I've usually seen the reference "out of sample data" used when discussing building or testing trading strategies, especially mechanical ones. When you build, test, and/or optimize a mechanical trading strategy you use a set of data. Then it's usually a good idea to test the system again on out of sample data, that not used to initially build the system. This is sometimes called walk-forward testing as well, although you could use previous data or data from another trading vehicle. The idea being if the system still performs well on the new data there's a better chance the method/system is more robust and not just curve-fitted to one data set. It's certainly not a guarantee, but if your method fails a walk-forward test chances are it will fail in real-time. Hope this helped..