9/26/24, 3:58 PM 8.

3 Methods with seasonality | Forecasting: Principles and Practice (3rd ed)

8.3 Methods with seasonality

Holt (1957) and Winters (1960) extended Holt’s method to capture seasonality. The
Holt-Winters seasonal method comprises the forecast equation and three smoothing
equations — one for the level ℓt , one for the trend bt , and one for the seasonal
component st , with corresponding smoothing parameters α, β ∗ and γ . We use m to
denote the period of the seasonality, i.e., the number of seasons in a year. For
example, for quarterly data m = 4, and for monthly data m = 12.

There are two variations to this method that differ in the nature of the seasonal
component. The additive method is preferred when the seasonal variations are
roughly constant through the series, while the multiplicative method is preferred
when the seasonal variations are changing proportional to the level of the series.
With the additive method, the seasonal component is expressed in absolute terms in
the scale of the observed series, and in the level equation the series is seasonally
adjusted by subtracting the seasonal component. Within each year, the seasonal
component will add up to approximately zero. With the multiplicative method, the

https://otexts.com/fpp3/holt-winters.html 1/9
9/26/24, 3:58 PM 8.3 Methods with seasonality | Forecasting: Principles and Practice (3rd ed)

seasonal component is expressed in relative terms (percentages), and the series is

seasonally adjusted by dividing through by the seasonal component. Within each
year, the seasonal component will sum up to approximately m.

Holt-Winters’ additive method

The component form for the additive method is:

y^t+h|t = ℓt + hbt + st+h−m(k+1)

ℓt = α(yt − st−m ) + (1 − α)(ℓt−1 + bt−1 )
bt = β ∗ (ℓt − ℓt−1 ) + (1 − β ∗ )bt−1
st = γ(yt − ℓt−1 − bt−1 ) + (1 − γ)st−m ,

where k is the integer part of (h − 1)/m, which ensures that the estimates of the
seasonal indices used for forecasting come from the final year of the sample. The
level equation shows a weighted average between the seasonally adjusted
observation (yt − st−m ) and the non-seasonal forecast (ℓt−1 + bt−1 ) for time t. The
trend equation is identical to Holt’s linear method. The seasonal equation shows a
weighted average between the current seasonal index, (yt − ℓt−1 − bt−1 ), and the
seasonal index of the same season last year (i.e., m time periods ago).

The equation for the seasonal component is often expressed as

st = γ ∗ (yt − ℓt ) + (1 − γ ∗ )st−m .

If we substitute ℓt from the smoothing equation for the level of the component form
above, we get

st = γ ∗ (1 − α)(yt − ℓt−1 − bt−1 ) + [1 − γ ∗ (1 − α)]st−m ,

which is identical to the smoothing equation for the seasonal component we specify
here, with γ = γ ∗ (1 − α). The usual parameter restriction is 0 ≤ γ ∗ ≤ 1, which
translates to 0 ≤ γ ≤ 1 − α.

https://otexts.com/fpp3/holt-winters.html 2/9
9/26/24, 3:58 PM 8.3 Methods with seasonality | Forecasting: Principles and Practice (3rd ed)

Holt-Winters’ multiplicative method

The component form for the multiplicative method is:

y^t+h|t = (ℓt + hbt )st+h−m(k+1)

yt
ℓt = α + (1 − α)(ℓt−1 + bt−1 )
st−m
bt = β ∗ (ℓt − ℓt−1 ) + (1 − β ∗ )bt−1
yt
st = γ + (1 − γ)st−m .
(ℓt−1 + bt−1 )

Example: Domestic overnight trips in Australia

We apply Holt-Winters’ method with both additive and multiplicative seasonality17

to forecast quarterly visitor nights in Australia spent by domestic tourists. Figure 8.7
shows the data from 1998–2017, and the forecasts for 2018–2020. The data show an
obvious seasonal pattern, with peaks observed in the March quarter of each year,
corresponding to the Australian summer.

https://otexts.com/fpp3/holt-winters.html 3/9
9/26/24, 3:58 PM 8.3 Methods with seasonality | Forecasting: Principles and Practice (3rd ed)

aus_holidays <- tourism |>

filter(Purpose == "Holiday") |>
summarise(Trips = sum(Trips)/1e3)

fit <- aus_holidays |>

model(
additive = ETS(Trips ~ error("A") + trend("A") +
season("A")),
multiplicative = ETS(Trips ~ error("M") + trend("A") +

season("M"))
)
fc <- fit |> forecast(h = "3 years")
fc |>
autoplot(aus_holidays, level = NULL) +
labs(title="Australian domestic tourism",

y="Overnight trips (millions)") +

guides(colour = guide_legend(title = "Forecast"))

Figure 8.7: Forecasting domestic overnight trips in Australia using the Holt-Winters
method with both additive and multiplicative seasonality.

https://otexts.com/fpp3/holt-winters.html 4/9
9/26/24, 3:58 PM 8.3 Methods with seasonality | Forecasting: Principles and Practice (3rd ed)

Table 8.3: Applying Holt-Winters’ method with additive seasonality for forecasting
domestic tourism in Australia. Notice that the additive seasonal component sums to
approximately zero. The smoothing parameters are α = 0.2620, β ∗ = 0.1646,
γ = 0.0001 and RMSE = 0.4169.
Quarter Time Observation Level Slope Season Forecast
t yt ℓt bt st y^t+1|t
1997 Q1 0 1.5
1997 Q2 1 -0.3
1997 Q3 2 -0.7
1997 Q4 3 9.8 0.0 -0.5
1998 Q1 4 11.8 9.9 0.0 1.5 11.3
1998 Q2 5 9.3 9.9 0.0 -0.3 9.7
1998 Q3 6 8.6 9.7 -0.0 -0.7 9.2
1998 Q4 7 9.3 9.8 0.0 -0.5 9.2
⋮ ⋮ ⋮ ⋮ ⋮ ⋮
2017 Q1 80 12.4 10.9 0.1 1.5 12.3
2017 Q2 81 10.5 10.9 0.1 -0.3 10.7
2017 Q3 82 10.5 11.0 0.1 -0.7 10.3
2017 Q4 83 11.2 11.3 0.1 -0.5 10.6
h y^T +h|T
2018 Q1 1 12.9
2018 Q2 2 11.2
2018 Q3 3 11.0
2018 Q4 4 11.2
2019 Q1 5 13.4
2019 Q2 6 11.7
2019 Q3 7 11.5
2019 Q4 8 11.7
2020 Q1 9 13.9
2020 Q2 10 12.2
2020 Q3 11 11.9
2020 Q4 12 12.2

https://otexts.com/fpp3/holt-winters.html 5/9
9/26/24, 3:58 PM 8.3 Methods with seasonality | Forecasting: Principles and Practice (3rd ed)

Table 8.4: Applying Holt-Winters’ method with multiplicative seasonality for

forecasting domestic tourism in Australia. Notice that the multiplicative seasonal
component sums to approximately m = 4. The smoothing parameters are
∗
α = 0.2237, β = 0.1360, γ = 0.0001 and RMSE = 0.4122.
Quarter Time Observation Level Slope Season Forecast
t yt ℓt bt st y^t+1|t
1997 Q1 0 1.2
1997 Q2 1 1.0
1997 Q3 2 0.9
1997 Q4 3 10.0 -0.0 0.9
1998 Q1 4 11.8 10.0 -0.0 1.2 11.6
1998 Q2 5 9.3 9.9 -0.0 1.0 9.7
1998 Q3 6 8.6 9.8 -0.0 0.9 9.2
1998 Q4 7 9.3 9.8 -0.0 0.9 9.2
⋮ ⋮ ⋮ ⋮ ⋮ ⋮
2017 Q1 80 12.4 10.8 0.1 1.2 12.6
2017 Q2 81 10.5 10.9 0.1 1.0 10.6
2017 Q3 82 10.5 11.1 0.1 0.9 10.2
2017 Q4 83 11.2 11.3 0.1 0.9 10.5
h y^T +h|T
2018 Q1 1 13.3
2018 Q2 2 11.2
2018 Q3 3 10.8
2018 Q4 4 11.1
2019 Q1 5 13.8
2019 Q2 6 11.7
2019 Q3 7 11.3
2019 Q4 8 11.6
2020 Q1 9 14.4
2020 Q2 10 12.2
2020 Q3 11 11.7
2020 Q4 12 12.1

The applications of both methods (with additive and multiplicative seasonality) are
presented in Tables 8.3 and 8.4 respectively. Because both methods have exactly the
same number of parameters to estimate, we can compare the training RMSE from
both models. In this case, the method with multiplicative seasonality fits the data
slightly better.

https://otexts.com/fpp3/holt-winters.html 6/9
9/26/24, 3:58 PM 8.3 Methods with seasonality | Forecasting: Principles and Practice (3rd ed)

The estimated components for both models are plotted in Figure 8.8. The small value
of γ for the multiplicative model means that the seasonal component hardly changes
over time. The small value of β ∗ means the slope component hardly changes over
time (compare the vertical scales of the slope and level components).

Figure 8.8: Estimated components for the Holt-Winters method with additive and
multiplicative seasonal components.

Holt-Winters’ damped method

Damping is possible with both additive and multiplicative Holt-Winters’ methods. A

method that often provides accurate and robust forecasts for seasonal data is the
Holt-Winters method with a damped trend and multiplicative seasonality:

https://otexts.com/fpp3/holt-winters.html 7/9
9/26/24, 3:58 PM 8.3 Methods with seasonality | Forecasting: Principles and Practice (3rd ed)

y^t+h|t = [ℓt + (ϕ + ϕ2 + ⋯ + ϕh )bt ] st+h−m(k+1)

ℓt = α(yt /st−m ) + (1 − α)(ℓt−1 + ϕbt−1 )
bt = β ∗ (ℓt − ℓt−1 ) + (1 − β ∗ )ϕbt−1
yt
st = γ + (1 − γ)st−m .
(ℓt−1 + ϕbt−1 )

Example: Holt-Winters method with daily data

The Holt-Winters method can also be used for daily type of data, where the seasonal
period is m = 7, and the appropriate unit of time for h is in days. Here we forecast
pedestrian traffic at a busy Melbourne train station in July 2016.

sth_cross_ped <- pedestrian |>

filter(Date >= "2016-07-01",
Sensor == "Southern Cross Station") |>

index_by(Date) |>
summarise(Count = sum(Count)/1000)
sth_cross_ped |>
filter(Date <= "2016-07-31") |>
model(

hw = ETS(Count ~ error("M") + trend("Ad") + season("M"))

) |>
forecast(h = "2 weeks") |>
autoplot(sth_cross_ped |> filter(Date <= "2016-08-14")) +
labs(title = "Daily traffic: Southern Cross",

y="Pedestrians ('000)")

https://otexts.com/fpp3/holt-winters.html 8/9
9/26/24, 3:58 PM 8.3 Methods with seasonality | Forecasting: Principles and Practice (3rd ed)

Figure 8.9: Forecasts of daily pedestrian traffic at the Southern Cross railway station,
Melbourne.
Clearly the model has identified the weekly seasonal pattern and the increasing trend
at the end of the data, and the forecasts are a close match to the test data.

Bibliography

Holt, C. C. (1957). Forecasting seasonals and trends by exponentially weighted

averages (ONR Memorandum No. 52). Carnegie Institute of Technology,
Pittsburgh USA. Reprinted in the International Journal of Forecasting, 2004.
[DOI]
Winters, P. R. (1960). Forecasting sales by exponentially weighted moving
averages. Management Science, 6(3), 324–342. [DOI]

17. Our implementation uses maximum likelihood estimation as described in Section

8.6 while Holt and Winters originally minimized the sum of squared errors. For
multiplicative seasonality, this will lead to slightly different parameter
estimates. Optimizing the sum of squared errors can be obtained by setting
opt_crit="mse" in ETS() . ↩︎

https://otexts.com/fpp3/holt-winters.html 9/9