This is a beta feature. Please email Richard on firstname.lastname@example.org with any feedback or bugs.
The URL for this report is secret but anybody with the URL can access it. Please be careful where you share it.
Loading decomposition details...
Error loading details
This page shows the various parts (known as components) of your forecast
Forecast Forge forecasts usually have three types of component:
Your forecast is the sum of the predictions for each component so you can get an idea of the importance of each by looking at the values on the y-axis of the below charts.
N.B. The values shown on these charts won't look right if your forecast uses a transformation; the charts show the forecast components for after the transformation has been applied but before then inverse-transform has been done.
The first chart shows your forecast along with what the forecast model would predict during the training period.
The trend represents the longer term patterns in your data adjusted for any seasonal variation and for any regressors.
Forecast Forge uses a piecewise linear trend which means the trend is made up of a series of straight line segments. The most recent of these segments is extended into the future to make the forecast. The uncertainty interval allows for the possibility of the trend changing in the future; this is learned from how much the trend has changed in the past.
Annual seasonality represents the patterns that recur every year.
Forecast Forge tries to smooth this pattern out to prevent accidentally mistaking a one off event as part of a repeating pattern. The downside is that this limits how "spiky" a pattern it can fit. If you have an annual pattern where some days are very different from the days either side of them you should use a regressor column for better results.
The weekly seasonality component learns the effect on your metric of each day of the week
This shows the effect of all the holidays in the country holiday database for the country that you chose.
The effect of a holiday is applied only to the day of the holiday itself. If a holiday is important for your business then you will probably see changes in the lead up to the holiday too. For example, if you sell fireworks then the weeks before Independence Day will also be good for business but Forecast Forge will only learn an "Independence Day effect" for the 4th July itself. If the annual seasonality component isn't learning this pattern then you can add a regressor column to help.
Regressors are the extra columns of data you add to help the forecasting algorithm make better predictions.
The chart below shows the total effect of all your regressor columns combined.
You can see the individual impact of each regressor on your forecast in the charts below.
Regressors are numbered from the left of your spreadsheet. So the leftmost column is Regressor 1 and then the one immediately to the right of that is Regressor 2 and so on.