Abstract
Five different forecasting models were used to examine a logo branded coffeemaker that is sold by Starbucks. A simple moving average was used for the first two models using 5 weeks past (figure 1) data and 3 weeks past data (Figure 2. The next method used was a exponential smoothing method with .4 alpha and 5 weeks data (Figure 3) and .2 alpha with 3 weeks data (Figure 4). The mean absolute deviation, mean absolute percent error and tracking signal were calculated based off of the total of all segments.
Simple Moving Average
Looking at Figure 1 and Figure 2 where a moving average forecast was used, the 3 and 5 week data were very similar in the mean absolute deviation, and percent error. The tracking signal is where they were the most different. Since the cumulative deviation (RSFE) was negative, the tracking number for the three week data was negative which show that forecasts are too high. If you look at the -4 data column, you can see that this number is significantly lower than the rest of the prior week’s numbers. This dip in sales is what caused the forecasts for the 3 week data to be higher while the forecasts for the 5 week data were lower causing a positive RSFE and therefore a positive tracking signal.
Exponential Smoothing
Figures 3 and 4 used the exponential smoothing forecast method. Comparing with the moving average, the 3 week data with the alpha of .2 most signifies the data that was recorded in Figures 1 and 2. The rule of thumb when switching from moving average is that alpha equal 2/(n+1). When using this rule our alpha would be .14 which is closest to that used in the 3 week data.
The last model was a forecast based on the aggregate demand rather than the demand of each individual sector with an exponential smoothing method with an alpha of .4. I used this method due to the percent error being the lowest on the first