In a problem with a predictive aspect, the goal is to say something about
the value of a response variate for one unit. In a typical time series application,
the problem may be to predict the value of the Dow Jones index at the close of the next trading day.
The target population is a single unit, the next trading day. The study population can be selected in many ways. 
In the example, we may decide to use a collection 
of trading days from recent history. Alternatively, we could use trading hours as
the study units with the response variate the value of the index at the end of each hour.
In either case, we need to think carefully about the units and 
what units to include in the study population. Presumably, we will not include units from a long time ago
because this is likely to induce study error. As noted above the choice of study population is an extra-statistical issue.
We need to describe the structure of the study population with an appropriate model.