Sales Forecast Case Studies Example

In this assignment I studied several methods of forecasting sales by the example of sales data for 2000 – 2009 (see table 1).

Sales for 2000 - 2009
Source: Joe F. Hair, Rolph E. Anderson, Rajiv Mehta, Barry J. Babin, Sales Management: Building Customer Relationships and Partnerships. Boston: Cengage Learning, 2008. Print.

In order to find expected sales in 2010, 5 forecasts were made using different quantitative methods:

Simple average for all periods.
Weighted moving average for 3 years.
Weighted moving average for 5 years.
Exponential moving average for 3 years.
Exponential moving average for 5 years.
In computing weighted moving average for 3 years the following weights were used: 0.433, 0.333, and 0.233. 5-years weighted average was computed using weights as follows: 0.3, 0.25, 0.2, 0.15, and 0.1.
Since forecasting using exponential moving average is more complex and a bit less intuitive, three forecasts were made for each period in order to show influence of parameter α. So the following values of α were used: 0.9, 0.5, 0.1.

Results of forecasts are reported in table 2.

Forecasted sales in 2010
It can be inferred from the calculations above that in case of mainly ascending time series simple average gives the least accurate results. Since real value of sales for 2010 is 2500, none of used methods produced results, close enough to real value. The reason is that all used techniques are backward-looking, because they only account for historical data and can not produce the values higher than those already observed. 3-year weighted moving average gives higher forecast than 5-year weighted moving average, since time series was almost always ascending, so taking into account older values produces lower value.
The table above shows that when using exponential moving average parameter α is crucial for forecasted value, because it represents the degree of decrease of older observations weights. So higher α mostly account for recent value and produce higher results while lower α gives smaller value. That`s why the closest value to real data was forecasted with the use of exponential moving average with α = 0.9