Compound DAO Treasury — Independent Risk Teardown

Everything below is built from public data: the DAO’s own April consolidation figures, the TMC’s published contract addresses, and balances we read on-chain on July 15 (data basis and limitations at the end). If anything in the data basis is wrong, say so and it will be corrected. With the manager evaluation window opening this week, independent numbers on the book seemed like useful input.

TL;DR (scenario losses are model estimates, not predictions; the on-chain findings are limited to the published addresses and snapshot described below; none of this is advice):

1. We could verify $56.5M of treasury assets on-chain today, against the April accounting’s $77.5M identified ($120M headline). Most of the gap is documented: $11.65M Elixir money that was never received, the $14M V4 tranche already paid out, and roughly $2M of vendor funding. About $11.8M (the $7.98M DAI line and the $3.8M Avantgarde mandate) we could not locate at any published address, which says something about the public map rather than about the funds.
2. Seventy percent of the verified book lives inside Compound’s own stack. The $28.2M managed vault turns out to hold its assets as deposits in Compound’s v3 USDC and WETH markets, on top of the $11.6M of v3 reserves. In the modeled protocol event (temporary freeze, 80% recovery, 20% haircut on the affected portion), that combined $39.6M dependency produces an estimated loss of $14.3M, a quarter of the treasury.
3. The mandate book the RFP describes is calmer than the treasury holding it. The $22.5M envelope at the published 70/30 allocation shows a simulated 1-in-20 year between −16% and −22% depending on model choice; the full treasury book sits at −22% to −30%.
4. The binding single-event stress is the crypto side, not the stablecoin side. A −30% crypto month costs the mandate book $2.0M against $1.5M for an SVB-scale depeg weekend; on the treasury book it is $6.6M against $3.4M. Neither approaches the 1-in-20-year numbers above.
5. Under crisis liquidity assumptions (5% of daily volume, order books only), $47.4M of the $56.5M exits to cash within a week. The slow assets are COMP (~230 days on our volume data) and WBTC (~175 days) — with an important caveat on WBTC below.
6. Crisis correlation stress does not materially change modeled VaR for either book (the change is less than 0.01%; details in §2). Stablecoins cannot decorrelate further, so the tail risk here is composition, not correlation, and the numbers in point 3 already contain it.

Snapshot date: 2026-07-15 (on-chain balances read at block ~25,539,543, ~17:20 UTC, managed-vault sweep ~19:50 UTC; prices at July 15 daily closes) Data basis: the DAO’s published April consolidation (comp.xyz), the TMC’s published contract addresses, direct RPC balance reads, and the governance record for address discovery Covered value: $56,477,660 across 24 positions at 7 venues — approximately 86% of the April identified book after removing the never-received Elixir claim and the documented V4 outflows

Backdrop

Three things make this month a reasonable time to put numbers on the Compound treasury:

1. The treasury manager RFP closed applications on June 24, and per the published timeline the TMC’s evaluation and recommendation window is July 15–22. The mandate hands an initial $20–25M to an external manager at a specified allocation. Whoever wins, the allocation itself has a risk profile, and nobody has published it.
2. In April, a delegate asked in the TMC thread for “a bit more about the composition and risk profile” of the consolidated treasury. Three months later that question has no quantitative answer in the thread.
3. The rsETH episode demonstrated in April how a single collateral incident propagates into reserves and market operations. To state it precisely: no realized-loss figure for Compound exists anywhere in the public record as far as we can find, and Gauntlet reported in April that WETH and wstETH depositors had incurred no losses to date. The open unwind positions on the WETH market were sized at roughly $12.1M and $6.0M of debt in Gauntlet’s July 8 update. We model no rsETH outcome here; the episode matters because it shows which way stress travels through the system.

Two books were run through the same scenario set: the treasury as we could verify it on-chain (Book T, $56.5M), and the mandate book as the RFP FAQ specifies it (Book R: $22.5M at 70% stablecoins, 18% WBTC, 12% WETH). Headline figures are model outputs under stated parameter choices; where a different defensible choice moves a number materially, that sensitivity is noted inline.

1 · What the treasury actually is (and what we couldn’t find)

Reconstructing the April consolidation on-chain today:

The verified book: 61.4% stablecoins (by design — the composition itself is not a finding), 12.2% WBTC, 12.2% ETH, 10.1% WETH, 3.1% COMP, and small wstETH/USDS positions. Two observations that are findings:

• The treasury holds almost no COMP (about $1.8M, mostly in one operations Safe). The old picture of a DAO treasury dominated by its own token is exactly backwards here; at today’s prices the treasury is on the order of 70% of COMP’s entire market capitalization. Treasury risk at Compound is protocol-survival math, not token-price math.
• Ownership state, described without interpretation: at the snapshot, the $28.2M vault’s owner is still the Governor Timelock, with no pending transfer; the smaller vendor-payments vault from Proposal 580 shows a transfer to the Treasury Timelock initiated but not yet accepted. This may simply reflect scheduled timelock actions we cannot see. We note it because the RFP’s initial envelope is defined by what the TMC actually controls at program start, and corrections here are particularly welcome.

2 · The mandate book, priced before the manager is picked

The RFP FAQ specifies the initial envelope: $20–25M, currently about 70% stablecoins (USDC and USDT), with the non-stable remainder around 60% WBTC and 40% WETH. We built the midpoint book — $22.5M as 60% USDC, 10% USDT (the FAQ gives no ratio; we mirrored the disclosed balances, which are heavily USDC), 18% WBTC, 12% WETH — and ran it through the same scenarios as the treasury. Results:

On the two simulation rows: the model-based run answers “what does this volatility structure do with no view on direction”; the bootstrap answers “what if next year statistically resembles the last one,” which was a bear year for the volatile third of these books. Neither is a forecast; the band between them is model risk, and it is the honest answer to any single-number VaR question.

The depeg parameters are choices, not predictions (USDC’s March 2023 close was about −10%). What the comparison shows is proportion: the volatile 30% of the mandate book carries the larger single-event loss and most of the simulated tail, despite the stablecoins being more than twice its size. Applicant stress answers about that 30% will tell the evaluation more than anything about the stablecoin side.

We also applied crisis-regime correlations to both books, and report it as a non-result: VaR moves by less than 0.01% (baseline diversification ratios 1.08 and 1.10). There is nothing left to decorrelate — the stablecoins already sit near zero correlation to the volatile side, and WBTC/WETH/ETH already move together. Whatever tail these books have is in the table above, in what they hold.

3 · The self-insurance question

Part of this treasury is the protocol’s own shock absorber, and a larger part than the reserve line suggests. The v3 reserves ($11.6M) exist to absorb bad debt. The $28.2M managed vault holds its USDC and WETH as deposits in the same v3 markets those reserves protect. A protocol incident severe enough to consume reserves is the same event that could freeze or impair the vault’s deposits.

Modeled as a temporary freeze with 80% eventual recovery and a 20% haircut on the affected portion:

The severity assumptions are deliberately moderate and the loss scales roughly linearly with them. The point is not that Compound’s markets are unsafe — the DAO knows its own protocol best. The point is that “diversify the treasury” and “deposit the treasury into our own markets” pull against each other, and the table sizes the trade: each 10 percentage points of the book moved outside the Compound stack reduces the modeled combined event by about $2M. Whether that trade is worth making against the yield given up is a judgment for COMP holders, not for us. It is also, concretely, a question the RFP evaluation could put to applicants, since the winning manager inherits exactly this structure.

At April’s published VaR floors, today’s USDC-market reserves sit only about $0.3M above the 2× buffer — the April withdrawal already took most of the excess. The reserve line in this book is mostly floor, not headroom.

4 · Exit liquidity in two regimes

Modeled as disciplined liquidation, fixed participation in daily volume with venue-depth slippage:

Two disclosures matter more than the day-counts. First, our volume data is a filtered venue set; it shows roughly 288 WBTC per day, while aggregate tapes across all venues show many times that. Take the broader tape and the WBTC day-counts shrink proportionally. Second, WBTC has a redemption path that order-book models ignore entirely: merchants redeem WBTC for BTC roughly 1:1, so “days to exit WBTC through order books” overstates real-world exit time if the DAO or its manager can access the merchant-redemption route. Read the WBTC and COMP day-counts as the cost of disciplined market exits; redemption shortens the WBTC one in practice, and nothing in the treasury’s current obligations forces exits on that clock anyway.

The stablecoin depth is the useful half of this table: in the crisis parameterization the USDC alone covers the RFP’s full initial envelope within a day.

5 · What the scenarios agree on

Run independently, the scenario families put the treasury’s stress losses in a consistent band:

1. Single-event stresses (depeg weekend, crypto month, protocol event) span $3.4M to $14.3M, with the Compound-stack dependency the largest of them by a factor of two.
2. The simulated bad year lands between −$12.4M and −$16.9M depending on model choice, and a >20% intra-year drawdown occurs in 51–79% of simulated paths.
3. The bear-market replay of the last nine months comes to −$11.1M on today’s composition, comfortably inside the simulated band.
4. Liquidity does not compound the problem: most of the book can leave through markets inside a week even under crisis assumptions, WBTC has a redemption path outside order books, and no current obligation forces the slow positions out on a deadline.

None of this says the current structure is wrong. A lending protocol’s treasury earning yield in its own audited markets is a coherent choice, and the reserve floor policy shows the DAO already thinks in these terms. What the numbers describe is a treasury whose dominant risk is a fat single dependency on its own protocol stack, held alongside a mandate process that is about to move $20–25M of it under external management. If the community finds these estimates useful, the two quantities worth tracking on a standing basis are the Compound-stack share of treasury assets (today: ~70%) and the crisis-liquid coverage of near-term obligations (today: ~$47M within a week, against whatever the V4 program and vendor streams draw).

Method & limitations

• Every headline figure is a model output under stated parameters (participation rates, shock sizes, recovery and haircut assumptions, covariance windows). Different defensible choices move results materially; treat the numbers as orders of magnitude.
• Public on-chain data only, read July 15: balances at the TMC’s published addresses, the two managed vaults identified through the governance record, and current v3 reserves read from the market contracts. ~86% coverage of the April identified book ex-Elixir. The unlocated $11.8M (DAI line, Avantgarde) and any off-chain or unpublished addresses are excluded; if those are stablecoin-heavy, our stability findings are slightly understated.
• The mandate book’s USDC/USDT split (60/10) is our assumption; the RFP names both without a ratio.
• Simulations: 25,000 paths, one-year horizon, seeded and reproducible; a zero-drift model-based run and an empirical bootstrap of the trailing 365 days. Distributional estimates, not forecasts. The 2021-crash replay was run but is reported only in passing: several current holdings lack 2021 price history and were beta-approximated, so we treat it as directional (−6.5%).
• Liquidity model: fixed participation of daily volume with venue-depth slippage, exchange order books only. It ignores WBTC merchant redemption and OTC, both of which materially improve real exits at treasury scale. Volume data for WETH and USDT uses aggregated market statistics rather than order-flow data.
• Protocol-event scenario: freeze with 80% recovery and 20% haircut on the affected portion is a parameter choice; no code audit, no probability estimate, and no claim about Compound’s actual security is implied.
• The two ownership observations in §1 are point-in-time reads of owner()/pendingOwner() and may not reflect scheduled governance actions. Corrections to any of the data basis are welcome and will be incorporated.
• Nothing here is investment advice, and we take no position on the RFP, the manager candidates, or any governance vote.