Multivariate dispersion models generated from Gaussian copula. (English) Zbl 0955.62054
Let \(H\) be an \(m\)-dimensional Gaussian CDF with covariance \(\Gamma\) and standardized margins \(H_i=\Phi\). The CDF
\[
C_H(u_1,\dots,u_m)=H(H^{-1}_1(u_1),\dots,H^{-1}_m(u_m)),\quad u_i\in[0,1],
\]
is called a Gaussian copula. Let \(F_i\) be one-dimensional dispersion model (DM) CDFs, i.e. they have PDFs of the form
\[
f(y_i;\mu_i,\sigma_i^2)=a(y_i,\sigma_i^2)\exp(-d(y_i;\mu_i)/(2\sigma_i^2)),
\]
where \(a\) and \(d\) are some fixed functions, and \(\sigma=(\sigma_i)_{i=1,\dots,m}\), \(\mu=(\mu_i)_{i=1,\dots,m}\) are parameters. Then the distribution \( C_H(F_1(x_1),\dots,F_m(x_m)) \) is called a multivariate dispersion model MDM\((\mu,\sigma,\Gamma)\). (Note that the margins of MDM are one-dimensional DMs \(F_i\)).
The author considers logit, probit, Poisson and Gamma MDM. Some asymptotic formulas for these models are obtained, e.g., asymptotic normality of normalised MDM when the covariance matrix tends to zero. Applications to longitudinal data and simulation results are presented.
The author considers logit, probit, Poisson and Gamma MDM. Some asymptotic formulas for these models are obtained, e.g., asymptotic normality of normalised MDM when the covariance matrix tends to zero. Applications to longitudinal data and simulation results are presented.
Reviewer: R.E.Maiboroda (Kyïv)
MSC:
62H05 | Characterization and structure theory for multivariate probability distributions; copulas |
62F12 | Asymptotic properties of parametric estimators |
62H12 | Estimation in multivariate analysis |
62J12 | Generalized linear models (logistic models) |