Treatment of conflictive forecasts: efficient use of non-sample information