ISOM 2700: Operations Management

Session 12: Revenue Management (Forecasting)

Dongwook Shin
Dept. ISOM, HKUST Business School
 Course Roadmap
Bottleneck
Little’s law
Utilization

Control chart
Acceptance sampling
Six sigma

Demand forecasting

Maximize
Profits

Decision tree method

Linear programming

Waiting time
Server utilization
1
Learning Objectives: Session 12
• Forecasting in Operations Management

• Quantitative techniques for forecasting

• Simple and weighted moving averages
• Exponential smoothing
• Linear regression

• Measuring forecast accuracy

• Qualitative techniques for forecasting

2
Revenue Management
• Revenue management is the application of disciplined
analytics that predict consumer behavior and optimize
capacity and price to maximize revenue

• Common approaches and tactics

• Passive role to forecast and respond to customer demand

• Focus of this class

• Active role to influence or prioritize customer demand

• Later

3
Principles of Demand Forecasting
• Forecasting is only forecasting
I see that you will…
• Not perfect

• The longer the forecast horizon,

the worse the forecast

• Aggregate forecasts are more accurate

• Good forecasts do not always require the use of complex

forecasting models
• “Simplicity is the ultimate sophistication”, Leonardo da Vinci

4
Components of Time Series Data

5
Components of Time Series Data: Trend
and Seasonal Variation

6
Components of Time Series Data:
Random Variation

7
Learning Objectives: Session 12
• Forecasting in Operations Management

• Quantitative techniques for forecasting

• Simple and weighted moving averages
• Exponential smoothing
• Linear regression

• Measuring forecast accuracy

• Qualitative techniques for forecasting

8
Simple Moving Average

Week Demand A t-1 + A t-2 + A t-3 +...+A t- n

1 650 Ft =
2 678 n
3 720
4 785
5 859 Question:
6 920
7 850 What are the 3-week and 9-week
8 758 moving average forecasts for
9 892 demand week 13?
10 920
11 789
12 844
9
How to Choose n?

10
Weighted Moving Average
• A weighted moving average allows any weights to be
placed on each element

Ft = w1 At -1 + w2 At -2 + ... + wn At -n
• Model
n

åw
i =1
i =1

• Choosing weights
• Most recent data with higher weighting
• If the data are seasonal, weights should reflect this
appropriately
• Experience and trial and error
11
Example

Question: Given the weekly demand and weights, what is

the forecast for the 4th period or Week 4?
Week Demand Weights:
1 650 t-1 0.5
2 678
t-2 0.3
3 720
4
t-3 0.2

Note that the weights place more emphasis on the

most recent data, that is time period “t-1”

12
Simple vs. Weighted Moving Average
• Simple Moving Average
• A forecast based on average past demand
• Assign equal importance to each component

• Weighted Moving Average

• A forecast made with past data where more recent data are
usually given more significance than older data

• A drawback for both methods

• Tuning the parameters (”n” and “weights”) is not easy
• Need to continually carry (potentially a large amount of)
historical data
13
Exponential Smoothing
Only three pieces of data
• Model: are needed to forecast the
future

Ft = Ft -1 + a ( At -1 - Ft -1 )
• !" = demand forecast for the coming time period
• !"#$ = demand forecast in the past time period
• %"#$ = actual demand in the past time period
• & = smoothing constant

14
Example

• Assume that the last month’s forecast was 1050 units

• It turns out that 1000 units were actually demanded

• Assume smoothing constant !=0.05

• Forecast for this month:

15
 The higher the
Effect of ! value of !,
the more closely
the forecast
follows the actual

16
 Does Exponential Smoothing Use Distant
Data?
Ft = Ft -1 + a ( At -1 - Ft -1 )
= (1 - a ) Ft -1 + aAt -1
= (1 - a )[(1 - a ) Ft - 2 + aAt - 2 ] + aAt -1
= (1 - a ) 2 Ft - 2 + a (1 - a ) At - 2 + aAt -1
= (1 - a ) 3 Ft -3 + a (1 - a ) 2 At -3 + a (1 - a ) At - 2 + aAt -1

a Past period Two periods ago Three periods ago

0.1 0.1 0.09 0.081
0.5 0.5 0.25 0.125
0.9 0.9 0.09 0.009
17
Linear Regression for Time Series
• Regression identifies the relationship as a function
between two or more correlated variables

• Linear Regression is the special class of regression where

the relationship between variables forms a straight line
Independent
Y=a+bX variable

Dependent
variable

• Time series forecasting: X represents time (day, week, …)

• Causal relationship forecasting: X may represents anything
other than time as a leading indicator
18
Linear Regression Analysis

• There are n data points: {(!" , $" ): % = 1, … , )}

• Forecast $*" = + + -!"
• Minimize squared error

6 6
8 8
min 3 $" − $*" = 3 $" − + + -!"
1,2
"45 "45

19
Least Squares Method
• The least squares method determines the parameters a and b such
that the sum of the squared errors is minimized
%
)
("
! &" − &
"#$

20
Least Squares Method
• Optimal coefficients a* and b* that minimize the sum of
the squared errors:

∗
∑%&' )% *% − ,)- *-
(
! =
∑(%&' )%. − ,)- .
∗ -
/ =*−! ) ∗-

- average of )%
• ):
- average of *%
• *:
21
Example: Credit Analysis
• A bank wants to know whether the credit amount is
higher for new customers or for old customers
Duration of Duration of
Credit Credit
credit credit
amount amount
(months) (months)
6 1169 15 1403
48 5951 24 1282
12 2096 24 2424
42 7882 30 8072
24 4870 24 12579
36 9055 24 3430
24 2835 9 2134
36 6948 6 2647
12 3059 10 2241
30 5234 12 1804
12 1295 10 2069
48 4308 6 1372
12 1567 6 426
24 1199
22
Example: Credit Analysis (continued)
20000

18000

16000

14000

12000
Credit Amout

10000

8000

6000

4000

2000 y = 146.3x + 213.22

0
0 10 20 30 40 50 60 70 80
Duration of Credit (in months)

We can also perform multi-variable linear regression:

Credit Amount = a + b1(Duration)+b2(Salary)+b3(Age)… 23
Discussion
• Advantages of Linear Regression

• Disadvantages of Linear Regression

24
Learning Objectives: Session 12
• Forecasting in Operations Management

• Quantitative techniques for forecasting

• Simple and weighted moving averages
• Exponential smoothing
• Linear regression

• Measuring forecast accuracy

• Qualitative techniques for forecasting

25
Measuring Forecast Accuracy
You are a marketing analyst for McDonalds and have the
following sales forecasts ($M) using two methods

Method 1 Method 2
Month Actual Demand
Forecast Forecast

1 100 60 100
2 100 130 100
3 200 200 150
4 200 270 200
5 400 340 250

26
Measuring Forecast Accuracy
• Forecast error is the difference between the actual value
and the predicted value
• Error = Actual – Forecast Measures the
average magnitude
of forecast errors

• Mean absolute deviation (MAD)

• MAD = ∑|Actual – Forecast| / n Measures the
bias of the
forecast

• Tracking signal (TS): Ratio of cumulative error and MAD

• TS = ∑(Actual – Forecast) / MAD

27
 Method 1: Tracking Signal

Month Actual Forecast Error ∑|Error| MAD ∑(Error) TS

1 100 60

2 100 130

3 200 200

4 200 270

5 400 340

Error = Actual – Forecast MAD = ∑|Error| / n TS = ∑(Error) / MAD

28
 Method 1: Tracking Signal

Month Actual Forecast Error ∑|Error| MAD ∑(Error) TS

1 100 60 40

2 100 130 -30

3 200 200 0

4 200 270 -70

5 400 340 60

Error = Actual – Forecast MAD = ∑|Error| / n TS = ∑(Error) / MAD

29
 Method 1: Tracking Signal

Month Actual Forecast Error ∑|Error| MAD ∑(Error) TS

1 100 60 40 40 40

2 100 130 -30 70 10

3 200 200 0 70 10

4 200 270 -70 140 -60

5 400 340 60 200 0

Error = Actual – Forecast MAD = ∑|Error| / n TS = ∑(Error) / MAD

30
 Method 1: Tracking Signal

Month Actual Forecast Error ∑|Error| MAD ∑(Error) TS

1 100 60 40 40 40 40

2 100 130 -30 70 35 10

3 200 200 0 70 23 10

4 200 270 -70 140 35 -60

5 400 340 60 200 40 0

Error = Actual – Forecast MAD = ∑|Error| / n TS = ∑(Error) / MAD

31
 Method 1: Tracking Signal

Month Actual Forecast Error ∑|Error| MAD ∑(Error) TS

1 100 60 40 40 40 40 1.00

2 100 130 -30 70 35 10 0.29

3 200 200 0 70 23 10 0.43

4 200 270 -70 140 35 -60 -1.71

5 400 340 60 200 40 0 0.00

Error = Actual – Forecast MAD = ∑|Error| / n TS = ∑(Error) / MAD

32
 Method 2: Tracking Signal

Month Actual Forecast Error ∑|Error| MAD ∑(Error) TS

1 100 100 0 0 0 0 0.00

2 100 100 0 0 0 0 0.00

3 200 150 50 50 16.7 50 3.00

4 200 200 0 50 12.5 50 4.00

5 400 250 150 200 40 200 5.00

Error = Actual – Forecast MAD = ∑|Error| / n TS = ∑(Error) / MAD

33
 Interpreting Tracking Signals

Method 2

3
2 Actual exceeds forecast
Tracking Signal

1
0
-1 Actual is less than forecast
Method 1
-2
-3
0 1 2 3 4 5
Month

34
Learning Objectives: Session 12
• Forecasting in Operations Management

• Quantitative techniques for forecasting

• Simple and weighted moving averages
• Exponential smoothing
• Linear regression

• Measuring forecast accuracy

• Qualitative techniques for forecasting

35
Qualitative Forecasting Methods (1)
• Market research
• Sets out to collect data in a variety of ways (surveys, interviews,
etc.) to test hypothesis about the market

• Typically used to forecast long-range and new-product sales

36
Qualitative Forecasting Methods (2)
• Historical analogy
• Ties what is being forecast to an existing product (such as
complementary product, substitutable product)

• Important in planning new products where a forecast may be

derived by using the history of a similar product

37
Qualitative Forecasting Methods (3)
• Panel consensus

• Free open exchange at meetings

• The idea is that discussion by the

group will produce better
forecasts than any one individual

• Participants may be executives,

salespeople or customers

38
Qualitative Forecasting Methods (4)
• Delphi method
• Experts respond to questions

• A moderator compiles results and formulates a new

questionnaire which is submitted to the group (perhaps with
new set of questions)

• There is a learning process for the group as it receives new

information and there is no influence of group pressure or
dominating individuals

39
Qualitative Forecasting Methods: Pros
and Cons
• Advantage of qualitative forecasting methods
• Does not require extensive historical data

• Disadvantage of qualitative forecasting methods

• Subjective

• When to use?
• When historical data are scarce or not available at all
• Use expert and/or customers opinion to predict future events
subjectively
• Example: sales of new product, environment and technology
change over the long term

40
Takeaways
• Understand the limitations of the qualitative &
quantitative forecasting methods
• Focus on the quantitative forecasting models
• Simple & Weighted moving averages
• Exponential smoothing
• Linear regression
• Measure accuracy of forecasting methods in terms of the
magnitude and direction of error

• Next class: Midterm exam review

41