+ Reply to Thread
Results 1 to 3 of 3

Thread: Best model time series regression

  1. #1
    Fortran must die
    Points: 58,790, Level: 100
    Level completed: 0%, Points required for next Level: 0
    noetsi's Avatar
    Posts
    6,532
    Thanks
    692
    Thanked 915 Times in 874 Posts

    Best model time series regression




    I am about to start using regression models to predict future trends. I am not sure how one choses the best of various alternative models. I assume you predict (holdout) historical data, but do you use AIC (as ARIMA does), MAPE (as some time series models do) or what?

    I won't be running linear regression since it is time series, probably regression with autoregressive error.
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

  2. #2
    Devorador de queso
    Points: 95,540, Level: 100
    Level completed: 0%, Points required for next Level: 0
    Awards:
    Posting AwardCommunity AwardDiscussion EnderFrequent Poster
    Dason's Avatar
    Location
    Tampa, FL
    Posts
    12,930
    Thanks
    307
    Thanked 2,629 Times in 2,245 Posts

    Re: Best model time series regression

    "ARIMA" doesn't use AIC, time series models don't use MAPE.

    The person running the model chooses that statistic, calculates it, and then interpret it, there is nothing in the model that chooses the statistic you use to assess the model itself. Some statistics are better depending on what your actual goal is (as I've mentioned plenty of times before )

    So identify what your main goal is and choose a criteria that best summarizes how well your model performs the task you want to do.
    I don't have emotions and sometimes that makes me very sad.

  3. #3
    Fortran must die
    Points: 58,790, Level: 100
    Level completed: 0%, Points required for next Level: 0
    noetsi's Avatar
    Posts
    6,532
    Thanks
    692
    Thanked 915 Times in 874 Posts

    Re: Best model time series regression


    What I meant is that these statistics, AIC for ARIMA, MAPE for univariate time series models, are often recommended in the literature to chose the best model among a variety of competing ones. Not that they are used by the software [Hal does not live here sadly] or that they were somehow inherent in the method.

    My goal is to select the best model among competing regression models - the one that predicts future data best . I could do this with a hold out data set (that is determine which model predict the last 12 months which is held out of the data when creating the model) when using MAPE. I have not seen hold out data sets addressed in the context of AIC.

    Of course a major problem for me, one that I suspect can not be avoided, is what predicts the last 12 months may not predict well the period after that as our process is undergoing frequent major changes this year. But I have to do the best I can - given that unavoidable problem.
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats