TERM PROJECT ON

FORECASTING
Course Name: Operations Management
Course Code: OPM501 (Summer 2023)

Submitted to
Dr. Abdullahil Azeem
Professor
Department of Industrial and Production Engineering (IPE)
Bangladesh University of Engineering and Technology (BUET)
Dhaka - 1000, Bangladesh.

Submitted by: Group 2

Adnan Asef (2022-3-95-060)
Farhana Akhter (2022-1-95-101)
S.M Shahinur Rahman (2021-3-91-010)
Tabassum ferdous Mumu (2022-2-95-005)
Rehnoma Tanjim Habiba (2022-3-95-061)

DATE OF SUBMISSION: 2nd September, 2023

 Acknowledgment
At first, we want to give thanks to Almighty for giving us enough spirit, energy and intelligence
and the opportunity to continue our project entitled “TERM PROJECT ON FORECASTING”. We
are grateful to our parents for their constant support.

We would like to give thanks with deep gratitude to our supervisor Dr. Abdullahil Azeem,
Professor, Department of MBA & EMBA Programs, East West University for his valuable and
helpful suggestions during the project work. His guidance helped us in writing and completing the
project.

Finally, we want to give respect to some important persons in our academic and personal life.
Without their guidance, this project work would not be possible. And we are also indebted to our
friends for giving their help and support to overcome the difficulties during the whole working
period.

1
Table of Contents
Chapter 1: Introduction ................................................................................................................................ 1
1.1 History....................................................................................................................... 1
1.2 The Product ............................................................................................................... 1
1.3 Report organization ................................................................................................... 2
Chapter 2: Monthly Sales Data ..................................................................................................................... 3
Chapter 3: Two Techniques from The Course Content................................................................................. 4
3.1 Weighted Moving Average ....................................................................................... 4
3.2 Exponential Smoothing ............................................................................................. 5
Chapter 4: Two Advanced Techniques ......................................................................................................... 9
4.1 Double Exponential Smoothing ................................................................................ 9
4.2 Non-Linear Regression ........................................................................................... 10
Chapter 5: Comparison between Four Different Techniques ..................................................................... 12
Table 5.1: Comparison between Four Different Forecasting Techniques .................................................. 12
Chapter 6: Accuracy of Forecasts................................................................................................................ 13
6.1 Forecasting Error for Weighted Moving Average ......................................................................... 13
6.2 Forecasting Error for Exponential Smoothing ........................................................ 14
6.3 Forecasting Error for Double Exponential Smoothing ........................................... 15
6.4 Forecasting Error for Non-Linear Regression ........................................................ 16
6.5 Comparison between MAD & MSE of Four Forecasting techniques .................... 17
Chapter 7: Results and Discussion .............................................................................................................. 18
Chapter 8: Conclusion ................................................................................................................................. 19
References .................................................................................................................................................. 20

2
 LIST OF TABLES
Table-2.1: Two Years Sales Data for Cylinder ......................................................................06

Table-3.1: Forecast calculation for Weighted Moving Average ............................................07

Table-3.2: Data for forecast calculation of Exponential Smoothing ......................................09

Table-4.1: Forecast calculation for Double Exponential Smoothing......................................13

Table-4.4: Forecast calculation for Non-Linear Regression ..................................................15

Table-5.1: Comparison between Four Different Techniques .................................................16

Table-6.1: Calculation of MAD & MSE for Weighted Moving Average ..............................17
Table-6.2: Calculation of MAD & MSE for Exponential Smoothing ................................... 18
Table-6.3: Calculation of MAD & MSE for Double Exponential Smoothing ...................... 19
Table-6.4: Calculation of MAD & MSE for Non-Linear Regression ....................................20
Table-6.5: Comparison between MAD & MSE of Four Different technique ........................21

3
 Chapter 1: Introduction

Omera LPG is a liquefied petroleum gas (LPG) brand incorporated in Bangladesh that has been
developed and owned by Omera Petroleum Limited, a subsidiary of MJL Bangladesh Limited.
Omera Petroleum Limited (OPL) has launched Omera LP Gas in Bangladesh considering the
growing demand of customers. OPL has started LPG venture in Bangladesh with the utmost
commitment to convenience, availability, functionality, and safety. Omera LPG is engaged in
importing, storing, bottling, and distributing LPG in residential, commercial, and industrial
sectors. Omera is the pioneer in Bangladesh in LPG transport through waterway. Omera has
invested in building three LPG carrying barges named M.T Omera Princess, M.T Omera Glory
and M.T Omera King based on Japanese Navigation & Communication Devices and European
Technology. Building these three barges is itself a revolution in the industry.

1.1 History
Natural gas dominated as the primary fuel source in the early decades of Bangladesh. The first
hydrocarbon exploration in East Bengal took place in 1911 and the first gas reserve was discovered
in 1955. The use of natural gas in industrial sectors began in East Bengal in 1955. In 1968, the
piped Natural Gas connection was introduced to households of East Pakistan.
In response to the demand for alternative fuel and the growth of the market, Omera Petroleum
Limited, a subsidiary of MJL Bangladesh Limited. has launched Omera LPG in the market on
March 31, 2015. Market analysis conducted by LightCastle Partners in 2019 concluded that
Bashundhara Group led the LPG market in Bangladesh with a 24% share, while Omera was the
second-largest provider, with a 20% market share.
Omera LPG, in partnership with Caritas Internationalis, distributed LPG cylinders among 12,000
refugee households to promote clean and reliable energy source for cooking and heating activities.
The project was also taken envisioning to environmental situation development and improved air
quality. Omera LPG has also been seen to take measurements against the COVID-19 Pandemic
situation by endorsing PPE to the National Central Jail of Bangladesh.
Since the very beginning, Omera has set up an ERP based platform Oracle to integrate all the
process efficiently where no other companies are using such system. Research and Development
is one of their core strength to cater the need of different segments. Omera is also the pioneer in
Sales Automation in the industry. Its specialized automation system ‘Sales Hawk’ has enabled the
sales team to conduct daily activities in a more efficient and smart way.

1.2 The Product

LPG is being embraced by industries as the most affordable clean energy solution. To cater this
growing demand, Omera LPG brings ‘Omera Priority’ offering uninterrupted LPG solutions to
industries.

1
LPG is a low-carbon fuel when compared to diesel or petrol, it also costs half the price of both
fuels. LPG is also called clean or green fuel. LPG (Liquid Petroleum Gas) is a mixture of propane
(C3H8) and butane (C4H10) gas suitable for vehicle fuel use. This fuel is also more efficient and
safer the engine. The level of air pollution and solid waste produced by LPG combustion is
significantly low.
Omera has supplied approx. 450 000 MT of LPG and supplied 3.6 Million LPG Cylinders in
Bangladesh retail market in first two years of their operations. They have maximized their reach
by having more than 35,000 retailers across the country which covers the 93% of the retail universe
of Bangladesh. Omera has a robust solution to secure the year-long uninterrupted supply of LPG
with the help of its VLGC operation.

1.3 Report organization

In Chapter-2, monthly sales data for Cylinder of Omera LP Gas Ltd on 2021 and 2022 is listed. In
chapter-3, we have calculated two techniques from the course content named weighted moving
average and exponential smoothing. In Chapter-4, we have calculated two advanced techniques
named double exponential smoothing and non-linear regression. In Chapter-5, the comparison
between four different techniques is given. In Chapter-6, the accuracy of forecasts is given.

2
 Chapter 2: Monthly Sales Data

We have collected the monthly sales data for the Cylinder of Omera LP Gas Ltd in
2021 and 2022.

Table 2.1: Two Years Sales Data for Cylinder

Omera LP Gas Ltd

Two Years of Sales Data for Cylinder
2021 2022
Month Total Month Total
January 49,950 January 76,600
February 52,350 February 74,700
March 54,750 March 79,900
April 54,750 April 82,100
May 55,750 May 80,640
June 58,750 June 82,840
July 59,150 July 85,040
August 59,150 August 84,580
September 60,150 September 86,120
October 65,750 October 87,660
November 62,950 November 90,760
December 68,950 December 89,860

3
 Chapter 3: Two Techniques from The Course Content

We have calculated two techniques from the course content named simple moving
average andweighted moving average in this chapter.

3.1 Weighted Moving Average

Weighted moving averages assign a heavier weighting to more current data points
since they aremore relevant than data points in the distant past. In the Weighted
moving average,

The forecast value,

Table 3.1: Forecast calculation for Weighted Moving Average

Sl No. Months of 2022 Total sales (Ai) Weight (Wi) Ai*Wi Forecast
1 January 76,600 1 76,600
2 February 74,700 2 149,400
3 March 79,900 3 239,700
4 April 82,100 4 328,400
5 May 80,640 5 403,200
6 June 82,840 6 497,040
7 July 85,040 7 595,280 81,964
8 August 84,580 8 676,640 83,038
9 September 86,120 9 775,080 83,991
10 October 87,660 10 876,600 84,974
11 November 90,760 11 998,360 86,647
12 December 89,860 50 4,493,000 88,579

4
 Figure 3.1: Chart for Weighted Moving Average

92,000 Chart for Weighted Moving Average

90,760
89,860
90,000
88,579
87,660
88,000
86,647
86,120
86,000 85,040 84,974
84,580
83,991
84,000 83,038
81,964
82,000

80,000

78,000

76,000
1 2 3 4 5 6
Actual 85,040 84,580 86,120 87,660 90,760 89,860
Forecast 81,964 83,038 83,991 84,974 86,647 88,579

5
3.2 Exponential Smoothing
Exponential smoothing is a forecasting method for univariate time series data. In
exponentialsmoothing,

The forecast value, Ft = Ft-1 + α (At-1 – Ft-1)

Where,
Ft-1 = Forecast of
previous monthFt =
Forecast of present
month At-1 =
Actual for the
previous month
For smoother demand,
α = 0.1 ~ 0.2 for
fluctuating demand, α
= 0.3 ~ 0.5

Table 3.2: Data for forecast calculation of Exponential Smoothing

Sl No. Months of 2022 Total sales Quantity Remarks

1 January 76,600
2 February 74,700
3 March 79,900
4 April 82,100
5 May 80,640
6 June 82,840
7 July 85,040 A1

8 August 84,580 A2
9 September 86,120 A3
10 October 87,660 A4
11 November 90,760 A5
12 December 89,860 A6

6
Here found no 1st forecast value so 1st actual value will be 1st forecast value
Forecast valueSimplifying the formula, we considered α = 0.2
Here, F1 = 81,964 (1st forecast value considered from the weighted moving average)

F2 = F1 + α (A1 – F1)
= 81,964 + 0.2 (85040 – 81,964)
= 82,579

F3 = F2 + α (A2 – F2)
= 82,579 + 0.2 (84,580 – 82,579)
= 82,979

F4 = F3 + α (A3 – F3)
= 82,979 + 0.2 (86,120 – 82,979)
= 83,607

F5 = F4 + α (A4 – F4)
= 83,607 + 0.2 (87,660 – 83,607)
= 84,418

F6 = F5 + α (A5 – F5)
= 84,418 + 0.2 (90,760 – 84,418)
= 85,686

Sl No. Months of 2022 Total sales Quantity Remarks Forecast

1 January 76,600

2 February 74,700

3 March 79,900

4 April 82,100

5 May 80,640

6 June 82,840

7 July 85,040 A1 81,964

7
8 August 84,580 A2 82,579

9 September 86,120 A3 82,979

10 October 87,660 A4 83,607

11 November 90,760 A5 84,418

12 December 89,860 A6 85,686

Figure 3.2: Chart for Exponential Smoothing

92,000
Chart for Exponential Smoothing
90,000

88,000

86,000

84,000

82,000

80,000

78,000

76,000
1 2 3 4 5 6
Actual 85,040 84,580 86,120 87,660 90,760 89,860
Forecast 81,964 82,579 82,979 83,607 84,418 85,686

8
 Chapter 4: Two Advanced Techniques
We have two advanced techniques named double exponential smoothing and simple
linear regression in this chapter.

4.1 Double Exponential Smoothing

The Double Exponential Moving Normal (DEMA) pointer was presented in January
1994 by Patrick G. Mulloy, in a piece within the "Specialized Investigation of Stocks &
Commodities" magazine: "Smoothing Information with Speedier Moving Averages."

By giving later numbers more weight, it tries to eliminate the typical lag associated with
moving averages. The name suggests that this is frequently done by using a twofold
exponential smoothing, although this is not the case. The word "double" refers to the fact
that an EMA's (Exponential Moving Normal) value is multiplied. The value "EMA of
EMA" is deducted from the already multiplied ema in order to maintain it consistent with
the real data and to eliminate the slack.

Double Exponential Smoothing is a general smoothing strategy to supply short-term

estimateswhen your data have drifted and don't have a regular component. This strategy
calculates energetic gauges for two components: level and trend. Double exponential
smoothing utilizes a level component and a trend component at each period. Twofold
exponential smoothing employments two weights, (also called smoothing parameters),
to upgrade the components at each period.

Formula,
Ft = a* At-1 + (1- a) *
(Ft-1 + Tt-1)Tt = b* (At-
1-Ft-1) + (1- b) * Tt-1
AFt = Ft + Tt
Where,
Ft = Unadjusted forecast
(before trend)Tt = Estimated
trend
AFt = Trend-adjusted forecast
Here, a = 0.5 and b = 0.5

9
 Table 4.1: Forecast calculation for Double Exponential Smoothing

Sl Months of Total sales Unadjusted Estimated Trend-adjusted

No. 2022 Quantity forecast (Ft) trend (Tt) forecast (Fat)

1 January 76,600
2 February 74,700
3 March 79,900
4 April 82,100
5 May 80,640
6 June 82,840 81,964 0
7 July 85,040 82,402 438 82,840
8 August 84,580 83,940 1,538 85,478
9 September 86,120 85,029 1,089 86,118
10 October 87,660 86,119 1,090 87,209
11 November 90,760 87,435 1,316 88,750
12 December 89,860 89,755 2,321 92,076

Figure 4.1: Chart for Double Exponential Smoothing

Chart for Double Exponential Smoothing

94,000
92,000
90,000
88,000
86,000
84,000
82,000
80,000
78,000
1 2 3 4 5 6
Actual 85,040 84,580 86,120 87,660 90,760 89,860
Forecast 82,840 85,478 86,118 87,209 88,750 92,076

4.2 Non-Linear Regression

Nonlinear regression is a form of regression investigation in which data is fit to a show
and after that communicated as a numerical work. Simple linear regression relates two
factors (X and Y) with a straight line (y = mx + b), whereas nonlinear relapse relates the
10
two variables in a nonlinear(bent) relationship.

The goal of the demonstration is to produce as few squares overall as possible. An indicator of
how far the Y observations deviate from the nonlinear (bent) work used to predict Y may be the
sum of squares. Finding the contrast between each Y point of the set's data and the fitted
nonlinear work comes first in the computation. Each of the contrasts is squared at that point. All
of the squared figures are combined in the final calculation. The work fits the information
targeted within the set better the smaller the total of these squared figures are. Logarithmic
capabilities, trigonometric capabilities, exponential capabilities, control capabilities, Lorenz
bends, Gaussian capabilities, and other fitting techniques are used in nonlinear relapse.
Table 4.2: Data for forecast calculation of Non-Linear Regression

Months of 2022 Actual Forecast

1 85,040 85,050

2 84,580 85,510

3 86,120 86,590

4 87,660 88,130

5 90,760 90,330

6 89,860 91,230

Figure 4.2: Chart for Non-Linear Regression Forecasting

Non Linear Regression

0 1 2 3 4 5 6 7
92,000 92,000
91,000 90,760 91,230 91,000
90,000 90,330 89,860
89,000 90,000
88,000 87,660 89,000
87,000 88,130 88,000
86,000 86,120
85,000 85,040 87,000
84,580 86,590
84,000 86,000
83,000 85,510
85,050 85,000
82,000
81,000 84,000
1 2 3 4 5 6

Actual Forecast

11
 Chapter 5: Comparison between Four Different Techniques

When it comes to predicting time series data, the choice of the right technique holds immense
influence over the precision and dependability of forecasts. This examination takes a deep dive
into four unique forecasting approaches: Moving Average, Exponential Smoothing, Double
Exponential Smoothing, Non-Linear Regression. Every method boasts its own set of advantages
and drawbacks, tailored to various data attributes, patterns, and prediction timeframes. The
comparison table of these four different forecasting methods are given below:

Table 5.1: Comparison between Four Different Forecasting Techniques

Months Weighted Exponential Double Non-Linear

Moving Smoothing Exponential Regression
Average Smoothing

July 81,964 81,964 82,840 85,050

August 83,038 82,579 85,478 85,510

September 83,991 82,979 86,118 86,590

October 84,974 83,607 87,209 88,130

November 86,647 84,418 88,750 90,330

December 88,579 85,686 92,076 91,230

12
 Chapter 6: Accuracy of Forecasts

For calculation the forecasting error we used two common methods: Mean Absolute Deviation
(MAD) and Mean Squared Error (MSE). These methods help assess the accuracy of a
forecasting model by quantifying the disparity between predicted values and actual observations.

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

• Mean Squared Error (MSE), MSE = ∑ (Actual − Forecast)² / n

6.1 Forecasting Error for Weighted Moving Average

Table 6.1: Calculation of MAD & MSE for Weighted Moving Averages

13
6.2 Forecasting Error for Exponential Smoothing

Table 6.2: Calculation of MAD & MSE for Exponential Smoothing

14
6.3 Forecasting Error for Double Exponential Smoothing

Table 6.3: Calculation of MAD & MSE for Double Exponential Smoothing

Months Exponential At-Ft |At-Ft| MAD |At-Ft|² MSE

Double
Smoothing

Actual (At) Forecast (Ft)

1,299 2,007,762
July 82,402 82,840 -438 438 191,844

August 83,940 85,478 -1,538 1,538 2,365,444

September 85,029 86,118 -1,089 1,089 1,185,921

October 86,119 87,209 -1,090 1,090 1,188,100

November 87,435 88,750 -1,316 1,316 1,730,540

December 89,755 92,076 -2,321 2,321 5,384,720

15
6.4 Forecasting Error for Non-Linear Regression

Table 6.4: Calculation of MAD & MSE for Non-Linear Regression

16
6.5 Comparison between MAD & MSE of Four Forecasting techniques

From the calculation, we get different accuracies for each of the forecasting techniques. The
comparison between MAD & MSE of Four Different techniques is given below table.

Table 6.5: Comparison between MAD & MSE of Four Forecasting techniques

SL NO. Different Forecasting Techniques MAD MSE

1 Weighted Moving Average 2,471.26 7024381.346

2 Exponential Smoothing 3,797.59 16,231,922.77

3 Double Exponential Smoothing 1,298.50 2,007,761.58

4 Nonlinear Regression 613.33 561,433.33

Figure 6.1: Chart for comparison between MAD & MSE of Four Forecasting techniques

17
 Chapter 7: Results and Discussion

The Weighted Moving Average method yields a relatively moderate MAD and a sizable MSE.
This suggests that while the average magnitude of errors is moderate, the squared differences
between forecasted and actual values are considerably high. It indicates that this method might
not be accurately capturing the variance in the data, resulting in larger errors overall.

Exponential Smoothing exhibits a higher MAD and an even larger MSE. This signifies that the
forecasting errors are comparatively larger, and the squared errors are significantly elevated. It
suggests that the Exponential Smoothing technique might not be effectively accounting for the
patterns and fluctuations in the data, resulting in less accurate predictions.

Double Exponential Smoothing showcases a lower MAD and a moderate MSE. These values
indicate that this method performs relatively better in terms of both the average magnitude of
errors and the squared differences. It suggests that this technique might be capturing the
underlying patterns in the data more accurately compared to the previous two methods.

Nonlinear Regression stands out with the lowest MAD and a considerably smaller MSE. These
values imply that this technique performs the best among the presented methods. It suggests that
the Nonlinear Regression method is effectively capturing the data's nuances and patterns, resulting
in smaller forecasting errors overall.

In summary, based on the MAD and MSE values, the Nonlinear Regression technique seems to
offer the most accurate predictions, followed by Double Exponential Smoothing, while Weighted
Moving Average and Exponential Smoothing exhibit larger errors. However, it's important to
consider other factors such as the specific characteristics of the data, the forecasting horizon, and
the underlying assumptions of each technique before making a final decision on the most suitable
forecasting approach.

18
 Chapter 8: Conclusion

Forecasting techniques play a pivotal role in predicting future trends, enabling informed decision-making
across various domains. In the realm of time series forecasting, the choice of technique wields the power
to determine the accuracy and reliability of predictions. Moving Average, a simple yet effective method,
provides a baseline for understanding trends by smoothing out short-term fluctuations. Exponential
Smoothing offers adaptability through weighted recent data, but its performance hinges on the extent of
data fluctuations. Double smoothing exponential building upon the principles of Exponential Smoothing,
incorporates an additional component to capture the trend, making it particularly effective for datasets
that exhibit both a consistent direction of change and recurring patterns. Nonlinear Regression is a
versatile forecasting technique that goes beyond linear relationships between variables.

In the realm of time series forecasting, the choice of technique wields the power to determine the
accuracy and reliability of predictions. As evidenced through Mean Absolute Deviation (MAD) and Mean
Squared Error (MSE) analyses, the forecasting methods' performance can be quantified, offering insights
into their accuracy. While simpler techniques like Moving Average and Exponential Smoothing provide
quick insights, they may fall short in accurately capturing complex data patterns.

In closing, this analysis underscores the crucial role that accurate forecasting plays in decision-making.
While Nonlinear Regression emerges as the most precise technique in this particular comparison, the
decision to choose a forecasting method should extend beyond MAD and MSE values. Contextual factors,
such as the nature of data, computational feasibility, and inherent assumptions, significantly influence
the most suitable technique. By understanding the strengths and limitations of each method,
practitioners can navigate the complex landscape of forecasting, making informed choices that lead to
more reliable predictions and strategic insights.

19
 References

Omera LPG. (n.d.). Retrieved from https://www.omeralpg.com/

Omera LPG. (n.d.). About Us. Retrieved from https://omeralpg.com/about

Wikipedia. (2023, August 25). Omera LPG. In Wikipedia. Retrieved August 25, 2023, from
https://en.wikipedia.org/wiki/Omera_LPG

Khan, M. J. (2019). Consumer behavioral intention on purchasing Omera LPG.

Sujana, S. I. (2019). Analysis of SCM of Omera Petroleum Limited: The future leader of the LPG
industry.

Omera LPG. (n.d.). Lab Test Reports. Retrieved from https://www.omeralpg.com/lab-test

20