Technology Forecasting Methods: Overview

Why bother forecasting technology?

The need to forecast technological change is an intrinsic part of the planning process for a wide variety of fields. With better forecasting, financial and engineering decisions can be optimized and policy can be adapted proactively instead of retroactively. To make an accurate forecast it is vital to use the correct method. As discussed in Statistical Basis for Predicting Technological Progress, a variety of rules describing how technology improves over time have been proposed over the years, though they had not been tested against large data-sets.

Historical Technology Forecasting Methods

The earliest proposal, made by Theodore Wright in 1936 (Wright TP (1936) Factors affecting the costs of airplanes. Journal of Aeronautical Sciences 10:302-328.) proposed that the cost of a unit of a technology decreases as a power law of cumulative production. Unfortunately, we were unable to access the original article. Gordon Moore, a co-founder of Intel, noted in a paper in 1965 that the number of components in integrated circuits had doubled every year from 1958 to 1965 and predicted that the trend had continued. The famous 18 month formulation often cited came later, from David House. In 1982, Charles Goddard at Bell Labs proposed a modification of Wright’s Law: that in addition to the relationship between manufacturing cost and cumulative production, that there is a corresponding relationship between manufacturing cost and annual production. He criticized Wright’s law due to its cumulative nature hiding cause and effect-the ‘historical learning curve’ disallows for further analysis of production costs. Instead, he focused on an annual control signal reflecting the changing nature of the industry. At the high level (specific math representation will be discussed as we review the advantages, disadvantages, and prediction power of each method), we can look at the methods in the following way (Nagy et. al):

  • Moore’s Law: The cost y of a given technology decreases exponentially with respect to time.
  • Wright’s Law: The cost y of a given technology decreases exponentially with respect to cumulative production.
  • Lagged Wright’s Law: The cost y of a given technology decreases exponentially with respect to cumulative production, but it takes some amount of time for cumulative production to be reflected in price.
  • Goddard’s Law: The cost y of a given technology decreases exponentially with respect to annual production, or the scale of production.

The cited paper also notes two additional synthesized methods- Nordhaus, which combines Wright’s Law and Moore’s Law; Sinclair, Klepper, and Cohen, which combines Wright’s Law and Goddard’s Law.

What We’re Covering

Over the next weeks, we will be examining the different methodologies used in forecasting, what data sets have been used for forecasting, and the reasons developments occurred. We’re also trying to get in touch with relevant researchers to provide our readers and clients with better insight into the process. Finally, we’ll carry out and recreate an analysis of our own to show how you might forecast technological change, explain how we did it, and where the issues are.