Forum Discussion

kukszi's avatar
kukszi
Helper I
3 years ago
Solved

Calculating Forecasting Accuracy

Hi there,

 

I am doing a rolling sales forecast every month, and I would like to measure the accuracy of my previous forecasts. What I would like to achieve is the following: currently, we are in the August forecasting cycle, and my July sales figures are actuals. I would like to measure how accurately I estimated the sales when I was doing the forecast for the month of July in the June forecasting cycle.

 

The demo file is saved here.

 

The "Matrix" tab shows that from Product A, in the 2022.06 forecasting cycle I have estimated 160.66 unit sales for the month of July. Now in August I know that the actuals for July were 175.43 units, so I have under-estimated the sales.

 

On the "Accuracy" tab I have a slicer where I can select that the current cycle is August, and I'm analyzing the July sales and comparing it with the forecast I made in June forecasting cycle.

 

I have two measures: "Previous forecast" (looking for the value of 160.66 for Product A) and "Actuals" (looking for the value of 175.43 for Product A). I have tried to play around the REMOVEFILTERS and FILTER functions, but couldn't figure out how to get the right values.

 

What's the best way to get the correct values in these two measures so I can continue the accuracy calculation?

 

Thanks,
kukszi

7 Replies

  • kukszi the number in your example doesn't match with the pbix file, for 2022.06, the total for product A 1493.26, where you are getting all these 160.66 and 175.43 numbers?

    • kukszi's avatar
      kukszi
      Helper I

      Hi parry2k ,

       

      Thanks for looking into my file. You are right, the full-year number is 1,493.26 units in the June forecasting cycle, but what I'm interested in is the July number from the June forecasting cycle (forecast before the period) and the July number from the August forecasting cycle (when it is already actualized). See the highlighted values on the below screenshot:

       

      Thanks,

      kukszi