More than Moore: Comparing Forecasts of Technological Progress

No Thumbnail Available
Farmer, J. Doyne
Nagy, Bela
Trancik, Jessika Ebba
Associated Organization(s)
Organizational Unit
Supplementary to
A key challenge in modeling technological innovation is to predict future costs based on historical data. This is of great interest to academics, as well as decision makers both in the private and public sectors. For example, many corporate strategies, industry roadmaps, and government policies are crucially dependent on the accuracy of forecasts of certain technological capabilities, e.g. in semiconductors, renewable energy technologies, etc. Recent research (Koh and Magee, 2006; Koh and Magee, 2008) has uncovered a set of long-term, approximately exponential empirical trends in performance of information technologies and energy technologies over time. These appear similar to the well-known Moore's law for integrated circuits; however, they typically span multiple technologies and many decades, in most cases extending back to the nineteenth century. These results suggest that time as an independent variable may be a useful predictor of cost or price measures that are often used as dependent variables to quantify the improvement of technologies over time. On the other hand, many cost forecasting methods employ cumulative production volume instead of time as the crucial explanatory variable, following Wright (1936) who identified experience as the most important factor affecting the cost of airplanes. This has spawned an extensive literature on learning curves (experience curves), providing substantial empirical evidence for a simple (often power law) relationship across a wide range of industries (e.g. Dutton and Thomas, 1984). Theory is lagging behind current practice, since at this time there is no generally accepted explanation of why we should see exponential or power law performance curves. A recent workshop on modeling technological innovation at the Santa Fe Institute generated a lively debate about what is a better "clock" for measuring improvement: calendar time or production experience? This debate has great relevance for public policy design to stimulate technological innovation, such as energy and climate policy. Another open question is how much of the so-called experience effect or learning effect is due to simple economies of scale. For instance, Goddard (1982) argues that by "combining annual production and time in the single variable, cumulative production, it hides more than it reveals", advocating modeling cost or price as a function of the annual production rate instead of cumulative production. One way to evaluate the competing models is to look at their prediction performance over time, since one could argue that in practice, the closer the forecasts are to the actual prices, the more useful the method is for the stakeholders. We compared the historical prediction accuracy of the various models proposed, using time vs. cumulative production vs. annual production (and all combinations of the three) in several different industries, including a number of information technologies and energy technologies. Initial results suggest that relying on time alone may be a risky strategy, since in many cases one may not achieve the desired accuracy. For example, one of our surprising results was that even for transistors (representing the quintessential technology described by Moore's law), using time as the only explanatory variable could have led one to a substantial bias. Erroneous forecasts included prices that were only one tenth or one hundredth (or less) of the actual values. This suggests that taking into account production data may lead to better forecasts. The significance of this work goes beyond identifying the most relevant factors in cost innovation. Our goal is to provide practical forecasting guidelines to managers and policy makers backed up by the empirical evidence to date. Our hope is that providing a simple summary of the alternatives, and an intuitive, visual way to evaluate them will lead to a more accurate view of the future and ultimately, to better decisions. H. Koh, C.L. Magee, A functional approach for studying technological progress: application to information technology, Technological Forecasting & Social Change, 73 (2006) 1061-1083. H. Koh, C.L. Magee, A functional approach for studying technological progress: extension to energy technology, Technological Forecasting & Social Change, 75 (2008) 735-758. T.P. Wright, "Factors affecting the cost of airplanes," Journal of Aeronautical Sciences, 3 (1936) 122-128. J.M. Dutton, A. Thomas, Treating progress functions as a managerial opportunity. Academy of Management Review, 9 (1984) 235-247. C.T. Goddard, Debunking the learning curve, IEEE transactions on components, hybrids, and manufacturing technology, Chmt-5 (1982)
Date Issued
Resource Type
Resource Subtype
Rights Statement
Rights URI