AI Glossary / Time series and signal processing / Intermediate
X-11 decomposition
Definition
X-11 decomposition is a seasonal-adjustment method that uses successive moving-average filters to separate a time series into trend-cycle, seasonal, and irregular components. Developed within the US Census Bureau's statistical-methods tradition, it remains available through later software such as X-13ARIMA-SEATS. It is principally used for monthly or quarterly series, not as a general solution for every seasonal frequency.
What it means
A business series can rise in December because its usual seasonal pattern is strong, even while underlying activity weakens. X-11 helps separate these movements so analysts can compare adjacent periods more meaningfully. It estimates components rather than observing a uniquely determined seasonal effect directly.
The name identifies a decomposition procedure, while modern software may surround that procedure with regression adjustments, ARIMA modeling, diagnostics, and revision analysis. Distinguish the X-11 filtering step from the full configured workflow. Also distinguish seasonally adjusted values from the trend-cycle: the adjusted series generally still contains irregular movements.
How it works
The procedure alternates estimates of trend-cycle and seasonality through several filtering passes. Removing an initial trend estimate helps reveal seasonal movements; estimating those movements permits a revised seasonally adjusted series and improved trend estimate. Different filters serve different components, and the seasonal pattern can evolve gradually over time.
An additive representation writes y_t = T_t + S_t + I_t. A multiplicative representation writes y_t = T_t times S_t times I_t, typically for positive series whose seasonal variation scales with their level. Seasonal adjustment subtracts S_t in the first case and divides by S_t in the second.
Production workflows may account for trading days, holidays, unusual observations, and extensions beyond the observed endpoints. Choices affect the result. Read diagnostics and record the specification; automation does not remove the need to assess whether adjustment is appropriate. New data can revise earlier estimated components, particularly near the end of the series.
Three practical examples
- A retailer adjusts monthly sales before comparing adjacent months, while retaining unadjusted sales for cash-flow reconciliation and seasonal staffing needs.
- A manufacturer examines quarterly production with a decomposition that separates recurring seasonal movement from broader changes, investigating factory closures as potentially unusual events.
- An AI analyst summarizes official X-11-based releases using their documented adjustment and revision policies, rather than interpreting every monthly rise as evidence of a new growth trend.
Think of it this way
Think of repeatedly separating strands in a braided rope. An initial separation helps expose another strand, and subsequent passes refine the result. The analogy captures iterative separation, but the real method relies on specified filters and assumptions; it does not reveal one indisputable decomposition hidden inside the observations.
Worked example
Consider an original interpretation exercise with hypothetical multiplicative components already estimated by a decomposition workflow. Month A has observed sales y_A = 132, seasonal factor S_A = 1.20, and trend-cycle T_A = 100. Its seasonally adjusted value is 132/1.20 = 110. Its irregular factor is 132/(100 times 1.20) = 1.10. Reconstruction gives 100 times 1.20 times 1.10 = 132.
Month B has y_B = 126, S_B = 1.05, and T_B = 120. Its adjusted value is 126/1.05 = 120, and its irregular factor is 1. The original series falls by (126 - 132)/132, approximately 4.55%, while the adjusted series rises by (120 - 110)/110, approximately 9.09%.
These different directions are possible because the seasonal factors differ. They do not establish why underlying demand changed. The example checks component arithmetic and interpretation only: two observations cannot estimate a credible X-11 seasonal pattern, and no X-11 software run or filter estimation is claimed. Factors centered around one are multipliers; subtracting 1.20 from 132 would apply the wrong operation.
Illustrative case study
Fictional case: AlderSupply builds an AI briefing around its monthly orders. A holiday month appears unusually strong, and the first narrative recommends permanent capacity expansion. The analyst asks for the original series, the seasonally adjusted series, and the trend-cycle estimate from the documented adjustment workflow.
The team also checks whether a newly introduced promotion is being treated sensibly and whether recent estimates are stable across releases. It rewrites the briefing to distinguish recurring seasonal demand, irregular variation, and the estimated trend. The case describes an analytical safeguard, not a measured financial benefit or a completed real-world X-11 implementation.
Common misconceptions and mistakes
X-11 is not identical to a single moving average, nor does it assume that each month's seasonal factor remains fixed forever. It does not automatically remove every unusual event or resolve all structural breaks.
The historical glossary's description of X-12 as the latest successor should not be treated as current software guidance. X-13ARIMA-SEATS includes an X-11 procedure alongside a distinct model-based SEATS option. Decomposition is not itself a full forecasting exercise, and seasonally adjusted changes are not causal estimates. Do not apply a monthly adjustment setup to daily observations without an appropriate method for their seasonal structure.
Business use
Use X-11 when a documented monthly or quarterly seasonal-adjustment workflow suits the data and reporting purpose. It can support comparisons across adjacent periods, official-statistics interpretation, and monitoring of recurring business activity.
Store raw observations, configuration, software version, and release date with the output. Compare revisions and residual seasonality before relying on the adjusted figures. For operational forecasting, retain the seasonal component: customers still arrive during seasonal peaks even when an analytical chart has removed that pattern. Evaluate forecast accuracy separately using a time-respecting test.
People also ask
Is X-11 obsolete?
Its historical standalone forms have successors, but the procedure remains available within modern software.
Does it always produce an additive decomposition?
No; additive and multiplicative forms are available.
Does seasonal adjustment equal trend estimation?
No. The adjusted series generally retains an irregular component as well as the trend-cycle.
Check your understanding
What is Month A's adjusted value?
Answer: 110, obtained by division by the seasonal multiplier.
Why can original and adjusted growth have opposite signs?
Answer: Changes in estimated seasonal effects alter the comparison.
Does the worked exercise demonstrate X-11 estimation?
Answer: No. It interprets hypothetical components and checks their arithmetic; estimation requires an adequate historical series and a specified procedure.
Related terms
- Seasonal adjustment
- trend-cycle
- moving-average filter
- X-13ARIMA-SEATS
- SEATS
- trading-day adjustment
- revision analysis
Learn next
Study additive versus multiplicative decomposition, then examine a documented monthly dataset in seasonal-adjustment software. Compare component reconstruction, residual seasonality, and successive data vintages. Finally, investigate how regression adjustments and endpoint extensions interact with the filtering procedure before selecting production settings.
More in Time series and signal processing
- Tracking signal - A tracking signal is a forecast-monitoring statistic highlighting persistent errors that may indicate bias or a changed forecasting environment.
- Transfer function - A transfer function describes how an input is transformed into an output by a specified linear system, including delayed effects.
- Trend analysis - Trend analysis examines how the underlying level or direction of a time series changes over a specified period.
- Trend chart - A trend chart displays a measured quantity in time or observation order so that changes, recurring patterns and unusual values can be inspected.
- Trend series - A trend series is an estimated time series that shows underlying movement after seasonal variation and irregular fluctuations have been reduced.
- White noise - White noise is a time-series process with a constant finite variance and no correlation between observations at different times.
Sources
Supplied historical glossary: https://robjhyndman.com/mwh3/FG4.pdf . X-11 entry consulted; its successor wording is historical. US Census Bureau, X-13ARIMA-SEATS: https://www.census.gov/data/software/x13as.html . Current program's X-11 and SEATS capabilities consulted. Hyndman and Athanasopoulos, decomposition chapter: https://otexts.com/fpp3/decomposition.html . X-11 overview and component relationships read. Direct legacy X-11 chapter URL failed; the consolidated chapter was available. Numerical exercise and case are original; no software implementation is claimed.
Previous: WYSIWYM (interaction technique) Next: X-bar-R chart