直观判断：依据自相关图与偏自相关图，例如自相关图是递减（或震荡递减）而偏自相关是某阶后突变为零的，大体上就是AR
再者就是AIC准则咯，越小越好


沃尔特 恩德斯（Walter Enders)
应用计量经济学——时间序列分析

看一下时间序列分析吧...

嗯，同2楼的方法。分析ACF和PACF图，AR model的TACF是拖尾的，TPACF是截断的；MA model的TACF是截断的，TPACF是拖尾的。不过因为我们实际分析的是ACF或PACF，所以可能不是绝对的截断，只要尾值很小正负变化我们就可以当它是截断的。另外，要是ACF和PACF都是拖尾的，那就要用到ARMA model或者ARIMA model了，这个的分析更复杂。我也只是略懂皮毛。

For a pure AR(p) model, the ACF decays slowly to zero, and the PACF become zeros for p+1,p+2.........
For a pure MA(q) model, the ACF become zeros for q+1,q+2,......, and the PACF decays slowly to zero.
So you can just take a look at the correlogram of the original series, and decide whether it is a Pure AR of Pure MA.
But this only applies to PURE MA or AR processes. Most likely, you will have a ARMA process, then it doesn′t work.


When your model is good enough, your residuals will be close to white noise, so you can validate your model by looking at the residuals, for example:
1 Q statistics (eviews)  
2 Correlogram: ACF and PACF of the residuals ( should not be significant for the first 10 to 20 periods)
3 DW statistics

or you can use AIC and BIC.

But first of all, make sure the series is stationary, this is very important.

Good luck!!!

谢谢诸位了！ 还有就是如果是几组独立序列的话除了分别处理，有没有统一处理的方法？