+ Reply to Thread
Results 1 to 3 of 3

Thread: "Best" Multiple (Multivariate) Nonlinear Regression?

  1. #1
    Points: 2,640, Level: 31
    Level completed: 27%, Points required for next Level: 110

    Location
    Los Angeles
    Posts
    6
    Thanks
    0
    Thanked 0 Times in 0 Posts

    "Best" Multiple (Multivariate) Nonlinear Regression?




    Hi all,

    I'm just getting started in R. I've had good luck finding the right (predictive) correlation weights using orthogonal polynomials in Java. Where to go from here? Quantile regression was introduced in 1978 and Multivariate Adaptive Regression Splines (MARS) was introduced in 1991. What newer/better methods should I be looking at?

    (If it helps I'm trying to find non-linear relationships between 10-12 predictors and 1 response variable. I'd stick with orthogonal equations except that the curse of dimensionality means I'm using only 1st and 2nd order polynomials so there's a good chance the fit is off.)

    Thanks,

    Axl

  2. #2
    Dark Knight
    Points: 6,762, Level: 54
    Level completed: 6%, Points required for next Level: 188
    vinux's Avatar
    Posts
    2,011
    Thanks
    52
    Thanked 241 Times in 205 Posts
    I haven't tried Quantile regression. So no idea.

    If you are using simple regression
    Using scatter plot and the residual plot , get the idea of the type of non-linear relationship.
    Then sometimes the transformation or introducing the polynomial term we can capture the relationship.

    There are different types of methods are there.. each one based on different different assumptions..
    Eg: Generalized linear model.Generalized additive models or mixed regression. etc
    what is the nature of your response variable?
    In the long run, we're all dead.

  3. #3
    Points: 2,640, Level: 31
    Level completed: 27%, Points required for next Level: 110

    Location
    Los Angeles
    Posts
    6
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Quote Originally Posted by vinux View Post
    There are different types of methods are there.. each one based on different different assumptions..
    Eg: Generalized linear model.Generalized additive models or mixed regression. etc
    what is the nature of your response variable?
    Thanks for the response. Good suggestion to use scatter plot and residual plot. I haven't done much visualization.

    If I'm understanding the question, all variables are continuous. To make things confusing though, I'm looking for overvalued/undervalued signals in stocks/commodities. So +1 for "buy" and -1 for "sell" could be considered either continuous or categorical (+0.95 - +1.00 means buy... -1.00 - -0.95 means sell... otherwise do nothing).

    Will look into GLM (generalized linear model), GAM (generalized additive model), and mixed models aka mixed effects. At first glance GAM looks like a good fit, we'll see...

+ Reply to Thread

           




Similar Threads

  1. Replies: 2
    Last Post: 10-29-2014, 01:33 AM
  2. Replies: 2
    Last Post: 09-14-2010, 04:54 PM
  3. "merge and delete" multiple databases
    By exceeder in forum SAS
    Replies: 4
    Last Post: 03-17-2010, 08:05 PM
  4. What means "1/X weighted multiple regression"
    By alex84 in forum Regression Analysis
    Replies: 1
    Last Post: 12-12-2009, 01:42 PM
  5. "Hepatitis" and "Golf" probabilities.
    By DAI&DF in forum Probability
    Replies: 7
    Last Post: 06-01-2008, 06:51 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