ENN570: Enterprise Resource Planning

Tutorial on Forecasting Problems Semester 2, 2022

i. The Tutorial Objectives

Upon the completion of this tutorial students can be able:
• to use the various forecasting tools
• to determine the forecasting accuracy

ii. Problems
- This tutorial includes 3 solved problems and 03 problems for practice at home.
iii. Reference books
• Operations Management (9th Edition) by Jay Heizer and Barry Render

• Operations Management by Lee J. Krajewski and Larry P. Ritzman

iv. Applied Formulas

Tutorial Problems
Problem 1: Actual monthly sales of a product is given in the following table:
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
20 21 15 14 13 16 17 18 20 20 21 23

(a) Forecast sales for January this year to January next year using A 3-month weighted moving average using
0.5, 0.3 and 0.2 weights from the latest sales data respectively.
(b) Forecast for November to next year January using exponential smoothing using α = 0.3 and an October

forecast of 18.
(c) Forecast for November to next year January using adjusted exponential smoothing assuming α = β = 0.3,

October forecast is 18, and trend is 1 unit.

(d) Forecast for January next year using trend projection with coefficient of correlations and coefficient of
determination.
(e) Calculate MAD and MAPE for question (a)

Solution 1:
Retrieving information from the problem
Monthly sales of a product are given, from which sales forecast for January the next year needs to be made
applying different forecasting tools.

Solving (a): Weighted moving average

Monthly sales (Actual)

Period Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan
Actual sale 20 21 15 14 13 16 17 18 20 20 21 23
3-month weighted moving
20 21 15.00 17.80 15.70 13.70 14.70 15.90 17.30 18.80 19.60 20.50 21.80
average-Forecast

Solving (b): Forecast applying Exponential Smoothing Technique by using 𝜶 = 𝟎. 𝟑

and an October forecast of 18

FNov = FOct + α (AOct – FOct) = 18 + 0.30(20-18) = 18.6

FDec = 18.6 + 0.30(21-18.6) = 19.32
FJan = 19.32+ 0.30(23-19.32) = 20.42
Solving (c ): Calculate the sales forecast applying Adjusted exponential smoothing using α = β = 0.3,
and October forecast and trend of 18 and 1 units respectively

Month At Ft = α (At-1) + (1- α )(Ft - 1 + Tt - 1) Tt = β(Ft – Ft – 1) + (1 – β) (Tt – 1)] TAF = Ft + Tt

Oct 20 18 1 19
Nov 21 0.3x20 + (1-0.3)x(18+1) = 19.3 0.3x(19.3-18) + (1-0.3)x1 = 1.09 19.3 +1.09 = 20.39
Dec 23 20.57 1.14 21.72
Jan 22.10 1.26 23.36
 Solving (d): Calculate sales forecast applying linear regression tool like Least Square Method
Month T(=x) At (= y) xy X2 Y2
Jan 1 20 20 1 400
Feb 2 21 42 4 441
Mar 3 15 45 9 225
Apr 4 14 56 16 196
May 5 13 65 25 169
Jun 6 16 96 36 256
Jul 7 17 119 49 289
Aug 8 18 144 64 324
Sep 9 20 180 81 400
Oct 10 20 200 100 400
Nov 11 21 231 121 441
Dec 12 23 276 144 529
Ʃx = 78
Ʃy = 218
Ʃxy = 1474
Ʃx2 = 650
Ʃy2 = 4070

∑ 𝑥𝑦−𝑛𝑥̅ 𝑦̅
𝑏= ∑ 𝑥 2 −𝑛𝑥̅ 2
=0.4
𝑎 = 𝑦̅ − 𝑏𝑥̅
= 18.167 – 0.4x6.5 = 15.56
For the next January, x = 13.
So, F13 = a + bx = 15.56 + (0.4) x (13) = 20.76

Solving (e) Forecasting error calculation for question a

Mean Absolute Deviation (MAD) ∑ |𝑨𝒄𝒕𝒖𝒂𝒍 − 𝑭𝒐𝒓𝒆𝒄𝒂𝒔𝒕|
𝑴𝑨𝑫 =
𝒏

Mean Absolute Percent Error (MAPE) |𝑨𝒄𝒕𝒖𝒂𝒍𝒊 − 𝑭𝒐𝒓𝒆𝒄𝒂𝒔𝒕𝒊 |

∑ 𝟏𝟎𝟎 ×
𝑨𝒄𝒕𝒖𝒂𝒍𝒊
𝑴𝑨𝑷𝑬 =
𝒏

We are doing the error analysis for the forecast value of January to December
 Actual Forecast Error (A-F) |𝐄𝐫𝐫𝐨𝐫| [|𝐄𝐫𝐫𝐨𝐫|÷Actual]×100

Period
January 20 20 0 0 0.00

February 21 21 0 0 0.00

March 15 15 0 0 0.00

April 14 17.8 3.8 3.8 27.14

May 13 15.7 2.7 2.7 20.77

June 16 13.7 -2.3 2.3 14.38

July 17 14.7 -2.3 2.3 13.53

August 18 15.9 -2.1 2.1 11.67

September 20 17.3 -2.7 2.7 13.50

October 20 18.8 -1.2 1.2 6.00

November 21 19.6 -1.4 1.4 6.67

December 23 20.5 -2.5 2.5 10.87

∑|𝐴𝑐𝑡𝑢𝑎𝑙 − 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡| 21
𝑀𝐴𝐷 = = = 1.75
𝑛 12

|𝐴𝑐𝑡𝑢𝑎𝑙𝑖 − 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡𝑖 |
∑ 100 × 124.52
𝐴𝑐𝑡𝑢𝑎𝑙𝑖
𝑀𝐴𝑃𝐸 = = = 10.38%
𝑛 12

From a conceptual standpoint MAD weights all errors evenly whereas MAPE weights according to relative
error.
Problem 2. In order to promote the sales of a certain commodity, TV advertisement is broadcasted. The
following monthly sales and TV advertisement frequency were recorded for the year (2020)
TV advertisement (Times
Month Sales ($ thousand)
per month)
Jan 17 15
Feb 12 10
Mar 16 13
Apr 14 15
May 24 25
Jun 25 27
Jul 26 24
Aug 22 20
Sep 24 27
Oct 30 44
Nov 25 34
Dec 17 17

Calculate
(a) Whether TV advertisements had any considerable influence on the sales?
(b) If yes, what was the expected sale in January 2020 with 40 times of TV advertisement?
(c) What could be expected and why if there was no advertisement in January

Solution 2:

Month TV Ad (say, x Sales (say, y) xy X2 Y2

)
Jan 17 15 255 289 225
Feb 12 10 120 144 100
Mar 16 13 208 256 169
Apr 14 15 210 196 225
May 24 25 600 576 625
Jun 25 27 675 625 729
Jul 26 24 624 676 576
Aug 22 20 440 484 400
Sep 24 27 648 576 729
Oct 30 44 1320 900 1936
Nov 25 34 850 625 1156
Dec 17 17 289 289 289
Ʃx = 252
Ʃy = 271
Ʃxy = 6239
Ʃx2 = 5636
Ʃy2 = 7159

Step 1: Solve (a) Calculate the coefficient of correlation

The high value of the coefficient of correlation indicates that the TV advertisement had a significant
influence on the sales of the commodity.
𝑟2 = 0.91672 = 0.84
Comment: This value of the coefficient of determination indicates that 84% of the variation in the sales can
be explained by this regression equation.

Step (2): Solve for (b) and (c): Develop the regression equation and calculate expected sales in January
2020 for 40 and 0 TV ad.

𝑎 = 𝑦̅ − 𝑏𝑥̅ = 22.583 − 1.593 𝑥 21 = −10.87, The regression equation y = - 10.87 + 1.593x

Expected sale in January with 40 TV ad, y = a + bx = -10.87 + (1.593)(40) = $ 52.85 K
Expected sale in January with 0 TV ad, y = a + bx = -10.87 + (1.593)(0) = $ (- $ 10.57) K

Comment: With no ad in the TV, the sales value is found to be negative which realistically does not carry any
sense. In reality, there might be no sales in January with an expected value of zero.

Problem 3: A computer distributer wants to develop monthly indices for sales. Data from past three years,
by month, are available. What would be the monthly demand for the fourth year?

Month Year 1 Year 2 Year 3

Jan 80 85 105
Feb 70 85 85
Mar 80 93 82
Apr 90 95 115
May 113 125 131
Jun 110 115 120
Jul 100 102 113
Aug 88 102 110
Sep 85 90 95
Oct 77 78 85
Nov 75 82 83
Dec 82 78 80

Solution 3:

Step (1): Retrieve information from the problem

Monthly sales data is given for the three consecutive years that indicate seasonal variations. Therefore,
seasonal indices for each month need to be calculated for the fourth year forecasting.

Step (2): Calculate the seasonal indices dividing the period or yearly average by the grand average
sales.
Forecast the total sale for the fourth year
 This forecasting is user’s choice, it may be moving average, weighted moving average or could be just a
growth rate of the last year and so.

Here we use 3 period (yearly) weighted moving average, such that fourth year total forecast is

𝐹4 =1204×0.5+1130×0.3+1050×0.2=1151

Step (4): Forecast monthly sales multiplying average monthly forecasted sale by the respective
seasonal index
Seasonal Year 4
Month Year Year Year
Indices Monthly Sale
1 2 3
Jan 80 85 105 0.957 (1151/12)x0.957=91.84
Feb 70 85 85 0.851 81.63
Mar 80 93 82 0.904 86.73
Apr 90 95 115 1.064 102.04
May 113 125 131 1.309 125.51
Jun 110 115 120 1.223 117.34
Jul 100 102 113 1.117 107.14
Aug 88 102 110 1.064 102.04
Sep 85 90 95 0.957 91.84
Oct 77 78 85 0.851 81.63
Nov 75 82 83 0.851 81.63
Dec 82 78 80 0.851 81.63

3 Problems for Practice

Problem 1: Howard Weiss, owner of a musical instrument distributorship, thinks that demand for bass
drums may be related to the number of television appearances by the popular group Stone Temple Pilots
during the previous month. Weiss has collected the data shown in the following table:
Period 1 2 3 4 5 6
Demand for Bass Drums 3 6 7 5 10 7
Stone Temple Pilot’s TV Appearances 3 4 7 6 8 5

(a) Graph these data to see whether a linear equation might describe the relationship between the group’s
television shows and bass drum sales.
(b) Use the least-square regression method to derive a forecasting equation.
(c) What is your estimate for bass drum sales if the Stone Temple Pilots performed on TV nine times last
month?
(d) Estimate the relationship between the sales variations and TV appearance.
Solution
Solve (a) Graph the demand with respect to the TV show
The observations tend to cluster about a straight line over the range shown and a linear equation might
describe the relationship between the group’s television shows and bass drum sales.
Solve (b) Use Least square regression method to derive a forecasting equation
Period TV Ad (say, x Demand (y) xy X2 Y2
)
1 3 3 9 9 9
2 4 6 24 16 36
3 7 7 49 49 49
4 6 5 30 36 25
5 8 10 80 64 100
6 5 7 35 25 49
Ʃx = 33
Ʃy = 38
Ʃxy = 227
Ʃx2 = 199
Ʃy2 = 268
The regression equation y = 0.665 + 1.03x
Solve (c) what is your estimate for bass drum sales if the Stone Temple Pilots performed on TV
nine times last month?

Expected sales in the last month 9 TV appearances, y = a + bx = 0.665 + (1.03)(9) = 9.935

Solve (d) Estimate the relationship between the sales variations and TV appearance.

r=0.823
𝑟2 = 0.8232 = 0.68. This means that 68% of the bass drum sales variation and TV appearances can be
explained by the derived regression equation.

Problem 2: Consider the following actual (At) and forecast (Ft) demand levels for a product:

Time Period (t) Actual Demand (At) Forecast Demand (Ft)

1 50 50
2 42 50
3 56 48
4 46 50
Using exponential smoothing method, find the forecast for time period 5. (Hints: The first forecast, F1,
was derived by observing A1 and setting F1 equal to A1).

Solution:
We need to find the smoothing constant α. We know in general that Ft = Ft–1 + α (At–1 – Ft–1); and t = 2,
3, 4. We choose either t = 3 or t = 4 [t = 2 will not let us find α because F2 = 50 = 50 + α (50 – 50) holds
for any α].

Let’s pick t = 3. Then F3 = 48 = 50 + α (42 – 50) =>50 - 8α = 48 => α = 0.25

Now F5 = 50 + α (46 – 50) = 50 – 4xα = 50 - 4x0.25 =50 -1 = 49. So the forecast for the time period 5
demand is 49 units.

Problem 3: The following two demand sets are to be used to test two different basic exponential smoothing
models. He first model uses α = 0.1, and the second uses α = 0.5. In both cases, the model should be
initialized with a beginning forecast value of 50; that is, the ESF forecast for period 1 made at the end of
period 0 is 50 units. In each of the four cases, (two models on two demand sets), compute the average
forecast error and MAD. What do the results mean?
Demand set I Demand Set II
Period Demand Period Demand
1 51 1 77
2 46 2 83
3 49 3 90
4 55 4 22
5 52 5 10
6 47 6 80
7 51 7 16
 8 48 8 19
9 56 9 27
10 51 10 79
11 45 11 73
12 52 12 88
13 49 13 15
14 48 14 21
15 43 15 85
16 46 16 22
17 55 17 88
18 53 18 75

19 54 19 14
20 49 20 16

Data Set I Alpha = 0.1 Data Set I Alpha = 0.5

Period Demand Forecast Deviation Abs Period Demand Forecast Deviation Abs
dev dev
1 51 50.00 1.00 1.00 1 51 50.00 1.00 1.00
2 46 50.10 -4.10 4.10 2 46 50.50 -4.50 4.50
3 49 49.69 -0.69 0.69 3 49 48.25 0.75 0.75
4 55 49.62 5.38 5.38 4 55 48.63 6.38 6.38
5 52 50.16 1.84 1.84 5 52 51.81 0.19 0.19
6 47 50.34 -3.34 3.34 6 47 51.91 -4.91 4.91
7 51 50.01 0.99 0.99 7 51 49.45 1.55 1.55
8 48 50.11 -2.11 2.11 8 48 50.23 -2.23 2.23
9 56 49.90 6.10 6.10 9 56 49.11 6.89 6.89
10 51 50.51 0.49 0.49 10 51 52.56 -1.56 1.56
11 45 50.56 -5.56 5.56 11 45 51.78 -6.78 6.78
12 52 50.00 2.00 2.00 12 52 48.39 3.61 3.61
13 49 50.20 -1.20 1.20 13 49 50.19 -1.19 1.19
14 48 50.08 -2.08 2.08 14 48 49.60 -1.60 1.60
15 43 49.87 -6.87 6.87 15 43 48.80 -5.80 5.80
16 46 49.19 -3.19 3.19 16 46 45.90 0.10 0.10
17 55 48.87 6.13 6.13 17 55 45.95 9.05 9.05
18 53 49.48 3.52 3.52 18 53 50.47 2.53 2.53
19 54 49.83 4.17 4.17 19 54 51.74 2.26 2.26
20 49 50.25 -1.25 1.25 20 49 52.87 -3.87 3.87

Average= 0.06 3.10 Average= 0.09 3.34

 Data Set II Alpha = 0.1 Data Set II Alpha = 0.5
Period Demand Forecast Deviation Abs Period Demand Forecast Deviation Abs
dev dev
1 77 50.00 27.00 27.00 1 77 50.00 27.00 27.00
2 83 52.70 30.30 30.30 2 83 63.50 19.50 19.50
3 90 55.73 34.27 34.27 3 90 73.25 16.75 16.75
4 22 59.16 -37.16 37.16 4 22 81.63 -59.63 59.63
5 10 55.44 -45.44 45.44 5 10 51.81 -41.81 41.81
6 80 50.90 29.10 29.10 6 80 30.91 49.09 49.09
7 16 53.81 -37.81 37.81 7 16 55.45 -39.45 39.45
8 19 50.03 -31.03 31.03 8 19 35.73 -16.73 16.73
9 27 46.92 -19.92 19.92 9 27 27.36 -0.36 0.36
10 79 44.93 34.07 34.07 10 79 27.18 51.82 51.82
11 73 48.34 24.66 24.66 11 73 53.09 19.91 19.91
12 88 50.80 37.20 37.20 12 88 63.05 24.95 24.95
13 15 54.52 -39.52 39.52 13 15 75.52 -60.52 60.52
14 21 50.57 -29.57 29.57 14 21 45.26 -24.26 24.26
15 85 47.61 37.39 37.39 15 85 33.13 51.87 51.87
16 22 51.35 -29.35 29.35 16 22 59.07 -37.07 37.07
17 88 48.42 39.58 39.58 17 88 40.53 47.47 47.47
18 75 52.38 22.62 22.62 18 75 64.27 10.73 10.73
19 14 54.64 -40.64 40.64 19 14 69.63 -55.63 55.63
20 16 50.57 -34.57 34.57 20 16 41.82 -25.82 25.82

Average -1.44 33.06 -2.11 34.02

Average Forecast Error Mean Abs Dev (MAD)

Data set Alpha = 0.1 Alpha = 0.5 Alpha = 0.1 Alpha = 0.5
I 0.06 0.09 3.10 3.34
II -1.44 -2.11 33.06 34.02

According to the calculated values of forecast error and MAD values, it is obvious that the data set II exhibit
more spread or fluctuations. The greater average forecast error or MAD for the smoothing constant values
(α =0.1 and 0.5) are in support of the conclusion. The average error values of .06 and 0.09 in data set I indicate
that the forecasts produced by both the exponential smoothing models have very little bias. However, the
forecasts produced by both exponential smoothing models on data set II exceed actual demand by 1 to 2
units per period. Of the two models, use of one with the smoothing constant value of 0.1 would be
appropriate for forecasting the demand for both data sets.
Additional 2 Problems (Try yourself)
Problem 1: As you can see in the following table, demand for heart transplant surgery at a general hospital
has increased steadily in the past few years:
Year 1 2 3 4 5 6
Heart transplant 45 50 52 56 58 ?
The director of medical services predicted 6 years ago that demand in year 1 would be 41 surgeries.
a) Use exponential smoothing, first with smoothing constant of 0.6 and then with one of 0.9 to develop
forecasts for years 2 through 6.
b) Use a 3-year moving average to forecast demand in years 4, 5, and 6.
c) Use the trend projection method to forecast demand in years 1 through 6.

Problem 2: Dr. Jerilyn Ross, a New York City psychologist, specializes in treating patients who are
agoraphobic (afraid to leave their homes). The following table indicates how many patients Dr. Ross has seen
each year for the past 10 years. It also indicates what the robbery rate was in New York City during the same
year.
Year 1 2 3 4 5 6 7 8 9 10
Number of 36 33 40 41 40 55 60 54 58 61
patients
Robbery rate 58.3 61.1 73.4 75.7 81.1 89.0 101.1 94.8 103.3 116.2
per 1,000
population

a) Apply linear regression to study the relationship between the robbery rate and Dr. Ross’s patient load.
a) Using trend analysis, predict the number of patients Dr. Ross will see in years 11, 12, and 13. How well
does the model fit the data?
b) If the robbery rate increases to 131.2 in year 11, how many phobic patients will Dr. Ross treat?
c) If the robbery rate drops to 90.6, what is the patient projection?
d) What is the model’s correlation coefficient? Determine the coefficient of determination and interpret
the value you have obtained. .