Seasonal adjustment in economic data

Seasonal adjustment removes estimated recurring calendar patterns to help compare adjacent periods.

Compare adjacent periods more carefully

Economic activity often has recurring calendar patterns: hiring around holidays, seasonal construction and predictable shopping cycles. Seasonal adjustment estimates these patterns and removes their effect from a series to make adjacent periods easier to compare. It does not remove every irregular event or guarantee that the remaining movement has one cause.

An unadjusted value is the observation under the program’s original publication basis. An adjusted value incorporates estimated seasonal factors and sometimes other calendar treatment. Both are useful, but they answer different questions and should remain separately labeled.

Keep the basis consistent

A monthly change should use the same adjustment basis at both endpoints. Do not divide a seasonally adjusted current index by an unadjusted previous index. The resulting ratio mixes measurement conventions rather than isolating the period’s price change. This site stores CPI adjusted and unadjusted inputs as distinct variants.

Twelve-month comparisons often use unadjusted values because the same month’s seasonal position is compared across years. This is not a universal rule: a program may publish adjusted yearly growth too. Read the exact source definition rather than assuming that every headline uses the same convention.

Adjusted histories can change

Seasonal factors are estimates. Re-estimation with more observations can change historical adjusted values even when the unadjusted source observation has not changed. A revision to an old period must be distinguished from an economic change between two observation periods.

A method or population change may create another break in comparability. Household-survey population controls, establishment benchmarks and industry classifications require source-specific notes. Seasonal adjustment alone cannot make differently defined populations or industries identical. A smooth-looking chart is not proof of an uninterrupted measurement history.

Record what you used

A defensible citation includes the series, frequency, adjustment, reference period, transformation and vintage or acquired snapshot. When a producer changes an adjusted history, the archived site file retains the earlier acquired version. The first site snapshot does not supply authentic vintages from periods before acquisition.

Where the methods come from

Modern seasonal adjustment at U.S. statistical agencies traces back to the Census Bureau’s X-11 program, developed in the 1960s, which estimated seasonal factors through a sequence of moving averages. Later versions added regression and time-series modeling to handle trading-day effects, holidays and outliers. The current Census Bureau software, X-13ARIMA-SEATS, is used by the Census Bureau, BLS and many other producers.

Each program configures the software for its own series. Choices include whether seasonal effects are treated as additive or multiplicative, which outliers to model and whether to adjust for the number of trading days in a month. Two producers adjusting similar data can therefore publish slightly different adjusted figures. The method is documented in producer notes, which are worth reading when an adjusted figure looks surprising.

When revisions to seasonal factors are scheduled

Many programs re-estimate seasonal factors on a regular schedule. For the CPI, BLS recalculates seasonal factors each year with the release of January data and revises seasonally adjusted indexes for the previous five years, while the unadjusted indexes stay unchanged. The payroll survey uses concurrent seasonal adjustment, which re-estimates factors each month as new data arrive, so recent adjusted months can shift slightly with each release.

These schedules mean an analyst who saved an adjusted series a year ago may hold numbers that no longer match the producer’s current history. The difference is a methodological revision, not a change in what happened during those months. Professionals who need reproducible results either store the date each file was acquired or use archived vintages from services that preserve past releases.

Common errors with adjusted data

A common error is to treat a large unadjusted monthly change as news when it reflects a normal seasonal pattern. Retail sales rise every December and construction slows every winter; comparing November with December in unadjusted data mostly measures the calendar. The opposite error is to treat an adjusted series as if every recurring pattern had been removed perfectly. Residual seasonality can remain.

Another error is combining adjusted and unadjusted data in one calculation, or quoting an adjusted level for a single month as though it were the actual count observed. An adjusted payroll level is an estimate of employment with typical seasonal swings removed, not a headcount from that month. Label the basis in every chart title and table header, and say which basis a quoted figure uses whenever both versions are published.

Worked example: removing a holiday pattern

Suppose a hypothetical retail sales series records 1,000 in November and 1,100 in December, unadjusted. Assume multiplicative seasonal factors of 0.98 for November and 1.08 for December.

  1. Unadjusted change: 1,100 ÷ 1,000 = 1.10, an apparent 10.0% increase.
  2. Adjusted November: 1,000 ÷ 0.98 ≈ 1,020.41.
  3. Adjusted December: 1,100 ÷ 1.08 ≈ 1,018.52.
  4. Adjusted change: 1,018.52 ÷ 1,020.41 ≈ 0.99815, a decline of about 0.19% once the usual holiday boost is removed.
  5. Mixed-basis error: dividing unadjusted December by adjusted November gives 1,100 ÷ 1,020.41 ≈ 1.078, a misleading 7.8% increase that matches neither basis.

Takeaway: An unadjusted 10% gain can correspond to a small adjusted decline. Keep both endpoints on the same adjustment basis.

Frequently asked questions

What does seasonally adjusted mean?

A seasonally adjusted series has had its typical recurring calendar patterns, such as holiday hiring or summer travel, estimated and removed. This makes it easier to compare one month or quarter with the previous one. Irregular events and the underlying trend remain in the adjusted data.

Should I use seasonally adjusted or unadjusted data?

Use adjusted data to compare adjacent months or quarters. Unadjusted data are often used for comparisons with the same month a year earlier, for contracts that specify unadjusted values, and when the actual observed level matters. Whichever you choose, use the same basis at both endpoints and label it.

What software do U.S. agencies use for seasonal adjustment?

The Census Bureau develops X-13ARIMA-SEATS, which the Census Bureau, BLS and many other producers use. It combines time-series models with moving-average filters to estimate seasonal factors, trading-day effects and outliers. Each agency configures it for its own series and documents the settings in technical notes.

Why did old seasonally adjusted numbers change?

Seasonal factors are estimates that are recalculated as new data become available. When a producer re-estimates them, adjusted values for earlier periods can change even though the unadjusted observations are the same. For the CPI, BLS revises five years of adjusted data each year with the January release.

What can we help you find?

Type at least two letters to start searching.

Browse economic guides Esc to close