Price Index

An index is defined by four choices, and each is a judgement rather than a discovery. Together they determine what question the index answers, and none of them is visible in the resulting number. Reading an index without its methodology is reading an answer without knowing the question.

The basket

Which items are included. This determines the index’s scope, and it draws a boundary: everything outside the basket is invisible to the index no matter how its price moves.

The base period

The reference the index is expressed relative to. It is an arbitrary choice with no economic significance, and changing it rescales the whole series without changing anything the series says.

The weights

How much each item counts. Usually derived from trade or consumption shares in some period, which is itself a choice. Weights are where an index encodes a view of what matters.

The formula

How the items are combined. Different formulae treat substitution and weight changes differently and produce different series from identical price data.

The price series used

Which prices represent each item. An index of international quotations and an index of retail prices for nominally the same basket measure entirely different phenomena.

The base period is the choice most often over-read. Because an index is conventionally expressed against a base, readers infer that the base was somehow normal or that a reading above it means prices are high. Neither follows. The base is a scaling convention, chosen for convenience, and an index above its base means only that prices are above what they were in an arbitrarily selected period.

The virtue of an index is compression: it reduces a great many prices to one series. The limitations all follow from the same property, because compression is lossy and the information it discards does not announce itself.

• It cannot tell you a price — an index is dimensionless and no commodity has a price of "an index number"
• It cannot tell you what any component did — a stable index is entirely compatible with violent, offsetting movements underneath it
• It cannot tell you about anything outside its basket, however economically important that thing is
• It cannot tell you what a party paid — real transactions occur at specific prices for specific lots, and an index has averaged all of that away
• It cannot tell you about levels across space — an index says a market moved, not whether it is expensive relative to anywhere else
• It cannot be compared with another index built on a different basket, base, or weighting, however similar the two sound

The second point is the one that most often produces confident error. An index moving little is routinely read as a market that was quiet, when it may equally be a market where large movements in opposite directions cancelled. The index records the net, and the net is not the story — it is one summary statistic about the story, and a lossy one.

An index faces a structural dilemma that has no clean solution. Weights derived from a base period describe the trade or consumption pattern of that period. As time passes, the pattern changes — different commodities matter more, substitution occurs, new products appear — and the index’s weights become progressively less representative of what it purports to measure.

Updating the weights fixes representativeness and breaks comparability: a series whose weights changed is not measuring the same thing before and after, so movements across the join are partly artefacts of the revision. Not updating preserves comparability and lets the index drift away from the reality it describes. Every index publisher must choose a position between these, and every position is a compromise rather than a solution.

For a reader, the consequence is practical. Long index series are the ones most susceptible to this, and precisely the ones people most want to use, since a long history invites comparison across decades. A reading compared with one from many years ago may be comparing two differently constructed measures wearing the same name, and only the publisher’s methodology documentation reveals whether it is.