I think one of the most critical issues with thinking about ADTs in terms of an algebra is that they lack inverse operations which are otherwise an extremely critical component of how we conceive mathematics. A disjoint union type is formally equivalent to addition, but disunion is never defined, we are given no formal representation of subtraction. The same is given for cognates of division, of logarithms and roots (what those might look like is an interesting, if oft. neglected manner of research)
There's (somewhat) good reasons we don't define these operations, of course. All the same, without being able to represent these inverse operations, the ease of which it maps to familiar notions of the typical algebras we're used to ends up thrown out the window.
The algebra metaphor makes a bit more sense when you move beyond ADTs into more advanced levels of type-theory. Dependent types are a fair bit beyond a mere algebra, instead being more like a complete metalanguage akin to higher-order logic.