Forecasting

Trend problem -Limitations

𝐖𝐡𝐚𝐭 𝐚𝐫𝐞 𝐭𝐡𝐞 𝐥𝐢𝐦𝐢𝐭𝐚𝐢𝐨𝐧𝐬 𝐨𝐟 𝐟𝐨𝐫𝐞𝐜𝐚𝐬𝐭𝐢𝐧𝐠 𝐮𝐬𝐢𝐧𝐠 𝐚 𝐭𝐫𝐞𝐧𝐝 𝐥𝐢𝐧𝐞?

 Always plot the data to understand the relationships between x and y.

 It is useful for linear relationship only. In case of curvilinear relationship, linear method will not give use
the accurate forecast!!

 This forecasting method is applicable for only next few periods.

 Deviation around the trend line assumed to be random and normally distributed.
 Extension of Exponential smoothing

 𝐖𝐡𝐞𝐧 𝐭𝐫𝐞𝐧𝐝 𝐢𝐬 𝐩𝐫𝐞𝐬𝐞𝐧𝐭 𝐭𝐡𝐞𝐧 𝐬𝐢𝐦𝐩𝐥𝐞 𝐞𝐱𝐩𝐨𝐧𝐞𝐧𝐭𝐢𝐚𝐥 𝐬𝐦𝐨𝐨𝐭𝐡𝐢𝐧𝐠 𝐰𝐢𝐥𝐥 𝐧𝐨𝐭 𝐠𝐢𝐯𝐞 𝐭𝐡𝐞 𝐫𝐢𝐠𝐡𝐭 𝐟𝐨𝐫𝐞𝐜𝐚𝐬𝐭.

 Simple exponential smoothing method is like moving average technique, which fails to respond to the
trend.

Week Actual Demand Forecast

1 200 F1 = 200
2 250 F2 = F1+ (A1-F1) = 200+0.4*(200-200)=200
3 300 F3 = F2+ (A2-F2) = 200+0.4*(250-200)=220
4 350 F4 = F3+ (A3-F3) = 220+0.4*(300-220)=252
5 400 F5 = F4+ (A4-F4) = 252+0.4*(350-252)=292

Why this forecast in not accurate?

 Trend adjusted exponential smoothing

 When trend is present, then how to use the simple exponential smoothing method.

 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑖𝑛𝑐𝑙𝑢𝑑𝑖𝑛𝑔 𝑡𝑟𝑒𝑛𝑑 (𝐹𝐼𝑇 ) = 𝐸𝑥𝑝𝑜𝑛𝑒𝑛𝑡𝑖𝑎𝑙𝑙𝑦 𝑠𝑚𝑜𝑜𝑡ℎ𝑒𝑑 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 (𝐹 )+

Exponentially smoothed trend (𝑇 )

𝐼𝑛 𝑡ℎ𝑖𝑠 𝑚𝑒𝑡ℎ𝑜𝑑, 𝑡ℎ𝑒 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑠 𝑜𝑓 𝑏𝑜𝑡ℎ 𝑡ℎ𝑒 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑎𝑛𝑑 𝑡ℎ𝑒 𝑡𝑟𝑒𝑛𝑑 𝑎𝑟𝑒 𝑠𝑚𝑜𝑜𝑡ℎ𝑒𝑑. 𝑇ℎ𝑢𝑠 𝑟𝑒𝑞𝑢𝑖𝑟𝑒
𝑡𝑤𝑜 𝑠𝑚𝑜𝑜𝑡ℎ𝑖𝑛𝑔 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟𝑠.

𝐹 =  ∗ (Actual demand of last period)+ (1- )(Forecast last period + Trend estimates last period)
𝐹 =  ∗ 𝐴 + (1- )(𝐹 + 𝑇 )
Where,  is the smoothing constant for the average.

𝑇 =  ∗ (Forecast this period-Forecast last period)+ (1- )(Trend estimates last period)
𝑇 =  ∗ (𝐹 − 𝐹 )+ (1- )(𝑇 )
Where,  is the smoothing constant for the trend .

𝐹𝐼𝑇 = 𝐹 + 𝑇
 Trend adjusted exponential smoothing -Problem 1

Week Actual Question. The actual demand of machines are shown in the table. After review of
Demand sales, a trend is visible. The firm knows that initial average forecast for month 1
1 10 (F1) was 8 units and trend for period 1 was 5 units.  and  are 0.2 and 0.4
respectively. Calculate the forecast for 7th month.
2 20
3 30
4 40  Step 1. Calculate the exponentially smoothed forecast average for period t,
𝐹 =  ∗ 𝐴 + (1- )(𝐹 + 𝑇 )
5 35
6 45  Step 2. Calculate the smoothened trend for period t,
𝑇 =  ∗ (𝐹 − 𝐹 )+ (1- )(𝑇 )
7 ?
 Step 3. Calculate the forecast including trend as,
𝐹𝐼𝑇 = 𝐹 + 𝑇
 Trend adjusted exponential smoothing - Problem 1

𝐹 =∗𝐴 + (1- )(𝐹 +𝑇 ) 𝑇 =  ∗ (𝐹 − 𝐹 )+ (1- )(𝑇 )

Week Actual
Demand 𝐹 = 0.2 ∗ 10+ (1- 0.2)(8 + 5)=12.4 𝑇 = 0.4 ∗ (12.4 − 8)+ (1- 0.4)(5)=4.76
1 10
𝐹𝐼𝑇 = 12.4+4.76=17.16
2 20 𝑇 = 0.4 ∗ (17.72 − 12.4)+ (1- 0.4)(4.76)=4.98
𝐹 = 0.2 ∗ 20+ (1- 0.2)(12.4 + 4.76)=17.72
3 30 𝐹𝐼𝑇 = 17.72+4.98=22.7
4 40 𝐹 = 0.2 ∗ 30+ (1- 0.2)(17.72 + 4.98)=24.16 𝑇 = 0.4 ∗ (24.6 − 17.72)+ (1- 0.4)(4.98)=5.74
5 35 𝐹𝐼𝑇 = 24.16+5.74=29.9
6 45 𝐹 = 0.2 ∗ 40+ (1- 0.2)(24.16 + 5.74)=32.92 𝑇 = 0.4 ∗ (32.92 − 24.16)+ (1- 0.4)(5.74)=6.94

7 ? 𝐹𝐼𝑇 = 32.92+6.94=39.86
𝐹 = 0.2 ∗35+ (1- 0.2)(32.92 + 6.94)=38.88 𝑇 = 0.4 ∗ (38.88 − 32.92)+ (1- 0.4)(5.74)=5.82
Do the same problem without 𝐹𝐼𝑇 = 38.88+5.82=44.7
using any trend just like
exponential smoothing forecast. 𝐹 = 0.2 ∗ 45+ (1- 0.2)(38.88 + 5.82)=44.76 𝑇 = 0.4 ∗ (44.76 − 38.88)+ (1- 0.4)(5.82)=5.84
Have you noticed any difference?
𝐹𝐼𝑇 = 44.76+5.84=50.6 Try to do in excel. Some variations may
come due to rounding off at each level.
 Trend adjusted exponential smoothing - Problem 1

Week Actual Forecast Exponentialy smoothing with trend adjustment forecast

Demand demand 50
1 10 8 40
2 20 17.6 30

Demand
3 30 22.7 20

4 40 29.9 10

5 35 39.86 0
1 2 3 4 5 6
6 45 44.7 Week

7 ? 50.06 Actual Demand Forecast demand

Draw this forecast and actual !!

 Seasonality

 Seasonality in data are regular movements in time series that relate

to recurring events such as weather or weekends etc.

Examples. sales of umbrella, movie tickets, power demand in summer.

 Steps to be followed during seasonality forecast.

1. Find average historical average demand of each season
2. Compute average demand over all months
3. Compute seasonality index
4. Estimate next period total demand
5. Divide the estimate of total annual demand by number of seasons. Then multiply it by seasonality index for
each month. This provides seasonal forecast.
 Seasonality Problem
Question. An automobile manufacturer wants to develop monthly sales forecast in Ranchi using last three years data as
given below. Forecast for the next 12 months with expected annual demand is 1200 units.

Month Year 1 Year 2 Year 3

Jan 80 85 75
Feb 70 80 90
March 75 75 85
April 85 90 100
May 110 120 125
June 115 125 110
July 105 110 115
Aug 90 95 100
Sept 85 80 90
Oct 90 85 85
Nov 80 80 90
Dec 75 80 85
 Seasonality Problem - Solution

Average period Average monthly Forcast based on annual

Month Year 1 Year 2 Year 3 Seasonal Index
demand demand demand of 1200 units
Jan 80 85 75 80 92 0.869565217 87
Feb 70 80 90 80 92 0.869565217 87
March 75 75 85 78 92 0.851449275 85
April 85 90 100 92 92 0.996376812 100
May 110 120 125 118 92 1.286231884 129
June 115 125 110 117 92 1.268115942 127
July 105 110 115 110 92 1.195652174 120
Aug 90 95 100 95 92 1.032608696 103
Sept 85 80 90 85 92 0.923913043 92
Oct 90 85 85 87 92 0.942028986 94
Nov 80 80 90 83 92 0.905797101 91
Dec 75 80 85 80 92 0.869565217 87

Insights:
 Think of these outcomes as percentage of total sales. Average sales without seasonality would be 92. But with
seasonality sales fluctuates between 85 to 129 .
 Seasonality Problem – Learning Exercise
Q. A automobile manufacturer wants to develop monthly sales forecast in Raipur region using last four years data
as given below. Next year expected annual demand is 1000 units.
Month Year 1 Year 2 Year 3 Year 4
Jan 80 85 75 78
Feb 70 80 90 85
March 75 75 85 90
April 85 90 100 85
May 110 120 125 105
June 115 125 110 120
July 105 110 115 90
Aug 90 95 100 85
Sept 85 80 90 75
Oct 90 85 85 70
Nov 80 80 90 75
Dec 75 80 85 70
 Seasonality Problem with Trend

Next, what happens when there is a trend in the seasonality??

Q. A hospital wants to improve its forecasting by applying both trend and seasonal indices using 36 months
collected data. It will then forecast “patient hours” over the coming years.
After using 36 months data, we get the following trend line and seasonality pattern is given below.
𝑦 = 10 + 2.5𝑥

Month Seasonality
Jan 1.04
Feb 0.95
Mar 1.05
Apr 1.02
May 0.98 Calculate the forecast for next 12 months (37-48 months).
Jun 0.95
Jul 1.1
Aug 1.04
Sep 0.94
Oct 1
Nov 0.96
Dec 0.98
 Seasonality Problem with Trend

Forecast with trend Forecast with Trend and seasonality

140 127.5 130 140 129.25 124.8
117.5 120 122.5 125 125 122.4 127.4
115
107.5 110 112.5
120 112.875112.2 110.25109.25 115.15
102.5 105 120 106.6
99.75
100 100

80 80
Hours

hours
60 60

40 40

20 20

0 0
37 38 39 40 41 42 43 44 45 46 47 48 37 38 39 40 41 42 43 44 45 46 47 48
Month Months

Insights:
 Forecasting of trend data with seasonality gives better forecast than trend only.
 Forecasting with trend have minimum value of 102.5 and maximum is 130 with increase in the trend line.
 Forecasting with trend + seasonal have minimum value of 99.5 and maximum is 129.25 with fluctuations as
per seasonal demand.
 Seasonality Problem with Trend – problem

Q. A computer manufacturing company used time-series based on point of sales data to forecast sales of next
4 months. Sales estimates are Rs. 10,00,000, 12,00,000, 14,00,000 and Rs. 13,00,000 respectively for four
months. Seasonal indices for four months are 1.10, 1.15, 0.9, and 0.8 respectively.

Ans:
To compute seasonalized effect,
𝑦 =𝑦 ∗ 𝑠𝑒𝑎𝑠𝑜𝑛𝑎𝑙 𝑖𝑛𝑑𝑒𝑥
Month 1: 10,00,000*1.10=11,00,000
Month 2: 12,00,000*1.15=13,80,000
Month 3: 14,00,000*0.9=12,60,000
Month 4: 13,00,000*0.8=10,40,000

Insights:
 Trend forecast is now adjusted with the seasonal index to reflects the seasonal changes.
Associative Forecasting Methods
 Associative Forecasting Methods

 Associative forecasting models usually consider several variable that are related to the predicted demand.

 Example: The sales of Apple laptops depends on multiple factors like Apple’s advertising budget, laptop price,
competitors prices, and even nation's economy.

 Hence, we consider Apple laptop sales as dependent variable here and others like Apple’s advertising budget,
laptop price, competitors prices, and nation's economy as independent variable.

 The operation manager’s job is develop the best statistical relationship between laptop sales and independent
variables.

 The common quantitative associative forecasting methods are:

 Correlation
 linear-regression
 Multiple regression
 Regression

 Linear regression

 Multiple linear regression

 Simple linear regression
 Linear regression is a linear approach for modelling the relationship between a dependent and one or more
independent variables.
 When one independent variable is used, then it is called simple linear regression.
 When two or more independent variables are used, it is called multiple linear regression.

𝑦 = 𝑎 + 𝑏𝑥̅

𝑾𝒉𝒆𝒓𝒆, 𝒚 is the dependent variable and 𝒙 is the

independent variable.

b is the slope of the regression line.

Coefficients a and b are calculated based on the following.
∑ ̅
b ∑ ̅
 Simple linear regression-problem
Question. Sales and profits of a food chain store is given in the following table. Predict the profit for a
store assuming sales of Rs. 6 billions.

Sales (in Profit (in

billions) millions)

1 2 ∑ ̅
b ∑ ̅
3 3
4 2.5
2 2
1 2
7 3.5
 Simple linear regression-problem
Question. Sales and profits of a food chain store is given in the following table. Obtain a regression line
for the data, and predict the profit for a store assuming sales of Rs. 6 billions.

Regression Line: 1.75+0.25x

Answer: Regression line as below.

Regression line 1.75+0.25x

When x=6, Profit 3.25 millions

Insights: The regression line indicate that profit increase at the rate of 0.25 million for every 1 billion of
increase in the sales.

Learning exercise: What will be the profit when sales is 8 billion?

When x=8, Profit 3.75 millions

 Linear Regression-Model fitness

 Two metrics are commonly used to test the model fitness of regression line.

 R-square
 Standard error of the regression

 R-square: is the proportion of the variance in the dependent variable that can be explained by the independent
variable.

 The standard error of the regression is the average distance that the observed values fall from the regression
line.

It measures the error from the dependent variable y to the Sales (in Profit (in
billions) millions)
regression line.
1 2
3 3
∑ 𝑦 − 𝑎 ∑ 𝑦 − 𝑏 ∑ 𝑥𝑦 4 2.5
𝑆 , = 2 2
𝑛−2
1 2
7 3.5
 Linear Regression-Model fitness
Sales (in Profit (in
billions) millions) xy x2 y2
(x) (y)
1 2 2 1 4 ∑ 𝑦 − 𝑎 ∑ 𝑦 − 𝑏 ∑ 𝑥𝑦
𝑆 , =
3 3 9 9 9 𝑛−2
4 2.5 10 16 6.25
2 2 4 4 4
Standard error: 0.306186
1 2 2 1 4
7 3.5 24.5 49 12.25
3 15 51.5 80 39.5 When x=6, Profit 3.25 millions

Insights: Its interpretation is similar to the standard deviation.

There is a 68.27% chance that profit will be lie between 3.25 +/- 0.30618 millions.

Learning exercise: What will be the probability that profit will exceed 3.55 millions, when sales will be 6 billions?

1-NORM.DIST(3.55,3.25,0.306,1) = 16%
 Multiple linear regression

Multiple Regression is an extension of simple regression model where more than one independent variables
can be used to understand its relationship on the dependent variable.

+ 

𝑊ℎ𝑒𝑟𝑒, 𝑦 is the dependent variable and 𝑥 , 𝑥 are the independent variables.

𝑏 and 𝑏 are the coefficients of the independent variables.

 is the error term.

Most widely used for predictive analytics.

 Multiple linear regression – problem 1
Question. Sales figures, advertisement cost, and profits of a food chain store is given in the following table.
Obtain a regression line for the data, and predict the profit for a store assuming sales of Rs. 6 billions and
advertisement cost of 0.6 millions.

Profit (in Sales (in Advertisem

millions) billions) ent cost (in
millions)

2 10 0.3
3 30 0.5
2.5 40 0.4
2 20 0.2
2 10 0.3
3.5 70 0.5
 Multiple linear regression – problem 1

Regression Statistics
Multiple R 0.977692361
R Square 0.955882353
Adjusted R Square 0.926470588
Standard Error 0.171498585
Observations 6

ANOVA
df SS MS F Significance F
Regression 2 1.911764706 0.955882 32.5 0.009267
Residual 3 0.088235294 0.029412
Total 5 2

Coefficients Standard Error t Stat P-value Lower 95%Upper 95%Lower 95.0%

Upper 95.0%
Intercept 1.029411765 0.261417483 3.937808 0.029179 0.197465 1.861359 0.197465 1.861359
x1 0.013970588 0.004877389 2.864358 0.064347 -0.00155 0.029493 -0.00155 0.029493
x2 2.867647059 0.918382059 3.122499 0.05237 -0.05505 5.790349 -0.05505 5.790349

Y = 1.029+0.0139*X1+2.867*X2 Y = 2.832
Design of Goods and Services
 Goods and Services
How you define goods and services?
 Goods include durable like cars, computers and nondurables like food and beverages.
 Services are offerings or experiences like education, software, healthcare.

 Leading companies generate a substantial portion of their sales

from products less than 5 years old.
 Car companies continuously developed new products, in spite
of continuing high sales of its existing models.
 Companies like Maruti or Tata Motors are continuously
innovating with new car models and variants even though even
though they are industry leaders in India.

How to manage a product or service ???

 Product Life Cycle

What is the product life cycle?

How cash flow, and profit varies over the life cycle of a product?
 Product Life Cycle

 Typically a firm has a negative cash flow while it develops a product. When the product is successful, those
losses may be recovered.

 Eventually, the successful product yield a profit prior to its maturity phase.
 Product Life Cycle

Strategies during product life cycle?

Introductory Phase: Product, process and suppliers are not matured at this stage. Operations managers were still groping
for the best manufacturing techniques.

Growth Phase: In the growth phase, product design has begun to stabilize, and effective forecasting of capacity
requirements is necessary. Adding capacity or enhancing existing capacity to accommodate the increase in product
demand may be necessary.

Maturity Phase: By the time a product is mature, competitors are established. So high-volume, innovative production
may be appropriate. Improved cost control, reduction in options, and a paring down of the product line may be effective
or necessary for profitability and market share.

Decline Phase: Management may need to be ruthless with those products whose life cycle is at an end. Dying products
are typically poor products in which to invest resources and managerial talent will be waste. Unless dying products make
some unique contribution to the firm’s reputation or its product line or can be sold with an unusually high contribution,
their production should be terminated.
 Product Development

How to develop a product? Product concepts are developed from

a variety of sources, both external and
internal to the firm.

 An effective product strategy

links product decisions with other
business functions, such as R&D,
engineering, marketing, and
Concepts that survive the product idea
finance. stage progress through various stages,
with regular review, feedback, and
evaluation in a highly participative
 Optimum product development environment to minimize failure.
depends not only on support from
other parts of the firm but also on
the successful integration of all
OM decisions, from product
design to maintenance.
 QFD

Quality Function Deployment (QFD)

QFD is a systematic approach that refers to

(i) Determining what are customers expectations
(ii) Translating those customers expectations into the design requirements

QFD is used early in the design process to determine what will satisfy the customer and where to deploy quality efforts .

One of the tools of QFD is the house of quality, a graphic technique for defining the relationship between customer
desires and product (or service).

Defining relationship is the first step in building a world-class production system. To build the house of quality, we
perform seven basic steps:
 House of Quality

Seven steps to build house of quality

Step 4. Step 1. Identify customers wants?

Step 2. Identify how goods or service will satisfy customer want?

Step 2. (product or process characteristics)

Step 3. Relate customer wants to product how (build a matrix

between these two)

Step 1. Step 3. Step 6 Step 4. identify relationships between firm’s how (how do firms’ how
tie together)

Step 5 Step 5. Develop importance ratings/weights (Using the customer’s

importance ratings and weights for the relationships shown in the matrix)
Step 7
Step 6. Evaluate competing products (How well do competing products
meet customer wants?)
Step 7 Step 7. Determine the desirable (target) technical attributes, your
performance, and the competitor’s performance against these attributes.
House of Quality-Example

Step 4.

Step 2.

Step 6
Step 3.
Step 1.

Step 5

Step 7
 House of Quality – Safety helmets

Develop a house of quality for safety helmets (XYZ)?

Thank You