It's interesting that you didn't bring up fico scores, or factor in any of their methodology.
Obviously fico scores are far from perfect, but one particularly interesting parameter would be history. An interaction or vote would lose weight as it ages, making it essential for you to maintain a good rating.
This would also make it tougher on people who want to manipulate such a system because creating dozens of accounts to vote for yourself would be very difficult. The fake accounts themselves have no history, making their votes worth nothing.
Credit scores are pretty far removed from the world of online communities -- it's just not practical to credit-check people who just want to use a forum or play an online game or so forth, but yeah, they are relevant from the point of view that they're a widely accepted form of reputation economy, and they've been around long enough to show how a reputation economy could behave, and what sort of social norms you might expect to arise. For instance, congress has decreed that people have a right to find out what their score is, and what data was used to calculate it.
Reputation bankruptcy in particular is an interesting concept. In practice, for most online communities it basically amounts to creating a new account and starting over, but you could have some kind of mechanism for someone to declare bankruptcy (i.e. remove all positive and negative ratings about that user) without having to change their identity.
I do think that negative ratings should expire eventually so that a single bad interaction doesn't have consequences that last forever, and if you expire bad ratings, that means you pretty much have to expire good ratings as well.
My approach to ignoring fake accounts is to treat reputation as fundamentally subjective. In other words, it doesn't make sense to say "Bob has twelve units of reputation" -- rather, you would say something like, "From the point of view of Alice, Bob has twelve units of reputation". (In our system, we actually use multidimensional scores, but I don't want to complicate the example.) The only thing that matters is how Bob is connected to Alice. If there are a thousand dummy accounts that rate Bob positively, it doesn't matter at all unless Alice has given a positive rating of one of those dummy accounts, or one of Alice's friends has, or one of her friend's friends, and so on, and the more hops there are between Alice and the dummy account, the weaker it's contribution to Bob's score will be.
