Introduction To Making Forecasts With Cross-sectional Data

[@diegogarcia-tu5xm]

Hey, thanks for your video master. Can you help me with the following question? I would appreciate a lot: Which is a good methodology for choosing an out of sample data in a cross sectional model ? For forecasting purpose, of course. In time series data, i do know it is appropriate using mobile window sizes, but i dont know a robust criteria for choosing the sample under a cross sectional context

[@weecology]

Unless there are complexities to the error structures in the data, the nice thing about cross-sectional approaches is that just standard random hold-out/cross-validation is fine

[@diegogarcia-tu5xm]

@@weecology do you refer trying different out-of-sample and random sets? I mean, make calculation of loss function for each random sample and comparing it. Right?

[@weecology]

@@diegogarcia-tu5xm Yes, both leave-one-out (leave out each point one at a time) and k-folds (divide the dataset into k chunks and leave each chunk out one at a time) cross-validation are good ways to go

[@diegogarcia-tu5xm]

Thank you !

[@weecology]

@@diegogarcia-tu5xm You're welcome!