«January 16, 2006 Howard Pack Kamal Saggi The Wharton School Department of Economics University of Pennsylvania Southern Methodist University 1400 ...»
sector to generate spillovers for both sectors thereby providing an inter-industry spillover rationale for the infant industry argument. However, the presence of such economies is not sufficient to justify intervention. As Succar notes, the discounted stream of productivity gains generated by LBD in the infant industry should outweigh the discounted stream of subsidies or else intervention is socially undesirable.3 The intuitive idea underlying Succar’s model is that the production of capital goods can enhance growth by acting as an “informal learning center where technical skills are required” thereby contributing to a country’s technical infrastructure.4 Such improvements in the skill base of workers complement investments in human capital and can help further the industrialization process in developing countries.
The distinction between firm and industry level LBD becomes quite important when one considers the fact that firms are heterogeneous in nature. Suppose some firms are more efficient at learning than others. Under such a scenario, optimal subsidies would necessarily have to be non-uniform and the government is unlikely to possess the information needed to implement an optimal subsidy program. Given the information problem, it might make sense for the government to adopt a uniform policy even though it may not be first best. While in theory one can design mechanisms that result in firms revealing their learning capabilities but the practical relevance of such mechanisms is far from clear.
As one might expect, there is more to the infant industry argument than the ‘simple version’ formalized by Bardhan and Succar. As Baldwin notes, notes that there are four versions of the infant industry argument that are a bit more nuanced: (a) acquisition of knowledge involves costs but yet knowledge may not be appropriable by an individual firm - this is the standard argument for subsidizing R&D; (b) firms may provide costly on-the-job training but may be unable to prevent the diffusion of such knowledge via movement of workers (i.e. there might be a free-rider problem in worker training) - while firm specific training involves no potential externality, general training Pack and Saggi (2001) explore the implications of the provision of free technology by the purchasers of a firm’s exports, a further complication.
It is not likely that this criterion has been satisfied by the European Airbus effort, widely considered a major example of a successful industrial policy. Furthermore, one also needs to account for the cost of distortions that are generated by the taxes needed to finance the subsidies paid.
can lead to externalities that would justify subsidies; (c) static positive externalities in the production of a good may justify trade protection and (d) determining profitability of a new industry might require a costly investment the results of which may become freely available to potential competitors – in other words, investment into new industries might result in informational externalities that make it difficult for investors to earn a rate of return that is high enough for the initial investment to be justified. This is precisely the argument that has been formalized by Hausmann and Rodrik (2003) although they call it the process of ‘self-discovery’ – i.e. the process of determining what you can produce profitably at world prices.
The infant industry argument does not really specify how learning occurs – i.e. it just assumes that dynamic scale economies will be somehow realized by the infant industry. Of course, learning is rarely exogenous and it usually requires considerable effort and investment on the part of firms (Pack and Westphal, 1986). If such investments are to be made, firms need to be able to appropriate the benefits of the knowledge gained.
As is well known, knowledge is a non-rival good and, once created, any number of agents can use it simultaneously. If firms cannot prevent the leakage of knowledge that is costly to create, then they will have little incentive to create such knowledge in the first place. In other words, property rights over knowledge may not be enforceable and this can create a rationale for government intervention (this is one reason why we have intellectual property rights protection).
As Baldwin (1969) notes, many types of knowledge acquisition are not subject to the externality described above since entrepreneurs can often prevent the leakage of their knowledge to potential competitors. Similarly, if there are only a few firms in the industry, inter-firm negotiations should help offset the externality problem (see Coase, 1960). But what if many rivals firms benefit from the investment undertaken by a knowledge acquiring firm and nothing can be done to prevent such diffusion? Is government intervention justified then? As is clear, trade protection is certainly not called for – a tariff does nothing to solve the basic externality problem. In fact, trade protection may very well worsen the problem. A production subsidy to the entire sector will also fail Succar’s emphasis on the capital goods sector is similar in spirit to Arrow’s learning by doing model and endogenous growth models such as Romer’s (1986) model that employed it as a building block.
to remedy the externality. Instead, what is needed are subsidies to initial entrants into the industry that help create new knowledge and discover better production technologies. As in the case of R&D subsides, governments should target the marginal rather than inframarginal research. In the case of new firms, it takes time to discover whether a new idea or technology is socially valuable and the adoption of a novel technology by others is in fact the strongest proof of its social value. Thus, such a policy of rewarding early entrants requires an accurate forecast of the social value of their inventions and discoveries – a process that can be fraught with failure. Not only that, given the uncertainty associated with new technologies, a delayed pattern of adoption might even be socially optimal.
A. Knowledge spillovers, dynamic scale economies, and industrial targeting Ever since David Ricardo, it has been well known that under free trade a country can increase its national income (and welfare) by moving resources into sectors in which its opportunity cost of production is lower than its trading partners. But is this prescription sufficient to generate economic growth? Perhaps not. Allocating resources according to comparative advantage can only ensure static efficiency and in no way guarantees dynamic efficiency. Succar (1987) argues “…the comparative advantage theory is a static construct that ignores forward linkages exist between present choices and future production possibilities. Therefore it cannot guide the pattern of international specialization when there are asymmetric learning opportunities associated with the production of different goods and/or use of certain techniques. Promotion of industries which generate substantial learning by doing economies should be an integral part of a strategy of human capital formation in LDCs.” In other words, Succar argues for some sort of industrial targeting although her model does not explicitly deal with this issue.
Even if one accepts the premise that certain industries are more likely to generate spillovers (based on knowledge diffusion or other factors), can policy be designed to encourage the ‘right’ industries? The ideal but rarely attained goal of industrial policy is the development of a general-purpose technology. DARPA, a small unit within the U.S.
Department of Defense that generated and financed a portfolio of projects, is widely credited with having been the key contributor to the development of the internet, the demand for this innovation being derived from the need to maintain communications during an assault on the U.S. This breakthrough was clearly fundamental and has social benefits many-fold the cost of the DARPA effort. This instance of success addressed a market failure, namely, the social benefits of research were much larger than the anticipated private benefits. Moreover, DARPA foresaw a potential need that may have escaped the purview of private firms. While the internet was a major technological breakthrough and suggests the potential gains from such an activity, it is useful to remember that, by their very nature, the discovery of such “general purpose technologies” is a rare event and less likely in low innovation intensity developing nations than in research rich developed countries.
The informational constraints facing policy-makers pursuing industrial policy are severe and any realistic model of industrial targeting needs to account for them. In a recent paper, Klimenko (2004) models industrial targeting as an optimal experimentation strategy of a government that lacks information about the set of industries in which the economy has comparative advantage with respect to rest of the world. He examines the set of industries in which a country will specialize as a result of such policy. In his model, for any set of targeted industries, it is possible to know whether or not a country will specialize in this set with positive or zero probability. He shows that an optimally designed industrial policy can actually lead a country to specialize in sectors in which it does not have comparative advantage. Depending on the beliefs of the policymaker, a country can end up abandoning the industries in which it has “true” comparative advantage.
Furthermore, Klimenko argues that the policy-maker may stop looking for better targets when the favored industries perform well enough. He interprets this outcome as a failure of industrial targeting policy even though it may not appear as such. In fact, he goes on to show that, despite the existence of market failures, the outcome of the learning process through private experimentation (without any assistance from the government) can even yield outcomes that are closer to the full information social optimum.
Klimenko’s rigorous analysis of this issue underscores our intuitive argument that the relevant counterfactuals are unavailable and what may appear to be a successful industrial policy may not be the first best outcome from a country’s perspective – merely doing something well need not imply one cannot be better at something else.
B. Coordination failures as a rationale for industrial policy The basic idea behind the coordination failure argument for industrial policy is that many projects require simultaneous investments in order to be viable and if these investments are made by independent agents there is little guarantee that, acting in their own self interest, each agent would choose to invest.5 As Scitovsky (1954) noted, reciprocal pecuniary externalities in the presence of increasing returns can lead to market failure because the coordination of investment decisions requires a signaling device to transmit information about present plans and future conditions and the pricing system is not capable of playing this role.
Pack and Westphal (1986) argue that such pecuniary externalities related to investments in technology are pervasive in industrialization. They provide an example of two infant industries (say A and B) where industry A produces an intermediate that is required in industry B and neither industry is profitable if it is established alone.
However, if both industries are established together, then both are profitable implying that it is socially optimal to indeed establish both. Of course, the problem is that without explicit coordination between investment decisions this outcome would not be obtained.
Okuno-Fujiwara (1988) presents a formal model of such interdependence between industries and the coordination failure that can result from such interdependence. He considers an economy with three goods: x, y, and z where good z serves as a numeraire and is produced under perfect competition with constant returns to scale. Good x is produced by a competitive industry and it requires good y as an intermediate. The technology for good y exhibits large economies of scale and the industry is assumed to be oligopolistic where the number of firms is endogenously determined to ensure zero profits in equilibrium. A coordination problem arises in the industry because the derived demand for the intermediate good y depends upon its price, which in turn determines incentives for entry into the intermediate sector. If y-producers anticipate low demand for their good, given the fixed costs of entry, few of them would want to enter implying a higher price for the intermediate which may then make industry x unsustainable – the key assumption here is that the intermediate good y must be locally supplied. On the other hand, if y producers are assured of a high demand for their product, more of them would Rodriguez-Clare (1996b) has shown that coordination failures can lead to ‘development traps’.
enter and such entry would lower its price which would then allow the high demand for the intermediate to be sustained.6 Okuno-Fujiwara (1988) shows that there is no unique equilibrium in a small open economy with the above production structure. In the bad equilibrium, the economy ends up specializing in good z where in the good equilibrium it produces both goods x and y and exports good x to the rest of the world (where the latter equilibrium Pareto dominates the former).
Turning to policy analysis, Okuno-Fujiwara suggests that three types of
traditional government intervention can help ensure that the good equilibrium is realized:
(a) the government can provide a production subsidy to either x or y industry (or both) thereby causing the two sectors to expand or (b) provide an export subsidy to the x sector;