Forecasting

MGT 3050 Management Science

Md Asnyat Asmat

Introduction

An essential aspect of managing any organization is planning for the future The long run success of an organization depends upon how well the managers anticipate or foresee the future & consequently develop the appropriate strategies

Md Asnyat Asmat

Introduction

There is some tendency for managers to forecast using their experience & intuition this type of forecast based upon human judgments are acceptable sometimes but not necessarily viable all the time

Md Asnyat Asmat

Forecasting Methods Quantitative Time Series Smooting Trend Projection Trend Projection Seasonal Adjustment Causal Regression Qualitative Delphi

Moving Average Exponential Smoothing

Simple Weighted
Md Asnyat Asmat 4

Introduction

We should forecast as accurate as possible need to use scientific forecasting method to help us make good decisions A number of forecasting methods have been developed
Quantitative methods or statistical forecasting method by utilizing historical data Qualitative or judgmental forecasting methods when historical data is not available or difficult to obtain

Md Asnyat Asmat 5

Introduction

Quantitative forecast methods

Past information about the variable being forecast is available Information can be quantified Assumption that the pattern of the past will continue into the future

Md Asnyat Asmat

Introduction

Use historical data to forecast the future

Historical data provide some pattern of the past (eg. sales), leading to a better prediction of the future (sales) A time series is a set of observations of a variable measured at successive point in time or over successive periods of time The objective of analyzing past data is to provide good forecasts or predictions of future values of time series

Historical data forms a time series

Forecasting procedure by using historical data is called time series method

Md Asnyat Asmat 7

Components of a Time Series

Trend component Cyclic component Seasonal component Irregular or Random

Md Asnyat Asmat

Components of Time Series

Trend Component

In a time series, the data may gradually increase or decrease the gradual shifting of upward or downward movement of the data over time is referred to as trend
If the time series data are recurring sequence of points or repeated after a certain number of periods, we call the time series has cyclical component
Md Asnyat Asmat 9

Cyclical Component

Components of Time Series

Seasonal component

The component of the time series that represents the variability in the data due to seasonal influences is called seasonal component
A time series that consist of components whose occurrence is totally unpredictable or having random variability is referred to as irregular component (ie. it does not follow any discernable pattern)
Md Asnyat Asmat 10

Irregular or random component

Time Series Smoothing Methods

Applicable only when the time series does not have any clear trend or seasonal influence the data are relatively stable over time

Moving average

Simple moving average Weighted moving average

Exponential moving average (also called exponential smoothing)

The objective is to smooth out the random fluctuations caused by the irregular components of the time series ( also called smoothing method)
Md Asnyat Asmat 11

Simple Moving Average

Simple Moving Average (s.m.a.) uses the average of the most recent n data values in the time series as the forecast for the next period
(most recent n data values)
Moving average = n

Md Asnyat Asmat

12

Simple Moving Average

Simple Moving Average (s.m.a.)

di
tn+1 = i=1 n n

d1, d2, ,dn are most recent data & n is number of periods

Md Asnyat Asmat

13

Eg. Demand of computers at Amir shop

Month January February March April May June July Demand 61 66 60 75 71 70 77

August September October

Md Asnyat Asmat

80 66 70
14

Forecast for November

Say n = 3

Forecast for November

= =

80 + 66 + 70 3 72

Md Asnyat Asmat

15

Forecast error = Actual data Forecast data 75 72 = 3

Assume actual demand for November is 75

Md Asnyat Asmat

16

Month

Demand

S.M.A Forecast

Forecast Error

Squared Forecast Error

January

61

February
March April May

66
60 75 71 (61+66+60)/3 = 62 (66+60+75)/3 = 67 13 4 169 16

June
July August September

70
77 80 66

69
72 73 76

1
5 7 -10

1
25 49 100

October
November December

70
75 67

75
72 70
Md Asnyat Asmat

-4
3 -3

16
9 9 394
17

Simple moving average

Forecast accuracy

Measured by computing the average of the sum squared errors called mean square error (MSE)
MSE for s.m.a. =
394 9

= 43.78

Md Asnyat Asmat

18

Weighted Moving Average

In simple moving average, each observation in the calculation receives the same weightage Weighted moving average involves selecting different weights for each data value & compute the weighted average of the most recent n data values as the forecast

Md Asnyat Asmat

19

Weighted Moving Average

Weighted Moving Average (w.m.a.)

w id i
w.m.a. n

wi
i=1

i=1 n

Md Asnyat Asmat

20

Month

Demand

W. M.A Forecast

Forecast Error

Squared Forecast Error

January February March April

61 66 60 75

3 60 2 66 1 61 62
3 2 1

13

169

May
June July August September October November

71
70 77 80 66 70 75

3 75 2 60 1 66 68
3 2 1

3
0 6 6 -11 -2 5

9
0 36 36 121 4 25

70 71 74 77 72 70
Md Asnyat Asmat

December

67

82

-5

25

21

Weighted moving average

Forecast accuracy

Measured by computing mean square error (MSE)

MSE for w.m.a. = 425 9

= 47.22

Md Asnyat Asmat

22

MSE for s.m.a. = 43.78

<

MSE for w.m.a. = 47.22

Md Asnyat Asmat

23

Exponential smoothing

Uses weighted average of past time series values as the forecast

select only one weight the weight of the most recent observation The weights for the other data values are computed automatically & get smaller and smaller as it moves away from the past

Md Asnyat Asmat

24

Exponential smoothing:
Mathematically: Ft+1 = Yt + (1- )Ft where Ft+1= new forecast for period t+1

Ft = old forecast for period t Yt = actual figure for period t = smoothing constant, (0 1)

Md Asnyat Asmat

25

Month

Demand

Exponential Smoothing Forecast ( = 0.4)

Forecast Error

Squared Forecast Error 25

January
February

61
66

61 (assumed)
(0.4 61) + (0.6 61) = 61 5

March
April May

60
75 71

(0.4 66) + (0.6 61) = 63

62 67

-3
13 4

9
169 16

June
July August September

70
77 80 66

69
69 72 75

1
8 8 -9

1
64 64 81

October
November December

70
75 67

72
71
Md Asnyat Asmat

-2
4 -6

4
16 36
26

73

Exponential smoothing

Forecast accuracy

Measured by computing mean square error (MSE)

MSE for exp. smoothing = 485 11 = 44.10

Md Asnyat Asmat

27

Selection of smoothing constant

Ft+1= Yt+(1- )Ft If the time series is stable (less variability), higher value of is preferred, otherwise lower Can find good value of by trial & error

Md Asnyat Asmat

28

MSE

0.1 0.2
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Md Asnyat Asmat

70.23 52.56
46.14 43.97 43.96 45.64 47.87 51.25 55.39 60.18
29

MSEs for the three smoothing methods

Smoothing Methods
Simple Moving Average
Weighted Moving Average Exponential Smoothing ( = 0.5)
Md Asnyat Asmat

MSEs
43.78
47.22 43.96

30

Time Series Trend Projection

In trend projection, we find out the trend and then project the trend in the future Applicable when data values exhibit a consistent increase or decrease over time This is called trend projection Statistical method least square method

Md Asnyat Asmat

31

Time Series Trend Projection

Linear equation of straight line ie. the line that passes through the scatter plot which minimizes the sum of squares of the vertical differences from the line to each of the actual observations

T1 = b0 + b1t
t = independent variable (eg. time) T = dependent variable (eg. demand) b0 = intercept of trend line b1 = slope of the trend line
Md Asnyat Asmat 32

Time Series Trend Projection

b1 =
b0 =
Where Y = t

tYt ( t Yt)/n

t2 ( t )2/n
Y b1t
Y1
n t n Actual value of time Yt = series in period t

n =

number of observations
Md Asnyat Asmat 33

Trend Projection
Month January February March April May June July Demand 52 57 56 60 65 69 75

Forecast for the month of August.

Md Asnyat Asmat 34

Yt

tYt

t2

1 2 3 4 5 6 7 t = 28

52 57 56 60 65 69 75 Yt = 434

52 114 168 240 325 414 525 tYt = 1838

1 4 9 16 25 36 49 t2 = 140

Md Asnyat Asmat

35

Trend and Seasonal Components:

Step 1: Compute seasonal indexes Step 2: Deseasonalize the data. Step3: Find out the trend for the deseasonalized data by trend projection and then forecast Step 4: Finally this forecasted figures have to be adjusted with the seasonal indexes.

Md Asnyat Asmat

36

YEAR

QUARTER 1

SALES 3

2
1 3 4 1

9
6 2 4

2
3 4 1

11
8 3 5 15 11 3
37

2 3 4
Md Asnyat Asmat

Year

Qtr

1 2
1 3 4 1 2 2 3 4 1

Sales (1000s) 3 9 6

4-qtr m.a.

Centered m.a.

Seasonal irreg.component

5.00 5.13 5.25 2 5.50 0.36 0.67 1.72 1.21 0.41 0.62 1.76 1.17

5.75
4 6.25 11 6.50 8 6.75 3 7.75 5 8.50 15 8.50 3 4 11 3
Md Asnyat Asmat 38

6.00 6.38 6.63 7.25 8.13 8.50

Quarter

Seasonal -Irregular Component Values 0.67 0.62

Seasonal Index (0.67+0.62)/ 2= 0.65 1.74

2 3
4

1.72 1.17
0.36

1.76 1.21
0.41

1.19
0.39

Md Asnyat Asmat

39

The final seasonal indexes are as follows:

Quarter
1 2 3 4

Index
0.65(4/3.97)=0.665 1.753 1.199 0.393

Md Asnyat Asmat

40

Deseasonalization of the Time Series

Simply divide the time series observations by the corresponding seasonal indexes.

Year
1

Quarter 1 2 3 4 1 2 3 4 1 2 3 4

Yt 3 9 6 2 4 11 8 3 5 15 11 3
Md Asnyat Asmat

St 0.655 1.753 1.199 0.393 0.655 1.753 1.199 0.393 0.655 1.753 1.199 0.393

Yt/St 4.58 5.13 5.00 5.09 6.11 6.27 6.67 7.63 7.63 8.56 9.17 7.63
41

t 1 2 3 4 5 6 7 8 9 10 11 12 Total=78

Yt 4.58 5.13 5.00 5.09 6.11 6.27 6.67 7.63 7.63 8.56 9.17 7.63 79.47

tYt 4.58 10.26 15.00 20.36 30.55 37.62 46.69 61.04 68.67 85.60 100.87 91.56 572.60

t2 1 4 9 16 25 36 49 64 81 100 121 144 650

42

Md Asnyat Asmat

Final forecast
t Trend Seasonal Forecast Index Quarterly Forecast

13 14 15 16

9179 9572 9966 10359

0.655 1.753 1.199 0.393

6012 16780 11949 4071

Md Asnyat Asmat

43

Linear Regression Analysis

Regression analysis is a statistical technique used to develop mathematical equation showing how variables are related
Variable that is being predicted is called dependent or response variable Variables being used to predict the value of the dependent variable are called independent or predictor variable

Md Asnyat Asmat

44

Linear Regression Analysis

Simple linear regression

Regression analysis involving one independent variable & one dependent variable
Regression analysis involving two or more independent variables

Multiple regression analysis

Md Asnyat Asmat

45

Linear Regression Analysis

Adv (1000s) 4 7 9 13 Sales (1000s) 23 49 48 58

16

61

Forecast for advertisement expenditure = $20,000

Md Asnyat Asmat 46

x 4

y 23

xy 92

x2 16

7
9

49
48

343
432

49
81

13
16 x = 49

58
61 y = 239

754
976 xy = 2597

169
256 x2 = 571

Md Asnyat Asmat

47

Correlation Coefficient

Correlation coefficient measures the strength of relationship between dependent variable and independent variable(s)
Denoted by r Lies between 0 and 1 (i.e. 0 r 1)

Formula for coefficient of correlation

r =

nxy - x y [nx2 (x)2 ] [ny2 (y)2 ]

Md Asnyat Asmat 48

Correlation Coefficient
x y xy x2 y2

4
7 9 13 16 x = 49

23
49 48 58 61

92
343 432 754 976

16
49 81 169 256

529
2401 2304 3364 3721

y = 239 xy = 2597 x2 = 571 y2 = 12319

Md Asnyat Asmat

49

Qualitative Forecasting methods

Jury of Executive Opinion Sales force composite Consumer market survey Delphi technique

Md Asnyat Asmat

50

Jury of executive opinion

Involves a small group of high profile managers who pool their best judgments to collectively make the forecast

Md Asnyat Asmat

51

Sales force composite

Bottom-up approach

Each sales person estimates future demand which are then passed through the organizations ladder & ultimately synthesized to make final forecast

Md Asnyat Asmat

52

Consumer market survey

Survey potential customers about their future purchasing plans Useful for new products

Md Asnyat Asmat

53

Qualitative approaches to forecasting

Delphi method Expert judgment Scenario writing Intuitive approaches

Md Asnyat Asmat

54

Delphi method

Develop forecast through group consensus

Panel of experts asked to respond to a series of questionnaires First responses tabulated & used to prepare second set of questionnaires Respondents asked to reconsider previous response in light of group information provided Exercise continue until some degree of consensus has been reached

Md Asnyat Asmat 55

Expert judgment

Based on judgment of single expert or consensus of a group of experts Often recommended when conditions in the past are not likely to hold in the future

Md Asnyat Asmat

56

Scenario writing

Consists of developing a conceptual scenario of the future based on a well defined set of assumptions

Different sets of assumptions lead to different scenarios

Md Asnyat Asmat

57

Intuitive approach

Based on the ability of the human mind to process information that is difficult to quantify Used in group works, a committee or a panel to develop new ideas or solve complex problems through a series of brainstorming sessions

Md Asnyat Asmat

58

Taco Bell
Number of outlets world-wide = 6,500. Annual sales revenue = $4.6 billion. Labor costs at Taco Bell are approximately 30% of every sales dollar.

52% of demand occurs in the 3-hour lunch time.

Taco Bell achieved labor savings of over $40 million from 1993 to 1996 using the new labor management system of which forecasting model was an integral part.

Md Asnyat Asmat

59

Md Asnyat Asmat

60

Md Asnyat Asmat

61

Md Asnyat Asmat

62