of Approaches to Technological Representation • Capital Investment and Growth Models • Option Value and Principal-Agent Models • Spillover Effects and Network Models • Sidebar: ‘Top-Down’ vs. ‘Bottom-Up’ Models • Discussion

3. I. Overview

4. Baseline Estimation is a Critical Step in R&D Benefit Assessment • R&D investment is intended to advance technological change, and to improve indicators of technological change • In order to assess the impact of R&D investment, it is necessary to understand and measure how technological change would develop in the absence of specific R&D investments • Baseline estimation cannot occur in a vacuum; it must be carried out in a robust analytic framework. In turn, such analytic frameworks need theoretical foundations and specific premises regarding how the processes of innovation and technological change operate in the economy • The purpose of this presentation is to outline major theoretical approaches that have developed in the field of economics, partly based on a literature survey conducted for the National Renewable Energy Laboratory (NREL).

5. II. NREL Research Task: Process and Methodology

6. Purpose of Study • NREL required a thorough survey of the major economics literature in order to develop the knowledge base on the role of technological change in economic growth. • The intent of the research task was to conduct a literature survey, select key papers from the survey results, and summarize them in a readily usable format. This allows for rapid access to the underlying literature base. • The study is one of a number of starting points for R&D benefit estimation, which in being reevaluated by many organizations partly in response to the requirements of the Governmental Performance and Results Act (GPRA).

7. Scope and Limitations • This literature review focused primarily on major economics journals: • American Economic Review • Journal of Economic Perspectives • Journal of Economic Literature • Economic Journal • Journal of Policy Analysis • Journal of Economic and Environmental Management • Resource and Energy Economics • Energy Journal • Another example of meta-data that can be useful in this context is “The Modern History of Energy Conservation: An Overview for information Professionals” by Donald R. Wulfinghoff. This document explains major information sources and their relative strengths and weaknesses, encompassing dedicated energy efficiency journals, trade and advocacy sources.

8. Research Methodology • ICF Consulting’s professional library and research staff carried out the initial phase of the literature review. • The Econolit on-line data service was searched using keywords such as ‘Coase Theorem’, ‘Technology’, ‘Research and Development’ and ‘Market Barriers’. • The initial search resulted in a citation list of over 300 candidate articles. This initial bibliography was narrowed to approximately 90 relevant cites, and these were then collected from university libraries and other sources. • Relevant articles were sorted into major categories and reviewed; key articles in each category were summarized for the draft report.

9. III. Summary of Approaches to Technological Representation

10. ‘Schools of Economics’ • Economic theory has advanced along a number of fronts, forming a series of theoretical approaches that can be termed competing schools of thought. • Understanding these competing schools of thought is a useful starting point for developing analytic frameworks and how technological change should be represented in baselines for R&D benefits measurement. • The themes outlined in this presentation are meant to be suggestive rather than definitive. Other categorizations can be equally useful; the intent here is to establish a starting point for discussion and a guide to more information.

11. Capital Investment and Growth Models (I) • By the 1950s growth theory had assumed a dominant role in macroeconomics. Key contributions were made by Arrow, Samuelson, and Solow. • During this period robust quantitative models were developed including general equilibrium models and optimal growth models. More detailed work on economic structure and the availability of measured data through national governments made more sophisticated arguments about the determinants of economic growth possible. • The dynamic of technology and productivity improvement is critical in economic growth. Other determinants of growth such as population and natural resources depend on physical stocks that can increase, but technological advances can not only increase in a simple incremental fashion, but also multiply in ways that allow for much faster growth in the overall economy.

12. Capital Investment and Growth Models (II) • Solow (1956 1957) examined how more flexible production functions, including a technical change term, worked in theory, while noting that he was using technical change “as a shorthand expression for any kind of shift in the production function [emphasis in original].” Solow demonstrated the importance of improvements in productivity, including improvements resulting from technological change. • Arrow (1962) posited endogenous technical improvement as a function of experience, making the observation that it is not simply the passage of time that results in technical change. ‘Learning by doing’ had been observed in industrial settings and Arrow formalizes this relationship. In this model technical change is a function of producing new capital equipment. • Mansfield (1961) introduced an early model for the firm’s technology adoption decision. In this model the rate of adoption depends on 1) the proportion of related firms already using an innovation (because of the reduced risk from others’ experience); 2) the estimated profitability of using the innovation; 3) the cost of adoption and 4) inter-industry variations in general propensities for adopting innovations.

13. Capital Investment and Growth Models (III) • Using an analytic framework based on the concepts advanced in this tradition will generally lead to a macroeconomic or industry-level model and to very broad, aggregate indicators of technological change. • Most models in this tradition would begin with aggregate productivity functions using some variant of the capital-labor-materials (KLM) or capital-labor-energy-materials (KLEM) specification. These production functions define the available production possibilities and can be used to estimate the optimal mix of inputs to production. • Productivity improvements can be though of as a ‘wild card’ multiplier for components of the productivity functions. Some theories treat all productivity improvements together as a single term while others attempt to distinguish multiple sources of productivity improvement such as technological change. • While very large R&D benefits (operating through the multiplier on production functions) are often estimated using this approach, the level of aggregation makes program-specific assessments difficult.

14. Option Value and Principal-Agent Models (I) • Coase (1960) considered the welfare tradeoffs introduced by entities whose economic activities cause harm to others, but who would in turn be harmed if their activities were limited. When transaction costs are present, the legal allocation of property rights can affect the efficiency outcome as well as the equity outcome. In such cases government intervention can raise total production and efficiency, particularly if large numbers of actors are involved. • Williamson (1979) introduced more specific concepts that apply to transaction costs and offered a framework in which the most economical structure for a given transaction can be evaluated. He defined three relevant attributes of transactions: uncertainty, frequency of exchange, and the degree to which investments are transaction-specific. • Dybvig and Spratt (1983) treated the benefits of adopting innovation or standards as a positive externality, i.e. a public good, showing that improvements in outcomes (Pareto optimality) will generally be available if government intervenes to encourage innovation. The degree of such sub-optimal outcomes depends on the private cost of the innovation and the amount of social benefit that can be gained.

15. Option Value and Principal-Agent Models (II) • The ‘market for lemons’ is first identified by Akerlof (1970) as an effect of incomplete information in the marketplace. The information contained in market prices is often an average that does not accurately reflect product quality. As a result, there is an incentive for producers to offer products of lower quality and withhold products of higher quality. The impact of this type of imperfect information is to restrict the availability of consumer products based on an insufficiently detailed pricing system. • Principal-agent information asymmetry was first studied by Aumann and Maschler for the US Arms Control and Disarmament Agency in the mid-1960s. Extensions of this approach include Maskin and Tirole (1990), who examined the problem of imperfect information and conclude that inefficiency will generically result when agents do not have the full set of information available. The lack of complete information is equivalent to a form of irrationality on the part of agents which restricts their ability to arrive at the efficient (Pareto-optimal) arrangement.

16. Option Value and Principal-Agent Models (III) • Establishing an analytic framework using this approach will generally lead to a focus on the role of information, especially market pricing and consumer’s ability to evaluate relevant product features. • Quantitative estimation of information asymmetry usually occurs on an industry or product-specific level, leading to indicators of the rate of diffusion of new technologies on a similar basis. This is an intermediate level of aggregation (relative to more disaggregate firm-level or transaction-specific indicators). • Specific models estimate the benefits of more rapid technology diffusion in different ways. In general these models require assumptions about the value of product quality to consumers. • R&D benefit estimates using these concepts will be program-specific, with an emphasis on product quality and the deployment of technology. These are not always the focus of particular R&D programs, so these are not always the most useful measures. In addition these models can be so specific to a single product market that they raise compatibility issues vis-à-vis other benefit estimates.

17. Spillover Effects and Network Models (I) • Based initially on studies of the telecommunications industry, spillovers and network effects were more broadly defined by Katz and Shapiro (1986). They extend the concept (consumer benefits of product use as an increasing function of other consumers’ use of compatible products) to industries without physical network characteristics. These include information technology and certain consumer goods, such as video cassettes and durable goods requiring specialized services. • Pepall (1992) developed a model of strategic product choice when consumer preferences combine both vertical and horizontal product differentiation. In such cases, the first mover advantage for producers is maximized when product differentiation is limited by preferences rather than technology (termed ‘niche markets’). As more consumers participate in the relevant market, follower firms offering a range of competing products increase consumer choices and lead to improved market benefits. • Godoe (2000) advanced a linkage between industries’ R&D intensity and their ability to create both incremental innovations and radical innovations. In this model institutional capacity for coordination, direction and leadership can anticipate and internalize the producer incentives offered by spillovers.

18. Spillover Effects and Network Models (II) • Analytic frameworks based on spillover effects and network characteristics focus on the returns to R&D and how specific firms react to these incentives. The emphasis is on groups of competing technologies within a particular market. • Network effects are important for consumers because their expectations about future product compatibility will influence their decisions about adopting current technologies. • Many models using this approach are concerned with strategic behavior and employ techniques from game theory, such as Cournot equilibrium, in order to specify market structure and strategic choices. Externalities arise when potential first mover firms lack the incentive to conduct R&D due to risks from follower firms. Public R&D investment can alter the incentives to private R&D investment. • R&D benefits estimated using this approach will focus on the effect of specific innovations on the full range of compatible products, including the chances for spillovers that improve the diffusion of compatible technologies. Such measures will be program-specific but are only useful when network characteristics apply.

19. Sidebar: Top-Down vs. Bottom-Up Models • Baseline estimation of technological change using computer simulation modeling has been an important element of research into the costs of policy responses to climate change. Some key references include Weyant et al (1990) and the IPCC Technical Assessment of 1995. These sources consider the ‘top-down’ vs. ‘bottom-up’’ debate explicitly. • ‘Top-down’ models are associated with macroeconomics and share the characteristics of growth theory approaches as discussed earlier in this presentation. ‘Bottom-up’ modeling draws on optimization techniques largely drawn from operations research, placing most of the literature outside the research developed in this study. • Optimization modeling is often used for electric power systems, because of theinterconnected nature of the transmission grid, prevalence of centralized dispatch for generation, relatively small number of decision agents, and high quality of data. • Integration of both approaches may be appropriate for thorough and consistent R&D benefit assessment, but is resource-intensive for individual programs.