#246 U_2 Theil fcast measure: don't calculate in a cross section

[Sven Schreiber]

In the $fcstats accessor (and printout?) gretl shows among other stuff Theil's U2 measure. As pointed out by Juan Carlos Estévez on the mailing list this presupposes a natural ordering of the data, which is not true in a cross section. Gretl should return U2 as NA in that case.

[Allin Cottrell]

I've committed to git a revision whereby fcast produces U2 if we know the data to be time series, otherwise U1. For the fcstats() function there's now an optional third argument to force or prevent production of U2. Doc is updated. This change is provisional since we might want to show both U1 and U2 in the time series case, although that would be slightly backward-incompatible.

[Sven Schreiber]

Am 14.03.2021 um 17:03 schrieb Allin Cottrell:

I've committed to git a revision whereby fcast produces U2 if we know the data to be time series, otherwise U1. For the fcstats() function there's now an optional third argument to force or prevent production of U2. Doc is updated. This change is provisional since we might want to show both U1 and U2 in the time series case, although that would be slightly backward-incompatible.

Thanks, Allin. I'll have to check in detail what "force or prevent"
means. It's true that both can be useful even in time series (the zero
value problem...).

cheers

sven

[Allin Cottrell]

"force": produce U2 even when we don't know the data to be time series. This may be useful when data are passed to fcstats in matrix form (so gretl doesn't know what they are).

"prevent": produce U1 instead of U2 in the case where the data are known to be time series and the default would be to give U2.

[Sven Schreiber]

OK great, closing this.