Parameter Uncertainty

Discussion in 'SP9' started by Edwin, Jul 16, 2014.

  1. Edwin

    Edwin Member

    Hi,

    Sweeting on page 260 introduces Parameter Uncertainty but I become confused when he explains how to allow for it.

    First he proposes using a least squares regression to parametrize and then computing a covariance matrix for the parameters. Then using this covariance matrix to simulate multivariate normal parameters 'which themselves are used in stochastic simulations'.

    1) I don't understand what he means by;-'which themselves are used in stochastic simulations'.
    2) If the parameters were wrong, then this information is carried onto the Covariance matrix used in the multivariate normals simulated. I don't see how parameter error is reduced?

    Infact, I feel it may be worsened if the distribution of parameters is 'fat-tailed' although talking about a 'distribution' for such few parameters is wrong...i'm obviously not understanding something here, please help.

    Or is it ok if he meant OLS by least squares regression i.e not GLM since the dependent variable has to normal?
     
    Last edited by a moderator: Jul 16, 2014

Share This Page