Consumer Price Index (CPI)

The Consumer Price Index is a family of price indexes published monthly by the Bureau of Labor Statistics, each measuring the change over time in what a defined population pays for a market basket of goods and services. The best-known member is CPI-U, which covers all urban consumers, and in the Internal Revenue Code the unqualified phrase means exactly that: section 1(f)(5) defines "Consumer Price Index" as "the last Consumer Price Index for all-urban consumers published by the Department of Labor."

The phrase "the CPI" is therefore a small trap. Alongside CPI-U, BLS publishes CPI-W for urban wage earners and clerical workers, and the Chained Consumer Price Index for All Urban Consumers, written C-CPI-U, which the Code defines at section 1(f)(6)(A) and which uses a method that allows for substitution between goods. Different legal regimes point at different ones, so the answer to "how much did the index rise" depends on which index and over which window. The phenomenon the index measures, including the distinction between headline and core readings and the reasons a household's own experience diverges from a national average, belongs to the entry on inflation. This entry is about the instrument.

The reason to keep the versions straight is that three of them govern three things in a single retiree's finances, and each is fixed by a different document.

Federal tax figures run on chained CPI. Section 1(f)(3)(A) sets the annual adjustment as the percentage by which "the C-CPI-U for the preceding calendar year" exceeds "the CPI for calendar year 2016," and section 1(f)(6)(B) defines the annual value as "the average of the C-CPI-U as of the close of the 12-month period ending on August 31 of such calendar year." That switch was made by section 11002 of the 2017 tax act, it reached essentially every indexed figure in the Code at once, and its effective-date provision says only that the amendments "shall apply to taxable years beginning after December 31, 2017," with no termination date attached. So chained indexing has been permanent since enactment rather than something later legislation had to preserve. The 2016 handover is bridged by a splice factor in section 1(f)(3)(B), which divides the chained index for 2016 by the traditional index for 2016, so the change of series did not itself reset any threshold.

The Social Security increase runs on CPI-W, and the reason is worth stating precisely because it is usually stated wrongly. The statute at 42 USC 415(i) refers only to "the Consumer Price Index ... as prepared by the Department of Labor" and names no version. The Social Security Administration explains the choice in its own annual determination: "At the time the Act was amended to provide automatic cost-of-living increases starting in 1975, only one CPI existed, namely the index now referred to as CPI for Urban Wage Earners and Clerical Workers (CPI-W). Although the Bureau of Labor Statistics has since developed other CPIs, we follow precedent by continuing to use the CPI-W." That is an agency practice resting on history, not a statutory command, which is a meaningful distinction in any argument about changing the measure.

Series I savings bond rates run on a third variant. Treasury states that it bases the inflation component "on changes in the non-seasonally adjusted Consumer Price Index for all Urban Consumers (CPI-U) for all items, including food and energy." So it is the unadjusted, all-items series, which is neither the chained measure the tax tables use nor the wage-earner measure Social Security uses.

Two features of how the numbers are produced matter more than they sound. The first is seasonal adjustment. The same month has both a seasonally adjusted and an unadjusted reading, month-over-month commentary conventionally uses the adjusted series while year-over-year comparisons and formula applications such as the savings bond rate use the unadjusted one, and a reader comparing two reports of "inflation last month" may simply be comparing the two versions. The second is revision. Chained CPI is uniquely subject to later revision, because its expenditure weights are estimated when first published and firmed up afterward, and section 1(f)(6)(A) locks the tax figures to "the latest values so published as of the date on which such Bureau publishes the initial value ... for the month of August for the preceding calendar year." The tables are therefore built on an unrevised snapshot and are never recomputed when the index is later corrected. The other series in this entry carry no such asymmetry.

Two further points about construction explain much of the confusion around the index. The first is how owner-occupied housing is handled. Shelter enters through rent and through owners' equivalent rent, an estimate of what an owner-occupied home would rent for. The price of a house does not enter, and the reason is that a house is treated as a capital asset rather than as something consumed: its price reflects both an investment value, which a cost-of-living index sets out to ignore, and the shelter it supplies, which is the part the index is after. Owners' equivalent rent is the attempt to isolate the second. The change was announced in 1981 and phased in one series at a time rather than all at once, reaching CPI-U with the data for January 1983 and CPI-W two years later. Before it, the measurement turned partly on mortgage interest rates, so a rise in rates by itself pushed measured inflation up. This is the best available explanation of why the index can look wrong to someone shopping for a first home in a rising market: what they are about to pay for the asset is not what the index is measuring.

The second is that the family includes a member built for older households, and it is not the one the benefit formula uses. In a 2019 Federal Register request for comment on the consumer inflation measures produced by federal statistical agencies, the Office of Management and Budget described an experimental Consumer Price Index for the Elderly, written CPI-E, which "uses the same price surveys and formulas as the CPI-U and CPI-W, but uses expenditure weights for households with a reference person or spouse aged 62 years or older," and stated that it "is an experimental index and is not currently used for official purposes." Legislation to substitute it for CPI-W is introduced in Congress after Congress, and anyone relying on its status should confirm it at the Bureau, which publishes the series.

The production cycle runs in one direction and each user picks its own window out of it. BLS collects prices, computes each index, and publishes monthly, releasing seasonally adjusted and unadjusted versions of the same period.

From there the three applications diverge. For federal tax figures, Treasury averages the twelve chained readings ending August 31, compares that average against the statutory base year, and the resulting percentage produces the following year's tables, which is why they appear in the autumn and why a December price shock cannot affect them. For Social Security, the comparison quarter is the third calendar quarter, and the Commissioner is required to publish the resulting determination in the Federal Register within 45 days of that quarter's close. For Series I savings bonds, Treasury sets a semiannual inflation rate on May 1 and November 1 from the unadjusted all-items series.

Three different windows, three different series, three different publication dates. A household that assumes one number drives all three will find the figures disagreeing every year, and none of them will be wrong.

What it does well

• Provides one long, consistent, publicly documented series, which is what makes indexing a statute to it workable at all.
• Published monthly and on a fixed schedule, so the figures that depend on it arrive predictably.
• Comes in versions matched to different populations and purposes rather than forcing one measure to serve every use.
• Is a real constraint on discretion. Once a formula points at a series and a window, an annual adjustment stops being an annual decision.

Its documented limits

• Every version describes an average for a defined population, so no version describes any particular household's prices.
• CPI-W, which sets the Social Security increase, covers only urban households where more than half of income comes from clerical or wage occupations and at least one earner was employed for at least 37 weeks in the previous 12 months, so a fully retired household sits outside its population by construction.
• Chained CPI has risen more slowly than the traditional measure every year since 2017, so figures indexed to it sit lower over time than the same figures indexed the old way would have.
• The tax figures are locked to an unrevised initial chained value and are never corrected when the index itself is later revised.
• The August 31 window means several months of recent price movement are outside next year's federal tax figures by construction.