In classical econometrics, we assume that the true parameters, say, alpha and beta, are fixed. they are points.

also, to evaluate different models, we need that model A is nested in Model B.

Bayesians assume the true parameter follows a distribution. Of course, a point a just a degenerate distribution.

Bayesian model selection is free from nesting constarints.

In order to estimate a model, you must have some prior(s) for the parameter of inetrest.

A lot of numerical methods have a close link to Bayesian methods.

Sometimes, a large fraction of the effort is in the selection of priors aiming to find a analytical posterior, or partly analytical results.

.................

One needs to be very good in Matrix Algebra to do Bayesian methods

[此贴子已经被作者于2005-11-3 20:06:17编辑过]

希望大家多提供这方面的资料哦

谢谢zhaosweden的解答！

可是还是不是太明白，我就是在有关矩阵的估计中看到这种方法，但采用这种方法有什么特殊意义呢，比如有分布的alpha and beta又意味着什么呢？在什么情况下需要采用这种方法呢？

'alpha and beta'

y = a + bx + e where a = alpha  b = beta and e = error

well, in a matrix system or we call it state space model, a regression can be represented as:

y = F*X + v  

F = F_hat + w

------->

F = [a b] and X = [1 x]'

bayesian method can be applied. a and b are two time-varying parameters.

[此贴子已经被作者于2006-11-27 22:22:59编辑过]