locked
Questions about time series RRS feed

  • Question

  •  

    Hi!

     

    First question: How many months of data do you need to make this algorithm to work in Excel 2007? And is there an issue about data types in Excel for this algorithm?

     

    I have found some odd behaviours regarding this. If I use the  DM sample Excel 2007 with time series data everything works fine. If I copy and paste data into Excel 2007, from another data source, I can get a forecast of repeating values, that is one value, that will be repeated for each month that I am trying to do a forecast.

     

    Should I avoid having time members for forecast dates in a column? Sometimes my forecast values will be placed below  my dates that do not have values. If I am forecasting months in 2008, with month values from 2007 and 2006 the forecast values will be placed below my 2008 empty months.

     

    Kind Regards

    Thomas Ivarsson

    Wednesday, March 5, 2008 9:39 PM

Answers

  • No, the input for Analyze->Forecast task is not sensitive to Excel data formats. It requires data in an excel table.

     

    If I understand correctly what you're describing, your data for 24 months have some fluctuations, but the predictions tend towards a smooth line. This might be because the model was not able to detect a strong statistical variation in your data which can be modeled as an auto regressive equation. Have you tried setting the "Periodicity of Data" for Analyze task or the PERIODICITY_HINT algorithm parameter in the forecast task and see if that makes a difference?

     

    You should be able to forecast 12 points when your input size is 24.

    Thanks

    Shuvro

     

    Thursday, March 6, 2008 6:38 PM
    Answerer

All replies

  • I assume you're using the Forecasting task in the Analyze ribbon of Excel. When you select the task, you can specify the columns you want to forecast from the underlying excel table. Data for those columns will be forecasted and added to the table. The number of rows added depends on the value of the option "Number of time units to forecast" whose default is 5 in the dialog. In most cases, if the number of steps you're predicting is larger than the original data, the accuracy of prediction will degrade, but with geometric series, you usually can run your predictions to any steps.

     

    Selecting the time column is optional and you can do that if you have a sort able numeric or date column in your excel table. If you leave it empty, the default sorting will be used and a time key will be generated for you.

     

    Please let me know any specific questions you have on this topic.

    Shuvro

     

    Thursday, March 6, 2008 2:45 AM
    Answerer
  • Thanks Shuvro. I am using both the forecast task and the time series task in Excel 2007 but my question is about the forecast task under analyze in Excel 2007.

     

    If I have sales for the last 24 months then I have values enough for making a prediction/forecast for the next coming 12 months? What I can see is that in these 12 prediction months the changes, in the previous 24 months will fade away, during the prediction months.

     

    In the input for the time series is this algorithm sensitive to Excel data formats like thousand separators?

     

    BR

    Thomas Ivarsson

    Thursday, March 6, 2008 6:48 AM
  • No, the input for Analyze->Forecast task is not sensitive to Excel data formats. It requires data in an excel table.

     

    If I understand correctly what you're describing, your data for 24 months have some fluctuations, but the predictions tend towards a smooth line. This might be because the model was not able to detect a strong statistical variation in your data which can be modeled as an auto regressive equation. Have you tried setting the "Periodicity of Data" for Analyze task or the PERIODICITY_HINT algorithm parameter in the forecast task and see if that makes a difference?

     

    You should be able to forecast 12 points when your input size is 24.

    Thanks

    Shuvro

     

    Thursday, March 6, 2008 6:38 PM
    Answerer
  • The data I have tried this on has strong seasonal variations and from what I understand the algorithm will adjust for that.

     

    Thanks for the help!

     

    Regards

    Thomas Ivarsson

    Thursday, March 6, 2008 7:00 PM