Top-coded values

Has anyone ever attempted to "fill-in" the top-coded values for the 5 year ACS PUMS data? For example, in 2009, all of the house values (variable: valp) greater than 4 million dollars in Hawaii get cutoff and it's impossible to tell if the observation you're looking at is a 50 million dollar home or a 4.01 million dollar home - since both look the same. The top-coded cutoff values are different for every year by state combination and I haven't had much luck trying to break through the top-coded value. I've mostly been running linear regressions on the data and am just trying to get a reasonable estimate for each observation that is top-coded. Any insight would be greatly appreciated.

[Updated on 2/24/2015 3:08 PM]