This is one of those classic papers where the result is so well-known I’d never bothered to actually look through the paper itself. Manuel Trajtenberg, in the late 1980s, wrote a great book about Computed Tomography, or CAT scans. He gathered exhaustive data on sales by year to hospitals across the US, the products/attributes available at any time, and the prices paid. Using some older results from economic theory, a discrete choice model can be applied to infer willingness-to-pay for various types of CAT scanners over time, and from there to infer the total social surplus being generated at any time. Even better, Trajtenberg was able to calculate the lifetime discounted value of innovations occurring during any given period by looking at the eventual diffusion path of those technologies; that is, if a representative consumer is willing to pay Y in 1981 for CAT scanner C, and the CAT scanner diffused to 50 percent market share over the next five years, we can integrate the willingness to pay over the diffusion curve to get a rough estimate of the social surplus generated. CAT innovations during their heyday (roughly the 1970s, before MRI began to diffuse) generated about 17 billion dollars of surplus in 1982 dollars.
That alone is interesting, but Trajtenberg takes this fact one step further. There has long been a debate about whether patent citations tell you much about actual innovation. We know from a variety of sources that most important inventions are not patented, that many low-quality inventions of little social value are patented, and that patents are used in enormously different ways depending on market structure. Since Trajtenberg has an actual measure of social welfare created by newly-introduced products in each period, and a measure of industry R&D in each period, and a measure counting patents issued in CT in each period (nearly 500 in total), he can actually check: is patenting activity actually correlated with socially beneficial innovation?
The answer, it turns out, is no. A count of patents, at any reasonable lag and any restriction to “core” CT firms or otherwise, never has a correlation with change in total social value of more than .13. On the other hand, patents lagged five months has a correlation of .933 with industry R&D. No surprise, R&D appears to buy patents at a pretty constant rate, but not to buy important breakthroughs. This doesn’t, however, mean patent data is worthless to the analyst. Instead of looking at patents, we can look at citation-weighted patents. A patent that gets cited 10 times is surely more important than one which is issued and never heard from again. Weighing patents by citation count, the correlation between the number of weighted patents (lagged a few months to give products time to reach the market) and total social welfare created is in the area of .75! This result has been confirmed many, many, many times since Trajtenberg’s paper. Harhoff et al (1999) found, using survey data, that each single patent citation for highly-cited patents is a signal that the patent has a additional private value of a million US dollars. Hall, Jaffe and Trajtenberg (2005) found that, using Tobin’s Q on stock market data holding firm R&D and total number of patents constant, an additional patent citation improves firm value by an average of 3%.