I am a fan of prediction markets too, after the OP is there more developments on this issue
Narayanan's group at Princeton developed independently a similar concept using with competing arbiters:
http://users.encs.concordia.ca/~clark/papers/2014_weis.pdf In their paper they compare their model with Truthcoin:
Voting as a Keynesian beauty contest. In this approach, each of N shares in a prediction market
confers one vote for market arbitration [7]. In the common case, a supermajority of voters ( k > 2N/3) vote
the same way, this is considered a consensus and the winning shares will be paid out.
Of course, voters holding losing shares have a nancial incentive to vote contrary to reality. To address
this dilemma, all market participants are required to post a bond in addition to the price of the shares
they purchase. Any voters who vote contrary to an outcome with reaches a 2N/3 consensus forfeit their
bond, disincentivizing voters to vote against the likely nal outcome. The market might still fail to reach
consensus if, for example, there are two possible outcomes and all participants holding shares in the losing
outcome form a coalition and refuse to provide any votes necessary for the market to reach consensus. To
disincentivize this, if the market fails to reach consensus after a certain time period then all participants
forfeit their bonds.
The functioning of such a system in practice is unknown. In the case of markets with a genuinely unclear
outcome, the system creates a \Keynesian beauty contest," with all participants incentivized to vote for the
outcome they believe others will consider correct, rather than their own fundamental beliefs.
Ignoring such cases though and assuming a market with a clear outcome, this system produces an iterated
game of chicken between coalitions of voters holding shares in each outcome. If either coalition is able to
convince the other that they are absolutely going to spend their
N/2 votes on their preferred outcome, the other side is incentivized to back down and concede to prevent losing their bond. It is only be repeated play, in which participants have a reputation to maintain that will be damaged if they vote for a patently incorrect
outcome, that this game can be avoided. Thus we think this approach is less desirable as it requires tracking
reputations for all participants in the market and not just a small number of adjudicators.