A MOVING AVERAGE

COL 1 COL 2 COL 3 COL 4 COL 5 COL 6

Year Quarter Period t Demand Dt Level Lt Forecast Ft

2022 1 1 15600
2022 2 2 18700
2022 3 3 26700
2022 4 4 28900 22475
2023 1 5 17700 23000 22475
2023 2 6 21200 23625 23000
2023 3 7 28600 24100 23625
2023 4 8 32300 24950 24100
2024 1 9 20000 25525 24950
2024 2 10 25000 26475 25525
2024 3 11 30000 26825 26475
2024 4 12 35000 27500 26825

1 The current forecast for Period 12 is 27500 and will be the same for all future

2 The Tracking Signal (TS) is well within the range of (±) 6, which indicates that t
does not contain any significant bias. It does, however, a fairly large MAD of 4

3 Given that the MAD is 4616, the estimate of standard deviation of forecast er
1.25 x 4616 = 5770. In this case, the standard deviation of forecast error is fair
to the size of the forecast. Hence 4 period Moving Average does not give us a
 COL 7 COL 8 COL 9 COL 10 COL 11 COL 12 COL 13
Forecast Absolute Percent
Error Et Error At MSE MAD Error MAPE TS

4775 4775 22,800,625 4775 27 27 1

1800 1800 13,020,313 3287.5 8 18 2
-4975 4975 16,930,417 3850 17 18 0.42
-8200 8200 29,507,813 4937.5 25 20 -1.34
4950 4950 28,506,750 4940 25 21 -0.33
525 525 23,801,563 4204 2.1 18 -0.27
-3525 3525 22,176,429 4107 12 17 -1.13
-8175 8175 27,758,203 4616 23 18 -2.78

be the same for all future periods say from period 13 to period 16.

±) 6, which indicates that the forecast using the 4- period Moving Average
er, a fairly large MAD of 4616 and MAPE 18 percent in period 12.

d deviation of forecast error, using a 4 period Moving Avearge is

tion of forecast error is fairy large relative
verage does not give us accurate forecast.
40000

35000

30000 f(x) = 1083.56643356643 x + 17931.8181818182

25000
Column F
20000
Linear (Column F)
15000 Linear (Column F)

10000

5000

0
1 2 3 4 5 6 7 8 9 10 11 12
B SIMPLE EXPONENTIAL SMOOTHING

Year Quarter Period t Demand Dt

2022 1 1 15600
2022 2 2 18700
2022 3 3 26700
2022 4 4 28900
2023 1 5 17700
2023 2 6 21200
2023 3 7 28600
2023 4 8 32300
2024 1 9 20000
2024 2 10 25000
2024 3 11 30000
2024 4 12 35000
24975
1 The Exponential Smoothing method is appropriate when demand has no observable tre
as seen in the graph. The trend line does not represent the demand .
Hence the following equation applies:
Systematic component of demand = Level
The initial estimate of L0 is taken to be the average of all historical data because deman
assumed to have no observable trend or seasonlity. Given the demand data for periods
we have the foloowing equation: n The current forecast for all future per
Ln = 1/n ∑Dt
i= 1

current estimate of level and is given as follows.

2 The current forecast for all future periods is equal to the current estimate of level and is
Ft+1 = Lt and Ft+n = Lt
3 After observing the demand (Dt+1) for period t+1, we revise the estimate of the level as
Lt+1 = αDt+1 + (1- α)Lt
where α is a smoothing constant for the level 0 < α < 1. The revised value of the level is
observed value of the level (Dt+1) for period t+1 and the old estimate of te level (Lt) in

4 The current estimate of the level is a weighted average of all the past observations of d
weighted higher than older observations. A gigher value of α, corresponds to a forecast
recent observations, whereas a lower value of α, represents a more stable forecast that
observations.

5 Restating the above table and starting with the initial forecst of period '0' (avearge of 12
compute Simple Exponential Smoothing.
 COL 1 COL 2 COL 3 COL 4 COL 5 COL 6 COL 7
Period t Demand D Level Lt Forecast Ft Error Et Absolute Mean Squared
ErrorAt Error MSEt

0 24,975
1 15600 24,038 24,975 9,375 9375 87,890,625
2 18700 23,504 24,038 5,338 5338 58,189,766
3 26700 23,823 23,504 -3,196 3196 42,198,515
4 28900 24,331 23,823 -5,077 5077 38,091,917
5 17700 23,668 24,331 6,631 6631 39,267,665
6 21200 23,421 23,668 2,468 2468 33,738,170
7 28600 23,939 23,421 -5,179 5179 32,749,944
8 32300 24,775 23,939 -8,361 8361 37,394,436
9 20000 24,298 24,775 4,775 4775 35,773,033
10 25000 24,368 24,298 -702 702 32,245,065
11 30000 24,931 24,368 -5,632 5632 32,197,432
12 35000 25,938 24,931 -10,069 10069 37,962,934
 40000

35000

30000

25000

20000

15000

10000

5000

0
1 2 3 4 5 6 7 8 9 10 11 12

d has no observable trend or seasonality

al data because demand has been

mand data for periods 1 through n,
ecast for all future periods is equal to the

estimate of level and is given as follows:

estimate of the level as follows:

ed value of the level is a weighted average of the

mate of te level (Lt) in period t.

past observations of demand, with recent observations

rresponds to a forecast that is more responsive to
ore stable forecast that is less responsive to recent

period '0' (avearge of 12 periods ) we now

COL 8 COL 9 COL 10 COL 11
MADt % Error MAPEt TSt

9375 60 60 1
7356 29 44 2
5970 12 34 2
5746 18 30 1
5923 37 31 2
5347 12 28 3
5323 18 26 2
5703 26 26 0
5600 24 26 1
5110 3 24 1
5158 19 23 0
5567 29 24 -2
 Column D
Column E
Linear (Column E)

7 8 9 10 11 12
 TREND-CORRECTED EXPONENTIAL SMOOTHING (Holt's Model)

Period t Demand Dt SUMMARY OUTPUT

1 15600
2 18700 Regression Statistics
3 26700 Multiple R 0.6247805213
4 28900 R Square 0.3903506998
5 17700 Adjusted R Square 0.3293857698
6 21200 Standard Error 5120.7800296
7 28600 Observations 12
8 32300
9 20000 ANOVA
10 25000 df SS MS
11 30000 Regression 1 167898619 167898618.88
12 35000 Residual 10 262223881 26222388.112
Total 11 430122500

Coefficients Standard Error t Stat

Intercept 17931.818182 3151.62224 5.6897105023
X Variable 1 1083.5664336 428.221138 2.5303898762

1 In this method the systematic component of demand is as follows:

Systematic Component of Demand = Level + Trend
This method involves the first step in making a regression analysis of the given data to find

2 We obtain an initial estimate of level of trend by running a linear regression between dem
Dt = b + at.
In this case running a linear regression between demand and time periods is appropriate b
has a trend but no seasonlality. The underlying relationship between demand and time is t
b ' measuares the estimate of demadn at period 't' = 0 and is an estimate of initial level L0
The slope measures the rate of change in demand per period and is the initial estimate of
In period 't', given estimates of Level Lt and trend Tn the forecast for future period is expre
Fn+1 = Lt + Tn and Ft+n = Lt + nTt
Herein we have found the Intercept as 17931 and Coefficient as 1083.
These two elements are L0 = 17931and T0 = 1083respectively.

3 After observing demand for period 't' we revise the estimates for level and trend as follow

L t+1 = αDt+1 + (1-α)(Lt + Tt)

and
 Tt+1 = β(Lt+1 - Lt) + (1-β)Tt

Where 'α' is a smoothing coefficient for the level , 0 < α < 1 and 'β' is a smoothing constant
We observe that in each of the two updates, the revised estimate (of level or trend) is a w
observed value and the old estimate.

4 Restating the Table we calculate the Trend -Corrected smoothing Coefficient as follows:

COL 1 COL 2 COL 3 COL 4 COL 5 COL 6 COL 7

Period t Demand D Level Lt Trend T Forecast Ft Error Et Absolute
ErrorAt
0 17931 1083
1 15600 18672.6 1015 19014 3414 3414
2 18700 19588.59 995 19687 987 987
3 26700 21195.21 1117 20584 -6116 6116
4 28900 22971.26 1249 22313 -6587 6587
5 17700 23568.28 1119 24220 6520 6520
6 21200 24338.23 1049 24687 3487 3487
7 28600 25708.43 1113 25387 -3213 3213
8 32300 27369.43 1223 26822 -5478 5478
9 20000 27732.95 1051 28592 8592 8592
10 25000 28405.45 975 28784 3784 3784
11 30000 29442.6 988 29381 -619 619
12 35000 30887.18 1079 30430 -4570 4570

5 Thus using Holt's Method the forecast for the next four periods is as follows:
F13 = L12 +T12 = 30887+1X1079=31966
F14 = L13 +T13 =30887+2X1079=33045
F15 = L14 +T14 = 30887+3X1079=34124
F16 = L15 +T15 = 30887+4X1079=35203

6 In this case, MAD =4447. Thus the estimate of standard deviation of forecast error using H
with α =0.10 and β = 0.20 is 1.25 x 4447= 5558. In this case. The standard deviation of fore
to other two cases is though less yet still fairly large.
 RESIDUAL OUTPUT

Observation Predicted Y
1 19015.385
2 20098.951
3 21182.517
4 22266.084
5 23349.65
6 24433.217
7 25516.783
8 26600.35
9 27683.916
F Significance F 10 28767.483
6.4028729254068 0.0298509429 11 29851.049
12 30934.615

P-value Lower 95% Upper 95% Lower 95.0% Upper 95.0%

0.0002011339739 10909.566213 24954.070151 10909.566213 24954.070151
0.0298509428975 129.43027768 2037.7025895 129.43027768 2037.7025895

e given data to find the intercept and Coefficient.

ssion between demand Dt and time aperiod t of the following form:

ods is appropriate because we have assumed that demand

emand and time is thus linear. The constant
ate of initial level L0
e initial estimate of the trend T 0.
ture period is expressed as follows:

and trend as follows:

smoothing constant for the period 0 < β<1.
vel or trend) is a weighted average of the

COL 8 COL 9 COL 10 COL 11 COL 12

Mean Squared MADt % Error MAPEt TSt
Error MSEt

11,655,396 3414 22 22 1
6,315,098 2201 5 14 2
16,680,338 3506 23 17 0
23,359,017 4276 23 18 -2
27,190,100 4725 37 22 0
24,684,857 4519 16 21 0
22,633,088 4332 11 20 0
23,555,574 4475 17 19 -2
29,141,097 4933 43 22 0
27,658,730 4818 15 21 1
25,179,170 4436 2 20 1
24,821,163 4447 13 19 0

ecast error using Holt's model

rd deviation of forecast error relative
 Period t Demand Dt Residuals
Residuals 1 15600 19015.38 -3415.385
-3415.385 2 18700 20098.95 -1398.951
-1398.951 3 26700 21182.52 5517.483
5517.4825 4 28900 22266.08 6633.916
6633.9161 5 17700 23349.65 -5649.65
-5649.65 6 21200 24433.22 -3233.217
-3233.217 7 28600 25516.78 3083.217
3083.2168 8 32300 26600.35 5699.65
5699.6503 9 20000 27683.92 -7683.916
-7683.916 10 25000 28767.48 -3767.483
-3767.483 11 30000 29851.05 148.951
148.95105 12 35000 30934.62 4065.385
4065.3846
D TREND-AND SEASONALITY-CORRECTED EXPONENTIAL SMOOTHING
(WINTER'S MODEL)

Winter's model assumes that the systematic component of demand is as follows:

Systematic Component of Demand = (Level +Trend) *Seasonal Factor

1 First we have to estimate the deseasonsiled demand as shown below:

Period t Demand Dt Deseasonalised Deseasonalised Seasonal

Demand Demand Factor
Col 1 Col 2 Col 3
1 15600 21851 0.71
2 18700 22572 0.83
3 26700 22737.5 23293 1.33
4 28900 23312.5 24014 1.41
5 17700 23862.5 24735 0.84
6 21200 24525 25456 0.98
7 28600 25237.5 26177 1.29
8 32300 26000 26898 1.43
9 20000 26650 27619 0.86
10 25000 27162.5 28340 1.06
11 30000 25000 29061 1.24
12 35000 19375 29782 1.42

The deseasonalised demand is derived in Column 3 as:

For period 3 = (C13+C17+2*SUM(C14:C16))/8

2 Next we take the regression of deseasonilised demand as shown above

The regression equation is Y = 11855 + 1690 *t
For period 1 it is 11855 + 1690*1 =13545

3
COL 1 COL 2 COL 3 COL 4 COL 5 COL 6
Seasonal
Period t Demand D Level Lt Trend T Factor St Forecast F

11855 1690
1 15600 14527 1788 0.47 6366
2 18700 16895 1846 0.67 10931
3 26700 18955 1868 1.16 21740
4 28900 20652 1851 1.66 34566
5 17700 23261 1926 0.47 10576
 6 21200 25510 1959 0.67 16875
7 28600 27328 1945 1.16 31863
8 32300 28781 1895 1.66 48592
9 20000 31271 1955 0.47 14418
10 25000 33430 1975 0.67 22261
11 30000 34928 1928 1.16 41070
12 35000 36067 1849 1.66 61180
0.47 17820
0.67 26642
1.16 48271
1.66 72146
mand is as follows:

SUMMARY OUTPUT

Regression Statistics
Multiple R 0.6643075844
R Square 0.4413045667
Adjusted R Square 0.3481886611
Standard Error 5032.4147695
Observations 8

ANOVA
df SS MS F Significance F
Regression 1 120023809.52381 120023809.52 4.7393038178 0.0723605989
Residual 6 151951190.47619 25325198.413
Total 7 271975000

Coefficients Standard Error t Stat P-value Lower 95%

Intercept 11855.952381 6836.0091340324 1.7343382884 0.1335516408 -4871.159384
X Variable 1 1690.4761905 776.518457338947 2.1769942163 0.0723605989 -209.5960254

COL 7 COL 8 COL 9 COL 10 COL 11 COL 12

Error Et Absolute Mean Squared MADt % Error MAPEt

ErrorAt Error MSEt

-9234 9234 85,263,986 9234 59.2 59.2

-7769 7769 145,614,772 8501 41.5 50.4
-4960 4960 170,214,887 7321 18.6 39.8
5666 5666 202,318,237 6907 19.6 34.7
-7124 7124 253,065,527 6950 40.2 35.8
-4325 4325 271,769,320 6513 20.4 33.3
3263 3263 282,417,271 6049 11.4 30.1
16292 16292 547,834,565 7329 50.4 32.7
-5582 5582 578,992,229 7135 27.9 32.1
-2739 2739 586,494,101 6695 11.0 30.0
11070 11070 709,037,253 7093 36.9 30.7
26180 26180 1,394,436,554 8684 74.8 34.3
17820 17820
26642 26642
48271 48271
72146 72146
 RESIDUAL OUTPUT

ObservationPredicted Y Residuals
1 20308.33 -2608.333
2 21998.81 -798.8095
3 23689.29 4910.714
4 25379.76 6920.238
5 27070.24 -7070.238
6 28760.71 -3760.714
7 30451.19 -451.1905
8 32141.67 2858.333

Upper 95% Lower 95.0% Upper 95.0%

28583.0641455515 -4871.159384 28583.064146
3590.54840637167 -209.5960254 3590.5484064

COL 13

TSt

-1.00
-2.00
-3.00
-2.36
-3.37
-4.26
-4.05
-1.12
-1.93
-2.47
-0.77
2.39