Exponential smoothing is a fancy way of saying “take a little of last period’s forecast and little of this period’s forecast to create a new forecast.” By defining it this way, exponential smoothing allows you to use trend as a factor. The equation looks something like this:
The Greek symbol alpha, or α, is a constant that you determine. It represents the contribution percent that the current period will provide in the forecast. The advantage of exponential smoothing is that you only need two time periods: the last time period and the current time period. With moving averages you have to maintain a lot of numbers so you can compute the average.
When we use VRS, the alpha constant becomes a variable depending on the trend, meaning it will automatically determine what the right alpha value should be.
For retail apparel basics, we have two approaches with exponential smoothing. One is seasonal, the other is non-seasonal. Seasonal will only use similar historical time periods that are the same as the time period we’re trying to forecast. Non-seasonal takes a rolling approach, similar to moving averages.
Let’s again try to forecast our retailer’s February sales for 2016, but this time we’ll use exponential smoothing. In this example, we’ll only take a look at seasonal. Let’s go ahead and assume our alpha constant is 15%, a good place to start.
(1 – 0.15) * 379,762+ 0.15 *373,141= 378,769
Our forecast of 378,769 is pretty close but still a little high. Clearly, 2016 trended down quite a bit and we’re going to need to come up with a better way to detect down trends sooner. Our model, so far, requires us to wait a whole year.
As you can see, both a moving average and exponential smoothing can be used both with pros and cons. The great thing about VRS is that it takes the best of both of these and combines it into one method that is incredibly reliable and best of all, automated. So let’s start building our VRS model.