Operations Management-II

Course details
 Course Instructor(s)
Abhishek Chakraborty ([email protected])
Alok Raj ([email protected])

Textbook:
“Operations Management” by William J Stevenson.
Session Topics
❖ Forecasting (AR)
❖ Scheduling (AR)
❖ Material requirement planning (AR)
❖ Aggregate requirement planning (AR)
❖ Project management (AR)
❖ Waiting linemanagement (AR)
❖ Quality management (AC)
❖ Supply Chain management (AC)
❖ Service management (AC)
❖ Theory of constraints (AC)
❖ Lean supply chain (AC)
Evaluation
❖ Assignments: 40
❖ Quiz: 40
❖ Class Quiz: 20

❖ There will be no make-up possible for a missed quiz and

the score in a missed quiz will be taken as 0.
 Forecasting
(Chapter3 )

Alok Raj
PODS Area
Office: Room No 14, 2nd Floor, Library Building, Tel. 3439
Email: [email protected]
Forecasting
 A statement or inference about the future, usually using
information about the past to make predictions about the
future.
Products/
Services
Demand

Good
Forecasting model
Relevance of forecasting
Year Passenger Vehicle Sales
2010 2501542
2011 2629839
2012 2665015
2013 2503509
2014 2601111
2015 2789678
2016 3047582
2017 3288581
2018 3377389
2019 2773575

https://www.wsj.com/articles/walgreen-shakeup-followed-bad-projection-1408494546, Aug. 19, 2014

Relevance of forecasting

https://covid19-projections.com/india
Types of forecasting
Strategic forecasts
❖ Decisions about overall directions.
❖ Used to help set the strategy of how demand will be met.
❖ 3-5 years
Tactical forecasts
❖ Market wise sales forecast , resource planning and budgeting
❖ 1-18 months
Operational forecasts
❖ Used to help the firm to operate on a day -to-day basis.
❖ 1 days -3 months
❖ Production, Inventory , scheduling
Principles of Forecasting

❖Forecasts are rarely perfect

❖Aggregate forecasts are more accurate

❖Forecasts are more accurate for shorter

than longer time horizons

A new Supply Chain Paradigm

❖ A shift from a Push System...

– Production decisions are based on forecast
❖ …to a Push-Pull System
 Push-Pull Supply Chains

Suppliers Customers
PUSH STRATEGY PULL STRATEGY

Low Uncertainty High Uncertainty

Push-Pull Boundary

The Supply Chain Time Line

Prediction is very difficult, especially it it’s about the future
—Neils Bhor
 Methods of forecasting
Demand Forecasting
(Independent demand) ❖ Independent
demand is demand for
Qualitative a finished product,
Quantitative
such as a car,
computer, a bicycle, or
Market
research Time series Causal Analysis a pizza.

Simple moving
average
Holt’s exponential
smoothing
❖ Dependent demand,
Delphi Method on the other hand,
is demand for
Simple Exponential Winter’s exponential
smoothing smoothing component parts or
subassemblies. For
Regression analysis Naïve
example, this would
be the wheels on the
bicycle, or the cheese
on the pizza.
Primary focus
Qualitative vs. Quantitative Forecasting

Qualitative Quantitative

• Subjective • Objective
Characteristic – Based on people’s – Based on numeric data
opinions and equations

• Can incorporate • Consistency

Strength expertise that is hard • Can consider large amounts
to codify of data

• Opinions can •Must have data

Weakness dominate/and or bias
the forecast
Qualitative
❖ Generally used to take advantage of expert knowledge.
❖ Useful when judgment is required, when products are new, or if the firm has little
experience in a new market.
❖ Text mining approach
 Qualitative Methods
Market research
❖ Forecasts derived from customer
responses.
❖ Ask customers about their purchase plans.
❖ What consumers say, and what they actually
do are often different: fairly inaccurate
forecasts.
❖ Used for long range and new product sales.

Delphi Approach
❖ Forecasts derived from panel of experts.
❖ It is an iterative method.
Quantitative Methods
❖ Time series data: Historical data to make forecasts
Demand ( in Lakhs) Passenger Vehicle Sales Air passenger
(In India) movements (in Lakhs)
35
4000000 450
30 3500000 400
25 3000000 350
300
2500000
20 250
2000000
200
15 1500000 150
10 1000000 100
500000 50
5 0
0
0

Level Level +Trend Level + Trend+ Seasonality

Do you think forecasting approach will be same for all type of products
Components of Demand

Seasonal
peaks Trend
component
Demand for product or

Actual
demand
line

Average
demand over
service

four years
Random
variation
Year Year Year Year
1 2 3 4

Demand=Systematic part (level +trend +seasonality)+ Random part (error)

Method 1 : The Naïve Method
● Ft=At-1 : Simplest Approach to Forecasting

Period Demand Forecast

Forecast of Demand
in Period 2
1 130 -
(This forecast made
after seeing
Demand in Period 2 155 130
1)
3 145 155
Forecast of Demand
in Period 5 4 160 145
(This forecast made
after seeing 5 151 160
Demand in Period
4)
6 143 151

7 143
 Method 2 : The Simple Average
● Ft=(A1+ A2+A3+…At-1)/(t-1)

Forecast of Period Demand Forecast

Demand in Period
2 (This forecast 1 130 -
made after seeing
Demand in Period
1) 2 155 130.00 130/1 =130

3 145 142.50 (130+155)/2 =142.50

4 160 143.33 (130+155+145)/3 = 143.33

Forecast of Demand
in Period 5 (This
forecast made after 5 151 147.50 (130+155+145+160)/4 = 147.5

seeing Demand in
Period 4) 6 143 148.20 (130+155+145+160+151)/5 = 148.2

7 147.33 (130+155+145+160+151+143)/6 = 147.33

The Moving Average Forecast
“The recent history is more relevant ”
● The Simple Average forecast uses ALL THE HISTORY of demands to
generate the forecast for the next period

● The (Simple) Moving Average forecast (order n) uses ONLY THE n

MOST RECENT period demands to generate the forecast for the
next period
Method 3 : The Simple Moving Average
● Forecast for period t = the average of demand in the past n
periods (from period t-1 tot-n)

A
At-t-11++ AA t-t- 22+A
+A t-t- 33+...+A
+...+A t-nt-n
F =
F t=
n

● At-1 is the actual sales in period t-1

 Method 3 : The Simple Moving Average(Cont’d)

Period Demand Forecast Using Using

Forecast of Demand n=3 n=4
in Period 2 (This
forecast made after 1 130 -
seeing Demand in
Period 1) 2 155 - Not enough history

Forecast of Demand
in Period 5 (This
3 145 - Not enough history

forecast made after

seeing Demand in 4 160 143.33 (130+155+145)/3 = 143.33

Period 4)
5 151 153.33 (155+145+160)/3 = 153.33

6 143 152.00 (145+160+151)/3 = 152.00

7 151.33 (160+151+143) /3 = 151.33

Method 3 : The Simple Moving Average (Cont’d)
● The nine-week average is smoother than three-week average.

165

160

155

150

145

140

135

130

125

120
1 2 3 4 5 6 7

Demand 2 period 3 period 4 period

 The Weighted Moving Average Forecast
❖The (simple) Moving Average forecast (order n) treats each
of the n most recent demands EQUALLY in generating the
forecast for the next period

❖The Weighted Moving Average forecast (order n) weights

each of the n most recent demands (possibly) DIFFERENTLY
in generating the forecast for the next period
Method 4: Weighted Moving Average (WMA)
● When a detectable trend or pattern is present, weights can be
used to place more emphasis on recent values

● Weights based onintuition

❖ Weights are values between 0 and 1
❖ Weights sum to 1.0
❖ Weights impact stability and responsiveness of the forecast

● Formula

Ft= w1At-1+ w2 At-2+w3 At-3+...+wn At-n

 Method 4 : Weighted Moving Average (WMA)

n=3:
Forecast of Demand
in Period 2 (This
Period Demand Forecast wt-1=0.5, w t-2=0.3,
forecast made after
wt-3=0.2
seeing Demand in 1 130 -
Period 1)
2 155 - Not enough history

3 145 - Not enough history

Forecast of 4 160 145.00 0.2(130)+0.3(155)+0.5(145) =

145.00
Demand in Period
5 (This forecast 5 151 154.50 0.2(155)+0.3(145)+0.5(160) =
154.50
made after seeing
Demand in Period
4) 6 143 152.20 0.2(145)+0.3(160)+0.5(151) =
152.50

7 148.80 0.2(160)+0.3(151)+0.5(143) =
148.80
Question-1
Day Number Sold
a) If a two-period moving average had been used to
1 25 forecast sales, what would the daily forecasts have
2 31 been starting with the forecast for Day 3?
3 29 b) If a four-period moving average had been used
4 33 determine what the forecasts would have been for
5 34 each day, starting with Day 5.
6 37 c) Use a three-period weighted moving average with
7 35 w1=0.2, w2=0.3, and w3=0.5 and forecast sales for
8 32 the 4th day onwards.
9 38 d) Plot the original data and each set of forecasts on
10 40 the same graph. Which forecast has the better
11 37 ability to respond quickly to changes?
12 32
Method 5 : Exponential Smoothing
Smoothing constant

● Ft = Ft-1 + α(At-1 - Ft-1)= α(At-1 )+(1- α) Ft-1

Forecast= Previous forecast + α(previous actual sales- previous forecast )

▪ At-1 is the actual sales in the previous period
▪ Ft-1 is the forecast for the previous period
▪ α: ranges from 0 to 1 and is subjectively chosen.
 Method 5 : Exponential Smoothing (Cont’d)

Forecast of Period Demand Forecast α=.2

Demand in Period
2 (This forecast
made after seeing
1 130 -
Demand in Period
1) 2 F2=αA1 + (1- α)F1
F2=(0.2)130 + (1- 0.2)F1
Problem!
Didn’t have a forecast for period 1 so how can we start the
exponential smoothing method
Options
(1) Use Naïve Forecast for Period 2 and then do exponential smoothing for periods
3,4,5,6,……
(2) Choose an initial forecast for period 1 (in some manner) and thendo
exponential smoothing for periods 2,3,4,5,6,……
 Method 5 : Exponential Smoothing (Cont’d)
● Using Naïve Method for Period 2
Period Demand Forecast
α=.2
Forecast of
Demand in Period
1 130 -
2 (This forecast
made after seeing
Demand in Period
2 155 130 Using Naïve/Simple average

1)
3 145 135.00 0.2(155)+(1-0.2)(130)=135.00

4 160 137.00 0.2(145)+(1-0.2)(135)=137.00

Forecast of
Demand in Period 5 151 141.60 0.2(160)+(1-0.2)(137)=141.60
5 (This forecast
made after seeing
Demand in Period 6 143 143.48 0.2(151) + (1-0.2)(141.60) =143.48

4)

7 143.38 0.2(143) + (1-0.2)(143.48) = 143.38

Method 5 : Exponential Smoothing (Cont’d)
● Forecast effects of Smoothing Constant α
Ft = Ft-1 + α(At-1 - Ft-1) Ft = α At-1 + (1- α)Ft-1

Ft = α At - 1 + α(1- α )At - 2 +…+ α(1- α)t-2A1 + (1- α)t-1F1

Weights on Actual Sales

Prior Period 2 PeriodsAgo 3 PeriodsAgo

α α(1 − α) α(1 − α)2

α = 0.10

α = 0.90
 Calculation
 Ft = Ft-1 + α(At-1 - Ft-1)

Ft = α At - 1 + α(1- α )At - 2 +…+ α(1- α)t-2A1 + (1- α)t-1F1

What is the Effect of Smooth Constant α?

Bio Magnetic
Ear Stickers
for Weight
Loss

● Large alpha reacts better to the change in demand.

→ use large alpha for volatile demand.
● Small alpha offers better smoothing.
→use small alpha for stable demand.
Summary: Exponential Smoothing
● Most popular
● Because…
❖ Formulating an exponential model is relatively easy
and it is surprisingly accurate.
❖ Little computation is required so the computer
storage requirements are small

● It is a special form of weighted moving average

❖ Weights decline exponentially with most recent
data weighted most
Responsiveness vs. Stability Trade-off

Responsiveness: Ability
Responsiveness: of forecast
Ability to respond
of forecast quicklyquickly
to respond to a
true
to change
a true in mean
change in level
meandemand
level demand

small n large α

Moving Average Exponential Smoothing

large n small α

Stability: Ability
Stability: Abilityofofforecast to ignore
forecast simple
to ignore random
simple random
variations
variations
Reality of Forecasting
●Forecasts are usually wrong!
❖ Aggregate forecasts are more accurate than
individual forecasts
❖ Long-range forecasts are less accurate than short -
range forecasts

●Forecasts should, therefore, be accompanied by a

measure of forecasting error.
Forecast Errors
● Forecast error is the difference between the forecast value
and what actually occurred.
❖ Forecast errort = Actualt –Forecastt
❖ et =A t − Ft

● Measures of Error
❖ Mean absolute deviation (MAD)
❖ Mean absolute percent error (MAPE)
❖ Mean squared error (MSE)
Measuring Forecast Errors: Mean Absolute
Deviation (MAD)
nn

A
t=1
t - Ft
t=1
MAD
MAD==
n
❖ The ideal Mean Absolute Deviation (MAD)is zero which would mean there is
no forecasting error at all.

❖ The larger the MAD, the less the accurate the resulting model.

❖ Mean Absolute Percent Error (MAPE) gauges the error relative to the average
demand.
▪ MAPE = MAD / Average Demand
Measuring Forecast Errors: Mean Forecast Error
(MFE)

Period Demand Forecast Error

1 130 - -

2 155 130.00 25.00

3 145 155.00 -10.00

(25+(-10)+15+(-9)+(-8))/5
= 2.60
4 160 145.00 15.00

5 151 160.00 -9.00

6 143 151.00 -8.00

Mean Forecast
Error = 2.60
Measuring Forecast Errors: Mean Absolute
Deviation (MAD)

Absolute
Period Demand Forecast Error Error
1 130 - - -

2 155 130.00 25.00 25.00

3 145 155.00 -10.00 10.00

4 160 145.00 15.00 15.00

5 151 160.00 -9.00 9.00

6 143 151.00 -8.00 8.00

MAD = 13.40
Measuring Forecast Errors: Mean Absolute
Percentage Error (MAPE)

Abs %
Period Demand Forecast Error % Error
Error
1 130 - - - -
2 155 130.00 25.00 =25/155=16.13% 16.13%
3 145 155.00 -10.00 =-10/145=-6.90% 6.9%
4 160 145.00 15.00 =15/160=9.38% 9.38%
5 151 160.00 -9.00 =-9/151=-5.96% 5.96%
6 143 151.00 -8.00 =-8/143=-5.59% 5.59%
MAPE 8.79%
Measuring Forecast Errors: Mean Squared Error
(MSE)
Squared
Period Demand Forecast Error

Error
1 130 - - -

2 155 130.00 25.00 625.00

3 145 155.00 -10.00 100.00

4 160 145.00 15.00 225.00

5 151 160.00 -9.00 81.00

6 143 151.00 -8.00 64.00

MSE = 219.00
Measure of Forecast Error
1
 Mean squared error (MSE): 𝑀𝑆𝐸𝑛 = 𝑛 σ𝑛𝑡=1 𝐸𝑡2
 The MSE penalizes large errors more significantly than small efforts because all
errors are squared.
 Lower MSE implies better prediction
1 𝑛
 Root Mean squared error (RMSE): 𝑀𝑆𝐸𝑛 = σ 𝐸2
𝑛 𝑡=1 𝑡
1
 Mean Absolute deviation (MSD): 𝑀𝑆𝐷𝑛 = 𝑛 σ𝑛𝑡=1 𝐸𝑡
1 𝐸𝑡
 Mean Absolute percentage error (MAPE): 𝑀𝐴𝑃𝐸𝑛 = 𝑛 σ𝑛𝑡=1 ×100
𝐷𝑡
 MAPE is dimensionless it can be used for comparing different models with
varying scales.
 It is a good measure of forecast error when forecast error when the underlying
forecast has significant seasonality and demand varies considerably from one
period to the next

MAPE and RMSE are the popular forecasting accuracy indicators

Method 6 : Regression

Year Demand t 𝒀𝒕

2014 25 1 26
2015 32
2 28
2016 24
3 29
2017 28 30.16
2018 26 4 31

2019 27 5 32
2020
6 35

Average=181/6=30.16
Regression
t Y tY t2
෍ 𝑦 = 𝑛𝑎 + 𝑏 ෍ 𝑡
1 26 26 1

2 28 56 4 ෍ 𝑦𝑡 = 𝑎 ෍ 𝑡 + 𝑏 ෍ 𝑡 2
3 29 87 9
181=6a+21b
4 31 124 16 663=21a+91b
5 32 160 25

6 35 210 36

Sum=21 181 663 91

 Method 7: Trend-Adjusted Exponential
Smoothing (Holt’s model)
 The trend adjusted forecast consists of two components
❖ Smoothed error
❖ Trend factor

𝐹𝑡+1 = 𝐿𝑡 + 𝑇𝑡
𝐿𝑡 = 𝛼𝐷𝑡 + 1 − 𝛼 𝐿𝑡−1 + 𝑇𝑡−1
𝑇𝑡 = 𝛽 𝐿𝑡 − 𝐿𝑡−1 + 1 − 𝛽 𝑇𝑡−1

Alpha and beta are smoothing Period t Dt Lt Tt Ft

constants 1 26 24.27 1.69 ---
2 28
Trend-adjusted exponential smoothing 3 29
has the ability to respond to changes in 4 31
trend 5 32
6 35
7
Trend-Adjusted Exponential Smoothing
(Holt’s model)
Period t Dt Lt Tt Ft

1 26 24.27 1.69
2 28 26.36 1.81 25.95

3 29 28.34 1.86 28.17

4 31 30.36 1.91 30.19

5 32 32.21 1.89 32.26

6 35 34.28 1.94 34.10

7 36.23
Demand forecasting at Asian paints

 Asian paints found that in certain districts of Maharashtra

there is spike in demand for 50-100 ml packs of deep
orange shade during a specific period of year
Method 8: Data with seasonality
Sales Year 1 Year 2 Year 3 Year 4 Year 5 25000
Comparison between Demand & Forecast
JAN 2,000 3,000 2,000 5,000 5,000

FEB 3,000 4,000 5,000 4,000 2,000 20000

MAR 3,000 3,000 5,000 4,000 3,000

15000
APR 3,000 5,000 3,000 2,000 2,000

MAY 4,000 5,000 4,000 5,000 7,000

10000
JUN 6,000 8,000 6,000 7,000 6,000

JUL 7,000 3,000 7,000 10,000 8,000

5000
AUG 6,000 8,000 10,000 14,000 10,000

SEP 10,000 12,000 15,000 16,000 20,000

0
OCT 12,000 12,000 15,000 16,000 20,000 1 3 5 7 911131517192123252729313335373941434547495153555759616365676971

NOV 14,000 16,000 18,000 20,000 22,000

DEC 8,000 10,000 8,000 12,000 8,000

Total 78,000 89,000 98,000 1,15,000 1,13,000

Steps
 Step 1: Evaluate the seasonality index for each time
period within a block

 Step 2: De-seasonalise demand data

 Step3: Determine trend and level components for the de-

seasonalised data series

 Step 4: Finalise the forecast model and use the model for
evaluating the forecast numbers for past data points.