Three numbers determine what roughly 71 million Americans get paid next year, and two of them already exist. July’s CPI-W reading came out on August 12, up 3.4% from a year earlier. August’s arrives in mid-September. September’s lands on October 13, and the Social Security Administration announces the 2027 cost-of-living adjustment the following day. That is the entire calculation. How the Social Security COLA is calculated is genuinely simple arithmetic, and the simplicity is the interesting part, because the formula does something narrower than most people assume it does.
The COLA measures three months, and none of them are next year
The Social Security Administration takes the Consumer Price Index for Urban Wage Earners and Clerical Workers, averages the July, August and September readings, and compares that average against the same three-month average from the prior year. The percentage change, rounded to the nearest tenth of a percent, is the COLA. It becomes effective in December and shows up in checks starting in January.
That window closes on September 30 and then governs payments for the twelve months of the following calendar year. By December of the benefit year, you are being compensated for prices that were measured fifteen months earlier. This is not a flaw in the design so much as the necessary consequence of using measured data instead of a forecast. The government has decided it would rather be accurate about the past than speculative about the future. When prices are accelerating, that choice costs beneficiaries money all year. When prices are cooling, as they appear to be now, it works in their favor.
The comparison year is not always last year
Here is the piece that even careful explanations skip. The COLA is not simply this year’s third quarter against last year’s third quarter. It is measured against the highest third-quarter average that has ever produced a COLA.
That distinction is invisible in ordinary years and decisive in strange ones. Benefits cannot fall, so when the index declines, the formula produces a zero rather than a cut. But the baseline does not reset downward with it. The old high stays locked in as the comparison point, and the next COLA only appears once the index climbs back above that peak. Beneficiaries in years like 2010, 2011 and 2016 received nothing, then received an adjustment measured from the previous high water mark rather than from the trough.
The result is a ratchet that protects the nominal check and quietly costs a little on the way back up. You never take a pay cut. You also never get credit for the recovery.
How the Social Security COLA is calculated says a lot about who it was built for
The W in CPI-W stands for wage earners. The index tracks the spending of urban wage earners and clerical workers, a working-age population, and its basket is weighted accordingly. Retirees are not the people it was designed to measure.
The Bureau of Labor Statistics has published an alternative since 2008, calculated back to 1982, called the CPI-E. The E is for elderly. The formula is identical; only the expenditure weights change. Medical care, housing and recreation carry more weight, while food, apparel, transportation and education carry less. Because health care prices have generally risen faster than the overall basket, the CPI-E has usually grown faster than the CPI-W. Social Security’s actuaries estimate the switch would add about 0.2 percentage points to the average COLA.
There is a real argument on the other side, and the Center for Retirement Research at Boston College has made it. The CPI-E is built on a much smaller sample, which makes it noisier. Its population is not the beneficiary population either: more than a fifth of Social Security beneficiaries are under 62, and more than a fifth of people over 62 collect nothing. Swapping the index is a permanent structural change made on thin data, and it would not touch the timing lag at all. So the CPI-W survives less because it describes retirees well than because it is measured well, which is a trade the formula makes quietly and permanently.
Medicare runs on its own calendar and gets paid first
Follow a single year all the way to the bank account and the gap between the announced number and the felt one stops being mysterious.
The Social Security Administration announced the 2026 COLA on October 24, 2025, at 2.8%. For the average retired worker that meant a monthly check rising by about $56, from roughly $2,015 to $2,071. Multiply by twelve and the headline promises $672 more over the year.
Then the other envelope arrives. The Centers for Medicare and Medicaid Services set the standard Part B premium for 2026 at $202.90, up $17.90 from $185.00 in 2025. Most retirees have that premium deducted straight out of the Social Security payment, before the deposit ever hits the bank. So the $56 raise becomes $38.10 in actual deposited dollars, and the $672 becomes $457.20. About 32% of the COLA went to Medicare before anyone made a spending decision. The Part B deductible moved too, from $257 to $283.
The hold-harmless provision exists for exactly this collision. In any year when the dollar increase in your Part B premium would exceed your dollar COLA, the premium increase is capped so your net check cannot shrink. That is a floor, not a refund. It stops the arithmetic from going negative and does nothing about a raise that gets mostly consumed.
A fully indexed benefit can still lose ground
Put the pieces together and the benefit is indexed in a specific, limited sense: adjusted for measured past inflation, in a basket built around working households, delivered on a lag, then reduced by a health premium set through a separate process on a separate calendar. The part that compounds is easy to miss. Each COLA is applied to the benefit you are already receiving, so a year in which the lag understates what prices actually did is not repaid later. It lowers the base that every subsequent adjustment is calculated from, and the shortfall carries forward for as long as the benefit is paid.
For 2027, the Senior Citizens League currently projects a 3.6% adjustment, which would be the largest in four years. July’s CPI-W came in at 3.4%, slightly below that projection, so the group is assuming a modest reacceleration across August and September. On a $2,071 check, 3.6% would be roughly $75 a month. What that is actually worth will not be knowable until the 2027 Part B premium is published this fall, and that number tends to arrive after the COLA headline has already faded.
The practical takeaway is not to do anything on October 14. It is to read the announcement as a gross figure rather than a net one, and to wait for the Medicare number before adjusting a budget around it. If you want a point of comparison for how differently inflation protection can be engineered, the I bond fixed rate resets on a six-month schedule using a different index entirely, and the mechanics of an out-of-pocket maximum show how another federal formula caps a health cost in advance instead of adjusting for it afterward.