Can an index be below one?

Yes; it indicates a below-baseline contribution in the multiplicative convention.

Should observations be divided or multiplied for adjustment?

Divide by multiplicative factors.

Are additive components dimensionless?

No; they use the original series' units.

AI Glossary / Time series and signal processing / Advanced

Seasonal index

Definition

A seasonal index summarizes the estimated contribution of a particular position in a recurring seasonal cycle. In a multiplicative model it is a factor, usually centered around one, that scales a nonseasonal level. In an additive model the corresponding seasonal component is expressed in the original units and is usually centered around zero.

What it means

A multiplicative index of 1.35 represents a seasonal contribution 35% above the model's nonseasonal baseline. An index of 0.80 represents a contribution 20% below it.

Seasonal indices describe a fitted or estimated pattern, not a guarantee about an individual future observation. Irregular variation, changes in level and a changing seasonal pattern can all affect what actually occurs.

How it works

Estimate seasonal contributions while accounting for the series' nonseasonal movement. One simple initialization method divides each observation by its own year's average, then averages the ratios for matching seasons across years. More sophisticated decomposition methods allow for trend and changing seasonality.

For multiplicative adjustment, divide an observation by its seasonal index. To restore seasonality to a forecast of the nonseasonal component, multiply by the appropriate index. Additive components instead use subtraction and addition.

Keep seasonal positions aligned. A third-quarter observation must be compared with third-quarter observations when estimating its index, not with whichever row happens to be nearby. Check the normalization convention and units before using a supplied set of indices.

Three practical examples

1. Capacity planning: convert a baseline workload forecast into forecasts for the expected busy and quiet parts of the year.

2. Performance comparison: adjust observations for estimated seasonal contributions before comparing nonseasonal movements.

3. AI forecast review: inspect whether seasonal multipliers are applied to the correct future months or quarters, using the intended scale.

Think of it this way

Think of a set of seasonal volume controls. A factor above one turns the baseline upward, a factor below one turns it downward and a factor of one leaves it unchanged.

Worked example

Consider invented quarterly activity observations for two complete years:

Year 1: [80,100,140,80].

Year 2: [108,120,156,96].

The first year's total is 400, so its quarterly mean is 100. The second year's total is 480, so its quarterly mean is 120. Use the simple year-relative initialization method for illustration, without claiming that two years establish a stable seasonal pattern.

Divide each observation by the mean for its own year. Year 1 produces ratios [0.80,1.00,1.40,0.80]. Year 2 produces [0.90,1.00,1.30,0.80]. Dividing every observation by the first year's mean would incorrectly mix the difference in annual levels into the second year's ratios.

Average corresponding quarterly ratios. The first-quarter index is (0.80 + 0.90)/2 = 0.85. The second is (1.00 + 1.00)/2 = 1.00. The third is (1.40 + 1.30)/2 = 1.35. The fourth is (0.80 + 0.80)/2 = 0.80.

These four indices sum to 4 and have mean 1. In this complete, equally weighted construction, that normalization follows from dividing each year's four observations by their annual mean. It is not an additional piece of evidence that the indices are accurate or that their pattern will continue.

On a base-100 display, the same indices are [85,100,135,80]. The number 135 means a multiplier of 1.35, not a multiplier of 135. It also does not mean growth of 135% from the preceding quarter.

Suppose a separate model supplies a flat nonseasonal forecast of 200 for each quarter of the next year. Restoring seasonality gives forecasts of 200 × 0.85 = 170, 200 × 1.00 = 200, 200 × 1.35 = 270 and 200 × 0.80 = 160. Their total is 800, matching four times the flat baseline because the factors sum to four.

If the nonseasonal forecasts differ by quarter, that total-preservation result need not hold. For example, baseline forecasts [200,210,220,230] total 860. Multiplying by the same indices gives [170,210,297,184], which total 861. A mean-one set of factors does not preserve every total when the quantities being multiplied differ.

For adjustment in the other direction, divide year-1 quarter 3 by its estimated index: 140/1.35 is approximately 103.70. Year-2 quarter 3 gives 156/1.35, approximately 115.56. The adjusted observations need not equal their annual means of 100 and 120 because each year's third-quarter ratio differs from the two-year average ratio.

The index is dimensionless, while the adjusted result retains the activity index's units. Dividing by 135 instead of 1.35 would give approximately 1.04 for the first adjusted observation, an error by a factor of 100.

For comparison, an additive seasonal component of 35 points would be added directly to the baseline. At baseline 200 it would produce 235, not 270. A multiplicative lift of 35% and an additive lift of 35 points coincide only at a baseline of 100; they are not interchangeable descriptions.

All observations and future baseline forecasts are invented. The arithmetic illustrates estimation conventions and use of indices, not a fitted operational forecast or a causal explanation for seasonal demand.

Illustrative case study

Fictional Cedar Services receives a quarterly file labeled 85, 100, 135 and 80. A spreadsheet treats those numbers as direct multipliers and produces forecasts one hundred times too large.

The analyst checks the stated base-100 convention, converts the numbers to factors and verifies the example's forecasts. Cedar also reviews whether the two-year pattern is appropriate for future planning. This fictional case illustrates a scaling check, not evidence of forecast accuracy.

Common misconceptions and mistakes

"An index of 135 always means multiply by 135." It may be a base-100 representation of 1.35.

"Mean-one factors always preserve annual totals." That requires additional conditions, as the varying-baseline example shows.

"A seasonal factor is a year-over-year growth rate." It describes a seasonal contribution relative to a baseline.

"Estimated indices are permanently fixed." They may be updated.

Business use

Label factors and base-100 indices explicitly. Record their estimation period, seasonal positions and whether they apply multiplicatively or additively. Validate forecasts after seasonality is restored, and investigate changes in the seasonal pattern instead of assuming historical factors remain suitable indefinitely.

Check your understanding

What is the third-quarter factor?

Answer: 1.35.

What is its base-100 representation?

Answer: 135.

What does it do to a baseline of 200?

Answer: It produces 270.

Related terms

Seasonal effect
Seasonal adjustment
Baseline
Normalization
Seasonal factor
multiplicative decomposition
additive decomposition
Holt-Winters method

Learn next

Use the same four factors with a flat baseline of 240. Calculate the quarterly forecasts and their total, then explain why the total equals four times the baseline in this specific calculation.

Baseline

Sources

NIST/SEMATECH, e-Handbook of Statistical Methods, Triple Exponential Smoothing, seasonal-index initialization: https://www.itl.nist.gov/div898/handbook/pmc/section4/pmc435.htm

Rob J Hyndman and George Athanasopoulos, Forecasting: Principles and Practice, seasonal components and Holt-Winters methods: https://otexts.com/fpp3/exponential-smoothing.html

Previous: Seasonal exponential smoothing Next: Seasonal subseries plot