Sequential testing is common for Time series.

Think SupF kind of tests, or a chow-test for structural break in the parameters over a grid of data points. So, we have many (as many as data points modulo few points from the start and end) statistics, and their correspondent P-values.

FDR is good for the independent case, what I just described yields highly correlated test statistics, so the FDR is too restrictive. That is, a rejection is a clear sign for a break, but no-rejection.. might be because we have many data points (so FDR_pv is small), I can search over more coarse grid to reduce the number of tests, so a max over a grid = {1,2,3,...,T} may not be rejected but a max over a grid = {1,3,5,...,T} may be rejected, which does not make too much sense.

Is anyone familiar, and may be able to give some references for FDR extension to the Time Series domain?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.