Does adjustment remove inflation?

Not automatically; inflation adjustment is a separate operation.

Are seasonal factors permanently fixed?

Not necessarily; estimates can be updated.

AI Glossary / Time series and signal processing / Advanced

Seasonal adjustment

Definition

Seasonal adjustment estimates and removes recurring seasonal effects from a time series so that nonseasonal movements are easier to examine. Depending on the procedure, related calendar effects may also be adjusted. The resulting series still contains irregular variation and should not be confused with a smooth estimate of the underlying trend.

What it means

A seasonal peak can make activity look stronger even when performance relative to the usual seasonal pattern has weakened. Adjustment provides another view of the observations, rather than replacing the original measurements with a uniquely correct set of numbers.

Seasonal effects are estimated. Their size can change, and unusual events can complicate their separation from other movements. Retain the original series, the adjustment method and the data version so that readers can understand how the adjusted values were produced.

How it works

In an additive decomposition, write an observation as y = T + S + R, where T is trend-cycle, S is seasonality and R is the remainder. Subtract S to obtain the adjusted value T + R.

In a multiplicative decomposition, y = T × S × R, so divide by S. Additive effects use the series' units; multiplicative factors are ratios. The choice must reflect the data and model, rather than whichever calculation gives a preferred result.

Estimate components from an appropriate history and review adjustment diagnostics. Procedures such as X-13ARIMA-SEATS provide modeling and diagnostic options. A list of manually chosen factors can illustrate the arithmetic, but it does not demonstrate that the factors have been estimated reliably.

Three practical examples

1. Business monitoring: compare successive periods of customer activity after accounting for a recurring seasonal pattern.

2. Workforce analysis: examine changes in employment without mistaking an expected annual hiring cycle for a new direction.

3. AI forecasting preparation: separate seasonal and nonseasonal behavior while keeping the adjustments available for reconstructing forecasts on the original scale.

Think of it this way

Think of judging a shop's performance against its usual busy and quiet periods. A large holiday total might be normal for that season. Removing the estimated seasonal lift helps assess the remaining movement, but it does not explain every unusual sale or make demand perfectly predictable.

Worked example

Consider an invented quarterly activity index with observed values 80, 110, 150 and 100. Suppose a separately established additive model supplies seasonal effects of -20, 0, 30 and -10 for the four quarters. These effects are stipulated for teaching; four observations alone are not presented as sufficient evidence to estimate a recurring annual pattern.

Subtract each seasonal effect from its corresponding observation:

Quarter 1: 80 - (-20) = 100.

Quarter 2: 110 - 0 = 110.

Quarter 3: 150 - 30 = 120.

Quarter 4: 100 - (-10) = 110.

The adjusted series is therefore [100,110,120,110]. Subtracting a negative seasonal effect raises the adjusted value. This makes sense: a period expected to be seasonally quiet is compared after removing that downward contribution.

From quarter 2 to quarter 3, the original index rises by 40, or 40/110 × 100 = approximately 36.36%. The adjusted index rises by 10, or 10/110 × 100 = approximately 9.09%. Both calculations are arithmetically correct, but they describe different series. Reporting 36.36% as growth after seasonal adjustment would mislabel the result.

Now suppose the model's trend-cycle estimates are [100,105,110,115]. The remainders are adjusted value minus trend-cycle: [0,5,10,-5]. Reconstruct quarter 3 as trend 110 plus seasonal effect 30 plus remainder 10, giving the observed 150. Reconstruct quarter 4 as 115 - 10 - 5 = 100.

The adjusted series falls from 120 to 110 between quarters 3 and 4, a decline of about 8.33%. Yet the stipulated trend-cycle rises from 110 to 115. There is no arithmetic contradiction: the remainder changes from 10 to -5. This illustrates why a single decline in adjusted observations does not, by itself, establish a declining trend.

The four seasonal effects sum to zero, so in this particular additive example both original and adjusted annual totals equal 440. This follows from summing y - S: if the seasonal effects sum to zero, the totals agree. Do not generalize that identity to every method, changing seasonal pattern or multiplicative adjustment.

For a separate multiplicative illustration, suppose an observation is 132 and its seasonal factor is 1.2. Its adjusted value is 132/1.2 = 110. Subtracting 1.2 would produce 130.8 and would apply the wrong operation. The factor says that the seasonal contribution multiplies the nonseasonal value by 1.2; it does not represent 1.2 index points.

If that factor is later revised to 1.1 while the original observation remains 132, the adjusted value becomes 120. The revised adjusted number need not mean that the original observation changed. It can reflect a different estimate of the seasonal contribution. These invented calculations illustrate bookkeeping and interpretation, not the accuracy of any fitted adjustment procedure.

Illustrative case study

Fictional Willow Services initially describes its third-quarter index increase as a major improvement. An analyst presents both the original and adjusted changes from the example and identifies the assumptions behind the factors.

Willow keeps the observed totals for operational reporting and uses the adjusted view to discuss recent movement. It also checks the trend-cycle and remainder before declaring a turning point. The case is fictional and does not establish that adjustment alone improves decisions.

Common misconceptions and mistakes

"Seasonal adjustment removes all noise." The remainder is retained.

"Adjusted data equal the trend." They need not.

"Every factor should be subtracted." Multiplicative factors are divisors.

"Revised adjusted values prove the original observations were wrong." Seasonal estimates can change.

Business use

Label original, adjusted and trend series clearly. Record the method and release date, and explain significant revisions. In historical forecasting evaluations, avoid using adjustments that depend on observations unavailable at the forecast date. When planning actual staffing or inventory, account for the seasonal demand that will still occur.

Check your understanding

Why does quarter 1 adjust upward?

Answer: Its negative seasonal contribution is removed.

What remains after additive adjustment?

Answer: Trend-cycle plus remainder.

What is 132 adjusted by a multiplicative factor of 1.2?

Answer: 110.

Related terms

Seasonality
time series decomposition
trend-cycle
remainder
calendar adjustment
X-13ARIMA-SEATS
STL
data revision

Learn next

Change quarter 3's observed value from 150 to 140, keeping its seasonal effect and trend-cycle fixed. Recalculate its adjusted value, remainder and adjusted growth from quarter 2. Explain which model assumptions this exercise holds constant.

Seasonal effect

Sources

Rob J Hyndman and George Athanasopoulos, Forecasting: Principles and Practice, Time series decomposition: https://otexts.com/fpp3/decomposition.html

U.S. Census Bureau, Seasonal Adjustment Questions and Answers: https://www.census.gov/data/software/x13as/seasonal-adjustment-questions-answers.html

U.S. Census Bureau, X-13ARIMA-SEATS Seasonal Adjustment Program: https://www.census.gov/data/software/x13as.html

Previous: Search tree Next: Seasonal Autoregressive Integrated Moving-Average