How to Design Better On-Chain Governance

The topic of on-chain governance has always been contentious. While off-chain governance is generally perceived as clunky, on-chain governance has allowed developers to build increasingly complex protocols allowing users to sway a network’s direction. But these are all essentially games which, if misconfigured or providing the wrong incentives, can steer the chain towards disaster.

In “What is Futarchy? — Trading the Future,” Freiderike Ernst, the cofounder of Gnosis, highlights the standard methodologies of on-chain voting. As the “one vote per person” paradigm is vulnerable to Sybil attacks on permissionless networks (one person can split their capital over multiple accounts and cast multiple votes), a user’s voting power is usually weighted by the token amount they hold. Lotteries and token curated registries use the same method to avoid Sybils.

Robin Hanson proposes a new governance model called futarchy, in which decisions are made based not on votes, but the results of prediction markets on the organization’s welfare measure, which is an indicator of the network’s growth or demise. Participants of the market will bet on the future value of the welfare measure.

Betting is usually implemented using outcome tokens, each of which represents one particular outcome of the market and whose monetary value is determined by the eventual welfare measure. Good predictions are rewarded and bad predictions result in losses.

Using outcome tokens, participants can even bet on the value of the welfare measure contingent on the implementation of the policy. For example, a participant can make a bet that pays a profit if the policy is implemented and the welfare measure increases by a certain amount, but is voided if the policy is not implemented.

For a publicly-traded company which chooses the stock price as their welfare measure and is considering firing their CEO, the upshot is that the organization obtains two predictions, the future stock price if the CEO is let go and the future stock price if the CEO is retained. As you can see from the chart below:

With futarchy, the decision that results in the highest possible welfare measure is implemented. As the final stock price prediction contingent on the CEO getting the ax is higher than the prediction contingent on the CEO being retained, the CEO is removed from the company. This takes all the emotion out of the decision process and allows the organization to make rational decisions based on what’s commonly referred to as the “wisdom of the crowd” to improve their values.

Implementing a market-maker to facilitate trades between participants poses some challenges. If we want to use futarchy to evaluate more complex contingencies, markets quickly rise to tens of hundreds of tokens. Here, the “thin market problem” rears its head: There are not enough participants to properly correct the probabilities of this many outcomes. The natural solution is an automated market maker (AMM).

A straightforward solution is the cost function implementation of the logarithmic market scoring rule. Unfortunately, this implementation does not allow ad-hoc changes of the liquidity, usually resulting in a market that is either too shallow to accommodate all participants or too deep to actually produce meaningful results. The liquidity-sensitive logarithmic market scoring rule (LS-LMSR) mitigates this problem, but the solution introduces new defects, the most severe of which is an arbitrage vulnerability which occurs in all scoring rule market makers except LMSR.

The crypto mainstay constant function market makers (CFMM) such as Balancer handle the liquidity aspect better by allowing LPs to dynamically deposit and withdraw liquidity, and are more familiar to crypto natives, but suffer from the same issue as LS-LMSR. However, it turns out that during their prediction market days, Gnosis appears to have found a CFMM implementation of the LMSR which combines the best of both worlds.