9/3/2023

Learning Objectives

• Understand the importance of forecasting

• What are the different types of forecasting
Forecasting methodologies
• Different types of models which we can build for
the forecasting
• Discuss and calculate various methods for
evaluating forecast accuracy

Why forecasting is necessary?

Defining Forecasting

• Planning and control for operations requires an • So, planning and control for operations requires
estimate of the demand for the product or the an estimation of the demand. So, whether it is a
manufacturing function or a service function you
service that the organization expects to provide need to have some kind of estimation that, what is
in the future going to be the demand, and that estimation is
called forecasting

• This forecasting is equally responsible for product

organizations, the manufacturing organization or
the service organization

• Eg- Car Mfg OR Hospital Setup

1
 9/3/2023

The Effect of Inaccurate Forecasting Time span for forecasting

• Current needs/short term (1 day-30 days)
• Intermediate range (1 month- 12 months)
• Long range plans (more than a year)

Basic categories of Forecasting Basic categories of Forecasting

• Extrapolative methods • Extrapolative methods-Time Series
Quantitative Method/analysis
• Causal methods
• Qualitative Considers historical data or past data

The limitation of the Time Series analysis,

is that we are forecasting for the coming
period but the information, the data which
you are using for that purpose that is of the
previous period

2
 9/3/2023

Basic categories of Forecasting Basic categories of Forecasting

• Causal methods - cause effect method • Qualitative
Considers demand based on various factors

Eg- Demand of a product is directly related to the Historical data not available
advertisement expenses

Y = a + bX, where Y is demand and X is those Not knowing the factors which are affecting
independent factors the demand

Eg- Y = a + b1 X1 + b2 X2 + b3 X3

Basic categories of Forecasting Extrapolative Methods/Time series Methods

• Qualitative forecasting types For our good Time Series analysis, we need to
identify that pattern in the historical data

1. Horizontal component
Delphi
Market survey 2. Trend
Brainstorming
3. Seasonal

4. Cyclic

3
 9/3/2023

Comparative analysis of components of Forms of Forecast Movement

historical data
Horizontal Trend Seasonal Cyclic

can very easily requires some require more efforts require

forecast amount of efforts extraordinary efforts

some minor will not be a smooth after a particular when demand is

fluctuations are curve. There will be interval demand will going to increase,
there but more or zigzag movements increase to a high that we do not know
less it is around a but overall effect is level and for rest of
straight line either increasing or the period demand
decreasing. remains to a low
level

Comparative analysis of components of Comparative analysis of components of

historical data historical data
Product life cycle Failure rate of a product

4
 9/3/2023

Time Series Forecasting Simple Moving Average method

• We can pick models based on
1. Time horizon to forecast

2. Data availability

3. Accuracy required

4. Size of forecasting budget

5. Availability of qualified personnel

Simple Moving Average method Time Series Forecasting

5
 9/3/2023

Weighted moving average method Weighted moving average method

• Weighted moving average method is a method
which is differentiating between different periods.
It gives different weight to the different periods’
demand

• While the simple moving average formula gives

equal weight, the weighted moving average
method permits an unequal weighting on prior
time periods

Weighted Moving Average

Weighted moving average method
• Adjusts moving average method to more closely
reflect data fluctuations
n
WMAn =  W i Di
i=1
where
W i = the weight for period i,
between 0 and 100
percent
 Wi = 1.00

6
 9/3/2023

Weighted Moving Average Example Weighted Moving Average Example

MONTH WEIGHT DATA MONTH WEIGHT DATA
August 17% 130 August 17% 130
September 33% 110 September 33% 110
October 50% 90 October 50% 90
3 3

November Forecast WMA3 = 

i=1
W i Di November Forecast WMA3 =  W i Di
i=1

= (0.50)(90) + (0.33)(110) + (0.17)(130)

= 103.4 orders

Simple exponential smoothing

• In exponential smoothing method, we

assign weights, but these weights are
decreasing in the exponential order
from the present period to the past
periods

7
 9/3/2023

Exponential Smoothing

• Averaging method
• Weights most recent data more strongly
• Reacts more to recent changes
• Widely used, accurate method
• Smoothing constant, α
• applied to most recent data

Simple exponential smoothing Exponential Smoothing

Ft +1 = Dt + (1 - )Ft
St = St-1 + α (D
(Dt- St-1) where:
OR Ft +1 = forecast for next period
St = α Dt + (1-
(1-α) St-1 Dt = actual demand for present period
Here, we are updating the base value, updating the base value Ft = previously determined forecast for
means, we are trying to find out the current base value and this present period
current base value is nothing but the forecast for the next
= weighting factor, smoothing constant
period.
St = Ft+1
Therefore

Ft+1 = α Dt + (1-
(1-α) Ft

8
 9/3/2023

Simple exponential smoothing Effect of Smoothing Constant

0.0  1.0

If = 0.20, then Ft +1 = 0.20Dt + 0.80 Ft

If = 0, then Ft +1 = 0Dt + 1 Ft = Ft

Forecast does not reflect recent data

If = 1, then Ft +1 = 1Dt + 0 Ft =Dt

Forecast based only on most recent data

Simple exponential smoothing

• Use Simple moving average method with an

average period (AP) of 3 days to develop a
forecast of the call volume in the Day 13

F13= (168+198+159)/3=175 calls

9
 9/3/2023

Use Exponential smoothing

• If a smoothing constant value of 0.25 is used
• Use Weighted moving average method with an and the exponential smoothing forecast for day
average period (AP) of 3 days and weights of 11 was 180.76 calls, what is the exponential
0.1, 0.3 and 0.6 for oldest to recent datum to smoothing forecast for Day 13?
develop a forecast of the call volume in the Day
13 Ft+1 = α Dt + (1-
(1-α) Ft
F13= α D12 + (1-
(1-α) F12
F13=0.1(168)+0.3(198)+0.6(159)=171.6 calls F12= α D11 + (1-
(1-α) F11
= (0.25x198) + (0.75x180.76)
=185.07

Exponential Smoothing (α=0.30)

PERIOD MONTH DEMAND F2 = D1 + (1 - )F1

1 Jan 37
F13= α D12 + (1-α) F12 2 Feb 40
3 Mar 41
= (0.25x159) + (0.75x185.07) 4 Apr 37
F3 = D2 + (1 - )F2
5 May 45
=178.55 6 Jun 50
7 Jul 43
8 Aug 47
9 Sep 56
10 Oct 52
11 Nov 55
12 Dec 54

10
 9/3/2023

Exponential Smoothing (α=0.30) Exponential Smoothing

FORECAST, Ft + 1
PERIOD MONTH DEMAND F2 = D1 + (1 - )F1 PERIOD MONTH DEMAND ( = 0.3) ( = 0.5)
1 Jan 37
= (0.30)(37) + (0.70)(37) 1 Jan 37 – –
2 Feb 40
= 37 2 Feb 40
3 Mar 41
3 Mar 41
4 Apr 37
F3 = D2 + (1 - )F2 4 Apr 37
5 May 45
= (0.30)(40) + (0.70)(37) 5 May 45
6 Jun 50
6 Jun 50
7 Jul 43 = 37.9 7 Jul 43
8 Aug 47
8 Aug 47
9 Sep 56
9 Sep 56
10 Oct 52
10 Oct 52
11 Nov 55
11 Nov 55
12 Dec 54
12 Dec 54
13 Jan –

Exponential Smoothing Exponential Smoothing

FORECAST, Ft + 1
PERIOD MONTH DEMAND ( = 0.3) ( = 0.5)
1 Jan 37 – –
2 Feb 40 37.00 37.00
3 Mar 41 37.90 38.50
4 Apr 37 38.83 39.75
5 May 45 38.28 38.37
6 Jun 50 40.29 41.68
7 Jul 43 43.20 45.84
8 Aug 47 43.14 44.42
9 Sep 56 44.30 45.71
10 Oct 52 47.81 50.85
11 Nov 55 49.06 51.42
12 Dec 54 50.84 53.21
13 Jan – 51.79 53.61

11
 9/3/2023

Exponential Smoothing with Trend and Exponential Smoothing with Trend and
Seasonality Seasonality
• Simple Exponential Smoothing model was
having only one component that was the base
component and the fluctuations are around this
base value and we wanted to smoothen those
fluctuation
• Simple Exponential Smoothing doesn’t have any
trend and seasonality
• Winters and Pegels, over the period of 10 years,
from 1960s to 1970, developed exponential
smoothing models by incorporating the fact of
trend as well as seasonality in the historical data

Exponential Smoothing with Trend and Exponential Smoothing with Trend and
Seasonality Seasonality
• Hence, we have 9 different types of models and
we can develop our extrapolative or exponential
smoothing models for all these 9 types of
models
• When we have trend also in our data, we will
use one more smoothing constant that is beta,
beta will be used for smoothing the fluctuations
of your trend data
• When we have seasonality also in our demand
data, you will use one more smoothing constant
that is gamma

12
 9/3/2023

Smoothing Constants Smoothing Constants

α= Smoothing constant for average • We have possibility of using either alpha alone,
β = Smoothing constant for trend you can use alpha and beta, you can use alpha
and gamma and you can use alpha, beta,
γ = Smoothing constant for seasonality gamma altogether
• Use of any smoothing constant depends upon
the type of demand data
• There is no restriction on the combination of
alpha, beta and gamma’s value, the only
limitation is the boundary condition of the values
will vary between 0 to 1

Adjusted Exponential Smoothing/Trend

Double exponential smoothing corrected exponential smoothing/Holt’s model
• Incorporating a trend component into • ADJUSTED FORECAST
exponential smoothing forecast is also known as AFt +1 = Ft +1+Tt +1
double exponential smoothing
• Two smoothing constants, alpha and beta are Where
used Ft +1 = forecast of average for next period
• One is for smoothing the fluctuations of your Ft+1 = α Dt + (1-α) Ft
base value, and second the smoothing the Tt +1 = forecast of trend for next period
fluctuation of your trend value
Tt+1= (Ft+1 - Ft) + (1 - ) Tt

13
 9/3/2023

Trend corrected exponential Trend corrected exponential smoothing

smoothing(α=0.5 and β=0.30) (α=0.5 and β=0.30)
PERIOD MONTH DEMAND T3 = (F3 - F2) + (1 - ) T2 PERIOD MONTH DEMAND T3 = (F3 - F2) + (1 - ) T2
1 Jan 37 1 Jan 37 = (0.30)(38.5 - 37.0) + (0.70)(0)
2 Feb 40 2 Feb 40 = 0.45
3 Mar 41 3 Mar 41
4 Apr 37 AF3 = F3 + T3 4 Apr 37 AF3 = F3 + T3 = 38.5 + 0.45
5 May 45 5 May 45 = 38.95
6 Jun 50 6 Jun 50
7 Jul 43 T13 = (F13 - F12) + (1 - ) T12 7 Jul 43 T13 = (F13 - F12) + (1 - ) T12
8 Aug 47 8 Aug 47 = (0.30)(53.61 - 53.21) + (0.70)(1.77)
9 Sep 56 9 Sep 56
= 1.36
10 Oct 52 10 Oct 52
11 Nov 55 11 Nov 55
12 Dec 54 AF13 = F13 + T13 = 12 Dec 54 AF13 = F13 + T13 = 53.61 + 1.36 = 54.97

Trend corrected exponential smoothing/Holt’s

smoothing
model Trend corrected exponential smoothing
(α=0.5 and β=0.30)
FORECAST TREND ADJUSTED FORECAST TREND ADJUSTED
PERIOD MONTH DEMAND Ft +1 Tt +1 FORECAST AFt +1 PERIOD MONTH DEMAND Ft +1 Tt +1 FORECAST AFt +1

1 Jan 37 1 Jan 37 37.00 – –

2 Feb 40 2 Feb 40 37.00 0.00 37.00
3 Mar 41 3 Mar 41 38.50 0.45 38.95
4 Apr 37 4 Apr 37 39.75 0.69 40.44
5 May 45 5 May 45 38.37 0.07 38.44
6 Jun 50 6 Jun 50 38.37 0.07 38.44
7 Jul 43 7 Jul 43 45.84 1.97 47.82
8 Aug 47 8 Aug 47 44.42 0.95 45.37
9 Sep 56 9 Sep 56 45.71 1.05 46.76
10 Oct 52 10 Oct 52 50.85 2.28 58.13
11 Nov 55 11 Nov 55 51.42 1.76 53.19
12 Dec 54 12 Dec 54 53.21 1.77 54.98
13 Jan – 13 Jan – 53.61 1.36 54.96

14
 9/3/2023

Trend corrected exponential smoothing Holt’s model

• Example 2. (Assignment-2)
An electronics manufacturer has seen demand
for its latest Smartphone increase over the past
6 months. Observed demand in thousands has
been D1=8415, D2=8732, D3=9014, D4=9808,
D5=10413 and D6=11961. Forecast demand for
period 7 using trend corrected exponential
smoothing with α=0.1 and β =0.2.

Time Series Decomposition Seasonal variation (additive or multiplicative)

• Chronologically ordered data are referred to as a

time series
• A time series may contain one or many elements
• Trend, seasonal, cyclical, autocorrelation, and
random
• Identifying these elements and separating the
time series data into these components is known
as decomposition
• Seasonal variation may be either additive or
multiplicative

15
 9/3/2023

Seasonal factor/index Example: (seasonal factor/index)

• The seasonal factor is defined as the ratio of • In past years, a firm sold an average of 1,000
amount sold during each season to the average units each year
of all seasons • 200 in spring
• 350 in summer
• It is the amount of correction needed in a time
• 300 in fall
series to adjust for the season of the year • 150 in winter
• Find the seasonal factors
• Using those factors, if we expected demand for
next year to be 1,100 units, compute demand
per period

Example : Finding Seasonal Factors Example 18.3: Forecast for Next Year

16
 9/3/2023

Linear Regression Analysis Linear Regression Analysis

• Regression can be defined as the functional xy - nxy

relationship between two or more correlated y = a + bx b =
x2 - nx2
variables a = y-bx
where
• Linear regression refers to the special class of a = intercept
regression where the relationship between where
b = slope of the line n = number of periods
variables forms a straight line x = time period
x
• The major restriction in using linear regression y = forecast for x =
n
= mean of the x values
demand for period x
forecasting is that the past data and future y
projections are assumed to fall in about a y = n = mean of the y values
straight line

Least Squares Example Least Squares Example

x(PERIOD) y(DEMAND) xy x2 x(PERIOD) y(DEMAND) xy x2
1 37 1 37 37 1
2 40 2 40 80 4
3 41 3 41 123 9
4 37 4 37 148 16
5 45 5 45 225 25
6 50 6 50 300 36
7 43 7 43 301 49
8 47 8 47 376 64
9 56 9 56 504 81
10 52 10 52 520 100
11 55 11 55 605 121
12 54 12 54 648 144
78 557 3867 650

17
 9/3/2023

Least Squares Example Least Squares Example

x = x = 78 = 6.5
12
y = y = 557 = 46.42
12
b = xy - nxy = b = xy - nxy = 3867 - (12)(6.5)(46.42) =1.72
x2 - nx2 x2 - nx2 650 - 12(6.5)2

a = y - bx a = y - bx
= 46.42 - (1.72)(6.5) = 35.2

Trend and seasonality corrected exponential

Linear trend line y = 35.2 + 1.72x
Forecast for period 13 y = 35.2 + 1.72(13) = 57.56 units
smoothing (Winter’s model)
Example:
• A hospital wants to improve its forecasting by
applying both trend and seasonal indices to 66
months of data it has collected. It will then
forecast “patient days” over the coming year

18
 9/3/2023

Trend and seasonality corrected exponential Trend and seasonality corrected exponential
smoothing (Winter’s model) smoothing (Winter’s model)
Determine the forecast till December i.e till 78th period
Using 66 months of inpatient days the following
equation was computed
y= 8090+21.5x
Where y= patient days
x=time in months
Based on this model the patients day forecast for
67th month should be ?

y67= 8090+(21.5X67)= 9530 days

Trend and seasonality corrected exponential Trend and seasonality corrected exponential
smoothing (Winter’s model) smoothing (Winter’s model)
The following table provides seasonal indices
based on the same 6 6months data.

19
 9/3/2023

Trend and seasonality corrected exponential Trend and seasonality corrected exponential
smoothing (Winter’s model) smoothing (Winter’s model)
Determine the combined trend and seasonal forecast
 Neither trend data nor the seasonal data alone
provide a reasonable forecast for the hospital.
 Only when the hospital multiplied the trend
adjusted data with respective seasonal index, it
obtain good forecasts
Thus for 67th month the patient days=
(Trend adjusted forecast)(Monthly seasonal index)
=(9530X1.04)
=9911

Advantage of combined Trend and Seasonal Trend and seasonality corrected exponential
Adjustments smoothing (Winter’s model)
 Question:
 With trend only, the September forecast is
If the slope of the trend line for patient days
9702 but with both trend and seasonal
is 22 and the index for December is 0.99,
adjustments the forecast is 9411
what is the new forecast for December
inpatient days?
 By combining trend and seasonal data the
hospital is better able to forecast inpatient
days and the related staffing and budgeting Answer: 9708 days
vital to effective operations

20
 9/3/2023

Accuracy of Forecasts Mean Absolute Deviation (MAD)

• Forecast Error (FE)
Difference between forecast and actual demand  Dt - Ft 
MAD = n
FE= Ft –Dt
The popular methods of error measurement where
t = period number
• Mean Absolute Deviation (MAD) Dt = demand in period t
Ft = forecast for period t
• Mean Absolute Percent Deviation (MAPD) n = total number of periods
 = absolute value
• Cumulative Error (E)

• Average error or bias

MAD Example MAD Calculation

PERIOD DEMAND, Dt Ft ( =0.3) (Dt - Ft) |Dt - Ft|
 Dt - Ft 
1 37 37.00 – –
2 40 37.00
MAD = n
3 41 37.90
4 37 38.83
5 45 38.28
6 50 40.29
7 43 43.20
8 47 43.14
9 56 44.30
10 52 47.81
11 55 49.06
12 54 50.84

21
 9/3/2023

MAD Example MAD Calculation

PERIOD DEMAND, Dt Ft ( =0.3) (Dt - Ft) |Dt - Ft|
 Dt - Ft 
1 37 37.00 – –
2 40 37.00 3.00 3.00
MAD = n
3 41 37.90 3.10 3.10 53.39
4 37 38.83 -1.83 1.83 =
5 45 38.28 6.72 6.72 11
6 50 40.29 9.69 9.69
7 43 43.20 -0.20 0.20 = 4.85
8 47 43.14 3.86 3.86
9 56 44.30 11.70 11.70
10 52 47.81 4.19 4.19
11 55 49.06 5.94 5.94
12 54 50.84 3.15 3.15
557 49.31 53.39

Other Accuracy Measures Comparison of Forecasts

Mean absolute percent deviation (MAPD)

FORECAST MAD MAPD E (E)
|Dt - Ft| Exponential smoothing (= 0.30) 4.85 9.6% 49.31 4.48
MAPD =
Dt Exponential smoothing (= 0.50) 4.04 8.5% 33.21 3.02
Adjusted exponential smoothing 3.81 7.5% 21.14 1.92
Cumulative error (= 0.50, = 0.30)
Linear trend line 2.29 4.9% – –
E = et
Average error

et
(E) =
n

22
 9/3/2023

Forecast Control Tracking Signal Values

DEMAND FORECAST, ERROR E =
Tracking signal PERIOD Dt Ft D t - Ft (Dt - Ft) MAD

1 37 37.00 – – –
2 40 37.00 3.00 3.00 3.00
• It helps to detect any drift in the forecasting system 3 41 37.90 3.10 6.10 3.05
4 37 38.83 -1.83 4.27 2.64
5 45 38.28 6.72 10.99 3.66
6 50 40.29 9.69 20.68 4.87
7 43 43.20 -0.20 20.48 4.09
(Dt - Ft) E 8 47 43.14 3.86 24.34 4.06
Tracking signal = MAD = MAD 9 56 44.30 11.70 36.04 5.01
10 52 47.81 4.19 40.23 4.92
11 55 49.06 5.94 46.17 5.02
12 54 50.84 3.15 49.32 4.85

Tracking Signal Values Tracking Signal Plot

DEMAND FORECAST, ERROR E = TRACKING
PERIOD Dt Ft D t - Ft (Dt - Ft) MAD SIGNAL

1 37 37.00 – – – –
2 40 37.00 3.00 3.00 3.00 1.00
3 41 37.90 3.10 6.10 3.05 2.00
4 37 38.83 -1.83 4.27 2.64 1.62
5 45 38.28 6.72 10.99 3.66 3.00
6 50 40.29 9.69 20.68 4.87 4.25
7 43 43.20 -0.20 20.48 4.09 5.01
8 47 43.14 3.86 24.34 4.06 6.00
9 56 44.30 11.70 36.04 5.01 7.19
10 52 47.81 4.19 40.23 4.92 8.18
11 55 49.06 5.94 46.17 5.02 9.20
12 54 50.84 3.15 49.32 4.85 10.17
6.10
TS3 = = 2.00
3.05

23