12 Modeling Induced Technological Change: An Overview – Technological Change and the Environment

Chapter 12

Modeling Induced Technological
Change: An Overview

Leon E. Clarke and John P. Weyant

12.1   Introduction

This chapter explores induced technological change (ITC) in the context of research and policy models of energy, the environment, and climate change. What elements of ITC can and should be included in these models? What should be left out and considered qualitatively in interpreting the models? How have modelers thus far included ITC? Are certain modeling approaches more amenable to ITC than others?

Almost everyone—researchers and policy makers alike—agrees that the response of technology to economic incentives and to policy over the coming decades may be crucially important in designing energy and environmental policies. Thus, questions about the optimal timing and stringency of greenhouse gas abatement policies have become increasingly concerned with assumptions about technological change in economic models. In addressing questions about the optimal timing of carbon abatement, Grubb et al. (1995) helped focus attention on the need to fully endogenize the rate and direction of technological change. Since then, Goulder and Schneider (1997, 1999), Goulder and Mathai (1998; also Chapter 9 in this volume), Grübler and Gritsevskyi (1998), Gritsevskyi and Nakicenovic (2000; also Chapter 10 in this volume), Nordhaus (Chapter 8 in this volume), and others have constructed policy models that include ITC.

The purpose of this chapter is not to critically evaluate particular model findings or to draw policy conclusions from them. It is to examine the methodology by which ITC has been modeled and to find inherent modeling limitations and opportunities for improvement, drawing on insights from the existing economic literature on technological change. We do not attempt a comprehensive literature review, only a selective one to highlight relevant ideas and present important distinctions. [Other reviews and related discussions can be found in Carraro (1997), Weyant (1997); Grübler et al. (1999); and Jaffe et al. (2000).] A critical question concerns how much confidence we can expect to place in models of long-term technological change, and how much of the analysis should rest on the qualitative insights of policy makers. Policy makers should ideally consider all important elements of ITC, but only some of these elements are amenable to inclusion in climate-change models.

Figure 12.1. Set of Issues Appropriate for ITC Models, as a Subset of a Larger Set of Issues.

Figure 12.1 is a simple visual representation of how one might consider the elements of ITC, just as modelers have already had to consider the complex elements of global social, economic, and climate systems. First, there is a set of possible phenomena and dynamics that are relevant to ITC, represented by the largest circle. But policy makers cannot consider everything: they must condense the system into something more manageable, even at a qualitative and intuitive level. Hence, the set of phenomena that are policy relevant is a subset of the larger set of issues, as illustrated by the mid-sized circle. Further, only a subset of the policy-relevant phenomena and issues is amenable to mathematical modeling: modelers cannot include everything that is important, just as economic and climate-change modelers more generally do not include everything that is potentially important. The smallest circle captures the set of policy-relevant phenomena amenable to mathematical modeling.

Hence, two issues present themselves to ITC modelers: first, which phenomena should modelers consider and which should they leave out as qualitative adjustments to models or as irrelevant, and second, how should modelers incorporate the elements they decide are important and amenable to modeling? An important implicit conclusion of this review is that communication between modelers and policy makers is essential. It is incumbent upon policy makers to understand that models capture only a portion of reality, to ascertain what is in and what is out, and to interpret model results accordingly. It is incumbent upon modelers to clearly convey this information to policy makers.

The remainder of this chapter is divided roughly into two parts. The first part reviews current thinking about technological advance in energy and environmental models. We begin with a definition of ITC in Section 12.2. We then present a selective review of the economic literature on technological advance in Section 12.3; discuss spillovers, including the empirical evidence on their magnitude, in Section 12.4; elaborate on the notion of innovation market failures more generally and the implications for climate-change modeling in Section 12.5; discuss prominent examples of state-of-the-art climate-change models with ITC in Section 12.6; and conclude with a summary discussion of ITC in energy and environmental models, based on the previous sections, in Section 12.7. The second part of the chapter, Section 12.8, looks at the limitations of current ITC modeling approaches and potential extensions. We explore notions such as complementary sources of technological change, heterogeneity, uncertainty, path dependence, and diffusion. Concluding remarks are presented in Section 12.9.

12.2   Definitions

What is ITC? This is actually a two-part question: what is “technological change”—what is it that is changing—and what does it mean for technological change to be “induced”? Broadly defined, a society's technology refers to the set of goods and services the society can produce and the methods, or the input combinations, by which it produces these goods and services. The introduction of the internal combustion engine was technological change: a new good and associated services became available. Subsequent improvements in the internal combustion engine are also technological change: cars are faster, more powerful, more comfortable, and more reliable than 100 years ago, and the methods of production are more advanced and efficient. And technological change is not just a matter of machinery. Fast food and the “Wal-Mart method” of retailing also constitute technological change.

Climate-change models capture technology through mathematical functions that specify the possible input–output combinations available to a society, a country, an industry, or a representative firm. Our definition of technology here should be consistent with this mathematical approach. In climate-change models, and economic models more generally, mathematical representations of possible input– output combinations are often referred to as “production functions.” This is the definition we will use here: technology is the production function. The essential point to remember is that for technology to change, the production function must change. Changing the set of feasible products and processes is technological change, choosing among products and processes is not.

What does it mean for technological change to be induced, or endogenous? One requirement is that technological change must respond to more than just the passage of time, for that would be exogenous technological change. But this requirement does not sufficiently pin down the notion of induced technological change. In one common paradigm, ITC refers to the fact that the research and innovation decisions of firms and individuals are influenced considerably by the private costs and rewards of innovation. For example, dreams of enormous wealth, some of them realized, are responsible for many of the technological advances coming out of Silicon Valley. Hence, technological change does not fall like manna from heaven, but is endogenous to the social and economic system. In this paradigm, technological advance is strictly a private sector phenomenon associated with private sector incentives. Governments can alter the rate and direction of technological advance by altering these incentives. Emissions taxes, for example, change the relative prices of polluting and nonpolluting technologies, and therefore the incentives to improve these technologies.

In this chapter we use a broader notion of ITC: technological advance responds to policy. The goal in taking this broader perspective is to emphasize that our ultimate concern is policy and, further, that policies affect technological advance through a multitude of mechanisms, not just through changes in the relative prices of polluting and nonpolluting goods brought on by emissions taxes. Governments have an extensive history of acting directly on technological advance through research and development (R&D) at national laboratories and universities, through research grants to private firms, and, more recently, through public–private research partnerships. Policies can also directly target the diffusion of technology—a salient concern in the context of global warming because technology in less developed countries lags significantly behind that in developed ones.

12.3   A Selective Review of the Economic Literature
on Technological Change

The economic literature on technological change is vast and diverse. It includes traditional neoclassical work, historical studies, “appreciative” theory, econometric efforts, and less mainstream mathematical work such as that in evolutionary economics. [See Ruttan (2001) for a thorough discussion of the ITC literature.] In this section, we will conduct a brief review of two particular strands of economic theorizing about technological change. First, we refer to “innovation theory” as the microeconomic study of innovation—roughly defined as the development of new products and processes—using neoclassical economic concepts. Innovation theory focuses on firms and industries. It explores firms’ incentives to improve technology (“We'll make millions!”) and the inefficiencies that result from firms’ failure to share the gains of their innovative endeavors (“We've got to patent this thing!”). The second category of conventional economic theorizing concerns macroeconomic growth. “Endogenous growth theory,” or “new growth theory,” borrows insights and methods from innovation theory and applies them within the context of neoclassical growth models. It looks at how innovation investments by private actors might prove to be a source of long-term economic growth. The new growth theory models of technological change are further abstracted from firm and industry behavior than is innovation theory, but as macroeconomic models, they are more consistent with climate-change models.

To be clear, innovation theory and endogenous growth theory are not the only important lines of economic research into technological change, but they are a good starting point for looking at efforts to date to include ITC in climate-change models. One reason is that they use the traditional neoclassical analytical structures that underlie much of the state-of-the-art climate-change work—they are working from the same manual as many climate-change modelers. Further, they are bound together by a common focus on one particular aspect of technological advance, namely, that technology is associated with knowledge, knowledge is a public good, and, hence, we should expect firms to spend less on improving technology than we as a society would like. In Section 12.8 we make up somewhat for our selectivity here. There, we give a broader view of technological change and the associated literature and discuss possible extensions to, and limitations of, current ITC models raised by the wider body of relevant research.

12.3.1   Innovation theory

Innovation theory asks why and how firms invest in innovation, how effectively they appropriate the fruits of their innovation investments through profits, and how efficient the resulting system is at the industry level. Technology and knowledge are often viewed as synonymous in this paradigm; since technological information is a public good, the private sector probably underinvests in innovation. The main contributions of innovation theory have been to emphasize both the importance of knowledge in markets for innovation and the private profit incentive as a key source of innovation activity, and to clarify the complex and imperfect competition in which innovations arise and are put to use.

In his seminal work, Arrow (1962a) demonstrated that neoclassical mathematical economic models could be used to explore firm behavior and incentives to innovate. He showed that an innovating monopolist and an innovating firm licensing technology into a competitive market have different incentives to innovate, and both have weaker incentives than would be socially optimal, even if none of their knowledge leaks out. In other words, market structure is important in technological advance, and private actors probably underinvest in innovation. Arrow further noted that a firm's ability to appropriate the gains from innovative activity may spur innovation, but it does so at the price of inefficiencies following innovation: “in a free enterprise economy, the profitability of invention requires a nonoptimal allocation of resources” (p. 617). If everybody shared everything they know, it would dilute incentives to innovate. These dual results of appropriability, its countervailing incentive and efficiency effects, are often referred to as a tension between dynamic and static efficiency. From Arrow's simple model has grown an extensive and increasingly complex line of mathematical theoretical research, relying heavily on game theory, into the interactions between firms before and after innovation. [Good reviews of seminal work in this literature can by found in Kamien and Schwartz (1982) and Reinganum (1989).]

Arrow also characterized technological information as being freely available and generally applicable, and he asserted that it is the primary resource for innovation. From this and related work has emerged a notion that basic science provides the technological paradigms, or raw material, for major cycles of innovation. Applied research draws from a pool of readily available technological information derived from basic science or basic research [see Evenson and Kislev (1975) for early work in this paradigm]. For example, basic research in physics helped set the stage for the development of transistors and, consequently, the computer on which this chapter was composed. The more basic the research, the less appropriable the resulting information and the lower the private incentive for investment. Consistent with this notion, the innovation literature frequently distinguishes between technological breakthroughs and the resulting, more incremental, product and process improvements. Innovation theory implicitly focuses on the second stage of the innovation process because it is the most predictable and it most clearly responds to market forces.

Innovation theorists have developed various notions of the interaction between market forces and technological advance. On the one hand, market-pull theory, generally attributed to Schmookler (1966), supposes that technological advance is largely the result of conscious responses to market forces: firms identify opportunities and then create the products or processes to take advantage of the opportunities. Schmookler hypothesized that major innovations create a new product frontier that is common knowledge. Identifying and bringing innovative products to market under the new paradigm is straightforward; the challenge to entrepreneurs is gauging and responding to market needs. Alternatively, technology-push theory, stated formally by Rosenberg (1976), asserts an opposite causal direction: technology evolves over time as the result of unpredictable product and process innovations. Technology-push theory emphasizes the uncertainty involved in innovation. Firms cannot predict the results of their innovative endeavors and therefore respond less directly to market forces than market-pull theory asserts.

In the 1980s, economic theorists were confronted by analytical barriers in trying to extend neoclassical models, and a number of researchers shifted their focus to empirical work. A primary new result of this generation of empirical work was the identification of a surprisingly large range of methods by which firms appropriate the returns from innovation. Instead of patents, the most commonly cited method of protecting innovations was reported to be a combination of learning curves, lead-time effects, and trade secrets.

In summary, innovation theory has demonstrated that profit incentives account for a major source of innovative activity, but that appropriating these profits leads to inefficient monopolistic behavior. Further, because knowledge is not fully appropriable, private markets probably underinvest in innovation. The simple lesson for climate modelers is that technological change is endogenous to the economic system—that is, it can be induced by policy—but the market response to technological opportunities is probably less than socially optimal.

12.3.2   Endogenous growth theory

Researchers in endogenous growth theory were interested in understanding how the neoclassical growth model might be extended or modified to better capture two important empirical observations: that economies have been able to maintain extended periods of economic growth and that a wide gap remains between rich and poor nations. Researchers in endogenous growth theory absorbed innovation theory's lessons about knowledge spillovers and incorporated them into the traditional growth models by making these models more rigorous about the manner in which human capital and technology change over time (Romer 1986, 1990; Grossman and Helpman 1991b; Aghion and Howitt 1992).

The new growth models, reviewed by Romer (1994), Grossman and Helpman (1994), and Jorgenson (1996), differ from traditional growth models in that they no longer assume exogenous technological advance. Instead, purposive, profit-seeking investments in knowledge are critical to the long-run growth process. In particular, the nonappropriable aspects of new technology created by profit-seeking firms create spillovers, or positive externalities. Endogenous growth theory has shown how positive externalities of this kind can lead to steady, long-run economic growth. A major contribution of Romer's important work (Romer 1990) was the integration of a neoclassical innovation theory model in the spirit of Arrow into a neoclassical growth model in the spirit of Solow. He demonstrated theoretically one combination of technological opportunities and market structure that might result in sustainable, long-run growth.

Because endogenous growth theory is still a relatively young field, it remains largely a theoretical structure, and computable models based on it are not yet available. Still, it provides important lessons for climate-change modeling. Endogenous growth theory emphasizes the importance of spillovers in modeling technological change and points to how such spillovers might be incorporated into aggregate economic models. From the premise that spillovers are a fundamental source of economic growth, it follows that any model of long-term technological change needs to incorporate spillovers.

12.4   Spillovers and Appropriability

Economists have long recognized that markets invest inefficiently in innovation. In Section 12.3 we looked at two particular strands of economic research held together by a common focus on one cause of this inefficiency, namely, that technology is associated with knowledge, and knowledge is a public good. There are other reasons for inefficiency, and we will touch on these in Section 12.5 and discuss them in more detail in Section 12.8. Nonetheless, because the notion of knowledge as a public good has taken on such a prominent role in economic theorizing about technological change, it is useful to explore the notion in more detail.

The “technology-is-knowledge-is-a-public-good” notion is captured by the dual phenomena of appropriability and spillovers. While appropriability and spillovers are widely recognized as fundamental aspects of technological change and economic growth, they are talked about and measured in many different ways. Here, we draw some important distinctions in the discussion of appropriability and spillovers, and then summarize the lessons from associated empirical work.

12.4.1   Spillovers versus appropriability

Spillovers and appropriability are two sides of the same coin: what innovators do not appropriate spills over. When we discuss one, we are implicitly discussing the other. But there are philosophical and practical reasons to distinguish between the two, and these reasons help to explain why innovation theory concentrates on appropriability and endogenous growth theory focuses on spillovers. The distinction has much to do with the underlying research agenda.

Endogenous growth theory observes economies at the macroeconomic level, developing models to account for aggregate-level growth. Growth is viewed as a consequence of firm innovation. Hence, spillovers are viewed as positive externalities: society gets more from innovation than do firms because of spillovers. As endogenous growth models develop more microeconomic rigor and detail, they are more explicit in modeling the individual investment incentives of firms and how innovation comes about (e.g., Romer 1990). The primary research focus, however, remains on innovation and spillovers as a combined source of macroeconomic growth, and, hence, on spillovers as positive externalities.

Innovation theory, on the other hand, examines firms at the microeconomic level, trying to understand their innovation incentives and therefore to describe their behavior. Innovation theory therefore places its emphasis on appropriability's implications for investment incentives prior to innovation and on its market structure effects immediately following innovation—the tension between static efficiency and dynamic efficiency. In this sense, spillovers are no longer a strictly positive externality because they hold back innovation.

12.4.2   Forms of spillovers

A second source of confusion about spillovers involves the form of the spillover. Just what is spilling over? (For simplicity, hereafter we refer simply to spillovers rather than to spillovers and appropriability.) At least as far back as Griliches (1979), many economists have—sometimes explicitly and sometimes implicitly—organized spillovers roughly into two categories: “rent spillovers” and “knowledge spillovers.”

The distinction is clearest if we imagine that research has two products: “blueprints” and “knowledge.” Blueprints are designs of specific products or processes that can be used to generate profits. For example, research at Goodyear might result in a new tire material that nets Goodyear a solid profit, justifying their research expenditure. Rent spillovers refer to firms’ inability to appropriate all the benefits associated with the use of the specific product or process described in the blueprint. Use of the knowledge embodied in the blueprint to create new blueprints or new knowledge falls under the heading of knowledge spillovers, which we will discuss shortly.

There are a number of ways rents might spill over. For one, innovators cannot appropriate the full social benefits of their blueprints unless they can perfectly price discriminate, even if they are able to hide the blueprint from their competitors and appropriate all the knowledge for themselves (Arrow 1962a). Further, other firms may imitate the product in the blueprint. For example, Pirelli may copy the Goodyear tire, thereby cutting into Goodyear's profits. If the two firms compete aggressively, the price of the new tire may be driven down, and not only will Goodyear fail to appropriate the full social returns, but the returns may accrue largely outside the tire industry. A portion of the returns will be embodied in quality improvements to tires, which might, in turn, increase the productivity of the trucking industry. This latter concern, where the innovative rents end up, is an important issue in statistical work on spillovers (again, see Griliches 1979). Empirical case studies (e.g., Mansfield et al. 1977; Trajtenberg 1990) focus largely on rent spillovers, attempting to determine the private and social returns from the use of particular innovations.

Blueprints are not the only product of research efforts. Research creates new knowledge, some embodied in blueprints and some not, that can be used by others as an input to new research. For example, inspection of the Goodyear tires may give Pirelli researchers ideas of how to improve their own tires without actually copying the Goodyear tires. Or researchers in a less closely related field might read in a professional journal about a coincidental discovery made while developing the new tires and use the ideas to improve their products or processes. In the new growth literature, the idea that researchers can build on the knowledge of others takes a number of forms. For example, it often takes the form of “quality ladders,” in which each innovation is an improvement to an existing product, as in Grossman and Helpman (1991a) and Aghion and Howitt (1992). In Romer (1990), it takes the form of an aggregate knowledge parameter—actually a count of the number of blueprints in the economy—that is added to with each new blueprint. Jones and Williams (1996) use a combination of these two approaches.

The distinction between rent spillovers and knowledge spillovers is an issue of some concern in statistical work. Griliches (1979) refers to the short-term productivity effects of rent spillovers as “pecuniary” externalities from declining real prices. He argues that their social product should be computable in principle from declining real factor prices, and, hence, that such spillovers are not knowledge spillovers and thus are not really spillovers. Regardless, rent spillovers are important to understand because they help to drive a wedge between social and private returns to R&D. They are also complex: the availability of new technologies may carry implications that go beyond mere cost reductions and quality improvements. New or different goods not only reduce other firms’ production costs, they alter other firms’ costs and opportunities for innovation. For example, the development of increasingly complex and higher-speed microprocessors by the semiconductor industry has had a profound impact on the types of products offered by the telecommunications and computer industries. In this way, the knowledge “embodied” in the new or cheaper good, the blueprint, does serve as a positive externality on innovation productivity, even if firms only use the good as an input and do not directly exploit the new knowledge embedded in it.

Regardless of the type of spillover, there are many ways that knowledge may be transmitted through an economy. Levin et al. (1987) conducted a particularly ambitious and important study of spillover channels (see also Levin 1988). They surveyed 650 R&D executives in 130 industries regarding the methods by which they learn about competitors’ technology. They included seven transmission mechanisms in the survey: licensing technology, patent disclosures, publications or technical meetings, conversations with employees of innovating firms, hiring of employees of innovating firms, reverse engineering, and independent R&D. Broadly speaking, they found that all methods of transmission were important under certain circumstances, although some were more important on average than others. Of interest, many of the most important modes of transmission required imitating firms to allocate significant resources to imitation, contradicting the notion that knowledge is “free.” [Cohen and Levinthal (1989) discuss this issue in more detail.]

12.4.3   Level of spillover

A third set of distinctions concerns the level at which spillovers occur. A first distinction is between intrasectoral and intersectoral spillovers. Intrasectoral spillovers occur within a particular industry as firms benefit from the innovation and development activities of competitors. If Intel discovers a way to pack more into their chips, they will care most about hiding the information from other chip manufacturers, AMD, for example. They will care little what Ford or General Motors might do with the information. This is the sort of spillover considered in innovation theory models because it is the most clearly associated with appropriability and it is the most predictable to innovators. Intersectoral spillovers, on the other hand, occur between industries, which may borrow products or ideas, or be stimulated by developments in related fields. Intersectoral spillovers are important for assessing the economy-wide impacts of innovation, but they are a more difficult modeling challenge, and one that may not be immediately appropriate for climate-change modelers.

Spillovers may also be distinguished by their geographic spread. Local spillovers occur within regions or countries. International spillovers work across national or regional boundaries. International spillovers are starting to be seen as a potentially positive feedback for R&D on environmental control technologies. For example, renewable energy technologies that are competitive in the markets of the Organisation for Economic Co-operation and Development (OECD) economies may also have large global benefits, allowing low-cost emissions reductions in developing countries. However, empirical studies provide evidence that many spillover externalities tend to be limited locally or nationally. A number of authors have suggested that institutional, political, and even cultural factors may prevent innovating firms or industries in one country from collecting innovative rents in another, especially in industries, such as energy, with a “national” character (see, e.g., Grossman and Helpman 1991b; Fagerberg 1994; Abramovitz and David 1995).

12.4.4   Empirical evidence: Spillovers and returns to R&D

In new growth theory, spillovers are the essential phenomenon that allows economies to maintain long-run growth. In innovation theory, spillovers are the main cause of private underinvestment in innovation, and appropriability is an important cause of monopoly behavior. Theory aside, spillovers are only a concern for energy and environmental models if there is empirical evidence that they are important. A large body of work has developed around this issue [see Griliches (1992) and Nadiri (1993) for surveys of the literature]. As one might expect, measurement of such a complex phenomenon has been difficult, and results have varied by method and application.

Nonetheless, there is a consensus among researchers that spillovers are significant and play an important role in technological advance. In the words of one expert,

In spite of these difficulties, there has been a significant number of reasonably well done studies all pointing in the same direction: R&D spillovers are present, their magnitude may be quite large, and social rates of return remain significantly above private rates. (Griliches 1992:S43)

Nadiri, in his survey of the literature, finds that estimates of firm-level returns average around 20 to 30 percent, whereas estimates of social returns average around 50 percent. Mansfield et al., in their seminal 1977 case study work, find average social returns to R&D of over 50 percent and private returns of about half that.

Two important lessons for climate modelers emerge from the empirical spillover literature. First, private returns to R&D are probably appreciably smaller than social returns, confirming the intuition from theoretical work. This means that the private sector R&D response to environmental policy—for example, an emissions tax—will be less aggressive than we as a society would like. It also opens the door for consideration of technology approaches to environmental problems, an option that has received only limited attention in climate-change models to date.

The second lesson is that a full accounting of spillovers in climate-change models is asking too much. The empirical work makes clear that spillovers are not confined within industries or countries, and, therefore, that models including ITC only within particular sectors or countries may miss intersectoral or international effects. Although these effects are significant, it may be inappropriate for particular climate-change models to include them. For example, if a climate-change modeler includes ITC in the energy industry and only the energy industry—a good first step to be sure—the modeler will neglect the effects of such R&D outside the industry. The point works in reverse as well: some level of technological advance within the energy sector will be exogenous, because it will derive from R&D elsewhere (we will have more to say on these complementary sources of R&D in Section 12.8).

12.5   Innovation Market Failures in ITC Models

There is a consensus among historians, empiricists, and theorists that markets do not invest efficiently in innovation and that underinvestment is significant enough to deserve the attention of policy makers (hence, the patent system). In other words, there are innovation market failures (IMFs), and these failures are important. By IMFs, we mean any characteristic of the market that causes private innovation investment or the results of innovation investments to be different from what would be most beneficial from a society-wide perspective. In the previous two sections we highlighted one cause of IMFs: because knowledge is a public good, markets are prone to underinvest in knowledge creation. But there are other reasons why markets invest inefficiently in innovation. One reason is that managers in firms may be more risk averse or prefer payoffs sooner than would be optimal from a social perspective. Few CEOs are excited about high-cost, high-risk projects that may not pay off for 20 years. Another reason is the limitations to rational behavior, as emphasized by evolutionary economic theory (see Nelson and Winter 1982). More will be said on these topics in Section 12.8, but readers should keep in mind that, henceforth, when we refer to IMFs, we are referring to the totality of factors that render market innovation decisions inefficient. We assume that IMFs most often reduce innovation investment below what would be best for society.

What are the implications of IMFs for climate-change modeling and policy? One implication is that optimal carbon taxes from models without IMFs—or without ITC at all, for that matter—may be biased. Another implication is that technology-based policy measures may be an important or useful component of climate policy. The strength of these implications rests largely on two issues: (1) how large the opportunities for technological change are and (2) how large the IMFs are. Neither of these is well understood at this juncture, but they are potentially important enough to justify the inclusion of ITC and IMFs in climate-change models. Below, we discuss some of the important challenges to incorporating IMFs in climate-change models. In Sections 12.6 and 12.7, we will discuss how ITC modelers have gone about tackling these and other challenges.

First, IMFs are complex. Even simple, two-period innovation theory models that focus exclusively on knowledge appropriability have enough complexity to limit strong general conclusions. In climate-change modeling the situation is many times more complex. For one, the time frames are often on the order of 100 to 200 years. Further, the scope of analysis is large enough to include intersectoral and international effects. Moreover, appropriability is not the only factor in markets for innovation. There are issues of private discounting and risk aversion that may be crucial for high-risk, high-cost, long-lead-time technologies—those “backstop” technologies that may ultimately play a large role in pollution reduction. Hence, a first challenge for climate-change modelers is to incorporate the enormous complexity of innovation systems without overwhelming the models or the intuition of those interpreting the models. It is our belief that climate-change modelers will be most effective if they capture this complexity in simple, heuristic, approximate models of nonoptimal innovation behavior rather than by attempting to capture all its nuances. More will be said in this regard in Section 12.7.

Figure 12.2. Stylized Representation of Decision Making in Energy and Environmental Models.

A second modeling challenge entails an asymmetry between control variables. Modelers often wish to ascertain optimal profiles for the carbon tax; this has been a primary focus of climate-change modeling to date. On the other hand, in the presence of IMFs, the time profile of innovation expenditures should be a socially nonoptimal, and likely a socially suboptimal, response to market prices including any environmental taxes. This asymmetry creates modeling problems. How can modelers determine optimal emissions policy while at the same time incorporating nonoptimal behavior with respect to innovation decisions? For example, how can a social planner act simultaneously optimally and nonoptimally?

Figure 12.2 is a stylized representation of the situation. Climate-change models have traditionally taken technological change as exogenous: they have worked within the dashed box of Figure 12.2, looking only at production and emissions decisions. How should researchers incorporate ITC into the system? One choice is to take innovation decisions as optimal—that is, under the control of a social planner. But this approach neglects IMFs. Another option is to take emissions policy (e.g., a carbon tax) as exogenous and then to model the nonoptimal market response. But this option does not provide optimal policy information. A third option is to draw a box around the whole system—that is, to determine optimal policy given nonoptimal innovation behavior. Modelers might use an iterative procedure: set a policy profile, model the system, change the carbon tax, model the system, and so forth until they reach a solution. But this approach is computationally challenging. Hence, modelers wishing to capture the full system face either this computational challenge or the challenge of developing an integrated framework in which a single decision maker acts both optimally and inefficiently.

Figure 12.3. Linear, Deterministic Model of Technological Change.

12.6   State-of-the-Art ITC Modeling

How have modelers incorporated ITC into climate-change models? It is constructive to classify ITC models into four types: cost-function models, neoclassical growth models, intertemporal general equilibrium models, and bottom-up energy systems models. This list runs in order of increasing technological detail, with cost-function models being the most abstract and bottom-up energy systems models being the most specific and rigorous about technology. In this section we discuss the mechanisms, purposes, and strengths and weaknesses of each type, along with prominent examples. We focus here on modeling methodologies and the importance of ITC, not on modeling extensions or implications for particular policy arguments. Table 12.1, at the end of this section, summarizes and compares the characteristics of five prominent models.

The innovation and growth literature emphasizes two forces behind technological change: technology evolves largely as a result of private investment incentives and appropriability of innovations, and it is enhanced by spillovers, or positive externalities, from these investments. Figure 12.3 is a stylized representation of this linear model of technological change. The models in this section do not all precisely follow the dynamic in Figure 12.3. For example, some models use experience-based technological advance instead of R&D, and some models do not include IMFs. Nonetheless, the dynamic in Figure 12.3 is a good starting place for thinking about state-of-the-art ITC modeling.

As we proceed through the types and examples of climate-change models in this section, we will focus on three key dimensions of ITC modeling. First, who are the decision makers? Most important in this regard, who makes the emissions decisions and who makes the innovation decisions? For example, social planning models assume an omniscient social planner, whereas private sector actors make decisions in general equilibrium models. Second, how are IMFs included in models? Do models explicitly capture interactions between firms, or are approaches more stylized? Third, how does technology improve? Do firms invest in R&D, or do they simply learn how to do things better the more often they do them? In drawing conclusions about the importance of ITC, we must keep in mind basic assumptions about these three dimensions.

12.6.1   Cost-function models

Cost-function models are the most abstract of the ITC models. They contract all emissions decisions and their economic consequences into a single abatement decision: how far should emissions be reduced from an exogenously given, baseline emissions path? Technological advance entails changes in the abatement cost function, which derives from the production function, making it cheaper to reduce emissions from the baseline path. With ITC, changes in the abatement cost function can be induced, or controlled, along with period-to-period abatement levels. The simplicity of cost-function models allows them to focus on general and theoretical conclusions.

Goulder and Mathai's (Chapter 9 in this volume) set of models is a prime and important example of cost-function modeling. [Tol (1996) uses a similar modeling framework.] As is typical in theoretical work, Goulder and Mathai capture ITC through changes in a “knowledge” parameter in the abatement cost function. Abatement costs at each point in time are a function of abatement levels and the knowledge level. They use two formulations to capture how knowledge increases over time: R&D and experience. In the R&D formulation, explicit expenditures on research increase the knowledge stock and thereby reduce future abatement costs. In the experience formulation, technological advance derives simply from the act of abating: the more we abate, the better we are at it. (Both of these mechanisms are discussed in more detail in Section 12.7.) An important characteristic of Goulder and Mathai's model is that an omniscient social planner makes all decisions: there are no IMFs. (Below, we discuss Nordhaus’ efforts to include IMFs in a social planning framework.)

Goulder and Mathai explore optimal abatement and emissions tax trajectories under both a cost-effectiveness criterion (how to achieve a target most cheaply) and a benefit–cost criterion (what is the optimal level of abatement?). Their work is a response to the “timing issue”—assertions that ITC calls for greater near-term abatement activity or tax levels relative to models in which technological change is exogenous. From first and second derivatives, Goulder and Mathai draw strong theoretical conclusions about ITC's effect on optimal abatement decisions, challenging assumptions that ITC calls for more aggressive up-front actions.

12.6.2   Neoclassical growth models

Neoclassical growth models are the most consistent with work in endogenous growth theory, as they typically employ the same aggregate formulation of technology based on inputs of capital and labor. Neoclassical growth models of climate change add an input to capture the impacts of emissions or emissions control. The social planner makes innovation decisions in neoclassical growth models.

The DICE model (Nordhaus 1994) is the most widely known of the neoclassical models. Nordhaus builds on DICE to create the R&DICE model, which incorporates ITC (see Chapter 8 in this volume). Both DICE and R&DICE represent production as a function of labor and capital, and both include a compact climate-change model that translates carbon outputs into a damage function reducing total output, and thus per capita consumption. The objective is to maximize discounted per capita consumption by controlling capital investment, carbon emissions, and, in the R&DICE model, R&D.

Nordhaus splits technological change into two components. The first component, which Nordhaus takes as exogenous, affects the productivity of labor and capital (it is the standard technology term in neoclassical growth models). The second component, technological advance in the carbon intensity of production, is endogenous. Nordhaus captures this endogeneity through a research productivity equation.

Because a social planner makes all decisions in neoclassical growth models, it is natural to assume that such models cannot include IMFs. Nordhaus demonstrates that this need not be the case. He captures IMFs in an approximate, heuristic fashion that maintains the computational advantages of the social planning framework: he simply raises the cost of research resources by a factor of four. This serves two purposes simultaneously. First, it puts a brake on carbon energy research; the market invests at a lower level than would be called for by the price of R&D services. Second, by including the inflated R&D cost in the objective function, Nordhaus accounts for the opportunity costs of carbon energy innovation—such innovation may pull resources from other sectors where their impacts are also supernormal. Although Nordhaus combines these two effects, they are potentially separable. The model could be solved using the inflated R&D costs and the final summary statistics calculated on uninflated values. This approach would capture intrasectoral IMFs, but not the opportunity costs of R&D.

Of importance, Nordhaus endogenizes only one particular sort of technological advance: departures from the assumed path of energy-efficiency improvements. This allows Nordhaus to isolate the effects of ITC in the energy sector by comparing his DICE and R&DICE results. He compares the results from a calibrated version of the R&DICE model (endogenous technological change but no substitution) with those from the DICE model (substitution and exogenous technological change).

12.6.3   Intertemporal general equilibrium models

Intertemporal general equilibrium models divide an economy into distinct sectors and then model economic activity, including interactions between the sectors, over time. The approach is computationally demanding and requires detailed data on the chosen sectors. A strength of general equilibrium models is that they allow researchers to explicitly consider interactions between sectors, something that is missed in the more abstract cost-function and neoclassical growth approaches. In addition, general equilibrium models are a more explicit representation of real market behavior than cost-function or neoclassical growth models, because the different sectors do not work together as an integrated entity. Two notable intertemporal general equilibrium models of the US economy related to pollution policy and ITC are those of Jorgenson and Wilcoxen (1993) and Goulder and Schneider (1997, 1999).

Jorgenson and Wilcoxen's model is distinguished by the econometric methods used to calibrate the model coefficients. Coefficients are based on extensive time-series data (1947–1985) of interindustry transaction tables. Jorgenson and Wilcoxen use the econometrically calibrated model to develop quantitative estimates of the impacts of environmental policy up to 2020. But the Jorgenson and Wilcoxen model does not include ITC as we have defined it in this chapter; in their model, changes to the production function(s) are exogenous. Prices and policy affect the economy's use of the changing production function, but not the production function itself.

Goulder and Schneider, on the other hand, do include ITC as we have defined it in this chapter. Representative firms can enhance their production capabilities by investing resources in knowledge accumulation. A strength of Goulder and Schneider's approach is their explicit inclusion of IMFs. They capture IMFs by modeling intrasectoral spillovers in a manner consistent with endogenous growth models. The accumulation of knowledge for representative firms is only partly appropriable, so markets as a whole underinvest relative to the social optimum. Goulder and Schneider do not attempt to ascertain optimal carbon taxes. Instead, they consider the effects of different carbon tax and R&D subsidy levels.

Another strength of the Goulder and Schneider model is the distinction they make between an alternative, or carbon-free, energy sector and a conventional energy sector. This distinction allows them to begin to address issues of technological heterogeneity, a topic we will return to in Section 12.8. In doing so, Goulder and Schneider must make assumptions about the direction of technological change in each industry, assumptions that influence their results in an interesting fashion. They assume that technological advance in the conventional energy industry cannot be energy/carbon reducing, only productivity improving. Hence, such advance increases carbon emission levels through cheaper, dirty fuels.

A final interesting component of the Goulder and Schneider formulation is the inclusion of an R&D services sector that supplies all R&D resources to innovating industries. Increases in alternative fuel research may have indirect effects on research in other sectors by reallocating a limited supply of R&D services. This raises the opportunity costs of environmental R&D. Nordhaus pursued the same notion, but less rigorously, in the R&DICE model by raising the costs of carbon energy R&D above the market price.

12.6.4   Bottom-up energy systems models

Whereas all the models discussed thus far capture technology through parameterized production functions of one sort or another, bottom-up energy systems models are specific about the individual technologies that serve the energy sector. They explicitly represent individual technologies or clusters of technologies, sometimes hundreds of them. Bottom-up models typically seek to minimize the costs of serving an exogenous energy demand by choosing which technologies to install, where, and when. A number of researchers have extended bottom-up models to include experience-based ITC. Three of these models are reviewed in

Seebregts et al. (1999). All fall within a social planning framework and do not include IMFs.

The defining characteristic of bottom-up energy systems models is their technological detail and the resulting ability to explore the impacts of technological heterogeneity. By distinguishing between individual technologies, they demonstrate discontinuities in aggregate energy systems characteristics as previously uncompetitive technologies—emerging technologies such as photovoltaic cells or fuel cells—reach market thresholds and begin to contribute to the energy system. They show that investment in a subset of technologies, rather than investment across the whole spectrum of technologies, drives long-term technological change. In other words, the allocation of innovative activity may be as important as the level of innovative activity. Investment in emerging environmental technologies, however it comes about, can also be viewed as an important hedging strategy against uncertainty.

Several authors have attempted to include uncertainty in bottom-up energy systems models, although the computational challenges are daunting. Grübler and Gritsevskyi (1998) minimize the computational requirements by simplifying down to three technologies: existing, incremental, and revolutionary. They go on to jettison the social planning assumption, and instead assume a set of regional actors and energy suppliers that must plan their technology deployment decisions under uncertainty (see Chapter 11 in this volume). They allow for ITC in 12 technologies. Gritsevskyi and Nakicenovic (Chapter 10 in this volume) include uncertainty in the MESSAGE model framework by reducing the full possible uncertain space to a set of 250 scenarios, and the possible technology installation scenarios to 520 “technology dynamics.”

12.6.5   Importance of ITC: Conclusions from state-of-the-art models

How important is ITC in climate-change models? Does the inclusion of ITC significantly alter the results from models with exogenous technological change? Most important, are the policy prescriptions different? Does ITC imply greater up-front abatement than in the exogenous models? Does ITC imply greater emphasis on R&D support? The results to date are less than definitive.

At a minimum, we should expect emissions to be more responsive to policy with ITC than without it. But how much more responsive? Nordhaus’ R&DICE results indicate that ITC is less important than technological substitution: reductions in carbon dioxide concentrations and in the global mean temperature with ITC alone are about half those from substitution alone. In Goulder and Schneider's general equilibrium model, a US$25/ton carbon tax noticeably increases alternative energy R&D, especially in the early years, but also dramatically decreases conventional energy R&D. In fact, conventional energy R&D drops to zero in the first three years of their model run! Goulder and Schneider find that ITC noticeably increases the emissions reductions from emissions taxes, thereby increasing gross costs, but also increasing net benefits because abatement is cheaper. The effect is significant: the gross domestic product (GDP) sacrificed to achieve the same cumulative abatement is approximately 25 percent lower with the presence of ITC than without it. However, the market is not as responsive as it might be because of limitations on the level of R&D resources.

How dramatically does the presence of ITC alter the optimal emissions and carbon tax profiles? In models with exogenous technological change (e.g., Wigley et al. 1996) it is best to push abatement toward the future because abatement costs will be lower the longer we wait. Some authors have suggested that ITC might alter this result. Goulder and Mathai, using their cost-function model, arrive at two strong theoretical conclusions in this regard. First, if innovation derives from R&D—and assuming no IMFs—the larger the potential for innovation, the further abatement should be pushed into the future. This is an intuitive result; the greater the potential to improve technology is, the better technology will be in the future—assuming we spend on the necessary R&D, and this is an important assumption—and the cheaper future abatement will be. If technological advance results solely from experience, on the other hand, the timing result is ambiguous. The lower future abatement costs are, the further abatement should be pushed into the future; but future abatement costs are a function of near-term abatement levels, which calls for more near-term abatement. Nordhaus's R&DICE model also addresses the timing issue. He finds that ITC has a negligible impact on the optimal carbon tax profile because ITC has a negligible impact on the path of climate change.

Another relevant policy issue is the importance of near-term government technology support. The Goulder and Schneider model directly compares R&D subsidies and emissions taxes. The authors find that the GDP costs of a 15 percent emissions reduction over the next century are approximately nine times higher with a targeted R&D subsidy alone than with a carbon tax alone. Hence, Goulder and Schneider indicate that R&D subsidies should be used as a complement to emissions taxes, not as a substitute for them.

The energy systems models also address issues of policy, albeit less directly. The models in Seebregts et al. (1999) make clear that the allocation of inventive effort is an important concern, something that is missed in models based on parameterized production functions. What matters in these models is the emergence of technologies that heretofore were uncompetitive. Whether these emerging technologies come to market through experience effects, as in the energy systems models, or through R&D investment, the conclusion remains the same: large technological changes will come from the emergence of new technologies.

An important, related result of Gritsevskyi and Nakicenovic's analysis is that there are many potential “local equilibrium” technology paths that might result from market forces, and hence significant uncertainty as to where the market may go. They found that 53 of their 520 technology dynamics result in risk-adjusted expected outcomes that are essentially identical in terms of energy systems costs. However, the dynamics are very different from a climate-change perspective, with some resulting in significantly greater carbon concentrations than others. This

Table 12.1. Characteristics of Five State-of-the-Art ITC Models.

Note: ITC = induced technological change; R&D = research and development. implies that a small initial push one way or another may have significant long-term effects on technology paths and climate change.

All in all, it is difficult to interpret the early ITC models because so much rests on modeling assumptions—most notably, the level of IMFs and the potential for technological advance—and on the mathematical structures of the models. Further, the sample size is still relatively small. What we can say is that the early models do not provide strong evidence for significantly lower abatement costs, different optimal carbon taxes, or changes in the optimal timing of abatement from models with exogenous technological change. However, there is more work to be done.

12.7   Fundamentals of ITC Modeling

Whether implicitly or explicitly, all models of ITC must address three central modeling issues: the innovation decision maker(s), IMFs, and the characteristics of technological advance. This section reviews the manner in which the state-of-the-art models deal with each of these elements.

12.7.1   Decision makers

While it may seem obvious, every ITC model needs at least one innovation decision maker. Models fall into one of two camps in this regard: the decision maker may be the social planner or it may be market actors. The nature of the incentives will vary depending on the decision maker. In social planning models such as those of Goulder and Mathai and Nordhaus, and the energy systems models, the incentive relates to social welfare. In models that attempt to capture market behavior, such as that of Goulder and Schneider, the market actors choose innovation levels, so innovative rents—profits—provide the incentive for technological advance.

12.7.2   Innovation market failures

IMFs are another essential matter in ITC modeling. Somehow, models must capture the fact that markets invest inefficiently in innovation. To fail to do so is to miss a phenomenon that historians and theorists alike deem critical to technological advance. How IMFs find their way into climate-change models depends on who is making the innovation decisions and what the models are trying to determine.

Social planning models aim to ascertain optimal policy; they put innovation and production decisions in the hands of the social planner. As we touched on in Section 12.5, a major challenge for these models is how to maintain the social planning framework's simplicity while including inefficient innovation markets. Goulder and Mathai and most energy systems models sidestep this issue by neglecting IMFs. Nordhaus, on the other hand, includes IMFs through a simple model adjustment. He maintains a central planner throughout, but approximates the vast and complex array of IMFs by increasing the cost of carbon-reducing R&D.

Market-based models, such as Goulder and Schneider's general equilibrium model, focus on market responses to policy rather than optimal policy itself. Because they try to model real market behavior, including inefficiencies, market-based models are conceptually more amenable to IMFs: there is no conflict between optimality in policy and nonoptimality in innovation decisions because optimal policy is not the goal.

So there is a challenge. Modelers wishing to determine optimal policy in the presence of IMFs must somehow combine two actors, the optimal policy maker and the suboptimal innovation decision maker, into a single integrated model. Coming from the social planning perspective, Nordhaus has attempted such a combination. Models coming from the market-based perspective could attempt such a combination through nested optimization, but no modeler has yet attempted this computational challenge.

A Proposal. IMFs are complex. We believe that it is acceptable to be approximate about IMFs: modelers should accept and use approximate, heuristic models of IMFs that result in socially nonoptimal, and typically suboptimal, innovation behavior. In other words, models should concentrate on the observation that markets underinvest in innovation and worry less about the economic rigor of the modeling approach used to achieve the underinvestment.

Nordhaus's R&DICE model is a good example of this approach. By increasing innovation costs in his social-planning framework, he forces the market to underinvest in innovation relative to the social optimum (if the price really were the opportunity cost, that is). The tie to the complexity of real-world innovation dynamics is tenuous at best, but the result ties well to the empirical observation that markets underinvest in innovation. Although the model of Goulder and Schneider is more rigorous than that of Nordhaus, it can also be viewed as using an approximate, heuristic tool. All IMFs in Goulder and Schneider's model boil down to “spillover” parameters, one for each industry. The higher the value of these parameters, the lower the investment in innovation, and vice versa. In the end, the result is identical to that of Nordhaus: individual industries underinvest in innovation and a single parameter controls the level of underinvestment. Whether researchers prefer one approach to the other depends largely on which better fits their intuition and their model structure.

It is tempting to try to bring the rigor of innovation theory into climate models, perhaps including more complex market and appropriability structures. We suggest, however, that attempts to model the complexity of IMFs in climate-change models may add confusion rather than clarity. In the end, we care most that markets do underinvest in innovation, not why they do so. We are not suggesting that the models of innovation theory and other explorations into the microstructure of innovation are not important concerns—they most certainly are, in a wide range of policy contexts, including climate change. We are saying that in long-time-frame, high-level analysis such as integrated assessment modeling, the goal is to capture the observations of these explorations.

12.7.3   Characteristics of technological advance

Climate-change modelers wishing to include ITC must be clear about two characteristics of technological advance: the mechanism by which technology advances and the manner in which advance alters technology.

R&D versus Experience

Technology advances through innumerable interactions between producers, users, designers, researchers, and so forth. For the purposes of modeling, however, a simple representation must be chosen. Two approaches have been used to date: R&D and experience.

In the R&D approach, technological advance occurs through the costly allocation of resources specifically to the task of innovation—that is, through R&D. If society allocates all its resources to production and nothing to R&D, then technology will stand still. The Goulder and Schneider and the Nordhaus models, and one of the models in Goulder and Mathai use the R&D approach. A difficulty with this approach is that the real-world effects of R&D are very difficult to discern with any precision, so modelers find themselves in a tight spot when they estimate parameters. [Recent econometric work by Popp (2001a, 2001b) attempts to make parameters less speculative using patent data as a link between R&D expenditures and industrial energy consumption.]

In the experience approach, technological advance is a happy consequence of the production and use of technologies (see Arrow 1962b). Technological advance is “free” in the sense that nobody needs to do R&D. For example, the more photovoltaic cells we produce, the less costly photovoltaic cells will become. An advantage of the experience approach is simplicity. Whereas decision makers in the R&D approach must decide on both production/abatement levels and on R&D expenditures, decision makers in the experience approach need only consider production/abatement levels. The experience approach is used in the energy systems models in Seebregts et al. (1999), Grübler and Gritsevskyi (Chapter 11 in this volume), and Gritsevskyi and Nakicenovic (2000; see also Chapter 10 in this volume), and in one of the models in Goulder and Mathai (Chapter 9 in this volume). (Chapter 7 in this volume provides a more thorough discussion of experience curves in energy models.)

Both the R&D and experience approaches tie well to real-world phenomena. On the one hand, governments and private firms spend billions of dollars on R&D, and it seems likely that this money is not entirely wasted. On the other hand, reductions in production costs do occur through “accidental” learning on the shop floor from routine, day-to-day operations: workers do simply learn how to do things better and faster. In fact, Kline and Rosenberg (1986) discuss industry studies that indicate that, in some cases, learning-by-doing improvements to processes contribute more to technological progress than the initial process development itself.

But neither approach is a complete picture of reality, so models based exclusively on one or the other are bound to miss something important. The experience approach, in particular, exudes a false veneer of precision because experience curves are so easy to estimate; all that is needed is a production history and a cost history. But experience curves miss an extensive history of public and private research expenditures and therefore tend to underestimate the costs of technological advance. Further, there are questions about the direction of causality. The experience literature has viewed the correlation between declining costs and increased production as evidence that the latter causes the former, but causality goes the other way, too. Decreasing costs from R&D spur technologies into new markets, thereby increasing cumulative production. Further, cost reductions may exhibit a strong correlation with the passage of time, and this applies to R&D as well.

There are similar problems on the R&D side. For one, the R&D approach misses experience effects. Moreover, investment in research should not be interpreted strictly as emanating from an R&D lab. Kline and Rosenberg (1986) argue that the notion that innovation is initiated by research of the scientific sort is wrong most of the time. Innovations evolve through cycles of design, testing, production, and marketing, all of which may draw on state-of-the-art knowledge and interact with research initiatives.

The point here is not that R&D is better than experience or vice versa: there is no “right” way to capture the determinants of advance. The point is that both approaches, R&D and experience, are approximations to aid in analysis, both are useful, and both miss important phenomena. Modelers can use the R&D approach to great effect knowing full well that experience is also important, and the same can be said about the experience approach. It is essential that modelers be clear on the implications of what is included in their models and what is not.

The Change in Technology

Kline and Rosenberg (1986) emphasize that there is no simple, single measure or dimensionality to innovation. We might think of innovations as new products or processes, substitution of inputs, or reorganization of production and distribution arrangements. How should technology be represented in ITC models and how should it change over time? This is as fundamental a question as any in ITC modeling.

At one extreme, the bottom-up models contain information about the manner in which individual technologies advance. A benefit of this approach is that the models provide information about which technologies might be important over the long haul, and therefore might be good candidates for government support. Further, modelers often receive information in terms of individual technologies; for example, the prospects for advance in clean coal. Bottom-up models are well placed to use this information. In addition, bottom-up models are explicit about heterogeneity (see Section 12.8). On the negative side, the detail of bottom-up models often comes at a severe computational price and may cloud intuition with complexity. Further, the technological detail may give the illusion that we know more about individual technologies than we really do, and therefore, that we can effectively “pick winners.”

At the other extreme, neoclassical growth models, such as Nordhaus’ R&DICE model, and stylized cost-function models, such as that of Goulder and Mathai, capture environmental technological advance through changes in a single parameter of the production function or the cost function. This approach greatly simplifies analysis and macroeconomic intuition at the expense of a keener understanding of which technologies emerge to cause the reductions.

Goulder and Schneider cut a middle ground between the two approaches by separately modeling distinct industries representing distinct sorts of technologies: an alternative, or emissions-free, energy industry and a conventional energy industry. Innovation in either industry reduces the costs of producing the particular sort of energy. The notion of heterogeneous technological advance, and particularly the notion of singling out innovation in backstop technologies, seems eminently worthwhile.

Regardless of the model type and innovation mechanism, it seems wise to start with ITC in the energy industry, leaving other technological change as exogenous. While it is true that intersectoral spillovers are real and important, models trying to include the complex interrelations between energy technologies and other technologies would probably be too cumbersome or abstract to be useful. It may also be worthwhile to consider two sources for energy- or carbon-saving improvements: decarbonization of energy services and reductions in the energy intensity of economic activities. The second source of technological advance is more troublesome to incorporate into ITC models, since it involves R&D efforts in sectors outside the energy industry—for example, energy-reducing improvements in transportation technology. Because endogenizing technological change across all industries might be too ambitious in many models, ITC modelers may consider improvements in the energy intensity of technologies outside the energy sector as exogenous.

12.8   Extensions and Limitations of ITC Modeling

Climate-change modelers have not attempted to incorporate the entire complex system of technological change into their models. Instead, they have narrowed their focus to a particular set of observations and phenomena, as every modeler must do to keep analysis manageable. ITC models in climate change have thus far focused on (1) deterministic innovation (2) in the energy or carbon sector (3) that results from private sector actions and (4) that is too slow as a result of IMFs. Many of the state-of-the-art models discussed in Section 12.6 deviate from this simple model in one way or another, but it still describes the main thrust of ITC modeling efforts to date and is a useful starting point for discussing extensions to and limitations of ITC modeling.

Figure 12.4. A More Complete Framework for Interpreting and Analyzing Technological Change.

One of the main conclusions to come out of the discussion here is that this simple conception is only one piece of a dynamic, uncertain, heterogeneous, context-sensitive, and path-dependent system, not all of which is amenable to modeling. This section, drawing on “appreciative theory” beyond conventional equilibrium economics, seeks to understand which extensions of the simple model are most helpful in improving upon exogenous models of technological change. Here, we discuss six sorts of extensions: complementary sources of technological change, heterogeneity, uncertainty, other IMFs, path dependence, inertia, lock-in, and diffusion. Figure 12.4 proposes a more complete framework for thinking about ITC. Though all these factors are critical in the formulation, analysis, and interpretation process, not all can be included explicitly in models.

12.8.1   Complementary sources of technological change

ITC in climate-change models is not an all-or-nothing proposition. Part of the modeler's task is to decide what should be endogenous and what should remain exogenous. The simple model described above focuses on private, energy sector innovation. Sources of change that are complementary to, or outside of, the simple ITC model may have complex complementary or feedback relationships with energy sector innovation. Some may be induced—that is, they may respond to policy—and are therefore sources and opportunities for technological change in their own right. Here we will highlight three complementary sources of technological advance.

The first complementary source is public sector R&D. ITC models spring from a set of innovation literature, innovation theory and endogenous growth theory, that focuses on private sector innovation. It is therefore natural for ITC models to neglect public sector sources of change. But publicly financed basic research as well as subsidies to private R&D have been discussed as central pieces of near-term climate-change policy. [If we think broadly enough, public sector R&D may also be induced. For example, Hayami and Ruttan (1985) show that public sector agriculture R&D was induced by differences and changes in relative factor endowments and prices.] The second complementary source is inter-sectoral spillovers. For example, metallurgical improvements in the past century made possible gradual improvements in electric power generation by allowing a steady rise in operating temperatures and pressures. To models that focus exclusively on the energy sector or restrict spillovers to within industries, such changes will be exogenous. The third complementary source of technological change is groundbreaking major innovations. At first glance, these major innovations may seem to fit well within the simple model: are they not simply bigger versions of the average innovation? To some extent they are, but there are also legitimate questions about the degree to which they respond to market forces. To the extent that they do not, they, too, are exogenous to the simple model, and they are highly uncertain.

ITC Modeling Implications. A first note is that modelers should not exclude technology policies—for example, government R&D and R&D subsidies—from the set of options they consider in their models. A more fundamental point in this section, and one that is fundamental to this chapter as a whole, is that modelers should not attempt to make everything endogenous, but rather should pick and choose what is most appropriate given their objectives and their model structures. Complementary sources of technological change tell us that some technological advance must ultimately remain exogenous in ITC models. What is exogenous and what is induced is a matter of model construction: it is a matter of how far we cast our net.

The empirical work of Newell (1997) and Newell et al. (1998; see also Chapter 5 in this volume) supports this notion. The authors used data from 1958 through 1993 to estimate the cost reductions and changes in energy-efficiency characteristics of domestic air conditioning and water heating equipment. They found evidence that the direction of technological advance responded to policy, and that about one-quarter to one-half of the improvement in mean energy efficiency since 1973 was associated with rising energy prices. At the same time, though, the authors found evidence that a large component of the cumulative energy-efficiency improvements occurring over the three decades consisted of “proportional,” or neutral, improvements in technology that were largely autonomous, or unresponsive to the set of policies they considered.

12.8.2   Heterogeneity

To ease analysis, economic modelers often simplify the diverse set of firms and technologies into representative firms and aggregate production functions. Here we discuss the implications of these simplifications for ITC modeling.

Heterogeneity in Firm Behavior

Evolutionary economics, drawing inspiration from the process of “creative destruction” outlined by Joseph Schumpeter over 50 years ago, highlights the importance of differences in firm behavior (see Nelson and Winter 1982; Nelson 1995). The issue is not how capitalism administers existing structures, but how it creates and destroys them in the dynamic process of growth and change. The neoclassical focus on the general equilibrium allocation of wealth takes a distinctly secondary role in the Schumpeterian world.

In evolutionary economic models, firms are heterogeneous, most notably in size and in organizational or behavioral structure. Firms are carriers of technologies (or routines or customs) that determine firm performance as a function of their environment. Routines represent at any time the best that firms know and can do in terms of standard operating procedures, investment behavior, innovation behavior, and so forth. An important implication of this focus on routines is that optimization is myopic and local. Routines are analogous to genes; the role of the market is to select from among the various routines. Consequently, the rise and fall of firms and nations, and changes in technology, are explained by the survival and proliferation of routines. Conditioned by a changing selection environment, firms evolve in a process that is partly stochastic, but not wholly random. As a consequence, the process of technological advance is strongly path dependent, with no unique, optimized equilibrium.

The point here is not that ITC modeling should be based on evolutionary economic models, but rather that a whole class of economic models, with a tradition dating back to before Schumpeter, has as its fundamental premise the differences between firms. Firm heterogeneity can also be gathered from a more traditional business strategy perspective. The notion of routines is analogous to the conscious differentiation of firms seeking competitive advantage over rivals.

The potential importance of firm heterogeneity implies that ITC models based on homogeneous firms may misrepresent industry innovative behavior. Imagine, for example, that carbon taxes change the selection environment. Those firms that are well positioned to capitalize on price changes will invest heavily in alternative energy or emissions-reducing technologies, while other firms will invest very little. The sum total of industry investment may easily be greater than if we assumed that numerous identical, “average” firms only invested a small amount: one firm in 10 that is well placed to respond may spend more than would 10 “average” firms in total.

ITC Modeling Implications. The implications of firm heterogeneity for ITC models are unclear. It is tempting to assert that the use of representative firms will underestimate industry responses to policy changes, as in the example just discussed. But the validity of this assertion depends crucially on what the representative firm represents. If it represents something resembling an “average,” then the assertion may be true. If, on the other hand, it is a representation of aggregate industry response, then there is less reason to expect bias one way or the other.

Technological Heterogeneity: Discontinuity in Technological Advance

Just as the representative firm hides heterogeneity in firm behavior, aggregate production functions hide technological heterogeneity. One potential implication of technological heterogeneity is discontinuous technological advance. Technological advance in parameterized production functions takes on only as many dimensions as there are parameters. For example, if a single knowledge parameter captures technological advance, then there is a single dimension for advance. In reality, though, advance has at least as many dimensions as there are individual technologies. As individual technologies advance, the aggregate production function changes.

Even if innovation is continuous and incremental in individual technologies, the aggregate production function's response to innovation investment may be nonlinear and may exhibit significant discontinuities. Discontinuities arise when previously uncompetitive technologies reach important market thresholds through cumulative incremental improvements and rapidly diffuse into the market. Prior to reaching a threshold, innovation investment in emerging technologies has a limited impact on aggregate production; after reaching a threshold, innovation expenditures have a more direct and significant impact. In economic terms, it is as if technologies sit on the interior of the production function and R&D efforts to bring them out toward the production function have little discernible impact at an aggregate level. To give a hypothetical example, R&D investment in photovoltaic cells over the past half century has had a relatively small impact on aggregate energy efficiency, but photovoltaic cells may continue to improve and someday compete with other electricity resources. If they do, R&D occurring at a point of rapid market acceptance will appear dramatically more efficient in reducing aggregate efficiency characteristics than early photovoltaic cell R&D. The relationship between aggregate efficiency and R&D, and the energy-efficiency characteristics in the energy sector as a whole, will undergo a discontinuous shift.

Chakravorty et al. (1997) emphasize the importance of heterogeneous technologies. Using a framework of optimal natural resource extraction, they study endogenous substitution between the energy resources of coal, oil, gas, and solar power, and the implications for climate policy. They find that carbon emissions will experience a sharp drop around 2050 as the costs of solar generation become more competitive. Although the results depend crucially on assumptions about solar development, they make clear that cost reductions in the backstop technology relative to those of fossil fuels may be at least as important as technological change in the energy sector as a whole.

The notion of heterogeneous technologies raises another notion: that of asymmetric IMFs. Improvements in sulfur dioxide scrubber technology may be less risky and may pay off sooner than investments in photovoltaic cells. To the extent that IMFs are based on technology characteristics such as risk and development time, and that technologies differ in these characteristics, the impacts of IMFs will be asymmetric: some technologies will be more susceptible than others. In particular, one would expect that emerging technologies would be disproportionately susceptible to these IMFs and, therefore, that advancement in emerging technologies might respond less vigorously to emissions taxes than would advancement in more mature and widespread technologies. We will have more to say on these topics later in this section.

ITC Modeling Implications. What do ITC models miss when they aggregate technology? First, these models will ascribe too much predictability and consistency to the innovative process by glossing over the continual process of emergence and obsolescence of technologies. Second, they will ascribe too great a private sector response to development of emerging environmental technologies, because these technologies are typically more susceptible to IMFs such as high private risk aversion and high private discounting. Third, and this seems the crucial point, aggregate models miss the central importance of emerging technologies and the associated notion that the allocation of innovative effort is important, not just the absolute level.

It seems feasible for ITC models to consider heterogeneous technologies, and some already do. Bottom-up models can explicitly consider hundreds of technologies and therefore give detailed representations of changes in aggregate production characteristics, including discontinuities and nonlinearities. Top-down models, such as the general equilibrium model of Goulder and Schneider, can distinguish between carbon-intensive and non-carbon-intensive industries. The level of aggregation is higher than in the bottom-up approach, but this is a great leap forward from models with a single, parameterized production function. As Chakravorty et al. demonstrate, technological advance in backstop (alternative energy) technologies may be more important than changes in conventional fossil fuel technologies. Hence, we believe that a good first step for modelers is to make a distinction between fossil-fuel-based and alternative energy industries or technologies.

12.8.3   Technological uncertainty

Technological change is an uncertain process. Uncertainty arises not only because the consequences of individual technological changes are so difficult to predict (see Rosenberg 1986), but also because we do not know what changes the future holds. One way uncertainty enters ITC models is by holding back private sector innovation: if private actors are excessively risk averse, they may bias their activities toward less risky projects. We will leave this issue for later, when we discuss “other IMFs.” Here we concentrate on uncertainty in the rate and direction of technological advance: we are unsure how technology will evolve in the future, and the further into the future we go, the less we know.

The fundamental concern for modeling the induced portion of technological change is, How will the production function respond to innovation investment (or cumulative experience)? There is enormous uncertainty about this response and it is not just a matter of “the right number”; the response will change over time as technologies come and go in the market. We suggest here three aspects of uncertainty in technological change that ITC models need to consider: uncertainty in the potential for individual technologies, heterogeneity and discontinuity in technology development, and major innovations.

The first uncertainty refers to our inability to predict how individual technologies will respond to R&D and to experience. The second uncertainty is at a higher level. When we use aggregate production functions, we miss the dynamic competition between technologies and the discontinuities when one technology displaces another. Both of these are “parameter” uncertainties, where the parameters refer to the response of technology to innovative effort or R&D. Both are important in ITC modeling.

The third uncertainty is different in nature. It concerns major, groundbreaking, Schumpeterian-style innovations. Mokyr (1990) draws a useful distinction between microinventions and macroinventions. Microinventions are “small, incremental steps that improve, adapt, and streamline existing techniques already in use, reducing costs, improving form and function, increasing durability, and reducing energy and raw material requirements.” Microinventions are consistent with the models of innovation theory and new growth theory because predictability is relatively high. On the other hand, macroinventions are those inventions in which radical new ideas emerge without precedent. These macroinventions do not seem to obey obvious laws, are not necessarily preceded by profit incentives, and “defy most attempts to relate them to exogenous economic variables.” In this sense, macroinventions may be exogenous in ITC models. Also, according to Mokyr, “the essential feature of technological progress is that the macroinventions and microinventions are not substitutes but complements.”

In the past, conventional energy-policy-oriented models have focused on time frames of up to 50 years, depending on the scope of analysis. In these time frames, it was perhaps justified to consider only continuous, incremental improvements in technology, such as those implicitly addressed in innovation theory and endogenous growth theory. In climate-change models, though, the scope is often extended to 2100 or beyond. Extrapolating the focus on microinventions into long-term models of technological change may introduce significant error. Witness, for example, the unprecedented emergence of nuclear power during the mid-twentieth century after, as late as the 1930s, leading scientists claimed power could never be harnessed from the atom. While microinventions and the “D” of R&D account for the majority of technological activity, long-term models of technological change are incomplete without consideration of macroinventions.

The distinction between microinventions and macroinventions concerns not only the magnitude of the technological response to innovative activity, but also the degree of causality. ITC models typically assume that technological advance will respond to the economic climate. For example, technological advance should respond to emissions taxes. This assumption is perhaps valid for microinventions, and we might expect an induced change in macroinventions as well, but historical observation indicates that major innovations are not necessarily preceded by a vast commitment of resources directed exclusively toward their development. They often arise as an unexpected byproduct of other innovative endeavors or from basic research. In a sense, they occur more or less randomly, though their occurrence gives rise to a subsequent large-scale commitment of scientific and technological resources to complementary microinventions. A classic example is the invention of the transistor (see Rosenberg 1994). Before the advent of the transistor in 1948, solid-state physics was an obscure subdiscipline. After this macroinvention, R&D communities in both universities and the private sector made large-scale commitments to exploit this new path of innovation.

ITC Modeling Implications. What uncertainties should ITC models include, and how should models include them? A first answer is that, again, some technological uncertainty pertains to technological advance that is largely exogenous in the context of climate models. Macroinventions may fall mainly into this category. On the other hand, some uncertainty pertains to the induced portion of technological advance. Most notable in this regard is uncertainty about innovation production function parameters and experience curve parameters that arises from uncertainty about individual technologies and from the use of aggregate production functions.

One approach to including this parameter uncertainty would be to base R&D production functions on expected values of uncertainty distributions. But this could be a mistake; point estimates can lead to erroneous conclusions in nonlinear systems. In the climate-change system, damage is a nonlinear function of climate change. Nordhaus (1994) showed the importance of accounting for uncertainty in climate-change models. Rather than using the expected values of the uncertain parameters, Nordhaus considered probability distributions on the major uncertain parameters in his DICE model. Using Monte Carlo simulation, he found that the optimal carbon tax more than doubles when uncertainty is taken into account, and the optimal control rate increases by slightly less than half. Similar phenomena may exist with respect to technological advance. Nordhaus’ approach could serve as a model for dealing with technological uncertainty in ITC models. Technological uncertainty is especially difficult, though, because it may only be resolved if we attempt to advance technologies: resolution may not be simply a function of time, as is often assumed in analytical models of uncertainty.

Regardless, the importance and prevalence of technological uncertainty indicate that energy policy decisions should be made in an incremental manner, making use of the gradual reduction of uncertainty and preserving options. Projecting technological characteristics far into the future is a daunting task, but this does not mean that we cannot make decisions in the face of uncertainty, for example, by using the principles of decision analysis.

12.8.4   Other IMFs

Innovation theory and new growth theory are based on the dual notions of spillovers and appropriability. Empirical and historical evidence tells us that these are, indeed, important aspects of technological change. There are, however, other IMFs that should be and are considered by policy makers and, therefore, should be addressed by policy modelers.

Risk Aversion and Discounting

Innovation takes time and is risky. To the extent that markets behave with different preferences for risk and time than we as a society would like, markets will invest in innovation differently than we as a society would like. These differences are not new concepts, and they are not limited to technological change. However, because of the time frames and risk associated with technological change, they may play a disproportionately important role. At the simplest level, in models with only a single dimension for advance—that is, where technology is represented by a single parameter, such as an aggregate knowledge parameter—these IMFs should simply hold back the rate of innovation.

Risk aversion and discounting begin to play a more important role when we consider technological heterogeneity, and emerging environmental technologies in particular. Some technologies will take longer to become competitive than others and some have greater risk. Differences between private and social preference for time and risk therefore affect not only the rate of technological change, but also its direction: markets choose inefficient levels of innovation effort and an inefficient allocation among potential technologies. Government energy research funding is often targeted along these lines, attempting to identify high-risk, long-time-frame technologies.

Because price-based policies such as emissions taxes cannot differentiate between technologies, they are unable to fully alter the allocation of private investment. This is one justification for technology instruments as part of the climate-change policy portfolio. At the same time, though, it is reasonable to ask whether these asymmetries should really be under the purview of climate policy. Why should governments target only environmentally related technologies that are high risk and far from competitive and not emerging nonenvironmental technologies?

ITC Modeling Implications. How can models address these deviations of private risk aversion and time preference from the socially preferred values? One way is simply to include them implicitly in approximate, heuristic models of IMFs. For example, we might increase the price of R&D resources in Nordhaus’ approach or we might adjust the spillover parameter upward in the Goulder and Schneider approach.

These two IMFs are challenging, though, because they may distinguish between heterogeneous technologies. If models differentiate between emerging and mature technologies, there may be ways to incorporate the effects. For example, modelers might ramp up the IMFs for emerging technologies relative to those for mature technologies, perhaps correlating IMFs to market share or some other measure of market position.

The Limits of “Rational” Modeling of Innovation

Not all investment activity can be captured by models of rational behavior. Some assert that a spirit that defies rational behavior often guides the entrepreneur, accepting a high probability of failure for a low-probability shot at success. Schumpeter went so far as to argue that the innovation process cannot be characterized by rational behavior:

[T]he assumption that business behavior is rational and prompt, and also that in principle it is the same with all firms, works tolerably well only within the precincts of tried experience and familiar motive. It breaks down as soon as we leave those precincts and allow the business community under study to be faced by—not simply new situations, which occur as soon as external factors unexpectedly intrude, but by—new possibilities of business action which are as yet untried and about which the most complete command of routine teaches nothing. (Schumpeter 1939:98–99)

Rosenberg, drawing on his experience as an economic historian, makes an equally strong statement:

The nature of the innovation process, the drastic departure from existing routines, is inherently one that cannot be reduced to mere calculation, although subsequent imitation of the innovation, once accomplished, can so be reduced. Innovation is the creation of knowledge that cannot, and therefore should not, be “anticipated” by the theorist in a purely formal manner, as is done in the theory of decision-making under uncertainty. (Rosenberg 1994:53–54)

While an important realization for the limits of ITC modeling, these observations are also discouraging because they call into question the predictive power of traditional economic models.

ITC Modeling Implications. While important, it seems a bit ambitious to incorporate notions like “entrepreneurial spirit” into ITC models. A more appropriate approach is for modelers to be clear that their models do not include such behavior and therefore greater uncertainty surrounds their results than otherwise would. It also seems, at this point at least, too ambitious to include models of routine-based behavior, such as those in evolutionary economics, in climate models. Routine-based behavior does, however, have one clear implication for ITC that might be amenable to models. Because firms generally search out routines that are similar to those that they already use, private markets will tend to innovate on technologies already in use. Hence, the market may have a sticky innovation response with respect to technological direction. The effect is similar to that of private risk aversion and time preference: it biases private sector innovative behavior toward dominant technologies.

12.8.5   Path dependence, “lock-in,” and inertia

Most economic historians and devotees of evolutionary economics will argue that there is much more involved in the evolution and diffusion of technology than merely prices, production functions, and knowledge stocks. In reality, technological change is highly conditioned on the past paths of major and discontinuous innovations, the development activities of firms, and existing capital stocks.

Technically, by path dependent we mean that a process is nonergodic—the sequence of historical events conditions future possibilities. Rosenberg explains the notion of path dependence applied to technology:

[T]he main features of the stock of technological knowledge available at any given time can only be understood by a systematic examination of the earlier history out of which it emerged. There is . . . a strong degree of path dependence, in the sense that one cannot demonstrate the direction or path in the growth of technological knowledge merely by reference to initial conditions. (Rosenberg 1994:10)

The notion of path dependence is important for making policy, on the one hand, because it suggests that technological change evolves with a great deal of inertia. On the other hand, path-dependent activity may propagate small changes in the system, so that small policy-induced changes today can result in substantial changes in the future.

There are a number of reasons for path dependence and inertia. At a first level, technological capital and R&D organizations are costly to redirect or replace. Capital stock turnover is another source of inertia. Even if less costly or more efficient technologies are available, old technologies may still be competitive because of sunk costs. Grubb (1996) estimates that for various components of the power generation and energy use infrastructure, capital stock turnover cycles range from 20 to 100 years. Grubb et al. (1995) and Ha-Duong et al. (1996) have emphasized the importance of this aspect of inertia in energy models. They argue that inertia tends to increase the optimal near-term abatement. Moreover, development activities tend to focus on existing capital, biasing development toward older technologies.

At a deeper level, inertia is generated by the costly nature and limited mobility of development activities. Rosenberg argues that there is an often underappreciated distinction between the availability and implementation of publicly known knowledge or information:

Development activities accounted for approximately 67 percent of total R&D spending (in the US, according to 1991 Science and Engineering Indicators). These figures, at the very least, suggest great skepticism about the view that the state of scientific knowledge at any time illuminates a wide range of alternative techniques from which the firm may make cost-less, off-the-shelf selections. It thereby also encourages skepticism toward the notion that is so deeply embedded in the neoclassical theory of the firm, that one can draw a sharp and well-delineated distinction between technological change and factor substitution. (Rosenberg 1994:13)

Indeed, technological change depends greatly on how firms have already geared up, or in the language of economics, the point on the production function at which they currently operate. Firms cannot instantaneously shift to alternative technologies, even if the shift involves only exploiting available but unfamiliar knowledge. This is similar to the notion of routine-based firms making localized searches for new knowledge in evolutionary economic models.

Arthur (1989) has demonstrated how the stochastic nature of the innovation process may lead to technology “lock-in,” even by inferior technologies. Development activities reinforce the lock-in, as they focus on existing technologies. The existing technologies suggest certain directions where research efforts can be usefully exercised, resulting in a series of minor improvements that may amount to significant change over the long term.

A prominent example of a path-breaking major innovation that shaped future development is the internal combustion engine. Its rapid development in the early twentieth century made possible numerous other innovations in the automotive and aircraft industries. Moreover, soon after its introduction, the internal combustion engine dominated research and engineering efforts in propulsion devices, even though the engine may not have been inherently technologically superior to the competing technologies of electric and steam power for its initial use in cars. Another example is the steam turbine's domination of electricity generation and the resulting focus of R&D on incremental improvements in that technology.

ITC Modeling Implications. An important implication of path dependence and associated phenomena is that the rate, and especially the direction, of innovation may respond more sluggishly to the economic climate than the neoclassical model of the firm would predict. But there is a deeper and more problematic concern associated with path dependence: actions and technological choices today are more important than conventional economic models would indicate because today's actions not only advance current technological characteristics, they also redirect the future path of technological change. What we do today affects how the economy will respond in the future.

Incorporating path sensitivity into ITC models is a challenge that is perhaps not worth the added complication. Realizing the distinction between changes in technology characteristics and technology paths, however, is important in interpreting model results and analyzing policy. Still, ITC models can be extended to better complement qualitative insights about path dependence. Models with experience effects already include inertia, because experience effects reinforce the lead associated with dominant technologies, making it harder for new technologies to emerge. Models with heterogeneous technologies might include asymmetric IMFs to produce a similar effect, and models may be more specific about capital stocks. Models might also include time lags to account for technological expectations, diffusion of innovations (discussed below), and the costliness of development activities. Lags in technology development would be particularly relevant to sudden changes in policy or carbon taxes, to developments following upon major innovations, or in the exploitation of spillovers.

12.8.6   Diffusion

In the simple model of ITC, firms invest in R&D to improve the production function. But what does the production function represent? If it represents the capital stock, and not the technologies society could use, then it includes the choice of technologies: it includes diffusion. The choice of ostensibly available technology may be as important as the frontier. The essential point here is that innovation does not in and of itself lead to the use of technology; the use of technology will lag behind innovation. [For a review of diffusion literature and models, see Karshenas and Stoneman (1995).]

But diffusion is not simply a matter of time lags. Rates of diffusion and the factors underlying diffusion vary by country and region, and policies should be tailored geographically to reflect this variation. The importance of diffusion is particularly dramatic in developing countries, since their technology lags well behind that of the developed countries. The gap might be closed, but the convergence process depends on numerous social factors (Fagerberg 1994).

A number of studies have looked at the spread of energy-efficiency technologies under the rubric, “the energy-efficiency paradox”—the slow diffusion of apparently cost-effective energy-efficiency technologies or, put another way, the apparently high discount rates used in evaluating energy-efficiency investments (see, e.g., Jaffe and Stavins 1994a, 1994b, 1995; and Energy Policy, issue No. 10, 1994). Studies show that market failures are not entirely to blame for the paradox—much of it is due to rational behavior in the face of uncertainty and advancing technology—but almost all studies point to some market failures. Notable among these are information market failures—equipment users may be unaware of the efficiency characteristics, the importance of the efficiency characteristics, or the durability of more efficient equipment. Appliance efficiency labeling and automobile mileage labeling are good examples of government policies aimed directly at information market failures.

Further, one consequence of “quasi-rational” behavior is the impediment of the diffusion of new technologies. Rosenberg (1982) has suggested several reasons why expectations about technology act to slow diffusion. Such expectations generally fall outside the scope of most economic models of ITC, in which firms decide to invest in knowledge without consideration of technology diffusion. First, expectations about the continued improvement and refinement of a technology, particularly the arrival or development of a major innovation, may lead to postponement of innovation activities or adoption. Firms are reluctant to invest in a fledgling technology when they expect substantial improvements to be forthcoming. As anyone who has bought a PC can attest, no one wants to feel burned by investing in a technology that is immediately rendered obsolete by subsequent improvements. Second, competition from new technology sometimes spurs development in old technology, making it more competitive and thus slowing diffusion of the new entrant. Similarly, because single breakthroughs seldom constitute complete innovations, decisions to adopt an innovation are often postponed “in situations that might otherwise appear to constitute irrationality, excessive caution, or over-attachment to traditional practices in the eyes of uninformed observers” (Rosenberg 1982).

ITC Modeling Implications. A first lesson for modelers is to be clear about just what the production function represents. If it represents the capital stock, then it implicitly includes diffusion. A second lesson is that technology diffusion can be induced in addition to innovation, but that diffusion of energy-efficiency improvements may respond more slowly to carbon taxes or other market-based climate policies than might be expected. Modelers may capture this effect through time lags. Further, other policies, most notably information diffusion policies, may play an important and effective role in climate policy.

12.9   Conclusions from a Broader Perspective

We can be confident about two aspects of ITC. First, it has the potential to be exceedingly important. A world with low-cost photovoltaic cells, solar hydrogen, fuel cell cars, and so forth will be very different from a world based largely on fossil fuels, as is today's world. ITC is important to the extent that today's policies influence which future world will emerge. Second, the real-world mechanisms underlying ITC are enormously complex. Together, these two aspects of ITC call for its consideration in climate policy, and environmental policy more generally, and present a modeling challenge.

In Section 12.6, we saw that current state-of-the-art climate-change models do not immediately suggest dramatic impacts from including ITC. In considering possible extensions to and limitations of the current models, this result needs to be qualified. A number of considerations, summarized in Table 12.2, suggest that current ITC models may miss important, policy-relevant nuances of technological advance. Current models may overestimate the speed of technological responses to policy; they may overestimate the flexibility of technological systems to change direction; they may miss important biases for and against particular technology paths (e.g., biases against emerging technologies); and they may miss the value of policies that directly stimulate innovation and diffusion. How might modelers respond to the challenges presented by the considerations in Table 12.2? What is important, what is best left to model interpretation, and what should be included in ITC models?

The most crucial element of ITC is that technological change responds to policy. All of the ITC models we reviewed in Section 12.6 have this fundamental element. It is in movements beyond that point that modelers must make more difficult choices. And no model can include everything. In considering what elements to include, we suggest that four extensions would be the most productive in the near term.

The first extension is the notion that private markets invest inefficiently in innovation. We suggest that it is inappropriate for large-scale models, such as climate models, to delve deeply into the microeconomic foundations of this inefficiency. Rather, we suggest that modelers use approximate, heuristic representations of IMFs that capture the fact that markets do invest inefficiently rather than worrying about why. The microeconomic foundations are important, and continued research in innovation theory may provide additional insights into the aggregate behavior of markets, but that is not the charge of the climate modeler. At least two modeling efforts to date—Nordhaus, and Goulder and Schneider—have explicitly included IMFs.

Table 12.2. Impact of Model Extensions and Limitations on the Current Generation of ITC Models.

Model extensions and limitations

Implications for ITC models

Complementary sources of technological advance Some technological advance should remain exogenous in energy and environmental models. Government R&D is a source of technological advance, and therefore a policy option.
Firm heterogeneity Firm heterogeneity implies greater complexity and uncertainty in market innovative behavior than representative firm models indicate.
Technological heterogeneity Aggregate models may miss important discontinuities and nonlinearities in technological advance. Aggregate models underestimate importance of emerging technologies.
Uncertainty in rate and direction of technological advance Models should explicitly consider technological uncertainty because climate systems are nonlinear.
“Other” IMFs: risk aversion and discounting Impacts of these “other” IMFs may be asymmetric. Models without asymmetric IMFs will overestimate technological advance in emerging technologies.
Less-than-rational innovative behavior Quasi-rational, routinized behavior implies less flexibility and responsiveness than traditional economic models indicate.
“Entrepreneurial spirit” decreases the link between incentives and technological advance, causing more technological change to be exogenous in models.
Path dependence, inertia, lock-in Path-dependent phenomena imply less flexibility and responsiveness in technological systems than traditional economic models indicate.
Path-dependent phenomena point to the value of policies to maintain technological diversity.
Diffusion Models without diffusion may overestimate the speed of ITC. Models without diffusion overlook the importance of policies to spur diffusion.

Note: IMF = innovation market failure; ITC = induced technological change; R&D = research and development.

The second extension is technological heterogeneity. If technological advance is to play an important role in the global society's response to climate-change concerns, it will most likely be through the development of emerging, low-emissions technologies that are currently unproven or uncompetitive with dominant, more-polluting technologies. We believe that it would be productive to explicitly capture this dynamic and its policy implications. To date, energy systems models [e.g., those models reviewed in Seebregts et al. (1999); Grübler and Gritsevskyi (1998; see also Chapter 11 in this volume); and Gritsevskyi and Nakicenovic (2000; see also Chapter 10 in this volume)] all explicitly consider technological heterogeneity. Top-down models generally do not. A notable exception is the Goulder and Schneider model, which makes the important distinction between technological advance in alternative energy and conventional energy. We think future top-down modelers would be wise to refine and expand upon this approach.

The third extension is uncertainty in how far and fast technology will advance, and how costly will it be. (We are not referring to the impact of uncertainty on markets for innovation; this is best captured through models of IMFs.) Uncertainty is a thorny issue. We know it is exceedingly important, whether it is in climate damages, costs, or technological advance, but it has proved difficult to model. The resolution of uncertainty over time—that is, learning—has proved especially difficult to handle using standard economic concepts. Nonetheless, it appears crucial enough that modelers would be well advised to consider it explicitly. One possibility might be to reduce the uncertain space into a manageable number of discrete possibilities and then to apply the tools of decision analysis or stochastic control.

The fourth, and last, extension is technological diffusion, both in time and between regions and countries. Empirical studies have shown that technology in developing countries often lags behind that in developed countries, and implementation in all countries lags behind R&D. Diffusion is important because it alters the responsiveness of technology to emissions or R&D policies, and it highlights the value of policies to speed diffusion.

In closing, we would like to emphasize again the crucial importance of communication and interpretation. It is essential for modelers to communicate to policy makers the scope of their models—what is in and what is out—and how reality might differ from model results. Similarly, policy makers must understand that models are models: they do not capture the full scope of reality. Models are an input to decision making, they are not the answer.


Abramovitz, M., and David, P., 1995, Convergence and deferred catch-up: Productivity leadership and the waning of American exceptionalism, in R. Landau, T. Taylor, and G. Wright, eds, Growth and Development: The Economics of the 21st Century, Stanford University Press, Stanford, CA, USA.

Aghion, P., and Howitt, P., 1992, A model of growth through creative destruction, Econometrica, 60(2):323–351.

Arrow, K., 1962a, Economic welfare and the allocation of resources for invention, in National Bureau of Economic Research, The Rate and Direction of Innovative Activity, Princeton University Press, Princeton, NJ, USA.

Arrow, K., 1962b, The economic implications of learning by doing, Review of Economic Studies, 29:155–173.

Arthur, B., 1989, Competing technologies, increasing returns, and lock-in by historical small events, Economic Journal, 99(March):116–131.

Carraro, C., 1997, Induced Technological Change in Environmental Models: Theoretical Results and Implementations, Working Paper, Department of Economics, University of Venice, prepared for the IIASA meeting on Induced Technological Change and the Environment, held in Laxenburg, Austria.

Chakravorty, U., Roumasset, J., and Tse, K., 1997, Endogenous substitution among energy resources and global warming, Journal of Political Economy, 105(6):1201–1234.

Cohen, W.M., and Levinthal, D.A, 1989, Innovation and learning: The two faces of R&D, The Economic Journal, 99:569–596.

Evenson, R.E., and Kislev, Y., 1975, Agricultural Research and Productivity, Yale University Press, New Haven, CT, USA.

Fagerberg, J., 1994, Technology and international differences in growth rates, Journal of Economic Literature, 32(3):1147–1175.

Goulder, L.H., and Mathai, K., 1998, Optimal CO2 Abatement in the Presence of Induced Technological Change, NBER Working Paper 6494, National Bureau of Economic Research, Cambridge, MA, USA.

Goulder, L.H., and Schneider, S., 1997, Achieving low-cost emissions targets, Nature, 389(4):13–14.

Goulder, L.H., and Schneider, S., 1999, Induced technological change and the attractiveness of CO2 abatement policies, Resource and Energy Economics, 21:211–253.

Griliches, Z., 1979, Issues in assessing the contribution of research and development to productivity growth, Bell Journal of Economics, 10:92–116.

Griliches, Z., 1992, The search for R&D spillovers, Scandinavian Journal of Economics, 94(Supplement):29–47.

Gritsevskyi, A., and Nakicenovic, N., 2000, Modeling uncertainty of induced technological change, Energy Policy, 28:907–921.

Grossman, G., and Helpman, E., 1991a, Quality ladders in the theory of growth, Review of Economic Studies, 58:43–61.

Grossman, G., and Helpman, E., 1991b, Innovation and Growth in the Global Economy, MIT Press, Cambridge, MA, USA.

Grossman, G., and Helpman, E., 1994, Endogenous innovation in the theory of economic growth, Journal of Economic Perspectives, 8(1):23–44.

Grubb, M., 1996, Technologies, energy systems, and the timing of CO2 emissions abatement: An overview of economic issues, in N. Nakicenovic, W. Nordhaus, R. Richels, and F. Toth, eds., Climate Change: Integrating Science, Economics, and Policy, IIASA Workshop Proceedings, International Institute for Applied Systems Analysis, Laxenburg, Austria.

Grubb, M., Chapuis, T., and Ha-Duong, M., 1995, The economics of changing course: Implications of adaptability and inertia for optimal climate policy, Energy Policy, 23(4/5):417–432.

Grübler, A., and Gritsevskyi, A., 1998, A model of endogenous technological change through uncertain returns on learning. http://www.iiasa.ac.at/Research/TNT/WEB/Publications/

Grübler, A., Nakicenovic, N., and Victor, D., 1999, Dynamics of energy technologies and global change, Energy Policy, 27:247–280.

Ha-Duong, M., Grubb, M., and Hourcade, J.C., 1996, Optimal Emission Paths Towards CO2 Stabilization and the Cost of Deferring Abatement: The Influence of Inertia and Uncertainty, Working Paper, CIRED, Montrouge, France.

Hayami, Y., and Ruttan, V., 1985, Agricultural Development: An International Perspective, The Johns Hopkins University Press, Baltimore, MD, USA.

Jaffe, A., and Stavins, R., 1994a, The energy paradox and the diffusion of conservation technology, Resource and Energy Economics, 16:91–122.

Jaffe, A., and Stavins, R., 1994b, Energy-efficiency investments in public policy, The Energy Journal, 15(2):43–65.

Jaffe, A., and Stavins, R., 1995, Dynamic incentives of environmental regulations: The effects of alternative policy instruments on technology diffusion, Journal of Environmental Economics and Management, 29:43–63.

Jaffe, A., Newell, R.G., and Stavins, R., 2000, Technological Change and the Environment, Resources for the Future Discussion Paper 00-47, RFF, Washington, DC, USA.

Jones, C., and Williams, J., 1996, Too Much of a Good Thing: The Economics of Investment in R&D, Working Paper, Department of Economics, Stanford University, Stanford, CA, USA.

Jorgenson, D., 1996,Technology in growth theory, in J. Fuhrer and J.S. Little, eds, Technology and Growth, Conference Proceedings, Federal Reserve Bank of Boston, Boston, MA, USA.

Jorgenson, D., and Wilcoxen, P.J., 1993, Energy, the environment and economic growth, in A. Kneese and J. Sweeney, eds, Handbook of Natural Resources and Energy Economics, North Holland, Amsterdam, Netherlands.

Kamien, M., and Schwartz, N., 1982, Market Structure and Innovation, Cambridge University Press, Cambridge, UK.

Karshenas, M., and Stoneman, P., 1995, Technological diffusion, in P. Stoneman, ed., Handbook of the Economics of Innovation and Technological Change, Blackwell Publishers, Oxford, UK.

Kline, S.J., and Rosenberg, N., 1986, An overview of innovation, in R. Landua and N. Rosenberg, eds, The Positive Sum Strategy: Harnessing Technology for Economic Growth, National Academy Press, Washington, DC, USA.

Levin, A., 1988, Appropriability, R&D spending, and technological performance, American Economic Review, Papers and Proceedings, 78:424–428.

Levin, A., Klevorick, R., Nelson, R., and Winter, S.G., 1987, Appropriating the returns from industrial research and development, Brookings Papers on Economic Activity, 3:783–820.

Mansfield, E., Rappaport, J., Romeo, A., Wagner, S., and Beardsley, G., 1977, Social and private rates of return from industrial innovations, Quarterly Journal of Economics, 77:221–240.

Mokyr, J., 1990, The Lever of Riches: Technological Creativity and Economic Progress, Oxford University Press, New York, NY, USA.

Nadiri, M., 1993, Innovations and Technological Spillovers, NBER Working Paper 4423, National Bureau of Economic Research, Cambridge, MA, USA.

Nelson, R.R., 1995, Recent evolutionary theorizing about economic change, Journal of Economic Literature, 33:48–90.

Nelson, R.R., and Winter, S.G., 1982, An Evolutionary Theory of Economic Change, Harvard University Press, Cambridge, MA, USA.

Newell, R.G., 1997, Environmental Policy and Technological Change: the Effects of Economic Incentives and Direct Regulation on Energy-Saving Regulation, PhD dissertation, Harvard University, Cambridge, MA, USA.

Newell, R.G., Jaffe, A., and Stavins, R., 1998, The Induced Innovation Hypothesis and Energy-Saving Technological Change, Resources for the Future Discussion Paper 98-12, Resources for the Future, Washington, DC, USA.

Nordhaus, W., 1994, Managing the Global Commons: The Economics of Climate Change, MIT Press, Cambridge, MA, USA.

Popp, D., 2001a, Induced Innovation and Energy Prices, NBER Working Paper W8284, National Bureau of Economic Research, Cambridge, MA, USA.

Popp, D., 2001b, The effect of new technology on energy consumption, Resource and Energy Economics, 23:215–239.

Reinganum, J., 1989, The timing of innovation: Research, development, and diffusion, in R. Schmalensee and R. Willig, eds, Handbook of Industrial Organization, Vol. 1, Elsevier Science Publications, Amsterdam, Netherlands.

Romer, P., 1986, Increasing returns and long-run growth, Journal of Political Economy, 94(5):1002–1037.

Romer, P., 1990, Endogenous technological change, Journal of Political Economy, 98:S71–S102.

Romer, P., 1994, The origins of endogenous growth, Journal of Economic Perspectives, 8(1):3–22.

Rosenberg, N., 1976, Perspectives on Technology, Cambridge University Press, Cambridge, UK.

Rosenberg, N., 1982, Inside the Black Box: Technology and Economics, Cambridge University Press, Cambridge, UK.

Rosenberg, N., 1986, The impact of technological innovation: A historical view, in R. Landua and N. Rosenberg, eds, The Positive Sum Strategy: Harnessing Technology for Economic Growth, National Academy Press, Washington, DC, USA.

Rosenberg, N., 1994, Exploring the Black Box: Technology, Economics, and History, Cambridge University Press, Cambridge, UK.

Ruttan, V.W., 2001, Technology, Growth, and Development: An Induced Innovation Perspective, Oxford University Press, New York, NY, USA.

Schmookler, J., 1966, Invention and Economic Growth, Harvard University Press, Cambridge, MA, USA.

Schumpeter, J., 1939, Business Cycles, Vol. I, McGraw-Hill, New York, NY, USA.

Seebregts, A., Kram, T., Schaeffer, G., Stoffer, A., Kypreos, S., Barreto, L., Messner, S., and Schrattenholzer, L., 1999, Endogenous Technological Change in Energy Systems Models: Synthesis of Experience with ERIS, MARKAL, and MESSAGE, Netherlands Research Foundation ECN, Petten, Netherlands.

Tol, R., 1996, The Optimal Timing of Greenhouse Gas Emission Abatement, the Individual Rationality and Intergenerational Equity, Working Paper, Institute for Environmental Studies, Vrije Universiteit, Amsterdam, Netherlands.

Trajtenberg, M., 1990, Economic Analysis of Product Innovations, Harvard University Press, Cambridge, MA, USA.

Weyant, J., 1997, Technological Change and Climate Policy Modeling, Stanford University, Paper prepared for the IIASA meeting on Induced Technological Change and the Environment, International Institute for Applied Systems Analysis, Laxenburg, Austria.

Wigley, T., Richels, R., and Edmonds, J., 1996, Economic and environmental choices in the stabilization of atmospheric CO2 concentrations, Nature, 379(6582):240–243.