The meaning of the causality window

This parameter is relevant for programs which estimate the forecast errors for a given model. Suppose one wants to forecast the element nj of a time series. Since one implicitly supposes that this element is unknown, one should not use any information about this element to build the model. Therefore one should exclude all delay vectors containing this element from the model building process. This is realized by the routine exclude_interval.
This exclusion of parts of the vectors from the model building process can be a severe problem if one doesn't do a one-step forecast, but a n-step forecast, with n large. In this case one has to exclude a large part of the data (roughly 2*n) and it could happen that the rest of the data is not sufficient for a reasonable model creation. By means of the -C flag one can therefore shorten this window. Setting the parameter to zero corresponds to a window of a one-step forecast.
Change this parameter only if you really know what you are doing!