1

I call "crop test" or whether my model passed the "crop test" when I remove data from my dataset, conveniently before some events in the data to check whether the historical predictions match the latest predictions in the cropped version of the data.

This way I check that there isn't a lookahead bias or, to put it differently, that future data isn't leaking into the model in any shape or form.

SkyWalker
  • 825
  • 1
  • 7
  • 12

1 Answers1

1

In time series forecasting, this is called "using a holdout sample", or simply "working with test data" (in contrast to the "training data" you use to fit your model). The point in time where the training data ends and the test data begins is the "forecast origin".

And of course, it's a good idea, since in-sample fit is a notoriously poor guide to out-of-sample performance.

Stephan Kolassa
  • 95,027
  • 13
  • 197
  • 357