Like some people are saying, it can be a decent indicator, but is no way a true measurement. Any skill measurement based on obscure algorithms is never going to be reliable. Other games have been doing it for over a decade with varying degrees of success.
In Battlefield 4, there was a "skill" stat, which was independent of Kill/Death Ratio (KDR) and Score Per Minute (SPM) stats. Between two stats alone can be used to measure how skilled a player is (KDR shows their effectiveness in combat, and SPM shows how much they actually contribute to the objective), but they still had a "skill" stat, that would flail wildly between games, increasingly slowly with good performance but plummeting with any bad performance, and no one knew what the heck it even really meant. Eventually they took it out of the game altogether because everyone agreed that it was useless.
Before that, I remember Halo 3's "TrueSkill" system, which was both better and worse. It was less volatile, but it also measured your skill against your past performances, which meant that sometimes increasing your rating was nearly impossible. Yet, if you were to create a new account after you've improved at the game, you could achieve a higher skill rating much more easily, since there were no past games to weigh it against. It also knocked you down a peg or two for each loss, but victories earned you very little.
GSP reminds me of it, with how wins get you very small increases but losses really crash your rank. I don't think that's a fair way of evaluating things, but I guess it's working to some degree. The people in Elite Smash seem to be genuinely good players, and those in the lower levels seem to belong there. So I don't think Smash's implementation is bad so far, as it does seem to be somewhat accurate.