to matched models, which have at times underestimated the true rate of price performance improvement in computers.14
Dr. Flamm’s second point was that some people have done credible studies suggesting that if one tries to calculate the percentage of the improvement in computers that comes from semiconductors, the answer depends on the weight of semiconductors in the value of computers. The problem is that national income accounts for the United States do not have good information for this weighting. According to Dr. Flamm, if the semiconductor content in computers is 15 to 20 percent, then about half the technological improvement in computers has come from semiconductors. If the content is as high as 40 percent, then virtually all the improvement is coming from semiconductors. This, he reminded his audience, has to be amended by the fact that computer numbers are probably too low, so other factors beyond semiconductors must play a role. It is a “giant mess” if you look only at the traditional numbers.
Dr. Flamm added a note from his personal experience in working on litigation questions. An interesting aspect of litigation work, he said, is the occasional access to proprietary data, which may be more detailed and more accurate than publicly available data. The public data show price performance improvements in personal computers in the range of 30 to 35 percent for the early 1990s. That is an uptick from a historical rate, which has been closer to 20 percent since the 1950s. Recently, however, unpublished hedonic studies of price performance indexes for personal computers have shown annual rates in the 40-45 percent range. He said he could not make public the proprietary data underlying these statements, but remarked that credible hedonic studies have shown these really dramatic and surprising price declines in personal computers after adjusting for other factors (Figures 10 and 11).
Where, he asked, is this large and surprising number coming from? He returned to Moore’s Law, and suggested there might be an economic corollary. This extension of Moore’s Law might argue that the industry is able to produce four times as many devices every three years per square centimeter of area. If one adds the assumption that wafer-processing costs remain roughly constant, the result is a minus 37 percent per year compound annual growth rate in the cost of the device.