一个time series/stochastic process可以看成是从[0,inf)乘以sigma field到R^n的映衬,such that这个映衬是both borel-measurable and sigma -measurable [当然,这个是cont time的,离散的话,[0,inf)换成N就可以了]。简单的说,一个time series/stochastic process是一个sequence of random variables。当然,random variable by defn是measurable的。
stationarity的含义是,你的这个variable (or some finite subsequence of variable) 是某个data generating mechanism (DGM) (that does not change over time) generate 出来的;也就是说,离开stationarity,你完全没法分析这个time series是哪个random variable generate出来的,i.e.,你完全没法分析这个time series/stoc. process. 这也是为什么做统计的要transform,做计量的要搞cointegration。
如果非要分strong or weak stationarity,以上的stationarity说的是strong stationarity。其实stationarity by defition就应该是strong stationarity。weak stationarity是strong stationarity的first/second order的evidence。所以其实stationarity by defn就是strong stationarity。 |