Demand Forecasting

Forecasting
• Forecasting is the process of estimating
future demand in terms of the quantity,
timing, quality and location for desired
products and services
 Types of Forecasts
• Long Term (2-10 years)
• Intermediate Term (1-24 months)
• Short Term (1-5 weeks)
 Long Term Forecast
• Time Horizon : 2 - 10 years
• Types of Products/services to offer
• Types & Sizes of Markets to serve
• Processes & Technologies to employ
• Plant location & size
 Intermediate Term Forecast
• Time Horizon : 1 - 24 months
• Size of Work Force to employ
• Kinds & amounts of inventories to maintain
• Amount of desired subcontracting when
needed
• Amount of desired overtime
 Short Term Forecast
• Time Horizon : 1 - 5 week
• Assignment of orders to special facilities
and personnel
• Dispatching to meet delivery times
 The Forecasting System
• Forecasting Outputs
• Forecasting Inputs
• Constraints
• Decisions
• Performance Criteria
 Forecasting System
Constraints Decisions
Data, Time, Experience, Selection of Data &
Funds Method
Inputs Forecasting Methods
Predictive Outputs
Internal
Data:Historical, Causal Estimates of
Subjective, Survey Time Series Long,Medium &
Routine Short Term Short Term
Environmental
Demand
Data: Social,
Economic, Forecast Error
Political, Performance Criteria
Technological
Accuracy, Stability,
Responsiveness, Objective,
Preparation Time
 Forecasting : Outputs
• Forecast of Expected Demand and not
future sales. Demand relates to orders
received, sales refer to shipments made.
• Output expressed in appropriate form -
marketing, production, finance.
• Translate demand for output units into
requirements for various production inputs.
 Forecasting System : Inputs
• Data from Internal and/or external sources.
• Historical Data in the form of a time series;
Expert opinions of an organization’s
personnel; Results of special surveys
• External sources : industry experts, private
consulting firms and government agencies.
• Internal for short/medium term, external for
long term.
Forecasting System : Constraints
• Time available to prepare a forecast
• Lack of relevant data from internal &
external sources.
• Quality of available data
• Expertise within the organization
• Available computing facilities
 Forecasting System : Decisions
• Decision with respect to data and method
• Data may be required in a particular form,
or may require adjustment or aggregation.
• For a long history of demand, care has to be
exercised for “how far back to go”
• Choice of method to prepare forecast will
depend on data available, time needed and
expertise that can be secured.
 Forecasting System :
Performance Criteria
• Accuracy of the forecast : Idle resources or
shortages
• Time required to prepare a forecast
• Benefit to cost ratio
 Forecasting Methods
• Predictive or Subjective (Estimates Survey &
Delphi Method)
• Causal (Regression Analysis, Econometric
Models & Input Output Models)
• Time Series (Trend, Seasonal, Cyclical,
Random)
• Routine Short Term Forecasting (Moving
Averages and Exponential Smoothing)
 Routine Short Term Forecasting

• Short Term forecasts are related to specific

product & services. It relies on historical data. The
method should be programmed on a computer for
frequent updates and forecasts for scheduling &
inventory control
• Simple & Weighted Moving Averages
• Exponential Smoothing
• Measurement & Control of Forecast Errors
 Simple & Weighted Moving
Averages
• Next Period’s demand is a simple moving
average. Equal or different weights to the
periods can be given.
• Fluctuations are smoothened with more
number of periods in the moving average (3, 5
or 7)
• Good when the demand is changing slowly.
• Can only forecast for one period ahead
 Moving Averages

Week Actual Forecast Absolute Forecast Absolute Forecast Absolute

Demand N = 3 wk Deviation N = 5 wk Deviation N = 7 wk Deviation
1 100.0
2 125.0
3 90.0
4 110.0
5 105.0
6 130.0
7 85.0
8 102.0 106.7 4.7 104.0 2.0 106.4 4.4
9 110.0 105.7 4.3 106.4 3.6 106.7 3.3
10 90.0 99.0 9.0 106.4 16.4 104.6 14.6
11 105.0 100.7 4.3 103.4 1.6 104.6 0.4
12 95.0 101.7 6.7 98.4 3.4 103.9 8.9
13 115.0 96.7 18.3 100.4 14.6 102.4 12.6
14 120.0 105.0 15.0 103.0 17.0 100.3 19.7
15 80.0 110.0 30.0 105.0 25.0 105.3 25.3
16 95.0 105.0 10.0 103.0 8.0 102.1 7.1
17 100.0 98.3 1.7 101.0 1.0 100.0 0.0
Total Absolute
Deviation 104.0 92.6 96.3
Mean Absolute
Deviation 10.4 9.3 9.6
Lowest
 Weighted Moving Averages
Equal Weights Weights 0.2,0.3,0.5 Weights 0.5,0.3.0.2
Week Actual Forecast Absolute Forecast Absolute Forecast Absolute
Demand N = 3 wk Deviation N = 3 wk Deviation N = 3 wk Deviation
1 100.0
2 125.0
3 90.0
4 110.0 105.0 5.0 102.5 7.5 105.5 4.5
5 105.0 108.3 3.3 107.0 2.0 111.5 6.5
6 130.0 101.7 28.3 103.5 26.5 99.0 31.0
7 85.0 115.0 30.0 118.5 33.5 112.5 27.5
8 102.0 106.7 4.7 102.5 0.5 108.5 6.5
9 110.0 105.7 4.3 102.5 7.5 110.9 0.9
10 90.0 99.0 9.0 102.6 12.6 95.1 5.1
11 105.0 100.7 4.3 98.4 6.6 102.0 3.0
12 95.0 101.7 6.7 101.5 6.5 103.0 8.0
13 115.0 96.7 18.3 97.0 18.0 95.5 19.5
14 120.0 105.0 15.0 107.0 13.0 104.0 16.0
15 80.0 110.0 30.0 113.5 33.5 106.0 26.0
16 95.0 105.0 10.0 99.0 4.0 109.5 14.5
17 100.0 98.3 1.7 95.5 4.5 103.0 3.0
Total Absolute
Deviation 104.0 106.7 102.5
Mean Absolute
Deviation 6.9 7.1 6.8
Lowest
 Exponential Smoothing
• Exponential smoothing takes the forecast of
the prior period and adds an adjustment to
obtain the forecast of the current period
• The adjustment is a proportion of the
forecast error (actual minus estimate)
• Forecastt= Forecastt-1 + a(Actualt-1 - Forecastt-1)
• a is a constant with value 0 to 1. Values of a are
0.1, 0.2, 0.3, or 0.5
 Exponential Smoothing

Week Actual Forecast Absolute Forecast Absolute Forecast Absolute

Demand a = 0.1 Deviation a = 0.2 Deviation a = 0.3 Deviation
1 100.0
2 125.0
3 90.0
4 110.0
5 105.0
6 130.0
7 85.0 85.0 85.0 85.0
8 102.0 85.0 17.0 85.0 17.0 85.0 17.0
9 110.0 86.7 23.3 88.4 21.6 90.1 19.9
10 90.0 89.0 1.0 92.7 2.7 96.1 6.1
11 105.0 89.1 15.9 92.2 12.8 94.2 10.8
12 95.0 90.7 4.3 94.7 0.3 97.5 2.5
13 115.0 91.1 23.9 94.8 20.2 96.7 18.3
14 120.0 93.5 26.5 98.8 21.2 102.2 17.8
15 80.0 96.2 16.2 103.1 23.1 107.5 27.5
16 95.0 94.6 0.4 98.5 3.5 99.3 4.3
17 100.0 94.6 5.4 97.8 2.2 98.0 2.0
Total Absolute
Deviation 133.8 124.5 126.1
Mean Absolute
Deviation 13.4 12.5 12.6
Lowest
 Causal Forecasting Methods
• Identify one or more variables which influence
demand. Select the form of relationships. Validate
the forecasting model to satisfy common sense and
statistical tests
• Regression analysis models - Two variable,
multiple variables
• Econometric Models
 Two Variable Regression Model
• Y = a + bX, Y= demand, X=causative factor
• Y = a + bX (Exponential)
• Y = a + bX + cX2 (Parabolic)
• Y = Y’ + Error
• Y’ = a + bX
• Method of Least Squares
• SumY = Na + bSumX
• SumXY = aSumX + bSumX2
• Value of a = (SumX2Sum Y – SumXSumXY)/(nSumX2 – (SumX)2)
• Value of b = (nSumXY – SumXSumY)/(nSumX2 – (SumX)2 )
• Coefficient of Determination r2 (where r = coeff of corelation)
• = (nSumXY – SumXSumY)/( SQRT{(nSumX2 – (SumX)2) (nSumY2 –
(SumY)2)}
• Explains the extent of total variation of Y by X
• Std Devn Sy = SQRT{(SumY2 – aSumY – bSumXY)/ n-2}
• Upper Limit = Y + t Sy t = 1.86 at 10%
• Lower Limit = Y - t Sy t = 1.86 at 10%
 Simple Linear Regression
Time Annual Time Annual
Period Sales Period Sales
Year X Y X2 XY Year X Y X2 XY
1 1 1000 1 1000 1 -5 1000 25 -5000
2 2 1300 4 2600 2 -4 1300 16 -5200
3 3 1800 9 5400 3 -3 1800 9 -5400
4 4 2000 16 8000 4 -2 2000 4 -4000
5 5 2000 25 10000 5 -1 2000 1 -2000
6 6 2000 36 12000 6 1 2000 1 2000
7 7 2200 49 15400 7 2 2200 4 4400
8 8 2600 64 20800 8 3 2600 9 7800
9 9 2900 81 26100 9 4 2900 16 11600
10 10 3200 100 32000 10 5 3200 25 16000

Total 55 21000 385 133300 0 21000 110 20200

Y = a + bX Y = a + bX
Sum Y = Na + bSumX Sum Y = Na + bSumX
Sum XY = aSumX + bSumX 2 Sum XY = aSumX + bSumX 2

21000 = 10a + 55b 21000 = 10a + 0

133300 = 55a + 385b 20200 = 0 + 110b

a = 913.32, b = 215.76 a = 2100, b = 183.64

Y = 913.32 + 215.76X Y = 2100 + 183.64X

11 11 3286.68 11 6 3201.84
12 12 3502.44 12 7 3385.48
 Subjective Forecasting Methods
• Rely on experience & opinions of people
inside or outside the organization.
• Employed when there is lack of time, data, or
introduction of a new product
• Estimates survey: Pooling of estimates made
by individual salesmen
• Delphi Method: Panel of experts respond to a
questionnaire.
 Estimates Survey
• Individual Salesmen submit estimates of demand
in their areas for a future period.
• These estimates are pooled at regional level and
adjusted for regional economic and demographic
factors
• Regional estimates are combined at headquarters
• Drawbacks are : recent experiences influence,
dominant personalities may divert from general
consensus, lack of measure for any accuracy.
 Delphi Method
• Panel of experts respond to a questionnaire
about future demand.
• Individual estimates are summarised and
returned to panel members for revision.
• Feedback allows arriving at a consensus
• Cost depends on panel composition and number
of rounds made.
• Used for Technological Forecasting
 Time Series

700

600

500

400
Demand

300

200

100

0
Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4
Quarter
 Time Series Forecasting Methods
• Time series data refers to a set of values of some
variable (Demand) measured at equally spaced time
intervals (Months, Quarters)
• Trend: long term growth or decline in the average level
of demand
• Cyclical: business cycle - large deviation of actual
demand values from those expected from a trend
• Seasonal: annually repetitive demand fluctuations
caused by weather, tradition.
• Random: irregular residual in the demand due to many
complex random forces in environment