When faced with measuring prices for products which undergo rapid quality change, international best practice is to develop hedonic price indexes when suitable source data are available. This is the approach being advocated by international agencies such as the Organisation for Economic Co-operation and Development³, the International Labour Organization and the International Monetary Fund.
A hedonic price index is any price index that utilises, in some manner, a hedonic function. In broad terms, a hedonic function identifies the relationship between the prices of different varieties of a product, such as differing models of personal computers, and the characteristics within them. By comparing prices and features of various computers, a hedonic regression model assigns values to each of the particular features that are identified as price determining (for example, processor speed, memory, disk capacity etc.).
Personal computers are an area of rapid technological change. Products available in the marketplace change frequently as new features are added and existing features improve. For example, the rapid change in hard disk size, random access memory, and clock speed of desktop personal computers is well documented. A further issue is that older models quickly become redundant. The net result of these changes is that over any two periods there are both new products and discontinued products, with the result that comparing like with like becomes difficult. This is of particular concern when it is observed that improved features on later models do not always result in a price rise, or a commensurate price rise that would be observed if the components were bought separately (again, a bigger hard disk drive is an example). The quality adjustment problem is applicable to all price indexes, not just those for personal computers. However, traditional approaches to solving this problem (for example, matched model approaches, explicit quality adjustments, or component level pricing, amongst others) are inadequate for these sorts of products.
The Producer Price Indexes use a form of hedonic index known as the 'consecutive two period chained time dummy double imputation hedonic price index' for use in price indexes for personal computers. This process sees a matched model price index applied for personal computers sold between consecutive periods, combined with a consecutive-period time dummy price index (this is produced by using regression techniques) to measure price changes for both discontinued and newly introduced products.
The double imputation method can best be thought of as a traditional matched model index with an explicit adjustment applied because of both the departure of superseded models and the introduction of new models. A key deficiency of the basic matched model approach is that it makes no provision for systematically including the effects of price and quality changes in models available in the marketplace, and determines price change by only considering those models which appear in the market in both periods of interest. In other words, any improvement in quality associated with the introduction of a new model will not be measured if only matched models are priced.
The double imputation price index counters this deficiency by implicitly imputing price movements for both superseded models and newly introduced models. This is where the term 'double imputation' arises. A hedonic regression model is run on the dataset each period and from this, a price factor is determined for each characteristic of the computer. Whilst this gives the ability to calculate the price for each specification, what is actually used in the imputation, is the time dummy variable. The time dummy variable is representative of the price change between the two periods taking into account the different characteristics of the computer. This is combined with the matched model index to create the double imputation index which will reflect the movement of the whole sample. The index is then considered representative of all transactions, since recently superseded models and new models are included in the determination of price change, in addition to products common to both periods.
Further, the implicit imputation process at the core of this technique uses a hedonic function to adjust for changes in the characteristics of both the new and superseded models; that is, the prices imputed are adjusted for quality change, and hence the resulting index measures pure price change.
The process utilises price data from Australian vendors of personal computers, and so is not only representative of the Australian marketplace but also avoids issues with both exchange rate fluctuations and arbitrarily lagging prices to take account of shipping times etc. The double imputation index uses a hedonic function based on characteristics of personal computers sold in the Australian marketplace, using prices in Australian dollars, and so furthermore does not rely on the restrictive assumptions underlying a universal hedonic function.
Any movements in the double imputation index can be decomposed into the movement due to changes in prices of the matched sample, and the movement due to changes in other products in the marketplace. Movements in the index can be explained in terms of changes in list prices of existing products and changes in quality of new products, and so the resulting measures are easily explainable to users.