In: Statistics and Probability
Using the data provided:
data:
Year | Quarter | Revenue |
1999 | Qtr1 | 1,939 |
Qtr2 | 2,373 | |
Qtr3 | 2,651 | |
Qtr4 | 3,111 | |
2000 | Qtr1 | 3,187 |
Qtr2 | 3,634 | |
Qtr3 | 3,702 | |
Qtr4 | 3,738 | |
2001 | Qtr1 | 3,627 |
Qtr2 | 3,916 | |
Qtr3 | 3,588 | |
Qtr4 | 2,932 | |
2002 | Qtr1 | 2,931 |
Qtr2 | 3,556 | |
Qtr3 | 3,812 | |
Qtr4 | 4,085 | |
2003 | Qtr1 | 4,570 |
Qtr2 | 4,189 | |
Qtr3 | 4,594 | |
Qtr4 | 4,576 | |
2004 | Qtr1 | 5,245 |
Qtr2 | 6,276 | |
Qtr3 | 6,558 | |
Qtr4 | 7,420 |
2004 | 2003 | 2002 | 2001 | 2000 | 1999 | |
Qtr1 | 5,245 | 4,570 | 2,931 | 3,627 | 3,187 | 1,933 |
Qtr2 | 6,276 | 4,189 | 3,556 | 3,916 | 3,634 | 2,373 |
Qtr3 | 6,558 | 4,594 | 3,812 | 3,588 | 3,702 | 2,651 |
Qtr4 | 7,429 | 4,576 | 4,085 | 2,932 | 3,738 | 3,111 |
Year | 25,508 | 17,929 | 14,384 | 14,063 | 14,300 | 10,068 |
Plot of the Raw Data
I don't find any seasonality effect in the data. Definitely, there is a trend. So, my choice is Holt's method. We can implement Holt's method in Minitab (or any other software) or in Excel spreadsheet. I prefer Excel.
Alpha = | 0.99 | Beta = | 0.1 | ||||
Period | Revenue (Yt) |
At | Tt | Period ahead |
Forecast | Absolute Error | Absolute % Error |
1 | 1939 | 1939.0 | 0.0 | ||||
2 | 2373 | 2368.7 | 43.0 | 1 | 1939.0 | 434.0 | 18.3% |
3 | 2651 | 2648.6 | 66.7 | 1 | 2411.6 | 239.4 | 9.0% |
4 | 3111 | 3107.0 | 105.8 | 1 | 2715.3 | 395.7 | 12.7% |
5 | 3187 | 3187.3 | 103.3 | 1 | 3212.9 | 25.9 | 0.8% |
6 | 3634 | 3630.6 | 137.3 | 1 | 3290.5 | 343.5 | 9.5% |
7 | 3702 | 3702.7 | 130.8 | 1 | 3767.8 | 65.8 | 1.8% |
8 | 3738 | 3739.0 | 121.3 | 1 | 3833.4 | 95.4 | 2.6% |
9 | 3627 | 3629.3 | 98.2 | 1 | 3860.3 | 233.3 | 6.4% |
10 | 3916 | 3914.1 | 116.9 | 1 | 3727.6 | 188.4 | 4.8% |
11 | 3588 | 3592.4 | 73.0 | 1 | 4031.0 | 443.0 | 12.3% |
12 | 2932 | 2939.3 | 0.4 | 1 | 3665.5 | 733.5 | 25.0% |
13 | 2931 | 2931.1 | -0.5 | 1 | 2939.7 | 8.7 | 0.3% |
14 | 3556 | 3549.7 | 61.5 | 1 | 2930.6 | 625.4 | 17.6% |
15 | 3812 | 3810.0 | 81.3 | 1 | 3611.2 | 200.8 | 5.3% |
16 | 4085 | 4083.1 | 100.5 | 1 | 3891.3 | 193.7 | 4.7% |
17 | 4570 | 4566.1 | 138.8 | 1 | 4183.6 | 386.4 | 8.5% |
18 | 4189 | 4194.2 | 87.7 | 1 | 4704.9 | 515.9 | 12.3% |
19 | 4594 | 4590.9 | 118.6 | 1 | 4281.8 | 312.2 | 6.8% |
20 | 4576 | 4577.3 | 105.4 | 1 | 4709.5 | 133.5 | 2.9% |
21 | 5245 | 5239.4 | 161.0 | 1 | 4682.7 | 562.3 | 10.7% |
22 | 6276 | 6267.2 | 247.7 | 1 | 5400.4 | 875.6 | 14.0% |
23 | 6558 | 6557.6 | 252.0 | 1 | 6515.0 | 43.0 | 0.7% |
24 | 7420 | 7413.9 | 312.4 | 1 | 6809.6 | 610.4 | 8.2% |
25 | 1 | 7726.3 | |||||
26 | 2 | 8038.7 | |||||
27 | 3 | 8351.2 | |||||
28 | 4 | 8663.6 | |||||
MAD = | 333.3 | ||||||
MAPE = | 8.5% |
Formulation
Determine Bias (With Tracking Signal)
Period | Revenue (Yt) |
Forecast | Absolute Error | Error | RSFE | Tracking Signal | Biased? |
13 | 2931 | 2770.4 | 160.6 | 160.6 | -443.0 | -2.76 | Y |
14 | 3556 | 2813.8 | 742.2 | 742.2 | 299.2 | 0.40 | N |
15 | 3812 | 3637.0 | 175.0 | 175.0 | 474.2 | 2.71 | Y |
16 | 4085 | 3938.2 | 146.8 | 146.8 | 621.0 | 4.23 | Y |
17 | 4570 | 4250.1 | 319.9 | 319.9 | 940.9 | 2.94 | Y |
18 | 4189 | 4820.3 | 631.3 | -631.3 | 309.6 | 0.49 | N |
19 | 4594 | 4269.8 | 324.2 | 324.2 | 633.8 | 1.95 | N |
20 | 4576 | 4762.9 | 186.9 | -186.9 | 446.9 | 2.39 | Y |
21 | 5245 | 4694.2 | 550.8 | 550.8 | 997.8 | 1.81 | N |
22 | 6276 | 5510.9 | 765.1 | 765.1 | 1762.8 | 2.30 | Y |
23 | 6558 | 6745.4 | 187.4 | -187.4 | 1575.4 | 8.41 | Y |
24 | 7420 | 6975.6 | 444.4 | 444.4 | 2019.8 | 4.55 | Y |
Considering an acceptable range of [-2,2] for the tracking signal, we cannot say that the forecast is free from bias.
Overall, the bias is high and the MAPE and MAD are also not very low (e.g. a MAPE below 5% would have been a good estimate). We thus propose a more advanced model such as ARIMA.