It makes no sense to take log of a return. log is only defined for positive numbers, but returns can be negative. You must specify the mean equation in (1) more carefully. Normally it's an ARMA process of return series rather than log price series. Unit root test is also necessary.

选2可能合理一些，取对数有一定的意义。如2楼所说，还应该考虑ARMA（DLM）。

[此贴子已经被作者于2007-3-11 10:45:12编辑过]

faint ... apparently 1) you have no idea what I was talking about; 2) you have no idea what you are talking about

log is meaningless in (2) ... take SSEC daily close for example

S_t = 2881.07, S_{t-1} = 2771.79, S_{t-2} = 3040.60

r_{t-1} = log(2771.79/3040.60) = -0.0926, r_t = log(2881.07/2771.79) = 0.0387

now ... what's meaning of log(r_{t-1}) = log(-0.0926)? ... log of negative number is NOT defined

[此贴子已经被作者于2007-3-11 11:34:19编辑过]

一般对收益率建模，得到收益率的办法是价格指数的自然对数的一阶差分，所以你得上面两种方法似乎都有问题，不符合大多数文献的做法。做东西，先进行文献综述很重要，先看懂别人的东西在研究。