Quarterly Demand for COKE

Year Quarter Period t Demand Dt

0
2008 1 1 256
2008 2 2 325
2008 3 3 645
2009 4 4 788 Determine the Suitable Forecasting Techniqu
2009 1 5 1900 Determine the Forecasts for Periods 13, 14,
2009 2 6 3087
2009 3 7 4523
2010 4 8 5342
2010 1 9 340
2010 2 10 276
2010 3 11 220
2011 4 12 135 See The Solution in other Workshe
2011 1 13
2011 2 14
2011 3 15
2012 4 16
 Method 1:
Moving Average
( Use When no observable Trend or Seasonality)

Applying Moving Average Method

Quarterly Demand for COKE No. of Demand Periods for Moving Average = 4
Absolute
Year Quarter Period t Demand Dt Level Forecast Error Dev MAD % Error MAPE TS
0
2008 2 1 256
2008 3 2 325
2008 4 3 645
2009 1 4 788 503.5
2009 2 5 1900 914.5 503.5 -1396.5 1396.5 1397 73.5% #VALUE! -1.0
2009 3 6 3087 1605 914.5 -2172.5 2172.5 1785 70.4% #VALUE! -1.2
2009 4 7 4523 2574.5 1605 -2918 2918 2162 64.5% #VALUE! -1.3
2010 1 8 5342 3713 2574.5 -2767.5 2767.5 2314 51.8% #VALUE! -1.2
2010 2 9 340 3323 3713 3373 3373 2526 992.1% #VALUE! 1.3
2010 3 10 276 2620.25 3323 3047 3047 2612 1104.0% #VALUE! 1.2
2010 4 11 220 1544.5 2620.25 2400.25 2400.25 2582 1091.0% #VALUE! 0.9

2011 1 12 135 242.75 1544.5 1409.5 1409.5 2436 1044.1% #VALUE! 0.6
2011 2 13 243
2011 3 14 243
2011 4 15 243
2012 1 16 243

As MAD and MAPE are fairly higher, this forecasting method is not appropriate for the given problem environment
Though the TS is within the acceptable range
Method 2:
Simple Exponential Smooting
( Use When no observable Trend or Seasonality)

Systematic Component of Demand = Level

Applying Exponential Smooting Method

Quarterly Demand for Coke Smoothing Constant for Level =0.1
Absolute
Year Quarter Period t Demand Dt Level Forecast Error Dev MAD % Error MAPE TS
0 1486.416667
2008 2 1 256 1363.375
2008 3 2 325 1259.53751363.375 1038.375 1038.375 1038 319.5% 320% 1.0
2008 4 3 645 1198.083751259.5375 614.5375 614.5375 826 95.3% 207.4% 0.7
2009 1 4 788 1157.075375
1198.08375410.08375 410.08375 688 52.0% 155.6% 0.6
2009 2 5 1900 1231.367838
1157.075375-742.924625742.924625 701 39.1% 126.5% -1.1
2009 3 6 3087 1416.931054
1231.367838
-1855.632163
1855.632163 932 60.1% 113.2% -2.0
2009 4 7 4523 1727.537948
1416.931054
-3106.068946
3106.068946 1295 68.7% 105.8% -2.4
2010 1 8 5342 2088.984154
1727.537948
-3614.462052
3614.462052 1626 67.7% 100.3% -2.2
2010 2 9 340 1914.085738
2088.984154
1748.9841541748.984154 1641 514.4% 152.1% 1.1
2010 3 10 276 1750.277164
1914.085738
1638.0857381638.085738 1641 593.5% 201.1% 1.0
2010 4 11 220 1597.249448
1750.277164
1530.2771641530.277164 1630 695.6% 250.6% 0.9
2011 1 12 135 1451.024503
1597.249448
1462.2494481462.249448 1615 1083.1% 326.3% 0.9
2011 2 13 1451
2011 3 14 1451
2011 4 15 1451
2012 1 16 1451
 Method 3:
Trend Correct Exponential Smoothing [HOLT'S Model)
( Use When no observable Seasonality)

Systematic Component of Demand = Level + Trend

Coke
Absolute Tracking
Demand Error E
Period (t) Level (L) Trend (T) Forecast (F) Error A MAD (t) % ErrorMAPE (t) Signal
(D) (t)
(t) (TS)
0 1380 16
1 256 1283 -7 1397 1141 1141 1141 446 446 1
2 325 1181 -26 1276 951 951 1046 293 369 0.9
3 645 1104 -36 1156 511 511 867 79 272 0.6
4 788 1041 -41 1069 281 281 721 36 213 0.4
5 1900 1089 -23 999 -901 901 757 47 180 -1.2
6 3087 1268 17 1066 -2021 2021 967 65 161 -2.1
7 4523 1609 82 1285 -3238 3238 1292 72 148 -2.5
8 5342 2056 155 1691 -3651 3651 1587 68 138 -2.3
9 340 2024 117 2211 1871 1871 1618 550 184 1.2
10 276 1955 80 2141 1865 1865 1643 676 233 1.1
11 220 1853 44 2035 1815 1815 1659 825 287 1.1
12 135 1721 9 1897 1762 1762 1667 1305 372 1.1
13 1730
14 1738
15 1747 Since Ft+n = Lt + nTt
16 1755

Min TS = -2.5 Max TS = 1.2

Estimate of change in demand(Trend) at t = 0 denoted by a 16 Estimate of demand at t = 0 or estimate of initial Level(L0)= 1380
Smoothing Constant of Level (α) = 0.1 Const for Trend (β)= 0.2
D= at + b

a = rate of change in demand per period = Initial estimate of the trend T0

b= estimate of demand period (t = 0) = estimate of the initial Level L0
As MAD and MAPE are fairly higher, this forecasting method is not appropriate for the given problem environmen
Though the TS is within the acceptable range
 Method 4:
Trend and Seasonality Corrected Exponential Smoothing [WINTER'S Model)
( Use When no both Trend and Seasonality is Observable)

Natural Gas.com
Deseasonaliz Absolut Trackin
Seasonal Forecas Error E MAD % MAP
Deseasonaliz ed Demand ( e Error g Signal
Perio Deman Level Factor 1 t (F) (t) (t) Error E (t)
ed Demand Trend (T) Rev ) A(t) (TS)
d (t) d (D) (L)
(D#)
0 1432 100
1 256 1485 96 1532 0.17 264 8 8 8 3 3 1
2 325 1528 90 1632 0.20 322 -3 3 5 1 2 -0.5
3 645 709 1575 86 1732 0.37 618 -27 27 12 4 3 -2.1
4 788 1260 1619 82 1833 0.43 731 -57 57 24 7 4 -2.4
5 1900 2090 1850 97 1933 0.98 1914 14 14 22 1 3 0.6
6 3087 3144 2121 114 2033 1.52 3393 306 306 69 10 4 4.4
7 4523 3518 2402 131 2134 2.12 5370 847 847 180 19 6 4.7
8 5342 2972 2702 148 2234 2.39 6814 1472 1472 342 28 9 4.3
9 340 2082 2742 137 2334 0.15 419 79 79 312 23 11 0.3
10 276 894 2754 125 2435 0.11 326 50 50 286 18 11 0.2
11 220 2746 111 2535 0.09 248 28 28 263 13 12 0.1
12 135 2721
98 2635 0.05
144 9 9 242 7 11 0.0
13 0.17 471
14 0.20 581
15 0.37 1122 Since, Ft+n = (Lt + nTt)St + n
16 0.43 1338

Periodicity = 4 -2.4 Max TS = 4.7

Estimate of change in demand at t = 0 denoted by a =100

Estimate of initial Demand or Initial Level L0 (b) = 1432
α = Smoothing Constant of Level = β = Smoothing constant of Trend 0.1
D= at + b 0.1
γ = Smoothing Constant for Season Factor
a = rate of change in demand per period = Initial estimate of the trend T0
b= estimate of demand period (t = 0) = estimate of the initial Level L0
As MAD and MAPE are relatively lower, this forecasting method seems to be quite environment
appropriate for the given problem
Applying this Model, the Forecasts for the next four periods are:
Period Forecast
13 471
14 581
15 1122
16 1338
with Mean Absolute Percentile Error at 11 %
and Mean Absolute Deviation of about 242 units approx.
Note:
In the given problem/context, the demand pattern is not uniform across all quarters and is unique to each period and quarter.
Demand behaviour is observed to be unique in each quarter.
Therefore, seasonal factor is unique to each period in a given quarter.
Hence seasonal factors of corresponding months should'nt be averaged.
The actual seasonal factors of corresponding months may be considered for computation of the forecasts.