Documente Academic
Documente Profesional
Documente Cultură
Forecasting Methods
horizontal pattern. 20
15
10 Sales (1000s of
gallons)
5
0
1 2 3 4 5 6 7 8 9 10 11
Week
Stationary Time Series
The term stationary time series are used to
denote a time series whose statistical
properties are independent of time. In
particular, this means that
1. The process generating the data has a
constant mean.
2. The variability of the time series is
constant over time.
Changes in business conditions often result
in a time series with a horizontal pattern that
shifts to a new level at some point in time.
For instance, suppose the gasoline
distributor signs a new contact to provide
gasoline in week 13. With this new contract, the
distributor naturally expects to see a substantial
increase in weekly sale starting in week 13.
Week Sales (1000s of gallons) Week Sales (1000s of gallons)
1 17 12 22
2 21 13 31
3 19 14 34
4 23 15 31
5 18 16 33
6 16 17 28
7 20 18 32
8 18 19 30
9 22 20 29
10 20 21 34
11 15 22 33
Gasoline Sales Time Series Plot After Obtaining
the Contract with the Vermont state Police
40
35
Sales (1000s of gallons)
30
25
20
15 Sales (1000s of gallons)
10
5
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Week
Trend Pattern
A trend pattern exists when the series
show a gradual shifts or movements to relatively
higher or o lower values over a longer period of
time. It is usually the result of long-term factors
such as populations increases or decreases,
shifting demographic characteristics of the
population, improving technology, and /or
changes in consumer preferences.
Sales
Year (1000s )
1 21.6 Bicycle Sales Time Series
2 22.9 35
3 25.5 30
25
4 21.9
Sales (1000s )
20
5 23.9 15 Sales (1000s )
6 27.5 10
7 31.5 5
8 29.7 0
1 2 3 4 5 6 7 8 9 10
9 28.6 Year
10 31.4
Year Revenue
1 23.1 Cholesterol Drug Rrevenue Time Series Plot ($ Millions)
120
2 21.3
3 27.4 100
4 34.6 80
5 33.8
Revnue
60
6 43.2 Revenue
7 59.5 40
8 64.4 20
9 74.2 0
10 99.3 1 2 3 4 5
Year
6 7 8 9 10
Seasonal Pattern
A trend pattern where movements in
historical data are analyzed over multiple years.
Seasonal patterns are recognized by observing
recurring patterns over successive period of
time.
UMBRELLAS SALES TIME SERIES
Year Quarter Sales Year Quarter Sales
1 1 125 4 1 109
2 153 2 137
3 106 3 125
4 88 4 109
2 1 118 5 1 130
2 161 2 165
3 133 3 128
4 102 4 96
3 1 138
2 144
3 113
4 80
80
Sales
60
40
20
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time Period
Trend and Seasonal Pattern
This pattern includes both the trend and
seasonal pattern.
QUARTERLY TELEVISION SET SALES TIME SERIES
Year Quarter Sales (1000s) Year Quarter Sales (1000s)
1 1 4.8 3 1 6
2 4.1 2 5.6
3 6 3 7.5
4 6.5 4 7.8
2 1 5.8 4 1 6.3
2 5.2 2 5.9
3 6.8 3 8
4 7.4 4 8.4
QUARTERLY TELEVISION SET SALES TIME SERIES PLOT
10
QUARTERLY TELEVISION SET
8
6
SALES (1000s)
4
Sales (1000s)
2
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
QUARTERLY TELEVISION SET SALES TIME SERIES
Year Quarter Sales (1000s) Year Quarter Sales (1000s)
1 1 4.8 3 1 6
2 4.1 2 5.6
3 6 3 7.5
4 6.5 4 7.8
2 1 5.8 4 1 6.3
2 5.2 2 5.9
3 6.8 3 8
4 7.4 4 8.4
The table above and the QUARTERLY TELEVISION SET SALES TIME SERIES PLOT
9
corresponding time series on the left
2 15.7 30
3 16.7 25
4 25
Sales (1000s )
20
5 18 15 Sales (1000s )
6 23 10
7 10 5
8 29.7 0
1 2 3 4 5 6 7 8 9 10
9 26 Year
10 17
FORECAST ACCURACY
Forecast accuracy is used to determine how well
a particular forecasting method is able to
reproduce the time series data that are
already available.
By selecting the method that is most accurate
for the data already known, we hope to
increase the likelihood that we will obtain
more accurate forecasts for the future time
periods.
Two Forecasting Methods:
1. Naïve Forecasting Method
An approach that uses the most recent data
value as the forecast for the next period. It is the
simplest method of forecasting.
2. Average of Past Values (Average of Historical
Values)
An approach that uses the average of all the
historical data available as the forecast for the next
period.
Consider the data below:
Period Demand
1 47
2 52
3 67
4 77
5 72
Naïve Average
Period Demand Forecasting Forecasting
1 47
2 52 47 47.00
3 67 52 49.50
4 77 67 55.33
5 72 77 60.75
Measures of Forecast Accuracy
1. Mean or Average of the Forecast Errors (MFE)
2. Mean Absolute Error (MAE)
3. Mean square Error (MSE)
4. Mean Absolute Percentage Error (MAPE)
n
∑
n n
et
∑
2
e
et t ∑
Y
100
MAE = t = k +1 MSE = t= k +1 t= k +1 t
n − k MAPE =
n − k n − k
FORECAST ERROR et
Forecast Error et - is the difference between the
actual and forecasted values for period t. It is
the key concept associated with measuring
forecast accuracy.
e t = Y t − Ft
where :
t = period
e t = error at period t
Y t = actual forecast
Ft = forecasted value
Measures of Forecast Accuracy
Mean or Average of the Forecast Errors (MFE)
n
∑e
t = k +1
t
MFE =
n−k
where :
t = period
e t = error at period t
k = number of periods at the beginning of the time series
for which we cannot produce a naive forecast.
n = total number of periods
Mean Absolute Error (MAE)
n
∑
t = k +1
et
MAE =
n−k
where :
t = period
e t = absolute value of error at period t
k = number of periods at the beginning of the time series
for which we cannot produce a naive forecast.
n = total number of periods
Mean square Error (MSE)
n
∑
t = k +1
e 2t
MSE =
n−k
where :
t = period
e 2t = squared error at period t
k = number of periods at the beginning of the time series
for which we cannot produce a naive forecast.
n = total number of periods
Mean Absolute Percentage Error (MAPE)
n
et
∑ 100
t = k +1 Yt
MAPE =
n−k
where :
t = period
e t = error at period t
k = number of periods at the beginning of the time series
for which we cannot produce a naive forecast.
n = total number of periods
FORECAST ACCURACY
To illustrate, consider the following Sales (1000s
data: Week of gallons)
1 17
The table on the right shows 2 21
the number of gallons of gasoline 3 19
(in 1000s) sold by a gasoline 4 23
5 18
distributor over the past 12 weeks. 6 16
Make a forecast about the data 7 20
using the naïve forecasting and the 8 18
average of the past, then compare 9 22
10 20
which of the two methods gives the 11 15
more accurate forecast using MAE, 12 22
MSE, and MAPE.
Absolute
Absolute Value of
Time Series Forecast Value of Squared Percentage Percentage
Week Value Forecast Error forecast Error Forecast Error Error Error
1 17
2 21 17 4 4 16 19.05 19.05
3 19 21 -2 2 4 -10.53 10.53
4 23 19 4 4 16 17.39 17.39
5 18 23 -5 5 25 -27.78 27.78
6 16 18 -2 2 4 -12.50 12.50
7 20 16 4 4 16 20.00 20.00
8 18 20 -2 2 4 -11.11 11.11
9 22 18 4 4 16 18.18 18.18
10 20 22 -2 2 4 -10.00 10.00
11 15 20 -5 5 25 -33.33 33.33
12 22 15 7 7 49 31.82 31.82
Total 5 41 179 1.19 211.69
n n
∑
t = k +1
et ∑ e 2t
∑
n
et
100
MAE = MSE = t = k +1 t = k +1 Y t
n−k n−k MAPE =
n − k
41 179 211.69
= = =
11 11 11
= 3.73 = 19.24%
= 16.27
Absolute Squared
Time Series Forecast Value of Forecast Percentage Absolute Value of
Week Value Forecast Error forecast Error Error Error Percentage Error
1 17
2 21 17.00 4.00 4.00 16.00 19.05 19.05
3 19 19.00 0.00 0.00 0.00 0.00 0.00
4 23 19.00 4.00 4.00 16.00 17.39 17.39
5 18 20.00 -2.00 2.00 4.00 -11.11 11.11
6 16 19.60 -3.60 3.60 12.96 -22.50 22.50
7 20 19.00 1.00 1.00 1.00 5.00 5.00
8 18 19.14 -1.14 1.14 1.31 -6.35 6.35
9 22 19.00 3.00 3.00 9.00 13.64 13.64
10 20 19.33 0.67 0.67 0.44 3.33 3.33
11 15 19.40 -4.40 4.40 19.36 -29.33 29.33
12 22 19.00 3.00 3.00 9.00 13.64 13.64
Total 4.52 26.81 89.07 2.75 141.34
n
∑ et
n n
∑ e 2t
et
t = k +1
∑ 100
t = k +1 Y t
MAE = t = k +1 MSE = MAPE =
n − k n − k n − k
26.81 89.07 141.34
= = =
11 11 11
= 2.44 = 8.10 = 12.85%
Comparing Naïve Forecasting Method and
Average of the Past Values
Naïve Method Average of the Past Values
MAE 3.73 2.44
MSE 16.27 8.10
MAPE 19.24% 12.85%
Moving Average =
∑ (most recent k data values)
k
∧
Y t +1 =
∑ (most recent k data values)
k
Y t - k +1 + ... + Y t −1 + Y1
=
k
t
∑Y
i = t − k +1
i
=
k
∧
Y t +1 = Forecast of the time series for period t + 1
Yi = Actual values of the time series in period i
k = number of periods of the time series data
used to generate the forecast
Moving Averages
To illustrate, consider the Sales (1000s
Week of gallons)
following data: 1 17
The table on the right 2 21
3 19
shows the number of gallons of 4 23
gasoline (in 1000s) sold by a 5 18
6 16
gasoline distributor over the 7 20
past 12 weeks. Compute the 3- 8 18
9 22
week moving averages of the 10 20
time series. 11 15
12 22
Moving Averages
Squared
Time Series Forecast Absolute Value Forecast Percentage Absolute Value of
Week Value Forecast Error of forecast Error Error Error Percentage Error
1 17
2 21
3 19
4 23 19 4.00 4.00 16.00 17.39 17.39
5 18 21 -3.00 3.00 9.00 -16.67 16.67
6 16 20 -4.00 4.00 16.00 -25.00 25.00
7 20 19 1.00 1.00 1.00 5.00 5.00
8 18 18 0.00 0.00 0.00 0.00 0.00
9 22 18 4.00 4.00 16.00 18.18 18.18
10 20 20 0.00 0.00 0.00 0.00 0.00
11 15 20 -5.00 5.00 25.00 -33.33 33.33
12 22 19 3.00 3.00 9.00 13.64 13.64
Total 0.00 24.00 92.00 -20.79 129.21
n n
∑ e
∑ e 2 n
e t
∑
t
t = k +1 t 100
MAE = MSE = t = k + 1
t = k + 1 Y t
n − k n − k MAPE =
24 n − k
= 92
9 = 129.21
9 =
= 2 . 67 9
= 10.22 = 14.36%
Forecasting Methods or Time Series Methods
A Weighted Moving Averages is a moving
average that involves selecting a different
weight for each data value in the moving
average and then computing a weighted
average of the most recent k values as the
forecast.
∧
Y t +1 = w t Y t + w t −1 Y t − 1 + ... + w t − k +1 Y t − k +1
∧
Y t +1 = Forecast of the time series for period t + 1
Yt = Actual values of the time series in period t
w t = weight applied to the actual time series value for period t
k = number of periods of the time series data
used to generate the forecast
Weighted Moving Averages
To illustrate, consider the Sales (1000s
Week of gallons)
following data: 1 17
The table on the right 2 21
3 19
shows the number of gallons of 4 23
gasoline (in 1000s) sold by a 5 18
6 16
gasoline distributor over the 7 20
past 12 weeks. Determine the 8 18
9 22
weighted moving average using 10 20
wt = 1/6, wt-1 = 2/6, wt-2 = 3/6. 11 15
12 22
. Note:
It should be noted that the most recent observation receives the
largest weight and the weight decreases with the relative age of
the data values.
Weighted Moving Averages using wt = 1/6, wt-1 = 2/6, wt-2 = 3/6.
Absolute
Time Value of Squared Absolute Value
Series Forecast forecast Forecast Percentage of Percentage
Week Value Forecast Error Error Error Error Error
1 17
2 21
3 19
4 23 19.33 3.67 3.67 13.44 15.94 15.94
5 18 21.33 -3.33 3.33 11.11 -18.52 18.52
6 16 19.83 -3.83 3.83 14.69 -23.96 23.96
7 20 17.83 2.17 2.17 4.69 10.83 10.83
8 18 18.33 -0.33 0.33 0.11 -1.85 1.85
9 22 18.33 3.67 3.67 13.44 16.67 16.67
10 20 20.33 -0.33 0.33 0.11 -1.67 1.67
11 15 20.33 -5.33 5.33 28.44 -35.56 35.56
12 22 17.83 4.17 4.17 17.36 18.94 18.94
Total 0.50 26.83 103.42 -19.17 143.93
Forecasting Methods or Time Series Methods
An Exponential Smoothing also uses a weighted
average of past time series values as a
forecast; it is a special case of the weighted
moving averages method in which we select
only one weight – the weight for the most recent
observation. The weights for the other data
values are computed automatically and become
smaller as the observations move farther into
the past. The exponential smoothing model
follows.
∧ ∧
Y t + 1 = α Y t + (1 − α ) Y t
∧ ∧
Y t + 1 = α Y t + (1 − α ) Y t
∧
Y t +1 = Forecast of the time series for period t + 1
Yt = Actual values of the time series in period t
∧
Yt = forecast of the time series for period t
α = smoothing constant (0 ≤ α ≤ 1)
Exponential Smoothing
To illustrate, consider the Sales (1000s
Week of gallons)
following data: 1 17
The table on the right 2 21
3 19
shows the number of gallons of 4 23
gasoline (in 1000s) sold by a 5 18
6 16
gasoline distributor over the 7 20
past 12 weeks. Find the simple 8 18
9 22
exponential smoothing using 10 20
an α of 0.20. 11 15
12 22
Time Series Squared
Week Value Forecast Forecast Error Forecast Error
1 17
2 21 17.00 4.00 16.00
3 19 17.80 1.20 1.44
4 23 18.04 4.96 24.60
5 18 19.03 -1.03 1.07
6 16 18.83 -2.83 7.98
7 20 18.26 1.74 3.03
8 18 18.61 -0.61 0.37
9 22 18.49 3.51 12.34
10 20 19.19 0.81 0.66
11 15 19.35 -4.35 18.94
12 22 18.48 3.52 12.38
Total 10.92 98.80
Trend-Adjusted Exponential Smoothing
A variation of simple exponential smoothing can be
used hen a time series exhibits a linear trend. It is called
trend-adjusted exponential smoothing or, sometimes, double
smoothing, to differentiate it from simple exponential
smoothing, which is appropriate only when data vary around
an average or have step or gradual changes.
If a series exhibits trend, and simple smoothing is used on it,
the forecast will all lag the trend
• if the data are increasing, each forecast will be too low;
• If the data are decreasing, each forecast will be too high.
Copyright ©2018 McGraw-Hill Higher Education. All rights reserved. No reproduction or distribution
without the prior written consent of McGraw-Hill Education
Trend-Adjusted Exponential Smoothing
Copyright ©2018 McGraw-Hill Higher Education. All rights reserved. No reproduction or distribution
without the prior written consent of McGraw-Hill Education
Trend-Adjusted Exponential Smoothing (cont.)
Copyright ©2018 McGraw-Hill Higher Education. All rights reserved. No reproduction or distribution
without the prior written consent of McGraw-Hill Education
Trend-Adjusted Exponential Smoothing
TAF t +1 = S t + Tt
where
S t = Previous forecast plus smoothed error
Tt = Current trend estimate
TAF t +1 = S t + Tt
S t = TAF t + α (At − TAF t )
Tt = Tt −1 + β (TAF t − TAF t−1 − Tt −1 )
Copyright ©2018 McGraw-Hill Higher Education. All rights reserved. No reproduction or distribution
without the prior written consent of McGraw-Hill Education
Process:
1. There must be a historical data.
2. Identify the α and β of the problem.
3. Compute the initial trend estimate and the starting forecast
of the current period.
Initial trend estimate = the average of the net changes from 1 – k
Initial Forecast = At-1 + initial trend estimate
4. Compute for the St, St = TAFt + α(At - TAFt).
5. Compute the next forecast TAFt+1 = St + Tt .
6. Compute for St and Tt. Let t + 1 be the next current period t.
Tt = Tt-1 + β(TAFt – TAFt-1 - Tt-1)
7. Continue the process 5 and 6 until the desired forecast is
computed.
8.
Example:
Cell phone sales for a Week Unit Sales
California-based firm over the last 10
1 700
weeks are shown in the following
2 724
table. Plot the data, and visually
check to see if a linear trend line 3 720
would be appropriate. If it is, use the 4 728
trend-adjusted exponential 5 740
smoothing to obtain forecasts for 6 742
periods 6 though 11, with α = 0.40 7 758
and β = 0.30. Based on the net 8 750
change from periods 1 t0 4. 9 770
10 775
Copyright ©2018 McGraw-Hill Higher Education. All rights reserved. No reproduction or distribution
without the prior written consent of McGraw-Hill Education
For purposes of illustration, the original data, the trend
line, and the two projections (forecasts) are shown on
the following graph:
Copyright ©2018 McGraw-Hill Higher Education. All rights reserved. No reproduction or distribution
without the prior written consent of McGraw-Hill Education
Week Unit Sales TAFt St Tt
1 700
2 724
3 720
4 728
5 740
6 742
7 758
8 750
9 770
10 775
Copyright ©2018 McGraw-Hill Higher Education. All rights reserved. No reproduction or distribution
without the prior written consent of McGraw-Hill Education