Price Falls Of Network Servers Have Been Dramatically Underestimated

Quality-adjusted prices for network servers – the computers that run the internet – have been falling twice as fast as we think, according to new research by Professor John Van Reenen, published in the February 2006 issue of the Economic Journal. Productivity growth has therefore also been faster than previously thought.

Productivity growth accelerated in the United States after 1995 and some commentators attribute this mainly to the rapid fall in the quality-adjusted prices of computers. People tend to focus on PCs when they think of computers. But in the corporate sector, the growth of spending on network servers outstripped that on PCs in the late 1990s. This study examines the market for network servers, which run the internet and most modern intranets. It finds that statistical offices across the world may have been dramatically underestimating the fall in real prices for servers since the mid-1990s. Rather than US prices falling at about 15% a year, they appear to have been falling around twice as fast. There is also some evidence that price falls may have been faster in the United States than in Europe.

Statistical offices try to adjust for quality by following the same model of computer server over time. So long as the quality does not change, this is a good way of controlling for quality and it is called the ''matched model'' method.

A big problem arises, however, when old models become obsolete and are withdrawn from the market and new, better quality models enter. It is not possible to ''quality correct'' for a model that no longer exists (nor can the change in price be followed for a good that has left the market). This is a version of ''selection bias'' – the models that the statistical agencies use to calculate the price index make it biased because it leaves out the big falls in price of the obsolete models.

One way to correct for this problem is through ''hedonic'' prices. This is a statistical technique that ''prices'' the characteristics of the component parts of servers (memory, speed, etc.). Thus, when a model exits or enters, it can be given a virtual price according to the bundle of characteristics from which the server is built.

This article argues for a combined or ''hybrid'' price index that uses the matched model approach when we observe it, but imputes using the hedonic method for periods when we do not observe price.

The article gives estimates for these different price indices. It finds that because a large number of servers enter and exit the market in any given year, there is a big bias from the standard matched model approach. Quality-adjusted prices for servers have been falling twice as fast as we think. Productivity growth has therefore also been faster than previously thought.

''The Growth Of Network Computing: Quality-Adjusted Price Changes For Network Servers'' by John Van Reenen is published in the February 2006 issue of the Economic Journal. John Van Reenen is professor of economics and director of the Centre for Economic Performance at the London School of Economics.