Hedonic Adjustment

Hedonic adjustment is a statistical method used by price-index compilers to separate a genuine price change from a change in product quality. When a new model of a laptop, washing machine or item of clothing replaces an older one in a price survey, statisticians use a hedonic regression to estimate how much of any price difference reflects better (or worse) features, and only count the remainder as inflation. The name comes from the Greek word for pleasure, reflecting the idea that a product's price reflects the bundle of benefits its characteristics provide.

Hedonic methods are used in the Consumer Price Index (CPI) in the United States, in the euro area's harmonised index and elsewhere, and in national accounts deflators for computers and software. They matter because a price index that ignores quality improvements overstates inflation and therefore understates growth in real terms.

Suppose a phone that cost 500 last year is replaced by a new model at 550, with a better camera and twice the storage. Has the price of "a phone" gone up by 50? Partly, but some of that 50 is paying for a better product, not the same product at a higher price.

Hedonic adjustment is the price-checker's way of asking what the extra features are worth. By looking at many phones and their prices, the statistician estimates, say, that the bigger storage and better camera are worth about 40. Only the remaining 10 counts as a price rise. If the new phone had fewer features, the adjustment would work the other way and the measured price rise would be larger.

A hedonic model regresses the prices of many varieties of a product on their measurable characteristics. A common form uses the logarithm of price:

• $P_i$ is the price of item $i$.
• $x_{1i}, \dots, x_{ki}$ are its characteristics, such as memory size, screen size, fabric type or energy rating.
• $\beta_1, \dots, \beta_k$ are estimated coefficients: the approximate percentage effect on price of one more unit of each characteristic.
• $\varepsilon_i$ is the error term for factors the model does not capture.

Worked example: if the model estimates that each extra 100 GB of storage adds about 4% to a phone's price, and the replacement model has 200 GB more than the old one, the quality difference is worth about 8%. If the observed price is 10% higher, the index records roughly a 2% price increase rather than 10%. Estimating such models is a standard task in econometrics.

The U.S. Bureau of Labor Statistics (BLS) has used hedonic models in the CPI since the late 1980s, starting with housing and apparel and later extending them to items such as televisions, computers and other electronics and some appliances. The Bureau of Economic Analysis applies hedonic price indexes to computers and peripherals in the national accounts, which affects measured gross domestic product (GDP) and productivity. Eurostat's rules for the Harmonised Index of Consumer Prices allow hedonic and other explicit quality adjustments, and national statistics offices in Europe apply them to selected products. Producer price measures such as a wholesale price index (WPI) and the GDP deflator can also embed hedonic indexes for specific goods. Quality adjustment is a separate step from seasonal adjustment, which removes regular calendar patterns rather than product changes. Other methods, such as option costing (pricing the added feature separately) or linking in the new model, are used where hedonic models are impractical.

The 1996 Boskin Commission report to the US Senate estimated that the CPI overstated the cost of living by about 1.1 percentage points a year, and named quality change and new products as a major source of that bias. Its findings encouraged wider use of hedonic and other quality adjustments.

Measured inflation feeds into many decisions. It guides central banks pursuing inflation targeting, it sets cost-of-living adjustments (COLA) for pensions such as Social Security, and it drives the indexation of tax brackets and inflation-linked bonds. Quality adjustment also shapes measured real output: rapid hedonic price declines for computers are one reason measured real investment in information technology grew so fast from the 1990s, linked to technological progress.

Hedonic adjustment is sometimes criticised as a way of "hiding" inflation. Statistical agencies reply that the adjustments go both ways and that, in practice, their net effect on the overall CPI is small; BLS research has found that hedonic adjustment changes the all-items index only slightly, because it applies to a limited share of the basket. More substantive limitations include:

• Unmeasured characteristics. A model can only adjust for features that can be quantified, so durability, design or reliability may be missed.
• Value to the buyer. A feature valued by the market may not be wanted by every consumer, so a "quality gain" may not feel like one.
• Model choice. Results depend on which characteristics are included and how often the model is re-estimated.
• Perception gap. Shoppers judge prices by their bills, which can make official inflation look too low to them. Surveys of household inflation expectations often run above measured inflation, a gap studied alongside money illusion in behavioural research.

• U.S. Bureau of Labor Statistics: Quality adjustment in the CPI
• Eurostat: Harmonised Index of Consumer Prices