Syntax
GROUP_HOLT_WINTERS(<number>;<number>;<number><number>;<number>;[<number><number>])
values: series values
α: data smoothing factor (between 0.0 and 1.0)
β: trend smoothing factor (between 0.0 and 1.0)
h: prediction range (how far the future should be predicted) - optional, default = 1
Description
This function computes Holt-Winters double exponential smoothing (non-seasonal) on a time series. Smoothing a time series helps remove random noise and leave the user with a general trend. Whereas in the simple moving average (GROUP_MOVING_AVERAGE) the past observations are weighted equally, exponential smoothing assigns exponentially decreasing weights over time. This gives a stronger weight to more recent values and can lead to better predictions.
This function can't be used referring to the same sheet, so make sure to use a different sheet than the value source sheet.
| Info | ||
|---|---|---|
| ||
How to set alpha (data smoothing) and beta (trend smoothing). Both can be set with values between 0.0 and 1.0) Alpha - Set a larger data smoothing value to reduce a greater amount of noise from the data. Use caution as setting the data smoothing value too high when your data doesn't have much noise can reduce data quality. |
...
In the following example we have price values per month over a two year period. Let's try to predict the first 5 months of the third year using double exponential smoothing.
- Add five additional rows in the date column to better visualize the results for 5 months of future predictions.
- Create a new worksheet in your workbook by duplicating the source sheet.
- Create a group key using GROUPBY(1).
- Sort the timeline using GROUP_SORT_ASC(#Monthly_Sales!Date).
- Click the Fx button on the formula line to display the formula builder (As of Datameer X 7.2, the formula builder is located in the worksheet inspector) and builder and select GROUP_HOLT_WINTERS
. - Use the monthly sales column for the value's argument.
...
After having completed the workbook, use a Line Chart to visualize your new smoothed data predictions compared to your actual data.
...