Additional insights from web3 reward emission research

We are currently undertaking a large-scale research effort to create better frameworks for token mechanics especially for web3 protocols in the hardware resource provisioning sector.

We have recently published part one, where we have compared emission schedules of various web3 networks, including of course The Graph. You can find the report here.

We wanted to share some additional data with the Graph community though, that is not in the report.

First off, we have classified The Graph’s token rewards to be in the category of fixed and constant emissions. For those the token rewards follow a constant rate (3% per year) emitted over time.

Despite its simplicity this approach is quite uncommon, at least there was no other network following this approach amongst the 16 protocols of Web3 infrastructure networks we analyzed. Compared to all of those the relative token reward emissions are the lowest and interestingly, on a level (0.2% per month) that other emission schedules also reach at later stages as shown in the following chart. It shows the average (blue line) and the interquartile range (blue shaded) of the monthly token rewards of all Web3 infrastructure networks together with the The Graph’s emissions highlighted:

Next we looked into the dollar value of those emissions, to compare how nodes are actually getting compensated - we uploaded that image into the repo linked at the end (as I am a new user).

Despite constant in rate and hence even increasing in absolute GRT amounts, The Graph’s monthly reward emissions in dollar-terms decreased in line with the price decline when the last bull market ended.
More relevant might be the comparison of token reward emissions to networks providing similar services, which is Covalent in our selected list. You find those charts in the repo as well.

Covalent is still in the process of ramping up (and detailing out) their rewards program for nodes and stakers since the start in May 2022. Covalent plans a maximum of 2% of total token supply for rewards per year (translating to 0.17% per month), hence quite similar to The Graph.

The dollar-amounts per node of those rewards are also added as chart, but here one needs to keep the caveats in mind that Covalent numbers include rewards for delegators and the fact that yet it is only Block-specimen production (see network-block specimen producer) that Covalent nodes do and get rewarded for

More data and information as well as the paper with further details on the published report are available in this repo. We hope this information yields some insightful information for The Graph community and are open to your feedback. What would be some additional data you would want to see?

1 Like

Thanks for sharing. Context is always helpful.

If you were looking to be inspired for a follow-on study, it would be interesting to also classify the mechanisms by their security and game theoretic properties. For example, are the issuance mechanisms sybil resistant/false-name proof:

And equally important, how efficient is the issuance to the health of the network:

Just a thought. Thanks for the perspective

wow - thanks a lot for those links. I didn’t come across those yet. And yes, security and game theoretic properties are for sure important aspects - have you researched in that area?

Yes, but nothing written up yet. The “Cost of Sybils” paper puts significant restrictions on how efficiently one can distribute emissions beyond stake. This is important because if stake is not perfectly correlated with platform “efficiency” (i.e. the ones with the most stake bring the most efficiency to the platform), stake based emissions inject inefficiencies.

curious to learn more in case you plan to write about it. On the tradeoff between efficiency and sybil-prevention: Have you taken a look into NYMs reward sharing scheme? I like their approach to have a parameter (alpha) that sets a lower bound for the cost of sybil and hence to allow for a spectrum (in combination with the other aspects e.g. reputation, cost declaration), but would like to get your take given you’re way deeper into that topic

Thanks for sharing! Let me take a look and get back to you.

Just on a quick scan, I like that the paper considers incentive compatibility of reporting c. Some actually skip yhsy important detail. And I agree with you regarding the alpha parameter. Since we have tons of results on how sybil resistance is hard/introduces a ton of inefficiencies, it is nice that they introduce a mechanism that is “approximately” sybil resistant. Two things that are not clear (though it may be clear, I was just reading too fast) are:

  1. There is no cost of capital in the model. In capital budgeting models (which I think staking is), there is usually an external cost of capital. I think here the assumption is that there is a fixed supply and thus the only decision is based on allocation proportions of stake. However, if an agent can change the total supply (at a cost or rate), it is not clear that the results hold. They might, I just don’t know.

  2. It is not clear whether an agent can influence M_i. In spirit, M_i is demand, but maybe a node can also increase demand on the other side thus manipulate M_i. I wonder if the truthfulness results still hold in that case. It would depend on the specifics of how demand is routed (which I don’t know in enough detail), but it’s another possible attack.

This of course assumes everything is technically correct, which I have no reason to believe otherwise, just didn’t have time to check.

Thanks for sharing!