Scribd
Upload
35 views

Holt-Winters Exponential Method of Forecasting

The document describes the Holt-Winters exponential smoothing method for forecasting time series data that exhibits trends and seasonality. It discusses additive and multiplicative models for handling seasonality and describes Winter's multiplicative exponential smoothing technique, which uses level, trend, and seasonal components with smoothing constants to forecast future values. An example application of the Winter's method to sales data is shown.

Uploaded by

Mahnoor Khalid
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
35 views

Holt-Winters Exponential Method of Forecasting

The document describes the Holt-Winters exponential smoothing method for forecasting time series data that exhibits trends and seasonality. It discusses additive and multiplicative models for handling seasonality and describes Winter's multiplicative exponential smoothing technique, which uses level, trend, and seasonal components with smoothing constants to forecast future values. An example application of the Winter's method to sales data is shown.

Uploaded by

Mahnoor Khalid
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 16

Lecture 6:

Holt- Winters Exponential Method


of Forecasting
Seasonality
• Seasonality if defined as the consistent month-to-month (or
quarter-to-quarter) differences that occur each year.
• The easiest way to check if there is seasonality in a time series is
to look at a plot of the times series to see if it has a regular
pattern of up and/or downs in particular months or quarters.
• There are basically two extrapolation methods for dealing with
seasonality:
• We can use a model that takes seasonality into account or;
• We can deseasonalize the data, forecast the data, and then adjust the
forecasts for seasonality.
Seasonality -- continued
• Winters’ model is of the first type. It attacks seasonality directly.
• Seasonality models are usually classified as additive or
multiplicative.
• An additive model finds seasonal indexes, one for each month, that we
add to the monthly average to get a particular month’s value.
• A multiplicative model also finds seasonal indexes, but we multiply the
monthly average by these indexes to get a particular month’s value.
• Either model can be used but multiplicative models are
somewhat easier to interpret.
Winter’s Model of Seasonality
• This method was first suggested by Holt's student, Peter
Winters, in 1960.
• The method is used when the data shows trend and
seasonality. A seasonal pattern appears to exist. Seasonality
can be additive or multiplicative.
Winter’s Model of Seasonality
• Winters’ model is very similar to Holt’s model - it has level
and trend terms and corresponding smoothing constants
alpha and beta - but it also has seasonal indexes and a
corresponding smoothing constant.
• The new smoothing constant controls how quickly the
method reacts to perceived changes in the pattern of
seasonality.
• If the constant is small, the method reacts slowly; if the
constant is large, it reacts more quickly.
Note
• Holt−Winter’s (additive and multiplicative) both smoothing
methods exists depending on the nature of seasonality, but
in this module we’ll work only with multiplicative smoothing
method.
Method for Computation
• The four equations used in Winter’s (multiplicative)smoothing method are:

• The exponentially smoothed series, or the current level estimate:


𝑌𝑡
𝐿𝑡 = 𝛼 + 1 − 𝛼 𝐿𝑡−1 + 𝑇𝑡−1 .
𝑆 𝑡−𝑠

• The trend estimate:


𝑇𝑡 = 𝛽 𝐿𝑡 − 𝐿𝑡−1 + 1 − 𝛽 𝑇𝑡−1 .

• The seasonality estimate:


𝑌
𝑆𝑡 = 𝛾 𝐿𝑡 + (1 − 𝛾)𝑆𝑡−𝑠 .
𝑡

• Forecast p periods into the future:


෣
𝑌𝑡+𝑝 = (𝐿𝑡 +𝑝𝑇𝑡 )𝑆𝑡−𝑠+𝑝 ,
Assumptions
• where
𝐿𝑡 = new smoothed value estimate of current level . usually we assume𝐿𝑠 =𝑌𝑠 .
s = length of seasonality.
α= smoothing constants for the level (0< α < 1).
𝑌𝑡 = actual value of series in period t.
𝛽 = smoothing constant for trend estimate (0< β <1).
𝑇𝑡 = trend estimate.
usually we assume𝑇𝑠 =0.
γ = smoothing constant for seasonality estimate (0< γ<1).
𝑆𝑡 = seasonal estimate.
usually we assume 𝑆𝑥 =1 for x=1,2,…,s.
p = periods to be forecast in the future.
෣
𝑌𝑡+𝑝 = forecast for 𝑝 periods into the future.
Assumptions
• One approach is to set the first estimate equal to the first
observation, the trend is then estimated to equal zero, and
the seasonal indices are set to 1.
Minitab
• Develops a regression equation, and uses constants from the
equation as initial estimates for the level and the trend. The
seasonal components are obtained from a dummy variable
regression using detrended data.
Minitab Instructions: STAT > TIME SERIES > WINTERS’
METHOD.
Winters' Multiplicative Model for Y

Actual
900 Predicted

800 Forecast
Actual
700 Predicted
Forecast
600
Y

500
Smoothing Constants
400 Alpha (level): 0.400
Gamma (trend): 0.100
300 Delta (season): 0.300

200 MAPE: 15.21


MAD: 63.55
100 MSD: 7636.86

0 10 20

Time
Model Accuracy
Autocorrelation function for Example 4.6 Residuals
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0

1 2 3 4 5 6

Lag Corr T LBQ

1 0.30 1.46 2.41


2 -0.03 -0.14 2.43
3 0.02 0.08 2.44
4 -0.11 -0.48 2.80
5 -0.26 -1.15 4.97
6 0.03 0.14 5.01

None of the coefficients appear to be significantly larger than zero, and the
small value of LBQ (5.01) shows that the series is random and model is
accurate .
Example
• The Acme Tool Company Sales for 5 quarters of year1996 and 1997 is
given below. Forecast the value of 1st quarter for 1998 using triple
exponential method with α=0.3, β=0.1 and γ=0.3 with length of
seasonality s=4 with an initial value of level= 500.
Year t 𝒀𝒕 𝑳𝒕 𝑻𝒕 𝑺𝒕
1996 1 500 1
2 350 1
3 250 1
4 400 500 0 1
1997 5 450 485.0000 -1.5 0.9784
6 350 443.4500 -5.505 0.9368
7 200 366.5615 -12.6434 0.8637
8 300 337.7427 -14.2609 0.9665
Example
We’ll compute the values for time period 5 as follows:
𝑌5
• 𝐿5 = 0.3 + 1 − 0.3 𝐿5−1 + 𝑇5−1
𝑆 5−4
450
• = 0.3 + 0.7 500 + 0 = 485
1

• 𝑇5 = 0.1 𝐿5 − 𝐿5−1 + 1 − 0.1 𝑇5−1


=0.1(485-500)+0.9(0) = -1.5

𝑌5
• 𝑆5 = 0.3 + (1 − 0.3)𝑆5−4
𝐿5
= 0.3(450/485)+0.7(1) = 0.978
Example
• And ෣
𝑌5+1 = (𝐿5 +𝑇5 )𝑆5−4+1 = 485 − 1.5 1 = 483.5

and so on ….. Clearly the forecast for 1st quarter of 1998 is

෣
𝑌8+1 = (𝐿8 +𝑇8 )𝑆8−4+1
= 337.7427 − 14.2609 0.9784 = 316.4946
Other Measures of Accuracy
• Mean Absolute Deviation 𝑛
1
𝑀𝐴𝐷 = ෍ 𝑌𝑡 − 𝑌෡𝑡
𝑛
𝑖=1
• Mean Squared Error 𝑛
1 2
𝑀𝑆𝐸 = ෍ 𝑌𝑡 − 𝑌෡𝑡
𝑛
𝑖=1
• Mean Absolute Percentage Error 𝑛
1 𝑌𝑡 − 𝑌෡𝑡
𝑀𝐴𝑃𝐸 = ෍
𝑛 𝑌𝑡
𝑖=1
• Mean Percentage error 𝑛
1 𝑌𝑡 − 𝑌෡𝑡
𝑀𝑃𝐸 = ෍
𝑛 𝑌𝑡
𝑖=1

You might also like