How the Social Security COLA is calculated, and why the official projection is already behind the data

Buried in the definitions at the top of section 215(i) of the Social Security Act is a sentence that most explanations of the cost-of-living adjustment skip. It says the Consumer Price Index for a quarter "shall be the arithmetical mean of such index for the 3 months in such quarter." That is the whole method. Three monthly numbers, averaged, compared against an earlier average, rounded to a tenth of a point.

No committee votes on it. No commissioner has discretion over it. The statute does the work, and the only inputs are prices that the Bureau of Labor Statistics has already published or has not yet published.

Right now, for the raise that lands in January 2027, it has not published any of them.

How is the Social Security COLA calculated?

The adjustment equals the percentage increase in the Consumer Price Index for Urban Wage Earners and Clerical Workers, the CPI-W, from the third-quarter average of the last year in which an adjustment took effect to the third-quarter average of the current year, rounded to the nearest tenth of one percent. If there is no increase, or the rounded increase is zero, there is no adjustment that year.

The Social Security Administration publishes the working for the current one, and it is worth reproducing because it is unusually easy to check. The computation uses the third quarter of 2024 as the base, because 2024 was the last year in which an adjustment became effective before the current one.

Take the two averages. Subtract, divide, multiply by a hundred, and the answer is 2.7649 percent, which rounds to 2.8. Every one of those six monthly figures can be pulled independently from the BLS series CWUR0000SA0, and they match the agency's table to the third decimal place. We checked them against the source rather than against the agency's summary of the source, which is a habit worth keeping on this beat.

That 2.8 percent became effective with December 2025 benefits, payable in January 2026, and it went to roughly 75 million people. The average retired worker's payment rose from $2,015 to $2,071 a month, about $56.

Which quarter is the raise measured against?

Not, as almost every explainer has it, the third quarter of last year. The measurement is against the third quarter of the most recent year in which an adjustment actually became effective, which the statute calls a cost-of-living computation quarter. In an ordinary year the two definitions give the same answer. In the year after a zero, they do not.

This matters more often than the obscurity of the drafting suggests. The adjustment was zero in 2009, zero in 2010 and zero again in 2015. In each of the years that followed, the base quarter stayed put rather than resetting to the intervening year, which meant beneficiaries were not permanently penalized for a period of falling prices. Prices had to climb back above the old high-water mark before anything was paid, and once they did, the measurement ran from that older peak.

The rule cuts both ways. It protects the recipient from a benefit cut, since the law provides no mechanism to reduce a payment when prices fall. It also delays the recovery, because the first year of renewed inflation is spent climbing back to a level already reached.

Why is a 2.8 percent raise called the 2026 adjustment when the agency files it under 2025?

Because the law dates the adjustment by the December in which it becomes effective, and the press dates it by the January in which the money appears. The Social Security Administration's own historical table lists 2.8 against the year 2025. The press release announcing the same number is titled "Social Security Announces 2.8 Percent Benefit Increase for 2026."

Both are correct. The confusion is real and it is worth naming, because it is the reason two apparently authoritative sources will hand you different numbers for the same year. The Trustees Report uses the statutory convention. Consumer coverage uses the calendar one. When a table says the 2026 adjustment is 2.7 percent, check which December it means.

When is the next adjustment decided, and by what?

By three price readings and nothing else. The base is now fixed at 317.265, the third-quarter 2025 average, because an adjustment became effective in 2025. The comparison quarter is July, August and September 2026, and the BLS release calendar sets the dates precisely.

Nothing before 12 August is data. Everything published in the meantime, including every estimate circulating now, is extrapolation from months that do not count.

What can be said today is what each outcome requires. Because the result is rounded to a tenth of a point, each possible adjustment corresponds to a band the quarterly average has to land inside. Measured against the 317.265 base, those bands are:

The last published CPI-W reading, for June 2026, is 327.075. Hold the index flat at that level for all three months and the adjustment is 3.092 percent, which rounds to 3.1.

What do the Trustees project, and why is it lower?

The 2026 Trustees Report puts the adjustment effective for December 2026 at 2.7 percent under its intermediate assumptions, 2.9 percent under low-cost and 2.3 percent under high-cost. From 2027 onward the intermediate assumption settles at a flat 2.4 percent a year, which is a modelling convention rather than a forecast of any particular year.

The gap between 2.7 percent and what the data now imply has a documented explanation, and the report supplies it without anyone needing to speculate. The report states plainly that the intermediate assumptions "were set in February 2026," and that actual economic data "were generally available through the third quarter of 2025 at the time the assumptions for this report were set."

So the 2.7 percent was fixed before the spring. Since then the CPI-W has gone from 317.014 in December 2025 to 327.075 in June, a rise of 3.2 percent in six months. For the intermediate assumption to hold, the index would have to average between 325.673 and 326.013 across the third quarter, roughly 0.3 to 0.4 percent below where it sat in June. That is not impossible. It is a decline, and the assumption was not built on one.

None of which makes the Trustees wrong. It makes them early. An annual report locks its inputs on a date, and the date was February.

Is the index about to fall?

It fell in June, which is the part of this that deserves care. The all-items index for urban consumers dropped 0.4 percent on a seasonally adjusted basis, the largest one-month decrease since April 2020, and the CPI-W itself fell 0.5 percent before seasonal adjustment. Energy did almost all of it, down 5.7 percent in the month after gains of 3.9, 3.8 and 10.9 percent in the three months before.

Twelve-month inflation still ran at 3.5 percent in June, down from 4.2 percent in May. Core inflation, excluding food and energy, was 2.6 percent and unchanged on the month.

The honest reading is that a scenario built on holding June's level flat is neither conservative nor aggressive. It is a marker, not a forecast, and the reason to state it is that it is checkable. Outside groups tracking the same data have published estimates near 3.7 and 3.8 percent, which assume the index resumes rising through the quarter, and we covered those estimates when they moved. They may well be closer than the flat case. The point of the bands above is that a reader can watch each print land and work out the answer without waiting for anyone's estimate.

How much of the raise does Medicare take back?

A meaningful share, and for most people it is deducted before the payment arrives. The standard Medicare Part B premium rose to $202.90 a month in 2026, an increase of $17.90 from $185.00, according to the Centers for Medicare and Medicaid Services. The annual Part B deductible rose $26, to $283.

Set that against the average retired worker's raise of about $56 a month. The premium increase consumed roughly 32 percent of it. A beneficiary receiving less than the average, and one in the disabled-worker population averaging $1,586 before the adjustment, saw a proportionally larger bite, because the premium increase is a flat dollar amount while the raise is a percentage.

Two qualifications matter. Roughly 8 percent of Part B enrollees pay income-related surcharges above the standard premium, and their increases were larger in dollar terms. And a hold-harmless provision, section 1839(f) of the Act, blocks a premium increase to the extent it would push a beneficiary's December payment, after the premium is deducted, below what they received in November. It protects people whose raise is smaller than the premium increase. It did not bind this year, because $56 comfortably exceeds $17.90, and it does not apply to those paying the income-related surcharge.

Does the index measure the prices retirees actually pay?

No, and this is a design feature rather than an oversight. The CPI-W is built from the spending of households where more than half of income comes from clerical or wage occupations and at least one earner worked at least 37 weeks in the previous 12 months. BLS puts that population at roughly 30 percent of the country.

Read the definition again. A household living on Social Security has no earner working 37 weeks, and no majority of income from wages. The index that sets the benefit is measured on working households, and the people receiving the benefit are, by the index's own definition, largely outside the population it measures.

The practical consequence is a weighting mismatch. Working households spend more of their money on transportation and less on medical care than retired households do. When medical prices run ahead of the general index, an adjustment keyed to wage earners understates what the retired population is facing, and when fuel prices spike, it overstates it. June was the second case: energy fell hard, and the index fell with it, on a weight heavier than a retiree's budget would carry.

An alternative already exists on paper. BLS calculates a research index called the R-CPI-E, built on the spending of Americans aged 62 and over, and the bureau states that official uses of it "have been considered by other government agencies, but not implemented due to the limitations noted below."

Those limitations are BLS's own and they are not trivial. The Consumer Expenditure survey is not designed to sample the over-62 population, so the weights come from roughly a fifth of the urban sample and carry higher sampling error. The geographic areas and the retail outlets priced are the ones chosen to represent all urban households, not older ones. The items priced are selected on the general population's spending. An index intended to track what retirees pay is assembled, at three separate stages, from a sample built to describe somebody else.

That is the honest state of the argument. The case for a retiree-weighted index is sound in principle, the instrument that would implement it is acknowledged by its own producer to be provisional, and the statutory reference remains the CPI-W. Until the text of section 215(i) changes, so does the answer.

Did the government shutdown affect any of this?

Not the current adjustment, and this is worth stating clearly because the assumption runs the other way. The 2025 lapse in appropriations stopped CPI collection from 1 October to 12 November 2025, and BLS was unable to gather that data retroactively. The October 2025 CPI-W was never published. Pull the series and there is a hole where the number should be.

The third quarter of 2025 was already complete by then. July, August and September were collected, published and final, so the 2.8 percent stood untouched. What the shutdown did move was the announcement: the September 2025 price data slipped, and the Social Security Administration announced the adjustment on 24 October rather than in the middle of the month.

The more durable effect is further out and almost nobody is tracking it. Consumer Expenditure survey collection also stopped, and BLS states that the missing October and November 2025 expenditure data affects "2027 CPI-U and CPI-W indexes." Expenditure data sets the weights inside the index, and the 2027 CPI-W is the index that will measure the adjustment effective December 2027. After expert panels convened with the National Association for Business Economics and the Committee on National Statistics in May 2026, BLS selected a survey weight adjustment approach and published the decision in June.

So a six-week gap in 2025 data collection reaches forward into the basket weights used to compute a benefit increase in 2028. The effect is probably small. It is also documented, deliberate, and disclosed in advance, which is more than can be said for most sources of error in this calculation.

The clause that can cap the raise below inflation

There is a second branch to the formula, dormant since it was written into law in 1983, and it does not appear in any of the standard explanations.

Section 215(i)(1)(C) sets out two cases. When the combined trust fund ratio is 20.0 percent or more, the applicable increase percentage is the CPI increase percentage, which is the familiar rule. When that ratio falls below 20.0 percent, the applicable increase becomes the CPI increase percentage or the wage increase percentage, "whichever is the lower." The fund ratio, defined a few lines later, is the combined balance in the two trust funds at the start of a year measured against what is expected to be paid out during it.

The 2026 Trustees Report projects that combined ratio, under intermediate assumptions, falling from 151 percent at the start of 2026 to 131, 111, 92, 74, 57, 40 and 23 percent in the years through 2033, and to 7 percent in 2034, the year the combined reserves are projected to be depleted. Twenty percent is crossed in 2034 on that path, and in 2032 under the high-cost assumptions. The Trustees' published trust fund ratio and the statutory definition are measured on closely related but not identical bases, so treat the crossing year as approximate rather than as a date.

The provision has a matching repair. If an adjustment is ever limited by the wage branch, section 215(i)(5) requires a catch-up increase in a later year when the fund ratio recovers above 32.0 percent, restoring what the compounded price-based increases would have paid.

The reason this has never fired is worth understanding, because it is also the reason it probably never will in the form its drafters imagined. The stabilizer was designed for a fund that thins gradually, giving the wage cap several years to operate before anything worse happened. On the current projection the ratio falls from 40 percent to 7 percent in two years. The trigger and the depletion arrive almost together, which leaves the mechanism very little room to do the job it was built for.

What it changes

For anyone already receiving a payment, nothing until January 2027. The 2.8 percent is fixed and in payment. The next figure is determined by three price readings on 12 August, 11 September and 14 October, and no announcement, projection or estimate before those dates changes the arithmetic.

For anyone planning against a number, use the bands rather than the headlines. The base is 317.265. Watch where the July, August and September CPI-W prints land, average them, and compare. The answer is available to a reader with a calculator on the morning of 14 October, at the same moment it is available to anyone else, which is an unusual property in financial forecasting and the direct consequence of the rule being written into statute rather than left to judgment.

And treat the 2.7 percent in the Trustees Report as what it is. Not a forecast of next year, but an assumption locked in February, on data that ended in September 2025, in a document whose purpose is a 75-year projection rather than a call on one quarter.

FAQ

How is the Social Security COLA calculated? It is the percentage increase in the CPI-W from the third-quarter average of the last year an adjustment took effect to the third-quarter average of the current year, rounded to the nearest tenth of one percent. The quarterly figure is the arithmetic mean of the three monthly index values. If the rounded increase is zero or negative, there is no adjustment.

What is the Social Security COLA for 2026? 2.8 percent, effective with December 2025 benefits and payable from January 2026. It came from a third-quarter 2025 CPI-W average of 317.265 against a third-quarter 2024 base of 308.729, an increase of 2.7649 percent before rounding.

When will the 2027 COLA be announced? The September 2026 CPI is scheduled for release on 14 October 2026 at 8:30am, and that print completes the measurement. The agency has announced on the day the September data lands: in 2025 that data was delayed to 24 October by the lapse in appropriations, and the announcement came the same day.

Why can the COLA be zero? Because the law provides for an increase when prices rise and no mechanism to reduce a benefit when they fall. If the third-quarter average does not exceed the base, the rounded increase is zero and no adjustment is paid. This happened in 2009, 2010 and 2015.

Does Medicare reduce the COLA? Not the adjustment itself, but the Part B premium is deducted from most payments before they arrive, so it reduces the net increase. The standard premium rose $17.90 in 2026, against an average retired-worker raise of about $56.

Is the COLA taxable? The increase is part of the benefit and follows the same rules as the rest of it. Whether any portion of a Social Security benefit is included in taxable income depends on combined income against the statutory thresholds, and a larger benefit can move a household across one of them.