Can we find Goldilocks? Musings on "two-tiered" staking, a native Liquid Staking Token design.

# Can we find Goldilocks? Musings on "two-tiered" staking, a native Liquid Staking Token design. *by [mike](https://twitter.com/mikeneuder)* *sept 11, 2023* $\cdot$ **Contents** 1. **Two-tiered staking** – *general description of the design.* 2. **Properties** – *exploring some of the high-level features of the system.* 3. **Goldilocks rate?** – *a thought experiment on the impact of the relative interest rates.* 4. **Relationship to today's LST ecosystem** – *how trust can be reintroduced.* 5. **Open questions** $\cdot$ ***Related work*** | **title** | **description** | | ---------- | ----------- | | [Liquid Staking Maximalism](https://notes.ethereum.org/bW2PeHdwRWmeYjCgCJJdVA) | Dankrad's proposal | | [Add Liquid Staking Module to the Cosmos Hub](https://forum.cosmos.network/t/signaling-proposal-draft-add-liquid-staking-module-to-the-cosmos-hub/10368) | Cosmos LSM proposal | | [Enshrined LST](https://mirror.xyz/arixon.eth/pE2nU_tSWeiTae2vSJ7a-tNK17kIkE_tqlpDf-neMPo) | Arixon's post | $\cdot$ ***Thanks*** *Many thanks to [Dankrad](https://twitter.com/dankrad) and [Vitalik](https://twitter.com/vitalikbuterin) for recent discussions on this topic.* ---  ### Two-tiered staking This idea was originally [proposed by Dankrad](https://notes.ethereum.org/@dankrad/r1xAbQXm2) in April 2023. The design allows delegation of `ETH` to collateralized node operators, which is similar to [RocketPool](https://rocketpool.net/) and the [Liquid Staking Module](https://forum.cosmos.network/t/signaling-proposal-draft-add-liquid-staking-module-to-the-cosmos-hub/10368) in Cosmos. The idea is summarized below. 1. ***Staked `ETH` is divided into two tiers, node operator bonds $(C_1)$ and delegated stake $(C_2)$.*** The node operator bond is slashable, while the delegated stake is not. 2. ***Each unit of $C_1$ is elligible for delegation of $g$ units $C_2$.*** This enforces the maximum ratio of slashable to non-slashable staked `ETH`. We will use $g=19$ to match Dankrad's post. 3. ***Each unit of staked $C_2$ creates a corresponding unit of a native liquid staking token.*** Each node operator has a different LST that is minted through delegation with them. Only non-slashable `ETH` can mint the LST. 4. ***$C_1$ earns more interest than $C_2$.*** Let $r_1$ and $r_2$ denote these rates respectively. This higher interest rate needs to cover the cost of node operation and the opportunity cost of having illiquid staked `ETH` (it is illiquid from the protocol perspective because it is slashable). 5. ***$C_2$ can be instantaneously re-delegated. $C_1$ must go through the activation and exit queues.*** The security properties of the consensus layer only depend on the rate-limiting of $C_1$. The redelegation of $C_2$ only has minor gas costs. The figure below captures this design. <img src=https://storage.googleapis.com/ethereum-hackmd/upload_31abcdd23713fe908ae81a1143f6dfdd.png width=100%> We have two delegators – `Delegator #1` & `Delegator #2`, and two node operators – `Node Operator A` & `Node Operator B`. Each node operator puts up a `1 ETH` bond to qualify for delegation, and these bonds are rate-limited and slashable. `Delegator #1` stakes `19 ETH` through the protocol to `Node Operator A` and receives `19 aETH` back. `Delegator #2` does the same with `17 ETH` and `Node Operator B` and receives `17 bETH`. The dashed lines represent the interest earned for each of the four participants. The delegators earn interest on the amount of delegated stake they provide at a rate of $r_2$. Node operators earn interest over the sum of their bond and their delegation at a rate of $r_1$. Note that `aETH` and `bETH` \*must be different\* ERC-20s. If they were the same, there would be no capital cost for a node operator to get a flash loan of the native LST and redelegate it to their node operators before repaying the loan. Thus the total supply of each node operator's respective LST is the amount of `ETH` presently delegated to them. ### Properties Let's examine some of the features of this design. ***Capital efficiency*** – Node operators have to supply `1 ETH` of collateral for every `19 ETH` that is delegated to them (a collateralization ratio of `1:19`). Additionally, this means that there are only `19` units of the LST in circulation for every `20 ETH` staked. RocketPool used a much higher collateralization ratio of `1:3`, and it resulted in the deposit pool regularly being full. See https://rocketscan.io/depositpool. In other words, there is never enough collateral from node operators to satisfy the demand for the liquid staking token. <img src=https://storage.googleapis.com/ethereum-hackmd/upload_e66d9ee54cdf3a18ac65a9962ed9004d.png width=50%> The Cosmos [Liquid Staking Module](https://forum.cosmos.network/t/signaling-proposal-draft-add-liquid-staking-module-to-the-cosmos-hub/10368) recommends a `1:250` ratio. There are differing opinions on this value and it would be a critical constant to decide on; a higher collateral requirement gives higher "economic security per unit of issuance" because it ensures that a larger fraction of the staked `ETH` is slashable. The obvious downside is the lower capital efficiency. ***Slashing penalties*** – By separating the delegated stake from the node operator collateral, the delegated stakers are not exposed to any slashing risk. Delegators can also freely move their stake from one node operator to another because their `ETH` is not part of the activation and exit queues. ***"Overpaying" for security*** – By paying issuance to both the node operator and the delegators, we essentially pay the same amount as today but for a reduced level of economic security (only the slashable staked `ETH` provides economic security). However, you could argue that LSTs of today have the same effect – LIDO node operators only have their reputational capital and future potential earnings at stake, so our notion of "economic security" is rather inflated already. The slashable `ETH` associated with LIDO node operators is owned by the `stETH` holders. Classic principle-agent problem. ***Interest rate tuning*** – As Dankrad mentioned in his initial proposal, this is an opinionated design. We would have to choose values for the interest rate of node operator collateral and delegated stake. If the difference between the two is too high, then forming LSTs based just on the node operator collateral portion of the stake could make sense (because of the outsized yield). If the difference is too low, the margins of the node operators are squeezed and we might have a more centralized operator set. We explore this further in the following section. ### Goldilocks rate? The setting of the interest rates $r_1$ and $r_2$ seems to be the key component of this proposal. > **(Observation 1)** $r_2$ is the true "risk-free rate" of Ethereum because it can be earned through the protocol and is not subject to slashing risk. As a result, either > (a) all `ETH` is staked (as $C_1$ or $C_2$), or > (b) the maximum-allowed amount of $C_2$ is staked. Under (1a), the ratio of $C_1$ to $C_2$ will depend on the relative sizes of $r_1$ and $r_2$. Under (1b), where not all `ETH` is staked, we can be sure that each unit of $C_1$ is matched with $g$ units of $C_2$ because $C_2$ earns risk-free interest. With this in mind, the figure below explores a few hypothetical stake distributions. <img src=https://storage.googleapis.com/ethereum-hackmd/upload_67416ebcc544ae3723119870e36ed027.png width=80%> Let's walk through the three cases. ***Case 1: $r_1$ too high.*** If $r_1$ is too high, the rewards earned from being a node operator outstrip the rewards of delegation. Thus, anyone who would delegate through the protocol will instead run a node (or use an out-of-protocol LST layer with trust assumptions – a.k.a. LIDO) to get access to the high interest. This is the situation we are in today because all of the staked `ETH` is slashable and there are large out-of-protocol LST providers. ***Case 2: $r_1$ too low.*** If $r_1$ is too low, there is not enough incentive for people to become node operators. As a result, all the operators will be fully delegated (because of **Obs. 1b**). Assuming $r_1$ is large enough to attract *some* amount of node operator collateral (we need this or the protocol doesn't have any stakers), this is likely a very centralizing force. The margins of node operation are compressed and only the highly efficient, large-scale operators would survive. ***Case 3: $r_1$ just right?*** In this hypothetical, $r_1$ is high enough to incentivize a large number of node operators, but not so attractive as to make it strictly dominate delegation by a significant margin. Based on **Obs. 1a**, all `ETH` must be staked because some of the delegations are not full. Let 25% of the total `ETH` supply be node operator collateral. The remaining 75% of `ETH` is delegated to mint different LSTs and earn interest at the rate of $r_2$. Delegators can freely move their `ETH` around to reward and/or punish node operators and get access to the node operator-specific LST. > **(Observation 2)** Delegators "vote" on a node operator distribution without taking on any risk. These delegations directly benefit the node operators through increased rewards. Thus delegators can allocate to the most value-aligned operators (non-censoring, geographically distributed) and have the freedom to redistribute instantaneously. The downstream effects of per-operator LSTs are not well understood. In an ideal world, the protocol-enforced exchange rate, instant redelegation, and DeFi liquidity on like-asset pools (e.g., Curve) would make the tokens approximately "fungible". Of course, there is always the risk that one LST accrues all the utility and we are in a winner-take-all market structure again. Perhaps a more modest goal is to end up in an oligopolistic market, where a few large LSTs exist and have "economic zones" built around them but no single token has an overwhelming majority. Even if trust is reintroduced (explored in the next section), the delegator's stake might not be fully slashable (e.g., an LST provider guarantees that only 1/10 of the delegator stake will be at risk); this has echos of fractional money rather than pure fiat money. ### Relationship to today's LST ecosystem It's worth comparing this design to the LST world we are in presently. Three critical value propositions of LST providers are: 1. ***Trust*** - The delegators trust their LST node operators not to get their `ETH` slashed. A large correlated slashing event would result in a devaluation of the LST because the LST would become undercollateralized. 2. ***Yield*** – The delegators want to earn interest on their `ETH`, but don't want to solo-stake. 3. ***Token utility*** - The delegators want to have liquidity and may choose to redeploy their LST to a DeFi protocol. > **(Observation 3)** Two-tiered staking eliminates the trust needed when a delegator chooses an LST provider by removing the slashing risk. By removing the trust, the LST providers would be forced to compete on yield and token utility. One obvious way to compete on yield is to reintroduce a trust layer by (i) using delegator `ETH` to fund the node operator bonds and (ii) distributing some of the node operator rewards back to the delegators. The figure below captures this. <img src=https://storage.googleapis.com/ethereum-hackmd/upload_20888b4960bda768c80f647714adddfb.png width=100%> Here, LIDO still serves as a broker of trust between the delegators and the node operators (and curates the node operator set to ensure the fungibility of the `stETH` token). The trust comes from the fact that some portion of the delegated `ETH` is now slashable again because the node operators are not putting up the protocol-enforced node operator bond. If we arrive in *Case 1* described above (only node operator bonds and no delegation), then all the `ETH` delegated to LIDO node operators will be slashable, and we revert to the norm of today. Notice that in the figure, the delegators are earning interest on their `stETH` at a rate of $r_2 + \epsilon$. LIDO can incentivize continual use of their protocol by kicking back some of the node operator rewards to delegators. This would probably be necessary to incentivize the delegators to take on the trust of the LIDO node operator set instead of just earning the $r_2$ rate risk-free by delegating directly through the protocol. ### Open questions Of which there are many... 1. *How interest rate sensitive are consumers of LSTs?* - Delegators will be presented with a choice. They could earn the "risk-free" rate of $r_2$ and delegate their stake to a small staker who they are most value aligned with. Alternatively, they could earn $r_2 + \epsilon$ by trusting an LST provider and increasing their risk. How big could $\epsilon$ be? We can think of this as the "cost of altruism". This also leads to the next question... 3. *Is the utility of LSTs derived from the liquidity they provide?* - LSTs provide interest and liquidity. If delegators get most of their utility from the liquidity of the LST (e.g., being able to use it as collateral money in DeFi), maybe they are less interest rate sensitive. 9. *What are the downstream effects of per node operator LSTs?* - With Curve pools and protocol-enforced exchange rates, the different LSTs should be approximately fungible. Could this result in a more distributed node operator set? If certain LSTs have much higher utility than others, could we at least end up in an oligopolistic world rather than a monopolistic one? 5. *Is there a dynamic way to set the interest rates?* - Ideally, the rates would be algorithmically determined to steer toward the Goldilocks rate. Setting a fixed rate seems quite fragile. Can we target some fixed amount of $C_1$ capital and adjust $r_1$ dynamically (e.g., find $r_1$ such that exactly 25% of the `ETH` supply is slashable node operator collateral)? How does this compare to trying to adjust the issuance schedule today? 6. *Are we OK with reducing the "on-paper" economic security of Ethereum significantly?* - This design makes the cost of running a 51% attack cheaper. Is that a tradeoff we are OK with? How does it compare to the situation today where a 51% attack could be launched by large pool operators who have access to massive amounts of capital, but only have their reputation and future earnings at stake? 7. *Is this even more centralizing the status quo?* - Because each node operator issues a different LST, will there be a very strong centralization pressure to use the same node operator to get access to the highest quality LST? 8. *What are the implications of steering towards a world with 100% of the `ETH` supply staked?* - Is it worse to have 20% of the supply staked, but a majority of the stake controlled by a single entity? Are we confident that the 100% staked `ETH` future would be more decentralized? Lots to think about... :-) <br><br>