Technological Change, Learning-by-Doing and the Structure of Production in the U.S. Machine Tool Industry

See National Research Council, 1983; and Krug, 1982.Google Scholar

The details and evidence are summarized in Ch. 5.Google Scholar

The cyclical volatility of output, employee-hours, and productivity in the U.S. machine tool industry, when compared to U.S. manufacturing generally, show the same cyclical pattern with larger amplitudes; and manufacturing is more volatile than the economy as a whole. See Tsai, 1991, pp.33.Google Scholar

This is the research conclusion by National Research Council, 1983, and also see Krug, 1982.Google Scholar

There are two separate multiproduct RVCF models; one for the metal cutting machine tool industry (SIC 3541), the other for the metal forming machine tool industry (SIC 3542). Products of the metal cutting industry are divided into 3 product categories: (1) Type MC-I, including boring machines (SIC 35411), drilling machines (SIC 35412), gear cutting machines (SIC 35413), and grinding, polishing, honing and lapping machines (SIC 35414). (2) Type MC-II: lathes (SIC 35415), milling machines (SIC 35416), machining centers (SIC 3541 A), station-type machines (SIC 354IB), and metal cutting machinetools, nee (nee: not elsewhere classified) (SIC 3541C) have been incorporated into the second category. (3) Type MC-III: machine tools for home workshops (SIC 35418), parts for metal cutting machine tools (SIC 35419), and machine tools, metalcutting types, nsk (nsk: not specified by kind) (SIC 35410) are in the third category. The product types in the metal forming machine tool industry are divided into two categories: (1) Type MF-I: punching, shearing, bending and forming machines (SIC 35421), presses, except forging (SIC 35422), and metal forming machine tools, n.e.c. (SIC 35423) are in the first category. (2) Type MF-II: parts for metal forming machine (SIC 35424), machine tools, metal forming types, nsk (SIC 35425) are in the second product category.Google Scholar

See Norsworthy and Jang (1992), ch. 3, for a discussion of the advantages of the demand equation over the share equation.Google Scholar

Conventional procedure is to estimate the cost function (2.1) jointly with three of the four share equations (2.2). McElroy (1987) proposed and demonstrated that the Additive General Error Model (AGEM) proved to be more advantageous in that it incorporates more information, and the error terms have direct interpretations as input quantities. Norsworthy and Jang (1992) argue that estimating the translog cost function with the demand equations leads to models that better represent the technology of production because the technology of production would be better identified with errors in input quantities rather than value shares. They extend the AGEM and obtain results that are superior to those obtained from estimating share equations in all industries they examine.Google Scholar

In the sense of Denny, Fuss and Waverman (1981).Google Scholar

A third order specification would be required to measure economies of scope among all three inputs.Google Scholar

Miguel A. Reguers, “An Economic Study of the Military Airframe Industry,” Wright-Patterson Air Force Base, Ohio, Department of the Air Force, October 1957, pp.Google Scholar

218.See Webre, Philip C. (1983) for a review of the learning curve literature.Google Scholar

Nerlove(1963).Google Scholar

Strictly speaking, learning effects are only partially irreversible in that they deteriorate through time, due to labor turnover and changes in the production process; in other terms, the “capital stock” of learning acquired through past production experience depreciates.Google Scholar

In recent studies, the time factor has been replaced by an index of learning: the cumulative output (Verdoorn, 1956), cumulative gross investment (Arrow, 1962), or cumulative manhours (Wright, 1956).Google Scholar

In some other industries, eg. for semiconductor manufacturing, learning effects appear to be incorporated into the technology of production. (Norsworthy and Jang, 1993 a)Google Scholar

For the detailed approach for internal rate of return and the service price of capital, see Fraumeni and Jorgenson (1986).Google Scholar

Serial correlation does not bias the point estimates of the coefficients.Google Scholar

Computation of long run Hicks-Allen partial elasticities of substitution from the RVCF is shown in Brown and Christensen (1981).Google Scholar

Berndt and Hesse (1986) and Kulatilaka (1987) reported the difficulties in obtaining numerical convergence with the translog RVCF and thus with computing estimates of long-run elasticities due to its nonlinear logarithmic form. It will be exacerbated if curvature conditions are close to being violated.Google Scholar

These requirements have in turn created new opportunities for responsive builders and for new plays from outside the traditional builder structure and technology base. For example, both Allen-Bradley and General Electric (GE) tried to translate success with programmable logic controllers to get into NC and CNC controls, and in GE’s case, an overambitious attempt to produce the entire “Factory of the Future.” New electronics-based players also include Digital Equipment Corporation (DEC), now working with Italy’s foremost builders, Coman, to link every element of the automated factory; and Big Eight accounting firms who believe their familiarity with information systems allows them to integrate factor systems. These system integrators are required to tap resources and skill bases outside the scope of traditional machine tool builders.Google Scholar

Based on the conclusion of Tsai and Norsworthy (1990).Google Scholar

This might result from the lack of degrees of freedom in the model when the index of learning curve effects enter into this three-product RVCF model. The estimation was successful for the metal forming industry, which has fewer outputs. Without more observations, it would be hard to measure the average rate of technical change in metal cutting machine tools to distinguish separate learning curve effects for each major product class.Google Scholar

In the data-rich environment provided by a pooled time series/cross-section of plant-level data, it would be possible to determine these depreciation rates directly by estimation.Google Scholar

For example, Norsworthy and Jang (1993b) find very low productivity growth in telecommunications equipment based on official statistics, whereas various quality adjustment methods show larger output growth rates that would lead to more rapid measured productivity growth.Google Scholar

Recall that declining employment in the industry means that the learning effect leads to higher costs and reduced productivity since cumulative employment, the source of the learning effect, is reduced.Google Scholar