CHAPTER 3 – DEMAND FORECASTING
Faculty of International Economic Relations
University of Economics and Law
1 Het begint met een idee
CONTENTS
Forecasting characteristics
Forecasting methods
Time-series forecasting models
Forecasting errors
Forecasting issues
Vrije Universiteit Amsterdam
FORECASTING CHARACTERISTICS
Source: foxbusiness.com Vrije Universiteit Amsterdam
FORECASTING CHARACTERISTICS
Forecast is always inaccurate should include expected value and
measure of error
Long-term forecasts are less accurate than short-term forecasts
Trumpet of Doom
Fo re c ast Erro r R ange o ve r Tim e
P e rc e ntage
Fo re c as t 0
Erro r
0 Tim e U ntil Fo re c as t Eve nt
Vrije Universiteit Amsterdam
FORECASTING CHARACTERISTICS
Aggregate forecasts are more accurate than disaggregate forecasts
Law of Large Number: As volume increases, relative variability
decreases
The farther up the supply chain, the greater the distortion of
information
Vrije Universiteit Amsterdam
FORECASTING METHODS
Qualitative: primarily subjective; rely on judgment and opinion
Time series: use historical demand only
Causal: use the relationship between demand and some other
factors to develop forecast
Simulation
o Imitate consumer choices that give rise to demand
o Can combine time series and causal methods
Vrije Universiteit Amsterdam
FORECASTING COMPONENTS (TIME SERIES)
Observed demand (O) =
Systematic component (S) + Random component (R)
Systematic component: expected value of demand
o Level: current deseasonalized demand
o Trend: growth or decline in demand
o Seasonality: predictable seasonal fluctuation
Random component: The part of the forecast that deviates from the
systematic component
Forecast error: difference between forecast and actual demand
Naïve Techniques & Averaging Techniques
Vrije Universiteit Amsterdam
MOVING AVERAGE
A t-1 + A t-2 + A t-3 +...+A t- n
Ft =
n
Ft = Forecast for the coming period
n = Number of periods to be averaged
A t-1 = Actual occurrence in the past period for up to “n”
periods
Used when demand has no observable trend or seasonality
Vrije Universiteit Amsterdam
MOVING AVERAGE
Week Demand 3-Week 6-Week
1 650 F4=(650+678+720)/3
2 678 =682.67
3 720 F7=(650+678+720
4 785 682.67 +785+859+920)/6
5 859 727.67 =768.67
6 920 788.00
7 850 854.67 768.67
8 758 876.33 802.00
9 892 842.67 815.33
10 920 833.33 844.00
11 789 856.67 866.50
12 844 867.00 854.83
Vrije Universiteit Amsterdam
MOVING AVERAGE
950
900
850
800
Demand
Demand
750
3 Week
700
6 Week
650
600
550
500
1 2 3 4 5 6 7 8 9 10 11 12
Week
fewer n more responsive more accurate
Vrije Universiteit Amsterdam
SIMPLE EXPONENTIAL SMOOTHING FORECAST
FFtt == FFt-1
t-1
+
+ (A
(A t-1
t-1
-
- F
F )
t-1)
t-1
Where :
Ft Forecast v alue for the coming t time period
Ft - 1 Forecast v alue in 1 past time period
At - 1 Actual occurance in the past t tim e period
Alpha smoothing constant
Vrije Universiteit Amsterdam
SIMPLE EXPONENTIAL SMOOTHING FORECAST
Week Demand Question:
Question:Given
Giventhe
theweekly
weeklydemand
demand
1 820 data,
data,what
whatare
arethe
theexponential
exponential
2 775 smoothing
smoothingforecasts
forecastsfor
forperiods
periods2-
2-
3 680 10
10using
usingalpha
alpha==0.10
0.10and
andalpha
alpha==
4 655 0.60?
0.60?
Assume
AssumeFF1=D
=D1
5 750 1 1
6 802
7 798
8 689
9 775
10
Vrije Universiteit Amsterdam
SIMPLE EXPONENTIAL SMOOTHING FORECAST
Week Demand 0.1 0.6
1 820 820.00 820.00
2 775 820.00 820.00
3 680 815.50 793.00
4 655 801.95 725.20
5 750 787.26 683.08
6 802 783.53 723.23
7 798 785.38 770.49
8 689 786.64 787.00
9 775 776.88 728.20
10 776.69 756.28
Vrije Universiteit Amsterdam
SIMPLE EXPONENTIAL SMOOTHING FORECAST
850
800
750
Demand
Demand
700
0.1
650
0.6
600
550
500
1 2 3 4 5 6 7 8 9 10
Week
Vrije Universiteit Amsterdam
TREND-ADJUSTED EXPONENTIAL SMOOTHING FORECAST
¿
¿
𝑇 𝑡 =𝛽 (𝐹 𝑡 − 𝐹 𝑡 −1 )+(1 − 𝛽)𝑇 𝑡 −1
𝐹 𝑡 :𝑒𝑥𝑝𝑜𝑛𝑒𝑛𝑡𝑖𝑎𝑙𝑙𝑦 𝑠𝑚𝑜𝑜𝑡ℎ𝑒𝑑 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑑𝑎𝑡𝑎 𝑠𝑒𝑟𝑖𝑒𝑠𝑖𝑛𝑝𝑒𝑟𝑖𝑜𝑑𝑡
𝑇 𝑡 :𝑒𝑥𝑝𝑜𝑛𝑒𝑛𝑡𝑖𝑎𝑙𝑙𝑦 𝑠𝑚𝑜𝑜𝑡ℎ𝑒𝑑𝑡𝑟𝑒𝑛𝑑𝑖𝑛𝑝𝑒𝑟𝑖𝑜𝑑𝑡
𝐴𝑡 :𝑎𝑐𝑡𝑢𝑎𝑙 𝑑𝑒𝑚𝑎𝑛𝑑 𝑖𝑛𝑝𝑒𝑟𝑖𝑜𝑑𝑡
¿
¿
Vrije Universiteit Amsterdam
TREND-ADJUSTED EXPONENTIAL SMOOTHING FORECAST
Assume that the forecast of Period 1 is 27 units and its trend is 0.
Alpha= 0.3; Beta= 0.6. The actual demand of Period 1 turned out to
be 30 units.
1. Calculate the forecast of Period 2.
2. The actual demand of Period 2 is 34 units. Calculate the forecast
of Period 3.
3. FIT1= F1+T1=27 T1=0 --> F1=27
4. FIT2=27.9+0.54=28.44
5. F2=0.3*30+0.7*27=27.9
6. T2=0.6(27.9-27)+0.4*0 = 0.54
Vrije Universiteit Amsterdam
LINEAR REGRESSION
^ ^𝑥
^ +𝑏
𝑦 =𝑎
^𝑦 : 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑓𝑜𝑟 𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒 𝑦
𝑥:𝑖𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒 𝑥,𝑢𝑠𝑒𝑑𝑡𝑜 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑦
𝑎^ : 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑𝑖𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡𝑡𝑒𝑟𝑚 𝑓𝑜𝑟 𝑡ℎ𝑒𝑙𝑖𝑛𝑒
^
𝑏:𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝑠𝑙𝑜𝑝𝑒𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 𝑓𝑜𝑟 𝑡ℎ𝑒𝑙𝑖𝑛𝑒
^
𝑏=
∑ 𝑥𝑦 − 𝑛 𝑥 𝑦
∑ 𝑥2 − 𝑛 𝑥2
^𝑥
^ = 𝑦 −𝑏
𝑎
Vrije Universiteit Amsterdam
LINEAR REGRESSION
Month Demand
1 8
2 12
3 25
4 40
5 50
6 65
7 36
8 61
9 88
10 63
11 ?
12 ?
13 ?
Vrije Universiteit Amsterdam
TREND-CORRECTED EXPONENTIAL SMOOTHING (HOLT’S MODEL)
𝐷𝑡 =𝑎𝑡 +𝑏
𝐹 𝑡 +1=𝐿𝑡 + 𝑇 𝑡
¿
𝑇 𝑡 +1= 𝛽( 𝐿𝑡 +1 − 𝐿𝑡 )+(1 − 𝛽) 𝑇 𝑡
𝐿𝑡 : 𝐸𝑠𝑡𝑖𝑚𝑎𝑡𝑒 𝑜𝑓 𝑙𝑒𝑣𝑒𝑙𝑖𝑛𝑝𝑒𝑟𝑖𝑜𝑑 𝑡
𝐿0 𝑎𝑛𝑑𝑇 0 𝑎𝑟𝑒 𝑜𝑏𝑡𝑎𝑖𝑛𝑒𝑑 𝑓𝑟𝑜𝑚𝑟𝑒𝑔𝑟𝑒𝑠𝑠𝑖𝑜𝑛!
Vrije Universiteit Amsterdam
TREND-CORRECTED EXPONENTIAL SMOOTHING (HOLT’S MODEL)
Japan National Tourist Organization has reported a constant increase in
number of visitors to Japan during the last ten years. For example, the
number of visitors to Japan from other Asian countries during the period of
2002–2007 has been 3,417,774; 3,511,513; 4,208,095; 4,627,478;
5,247,125; and 6,130,262 annually. Forecast the number of visitors for 2008
using trend-corrected exponential smoothing with Alpha = 0.1, Beta = 0.2.
Vrije Universiteit Amsterdam
TIME SERIES WITH SEASONALITY (WITHOUT TREND)
1. Compute the average historical demand each season
2. Compute the average demand over all months
3. Compute a seasonal index for each season (Step 1/ Step 2)
4. Estimate next year’s total annual demand
5. Estimate demand for each month in the next year (Step 4/n * Step 3)
Vrije Universiteit Amsterdam
TIME SERIES WITH SEASONALITY (WITHOUT TREND)
2
Month Year 1 Year 2 Year 3 1 3 TB 36 THÁNG =94
Jan 80 85 105 SEA
90
SON
Feb 70 85 85 80
AL
Mar 80 93 82 INDE
X
Apr 90 95 115
May 113 125 131 Year 4’s forecast: 1,200 units
Jun 110 115 120 4 --> TB MỖI THÁNG =100
Jul 100 102 113
Aug 88 102 110
Sep 85 90 95
Oct 77 78 85
Nov 75 82 83
Dec 82 78 80
Vrije Universiteit Amsterdam
TIME SERIES WITH SEASONALITY (WITHOUT TREND)
140
120
100
80
60
40
20
0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Year 1 Year 2 Year 3
Vrije Universiteit Amsterdam
TREND- AND SEASONALITY-CORRECTED EXPONETIAL SMOOTHING (WINTER’S
MODEL)
𝐹 𝑡 +1=(𝐿𝑡 +𝑇 𝑡 )𝑆𝑡 +1
¿
𝑇 𝑡 +1= 𝛽( 𝐿𝑡 +1 − 𝐿𝑡 )+(1 − 𝛽) 𝑇 𝑡
𝑆 𝑡 +𝑝+1=𝛾 (𝐷𝑡 + 1 /𝐿𝑡 +1 )+(1 −𝛾)𝑆 𝑡 +1
¿
¿
¿
Vrije Universiteit Amsterdam
TREND- AND SEASONALITY-CORRECTED EXPONETIAL SMOOTHING (WINTER’S
MODEL)
1. De-seasonalize demand
2. Run linear regression to estimate level and trend
3. Estimate seasonal factors
4. Calculate the forecast for any period
Vrije Universiteit Amsterdam
FORECAST ERRORS
e t = A t Ft
Where :
et Forecast error for period t
At Actual demand for period t
Ft Forecast for period t
Vrije Universiteit Amsterdam
FORECAST ERRORS
Mean Squared Error Mean Absolute Deviation (MAD)
n
n
|e |
e 2
t MAD t 1
t
MSE t 1 n
n
Mean Absolute Percentage Error (MAPE)
100 n et
MAPE
n t 1 At
Vrije Universiteit Amsterdam
SELECTING THE SMOOTHING CONSTANT
Period Demand Simple exponential smoothing forecast:
1 180 Alpha = 0.1
2 168 Alpha = 0.5
3 159
Forecast of Period 1 = 175
4 175
Which alpha is preferred?
5 190
Hints: MSE and MAD
6 205
7 180
8 182
Vrije Universiteit Amsterdam
TRACKING SIGNAL
e t
TS t 1
MAD
If TS [6, 6] There is bias in the forecast
Vrije Universiteit Amsterdam
TRACKING SIGNAL
3.0
2.5
Period Demand Forecast 2.0
1 90 100 1.5
2 95 100 1.0
3 115 100 0.5
0.0
4 100 110 0 1 2 3 4 5 6 7
-0.5
5 125 110
-1.0
6 140 110 -1.5
-2.0
-2.5
Vrije Universiteit Amsterdam
FORECASTING REMARKS
No single technique works best in every situation
Should use more than one forecasting method
Factors to consider: cost, accuracy, the data availability & software,
time needed to gather & analyze data
The higher the accuracy, the higher the cost
Vrije Universiteit Amsterdam
FORECASTING PROCESS
Determine purpose of forecast
Establish a time horizon
Select a forecasting technique
Obtain, clean, and analyze data
Make a forecast
Monitor the forecast
Vrije Universiteit Amsterdam
CHAPTER 3 – DEMAND FORECASTING
THANK YOU!
33 Het begint met een idee
