Prior to 2005, you will want to use HUHHNUM, in place of HRHHID2, to link ASEC samples. Unfortunately, this is not guaranteed to give you a unique match across samples, as the ASEC includes an oversample of households from adjacent months. IPUMS-CPS is currently working on a variable called MARBASECID to facilitate linking across ASEC samples (see this previous answer). In the meantime, you can use person-level characteristics such as SEX, AGE, and RACE to help improve your match rate.
Hope this helps.