Stratification and price index computation

How to stratify dynamic populations of articles for price index computations.
Because product populations are dynamic, it is not possible to follow the price development of products at the GTIN level and compute price indices on this basis. In reality each GTIN has a finite lifespan (its time on the market). Products are also re¬newed by replacing them with others. Sometimes these replacements are simply relaunches of an earlier product, and they are basically as old wine in new bottles. And they come with a new price, typically higher than the product they replace. If all GTINs are treated as separate products price jumps due to relaunches will be missed. In order to avoid this one could try to link relaunches to their predecessors. But this is not so easy as it may seem. Another option is to classify the items on the basis of some common, key features, and consider subgroups of COICOP groups as strata and follow their price development. In order to do this, however, these features should be extracted from the product descriptions available in scanner data or internet data. The present paper discusses which properties stratifications of product populations should have.