Thank you for puting all these together. This series was exactly what I was looking for. I would highly appreciated it if you could suggest what kind of modeling strategy you would follow if you had say thousands of 'industries' (as opposed to 20 here) in the panel data. Would you still go with one-hot-encoding (dummy encoding) the industies? If not what would be the solution to this. Thanks.