ENA Fee Switch Analysis
Blockworks Advisory supports the architecture, the supply anchors and the rung levels of the fee switch design. We have backtested the proposed buyback design day by day across 705 days of realized protocol revenue, USDe supply and ENA prices. The take comes off gross protocol revenue and is split pro rata across everything that revenue funds, so the sUSDe distribution and the two reward lines each fall by the take rate. On a 14 day supply average the design captures $52.7M a year while active, a buyback yield of 3.36 percent on ENA’s market capitalization, and it absorbs about a tenth of scheduled ENA gross unlocks. We recommend two changes and two remarks which would improve the schedule and they are set out below.
Background
The Ethena fee switch converts a share of protocol revenue into ENA buybacks. The Risk Committee agreed three success metrics for considering it in late 2024, which included USDe circulating supply above $6B, cumulative protocol revenue above $250M and USDe adoption across the top five centralized derivatives exchanges, where member recommendations ranged from three of five to four of five. That framework also set a priority order for yield distribution, placing Reserve Fund capitalization first, sUSDe competitiveness second and sENA revenue sharing on the remaining yield.
The design now replaces the discrete gate with a schedule where a take on gross protocol revenue rises with USDe supply as follows: five percent at $7.5B, 10 percent at $10B, 15 percent at $15B and 20 percent at $20B. The proposal contains no Reserve Fund term and no competitiveness condition, and it uses a protocol APY of 6 percent to illustrate the resulting dollar figures. At today’s USDe supply of $4.07B the schedule takes nothing, so nothing in this document is immediate.
The source of the take sets up everything that follows. It comes off gross protocol revenue and is split pro rata across everything that revenue currently funds. Over the last 30 days $175M a year was distributed across three live lines: 34.3 percent to sUSDe staking, 34.3 percent to partner payouts and 31.4 percent to Aave liquid leverage. A 10 percent take removes $6.0M a year from the sUSDe distribution and $11.5M from the two reward lines. That proportional cut is the mechanic every result below rests on and it does not depend on the shares.
The tests use 705 days of Ethena data, from 19 September 2024 to 24 August 2026, covering USDe circulating supply, protocol revenue, sUSDe and sUSDS yields, and ENA prices. The design is backtested day by day on realized revenue and realized prices under each candidate supply specification.
Key takeaways
First, the take is spread across three live claims on revenue. Over the last 30 days the distribution ran at $175M a year, split 34.3 percent to sUSDe staking, 34.3 percent to partner payouts and 31.4 percent to Aave liquid leverage. A pro rata take reduces each by the rate, so a 10 percent take removes $6.0M a year from sUSDe and $11.5M from the two reward lines.
Second, the programme is a revenue sharing mechanism. Backtested on a 14 day window it captures $52.7M a year while running and $8.82M a year spread over the full panel, buyback yields of 3.36 and 0.56 percent on ENA’s $1.57B market capitalization. Against scheduled gross unlocks of $512M a year to April 2028 that is about a tenth at the running rate and under two percent across the cycle.
Third, a 14 day trailing average is the right supply measure where supply first crossed $7.5B on 27 July 2025. A 14 day average would have started the buyback seven days later and a 90 day average alternative 34 days later, and the shorter window captures more, $8.82M against $7.66M per panel year, with no additional rung switching. A single day’s supply can be moved by one large mint, which is why the average is preferable to the daily print.
Fourth, the schedule reads only supply and supply moves more slowly than yield. In October 2025 the yield ratio fell below parity within days while supply held above the $10B rung for weeks and the schedule kept taking. Across the 118 active days the take pushed sUSDe below sUSDS on 11 of them, all landing between 0.90 and 1.00, and the worst post take ratio of 0.44 fell on a day that stood at 0.49 before any take was applied. We recommend the disclosure and monitoring of this exposure. A parity floor would remove all 11 days and it gives up more capture than the exposure is worth.
The sections that follow take each finding in turn, beginning with what the take removes and from where.
What the take costs
The allocation of USDe fees has three live lines. Over the last 30 days the sUSDe distribution took 34.3 percent of what was distributed, partner payouts 34.3 percent and Aave liquid leverage 31.4 percent. A pro rata take reduces each of them by the take rate. The shares matter for the dollar impact on each line where a staker, a 10 percent take is a 10 percent cut to sUSDe APY whatever the allocation looks like.
The allocation has shifted. Across the full history sUSDe staking absorbed 65.4 percent of everything distributed. Over the last 30 days it took 34.3 percent, because a smaller fraction of USDe is staked than in earlier periods and reward spending has grown to fill the gap. Table 1 sets out what a take at each rung removes from each line at today’s shares.
| Rung and take rate |
From sUSDe |
From partner payouts |
From Aave liquid leverage |
Total |
| $7.5B, 5% |
$3.0M |
$3.0M |
$2.8M |
$8.8M |
| $10B, 10% |
$6.0M |
$6.0M |
$5.5M |
$17.5M |
| $15B, 15% |
$9.0M |
$9.0M |
$8.3M |
$26.3M |
| $20B, 20% |
$12.0M |
$12.0M |
$11.0M |
$35.1M |
Table 1. Annual amount removed from each line at each rung, on the distribution of the last 30 days.
Two features of the allocation are worth noting. Mint fees, $8.4M a year and 4.7 percent of gross revenue, are reported as a separate line and are not part of the distribution, so they are excluded from the shares above. The Reserve Fund has received no allocation since 12 December 2024, having taken $35.7M cumulatively before that. That is consistent with the fund being adequately capitalized against current exposures and it means the first claim in the 2024 priority order is already satisfied at no ongoing cost. Therefore, the question the schedule decides is the second claim against the third, sUSDe competitiveness against ENA buybacks. The rest of this note addresses it and the evidence bears on how tightly that second claim needs to be protected.
Having established what the take costs, the question becomes which supply measure should trigger it.
The supply window
Four specifications were tested against the same rung levels and the same revenue series. Table 2 reports that the differences are small and they run in favor of the shorter windows.
| Supply measure |
Days to first buyback |
Active days |
Rung changes |
Capture, panel year |
Capture, while active |
| Daily target |
0 |
117 |
4 |
$8.79M |
$53.0M |
| 14 day average |
7 |
118 |
4 |
$8.82M |
$52.7M |
| 30 day average |
14 |
120 |
4 |
$8.64M |
$50.8M |
| 90 day average |
34 |
141 |
4 |
$7.66M |
$38.3M |
Table 2. Days to first buyback is measured from 27 July 2025, when daily supply first crossed $7.5B. Capture per panel year spreads the total over all 705 days and capture while active annualizes it over the active days only.
Rung stability is the same across all four specifications. Each changes rung four times across the panel and each turns the mechanism on once and off once. The 90 day average is active on more days only because it lags the drawdown, and those extra days fall in a lower revenue period, which is why it captures less. The chart below shows where that lag arises.
The shorter window carries one consequence. The 14 day average peaked at $14.65B in October 2025, which is 2.3 percent short of the $15B rung, while the 90 day average peaked at $12.38B, which is 17.5 percent short. A 14 day window makes the upper rungs reachable on a supply spike that a longer window smooths away and those rungs carry no backtest. The daily target reaches the first rung a week sooner and captures almost the same amount. The reason to prefer the average is that a single day’s supply can be moved by one large mint, while a 14 day average requires the supply to persist. We recommend a 14 day trailing average of USDe supply and note that the $15B and $20B rungs are untested. What those rungs would deliver in dollars depends on the revenue assumption, which is the next question.
Sizing
Protocol APY, defined here as 30 day annualized revenue divided by USDe supply, converts a supply threshold into a dollar figure. It has exceeded 6 percent on 121 of the 705 panel days, or 17 percent of the sample, so the assumption is not arbitrary even though it does not describe the median day. The protocol’s diversification of backing assets could return protocol APY to that level. It is not where the protocol sits at the date of writing and the gap between the two is what Table 3 quantifies.
| Rung |
Take rate |
At 6% illustrative |
At 4.60% realized |
| $7.5B |
5% |
$22.5M |
$17.3M |
| $10B |
10% |
$60.0M |
$46.0M |
| $15B |
15% |
$135.0M |
$103.5M |
| $20B |
20% |
$240.0M |
$184.1M |
Table 3. Annual buyback at each rung. The two upper rungs have never been reached on any supply measure.
Those rung figures describe a running programme. What the backtest delivers over a full cycle is smaller, because the design spends most of the period switched off. Backtested on a 14 day window it captures $17.04M in total. That figure can be annualized two ways and the difference matters. Spread over all 705 panel days it is $8.82M a year, a buyback yield of 0.56 percent on ENA’s current $1.57B market capitalization. Annualized over the 118 days the design would have been running it is $52.7M a year, a yield of 3.36 percent. The first number describes the programme across a full cycle including the periods it sits idle and the second describes it while it is working.
Scale is best judged against the supply the buyback has to absorb. ENA gross unlocks continue to April 2028 at 267,187,500 tokens a month, worth roughly $42.7M a month or $512M a year at the current price. At its while active run rate the buyback offsets about 10 percent of that, and at the panel average rate under 2 percent. Therefore, the programme is a revenue sharing mechanism and settles what the design delivers. The remaining question is what it costs sUSDe.
Competitiveness
The schedule is grounded on supply and it moves more slowly than yield. Yield can collapse in a day when the deposit rate or the funding environment turns, while supply leaves through a seven day cooldown and unwinds slowly, so there is a window in which the ability to pay has gone and the schedule is still taking. Here, the context frames how much that matters. sUSDe sat below sUSDS on 243 of the 705 panel days with no fee switch running at all, so competitive pressure is a market condition the design inherits and does not create. What the design controls is whether it adds to that pressure.
Over the 118 active days the take pushed sUSDe below parity on 11 days that had started above it and 25 days were already below before any take. The 11 breaches are marginal. Their pre take ratios run from 1.001 to 1.103 and their post take ratios from 0.90 to 1.00, with six of the 11 above 0.95. The worst post take ratio of 0.44 was reached in the October 2025 collapse, on a day when the ratio stood at 0.49 before any take was applied. The severe state is a market condition where the design deepened at the margin and did not produce.
A floor at parity, capping the take so that post take sUSDe APY never falls below sUSDS APY, removes all 11 breaches. It costs 27.7 percent of gross capture, $4.72M of the $17.04M the design captures across the panel, binds on 36 of the 118 active days and reduces the take to zero on 25 of them. Those 25 are the days that began below parity, where no positive take is compatible with the floor. The floor lifts the mean active yield ratio by 0.020, which on the supply coefficient in Table 5 is associated with $30M to $54M of additional USDe on mean active supply of $11.4B, or $1.4M to $2.5M a year of revenue at a 4.60 percent protocol APY. It gives up $4.72M to buy that. On the coefficients this note uses elsewhere the floor costs two to three times what it returns and the staked share regression, which is the direct test of the harm a floor prevents, finds nothing.
Lowering the rungs is the obvious alternative and it works less well. Cutting every rate by a quarter leaves nine breaches and gives up 25 percent of capture, while halving them leaves five breaches and gives up half. The reason is in the headroom distribution: on the days that began competitive, the maximum take compatible with parity had a median of 22.6 percent, a tenth percentile of 6.4 percent and a minimum of 0.1 percent. Most days had far more room than the schedule uses and a few had none, so a flat cut penalizes the many to protect the few while a floor acts only where it is needed. Neither bound is at no cost during the observed history, which is why we recommend disclosure and monitoring in place of a standing cap.
Whether a weaker ratio costs anything is a separate question and we test it two ways. The explanatory variable is the contemporaneous ratio of sUSDe APY to sUSDS APY. Each dependent variable is measured 30 days forward, so the estimate describes what follows a given level of competitiveness. Consecutive windows overlap by 29 days and all standard errors are Newey West with 30 lags. The ratio is winsorized at 3.0 to limit the influence of the December 2024 sUSDe spike, and all specifications use 675 daily observations. Table 4 reports the first test.
| Yield ratio |
n |
Mean 30d change |
95% interval |
p |
| Below 0.95 |
190 |
−2.54% |
−10.37 to +5.30 |
0.53 |
| 0.95 to 1.00 |
53 |
−0.65% |
−5.73 to +4.42 |
0.80 |
| 1.00 to 1.25 |
211 |
+1.79% |
−2.48 to +6.06 |
0.41 |
| 1.25 and above |
221 |
+0.56% |
−8.09 to +9.22 |
0.90 |
Table 4. Forward 30 day change in the staked share of USDe by yield ratio bin. HAC standard errors, 30 lags.
The result is stakers do not visibly leave. The yield ratio explains less than one percent of the variation in the staked share, none of the four bins is significant and the largest point estimate, a 2.54 percent outflow below parity, is small in itself. The intervals are wide, the below parity bin does not exclude an outflow of 10.4 percent and the estimate is pooled over a period in which the revenue mix changed. Within those limits there is no measurable evidence that a weaker sUSDe drives stakers out.
Total supply is a different matter. Table 5 reports both dependent variables together.
| Dependent variable and control set |
(1) |
(2) |
(3) |
| Staked share of USDe, coefficient |
0.028 |
0.038 |
0.034 |
| HAC standard error |
(0.038) |
(0.039) |
(0.035) |
| R-squared |
0.008 |
0.069 |
0.076 |
| USDe supply growth, coefficient |
0.264 |
0.267 |
0.240 |
| HAC standard error |
(0.066) |
(0.067) |
(0.059) |
| R-squared |
0.202 |
0.205 |
0.314 |
| Protocol APY control |
no |
yes |
yes |
| Log USDe supply control |
no |
no |
yes |
Table 5. Coefficient on the sUSDe to sUSDS yield ratio. Both dependent variables are forward 30 day changes. Newey West standard errors with 30 lags, 675 observations in every column.
The yield ratio shows no relationship with how USDe splits between staked and unstaked, but it moves closely with how much USDe there is. On composition the coefficient is 0.03 against a standard error of 0.04. On total supply it is 0.24 to 0.27 with a t-statistic near 4.0 throughout. A 10 percent pro rata take moves the median active ratio from 1.178 to 1.060, which on column (3) is associated with 2.8 percentage points of slower 30 day supply growth, falling to 1.5 points once past supply growth enter as controls. These are predictive associations estimated on observational data and should not be read as causal effects. The direction is the point. A take that weakens sUSDe is associated with slower growth in the variable the schedule itself triggers on. These coefficients are estimated on 675 overlapping windows drawn from a period in which no take was applied, so carrying them into the schedule is an extrapolation.
Risk considerations
The schedule carries a competitiveness exposure that should be disclosed and monitored. On 11 of the 118 active days the take pushed sUSDe below sUSDS from a starting point at or above it, with post take ratios between 0.90 and 1.00, and the worst observed post take ratio of 0.44 fell on a day that stood at 0.49 beforehand. We recommend the post take yield ratio is published alongside the take each period, so that a shift in that distribution is visible and can be acted on before it becomes material. On this history the exposure does not justify a standing cap and a cap that binds is itself costly. Any floor at or above parity zeroes the take on the 25 days that began below parity and a programme that stops is harder for holders to follow than one that runs continuously.
The trigger is procyclical where a 30 day changes in supply correlate with 30 day changes in the ENA price at 0.58, so the design buys into strength. On the 14 day specification the backtest paid an average of $0.5369 per ENA against $0.4761 for an equal daily spend across the same active days, a penalty of 12.8 percent. It is a property of any growth triggered buyback and it is not specific to this design.
The allocation of revenue has moved. sUSDe staking took 65.4 percent of everything distributed since inception and 34.3 percent over the last 30 days, while Aave liquid leverage went from nothing to 31.4 percent. A schedule expressed as a share of gross revenue inherits whatever that split happens to be on the day it activates.
Conclusions
We support the architecture, the supply anchors and the rung levels. The design links ENA value accrual to the growth of the core product in a way that is public and easy to follow and that is the right foundation. Our recommendations concern how the take is triggered, disclosed and presented, and are as follows.
First, adopt a 14 day trailing average of USDe supply as the trigger instead of the daily target. It starts the buyback seven days sooner than a 30 day average and 27 days sooner than a 90 day average, captures more, shows no additional rung switching, and unlike the daily target it cannot be moved by a single large mint.
Second, disclose and monitor the competitiveness exposure. In the observed history the take pushed sUSDe below sUSDS on 11 of the 118 active days, with post take ratios between 0.90 and 1.00 and the worst post take ratio of 0.44 fell on a day that stood at 0.49 before any take. We recommend the post take yield ratio is published alongside the take, so that a change in that distribution is visible. We do not recommend a standing yield floor. A floor at parity removes all 11 days at a cost of 27.7 percent of gross capture, which is more than the associated supply effect is worth on our own estimates and the staked share regression finds no measurable outflow from a weaker ratio.
Third, mark the $15B and $20B rungs as untested, because a 14 day window brings them within reach of a supply spike and no backtest covers them.
Fourth, present the rung figures at realized protocol APY, with 6 percent labelled as an illustration, so the published numbers are not read as expected buyback volume.
These results rest on the take being pro rata. A take carved only from USDe reward programmes would leave sUSDe yield unchanged, the 11 breaches would not arise and the competitiveness exposure described here would not apply.
