Differences in Hedonic and Matched-Model Price Indexes: Do the Weights Matter? (PDF)

This note uses scanner data for over 60 segments of consumer information technology (IT) and electronic goods to construct matched-model indexes. Virtually all of the segment-level indexes constructed with these data show price declines that reflect quality increases—a typical exception is floppy disks, a category that shows price declines that reflect falling average prices. Our first pass at these data show that in all but nine of the categories, unweighted geometric mean price indexes falls faster than weighted superlative indexes (Fisher and Tornquist). Part of the reason for this appears to be that, within each segment, goods with relatively low market shares tend to show faster price declines than those with high market shares. Although it would be interesting to explore whether life-cycle effects over the life of each good also contribute to this result, the time-series dimension of our data is short and precludes an analysis of pricing over the product cycle.

Because dummy variable hedonic measures (DV) are also unweighted, our preliminary finding suggests that, all else held equal, DV indexes will tend to show faster price declines than their superlative counterparts. Of course, hedonic techniques arguably do a better job of capturing quality change and will tend to show faster price declines for that reason. Our only point is that maybe the weights matter too.

Ana M. Aizcorbe and Yvon H. Pho