Shouldn't he test his model on another dataset besides the one used to generate it? Maybe it's overfitted. It would have been nicer to see the whole thing devel

Shouldn't he test his model on another dataset besides the one used to generate it? Maybe it's overfitted. It would have been nicer to see the whole thing developed with only a decade or two of data, then shown to work for all the other decades too.