The proposed measure is highly correlated with U-3, so as time-series they should basically tell the same story. If the assertion is "U-3 doesn't predict this phenomenon but this other measure does" it's likely to be wrong since the signals are roughly equal to a constant factor. For the entire data range depicted in the paper this property holds. Is it possible that back in $GOOD_OLD_DAYS this isn't true? Well I'd like to see the data but I don't have time to chase it down and none has been offered to support that claim.