Section II.

Analysis of Forecasting Methods

For this analysis, we used different methods to forecast the historical product demand for a
manufacturing company with footprints globally. The company provides thousands of products within
dozens of product categories. There are four central warehouses to ship products within the region it is
responsible for. We utilized four different forecast methods to forecast the demand, namely, Moving
Average, Exponential Smoothing, Linear Trend and Holt's method. All of the calculations were
performed using various tools on Microsoft Excel. Furthermore, a select few outliers were omitted from
the data during the calculations to provide more accurate forecasts.

We utilized four different forecast methods to forecast the sales, namely, Moving Average, Exponential
Smoothing, Linear Trend and Holt's method. All of the calculations were performed using various tools
on Microsoft Excel. Furthermore, a select few outliers were omitted from the data during the
calculations to provide more accurate forecasts.

Section II-A. Historical product demand forecast (year-2016)

 Moving
Holt's
Linear
Expo. Method
average
Trend of
for
Smoothing of
year
year
of year
2016
yeat2016
2016
2016
400000
300000
The four different demand forecasts for the
200000
200000 company is shown in Figure 1. For the exponential
100000
100000 smoothing, we initially used an alpha of 0.5. We
00
y y h il y e y t r r r r then used the solver method on excel to optimize
u aarry uaarry arrcch Apprril Maay Junne Juully guusst bbeer obbeer bbeer bbeer
nu ru
JJaan Feebbr M
a A M Ju J u g em
AAu pt em Occtto veemm ceemm the alpha by minimizing the Mean Squared Error
M t O Noov Deec
F SSeep N D (MSE). We obtained the following: Optimized
Alpha = 0.564. To analyze the effectiveness of
Actual
Actual Demand
Demand Forecasted
Forecasted Demand
Demand each of these four forecasting methods, we
performed five different types of error
calculations, namely Bias, Mean Absolute Deviation (MAD), Mean Squared Error (MSE), Mean Absolute
Percent Error (MAPE), and Tracking Signal. The values of these five errors for each of the four
forecasting methods for the year 2016 are described in Table 1.

From the table, it is evident that for Bias, the Linear Trend method resulted in the lowest error value of 0
whereas the Holt's method resulted in the highest error value of 0.7.

In terms of the Mean Absolute Deviation (MAD), the Exponential Smoothing method resulted in the
lowest error value of 326.8 while the Holt's method resulted in the highest error value of 843.4.

As for the Mean Squared Error (MSE), the Expo. Smoothing method resulted in lowest error value of
827429.9 and the Holt's method resulted in the highest error value of 3975529.1.

Average for year2016

Errors Moving Average Expo. Smoothing Linear Trend Holt's method
Bias -0.2 -0.3 0.0 0.7
MAD 404.1 326.8 759.3 843.4
MSE 1351733.2 827429.9 3297628.7 3975529.1
MAPE 953.0 829.4 2830.1 2645.6
Tracking Signal 0.0 -0.1 0.0 -0.5

In terms of the Mean Absolute Percent Error, the Exponential Smoothing method resulted in the lowest
error of 829 and the Holt's method resulted in the highest error value of 2645.6.

As for the Tracking Signal, the Linear Trend method resulted in the lowest error value of 0 and the
Exponential Smoothing method resulted in the highest error value of -0.1. From this discussion, it is
evident the Exponential Smoothing method resulted in the lowest error value for four out of five of the
error calculations. Thus, we can conclude that Exponential Smoothing method has provided the best
demand forecast for the year of 2016.
 Section II-B. Historical product demand forecast
Expo.
Holt's
Linear
MovingSmoothing
Method
Trend of
for
average of
year
of year
yeat
2015
year 2015
2015
2015 (year-2015)
300000
200000
The four different demand forecasts for the
100000
company is shown in Figure
0
0
rryy rryy chh riill ayy nee ullyy sstt beerr beerr beerr beerr
nua rruuaa aarrc Appr Ma JJun Ju ugguu m b toob mb mb The values of these five errors for each of the four
JJaan Feebb M
M Au ptteem OOcct ovveem ecceem
F SSeep NNo DDe forecasting methods for the year 2015 are
described in Table. From the table, it is evident
Actual
Actual Demand
Demand Forecasted
Forecasted Demand
Demand that for Bias, the Moving average resulted in the
lowest error value whereas the Holt's method
resulted in the highest error value of 41.4.

In terms of the Mean Absolute Deviation (MAD), the Exponential Smoothing method resulted in the
lowest error value of 312.5 while the Holt's method resulted in the highest error value of 802.6.

As for the Mean Squared Error (MSE), the Expo. Smoothing method resulted in lowest error value of
612716.2 and the Holt's method resulted in the highest error value of 2741427.8.

Average for year2015

Errors Moving Average Expo. Smoothing Linear Trend Holt's method
Bias -0.2 0.6 0.0 41.4
MAD 393.5 312.5 647.4 802.6
MSE 1115938.0 612716.2 1998795.8 2741427.8
MAPE 666.0 654.6 3047.9 2129.3
Tracking Signal 0.0 0.0 0.0 1.9

In terms of the Mean Absolute Percent Error, the Exponential Smoothing method resulted in the lowest
error of 654.6 and the Linear Trend resulted in the highest error value of 3047.9.

As for the Tracking Signal, all methods except the Holt's method resulted in the lowest error value of 0
and the Holt's method resulted in the highest error value of 1.9.

From this discussion, it is evident the Exponential Smoothing method resulted in the lowest error value
for four out of five of the error calculations. Thus, we can conclude that Exponential Smoothing method
has provided the best demand forecast for the year of 2015.
 Section II-C. Historical product demand forecast
Holt's
Linear
Expo.
Moving Method
Smoothingfor
Trend of
average of
ofyear
year
year
yeat 2014
2014
2014 (year-2014)
200000
160000
160000
150000
120000
100000
120000
80000
50000
80000
40000 0
0ry ry ch ril ay ne ly st er er er er
40000nua y rua y ar h Ap il M y Ju e July ug u t mb r t ob r mb r mb r The four different demand forecasts for the
Ja uarFeb uar Marc Apr Ma Jun Ju A guspte beOc obeove beece be
r
n
Ja 0Fyeb y M h AuSe tem ct N emD em company is shown in Figure.
r r c ril ay ne uly espt erO Neorv Deerc er
a a r Ap M Ju J Su b b b b
nu ru Ma ug em cto em em The values of these five errors for each of the four
J a F eb Actual Demand AForecasted
pt O Demand
ov ec
Forecasted Demand Se Actual Demand
NDemand
D
Actual Demand Forecasted forecasting methods for the year 2014 are
described in Table.

From the table, it is evident that for Bias, the Holt's method resulted in the lowest error whereas the
Moving Average resulted in the highest error.
In terms of the Mean Absolute Deviation (MAD), the Exponential Smoothing method resulted in the
lowest error while the Holt's method resulted in the highest.

As for the Mean Squared Error (MSE), the Expo. Smoothing method resulted in lowest and the Holt's
method resulted in the highest error.

Average for year2014

Errors Moving Average Expo. Smoothing Linear Trend Holt's method
Bias 2 -1 0 -14
MAD 321 254 608 657
MSE 586495 345477 1320558 1657499
MAPE 917 796 2840 2484
Tracking Signal 1 -1 0 -3

In terms of the Mean Absolute Percent Error, the Exponential Smoothing method resulted in the lowest
error and the Linear Trend resulted in the highest error.

As for the Tracking Signal, the Holt’s method resulted in the lowest error and the Moving average
method resulted in the highest error.

From this discussion, it is evident the Exponential Smoothing method resulted in the lowest error value
for four out of five of the error calculations. Thus, we can conclude that Exponential Smoothing method
has provided the best demand forecast for the year of 2014.
 Section II-D. Historical product demand forecast
Expo.
Moving
Holt's
Linear
Smoothing
Method
average
Trend of
for
of
of
year
year
year
yeat
2013
2013
2013
2013 (year-2013)
200000
150000
100000
The four different demand forecasts for the
50000
company is shown in Figure
0
ry ry ch ril ay ne ly st r r r r
ua rua ar Ap M Ju Ju gu mbe tobe mbe mbe The values of these five errors for each of the
n M u
J a F eb A pte Oc ve ce
S
e No De four forecasting methods for the year 2013 are
described in Table. From the table, it is evident
Actual Demand Forecasted Demand that for Bias, the Holt's method resulted in the
lowest error whereas the Linear Trend resulted
in the highest error.

In terms of the Mean Absolute Deviation (MAD), the Exponential Smoothing method resulted in the
lowest error while the Holt's method resulted in the highest error.

As for the Mean Squared Error (MSE), the Expo. Smoothing method resulted in lowest error and the
Holt's method resulted in the highest error.

Average for year2013

Errors Moving Average Expo. Smoothing Linear Trend Holt's method
Bias -2 -4 0 -72
MAD 437 313 672 824
MSE 759610 368931 1339542 1826984
MAPE 566 534 2108 1927
Tracking Signal 0 -1 0 -7
In terms of the Mean Absolute Percent Error, the Exponential Smoothing method resulted in the lowest
and the Linear Trend resulted in the highest error.

As for the Tracking Signal, the Holt’s Method resulted in the lowest error value and the Linear Trend
resulted in the highest error.

From this discussion, it is evident the Exponential Smoothing method resulted in the lowest error value
for four out of five of the error calculations. Thus, we can conclude that Exponential Smoothing method
has provided the best demand forecast for the year of 2013.

Section II-E. Historical product demand forecast (year-2012)

 Expo.
Holt's
Linear
MovingSmoothing
Method
Trend of
for
average of
year
of year
yeat
2012
year 2012
2012
2012
300000
200000 The four different demand forecasts for the
company is shown in Figure
100000
0
0 The values of these five errors for each of the four
ryy ry ch ril ay unne Julyy ustt berr berr berr berr
nua rua ar AAp M M Ju J ug m to m m forecasting methods for the year 2012 are
Jan Feeb M AAu pte Oc ve ce
described in the Table.
See
p No DDe

From the table, it is evident that for Bias, the

Actual Demand Forecasted Demand
Holt's method resulted in the lowest error value of
-344 whereas the Linear Trend method resulted in
the highest error value of 0.
In terms of the Mean Absolute Deviation (MAD), the Exponential Smoothing method resulted in the
lowest error while the Holt's method resulted in the highest error.

As for the Mean Squared Error (MSE), the Expo. Smoothing method resulted in lowest error and the
Holt's method resulted in the highest error.

Average for year2012

Errors Moving Average Expo. Smoothing Linear Trend Holt's method
Bias -5 -17 Linear Trend -344
MAD 291 234 555 1053
MSE 296018 209121 836098 9365925
MAPE 845 663 3125 6072
Tracking Signal -1 -5 0 -10

In terms of the Mean Absolute Percent Error, the Exponential Smoothing method resulted in the lowest
error and the Holt's method resulted in the highest error.

As for the Tracking Signal, Holt's method resulted in the lowest error value of -10 and the Linear Trend
resulted in the highest error value of 0.

From this discussion, it is evident that both Holt’s method and the Exponential Smoothing method
resulted in the lowest error value for four out of five of the error calculations. Thus, we can conclude
that both Holt’s method and Exponential Smoothing method has provided the best demand forecast for
the year of 2012.