sunnytamu

For 2 asynchronous high frequency time series of prices, e.g., S&P and NASDAQ, please design a method to check if one is leading another? If in fact one leads the other, how to determine which one and when?

谢谢

abiao

what do u mean leading? there is some way to check dependence, for instance, cointegration, Copula, etc. I dont know what leading means.
when leading? you ll be millionaire if you know that.

luqi

the simplest way is that you can use correlation r, if found r is near to 1 or  
-1, it means two time series are averagely strongly related. then use the method as following, divide the time series into sub-series for both, then check
r(t,delt_t)=[y(t)-E(y(t))][x(t-delt_t)-E(x(t-delt_t))], r is a function of t and delt_t , E is expected value.
last step choose the maximum value of r, the corresponding t is the time you want and delt_t is the time interval it happens.

vanney

Maybe you can take advantage of the VAR model or Granger Causality Test to examine the the lead-lag relation between  returns of S&P 500 and Nasdaq.  For detail, you 'd better refer the book "time series analysis" written by Halmilton (1994).

deanwj

格兰杰因果关系检验可以做，不过还是用到VAR，向量自回归模型

longarrow_21

引用:

原帖由 vanney 于 2008-8-23 10:52 AM 发表
Maybe you can take advantage of the VAR model or Granger Causality Test to examine the the lead-lag relation between  returns of S&P 500 and Nasdaq.  For detail, you 'd better refer the book "time ser ...

Yes, that's the normal way you can follow.

renjizhao

这个正是我这学期projet的题目。
具体还没开始做呢，大概会从cointegration入手，还有种方法是用小波分析，还一点都不懂。
欢迎继续讨论啊。