iSixSigma

Measuring Forecast Accuracy

Six Sigma – iSixSigma Forums Old Forums General Measuring Forecast Accuracy

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
    Posts
  • #29776

    Cecille
    Participant

    Help!  I forecast inventory for an airline and after 9/11 have been tasked to evaluate the accuracy of our forecast. I’ve taken the percent deviations, etc….  But I’d like to implement a process where a tracking signal will alert me when my forecast is going out of control.  My data is a monthly forecast versus monthly actual demand.  Any suggestions?  Thanks in advance,  

    0
    #76855

    melvin
    Participant

    Cecille:
    Sounds like you’ve developed a model – be it linear or other, based on past process data.  i.e. – into this model, you enter an input variable(s) and out from the model pops a forecasted inventory.
    If you continue to use the model and, along with the inputs used to feed the model, you keep track of the actual inventory value (what the inventory really was), you can then continually analyze the effectiveness of the model (in Minitab or other statistical software) by comparing the “residuals versus fits”.  All this does is compares the predicted result of your model with the actual results (what really happened) and “flags” any unusual occurrences (usually based on a 95% confidence interval).
    Doing this on a continuous basis can serve as somewhat of a control chart in that “special” or “unusual” occurrences are flagged and signal a possible change in the process or – in your case, a possible move toward your model not “fitting” your process anymore.
    Without knowing more about your exact process, this is the best I can offer.

    0
    #76856

    Mares
    Participant

    Cecille,
    My first thought is to use I-MR control charts: in the I chart you draw monthly forecast: Y(t). In the MR Chart you draw moving averages: Y(t)-Y(t-1).
    The I chart represents month to month variation. Any out of control point should trigger focused on that: special causes, seasonability (not considered before), new relevant incidents.
    The MR chart in an estimation of within month variability. Any out of control point there should trigger actions focused on inventories predictability: causes of deviation, lack of communication, etc.
    Regards… Adrian

    0
    #76862

    Cecille
    Participant

    Hi Bob,  Thanks for the input.  I do use different time series forecast models, depending on the demand.  Approx 50% of my inventory is using a moving average model, approx 20% patterned, etc.  I plan on measuring accuracy separately for each model.  I’d like to have a method for calculating the control limits (including warning limits) instead of using an arbitrary point.  The only thing I’ve done with residuals is taking the absolute value and using it come up with an overall absolute percent deviation but I will try your suggestion.  Thanks again for your input, Cecille. 

    0
    #76863

    Cecille
    Participant

    Hi Adrian, Thanks for your suggestion!  I haven’t used I-MR charts but from your description of it, it sounds extremely promising for what I want to accomplish.  I really appreciate your help, Cecille

    0
Viewing 5 posts - 1 through 5 (of 5 total)

The forum ‘General’ is closed to new topics and replies.