FORECASTING

– Part 5 & 6
Dr. A NOORUL HAQ
Contents:
• Forecasting Techniques
o Double Exponential Smoothing
• Seasonal Influences

2
Double Exponential Smoothing
• This method is used when the demand series has a trend.
• When a trend is present, the average of the series is
systematically increasing or decreasing over the time.
• Single exponential smoothing approaches must be modified.
• Otherwise, the forecast will always be below or above the actual
demand.
• An estimate of the current trend is the difference between the
simple averages of the series computed for current period, and
the average computed for the last period.

3
• To obtain a better estimate of a long term trend the effects of
random causes can be reduced by averaging the current
estimate.
• The method of arriving at the estimate of trend is similar to the
method used to get the estimate of average with single
exponential smoothing.
• The method for incorporating a trend in an exponentially
smoothed forecast is called ‘double Exponential smoothing.’
• Because in this method, the estimate of average is smoothed as
well of the estimate of the trend.
4
The calculations are given below.

• Average, At = αDt + (1- α) (At-1 +Tt-1)

• Current estimate of trend, CTt = At - At-1

• Average trend, Tt = β CTt + (1- β) Tt-1

• Forecast, Ft+1 = At + Tt

5
• where, At = exponentially smoothed average of series in period t.
CTt = current estimate of trend in period t
Tt = exponentially smoothed average of trend in period t
Ft+1 = forecast for the next periods.
α = smoothing parameter with a value between 0-1
β = smoothing parameter with a value between 0-1
• An initial estimate for average trend is needed to get started.
• These estimates can be derived from past data or based on past
experience guess can be made.

6
Problem 1:
For the demand data shown in the following table, calculate the forecast with
α=0.2, β=0.2, T0 = 9 and A0 =480.
Month Demand
t Dt
March 460
April 510
May 520
June 495
July 475
August 560
September 510
October 520
November 540
December 550
January 555
February 569 7

March -
Solution 1:
Month Demand Average Trend Forecast
t Dt At Tt Ft+1= At + Tt
- - 480 9.00 -
March 460 483.2 7.84 489
April 510 494.8 8.60 491.0
May 520 506.7 9.30 503.4
June 495 511.8 8.40 516.0
July 475 511.2 6.60 520.2
August 560 526.2 8.30 517.8
September 510 529.6 7.30 534.5
October 520 533.6 6.60 536.9
November 540 540.2 6.60 540.2
December 550 547.4 6.80 546.8
January 555 554.4 6.80 554.2
February 569 562.7 7.10 561.2 8

March - - - 569.8
Solution 1:
March:
Amarch = αDt + (1- α) (At-1 +Tt-1) = (0.2*460) + (0.8*(480+9))
= 92 + 391.2
= 483.2
Tmarch = 0.2*(483.2-480) +(0.8*9)
= 0.64 + 7.2
= 7.84
Fapril = Amarch + Tmarch = 483.2 + 7.84
= 491.04
9
April:
Aapril = αDt+(1-α)(At-1+Tt-1) = 0.2*510 + 0.8*(483.2+7.84)
= 102+392.83
= 494.83

Tapril = 0.2*(494.8-483.2) + 0.8*7.84

= 2.32 + 6.27
= 8.59

Fmay = Aapril + Tapril = 494.83 + 8.60

= 503.4 10
Problem 2:
Compute the double exponential forecast for week 1 for a firm with the following
data. Assume the initial forecast as ‘600’ and corresponding initial trend as ‘0’. Let
α=0.1 and β=0.2.

Week Demand
t Dt
week 1 650
week 2 600
week 3 550
week 4 650
week 5 625
Week 6 675
week 7 700
week 8 710 11
Solution 2:

Week Demand Average Trend Forecast

t Dt At Tt Ft+1= At + Tt
- - 600.00 0 -
week 1 650 605.00 1.000 606.00
week 2 600 605.40 0.880 606.28
week 3 550 600.65 -0.246 600.40
week 4 650 605.36 0.742 606.10
week 5 625 607.99 1.120 609.11
week 6 675 615.70 2.440 618.14
week 7 700 626.33 4.080 630.41
week 8 710 738.37 5.670 644.04

12
Seasonal Influences
• Many organizations experience seasonal demand for their
product.
• Volume of letters processed by postal department increases
dramatically during Christmas holiday period.
• Demands for products such as clothing, room heater/room
cooler, air conditioners are all have seasonal influences.
• A number of methods are available for forecasting time periods
with seasonal influences.
• One such method is multiplicative seasonal method.
13
Multiplicative Seasonal Method
• This method is a simple one and introduces the notion of
seasonal factors.
• The working of this method is explained with an example.
• Consider the data given in the following table which experiences
seasonal influences.
• In this data, the demand is very low in the first quarter and peak
in the third quarter.
• The total demand of year 1983 was 1000 units or an average of
250 units per quarter.
14
 Calculation of seasonal factors:

1983 1984 1985 1986 Average

Seasonal
Demand Seasonal Demand Seasonal Demand Seasonal Demand Seasonal
Quarter Factor
factor factor factor factor

1 45 0.18 70 0.23 100 0.22 100 0.18 0.20

2 335 1.34 370 1.23 585 1.30 725 1.32 1.30
3 520 2.08 590 1.97 830 1.84 1160 2.11 2.00
4 100 0.40 170 0.57 285 0.63 215 0.39 0.50
Total 1000 - 1200 - 1800 - 2200 - -
Quarterly
Average 250 300 450
Demand
15
• Seasonal factor can be calculated for each quarter by dividing the
actual demand in a quarter by the average quarterly demand for
that year.
• For example, in 1983, the first quarter demand was 18% of
average demand of 250 units.
• This gives an average estimate of the seasonal factor for the first
quarter in one year.
• This process is repeated for each year of the data given.
• The above table gives the estimates of each quarter seasonal
factors for the year 1983-1986.
16
• A simple way to arrive at one seasonal factor for each quarter is to
average the four estimates.
• This average seasonal factor is given in the last column of the
table.
• To use this method, the projected demand of 1987 is needed.
• The projection can be made in a number of ways.
• Causal method or qualitative method can be used for this
projection.
• Assume that the projection of demand for 1987 is 2600 units.
• Now the estimate of average quarterly demand is 2600/4 = 650
units.
17
• Then the quarterly forecast can be made by multiplying the
seasonal factor for each quarter by the projected average totally
demand.
• F1987,1 = 650*0.20 = 130,
• F1987,2 = 650*1.30 = 845,
• F1987,3 = 650*2.00 = 1300,
• F1987,4 = 650*0.50 = 325.

18
• At the end of each year the seasonal factor for that year is
calculated.
• Average seasonal factor for each quarter can be updated.
• By calculating the average of all seasonal factors for that
quarters.
• To have a control over the relevance of past demand pattern
average seasonal factor for each quarter can be updated.
• By calculating a moving average or single exponential smoothing
average.

19
Problem 2:
A courier service experiences a seasonal pattern of daily mail volume every week. The following
data for two weeks is expressed in thousands of pieces of mail.
Day Week-1 Week-2
Monday 20 15
Tuesday 30 30
Wednesday 35 30
Thursday 49 47
Friday 70 70
Saturday 15 10
Sunday 5 8

i) Calculate seasonal factors for each day of the week.

ii) If the courier service manager estimates that there will be 2,30,000 pieces of mail to sort on
next week. Forecast the volume for each day of the week. 20
Solution 2:

Day Week-1 Week-2 Average

Demand Seasonal Demand Seasonal seasonal
factor factor factor

Monday 20 0.625 15 0.50 0.563

Tuesday 30 0.938 30 1.0 0.969
Wednesday 35 1.094 30 1.0 1.047
Thursday 49 1.531 47 1.567 1.549
Friday 70 2.188 70 2.333 2.261
Saturday 15 0.469 10 0.333 0.401
Sunday 5 0.156 8 0.267 0.212
Total 224 210
Daily average 32 30
demand
21
Solution 2:
Estimate of mail for next week = 2,30,000 pieces.
Estimate of average daily mail = 2,30,000 / 7 = 32,857
Therefore, projected volume of mail for next week is,

FMonday = 32857*0.563 = 18498

FTuesday = 32857*0.969 = 31838
FWednesday = 32857*1.047 = 34401
FThursday = 32857*1.569 = 50895
FFriday = 32857*2.261 = 74290
FSaturday = 32857*0.401 = 13176
FSunday = 32857*0.212 = 6966
22
END

23