 You just saw how to implement weighted moving average and exponential smoothing on the time series. Those techniques are useful but limited in that you can only effectively forecast for the next period. When you need an extended forecast or when you have a data set that seems to have a repeating pattern such as our data does, you need a more powerful forecasting technique. One such technique is the Holt Winners or double exponential smoothing approach. Until the most recent version of Excel there was not a built-in tool for making forecasts using Holt Winners. So we will learn the steps of building a Holt Winners forecast using basic Excel functions and formulas. A bit later in this assignment you will see the great new tool in Excel 2016 that makes this technique literally one click. Looking closely at our data you can see there is a bit of a trend upwards as well as the seasonality, the repeating pattern. Inspecting the graph you can see the peaks occur each December which means our pattern repeats on a 12 month cycle. In the Holt Winners technique we break down the sales in our case into three components, a base level, a trend, and the seasonality portion. When we have fully developed our model we will recompose the sales by summing all three components. Holt Winners is called exponential smoothing because we use smoothing factors, one for each of the three components, to determine how much of each we will incorporate in the final forecast model. They are named alpha, beta, and gamma. Let's jump into Excel and do this. Here's our data. If you recall we have sales and we have a number of months. I'm going to scroll down. We have 60 months of data and we want to forecast ahead for the next year. First we need to seed the seasonality factor S. We will do this by taking the value of the first period over here, period one, and subtracting from it the average of the first 12 months because we have a 12 month cycle. I'm going to click in that cell, hit equal, first month's actual sales minus average of those 12 months. Close that and hit enter. That gives us our first estimate of the seasonal portion of our data. Before I copy this down I need to lock this range. Remember the function F4 key and click there and F4 key. That locks the average range so I can drag it down that whole column. I'm going to click the cell, get my little plus, and drag it down and that will give me the seasonal factor for the first 12 months, our estimate. Notice that the first month is $45,000 roughly above the average and we've got one here that's what $46,000 below the average. If you sum these up logically they would total to zero because we're looking at the differences from the averages for those 12 months. Next we will seed the baseline L beginning with the last month of the first year. This is a simple step. We're just going to take the actual value for that month and subtract the seasonal component for that month to give us an estimate of the base level for month 12 of the first year. Finally we're going to seed the trend was zero which is a commonly used approach. Remember these are initial estimates and as we go along the whole winners approach will continue to improve them as we develop it down the column. Now we have our seeds that we can use for year two. We will start with a base level L for month 13 and find it in using one of these formulas that I've got hidden under the graph. Let me move the graph out of the way and we'll see the first formula. The base level L for time t is equal to alpha, our smoothing factor, times the actual at time t which would be that cell minus the seasonality for one period back and our period member is 12 months so we'd go all the way back to month one. And then we would add to it one minus alpha like we did with the exponential smoothing times the level from the prior month plus the trend from prior month which is zero although I erased it accidentally that is equal alpha times and we want to go to the actual for month 13 minus the seasonal value for the prior period which would be that cell plus one minus alpha close that times the prior, oops, prior level plus our trend which is just zero and then I will close that and that gives us our value for the base level from November of 08 which is month 13. That was a bit tedious but the good news is we can just copy these formulas down through the end of our actual data. Now we can begin to get the actual forecast we need. We start by finding the forecaster month 61. It is equal to the preceding month's base plus the preceding month's trend times a number of periods which is just month 61 minus month 60. Again we need to lock down month 60 as we need to use it to drag down the formula. Finally we need to add the seasonal factor which is just the seasonal factor for the preceding period which is November 11. Then we can just drag down there 12 months to get our forecast. I've added the new data points, the 12 months forecast to our chart here in orange. You can see it gives us a pretty good feel that it's going to give a good forecast for each of the 12 months coming up. Even though this seems like a tedious process, the whole winners exponential smoothing, double exponential smoothing really is a good tool to have. The new versions of Excel, Excel 2016, we'll let you do this in one click and we'll show you how to do that shortly. But just remember that any of you have an older version of Excel, you can still do whole winners manually and it's really not that difficult if you follow these steps.