Some time ago Robert launched a new product some years ago at first, sales were good but now the figures are causing concern. Robert wants a more accurate sales forecast to produce detailed cash forecasts.

Since there is some seasonality present in the raw data, the series for sales shown below represents the underlying trend based on an averaging process:

Year Quarter Trend point Sales (cartons)
X y
2016 3rd 1 10,000
2016 4th 2 10,790
2017 1st 3 10,920
2017 2nd 4 11,000
2017 3rd 5 11,050
2017 4th 6 11,080
2018 1st 7 11,085
2018 2nd 8 11,095
2018 3rd 9 11,120
2018 4th 10 11,130

On average, quarters, 1 and 3 are 5% and 6% respectively above trend whilst quarters 2 and 4 are respectively 2% and 9% below trend. Some preliminary calculations on the above ten observations have been carried out and the results are summarised below:

Results from ten periods’ observations:

Linear regression y = a + bx Slope = 82.67 Intercept = 10,472.33 Coefficient of determination = 0.535

It is required to make forecasts of sales for quarters 3 and 4 in 2019 and for quarters 1 and 2 in 2020 but there is some discussion on whether the ten-period data shown above are suitable for forecasting or whether only the last five periods would provide a better basis for forecasting. Linear analysis of the last five periods only gives the following intermediate results:

Results of last five periods’ observations:

y = 555.10

x2 = 330

y2 = 61,627.40

xy = 4,442.15

Note: the y values have been scaled down by 100 times for ease of calculation.

Required:

a. Forecast the sales of the four quarters required using the ten-period observations results.

(6 Marks)

b. Prepare similar forecasts based on the last five periods’ observations.

(8 Marks)

c. Explain which forecasting basis produces the better forecast. (6 Marks)

a.

Year Quarter Trend point Sales trend (cartons)
x y = a + bx
2019 3 13 10,472.33 + (82.67 × 13) = 11,547
2019 4 14 10,472.33 + (82.67 × 14) = 11,630
2020 1 15 10,472.33 + (82.67 × 15) = 11,712
2020 2 16 10,472.33 + (82.67 × 16) = 11,795

The seasonally adjusted forecasts are therefore as follows:

Year Quarter Seasonally adjusted forecast sales (cartons)
2019 3 11,547 × 1.06 = 12,240
2019 4 11,630 × 0.91 = 10,583
2020 1 11,712 × 1.05 = 12,298
2020 2 11,795 × 0.98 = 11,559

b. We need to calculate the regression line of the last five periods’ observations.

This line is y = a + bx

where b = n xy − x y / n x2 − x 2 = 5 × 4,442.15 −(40 × 555.1) / 5 × 330 − 40 × 40 = 6.75 /50 = 0.135

a = 555.1 /5 −0.135 × 40 /5 = 111.02 −1.08 = 109.94

We need to multiply the “a and b’’ values back up by 100 to compensate for the fact that the y values were scaled down by 100 before the sigma calculations were carried out.

So, the regression line is y = 10,994 + 13.5x

Year Quarter Trend point Sales trend (cartons)
x y = a + bx
2019 3 13 10,994 + (13.5 × 13) = 11,169.5
2019 4 14 10,994 + (13.5 × 14) = 11,183.0
2020 1 15 10,994 +(13.5 × 15) = 11,196.5
2020 2 16 10,994 + (13.5 × 16) = 11,210.0

The seasonally adjusted forecasts are therefore as follows:

Year Quarter Seasonally adjusted forecast sales (cartons)
2019 3 11,169.5 × 1.06 = 11,840
2019 4 11,183.0 × 0.91 = 10,177
2020 1 11,196.5 × 1.05 = 11,756
2020 2 11,210.0 × 0.98 = 10,986

c. One common method of deciding on which forecasting basis produces the better result is to compare the coefficients of determination. With ten periods’ observations we are told that r2 = 0.535.

With five periods’ observations we have to calculate the coefficient of determination from the equation given in the tables provided.

r2 = ( n xy − x y )2 / ( n x2 − x 2 ) ( n 𝑦2 −( 𝑦)2 ) = 5 × 4,442.15 − 40 × 555.1 2 / 5 × 330 − 40 × 40 5 × 61,627.4 −(555.1 × 555.1) = 6.752 /50 × 0.99 = 0.92

Prima facie the forecasting method using five periods’ observations is much better than the other using ten periods, since the coefficient of determination of the former is 0.92 compared with 0.535 for the latter.

We can explain 92% of the variations in sales by the passage of time using the former method but can explain only 53.5% of the variations in sales using the latter method.

online
Knowsia AI Assistant

Conversations

Knowsia AI Assistant