May be I´m wrong, but I suppose that Theil statistics need ordered data to compute Y_t-Y_t-1.

But perhaps I´m missing something.

Ah, so your topic is probably more closely related than I thought. I think I see what you mean. Indeed the U_2 measure does not make sense for unordered data. Maybe another reason to add the U_1 measure, which just compares the actual and forecast within each observation. (With a different interpretation, of course - I'm not saying that the two variants mean the same thing.)

And in addition I think I agree that U_2 shouldn't be done in a cross section.

thanks

sven