Learning curves
Import Subsidies?

An important form of technological progress is learning-by-doing, in which a firm's productivity increases with production experience. One way of formulating this effect is to assume that a firm's cost per unit of production declines as its cumulative output rises. Such a relationship is called a learning curve. Learning creates a natural monopoly, since industry-wide unit cost falls fastest if production is restricted to a single firm. Learning also involves sunk costs. A firm cannot 'unlearn' and recoup the expenditure it has incurred acquiring production experience.

To date economic theory has told us very little about the industrial structure one might expect in the presence of a learning curve, although such an understanding is essential to the formulation of economic policy. In Discussion Paper No. 80, Joseph Stiglitz and CEPR Programme Director Partha Dasgupta explore this relationship between learning-by-doing, industrial structure and industrial and trade policy.

Dasgupta and Stiglitz first analyze the effects of learning on the production and pricing policies of a nationalized industry. They find that learning provides a nationalized industry with a reason for pricing its product below current marginal production costs: the learning curve means that today's higher production leads to lower costs tomorrow. If the nationalized industry must break even over its planning horizon, then setting price below current marginal cost is still the right policy in the initial period, though in later stages price must exceed the then current unit cost so as to cover the losses in earlier years. Dasgupta and Stiglitz also derive production and pricing rules that a protected private monopolist would follow, and they show that monopoly output is less in each period than the optimal output of a revenue-constrained nationalized industry.

The authors then examine the more interesting case in which market structure is not given but is influenced by the learning process. Suppose an incumbent firm is threatened by a rival which can enter the market now or at any time in the future. Entry by the rival would involve it in fixed costs that must be sunk. Dasgupta and Stiglitz show that the potential entrant has absolutely no effect on the behaviour of the incumbent firm. In the presence of sunk costs, the invisible hand is not merely weak, it is paralysed.

Dasgupta and Stiglitz then examine government policies for such an industry, exploring the implications of two anti-trust policies often advocated in the United States. They first analyse the effect of the imposition of a ceiling on the market share that a given firm can enjoy. They also examine the effect of the government's setting current marginal costs as the lower bound on the price that the incumbent firm can charge - the rationale being that a lower price shows predatory pricing. Dasgupta and Stiglitz find that both these policies can reduce social welfare in a wide variety of circumstances.

Dasgupta and Stiglitz also examine optimum trade policies in the presence of learning-by-doing. Learning gives government an argument for subsidizing domestic production, not for imposing an import tariff on foreign competition. In fact, if foreign firms face a learning curve and if domestic production is possible only at a high cost, there is a case for subsidizing imports!


Learning-by-Doing, Market Structure and Industrial and Trade Policies
Partha Dasgupta and Joseph Stiglitz

Discussion Paper No. 80, October 1985 (ATE)