Using a 'magic value' to represent an unknown value generally does not seem to be a good idea.  Most of us have been in the situation where the user has states categorically that a certain case will never exist only for it to materialise a couple of years down the track as a legitimate case.  If for some reason some particular end user interface cannot handle null values(purists will say null is a state rather than a value), then provide a view which CASTs the null to some value acceptable to that user.
Early drafts of the ISO (ANSI for Americans) SQL standard incorporated the concept of 'multi-valued' null to allow the differentiation of missing values, known unknowns not relevant etc... but the concept was dropped as none of the implementors were interested in providing it.  That proposed facility originated for the earlier ANSI work on multiple types of null