# Token-holder votes on supply and fees show no reliable price drift

> Proposals to burn tokens, share fees or cut emissions were not followed by a reliable 24-hour price drift; the average gain rested on a few outliers, and the mirror short on supply proposals lost money on average.

E819 · Verdict: KILL · Cause: Concentrated · Market: Crypto & DeFi · Mechanism: Public intent · Horizon: 24h · Tested: October 2026

Source: https://alphanotfound.eu/experiments/E819-governance-proposals-token-price-drift

Token communities vote on proposals that change how much of a token is created, burned or sold. The belief is that holders learn of a buyback, burn or fee share hours late and buy slowly. We took 199 such proposals in 43 token communities and bought each token 15 minutes after the vote was posted, hedged against Bitcoin's move, for 24 hours. The average was +40.1 bp (bp = hundredths of a percent) after costs, but the plausible range ran from -21.7 to +123.4 bp, the median trade lost, and the average was no better than after random proposals. The mirror short on 49 supply proposals lost 5.3 bp.

## Key numbers

- **Burn, buyback and fee-share proposals:** 199 (43 token communities, Sep 2020 to Oct 2026)
- **Average net per trade:** +40.1 bp (90% interval -21.7 to +123.4 bp; median trade -74.1 bp after costs (-54.1 before))
- **Minus random proposals:** +39.4 bp (90% interval -32.5 to +129.9 bp)
- **Supply proposals, short:** -5.3 bp (49 trades, 90% interval -103.1 to +77.4 bp)

## The idea being tested

**Belief:** A token proposal that makes the token scarcer or more valuable is not priced the moment it is posted, so the price keeps rising for a day.

**Why people believe it:** Proposals appear on an off-chain governance voting platform first and reach headlines hours later. Anyone reading the platform could act before the wider market, and the same holds in reverse for proposals that sell treasury tokens, unlock supply or raise emissions.

**Why it would persist:** Hundreds of communities post proposals every week and few traders read them, so the argument goes that nobody is paid to watch them all.

## How it was tested

We downloaded every proposal in 100 token communities and sorted the titles with a fixed list of keywords: burn, buyback, fee share or emission cut on one side; treasury sale, unlock or emission increase on the other. For each proposal we simulated buying (or shorting) the token on a large crypto exchange 15 minutes after it was posted and closing a day later. Every trade was hedged against Bitcoin and charged trading costs. We compared the result with trades placed after random proposals of the same communities, and with trades placed when the vote ended instead of when it began.

**Pre-registered:** yes

**Data:** Proposals (title, creation and end time, votes) of 100 token communities (off-chain governance voting platform, public query interface (anonymous); per proposal; 2020-09..2026-10); 5-minute opening prices of the token against USDT (perpetual contract, or spot where no perpetual existed) (largest USD-margined crypto perpetual venue and its spot market, public market data; 5 minutes; 2020-09..2026-10)

**Universe:** 100 token communities, picked by hand from names and symbols only, each with a token traded against USDT on the venue at some point. 22,471 proposals in total.

**Signal:** Lower-cased proposal title matched against fixed keyword lists. Supply-reducing class (UP): fee switch, buyback, burn, revenue or fee share or distribution, repurchase, value accrual, activating or enabling a fee, or a cut in emissions, inflation or issuance. Supply-adding class (DOWN): treasury sale or diversification, OTC, token sale, selling from the treasury, increasing or extending emissions, inflation or incentives, unlock. UP wins where both match. Titles containing "test" and flagged proposals were dropped. One event per community and class per 7 days (the first). Control class: all other proposals of the same communities, one per community per 7 days, then a random sample of at most 800. Result: 237 UP, 105 DOWN and 800 control events, of which 199, 49 and 609 had a price (the token was not listed on the venue at that time for the rest).

**Execution:** UP: long, DOWN: short. Entry at the open of the first 5-minute bar at least 15 minutes after the proposal was created, exit 24 hours later. Perpetual contract where one existed, otherwise spot (long only). DOWN was traded on perpetuals only. Simple returns on notional. BTC-adjusted return = token return minus the Bitcoin perpetual return over the same window (beta 1; the Bitcoin leg is not charged).

**Costs:** 14 bp round trip on the perpetual, 20 bp on spot. Funding ignored.

**Controls:** Random proposals of the same communities (decides the verdict). Placebo: the same UP and DOWN proposals traded from 15 minutes after the vote ended instead of after it was created.

**Sample split:** Rule and criteria fixed before any price was fetched. Results by calendar year reported descriptively.

**Pass criterion:** UP long: at least 40 events; mean BTC-adjusted net above 0 with a community-clustered 90% bootstrap interval above 0; median BTC-adjusted gross above 0; UP minus control interval above 0; vote-end placebo mean below half the creation-time mean. DOWN short: the same tests on the short return.

**Kill criterion:** Any failed test for a class. Fewer than 40 events would have been inconclusive.

**Power:** The clustered 90% interval for the UP long is 145 bp wide (half-width about 72 bp): effects smaller than that cannot be told from zero.

## Result

**Why it was rejected:** Both classes were rejected. For UP the interval included 0, the median trade lost and the difference to random proposals included 0; for DOWN the interval included 0 and the difference to random proposals included 0.

**Observations:** 199

**Net mean:** +40.1 bp

**Median:** -74.1 bp

**Worst:** -2,586 bp

- UP long, 199 proposals in 43 communities: +40.1 bp BTC-adjusted net per trade after costs, with a community-clustered 90% interval of -21.7 to +123.4 bp. The median trade lost 74.1 bp after costs (54.1 bp before costs), and 40.7% of the trades gained after costs. The worst trade lost 2,586 bp.
- Random proposals of the same communities, traded the same way (609 trades in 78 communities), averaged +0.7 bp net with a median of -13.8 bp before costs. UP minus random is +39.4 bp (difference of the two means; 90% interval -32.5 to +129.9 bp).
- Control timing: the same 200 UP proposals traded from vote end instead of creation averaged -18.8 bp net.
- The average depends on a handful of trades. Removing the five best of the 199 turns +40.1 bp into -36.5 bp.
- By year the UP average changes sign: 2020 +721.7 bp (only 6 trades), 2021 -125.0 (30), 2022 +42.5 (66), 2023 -72.6 (41), 2024 +281.5 (24), 2025 +27.9 (18), 2026 +21.9 (14). The largest average comes from 6 trades, and the positive years are no more reliable than the negative ones given the interval above.
- One token with frequent burn votes accounts for 55 of the 199 UP trades.
- DOWN short, 49 perpetual trades in 21 communities: -5.3 bp net per trade, 90% interval -103.1 to +77.4 bp. DOWN minus random (perpetual trades only, 426 of the 609): +21.8 bp, 90% interval -86.8 to +112.3 bp.
- The DOWN median is positive (+136.3 bp before costs; 59.2% of the shorts gained after costs) while the mean is negative. A positive median with a negative mean means most shorts made money and a few lost far more: the worst trade lost 2,297 bp when a token rose. The mean is what a trader earns, so the positive median does not count.
- Title matching is rough: a proposal to deprecate buybacks and a proposal to restore a buyback counter on a website were both classed as UP among the 237 UP matches, so some UP events are not what the keyword list meant.

### Both intervals include zero (bp)

Mean net return per trade after costs, measured as the token's move minus Bitcoin's move, with 90% intervals that treat each community as one unit (resampled many times). Passing required each interval to sit above 0; none does, and the UP interval is wide enough to hold both a loss of 22 bp and a gain of 123 bp.

|   | Mean | 95% low | 95% high |
| --- | --- | --- | --- |
| UP long (199 proposals) | 40.1 | -21.7 | 123.4 |
| UP minus random proposals | 39.4 | -32.5 | 129.9 |
| DOWN short (49 proposals) | -5.3 | -103.1 | 77.4 |
| DOWN minus random proposals | 21.8 | -86.8 | 112.3 |

### The sign flips from year to year (bp)

Mean net return of the UP long by creation year, number of trades in brackets. Years differ in sign and size, and each year has at most 66 trades, so no year is evidence by itself; per-year figures are descriptive, not tests. 2020 (+721.7 bp, 6 trades) is left off the chart because six trades say nothing.

|   | UP long, net |
| --- | --- |
| 2021 (30) | -125.0 |
| 2022 (66) | 42.5 |
| 2023 (41) | -72.6 |
| 2024 (24) | 281.5 |
| 2025 (18) | 27.9 |
| 2026 (14) | 21.9 |

### Shorts: many small wins, a few large losses (% of trades)

Share of the 49 DOWN short trades and the 199 UP long trades by net return in bp. The DOWN shorts win more often than not but their left tail is long, which is why the median is positive and the mean is below zero. The UP long has no such pattern: it loses more often than it wins.

| Bin | DOWN short (49 trades) (count) | UP long (199 trades) (count) |
| --- | --- | --- |
| < -800 | 5 | 7 |
| -800..-400 | 2 | 18 |
| -400..-200 | 4 | 32 |
| -200..-100 | 5 | 37 |
| -100..-50 | 2 | 9 |
| -50..0 | 2 | 15 |
| 0..50 | 1 | 8 |
| 50..100 | 3 | 5 |
| 100..200 | 6 | 15 |
| 200..400 | 10 | 25 |
| 400..800 | 8 | 18 |
| > 800 | 1 | 10 |

## Interpretation

**Why it failed (interpretation, not established):** The proposals usually follow weeks of discussion on community forums, so the vote itself is old news, and most are small next to the token's daily volume. The few large moves come from single tokens in single years, which a clustered interval correctly treats as little evidence. A positive mean from a few big winners, a negative median and a sign that changes with the year is also what noise looks like.

**Lesson:** A positive average with a negative median is a warning that a few trades carry the result, and it needs the same test as any other claim: an interval, a control and a year-by-year check. A signal read from text, here a keyword list on titles, adds its own noise (a proposal to remove a buyback was counted as a buyback), so a weak result from it says little about the underlying idea in either direction.

**Scope:** Covers proposal titles of 100 hand-picked token communities, matched by keyword, traded 15 minutes after posting for 24 hours against a Bitcoin hedge, September 2020 to October 2026. Tokens not listed on the venue at the time are not included.

## Reproducibility

**Steps:** 1) Download all proposals for the 100 communities from the governance platform's public GraphQL interface and keep title, creation time, end time and flag. 2) Drop titles containing "test", flagged proposals and anything created outside September 2020 to October 2026; match the lower-cased title against the UP and DOWN keyword lists (UP first), else control; keep the first event per community and class in each 7-day span and sample at most 800 controls with a fixed seed. 3) For each event take the open of the first 5-minute bar at least 15 minutes after creation and the open 24 hours later on the perpetual (spot if there was none; DOWN only on perpetuals); subtract the Bitcoin perpetual return over the same window; subtract 14 bp (perpetual) or 20 bp (spot). 4) Bootstrap by community with 4,000 resamples for the 90% intervals, resampling the event and control groups independently for the differences; repeat from vote end for the placebo.

**Grade:** COMPLETE

---
AI-generated research, checked but not peer-reviewed. Historical, experimental research, not investment advice. Dataset Terms of Use: https://alphanotfound.eu/terms