Comparative analysis between different forecasting methods

Course: Managing Operations and Supply Chain (P501)

Prepared for
Dr. Abdullahil Azeem
Professor
Department of Industrial and Production Engineering
Bangladesh University of Engineering and Technology

Prepared by
Group 06
33- Ahmed Abdullah Albab
35- Md.Reazul Mumin
27(58D)- Towkir Ahmad

Institute of Business Administration, University of Dhaka

Date: December 03, 2021
December 03, 2021

Dr. Abdullahil Azeem

Professor
Department of Industrial and Production Engineering
Bangladesh University of Engineering and Technology

Subject: Submission of Assignment

Sir,
We, group 06, are glad to present to you our assignment titled “Comparative analysis between
different forecasting methods” as part of the Managing Operations and Supply Chain (P501)
course requirement. The main purpose of this assignment was to prepare a comparative analysis
of actual data and forecasted data as well as making comparison among the forecasting methods.
In the development of this paper, we have tried our level best to follow the guidelines you provided
us with and to comply with your lofty standards. We sincerely hope that this paper meets your
expectations. Please note that this report has been prepared under your supervision. Under no
circumstances will this report be produced for any other MBA (IBA) course ever. No part of this
report will be shared or republished without your authorization.

Sincerely yours,
33- Ahmed Abdullah Albab
35- Md.Reazul Mumin
27(58D)- Towkir Ahmad
Table of Contents

Introduction:.................................................................................................................................................. 1
Objectives: .................................................................................................................................................... 1
Methodology: ................................................................................................................................................ 1
Important Terms: .......................................................................................................................................... 1
The forecasting process: ............................................................................................................................... 2
Holt Method (Double Exponential method) ................................................................................................. 2
Holt- Winter Method: ................................................................................................................................... 3
Multiplicative Decomposition: ..................................................................................................................... 4
Comparison: .................................................................................................................................................. 7
Conclusion: ................................................................................................................................................... 8
Introduction:
People, societies, products and processes generate lot of data. Data science and business analytics
have evolved as more and more data becomes available at higher resolution and at higher
frequency. Quantitative forecasting is the science of using time series data for generating forecasts.
In other words, extrapolating a series of measurements into the future. Forecasting plays and
essential role in decision making in almost any environment we can imagine where data are
collected over time. For example: forecasting the daily or monthly sales or total no. of customers
in a restaurant or other service provider can help with recruitment, purchase and other
administrative and inventory management decisions as well as cash flow planning and space
design process and choice with technology to use. In a developing country life Bangladesh,
forecasting is very important. One example can be estimating the future demands of transportation
system, compare them with present capacity and finally make decision what can be done to meet
future demands. In today’s Internet of things (IoT), automated sensor data can help forecast usage
of offices, restrooms and other facilities. Such forecasts can help with staffing, replenishment and
other service operations. It can also support environmentally minded decision making and design.

Objectives:
• Forecast of sales information for final six months of our selected timeframe and compare
the forecasted data with actual data using three different techniques.
• Compare among the techniques used to forecast.

Methodology:
In this report, we have used the monthly sales data of a manufacturing company for the past 24
months (December, 2018 - November, 2020) which is a secondary dataset. The forecasting
techniques we have used are:
• Holt’s Method (Double exponential smoothing)
• Holt- Winter’s Method
• Multiplicative Decomposition

Important Terms:
Time Series forecasting: Time Series forecasting refers to generating a future value for the series
of interest.
Forecast Error: The difference between the actual value and forecasted value calculated for the
same time frame
Descriptive goal: Time series analysis describe the patterns in historical data.
Predictive goal: Time series forecasting predict future values of the series.

Page | 1
The forecasting process:
Forecasting is not a function in a software or a button to run but includes an entire
set of steps which are often not maintain same steps from start to end. It starts
with defining the goal, getting the data, then exploring and visualizing it. Finally
apply different forecasting methods, compare them and chose one. Often it can
be the case where we don’t have the right kind of data to solve the problem and
in that case we may need to go back and collect new data. Sometimes when
comparing among models no model seems to give satisfactory result and so we
have to try new models.

Holt Method (Double Exponential method)

Now, we're focusing on Holt exponential smoothing. The main idea in advanced exponential
smoothing is to take the simple exponential smoothing setup, and to expand it to capture trend,
and or seasonality. We can use advanced Exponential Smoothing, either for forecasting a series
with a trend or with seasonality, or with both.
The idea is to take simple exponential smoothing, but to add a trend component. Before we talk
about the method itself, let's think about what it means. We're making an assumption here that the
series has a level, it has trend. And of course it has noise, but there is no seasonality. Therefore,
when we're talking about the forecast, we're going to take an estimated level and combine it with
our estimated trend at the most recent time point. Here's what the Double exponential smoothing
formula looks like:

𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 = 𝐸𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝑙𝑒𝑣𝑒𝑙 + 𝑇𝑟𝑒𝑛𝑑 𝑎𝑡 𝑚𝑜𝑠𝑡 𝑟𝑒𝑐𝑒𝑛𝑡 𝑡𝑖𝑚𝑒 𝑝𝑜𝑖𝑛𝑡

𝐹𝑡+𝑘 = 𝐿𝑡 + 𝑘 𝑇𝑡

We’re going to take an estimated level and combine it with our estimated trend at the most recent
time point. This is our most recent level estimate. And our most recent trend, and k is how many
steps into the future.
We're trying to forecast we're going to have two updating equations, one for the level and one for
the trend, because we're learning both of them from the data. The updating equation for level is
similar to what we had in simple exponential smoothing, we're going to take the most recent data
point and use it to update our previous estimate of the level at time t minus one. The only difference
is that here we also add the trend and take it into account. What this equation is showing us is that
we're basically adjusting the previous level by adding a trend.
𝐿𝑡 = α𝑌𝑡 + (1 − α)(𝐿𝑡−1 + 𝑇𝑡−1 )
𝑤ℎ𝑒𝑟𝑒 𝐿𝑡−1 𝑖𝑠 𝑡ℎ𝑒 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑙𝑒𝑣𝑒𝑙 𝑎𝑛𝑑 𝑇𝑡−1 𝑖𝑠 𝑡ℎ𝑒 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑡𝑟𝑒𝑛𝑑

Page | 2
The second equation is going to update our estimate of the trend. So we're going to take the trend
in the previous time period and update it by looking at the difference between the most recent level
estimates. We're updating the previous trend by using the difference between the most recent level
values.

𝑇𝑡 = β (𝐿𝑡 − 𝐿𝑡−1 ) + (1 − β)𝑇𝑡−1

𝐻𝑒𝑟𝑒 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑡𝑟𝑒𝑛𝑑 𝑖𝑠 𝑢𝑝𝑑𝑎𝑡𝑒𝑑 𝑢𝑠𝑖𝑛𝑔 𝑡ℎ𝑒 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑡ℎ𝑒 𝑚𝑜𝑠𝑡 𝑟𝑒𝑐𝑒𝑛𝑡 𝑙𝑒𝑣𝑒𝑙 𝑣𝑎𝑙𝑢𝑒𝑠.

This setup allows the trend to vary and change in shape over time. The smoothing constant β is
controlling this Speed of adjusting the trend. So if the trend changes very quickly during the series,
we might want to learn it faster.
To choose the smoothing constants, alpha and beta, we can use default values that most software
will have. Those are based on empirical experience, which range in between 0.1 to 0.2. The other
approach is to try and find α and β that minimize some error metric, such as MSE, RMSE, or
MAPE, or any other type of metric over the training set. If you do that, you have to be careful of
overfitting because if you use those optimized alpha and beta, you might not be able to generalize
very well into the future.

Holt- Winter Method:

With Holt- Winter’s method, we'll be able to capture additive and multiplicative trend, as well as
additive and multiplicative seasonality. Winters method, also called Holt winters method, and even
sometimes called triple exponential smoothing takes the idea of Holt's method and adds a seasonal
component to create an even more complex system. Here’s what the Holt winters exponential
smoothing looks like.

𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 = 𝐸𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝑙𝑒𝑣𝑒𝑙 + 𝑇𝑟𝑒𝑛𝑑 + 𝑆𝑒𝑎𝑠𝑜𝑛𝑎𝑙𝑖𝑡𝑦 𝑎𝑡 𝑚𝑜𝑠𝑡 𝑟𝑒𝑐𝑒𝑛𝑡 𝑡𝑖𝑚𝑒 𝑝𝑜𝑖𝑛𝑡

𝐹𝑡+𝑘 = 𝐿𝑡 + 𝑘𝑇𝑡 + 𝑆𝑡+𝑘−𝑀

The assumption here is that our series contains level trend, seasonality with M seasons, and of
course noise. Therefore, our forecasting equation now combines estimates of our level of trend
and of the seasonal components.
In the Holt winters Exponential Smoothing, we have three smoothing constants, and therefore
three updating equations. In the level equation, there is one part that needs an explanation. S is a
seasonal component. So, when we divide Y by S, it means we are deseasonalizing the value of Y.

Page | 3
 𝑌𝑡
𝐿𝑡 = α + (1 − α)(𝐿𝑡−1 + 𝑇𝑡−1 )
𝑆𝑡−𝑀

In the level equation, we're updating the previous level Lt-1 by adding the previous trend estimate
Tt-1, and then combining with a deseasonalized value of Yt.
𝑇𝑡 = β(𝐿𝑡 − 𝐿𝑡−1 ) + (1 − β)𝑇𝑡−1

In the seasonality updating equation, we see that Yt is divided by the level component Lt, this
gives the detrended value of Y. So the seasonality is updated by combining the most recent
seasonal component St-m with a D trended value of Yt.

𝑌𝑡
𝑆𝑡 = 𝛾 + (1 − 𝛾)𝑆𝑡−𝑀
𝐿𝑡

We have three smoothing constants when we're talking about Holt winters method, and these
control the speed of learning from the data and answer questions such as How fast do we update
the level half or How fast do we update the trend and the seasonal components.

Multiplicative Decomposition:
The multiplicative decomposition or time series decomposition is one of the commonly used as
well as important foretasting methods. Here seasonal indices will allow us to understand when
there are short- term fluctuations and cyclic factor will allow us to understand highlighting the
importance of seasonal indices and cycle factors. The process is narrated below:
Step 1: First, we need to remove the short term fluctuation by using centered moving average
(CMA) method. Now the formula will be different for monthly data:

𝑌𝑡−6 + 𝑌𝑡−5 + ⋯ … … … . 𝑌𝑡 + 𝑌𝑡+1 + ⋯ … … … . +𝑌𝑡+5

𝑀𝐴𝑡 =
12

If the number of periods is odd, the moving average is automatically centered, so no further
adjustment is necessary for even number of periods, to center the moving average, a two-period
moving average is calculated. The formula is:

𝑀𝐴𝑡 + 𝑀𝐴𝑡+1
𝐶𝑀𝐴𝑡 =
2

Page | 4
Step 2: Next we need to calculate the Seasonal Factor or Degree of seasonality. It is basically the
ratio of actual data (Yt) and Centered moving average (CMAt) The formula is:

𝑌𝑡
𝑆𝐹𝑡 =
𝐶𝑀𝐴𝑡

Step 3: Next we have to calculate Seasonal Indices (SI) from Seasonal Factors (SF). SI is the
simple average of all like seasons. Then we have to calculate normalized SI:

𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑟𝑖𝑜𝑑𝑠 × 𝑆𝑒𝑎𝑠𝑜𝑛𝑎𝑙 𝐼𝑛𝑑𝑒𝑥

𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 𝑆𝐼 =
𝛴(𝑆𝑒𝑎𝑠𝑜𝑛𝑎𝑙 𝐼𝑛𝑑𝑒𝑥)

Once the SI is known, we can calculate deseasonalized data which is a ratio of actual data and
normalized SI.

𝑅𝑎𝑤 𝑑𝑎𝑡𝑎
𝐷𝑒𝑠𝑒𝑎𝑠𝑜𝑛𝑎𝑙𝑖𝑧𝑒𝑑 𝑑𝑎𝑡𝑎 =
𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 𝑆𝐼

Step 4: Once the seasonality is removed from the data. Unseasonalized forecast can be calculated
by estimating a simple linear equation:

𝑓(𝑇𝑖𝑚𝑒) = 𝑎 + 𝑏(𝑇𝑖𝑚𝑒)

Where ‘a’ is the intercept of unseasonalized data and ‘b’ is the slope of unseasonalized data.
Step 5: Finally we can calculate the Seasonalized forecast by multiplying the unseasonalized
forecast data with Seasonal indices (SI) and calculate error by comparing forecasted data with
actual data.

Page | 5
Comparison:

Holt- Winter’s Multiplicative

Parameters Holt’s Method
Method decomposition
Mean Squared Error
13398279249 4256251196.15 562189757952.93
(MSE)

Root Mean Square Error

115750.9363 65239.95092 749793.1434
(RMSE)

Forecast to actual (%)

94.71 % 126 % 238%
June-2020

Forecast to actual (%)

116.81 % 82 % 107%
July-2020

Forecast to actual (%)

106.44 % 151 % 102%
August-2020

Forecast to actual (%)

136.32 % 94 % 64%
September-2020

Forecast to actual (%)

103.94 % 116 % 86%
October-2020
Forecast to actual (%)
November-2020 126.09 % 85 % 65%

Page | 6
Comparison of MSE

Comparision of MSE for different methods

2.5E+10
2E+10
1.5E+10
MSE

1E+10
5E+09
0
Holt’s Method Holt- Winter’s Method Multiplicative
decomposition
Methods

Comparison of RMSE

Comparision of RMSE for different methods

200000

150000
RMSE

100000

50000

0
Holt’s Method Holt- Winter’s Method Multiplicative
decomposition
Methods

Comparison of Forecast to Actual (%)

Comparision of Forecast to actual (%) for different methods

250.00%
Forecast to actual( %)

200.00%
150.00%
Holt’s Method
100.00%
Holt- Winter’s Method
50.00%
Multiplicative decomposition
0.00%
Jun-20 Jul-20 Aug-20 Sep-20 Oct-20 Nov-20
Months

Page | 7
Conclusion:
The goal of forecasting is to identify the possible future scenarios and take necessary measures.
Although the main point is to be able to communicate generated forecast very clearly. It means the
actual and forecasted scenarios should be easily distinguishable. To convey the uncertainty, the
best approach is to use prediction intervals. Also it is very important to use more than one
techniques to forecast and provide necessary explanations so that the user can easily select the best
option. To monitor the performance of forecast, keeping track of errors by comparing the
forecasted data with the actual data can help to generate better forecast in future. It will also help
to identify any deteriorations and fix the model accordingly.
In our report, we have used Holt trend, Holt- Winter and multiplicative decomposition models to
forecast our monthly sales data. For our analysis is evident that multiplicative decomposition
model presents the highest error. On the other hand, Holt- winter method provides the minimum
error. This is because this model consider both trend and seasonality while Holt’s method considers
only trend.

Page | 8