Allow for ids to be defined with multiple variables

[eloualiche]

More of a wish than a necessity.
Sometimes my id is composed of two variables to uniquely identify one observation for one date.

One such example would be in finance stockholding dataset: individual i holds N shares of stock j at time t.
How many were they holding at time t-1?

ideally

lag!(df, [:stock_id, :investor_id], :date, :portfolio_holding)

Since my understanding is that this package uses PanelShift.jl, this could be built in.
Their version of a shift is (their readme)

function panellag!(df, id, t, x, newx, n=oneunit(df[1, t] - df[1, t]); checksorted=true)
    return transform!(groupby(df, id), [t, x] => ((t, x) -> tlag(t, x, n; checksorted=checksorted)) => newx)
end

So the groupby argument can be either one or an array of ids.

Thank you!

[eirikbrandsaas]

Thank you for the suggestion, yes that should be easy to add in. For now, as I bet you know, you could just create the new panel ID first, so it is unique based on the two variables you have, and then use the package as is.

That said, I really need to revisit this package. It is very slow for even moderate sizes and doesn't work for all time formats. I was really hoping that by now somebody had written a better package, but I think PanelShift.jl is still the closest, and this really is just a convenience wrapper around that package.

[eloualiche]

Thank you for the update.
The package is good. As long as it is fairly stable, I have not had major issues with it. To be honest I do switch back and forth with PanelShift depending on what I need (and what I remember).

Thanks for the great work!