回购与直接分配：价值传递的形式重要吗？

This cp0x study asks whether the form in which a DeFi protocol passes value to its governance token is linked to the token's later price against ETH. It compares direct distributions of fees, buybacks with distribution of the purchased tokens (buy-and-distribute) and buybacks into the treasury, using seven dated transitions in seven protocols between February 2022 and May 2025 plus a two-step case at Yearn. Each token was measured over 90 calendar days after the first actual flow to holders, against a matched "twin" token where one qualified and otherwise against a basket of COMP, CRV and ENS. Six of the seven cases came in below the main control, one of them practically at zero (Sushi, -0.3 percentage points), and the only positive case was ether.fi at +9.8 points against the basket. Seven cases are far too few to show that the forms are equivalent, but the study found no historical support for the view that the market rewards a more direct flow.

Key findings

• In this study (seven DeFi value-transfer transitions, 2022 to 2025), no form of value transfer showed a consistent 90-day price advantage against ETH-relative controls, and six of the seven cases were below the main control.
• Both introductions of direct distributions from zero underperformed their control over 90 days, Balancer (April 2022) by 11.2 percentage points against its twin COMP and Jito (January 2025) by 2.3 points.
• The only clean switch from buy-and-distribute to direct distributions found by this study, Trader Joe in February 2022, trailed the control basket by 36.3 percentage points over 90 days while JOE lost 51.6% against ETH.
• Both switches from direct distributions to buy-and-distribute were negative over 90 days, Gains Network (July 2024) at -31.8 percentage points against its twin and the basket and GMX (October 2024) at -81.9 points against the basket.
• In this study an early signal in favour of direct distributions disappeared after checking what holders actually received, which moved Sushi and ether.fi to buy-and-distribute, and after an external review replaced row-count windows with calendar windows.
• Only three of the seven cases had a qualifying twin token, all three twins were COMP, and in all three the 90-day sign against the twin matched the sign against the basket.
• A frozen search of 164 DEX and derivatives protocols found no new switches from direct distributions to buy-and-distribute, but 144 of the 164 had sources too thin to classify even as negative cases, so this does not show that such switches are rare.

Study at a glance

Question and motivation

DeFi protocols pay holders part of revenue directly, buy the token and hand it to stakers, or buy it into the treasury. Tokenomics debates often claim that the market rewards the more direct cash flow.

The study tested whether that claim has historical support. It also asked whether introducing value transfer from zero behaves differently from replacing a working form. The unit of observation is one dated change in the economic link between the token and protocol revenue.

Data

Inclusion required a confirmed form before and after, a confirmed date of first actual flow to holders, and prices. Kyber was excluded because its change of form (spring 2021) and full flow (around July 2022) were more than a year apart. Vertex was left out because its first flow could not be dated within December 2024 to February 2025. Pendle (a 30% emission cut in the same package), Impermax (a hybrid) and Lista (unverified prior form, no prices) were also excluded. Changes involving treasury buybacks, such as Yearn and dYdX, are "ladder" cases outside the four categories.

Method

1. Classify each change. Category A is direct distributions introduced from zero, B is buy-and-distribute replaced by direct distributions, C is the reverse of B, and D is buy-and-distribute introduced from zero.
2. Date the event as the first day the new flow reached holders, not the vote, using on-chain evidence where no date was published.
3. Express prices in ETH. The value on a date is the last observation on or before it and at most 7 days old. Token and controls share calendar windows of 30 and 90 days.
4. Choose the twin with the smallest distance, the absolute gap in drawdown from the 365-day maximum plus 0.5 times the absolute gap in 90-day pre-event return. A twin needs 365 days of history and distance at most 0.25.
5. Exclude from the twin pool the study's own tokens, LDO, SNX, tokens with their own dated event from 45 days before the event to the end of the window, and tokens with a disqualifying event such as the Vertex shutdown.
6. Use the basket fixed in advance when no twin qualifies, report both controls, and flag whether 90-day signs agree.
7. Report the arithmetic difference of simple returns in percentage points, which overstates differences when moves are very large.
8. Record confounders such as unlocks, emission changes and redesigns. Rows with a strong confounder stay visible but are not strict evidence.
9. Do not build a pooled effect until categories B and C have enough events, and do not read ladder steps as a monotonic scale.
10. Search for more cases with targeted checks, pattern searches and a frozen procedure over a pre-fixed universe. A template counts as validated only if it finds a known positive control without its name.

Results

Seven transitions against controls

The table shows the 90-day outcome of each main case against its twin and against the basket, in percentage points.

Six of seven are below the main control. Four of the seven could be compared only with the basket, and all seven tokens fell against ETH over 90 days.

Yearn, one protocol changing form twice

Yearn is a ladder case outside the four categories.

Dropping direct payouts coincided with a poor outcome, and distributing bought-back YFI was strongly positive in the first month. The 90-day sign of that step depends entirely on the control.

In the current calculation the three replacements of a working mechanism sit further below control than the introductions from zero. This question arose during the study and its picture changed twice with the control method, so it is not a finding.

Search for more transitions

The table summarises the search for further cases.

The procedure recovered both known C cases, but its low recall means the absence of new cases is not evidence of rarity. The reverse direction B could not be validated at all.

Robustness and replication

The corrections note records how the headline numbers changed.

The main conclusion survived, but ether.fi's best result proved an artefact of a closed twin.

Limitations

• Each category has one or two events, too few for a statistical comparison. Absence of evidence here does not mean the forms are equivalent.
• Only three of seven events have a qualifying twin, all COMP, so the twin layer adds little beyond the basket, and control quality differs between groups.
• Most changes came with unlocks, emission changes or redesigns, so results are associations, not causes.
• The frozen C query was phrased around former ETH payouts and may miss switches from USDC or SOL payouts. The frequency of B switches is not estimated.
• Analysis rules were frozen during the study, after early results, so pre-registration is partial.
• Balancer and Jito were in effect fee switch activations. Such cases are rare and fee switch activations were not tested systematically.
• Price outcomes do not capture transparency, alignment, tax and legal effects or treasury discipline.

What this means for DAOs

The study gives no empirical basis for choosing direct distributions over a buyback, or the reverse, to get a better token price. This matches the overall conclusion of our other cp0x studies of token value. The form is better argued on transparency, alignment, tax, legal and treasury grounds.

Frequently asked questions

Do direct fee distributions raise a token's price more than buybacks?

This study found no evidence of that in seven DeFi cases from 2022 to 2025. Both introductions of direct distributions were below control over 90 days (Balancer -11.2 pp, Jito -2.3 pp), and the one switch to direct distributions, Trader Joe, was 36.3 pp below the basket.

Do buybacks with distribution do better?

No consistent advantage appeared. Introductions of buy-and-distribute were close to the basket (Sushi -0.3 pp, ether.fi +9.8 pp), while both switches to buy-and-distribute were negative (Gains -31.8 pp, GMX -81.9 pp over 90 days).

What happened to the early signal in favour of direct distributions?

It disappeared once the study checked what holders actually received, because Sushi and ether.fi holders got purchased tokens. External review then corrected the windows and a closed twin, and Balancer moved from +24.2 pp to -11.2 pp against its twin.

Does the search of 164 protocols show that such switches are rare?

It does not, because the automated screen left 144 of 164 DEX and derivatives protocols as no_low_recall, 129 of them with no historical evidence. It covered only switches from direct distributions to buy-and-distribute, and no procedure recovered the known reverse case.

Why compare against a twin and a basket?

Prices in ETH remove market moves and controls remove sector moves. Only three of the seven cases had a twin within distance 0.25, so the COMP, CRV and ENS basket carries most comparisons and both are shown.

Data and code

All data, code, the frozen method and the review for this study are on GitHub.

Prices are not redistributed, so run code/fetch_prices_stage3.py first and then follow code/README.md.

Version history