AGP Discussion: ANV Reward Policy

Discussion thread for https://github.com/aragon/AGPs/pull/65/files#diff-10798423ab605414e776dca4684590ad

This is a pre-proposal right now and I hope to get feedback on both the general idea as well as opinions on specific details (amounts, duration, specificity).

3 Likes

Wouldn’t this incentivize random voting? I mean as in participants randomly voting, instead of incurring in the time cost of informing themselves, just to get the ANT reward

1 Like

I support this, as it stands there is not enough incentive to vote if you are a smallholder, this for sure helps align incentives better in addition to the benefit of making the ‘bad voter’ attack more difficult

possibly, I don’t think it is that much of an issue in practice but it would really depend on how much the reward was. Also, this could be minimised if combined with the lock voting proposal

I also have concerns with this, as Luis mentions. I think that the Reward Policy should come along with some kind of locked voting mechanism, like Aaron suggests.

Yes this is definitely the case, its just not clear the magnitude of this effect is relative to voting thoughtfully. I would say that it should definitely change voters behavior (which is good because that’s the goal), but the extent at which voters vote randomly, tend to defer to sentiment, or to the positions of people they respect is hard to guess.

I suspect that most people would prefer to delegate their vote (either automatically or simply by asking someone they trust their opinion before voting) than to vote randomly.

I think this could definitely be interesting in combination with locked voting, but I don’t agree that they should necessarily be coupled.

One way to think about this proposal is that the network is paying for additional security of the voting process by targeting a higher participation rate, making it more difficult to quickly buy/sell after the vote snapshot period and not be exposed to the results of the vote.

In this sense, this proposal is an alternative approach to mitigating this type of behavior, that while complementary to the locked voting proposal, may be sufficient to deter such behavior without implementing something like locked voting.

Also due to the timing, I expect that this will stay in a draft stage and not be submitted for ANV3, its not ready for a full proposal, but it seemed like it would be worthwhile to start the discussion while everyone is paying attention. :slight_smile:

I generally do not support proposals that pay people to vote. What I want is not for people to vote, but to vote well and it’s not clear that this proposal would lead to this outcome. While paying people to vote could certainly increase turnout, and thus make Bad Voter attacks more difficult to mount, I believe it would do so at the expense of good decision-making (and at worst, make the Bad Voter attack doubly profitable since the attacker could get paid to vote badly).

Edit: While I do not generally support proposals that pay people to vote, if there is interest in experimenting as a matter of empirical data gathering then I offer the suggestion to try the experiment somewhere else where the stakes are lower, such as the CFDAO or in another community org. Also, to weed out complete mercenaries, perhaps the reward could be randomized so that there isn’t an expectation that everyone gets paid. And finally, whatever the reward ends up being, it should be much less than $1 million worth of ANT per year, which is a huge amount of money relative to our treasury balance to spend on something so experimental.

It’d definitely be cool to experiment with rewards + locking. Rewards without locking incentivize random voting, but rewards with locking incentivize long-term views + participation

People need to be incentivised to vote full stop. if a high quorum is a metric we want to optimise, then we need to change the incentive structure for this to happen. financial incentives as we know are powerful but blunt instruments, however, it doesn’t mean we shouldn’t use them.

To your point about good decision making, delegative voting should this a non-issue (with good UX). it should be easier to delegate than to random vote. furthermore staking further incentivises good decision making, be that delegating or thoroughly reviewing proposals.

just to throw another idea into the mix. if ANT does transition to an inflationary model, then the payout could come from the inflation, in which case you wouldn’t be paying voters you would be penalising non-voters which IMO is a superior solution as non-voters are free riding from the curation and security and security provided by the voters

I think it’s an oversimplification to suggest that providing an incentive to vote would result in poor decision making in aggregate. While some voters may respond to the incentive by voting casually without fully considering the options, or even go so far as automating their votes to ensure they are eligible to the reward without considering the options at all, its not clear that this would be the dominant strategy.

My expectation is that a portion of ANT holders would respond to an incentive to vote by actively engaging in the governance processes. With delegation on the horizon many may choose to simply engage enough to find someone who they feel aligned with and delegate their vote. In either case we would expect to see increased discussions on issues in the forum and other channels, and a much broader and diverse population of participants in the process. Individually they may not be as qualified to make “good decisions” as a small handful of highly engaged experts, but it doesn’t seem obvious to me that as whole decisions quality would be made worse as a result.

I’m not opposed to experimenting with this in another organization, but would like to clarify how you are quantifying the risk? Lock voting, which you presented as a potential change to the AGP process is also a significant change the incentive structure of voting, is it prudent to also experiment with that somewhere else first? If not, why not?

This seems like it adds a lot of complexity and at a surface level I’m not sure how it would impact the incentives of “complete mercenaries” as the expected value of a lottery should essentially equivalent the expected value of a straightforward reward.

The proposal isn’t to create an ongoing policy that would renew after a year, but specifically to allocate up to 1$ million in rewards over the next 12 months. The policy could and likely should be amended or changed in subsequent ANVs.

The reason I choose 1Million as a place to start the discussion (besides it being a nice round number) is because ANT’s market cap is currently 24M and so 1Million dollars in rewards over the course of a year would be approximately 4%. If we compare this reward mechanism to PoS or DPOS reward mechanism, which are probably the closest analogue, 4% doesn’t seem particularly high or super low.

I’m trying to avoid going into locking discussions in this thread, but I see how it makes sense to talk about both proposals together if locking is being proposed as a solution to minimize negligent/random voting.

My main concern with introducing a locking requirement for participation is that it would increase the cost to participate and exclude users from participating because of how their ANT is being stored–Specifically I’m trying to avoid locking because I think that with Aragon Fundraising, organizations will have ANT held in a bonding curve reserve pool and will not be able to move their ANT into a locking contract. If we implement locking (and locking requires moving ANT into a lock contract) it would be impossible for ANT that is in the bonding curve reserve contract to be used to vote. If Aragon Fundraising is successful, and the organizations that launch on the platform are successful, this could end up locking a significant amount of ANT in these bonding curve reserve contracts.

For a bit of background, right now the default template for org creation would have both a DAI and ANT bonding curve, and the reserve ratio (aka connector weight) for ANT would be 0.01 (some additional context on this would work is here). An organization using this configuration that has a market cap of 1Million, would be holding $10k worth of ANT in their bonding curve reserve. If our goal is for the AGP process to be representative of the community and users of Aragon, figuring out how to make the AGP process work without excluding these users should be a prominent design goal…

If we can figure out a way to do locking that doesn’t exclude organizations that have ANT locked in a bonding curve from participating I think that locking because a much more interesting avenue to explore. One possibility would be to implement the locking functionality in ANT token controller contract, though this might require a token migration.

I’m not sure how complex it would end up being but Instead of locking, we could consider a Vested Rewards structure. Where the rewards you receive for voting are not immediately accessible but instead vested over time (perhaps 4 years with a one year cliff), and revoked or reduced if your balance of ANT decreases during the vesting period. This would provide a similar long-term incentive to hold ANT and be subject to results of votes, but without a strict locking mechanism.

3 Likes

instead of giving the 4% to ANT holders to Aragon projcetcs which increase the demand for ANT token more then 4% a year (perhaps in the first and second year not but longer term (see also aragon manifesto)) then why giving it to ANT holders for voting?

Very good point

Loving this idea!

1 Like

It’s not clear that it wouldn’t be either. And if we can’t be sure that there will be better decision-making if we pay voters, then why do it? Where is the data to show that paying voters results in better decision-making?

Unlike paying voters to vote, locked voting doesn’t introduce an incentive for voters to vote randomly simply for the benefit of getting paid – in fact, quite the opposite, since there is actually an opportunity cost to locking their ANT to vote and they will be financially exposed to the outcome of the vote for at least as long as they are required to lock their ANT for.

When quantifying the risk I’m thinking “if implemented, how many voters will this attract who will just vote randomly or semi-randomly (reading the proposal but not really digging in to fully understand it) vs how much will this increase turnout and make the Bad Voter attack more expensive vs how much better will decision-making be overall”. I don’t have enough data to go on to fill in the blanks in that equation.

I can imagine scenarios where it goes well (higher turnout, happy and informed voters, little/no random voting) and I can also imagine a scenario where it goes wrong (higher turnout but only whales turn out in greater numbers, mercenaries vote randomly just to get the rewards and then dump their ANT back on the market, proposals that shouldn’t pass get approved and/or proposals that should pass get rejected due to random/ low-info voting). Simply due to the incentives I see the random voting outcome as being more likely than not.

Economically rational actors might approach a random reward system differently than they would approach a guaranteed reward system. It depends on the parameters of how each system is set up.

The reason I choose 1Million as a place to start the discussion (besides it being a nice round number) is because ANT’s market cap is currently 24M and so 1Million dollars in rewards over the course of a year would be approximately 4%.

I am comparing the size of the proposed reward relative to the AA treasury since that is where the ANT is coming from, and this would be about 15% of ANT currently held by the treasury. In addition to my general opposition to paying people to vote, I think this is too much ANT to spend in one year on something that doesn’t guarantee or show strong signs of improving the status quo.

In general the mechanism could also create a “rich get richer” effect if it ends up simply attracting more whales and does little to attract the smaller holders of, say, <10,000 ANT, who decide the rewards still don’t make it worth their time to vote, even if they don’t actually have to make an informed decision when they vote. I wouldn’t consider the rich get richer effect a problem if the rich were actually providing some value to the system in exchange for the reward but since they get paid whether they vote well or not I don’t think they are deserving of such a reward.

So in addition to the random voting problem, there’s this potential rich get richer problem, and we’re proposing spending a large chunk of the AA’s ANT reserve despite these problems. All of this adds up to a “no” from me. While I support the right to experiment, I don’t support using the AGP process as the lab for this experiment and I don’t support spending 15% of the AA’s ANT reserves on the experiment.

So the risks are not the same but both do have risk. Lock voting in the current proposal significantly disincentivises voting which its self has its risks associated. the ‘Bad Voter’ attack is a serious problem, but IMO low voter turnout is right up there and more urgent

4879.1452 ANT was used for voting in recent upgrade vote. granted it was not contentious (100% voted yes), but this is kind of the problem. Voting has a cost the time and attention it takes to review and participate is a cost to those who vote. so Why should those providing this valuable curation bare it when the network as a whole is the beneficiary? Why would I (non-active token holder) spend my own resources when someone else is going to do that on my behalf for free?. so maybe a better way of reframing this is

  1. How do we incentivise good decision making
    AND
  2. How do we disincentivise bad/non-decision making, I put these together because they are both have a negative effect

it is for this reason including point 2 above why I am more in favour of inflationary reward

1 Like

(Not to make this thread about locking but…) Locking supports both of these goals: good decision making is incentivized because if voters govern well then ANT will increase in value, and it disincentivizes bad decision making because if voters govern poorly then their locked ANT will go down in value.

Non-decision making to me is neutral, I’m not going to judge someone for choosing not to vote or delegate their vote. The quantity of voters is to me less important than the quality of the decisions made. Increasing turnout at the expense of decision-making quality is not a trade-off I think is worth making. There are reasons to believe paying people to vote will incentivize random voting, whereas I have not seen equally convincing reasons to believe that it won’t, or that there will be a net benefit in any case.

That said, I think locking will also increase turnout among eligible voters: if you have taken the time to lock your ANT specifically for the purposes of being eligible to vote, you will probably also take the time to vote. There is empirical data to support this claim when we look at systems like Decred or Tezos where not every coin holder is eligible to vote, only coin holders who stake or “activate” their tokens are eligible to vote, and their turnout rates are relatively high (33.5% and 87% in the most recent votes, respectively).

I would prefer to experiment with a pure lock mechanism in the AGP process first and measure its effects on turnout and decision-making quality before adding rewards.

While an inflationary reward would resolve my concerns about spending down the AA treasury, it opens up new questions about the value of increasing turnout (if it even has that effect) vs the cost of diluting non-voters and the market effects caused by such inflation. Along these lines, I should point out that an inflationary reward doesn’t solve the rich get richer problem, and it could actually make it worse by further diluting small holders who choose to not vote for whatever reason.

I agree with you that the potential result for voting rewards could either be positive or negative. I’d lean toward the positive outcome being much more likely, but I also agree that there is insufficient data that is directly relevant to our situation to draw a conclusion from. That’s why I would support an experiment like this, precisely because I don’t think it’s reasonable to make conclusions otherwise.

If there is a safer, less risky way to validate the hypothesis let’s do that! Unfortunately I think that it may be challenging to scale this down much without making it difficult to draw conclusion about how the process would impact future AGPs.

The AGP process is itself an experiment. And all changes are going to carry some risks, even choosing to maintain the status quo is risky.

Some of experiments may be more risky than others, and some may be viable to try at smaller scales while others would be more difficult to draw useful conclusions from without implementing at scale.

As an ANT holder and constituent of the community a bounded (by time and expense) experiment feels less risky to me than permanently altering participation/eligibility requirements.

In order to minimize risks of experimentation with important but potentially risky process changes like Locked Voting or Voting Rewards we could consider adopting a bicameral approval process–Where a proposal must pass a vote in which all ANT holders can participate in, and a vote that only ANT holders which opt to lock their ANT can participate in. This would also allow us to implement ANV rewards (for non locked voting) to encourage greater participation and community legitimacy, while minimizing the risk that random or negligent voting could result in bad proposals being passed.

It doesn’t make sense to me that we would compare this to the current holdings of ANT in the Associations treasury. If you want to compare it to the Associations treasury you should use the total value of treasury because if the Association wants acquire more ANT to maintain a specific allocation of ANT they are able to do so. Currently this is is ~45M so 1M over the course of a year would be ~2%.

I think the basis for this argument comes from the same amount of “work” being required regardless of stake, but the reward scaling with stake. Which I think is a valid point, but I would counter that there is value being created here.

Even if all they are doing is delegating, increasing the participation rate increases the security and legitimacy of the voting process by making more difficult for people to swing votes with low amounts of stake.

It also seems reasonable to assume that a whale attracted to vote for the sake of earning rewards is much less likely to vote negligently, precisely because their position will have a greater impact on the result.

It’s interesting that you use these as examples, as in both cases voters are being rewarded. It doesn’t really seem reasonable to claim that it is the locking (alone) that is encouraging them to vote, but rather that they are being rewarded for doing so that results in high participation rates.

Do you expect that introducing locking to the voting process without also introducing rewards would result in greater participation?

Another example of participation in a network is Livepeer, which has a minimal lock period (7 day unbonding) and large inflationary reward. At the moment there are also no slashing conditions for delegators (though this may have changed recently or will change soon), and yet they don’t seem to have had an issue with random voting so far.

I don’t any of these can be directly compared to the AGP process, but I think the livepeer example (due to not having direct slashing conditions for delegators and relatively small lock period) is probably the closest parallel to what the AGP process would look like with ANV rewards + delegation.

I think if we are able to show the value of voting rewards without inflation, it makes the argument for implementing them with inflation much more sound. I see using treasury funds for ANT rewards as a reasonably safe way for us to emulate what inflationary rewards would look like, but without needing to adjust how the token currently works directly.

As far as the value of increasing participation rate through ANT voting rewards (inflationary or treasury-funded) at the expense of diluting value for passive holders… it seems that this could be beneficial for a number of reasons… makes the voting process more secure (particularly as ANT liquidity increases), makes a stronger case for ANT being used as a utility token because it is more valuable when it is actively used, it gradually shifts the distribution of ANT to users who are actively engaged in the community which can increase social cohesion.

2 Likes

I hold a small amount of ANT since 2017. I voted for the first time last week for the upgrade of the governance dApp. And I can tell you, voting has a material cost. You have to pay attention to the relevant dates, make sure your ANT is in an account that will be able to vote, you have to take the time to go to the voting app, cast multiple votes (in this trivial vote I had to vote four times), pay as many tx fees, and repeat the whole process if your ANT is stored in more than one account.

And I’m not even talking about the research you have to do to be able to vote, especially if you’re not very familiar with the recent innovations of this very changing ecosystem. You would need to start by understanding the difference between the Aragon Association, Aragon One and the Aragon Network, for instance. Stuff that you guys take for granted, but that are very confusing for outsiders.

So I think a reward for that effort makes a lot of sense. Even if small ANT holders like me would receive an almost negligible amount, symbolically it would still count as a “thank you” for taking the time and effort to do it.

Also, from a more conceptual perspective, I believe that democracy should be everybody’s part-time job. Instead of having a few individuals with very high salaries who manage the system while the rest of the population sporadically vote without receiving anything in exchange (as in our current nation states), participating in democratic processes should be regarded as the job it is.

6 Likes

I’m imagining something like allocating 10,000 ANT to reward voters that vote on CFDAO proposals for the next year.

A bicameral process is interesting. If we can reduce the number of transactions voters need to sign to participate, so that we can reduce the friction involved in having to vote in two systems, then something like this could be worth trying.

I chose to compare it that way because the AA treats its ANT differently than it treats its other crypto/fiat reserves, specifically in how they distribute it. And if the AA wanted to replace ANT spent on this proposal then that would require spending from their crypto/fiat reserves to do so, which means taking funds that they would normally have spent on grants etc and allocating it to buy ANT instead. But ok fair enough. This proposal costs ~2% of the current treasury value. However you slice it, $1 million is a lot to spend on this.

In Decred and Tezos stakers are rewarded but not for voting on governance proposals, only for making/ signing blocks accepted by the network. These are two different things that stakers in these networks do. They would get paid whether they voted on governance proposals or not, and yet they still have these high turnout rates for voting on governance proposals.

What is the participation rate in Livepeer? (I guess the more accurate question would be “what percent of tokens is bonded or staked?”)

Superficially Livepeer looks similar, but functionally they are quite different. The things that LPT holders “vote” on (I guess just who to bond to?) are quite different than the things that ANT holders vote on. As such the consequences of random voting are different as well. So although they do look the same on the surface I’m not sure we could use Livepeer as an informative case regarding whether or not we should implement rewards in Aragon.

I agree that if we do decide to try this experiment that using treasury ANT to fund it rather than inflation is the better approach.

I’m going to keep getting hung up on the fact that it shifts the distribution of ANT to those who vote, not necessarily those who vote well. And I don’t think that’s what we want to optimize for.

Shareholders in a company don’t get paid to vote either. Their incentive to vote comes from the fact that if they govern poorly then the company could fail and their shares will be worthless. I can think of many similar examples where voters don’t get paid and there are still good outcomes. I cannot think of a single example where voters get paid and it has led to long-term good results.

Given the limited budget of the CFDAO, proposals are unlikely to have any meaningful impact on the price of ANT. In contrast when applied at scale, the proposals can have a meaningful impact on the price of ANT, which changes the payoff matrix of making an educated vote versus a negligent vote significantly.

When there is very little material difference between proposals which should or should not pass (eg CFDAO proposals which are unlikely to impact ANT because they represent such a small budget). The educated voter spends time considering a proposal which is a cost to themselves, and of course the rewards are equivalent whether the voter chooses to spend time considering a proposal or not. The negligent voting will have a higher payoff every time, if we assume that there is no significant cost to passing bad proposals or not passing good proposals.

But if we implement rewards when the outcome of votes actually matters, negligent voting doesn’t makes sense, especially for large holders. The more stake you have, the more likely you are to be decisive if you participate, and if you participate negligently you increase the risk of incurring costs that actually outweigh the reward. The reward encourages people who may have been satisfied to “delegate their vote” through abstinence to actively participate, but it doesn’t change the fact that people are exposed to cost of bad proposals passing.

Implementing this at scale versus in a sandbox like the CFDAO is like the difference between saying pick “option A or option B and I will give you a dollar”, versus “pick option A or B and I’ll give you a dollar, but if your pick incorrectly you lose 10 dollars and if you pick correctly you get an additional 10 dollars”. In the first scenario you would pick randomly, in the second scenario you would pick carefully.

Ideally we would have a reward amount which is significant enough that it encourages lots of people to participate, but not so large that it dwarfs the payoffs of the actual proposal outcomes.

The point I’m making is not that they are being paid to vote, but that they are being paid to be active participants in their respective networks, and so a large amount of token holders have become active participants.

It doesn’t seem too surprising to me that you would have a higher voter participation rate, if you have a higher overall ratio of active to inactive participants in the community/network.

In the case of Lock Voting though, it doesn’t seem like there is any reason to suspect that a locking requirement would increase the ratio of active to inactive ANT holders, or the participation rate in votes. Do you disagree?

You may find the scout dashboard for Livepeer interesting. It looks like they are currently at ~46% participation rate, and are targeting 50%. (they use a dynamic inflation mechanism to target a specific bonded percentage).

Agreed that it’s different. Although with delegation coming to the Voting app soon™, I’m not sure how much different delegating to a transcoder is versus delegating to a community member is if an ANT holder wanted to be minimally involved.

Having a diverse and actively engaged community of voters is something that I think we should be optimizing for–we should be striving for participatory governance to set our values and create alignment and cohesion around our mission, culture, and goals. If we are interested in ensuring that decisions are optimized around some objective metric like the price of ANT or some other measurable goal I would focus on augmenting the voting process with mechanisms that involving betting/predicting on outcomes of proposals (like futarchy), rather than try to restrict voting to only “experts” who will vote “well”.