I have been reading "Doing Data Science" by O'Neil and Schutt (and, by the way, I think it is very good). One section describes a process for evaluating the predictive power of a model by replaying a series of past events, using all of the data prior to time t to predict what will happen at time t (for all t).
This is such a fine idea that it seems like it must have a name. The book refers to it as a "causal model" because it obeys the notion that causality can only go forward in time. Is this a commonly-used term for this idea? A Google search reveals that "causal modeling" is more often used for models that are intended to show causality (beyond just correlation) so that's a different thing entirely.
[link][1 comment]