Everything below is built from the on-chain treasury balance (data basis and limitations at the end). If you find an error in the data, let me know and I’ll correct it. If anyone at Intersect or elsewhere wants this re-run with different assumptions, different mixes, or the exact tier structure, happy to do that as well.
Backdrop to the analysis: back in March, the “Investment and Risk Framework for the Cardano Treasury” thread asked for exactly this kind of work (VaR and expected-shortfall models, stress scenarios, allocation analysis). Four months later that request is still open, the treasury’s dollar value has roughly halved, and this month DReps are voting on budget withdrawals, a 120M ADA single action, and a proposal to raise the annual spend limit. Those votes all draw on one treasury pot and these are risk numbers for that pot.
TL;DR — four model estimates. Every one of them depends on the assumptions listed at the end. None of them is a fact or a prediction, and none of it is advice.
As of July 13, the treasury stands at 1.474B ADA with a value of $233M. For this, we found the following:
- Across two deliberately different Monte Carlo models, the simulated 1-in-20 bad year for the pot’s dollar value lands anywhere between −73% and −94%. The distance between those two numbers is model risk, and it’s the strongest argument I know for the written risk framework the March thread proposed.
- What diversification buys: an illustrative 10% BTC+stablecoin mix improves the 1-year 95% expected shortfall by about $14M on the $233M pot, and softens an FTX-week repeat by about 1.3 points. A 5% mix buys about half of that. This is real money, though as section 2 shows this doesn’t change the pot’s basic character.
- The 350M ADA annual budget bought $55M of real-world goods and services (using July 13 prices). At the year-end median of the bootstrap model it buys $12M. The treasury can always pay the ADA; what the ADA pays for is the open question. Section 1 runs the same arithmetic on the withdrawal asks being voted on this month.
- Converting $100M of ADA to anything else, selling no more than 5–10% of each day’s trading volume so the sale itself doesn’t move the price, takes roughly two months in a normal market and close to a year in a crisis on our volume data, with $27–41M of modeled price impact along the way. The day-count depends heavily on which daily-volume figure you accept for ADA; section 3 shows both bases.
Snapshot: treasury balance 1,473,782,005 ADA at epoch 642, read from the chain (Koios) on 2026-07-13; ADA at $0.158136 (July 13 close). Coverage is 100% of the pot by construction. Books modeled: A = the pot as it is (100% ADA). B and C = illustrative 95/5 and 90/10 mixes, with the non-ADA share split evenly between BTC and USDC. B and C match the 5–10% range that has been floated publicly; they are constructions for measurement, endorsed by nobody.
1. Monte Carlo Analysis
We ran two MC simulation models, same pot, same seed, 25,000 paths each over 365 days. One (GBM) takes ADA’s measured volatility and assumes no view on direction. The other (an empirical bootstrap) resamples the last year’s actual daily returns, which for ADA was a brutal year, so it carries that year’s downtrend into the future.
| 1-year outcome, $233.1M start | GBM | Bootstrap |
|---|---|---|
| 1-in-20 bad year (5th percentile) | −73% | −94% |
| Median | +0.4% | −79% |
| 1-year 95% VaR | $171M | $219M |
| Expected shortfall (95%) | 80% | 96% |
| Median worst drawdown within the year | 59% | 84% |
| Paths with a 20%+ drawdown | 100.0% | 100.0% |
GBM answers “what does roughly 80% annualized volatility do to a single-asset pot, direction-neutral?” The bootstrap answers “what if the coming year statistically resembles the last one?” Its −79% median is a replay of a terrible year, not a forecast. Reality gets to pick something in between. Two things hold across both columns, though: a 20%+ drawdown happened in every single simulated path, and the 1-in-20 number ranges over twenty-one points of the pot’s value depending on model choice alone. Any budget policy sized off one VaR number inherits that spread silently. A written framework with stated models and stated confidence levels (what the March thread asked for) is how you stop inheriting it silently.
Purchasing power of the 350M ADA annual budget (arithmetic on the same simulations; this is about what the budget buys, never about whether it can be paid):
| Today | GBM, bad year (p5) | GBM, median | Bootstrap, bad year | Bootstrap, median | |
|---|---|---|---|---|---|
| 350M ADA buys | $55.3M | $14.7M | $55.6M | $3.3M | $11.8M |
If the proposal to raise the limit passes, scale the row up proportionally; the shape of the risk doesn’t change, only the notional.
Two of the withdrawal asks on the ballot this month, run through the same percentile prices. Same reading as the budget row: this is what the requested ADA is worth at simulated year-end prices
| Today | GBM, bad year (p5) | GBM, median | Bootstrap, bad year | Bootstrap, median | |
|---|---|---|---|---|---|
| 120M ADA (the Cardano PRIME action) | $19.0M | $5.0M | $19.1M | $1.1M | $4.0M |
| 25.4M ADA (Intersect operations) | $4.0M | $1.1M | $4.0M | $0.24M | $0.86M |
2. Diversification
The 5–10% diversification idea has been on the table for a year and here are the number on what it does. Using illustrative BTC+USDC mixes at July 13 prices we get the following:
| A: 100% ADA | B: 95/5 | C: 90/10 | |
|---|---|---|---|
| 1-in-20 year, GBM | −73.4% | −70.8% | −68.1% |
| 1-in-20 year, bootstrap | −94.1% | −91.1% | −88.0% |
| 1-yr 95% expected shortfall (GBM) | $187M | $180M | $174M |
| FTX-week replay (Nov 2022) | −24.1% | −23.4% | −22.8% |
| Oct-2025→Feb-2026 bear replay | −67.0% | −64.7% | −62.4% |
So a 10% mix is worth about $13–14M of 1-year tail relief on today’s pot, and about 2–5 points in a replayed crash. That’s real money.
But it doesn’t change the pot’s basic character, and the crisis-regime correlation run shows why: at 5–10% the mixed books have a diversification ratio of about 1.00, statistically almost the same animal as pure ADA, because BTC falls alongside ADA in exactly the scenarios the mix is meant for.
As for the question of size of the mix, the numbers can inform but not settle.
3. Conversion of $100M
The open $100M conversion idea proposed a year ago has a practical question attached that I haven’t seen priced anywhere: what does selling that much ADA cost?
Modeled as patient selling, capped at a fixed share of each day’s volume so the sale itself doesn’t crater the price:
| 10% of volume, normal market | 5% of volume, crisis conditions | |
|---|---|---|
| Days to convert $100M | 55 days | 330 days |
| Modeled price impact | $27M | $41M |
One caveat matters a lot here: those day-counts depend on which daily-volume figure you use for ADA. The table uses the strictest one there is: what actually trades on the single deepest ADA market, the ADA/USDT pair on Binance, about $24M a day. It deliberately ignores every other exchange, so read it as a floor. CoinGecko’s figure for the same days is $240–470M, roughly ten times more, and is best read as a ceiling. A real seller would reach more than one venue but not all of them, so the truth sits between the two. Take the ceiling and the whole job takes a tenth as long (a crisis conversion becomes ~33 days rather than 330). Two things hold on either basis: a $100M conversion is a managed program measured in weeks to months, not a treasury action that happens at a click, and it gets slower and more expensive precisely in the conditions where it would feel most urgent. The March thread’s tiered-allocation idea is one way to make it a scheduled program instead of an emergency one; this analysis prices the problem, not the solution.
What the numbers say together
A modest diversified mix of the 100% ADA position would soften the tails by low-tens-of-millions but wouldn’t change the pot’s basic character, and physically converting any large slice is itself a months-long market operation. None of that is a criticism of the design; it’s what a protocol-native treasury is. It does mean the three quantities worth tracking on a standing basis are the purchasing power of the annual budget at stated percentiles, the model-risk band on the pot’s yearly value, and the realistic conversion timeline at a stated volume basis.
Method and limitations
- Every headline figure is a model output under the stated assumptions. Different defensible parameter choices move the results materially. Treat the numbers as orders of magnitude.
- Treasury balance read from the chain at epoch 642 (Koios); prices from July 13 closes. The pot takes in 4–5M net ADA per epoch, which a snapshot simulation ignores; over a year that understates the ADA balance by roughly 25–30M (2%), in the treasury’s favor.
- Monte Carlo: 25,000 paths, 365 days, seed 42, reproducible. GBM uses zero drift; the bootstrap resamples the trailing year’s actual returns and therefore carries its downtrend. Neither is a forecast; the spread between them is reported as model risk on purpose.
- Books B and C are illustrative constructions (BTC+USDC halves), not any proposal’s actual specification. Cardano-native stablecoins were not used because their price-series identity in our data source couldn’t be confirmed; USDC stands in, labeled.
- Liquidity model: fixed participation of daily volume with venue-depth slippage, order books only, with Binance ADA/USDT volume (~$24M/day) as the participation base — the single-deepest-market floor, roughly 10× below CoinGecko’s all-venue figure; both bases are shown in section 3. OTC desks, auction mechanisms, and negotiated blocks are not modeled and would change the picture for a real program.
- Snapshot-based. No protocol-code audit. Not investment advice, and no position on any open governance action.