New Growth Theory and Increasing Returns

According to new or endogenous growth theory, economic growth can be understood as a process of learning-by-doing, within a firm, within an industry, and within a metropolitan area as well. (Arrow 1962, Romer 1986, 1993, Lucas 1988, Krugman and Obstfeld 1997.)

By way of counterpoint, what might be termed “exogenous growth theory” sees rising output per capita as resulting from externally given increases in the quantities of labor and capital. This in outline is the constant-returns-to-scale approach modeled by Dale Jorgenson. In Jorgenson’s growth accounting (a classic example of what intricate model refinements can accomplish), virtually all U.S. economic growth in the 20^th century can be statistically explained by the increases in quantities of the factors of production—when the “quantity” of labor is specified in ways that include increasing education for workers. This, as if there were no technological or organizational change, no economies of scale, but only constant returns in an environment altered solely by investment and labor-force growth.

In endogenous growth models, on the other hand, growth over time entails increasing returns to scale for a metropolis or a national economy. A proportionate increase in labor and capital gives rise to more than proportionate gains in output. The explanation lies in better “recipes,” as Romer terms innovations, and in spillovers that operate over time, enhancing skill and productivity levels throughout the economy.

Learning-by-doing within a firm means that current unit costs are a function of experience (as measured by the firm’s total cumulative past output). Given the learning curve for a single firm, then imitation of successful firms on the part of other firms in the industry spreads the “learning” around, such that the industry can benefit from falling-forward supply curves. The process links unit costs to cumulative industry output within a country. The ease of imitation and learning then increases within spatial agglomerations—which in turn can be understood as “little nations” (my phrase) benefiting from increasing returns. (Krugman and Obstfeld 1997, p. 154.)

In sum, interpreted variously in terms of Arrow’s learning curves, Romer’s recipe for growth, or Lucas’s vector-autoregressive (VAR) time-series specifications, the key ideas are learning-by doing and cross-fertilization over time.