# Storing Projected Values

Options
Member

I have a graph setup giving me daily data. (e.g. sales). I've used as part of the graph "Multi-Period Projection" and set it for 5 days out. What I want to know is how accurate this is to the actual. Today is 1/31/2023 and sales was \$1M, and the prejection for 2/3/2024 is \$1.3M, I want to know how close the projection was to the actual for that day. To do this, I need to know what the project was today and then compare to the actual. How do I store/retain the projecte amount?

• Coach
Options

Long answer: Projected values update in real time as new data becomes available. You could store the projected values in a beastmode or in a webform and join them back to your data, but at what point would you consider them locked in? 1 day out? 3 days out? It sounds like perhaps this is just a curiosity thing in which case I would just jot them down and see how well they perform. If you actually need to measure the variance to the projected values I would recommend projecting your own values using linear regression with scripting tiles (python, R) in magic ETL or in a Jupyter Workspace.

• Coach
Options

What would be functionally equivalent is to retroactively calculate what the projections would have been for your entire data set, so you can see historically how your actuals compare to your forecasts. If you're using linear regression, I played with an approach where I tool @marcel_luthi 's data set-up from this thread on rolling averages:

And create a scatter where the x-axis is pre-existing date, the y-axis is pre-existing value, and the series is the date. Then you can see all the past regressions (and filter for which regression you want):

Disadvantage, is ideally I would be able to connect these points with a line graph, but as far as I can tell, you can't have multiple regression lines on any of the line graph options.

• Coach
Options

Long answer: Projected values update in real time as new data becomes available. You could store the projected values in a beastmode or in a webform and join them back to your data, but at what point would you consider them locked in? 1 day out? 3 days out? It sounds like perhaps this is just a curiosity thing in which case I would just jot them down and see how well they perform. If you actually need to measure the variance to the projected values I would recommend projecting your own values using linear regression with scripting tiles (python, R) in magic ETL or in a Jupyter Workspace.

• Member
Options

I understand what you are stating. Just hoping for some easy way to do this without someone remembering to write them down. I would say we could do it for a week or two and then just do a check every other month for about a week.

• Coach
Options

What would be functionally equivalent is to retroactively calculate what the projections would have been for your entire data set, so you can see historically how your actuals compare to your forecasts. If you're using linear regression, I played with an approach where I tool @marcel_luthi 's data set-up from this thread on rolling averages:

And create a scatter where the x-axis is pre-existing date, the y-axis is pre-existing value, and the series is the date. Then you can see all the past regressions (and filter for which regression you want):

Disadvantage, is ideally I would be able to connect these points with a line graph, but as far as I can tell, you can't have multiple regression lines on any of the line graph options.