Measuring Forecast Bias at Aggregate Level

edited November 2022 in Intelligence

Hi, I would like to ask if how can we measure bias at the aggregate level. Do I just use the Total Actual Values vs Total Forecast Values to calculate for the Bias%?

If yes, how can it consider possible scenario of offsetting biases at the item-level (e.g., one item has +30% bias, while the other item has -30% bias), if the Aggregate level will show a 0% bias?


Best Answer

  • Vinay VaradarajM
    edited November 2022 Answer ✓

    Hi @afdelarosa,


    Forecasts are said to be more accurate at aggregate levels: This is because Total of actual is compared with Total of Forecast.
    Hence, in my opinion you need to calculate BIAS in the same way as well (total actual vs total forecast). However, you're right in saying that positive & negative bias will nullify each other, and that it will not be a true value overall.

    In this case, I think that it will be helpful to use additional measures which are absolute in nature, such as MAD, MAPE etc. There could be other measures such as RMSE, wMAPE depending on what suits the problem statement best.


    Hope this helps!



  • Hello @vinayvm <a href="">Minesweeper


    Thank you for sharing these useful things with the project!