In this video, learn to link formulas to smoothing constants and damping factors.
- [Instructor] Here is a worksheet with the baseline…all setup to optimize the smoothing constant…and thereby minimize the total root mean square error.…Recall from previous lessons…that the smaller the root mean square…the more accurate the forecasts…for the actual observations in the baseline.…The forecasts in column C are made using the smoothing form…of the exponential smoothing formula.…For simplicity, I've set the first forecast for June 2…in cell C4 equal to the first actual observation in cell B3.…
I have also set the starting value…for the smoothing constant or alpha in cell K2 to 0.1.…Notice that the damping factor or one minus alpha in cell K3…is therefore 0.9.…With these settings and of course…with the actual baseline values shown in column B,…we get a root mean square error in cell H4 of 3,112.…We also get a one step ahead forecast in cell C22 of 4,037…because the current smoothing constant of 0.1…was chosen virtually at random.…
We can almost certainly improve…the accuracy of the forecasts by optimizing its value…
- Demonstrate how to evaluate a baseline using a correlogram.
- Identify the drawbacks of using Microsoft Excel’s exponential smoothing tool.
- Explain the different ways you can initialize the first forecast.
- Compare the average raw deviation forecast with the mean absolute deviation forecast method.
- Break down the reasons to use R instead of Excel for exponential smoothing.