+ Reply to Thread
Results 1 to 3 of 3

Thread: Combining regression with time series

  1. #1
    Points: 1,164, Level: 18
    Level completed: 64%, Points required for next Level: 36

    Posts
    5
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Combining regression with time series




    Using an existing time series of supermarket sales data I need to make a forecast. Basically a simple exponential smoothing model explains future values fairly well. But besides being dependent on the past values there is also a strong dependency on an external variable, the amount of spend on newspaper advertising, and I would like to include that in the model.

    Ideally I would like to use a least-squares regression with two regressors, one for the time series component (exponential smoothing, AR, MA, or any other suitable method) and one for ad spending.

    So far I have two ideas but both have flaws:

    1) An AR(1) process plus the ad spending variable would be easy to model with least squares but AR(1) uses only the previous day's information and I already know that more past terms should be included, such as in exponential smoothing.

    2) I could first apply exponential smoothing to get a smoothed time series and then do a least squares regression on the ad spending with it. But then the two effects would be calculated separately and I am afraid of killing some of the ad spending effects if I smooth the time series before it.

    Any ideas for a simple model with a predictor for the time series component and one for the ad spending?

  2. #2
    Points: 224, Level: 4
    Level completed: 48%, Points required for next Level: 26

    Posts
    1
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Combining regression with time series

    Consider using two regressors: exp-smoothed series with lag 1 and the external variable.

  3. #3
    Points: 651, Level: 13
    Level completed: 2%, Points required for next Level: 49

    Location
    Groningen, Netherlands
    Posts
    43
    Thanks
    2
    Thanked 2 Times in 2 Posts

    Re: Combining regression with time series


    I don't know how it works yet, I will figure it out myself (for a real world situation) in a couple of weeks but; you know how to estimate an Arima model i guess. You can add extra independent variables by using so called ARMAX models, (X standing for the independent added variables). I guess, if you know how to estimate and interpret Arima models you could also do the same for ARMAX model. Spss has a function where you can add these extra variables.

+ Reply to Thread

           




Similar Threads

  1. which regression model for time series?
    By Sylviastat in forum Regression Analysis
    Replies: 17
    Last Post: 03-02-2011, 10:59 AM
  2. Replies: 1
    Last Post: 01-20-2010, 04:59 PM
  3. Regression/Time Series
    By iknowu99 in forum Regression Analysis
    Replies: 11
    Last Post: 07-30-2009, 03:28 PM
  4. specifying (time series) regression modell
    By panta_rhei in forum Statistics
    Replies: 0
    Last Post: 10-28-2007, 03:53 PM
  5. time series regression
    By zingbergeur in forum Psychology Statistics
    Replies: 1
    Last Post: 05-04-2006, 07:07 PM

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