This paper is about demand forecasting. A scenario is presented, and three techniques are used. These are simple moving average, weighted moving average, and the exponential smoothing technique. Then, at the conclusion of the different calculations, it is argued as to which of these techniques performed better, and how to improve the testing model.
LOGISTICS
The objective of this case is to determine the best method for forecasting tire sales. The first method is a three-month moving average with normal weights (1/3 each). The second is a three-month weighted moving average. The third method will be with exponential smoothing.
The company has very little past sales data with which to work, only 2011 and part of 2010. This presents some issues with respect to forecasting, so some techniques to estimate the forecast sales is going to need to be conducted.
Using a simple three-month moving average for 2012, the results are as follows, and they trend to 11130 as the number.
Last Year
period moving avg
Total Demand
Using a weighted three months average, with the nearest month being .6 and the farthest being .1, we get the following forecast for the coming year: The middle would be 0.3 in the weights. This figures trend towards 11192 as the number.
Month
Last Year
period moving avg
January
11177.0
February
11205.1
March
11189.2
April
10691
11192.7
May
11192.9
June
11192.5
July
10192
11192.6
August
10751
11192.6
September
11192.6
October
10193
11192.6
November
11599
11192.6
December
11130
11192.6
3. The forecast using exponential smoothing is as follows:
Month
ES
Forecast
January
February
10152.4
March
April
May
June
July
August
September
October
November
December
This figure is low, inherently, because the starting point was lot. The starting point was reasonable based on the company's performance last January, but that level is lower that were the company usually is. Thus given a different starting point the model might have performed better.
4. To determine the best forecast method in this situation there are some different metrics that can be used. The mean absolute deviation measures how much off from the projections the actual sales were under the three different forecasting techniques. This measures the amplitude of the forecast error for each month to determine how much each technique was off each month.
Thus, according to these figures, the best-performing system for forecasting is the simple moving average method. This method, with no weights or smooth, outperformed. In a sense, this is not particularly surprising. While in the real world the tire business varies by season, in this example the demand numbers appear to have been randomized. Each of these methods is essentially the same, but for the weights of the months in question. However, since the demand function is not experiencing linear growth, none of these approaches can be expected to be particularly strong with respect to the forecast. Simple moving came closest to predicting the actual demand, so in this instance it was the most reliable.
You’re 85% through this paper. Sign up to read the full paper.
Sign Up Now — Instant Access Already a member? Log inAlways verify citation format against your institution’s current style guide requirements.