Does the method estimate a different seasonal pattern every year?

The classical procedure repeats one estimated pattern across cycles.

Why do endpoint values disappear?

Centered moving averages require observations on both sides of the estimated point.

Is the remainder necessarily random?

No. It can retain omitted structure, outliers or changes the procedure does not capture.

Can the basic multiplicative version handle zeros without care?

Division and relative-factor interpretation require suitable data; zeros and negative values can make that form inappropriate.

AI Glossary / Time series and signal processing / Intermediate

Classical decomposition method

Definition

The classical decomposition method separates a time series into estimated trend-cycle, seasonal and remainder components using moving averages and seasonal averaging. In the additive form, the components add to the observed value; in the multiplicative form, they multiply. The classical method assumes a seasonal pattern that repeats with the same structure across cycles.

What it means

An operational series may combine gradual movement, recurring seasonal effects and irregular variation. Decomposition helps display those features separately so users can ask clearer questions about the observed changes. For example, a high quarterly value may reflect the usual seasonal pattern rather than a new rise in underlying activity.

The components are model-based estimates, not directly observed causes. The trend-cycle combines longer-term movement rather than necessarily identifying a separate economic cycle. The remainder includes what the procedure leaves behind and should not automatically be called independent noise. Classical decomposition is also a description of historical data, not a complete forecasting model on its own.

How it works

Specify the seasonal period m. Estimate the trend-cycle with an m-period moving average when m is odd, or a centered two-by-m moving average when m is even. Detrend by subtracting the trend-cycle for the additive form or dividing by it for the multiplicative form.

Average the detrended observations for each seasonal position and normalize the seasonal values. Additive seasonal effects sum to zero across a cycle; multiplicative factors average one. Repeat the estimated seasonal pattern through the series, then calculate the remainder by subtraction or division. Standard centered moving averages leave the first and last observations without trend estimates. Any endpoint extrapolation or other modification must be identified rather than presented as the unmodified procedure.

Three practical examples

1. Quarterly reporting: A planner separates recurring seasonal variation from a stable underlying level to explain why quarterly totals differ even without a new business development.

2. Forecast preparation: An analyst examines a seasonally adjusted series before choosing a forecasting model, while checking whether the fixed seasonal assumption is credible.

3. Data investigation: A team studies large remainder values after decomposition to locate unusual events or possible recording errors, without assuming that every remainder is an anomaly requiring removal.

Think of it this way

Think of separating a repeated decorative pattern from the broad shape of a fabric and the small irregularities left over. The method estimates each layer according to a rule. If the repeated pattern itself evolves, forcing one fixed pattern onto the whole fabric can push meaningful changes into the remainder.

Worked example

Use an original quarterly additive example with twelve values: 90, 100, 110, 100, repeated three times. The seasonal period is four. Every four-quarter moving average equals 100. Centering adjacent four-quarter averages therefore gives trend-cycle estimates of 100 for periods 3 through 10.

The available detrended values follow seasonal positions: first-quarter effects are -10, second-quarter effects 0, third-quarter effects +10 and fourth-quarter effects 0. Averaging by quarter yields those same four effects, and their sum is already zero. No further centering adjustment is needed.

At period 7, a third quarter, the observed value is 110. The estimated remainder is 110 - 100 - 10 = 0. All other periods with an available trend estimate also have zero remainders. The standard centered calculation does not provide trend or remainder estimates for periods 1, 2, 11 and 12. This deliberately exact pattern demonstrates the algorithm; real data rarely separate this neatly, and it does not prove that the pattern will continue.

Illustrative case study

Fictional case: Juniper Services uses classical decomposition to explain quarterly customer-contact volumes. Initially, the seasonal pattern is stable enough to support a clear historical presentation. Later, a new subscription policy changes when customers contact the service, shifting the seasonal shape.

The analytics team notices systematic remainder patterns and compares the estimated seasonality across shorter periods. It moves to a method that can accommodate changing seasonal behavior and labels the resulting components clearly. Managers learn that the original decomposition was useful for its historical window but should not be treated as a permanent seasonal rule for the business.

Common misconceptions and mistakes

Classical decomposition is not the same as every method called seasonal decomposition. STL and other procedures use different smoothing choices and can allow different behavior. Additive and multiplicative seasonal components also have different interpretations: units versus relative factors.

Do not ignore missing endpoint estimates or silently fill them with values from an unrelated procedure. A zero average seasonal effect does not imply that individual seasons have no effect. Centered decomposition uses later observations to estimate historical components, so applying it across an entire dataset before time-based validation can leak future information. It is also sensitive to unusual observations and can oversmooth abrupt changes.

Business use

The method offers an accessible explanation of trend and seasonality for reporting and model exploration. It can be a useful teaching tool and a simple baseline when seasonal effects are reasonably stable.

For operational use, check whether seasonal magnitude is approximately constant in units or proportional to the series level. Assess whether the pattern changes over time and whether endpoints are central to the decision. If the task is real-time forecasting, use a procedure evaluated with only information available at each forecast date. Historical decomposition plots should not be mistaken for forecasts that could have been issued at the time.

Check your understanding

What is the example's trend estimate at period seven?

Answer: It is 100.

What is its third-quarter seasonal effect?

Answer: It is +10 units.

Why is period twelve's standard trend estimate unavailable?

Answer: The centered moving average would need future observations beyond the series endpoint.

Related terms

Seasonal adjustment
Moving average
Seasonal index
trend-cycle
additive decomposition
multiplicative decomposition
STL
remainder

Learn next

Learn next how moving-average centering works and how STL differs from the classical method. Then examine how decomposition can be embedded in a forecasting workflow without using future observations prematurely.

Moving average

Sources

Hyndman and Athanasopoulos, Forecasting: Principles and Practice, Classical decomposition: https://otexts.com/fpp2/classical-decomposition.html . The author reference states the additive and multiplicative procedures and endpoint limitations. The quarterly sequence and component calculations here are original and independently checked; no endpoint extension is assumed.

Previous: Class interval Next: Classical planning