Whilst this may be a fundamental and basic question, I need to pin down a solid understanding of exactly what in-sample forecasting is and entails. I understand out-of-sample as taking a period of training data and forecasting future values based on the regression of data from that period. However, in-sample seems to be seldom defined without reference to not being out-of-sample.
Take the example of a time series from 1900-2000 for which I have a regression specification. I want to make in-sample forecasts for 1990-2000. What does this entail?