Alternative price indices for computers in the Netherlands using scanner data

Using a scanner data set that covers nearly all computer sales in the Netherlands for a period of three years, this paper investigates whether there is a significant difference between a matched model index and a hedonic imputed index, which also takes non-matched observations into account. The result of this study was that this does not appear to be the case. The lack of significance of the difference can be attributed to two reasons: the high share in sales values of the matched items, and the mediocre fit of the hedonic model. Given the fact that an earlier study based on a different data set (Van Mulligen, 2002) also pointed out that the difference between a matched model index and a hedonic imputed index is small, we draw the conclusion that making explicit adjustments for non-matched items is not necessary. Although the official CPI for computers also uses the matched model methodology, it appears to introduce a substantial downward bias in the actual quality adjusted price index, due to biased sampling and the lack of representative weighting of individual items. The main area for improvement of the CPI for computers (and possibly other durables as well) lies therefore in more frequent sampling, weighting and chaining of the indices rather than making explicit quality adjustments.