Evaluating a model by repeatedly training on past data and testing on subsequent future periods, simulating real deployment.