Forecasting Demand For Services

Sachin Modgil IMI-K

 Introduction
• Forecasting techniques allow us to translate the
multitude of information available from databases into
strategies that can give services a competitive
advantage.
• Three types of forecasting models (Subjective, Causal
and Time series based)
• The horizon of forecasting becomes shorter as we move
from subjective to Time series methods.
Classification of Forecasting Models

Forecastin
g methods

Time
Subjective Causal
series
models Models
models
Classification of Forecasting Models
Time
Subjective Causal
Series

Delphi Moving
Regression
Method average

Cross-
Exponential
Impact Econometric
smoothing
analysis

Historical
analogy
Characteristics of Forecasting Methods
 Delphi Method
• Developed at the Rand Corporation by Olaf Helmer, the
Delphi method is based on expert opinion.
• In its simplest form, persons with expertise in a given
area are asked questions, and these individuals are not
permitted to interact with each other.
• Typically, the participants are asked to make numerical
estimates.
• For example, they might be asked to predict the
highest NIFTY/ BSE average for the coming year.
 Delphi Method
• Developed at the Rand Corporation by Olaf Helmer, the
Delphi method is based on expert opinion.
• In its simplest form, persons with expertise in a given
area are asked questions, and these individuals are not
permitted to interact with each other.
• Typically, the participants are asked to make numerical
estimates.
• For example, they might be asked to predict the
highest NIFTY/ BSE average for the coming year.
 Delphi Method
• The test administrator tabulates the results into quartiles
and supplies these findings to the experts, who then are
asked to reconsider their answers in light of the new
information.
• Additionally, those whose opinions fall in the two outside
quartiles are asked to justify their opinions.
• All the information from this round of questioning is
tabulated and once again returned to the participants.
• On this occasion, each participant who remains outside the
middle two quartiles (i.e., the interquartile range) might be
asked to provide an argument as to why he or she believes
those at the opposite extreme are incorrect
 Delphi Method
An example of the Delphi method can be seen in a study of the
nuclear power industry. Ninety-eight persons agreed to participate in
this study.

It is desirable that utilities be permitted to integrate capital

investment costs more aggressively into rate structures .
 Cross-Impact Analysis
• Cross-impact analysis assumes that some future
event is related to the occurrence of an earlier
event.
• As in the Delphi method, a panel of experts
studies a set of correlations between events
presented in a matrix.
• These correlations form the basis for estimating
the likelihood of a future event occurring.
 Historical Analogy
• Historical analogy assumes that the introduction and
growth pattern of a new service will mimic the pattern
of a similar concept for which data are available.
• A famous use of historical analogy was the prediction of
the market penetration by color television based on the
experience with black-and-white television only a few
years earlier.
• For example, growth in the demand for housekeeping
services could follow the growth curve for child-care
services.
 Regression Models

Y  a0  a1 X 1  a2 X 2  ........  an Xn
a 1 , a 2 , . . . , a n are coefficients
 Econometric Models

• Econometric models are versions of regression

models that involve a system of equations.

• The equations are related to each other, and the

coefficients are determined as in the simpler
regression models
 Time Series Models

• N-Period Moving Average

• Simple Exponential Smoothing
• Exponential Smoothing with Trend Adjustment
• Exponential Smoothing with Seasonal Adjustment
 1. N-Period Moving Average
The hotel owner (100 room) has noted increased occupancy for the
past two Saturdays and wishes to prepare for the coming weekend
(i.e., August 17), perhaps by discontinuing the practice of offering
discount rates. Do the higher occupancy figures indicate a change in
the underlying average occupancy?
Saturday Occupancy
July, 6, 2019 79
July, 13, 2019 84
July, 20, 2019 83
July, 27, 2019 81
August, 3, 2019 98
August, 10, 2019 100
August, 17, 2019 ???
 1. N-Period Moving Average

Consider the 3 period moving range

Saturday Period Occupancy 3-priod Foreca
Rate moving range st
July, 6, 2019 1 79
July, 13, 2019 2 84
July, 20, 2019 3 83 82
July, 27, 2019 4 81 83 82
August, 3, 2019 5 98 87 83
August, 10, 2019 6 100 93 87
August, 17, 2019 7 93
 2. N-Period Moving Average
A XYZ refrigerator supplier has experienced the following
demand for refrigerator during past five months. Find out
the demand forecast for the month of July using five-period
moving average & three-period moving average using
simple moving average method.

Month Demand
February 20
March 30
April 40
May 60
June 45
 2. N-Period Moving Average

Consider the 5 period moving range

Saturday Period Occupancy 5-priod Foreca
Rate moving range st
February 1 20
March 2 30
April 3 40
May 4 60
June 5 45 39
39
 2. N-Period Moving Average

Consider the 3 period moving range

Saturday Period Actual 3-priod Foreca
Demand moving range st
February 1 20
March 2 30
April 3 40 30
May 4 60 44 30
June 5 45 49 44
49
 3. N-Period Moving average
Month (2018) Demand
Jan 200
Feb 190
Mar 180 Find Out
Apr 210 forecast
May 195
Jun 175 demand of
Jul 170 mobile phone
Aug 172 for month of
Sep 178 Jan, 2019 ,
Oct 230
Nov 225 employing 3-
Dec 224 months moving
average.
 N-Period Moving average
Month (2018) Demand (At) 3 –period moving range Forecast (F)
Jan 200    
Feb 190    
Mar 180 190  
Apr 210 193.33 190.000
May 195 195 193.333
Jun 175 193.33 195.000
Jul 170 180 193.333
Aug 172 172.33 180.000
Sep 178 173.33 172.333
Oct 230 193.33 173.333
Nov 225 211 193.333
Dec 224 226.33 211.000
January, 2019 226.333
 4. Weighted Moving average
Weightage of P1 = 0.25, Weightage of P2 = 0.35 and P3 = 0.40
Month (2018) At W-moving average Forecast (F)
Jan 200    
Feb 190    
Mar 180 188.5  
Apr 210 194.5 188.500
May 195 196.5 194.500
Jun 175 190.75 196.500
Jul 170 178 190.750
Aug 172 172.05 178.000
Sep 178 173.9 172.050
Oct 230 197.3 173.900
Nov 225 215 197.300
Dec 224 225.85 215.000
Jan, 2019 225.850
 4. Weighted Moving average
Weightage of P1 = 0.60, Weightage of P2 = 0.40

Week Demand
1 51 Calculate the
2 68 forecast through
3 68 2 period weighted
4 70 moving average.
5 73
6 75
7 76
8 77
 4. Weighted Moving average
Weighted-moving
Week At average Forecast (F)
1 51
2 68 57.8
3 68 68 57.8
4 70 68.8 68
5 73 71.2 68.8
6 75 73.8 71.2
7 76 75.4 73.8
8 77 76.4 75.4
76.4
 Simple Exponential Smoothing

Simple exponential smoothing is the time series

method most frequently used for demand forecasting.
Simple exponential smoothing also “smooths out”
blips in the data, but its power over the N -period
moving average is threefold:
(1) old data never are dropped or lost,
(2) older data are given progressively less weight, and
(3) the calculation is simple and requires only the most
recent data.
 Simple Exponential Smoothing

St = St-1 + ἀ (At – St-1)

St is the smoothed value for period t,
At is the actual observed value for period t, and
ἀ is a smoothing constant that usually is assigned a value
between 0.1 and 0.5.
Example-1- Exponential Smoothing
Given ἀ = 0.5
Saturday Occupancy
July, 6, 2019 79
July, 13, 2019 84
July, 20, 2019 83
July, 27, 2019 81
August, 3, 2019 98
August, 10, 2019 100
August, 17, 2019 ???
Example-1-Exponential Smoothing
 Example-2-Exponential Smoothing

Month (2018) At (Actual Demand)

Jan 200
Feb 190
Mar 180 Given ἀ = 0.2
Apr 210
May 195
Jun 175
Jul 170
August  
 Example-2-Exponential Smoothing

Month At- Abs(At- (At- Abs(At-

(2018) At St Ft Ft Ft) Ft)^2 Ft)/At
Jan 200 200          

Feb 190 198 200 -10 10 ` 5

Mar 180 194 198 -18 18 324 10
Apr 210 198 194 16 16 256 7
May 195 197 198 -3 3 9 1
Jun 175 193 197 -22 22 484 13
Jul 170 188 193 -23 23 529 13
August     188 -60 15 284 8
CFE MAD MSE MAPE
 Exponential Smoothing with Trend
Adjustment
The trend in a set of data is the average rate at which the
observed values change from one period to the next over time.
The changes created by the trend can be treated using an
extension of simple exponential smoothing.

St = St-1 + ἀ (At – St-1)

Exponential Smoothing with Trend Adjustment
Ex.1-Exponential Smoothing with Trend
Adjustment

Week (t) Load (At)

1 31 Given ἀ = 0.5
2 40 β= 0.3
3 43
4 52
5 49
6 64
7 58
8 68
Exponential Smoothing with Trend Adjustment
Ex.2-Exponential Smoothing with Trend
Adjustment

Week Demand
1 51 Given ἀ = 0.4
2 68 β= 0.2
3 68
4 70
5 73
6 75
7 76
8 77
 Ex.2-Exponential Smoothing with Trend
Adjustment

Week Forecast Abs (At-

(t) At St Tt Ft Error (At- Ft)
Ft)
1 51 51 0       0.4 =ἀ
2 68 57.8 1.36 51 17 17 0.2 =β
3 68 63.58 2.686 59 9 9
4 70 68.133 3.246 66 4 4
5 73 72.18955 3.489 71 2 2
6 75 75.33939 3.387 76 -1 1
7 76 77.3634 2.978 79 -3 3
8 77 78.6709 2.477 80 -3 3
5
MAD