do not generate statistic if data is 'wrong'
CBO works by its own rules which makes decisions using statistics. If statistics are wrong, then it builds wrong plans. Recently I had this problem:
table had a lot of test records with the same column f.k. value. CBO desided not to use the index just because it found out this index has only 7 distinct values for 500.000 records.
So, because w/o statistics CBO uses rule, then it is better to clean statistcsif one is not sure about them: CBO features still are available, like partition pruning, but no messing with 'wrong' data.