Demand Forecasting

© 2007 Pearson Education 7-1

 Outline
• The role of forecasting in a supply chain
• Characteristics of forecasts
• Components of forecasts and forecasting
methods
• Basic approach to demand forecasting
• Time series forecasting methods
• Measures of forecast error
 Forecasting Prediction
1 Forecasting involves the projection of Prediction involves judgment in
the past into the future. management after taking all available
information into account.
2 Forecast involves estimating the level Prediction involves the anticipated change
of demand of a product on the basis of into the future. It may include even new
factors that generated the demand in factors that may affect future demand.
the past months.
3 Forecasting is more scientific. Prediction is more intuitive.
4 It is relatively free from personal It is more governed by personal bias and
bias. preferences.
5 It is more objective. It is more subjective.
6 It is generally called as “Throw It is generally called as “Saying
Ahead” technique. Beforehand” technique.
7 Error analysis is possible. Prediction does not contain error analysis.
8 Forecasting is reproducible, i.e., every It is non-producible.
time same result would be obtained by
any particular technique.
 Role of Forecasting
• The basis for all strategic and planning decisions in
a supply chain
• Examples:
– Production: scheduling, inventory, aggregate planning
– Marketing: sales force allocation, promotions, new
production introduction
– Finance: plant/equipment investment, budgetary
planning
– Personnel: workforce planning, hiring, layoffs
• All of these decisions are interrelated
 Characteristics of Forecasts
• Forecasts are always wrong. Should include expected
value and measure of error. Ex.100-1900 and 900-
1100
• Long-term forecasts are less accurate (large standard
deviation) than short-term forecasts (forecast horizon
is important)-
• Aggregate forecasts are more accurate (small
standard deviation) than disaggregate forecasts.
Ex – GDP (2% accuracy) - Sector-Industry-Firm.
 Forecasting Methods
• Qualitative Methods: primarily subjective;
rely on judgment and opinion-no or little
historical data
Expert Opinion
Market Survey
Delphi Method
Historical Analogy
Quantitative methods

Time Series: use historical demand only

Static
Adaptive

Causal: use the relationship between demand and some

other factor to develop forecast (Interest rates-cement,
power cuts-inverter, coffee-tea)

Simulation:
Imitate consumer choices that give rise to demand.
Can combine time series and causal methods.
What if we do not forecast?

Cost of forecasting ????

What if we do not forecast?

Cost of forecasting : labor, material, capital, capacity etc.

Total
Cost Cost Cost of increased forecasting activity

Cost of poor/inaccurate forecasting

No Expensive
method method
 Components of time series: T+C+S+I

Demand
Demand

Random
movement

Time Time
(a) Trend (b) Cycle

Demand
Demand

Time Time
(c) Seasonal pattern (d) Trend with seasonal pattern
Variations in Time series.
Trend: long term direction of the series. It describes historical
pattern, helps in projecting past pattern into future, eliminate trend
from time series (other components can be studied separately).

How to find out trend?

Variations in Time series.
Trend: long term direction of the series. It describes historical pattern, helps in projecting past
pattern into future, eliminate trend from time series (other components can be studied separately).

-One way to describe the trend component is to fit a line visually to a set of points on a graph.

- Fitting the linear trend by the least square method.

Estimated (or
predicted) Y Estimate of the Estimate of the
value for regression regression slope
observation i intercept

Value of X for

Ŷi  a  b X i observation i
Translating, or coding, Time
Ex. What was the sales in 1996?

X Y Coded small
1988 98 (IDV)
1989 105 (IDV ) X (Y) (x)
1990 116 1988 98 -7
1991 119 1989 105 -5
1992 135 1990 116 -3
1993 156 1991 119 -1
1994 177 1992 135 1
1995 208 1993 156 3
= - 29875.50 1994 177 5
1995 208 7

b= Sum (xY)/sum(x*x)=7.536
=15.071 a= Y =139.25

Sales(1996)= -29875.50+15.071 *1996=206.21 Sales(1996)= 139.25+7.536*9= 207

Year Car sold(Y) X (per capita
income)

1 9.5 120
2 11 135
3 12 130
4 12.5 150
5 14 170
6 16 190
7 18 220

Find Y8 and standard error

Standard error of
forecasting
•The “t” distribution is a continuous probability distribution of the z-
score when the estimated standard deviation is used in the denominator
rather than the true standard deviation.

•The “t” distribution, like the normal distribution, is bell-shaped and

symmetric, but it has heavier tails, which means it tends to produce
values that fall far from its mean.

•“t” - tests are used in statistics to estimate significance.

Y8 (at 95% CL)=20.55+/-2.571*.57 = 19.070 to22.02
Find Y9, Y10.
Use of second degree in trend in time series.

Ŷ = a+bx+c x 2

Ŷ = estimate of the dependent variable

a,b,c = numerical constants
x= coded value of time variable

Ex. What would have been the sales in 1996?

  Year (X)  Sales
1991 13
1992 24
1993 39
1994 65
1995 106
   Year (X)  Sales (Y)
1991 13
1992 24
1993 39
1994 65
1995 106

X Y x x2 x4 xY x2*Y
1991 13 -2 4 16 -26 52
1992 24 -1 1 1 -24 24
1993 39 0 0 0 0 0
1994 65 1 1 1 65 65
1995 106 2 4 16 212 424
9965 247   10 34 227 565
, 247 = 5a+10c
,565=10a+34c

= 227/10=22.7
a=39.3,b=22.7,c=5.07

Ŷ = a+bx+c x 2
2
= 39.3+22.7x+5.07 x
Y(1996,x=3) = 39.3+22.7*3+5.07*9 = 153.03
The number of faculty-owned personal computers at the University of
Ohio increased dramatically between 1990 and 1995:

Year : 1990 1991 1992 1993 1994 1995

Number of PCs: 50 110 350 1,020 1,950 3,710

(a)Develop a linear estimating equation that best describes these data.

(b) Develop a second-degree estimating equation that best describes

these data.

(c) Estimate the number of PCs that will be in use at the university in
1999, using both equations.

(d) If there are 8,000 faculty members at the university, which

equation is the better predictor? Why?
Cyclical variation: Is the component of time series that tends to oscillate above and
below the secular trend line for periods longer than 1 year.

Residual method: When we look at time series of annual data, only the secular trend,
cyclical and irregular components are considered. (This is true because seasonal variation
makes a complete, regular cycle within each year and thus does not effect one year any more
than another).

If we use a time series composed of annual data, we can find the fraction of the trend by
dividing the actual value (Y) by the corresponding trend value ( Ŷ) for each value in the
time series. We then multiply the result of this calculation by 100. This gives us the measure
of cyclical variation as a percent of trend.

Percent trend = ( Y/ Ŷ )*100

Ex.: Find cyclical variation

1988 7.5
1989 7.8
1990 8.2
1991 8.2
1992 8.4
1993 8.5
1994 8.7
1995 9.1

Soln. Estimated line = -390+0.200X

Estimated line (1988) = -390+0.200*1988=7.6, similarly for other years.

X Y Ŷ 
1988 7.5 7.6
1989 7.8 7.8
1990 8.2 8
1991 8.2 8.2
1992 8.4 8.4
1993 8.5 8.6
1994 8.7 8.8
1995 9.1 9
 X Y Ŷ   % of Trend = (Y/ Ŷ )100
1988 7.5 7.6 98.68
1989 7.8 7.8 100
1990 8.2 8 102.5
1991 8.2 8.2 100
1992 8.4 8.4 100
1993 8.5 8.6 98.84
1994 8.7 8.8 98.86
1995 9.1 9 101.1

Graph of percent of trend

 Relative cyclical residual: Another measure of cyclical variation . In this method , the percentage
deviation from the trend is found for each value.

Relative Cyclical Residual = (( Y - Ŷ )/ Ŷ ) *100

1 2 3 4 5= col. (4)-100
% of
X Y Ŷ
  Trend RCR
1988 7.5 7.6 98.68 -1.32
1989 7.8 7.8 100 0
1990 8.2 8 102.5 2.5
1991 8.2 8.2 100 0
1992 8.4 8.4 100 0
1993 8.5 8.6 98.84 -1.2
1994 8.7 8.8 98.86 -1.1
1995 9.1 9 101.1 1.1
Graph of relative cyclical residual
Graph of percent of trend
Graph of relative cyclical residual
Ex. Find seasonal Indices?
Year (1) Quarter (2) Occupancy (3)

2016 I 1861
     
  II 2203
     
  III 2415
     
  IV 1908
     
2017 I 1921
     
  II 2343
     
  III 2514
     
  IV 1986
     
2018 I 1834
     
  II 2154
     
  III 2098
     
  IV 1799
     
2019 I 1837
     
  II 2025
     
  III 2304
     
  IV 1965
     
2020 I 2073
     
  II 2414
     
  III 2339
     
  IV 1967
Ex. Find seasonal Indices?
Year (1) Quarter (2) Occupancy (3) 4 quarter MA (4) 4-quarter centered MA (5) % of actual MA values (6)= (3)/(5)*100

2016 I 1861      
           
  II 2203      
      2096.75    
  III 2415   2104.25 114.767732
      2111.75    
  IV 1908   2129.25 89.60901726
      2146.75    
2017 I 1921   2159.125 88.97122677
      2171.5    
  II 2343   2181.25 107.4154728
      2191    
  III 2514   2180.12 115.3147533
      2169.25    
  IV 1986   2145.62 92.56065846
      2122    
2018 I 1834   2070 88.59903382
      2018    
  II 2154   1994.62 107.9904944
      1971.25    
  III 2098   1971.62 106.4099573
      1972    
  IV 1799   1955.87 91.97952829
      1939.75    
2019 I 1837   1965.5 93.46222335
      1991.25    
  II 2025   2012 100.6461233
      2030.75    
  III 2304   2062.25 111.722633
      2091.75    
  IV 1965   2140.37 91.80655681
      2189    
2020 I 2073   2193.37 94.51209782
      2197.75    
  II 2414   2198 109.8271156
      2198.25    
  III 2339      
           
  IV 1967      
 % of actual MA values

 Year QI QII QIII QIV

2016     114.8 89.6
2017 89 107.4 115.3 92.6
2018 88.6 108 106.4 92
2019 93.5 100.6 111.7 91.8
2020 94.5 109.8    
Mean of
Remaining 91.25 107.7 113.25 91.9

Modified Mean  
  QI 91.25
  QII 107.7
  QIII 113.25
  QIV 91.9
Total
Quarter Unadjust X Adjusting Seasonal
indices=404.1 ed constant Index
Indices
QI 91.25 X 0.9899 90.3
Adjusting constant = 400/404.1 = .9899 QII 107.7 X 0.9899 106.6
QIII 113.25 X 0.9899 112.1
QIV 91.9 X 0.9899 91
        400
Ex. Solve the following problem
Year (1) Quarter (2) Occupancy (3) 4 quarter 4-quarter centered % of actual MA values
MA (4) MA (5) (6)= (3)/(5)*100
2017 I 87      
           
  II 106      
      101    
  III 86   100.75 85.3598
      100.5    
  IV 125   101 123.7624
      101.5    
2018 I 85   101.125 84.0544
      100.75    
  II 110   101 108.9109
      101.25    
  III 83   101.125 82.0766
      101    
  IV 127   100.375 126.5255
      99.75    
2019 I 84   100.25 83.7905
      100.75    
  II 105   100.875 104.0892
      101    
  III 87   101.5 85.7143
      102    
  IV 128   101.875 125.6442
      101.75    
2020 I 88   101.875 86.3804
      102    
  II 104   101.5 102.4631
      101    
  III 88      
           
  IV 124      
   I II III IV
2017     85.4 123.8
2018 84.1 108.9 82.1 126.5
2019 83.8 104.1 85.7 125.6
2020 86.4 102.5    

Modified Median  
2017 QI 84
2018 QII 104
2019 QIII 85
2020 QIV 125
398
Quarter Unadjusted X Adjusting Seasonal
Indices constant Index
QI 84 X 1.005 84.42
Adjusting constant = 400/398 = 1.01 QII 104 X 1.005 104.52
QIII 85 X 1.005 85.42
QIV 125 X 1.005 125.62
  398     399.99
A problem involving all components of time series
Year (1) Quarter (2) Occupancy (3)
2016 I 16
     
  II 21
     
  III 9
     
  IV 18
     
2017 I 15
 
 
 
II
 
20 1. Deseasonalise the time series
 
 
 
III
 
10 2. Develop the trend line
 
  IV
   
18 3. Find cyclical variation around trend line
     
2018 I 17
     
  II 24
     
  III 13
     
  IV 22
     
2019 I 17
     
  II 25
     
  III 11
     
  IV 21
     
2020 I 18
     
  II 26
     
  III 14
     
  IV 25
Year (1) Quarter (2) Sales (3) 4 quarter MA (4) 4-quarter centered MA (5) % of actual MA
values (6)=
(3)/(5)*100
2016 I 16      
           
  II 21      
      16    
  III 9   15.875 56.6929
      15.75    
  IV 18   15.625 115.2000
      15.5    
2017 I 15   15.625 96.0000
      15.75    
  II 20   15.75 126.9841
      15.75    
  III 10   16 62.5000
      16.25    
  IV 18   16.75 107.4627
      17.25    
2018 I 17   17.625 96.4539
      18    
  II 24   18.5 129.7297
      19    
  III 13   19 68.4211
      19    
  IV 22   19.125 115.0327
      19.25    
2019 I 17   19 89.4737
      18.75    
  II 25   18.625 134.2282
      18.5    
  III 11   18.625 59.0604
      18.75    
  IV 21   18.875 111.2583
      19    
2020 I 18   19.375 92.9032
      19.75    
  II 26   20.25 128.3951
      20.75    
  III 14      
           
  IV 25      
   I II III IV
2016     56.7 115.2
2017 96.0 127.0 62.5 107.5
2018 96.5 129.7 68.4 115.0
2019 89.5 134.2 59.0604 111.2583
2020 92.9032 128.4    
 Total of
Remaining 188.9  258.1  121.6  226.3 

Modified Mean    
  QI 94.45  
  QII 129.05  
  QIII 60.80  
  QIV 113.15  
397.45 Quarter Unadjusted X Adjusting Seasonal Index
Indices constant
QI 94.45 X 1.0064 95.05448
QII 129.05 X 1.0064 129.87592
QIII 60.8 X 1.0064 61.18912
400/397.45=1.0064 QIV
 
113.15
397.45
X
 
1.0064
 
113.87416
399.99368
Year (1) Quarter (2) Sales (3) Seasonal Index (4) Depersonalized Sales (5)
=(3)/(4) * 100
2016 I 16 95.05 16.83
         
  II 21 129.87 16.3
         
  III 9 61.18 14.8
         
  IV 18 113.87 15.9
         
2017 I 15 95.05 15.9
         
  II 20 129.87 15.5
         
  III 10 61.18 16.4
         
  IV 18 113.87 15.9
       
2018 I 17 95.05 18.0
         
  II 24 129.87 18.6
         
  III 13 61.18 21.4
         
  IV 22 113.87 19.4
       
2019 I 17 95.05 18.0
         
  II 25 129.87 19.4
         
  III 11 61.18 18.1
         
  IV 21 113.87 18.6
       
2020 I 18 95.05 19.1
         
  II 26 129.87 20.1
         
  III 14 61.18 23.0
Now calculate trend b/w deseasonlized demand (DD)and time

DD (Y) Time (X)

16.83 -19
   
16.3 -17
   
14.8 -15
 
15.9
 
-13
a= 18, b=0.16
Ŷ = 18+0.16 X
   
15.9 -11
   
15.5 -9
   
16.4 -7
   
15.9 -5
   
18 -3
   
18.6 -1
   
21.4 1
   
19.4 3
   
18 5
   
19.4 7
   
18.1 9
   
18.6 11
   
19.1 13
   
20.1 15
   
23 17
   
21.95 19
Cyclical variation
Percent of trend =
DD (Y) a+bX = Ŷ 100* Y/Ŷ
16.83 18+.016(-19) 14.96 112.9679
16.3 18+.16 (-17) 15.28 106.6754
14.8 18+.16 (-15) 15.6 94.87179
15.9 18+.16 (-13) 15.92 99.87437
15.9 18+.16 (-11) 16.24 97.9064
15.5 18+.16 (-9) 16.56 93.59903
16.4 18+.16 (-7) 16.88 97.1564
15.9 18+.16 (-5) 17.2 92.44186
18 18+.16 (-3) 17.52 102.7397
18.6 18+.16 (-1) 17.84 104.2601
21.4 18+.16 (1) 18.16 117.8414
19.4 18+.16 (3) 18.48 104.9784
18 18+.16 (5) 18.8 95.74468
19.4 18+.16 (7) 19.12 101.4644
18.1 18+.16 (9) 19.44 93.107
18.6 18+.16 (11) 19.76 94.12955
19.1 18+.16 (13) 20.08 95.11952
20.1 18+.16 (15) 20.4 98.52941
23 18+.16 (17) 20.72 111.0039
21.95 18+.16 (19) 21.04 105.038
 Components of an Observation
Observed demand (O) = Systematic component (S) + Random component (R)
Level (current deseasonalized demand)

Trend (growth or decline in demand)

Seasonality (predictable seasonal fluctuation)

• Systematic component: Expected value of demand

• Random component: The part of the forecast that deviates
from the systematic component
• Forecast error: difference between forecast and actual demand
 Time Series Forecasting Methods
• Goal is to predict systematic component of demand
– Multiplicative: (level)(trend)(seasonal factor)
– Additive: level + trend + seasonal factor
– Mixed: (level + trend)(seasonal factor)

 Static: (companies estimate various parts like; level,trend,seasonality of (S) once

and do not update even if they observes new demand)
 Adaptive: (companies estimate various parts like; level,trend,seasonality of (S)
and update after every new demand is observed )
– Moving average
– Simple exponential smoothing
– Holt’s model (with trend)
– Winter’s model (with trend and seasonality)
 Static Methods
• Assume a mixed model: We should predict S and estimate R
Systematic component = (level + trend)(seasonal factor)
The forecast in period t for demand in period t + 1
Ft+1 = [L + (t + 1)T]St+1

L = estimate of level for period 0 (The deseasonalised demand

estimate for period 0)
T = estimate of trend (Increase or decrease for demand period)
St = estimate of seasonal factor for period t
Dt = actual demand in period t
Ft = forecast of demand in period t
 Static Methods
• Estimating level and trend
• Estimating seasonal factors
 Estimating Level and Trend
• Before estimating level and trend, demand data must
be deseasonalized
• Deseasonalized demand = demand that would have
been observed in the absence of seasonal fluctuations
• Periodicity (p)
– the number of periods after which the seasonal cycle
repeats itself
– for demand at Tahoe Salt (Table ) p = 4
 Time Series Forecasting
Forecast demand for the next four quarters.
Quarter Demand Dt
II, 1998 8000
III, 1998 13000 50,000
IV, 1998 23000 40,000

I, 1999 34000 30,000

20,000
II, 1999 10000
10,000
III, 1999 18000 0
IV, 1999 23000
I, 2000 38000
II, 2000 12000
III, 2000 13000
IV, 2000 32000
I, 2001 41000
 Deseasonalizing Demand

[Dt-(p/2) + Dt+(p/2) + S 2Di] / 2p for p even

Dt = (sum is from i = t+1-(p/2) to t+1+(p/2))

S Di / p for p odd
(sum is from i = t-(p/2) to t+(p/2)), p/2 truncated to lower integer
 Quarter Demand Dt

Deseasonalizing Demand II, 1998

III, 1998
IV, 1998
8000
13000
23000
I, 1999 34000
II, 1999 10000
III, 1999 18000
IV, 1999 23000
I, 2000 38000
II, 2000 12000
For the example, p = 4 is even III, 2000
IV, 2000
13000
32000

For t = 3:
I, 2001 41000

D3 = {D1 + D5 + Sum(i=2 to 4) [2Di]}/8

= {8000+10000+[(2)(13000)+(2)(23000)+(2)(34000)]}/8
= 19750
D4 = {D2 + D6 + Sum(i=3 to 5) [2Di]}/8
= {13000+18000+[(2)(23000)+(2)(34000)+(2)(10000)]/8
= 20625
Period t Demand Dt Deseasonalised demand
1 8000
2 13000
3 23000 19750
4 34000 20625
5 10000 21250
6 18000 21750
7 23000 22500
8 38000 22125
9 12000 22625
10 13000 24125
11 32000
12 41000
 Deseasonalizing Demand
Then include trend
Dt = L + tT
where Dt = deseasonalized demand in period t
L = level (deseasonalized demand at period 0)
T = trend (rate of growth of deseasonalized
demand)
Trend is determined by linear regression using deseasonalized demand
as the dependent variable and period as the independent variable (can be
done in Excel)
X Deseasonalised
Demand
3 19750
4 20625
5 21250
6 21750
7 22500
8 22125
9 22625
10 24125

Deseasonalised demand Dt for any period t is thus as follows,

L = 18,439 and T = 524

Dt = L + tT = 18439+524t
Calculate deseasonalised demand from regression.

Dt = L + tT = 18439+524t,

Period t Demand(Dt) Dt Dt (Reg.)

1 8000 18963 18439+524*1 =18963
2 13000 19487
3 23000 19750 20011
4 34000 20625 20535
5 10000 21250 21059
6 18000 21750 21583
7 23000 22500 22107
8 38000 22125 22631
9 12000 22625 23155
10 13000 24125 23679
11 32000 24203
12 41000 24727
 Time Series of Demand

50000
40000
Demand

30000 Dt
20000 Dt-bar

10000
0
1 2 3 4 5 6 7 8 9 10 11 12
Period
 Estimating Seasonal Factors

St = Dt / Dt = seasonal factor for period t

S2 = 13000/19487 = 0.67

The seasonal factors for the other periods are calculated in

the same manner
 Estimating Seasonal Factors

t Dt Dt Dt/Dt S
1 8000 18963 8000/18963 0.42
2 13000 19487 13000/19487 0.67
3 23000 20011 23000/20011 1.15
4 34000 20535 34000/20535 1.66
5 10000 21059 10000/21059 0.47
6 18000 21583 18000/21583 0.83
7 23000 22107 23000/22107 1.04
8 38000 22631 38000/22631 1.68
9 12000 23155 12000/23155 0.52
10 13000 23679 13000/23679 0.55
11 32000 24203 32000/24203 1.32
12 41000 24727 41000/24727 1.66
 Estimating Seasonal Factors

The overall seasonal factor for a “season” is then obtained by

averaging all of the factors for a “season”
If there are r seasonal cycles, for all periods of the form pt+i,
1<i<p, the seasonal factor for season i is
Si = [Sum(j=0 to r-1) Sjp+i]/r
In the example, there are 3 seasonal cycles in the data and p=4,
so
S1= (S1+S5+S9)/3= (0.42+0.47+0.52)/3 = 0.47
S2=(S2+S6+S10)/3=(0.67+0.83+0.55)/3 = 0.68
S3=(S3+S7+S11)/3= (1.15+1.04+1.32)/3 = 1.17
S4=(S4+S8+S12)/3= (1.66+1.68+1.66)/3 = 1.67
 Estimating the Forecast
Using the original equation, we can forecast the next
four periods of demand:
Ft+1 = [L + (t +1)T]St+1

F13 = (L+13T)S13 = [18439+(13)(524)](0.47) = 11868

F14 = (L+14T)S14 = [18439+(14)(524)](0.68) = 17527
F15 = (L+15T)S15 = [18439+(15)(524)](1.17) = 30770
F16 = (L+16T)S16 = [18439+(16)(524)](1.67) = 44794