Alex North@Alex North and @Nicola.
Background and problems
- Some SPs have difficulty borrowing the pledge collateral required to commit new sectors, especially FIL+, and probably more-so for longer commitments after SDM. Demand for borrowed tokens exceeds supply, and this constrains onboarding.
- The network is experiencing inflation as the supply inflows of block rewards, SAFT vesting and pledge unlock exceed the outflows of pledge locking and gas burn. (This compounds the capital access problem, as fewer parties want to own and lock FIL while inflationary, and also pledge requirements increase with inflation.)
- We hope that off- and on-chain FIL leasing solutions will develop to close the gap. These will take time to mature and grow, but we don’t want to crowd out such developments.
Proposal
The basic mechanism is:
- An SP is permitted to onboard or extend the commitment to sectors with less than the nominal pledge collateral requirement. The maximum allowable under-collateralisation is the shortfall fraction, (calculated for the SP as a whole, not per-sector). E.g. if the maximum shortfall fraction is 30%, the SP must provide 70% of the sector’s initial pledge. The SP is constrained so their total collateral shortfall across all sectors cannot exceed the shortfall fraction. The maximum shortfall fraction is calculated to be that amount that could be repaid in full by garnishing expected block rewards (vested, and after fees) within a target term (e.g. 1y, the minimum sector expiration).
- An SP with a pledge shortfall continuously accrues a shortfall fee for maintaining this shortfall over time, e.g. 20% p.a. This fee is paid by burning tokens from the block rewards earned by the SP (from the part vests immediately). The shortfall fee rate could be a constant, or a dynamic value that scales with each SP’s shortfall. The maximum shortfall fee is bounded by the expected block rewards to be earned and immediately vested.
- An SP with any shortfall has some fraction of their vested rewards automatically redirected into pledge to reduce the shortfall. The repayment rate is at least sufficient to repay the entire pledge shortfall within the target term. This mandatory repayment is the key to this proposal’s positive impact on inflation.
When a sector expires, only the fraction of target pledge that has been satisfied (initially or from repayments) is released to the SP, i.e. actual pledge is released in proportion to the expired sector’s share of the total pledge requirement. The SP’s shortfall fraction remains constant, and the SP could then onboard one new sector with that released pledge while maintaining that shortfall fraction (if the pledge requirement has remained constant).
Summary of intuitions
Problem (1) of difficulty sourcing pledge tokens is addressed by reducing collateral requirements. SPs will be able to commit more storage & deals with the tokens available to them, but receive a lower net reward after paying fees, compared to SPs that deposit the full pledge.
Problem (2) of inflation is addressed by both the shortfall fees and, much more significantly, by the automatic redirection of vested rewards into pledge.
Problem (3) of crowding out is mitigated by requiring each marginal sector to still provide a large portion of its required pledge, and by setting generally high shortfall fees. Demand for leased tokens will remain high, and leasing externally will usually be more economical than taking a shortfall (but doing both at once is likely).
It is rational for an individual SP onboarding storage to take on a pledge shortfall because they can then get more power for a fixed amount of pledge. However, any increased share of rewards may be competed away by other SPs doing the same thing. The rationality of maintaining a shortfall (when pledge funds are limited) induces a large fraction of block rewards to be re-committed to pledge before they vest, so never enter the circulating supply.
Summary of impacts
To give a guide to the scale of impact, rough analysis suggests that given Feb 2023 network conditions, a high bound is:
- A maximum shortfall of ~35% could let SPs onboard 53% more QAP for the tokens available to them (18EiB → 28EiB, over time). This shortfall is repayable with a max repayment rate of 75% of gross rewards (i.e. all vesting rewards) over one year.
- A shortfall fee of 33% p.a. could result in burning up to 65K FIL/day, at max shortfall (up from ~2K/day gas burn recently). This is burn rate of 5.6% p.a. of circulating supply (up from 0.17%). SPs would still have a 53% p.a. return on pledge requirement (down from 71%).
- A maximum repayment rate of 75% of total rewards could remove up to 196K tokens/day from net emissions, if all SPs took a large shortfall. The remaining inflation after fees and repayments would be 216K FIL/day, or 18% p.a. (down from 478K/day and 39% p.a.). I.e. basically halve inflation.
Design details
Shortfall term and amount
A network parameter of the expected shortfall repayment time is set, say to the minimum sector commitment duration (1y after SDM). This expected duration, plus the repayment rate (below) determines the maximum collateral shortfall amount for any SP: no higher than the amount of expected block rewards to be earned and available to redirect into pledge over the life of the sector. The intuition here is that every sector should reach full collateralisation before it expires.
This maximum shortfall fraction is likely in the range 25-35% today (depending on maximum repayment rate chosen). This value will decrease over time as the block reward decays. This leads to a natural phasing out of the collateral shortfall mechanism in the long term, as the max shortfall shrinks toward zero.
Upon introduction, we may need to ramp up the maximum shortfall slowly to avoid incentives to terminate and re-onboard sectors.
Repayment rate & shortfall fee
We take as a design principal that an SP’s total shortfall must be repayable from expected rewards, after fees, within the target term of 1 year. This means an upper bound on the repayments + fees is an SP’s total rate of reward.
An SP will presumably need some free token flow in order to pay operational expenses, but this amount will vary between SPs. A good mechanism will let a growing SP take on a shortfall up to their ability to maintain operations.
If we take an upper limit of 100% of expected rewards being directed toward either repayments or fees when an SP takes on the maximum shortfall, then we see network parameters of the maximum fraction of rewards directed toward repayments or fees (adding up to ≤ 1.0). At lower shortfall levels, a lesser fraction of rewards are directed to both repayments and fees, and the remainder is free for the SP’s opex.
These parameters turn out to directly imply a maximum shortfall fee. Or, said another way, a target fee rate determines the split between repayments and fees. Today 75% of an SPs rewards are vested for six months, and 25% released immediately. If we take that split as the repayment/fee split, the implied fee rate is 33.3% (i.e. the ratio). The fee rate may be increased by burning a greater part of their reward (but note this would also reduce the max shortfall, since fewer tokens are available for repayments within the fixed term).
It is also necessary to set a minimum repayment rate in order to have shortfalls actually paid back. Otherwise, as the shortfall is paid back, the repayment rate will shrink too like Xeno’s paradox. E.g. we could set at least 25% of block rewards are used to repay any level of shortfall.
An SP can take as much additional shortfall as they wish to reach breakeven on their total operational return. If they do, all block rewards they earn are either burnt as fees, redirected into pledge, or necessarily spent on operations.
Dynamic fees (optional)
Whatever fixed fee is set, there remains a possibility that external interest rates will rise above it, leading to substitution of a shortfall for leasing. Similarly, the fee could be set too high above sector net returns (i.e. after costs) to be useful, even though tokens for leasing are scarce. As an optional extension, the shortfall fee could change dynamically over time in response to utilisation of the shortfall. If allowable shortfalls are nearly fully utilised, then the fee may be lower than leasing rates (or is at least well clear of minimum net returns), and could be raised. Conversely, if there is ~no utilisation, the fee rate may be too high to be useful and could be lowered toward some floor.
Proposed parameters
Here are some concrete proposed parameters and functions for analysis. They may be changed if we can quantify better values.
- RepaymentTerm: EPOCHS_IN_YEAR, the (expected) minimum sector commitment duration
- MaxRepaymentRewardFraction: 75%, the maximum fraction of earned rewards to redirect into shortfall repayments. The value 75% is chosen as equal to all of an SP’s vesting rewards.
- MaxFeeRewardFraction: 25%, the maximum fraction of earned rewards to charge as shortfall fee. The value 25% is chosen as the amount of earned rewards that don’t vest, so can be immediately used.
These imply:
- MaxRepaymentRate [FIL/epoch]: EpochReward * MaxRepaymentRewardFraction
- MaxNetworkShortfall [FIL]: MaxRepaymentRate * RepaymentTerm
- MaxFeeRate [FIL/epoch]: EpochReward * MaxFeeRewardFraction
For an SP with an actual MinerShortfall amount, given an EarnedReward:
- MaxMinerShortfall [FIL]: MaxNetworkShortfall * MinerPower/NetworkPower
- Repayment: (0.25 + 0.75 * sqrt( MinerShortfall/MaxMinerShortfall )) * MaxRepaymentRewardFraction * EarnedReward
- Fee: (MinerShortfall / MaxMinerShortfall) * MaxFeeRewardFraction * EarnedReward
Other impacts
Pledging more tokens
This proposal lets SPs onboard more power for the tokens they have available, it does not decrease the amount that they will lock.
- SPs still need actual tokens for at least ~2/3 (and increasing) of the required pledge, and
- The shortfall fee is expected to be higher than fees on leasing in most circumstances
For a shortfall to be preferable to leasing tokens for collateral, both (a) more than 2/3 of the demand for pledge must be available and met by FIL holders at fees compatible with positive sector net returns, and (b) the leasing fee must be higher than the shortfall fee (e.g. 33%). These are unlikely to happen at the same time: abundant supply will generally lead to low fees on leased tokens.
The shortfall fee quantifies how much more the network prefers locking an otherwise free token over a future/vesting one.
Decay over time
The mechanism of using block reward and a target repayment period to set the maximum shortfall means this mechanism naturally decays over time. As the block reward reduces, the maximum shortfall reduces, which reduces the maximum fees earned too. The impact on inflation remains strong if the repayment rate has a floor.
This decay is entirely appropriate, as in the long run we expect (1) leasing markets to mature and resolve the problems of access to collateral, (2) inflation to decrease both with the block rewards, and when SAFT vesting expires. This proposal cannot induce deflation or perpetual burning (but of course other mechanisms can and will).
Potential impact on consensus security
At first glance, reducing pledge per unit power seems to threaten consensus security, allowing a cheaper buying of power. However, this is only the case if unilaterally adopted. If most SPs take on similar levels of collateral shortfall, then while the total power:collateral ratio increases, no individual SP gains a significantly larger share of power or influence over block production. The raw cost of share of consensus power does not change much. Given upper bounds on the shortfall fraction, consensus risk seems minimal.
Potential for failure to repay shortfall
Because repayments are taken from earned rewards, it’s possible for an SP that earns much less than expected rewards to not reach full pledge before a sector is eligible for expiration. This could happen if
- an SP accidentally or intentionally does not produce blocks when eligible
- new miner actors have less than the minimum consensus power and are not eligible
- network power accelerates much faster than an SP’s share of that power, such that their earning rate slows down well below expectations
An un-earned reward is closely equivalent to a burn of that reward. With the current construction of the built-in reward actor, un-earned rewards will accumulate indefinitely and never enter the supply (TODO: confirm this). Protocol changes in the future could possibly alter the minting algorithm to eventually emit these tokens. Thus, from a network perspective, missing a reward has better supply impacts than earning and locking it anyway.
The case of new miner actors below the minimum consensus power needs analysis. Intuition suggests that such SPs earn no consensus power or rewards from the network, so pledging the minimum (2/3) requirement is already sufficient for incentive alignment and security.
The situation of rapid network growth reducing rewards below expectations needs analysis. At the high level this seems like a desirable outcome for the network, and any potential under-collateralisation by SPs left behind as something of a champagne problem. A related situation is the reduction of rewards to simple minting, when raw byte power is below baseline (as it currently is).
Alternative options
- The shortfall fees could be paid from the SP’s balance, instead of from actually earned rewards. This means the fees would accrue regardless of SP’s actual income.
- An SP could then go into fee debt if unpaid, and stop earning rewards, which could spiral out of control.
- Would be tough on new/small SPs, with more variable rewards, and time to cross the 10TiB minimum consensus power.
- Maintains high incentive to actually produce blocks, whereas proposal as written effectively reduces per-block rewards when a shortfall is taken.
- The decision to pay fees only from earned rewards leans towards safety for the SP, while the network may burn less tokens if something goes bad.