One thing that's always puzzled me about Blizzard's Arena system is that Blizzard resets ratings at the beginning of every season. Most other rating systems go out of their way to avoid situations like that.
The concept of ratings is that a person has a "true skill level". Ideally, her rating accurately reflects her true skill. And as her true skill increases and decreases, so does the rating. The problem is that when a person first enters the rating system, her rating does not match her true skill, but through winning and losing many games, the two will eventually match.
As such, most rating systems treat that entry point a little differently. In chess, people are often given a "provisional" rating, which lasts for about 25 matches or so. While they have a provisional rating, rating changes are calculated differently. Microsoft's Trueskill system actually has a second value paired with the rating, which is a measure of how "confident" the system is that the rating actually reflects a player's true skill. As you play more and more games, the system becomes more and more confident that your rating is correct.
Blizzard's rating system is very unusual in that it returns everyone to that low confidence state every so often. The usual reason given is that it gives everyone a fresh start. But in reality, your true skill level doesn't change that much that quickly. All the reset does is cause people to end up playing matches against teams of wildly differing skill level. A team that should be rated 2000 is now rated 1500, and is going to steamroll most teams it will encounter at the start of the season. This happens until the ratings shake out and people are restored to their true skill levels.
I think there's a different reason Blizzard resets the ratings. The Arena rating system is meant to be a zero-sum system. If my team gains 20 points, your team loses 20 points. However, in its current incarnation, the Arena system is vulnerable to rating inflation. What happens is a low-ranked team (say 1200 rating), gets tired, dissolves, and reforms as a new team. The new team enters at 1500 points, meaning that 300 points are added into the system. They will probably eventually fall back to 1200 and the process may begin anew.
By resetting the ratings, Blizzard clears out the excess ratings added into the system and restores the system to its zero-sum balance. Unfortunately it has the side effect of ensuring that ratings don't match the true skill of the teams for a few weeks after the reset. And it causes heavy load and long queue times on the servers as the higher ranked teams seek to restore their correct rating.
It's sort of amusing, but chess actually has the same problem, only in the opposite direction. I have a friend who is heavily involved with the Chess Federation of Canada. According to him, one of the main problems with their rating system is the existence of chess schools or camps for youth. What happens is that during the summer, the kids play constantly against each other and end up being pretty good because of the practice and training (not Grandmaster-good, but better than average).
Then at the end of the summer, they will play in a couple of rated tournaments. Because their entry rating is lower than their true skill level, they end up taking a lot of rating points from the other people. However, after that summer they stop playing tournament-level chess, taking those points with them, and the chess rating system suffers from point deflation.
To combat this, the equations used by the CFC that govern rating changes have a very small bonus term, which increases the amount of rating in the system, hopefully restoring the balance and keeping the amount of rating in the system constant.
It's an amusing parallel to the situation faced because Blizzard did team-based ratings and allowed teams to be dissolved and rebuilt. A personal rating system, such as is being introduced with Season 3, tends to be more robust because it cannot be reset easily, and the system is not quite as vulnerable to inflation.