The treasury question
Most on-chain businesses hold their treasury in one of four ways: their own token, ETH or BTC, spot stables, or stablecoin yield (Aave, Compound, similar). Each choice has a real cost over a full cycle, and the data is public.
The bear from late 2021 through 2022 is the most useful recent natural experiment. ETH drew down ~94% peak-to-trough; most 2021-era project tokens are still down 80–95% from their peaks; pure spot stables earned nothing while everything around them moved. Treasuries that didn’t survive that drawdown are the silent majority of the story.
This piece compares the dominant on-chain treasury patterns against a regime-based 3-state allocation across the full 8+ year window, and decomposes where the value comes from.
The seven comparison points
Daily-rebalanced backtests for the strategies on-chain businesses actually pick between. Numbers from a deterministic, open-source pipeline; sources, code, and live snapshot at /regime.
| Treasury approach | CAGR | Max DD | Sharpe | Smart-contract / token risk |
|---|---|---|---|---|
| Regime-based 3-state (ETH+SP500+cash) | +24.7% | -44% | 0.87 | None — spot positions only |
| 50/50 ETH+SP500 HODL | +22.1% | -73% | 0.67 | None |
| 100% ETH HODL | +13.8% | -94% | 0.58 | None |
| 100% SP500 HODL | +12.8% | -34% | 0.72 | None |
| 100% cash — 3-month T-bill | +2.6% | 0% | — | None |
| DeFi stable yield (Aave-style) | ~5–8%* | ~0%† | — | Smart-contract + depeg + protocol risk |
| Own / ICO token treasury | typically negative | −80 to −95% | — | Severe — concentrated, illiquid |
* DeFi stable yields are time-varying and not backtested here (lending rates and protocol health change across cycles). † Excludes catastrophic events (depegs, exploits) which have caused major drawdowns in this category.
8 years on a single chart
Same starting capital ($1), same end date (May 2026), log scale so the doublings are visually comparable.
What the chart shows
Three things are visible:
- ETH HODL took a 94% drawdown. For a business treasury that needs to make payroll, that is a non-survivable event. The line eventually recovers, but the practical question is whether the company is still operating when it does.
- Spot stables yielded +2.6%. Zero drawdown is great; almost zero growth is a real opportunity cost over an 8-year window where the alternative grew 5–10× depending on which one was picked.
- The regime-based 3-state composite reached 6.12× with a -44% max drawdown. Better CAGR than any HODL baseline and roughly half the worst drawdown of the 50/50 blend.
Where the returns come from
The composite’s outperformance decomposes cleanly into two stacked effects, each valuable in its own way.
The green band — diversification — is the steady-state engine. Owning a balanced portfolio captures the long-term return premium of risky assets at a much lower volatility than concentrating in any single one. The band widens continuously across the full window, including through 2022, because two imperfectly-correlated assets average out each other’s drawdowns.
The cyan band — regime overlay — is the tail-event protection. It expands when the classifier moves the portfolio to cash ahead of and during the 2022 drawdown, locking in capital that compounds forward. Both effects are real value sources for a treasury, and both contribute meaningfully to the final result.
The attribution, in numbers
The same decomposition as a 2×2 table. Read across rows for the diversification effect (100% ETH → 50/50 blend); read down columns for the regime overlay effect (HODL → composite).
| No regime (HODL) | With regime (Composite) | Δ from regime | |
|---|---|---|---|
| 100% ETH | +13.8% / 0.58 / -94% | +31.4% / 0.77 / -63% | +17.6pp CAGR · +0.19 Sharpe · +30pp DD |
| 50/50 ETH+SP500 | +22.1% / 0.67 / -73% | +24.7% / 0.87 / -44% | +2.6pp CAGR · +0.21 Sharpe · +29pp DD |
| Δ from diversification | +8.3pp CAGR · +0.09 Sharpe · +20pp DD | -6.7pp CAGR · +0.10 Sharpe · +19pp DD | — |
Each cell: CAGR / Sharpe / Max drawdown. Both effects contribute to Sharpe and drawdown protection meaningfully. Diversification adds the larger CAGR step in the path from ETH to the composite; the regime overlay adds the larger Sharpe step and a deep drawdown cut.
Performance through each cycle phase
The 8-year aggregate makes the headline easy to read. The phase-by-phase view shows when each strategy was earning its return, which matters for a treasury that operates through every part of a cycle, not just the average.
| Phase | Composite | 50/50 HODL | 100% ETH HODL | Cash |
|---|---|---|---|---|
Pre-2020 + COVID Feb 18 → Dec 20 | +30.3% / -35% | +14.4% / -60% | -5.1% / -87% | +1.4% / 0% |
2021 mania Dec 20 → Nov 21 | +247.9% / -8% | +251.4% / -8% | +645.5% / -18% | +0.0% / 0% |
2022 drawdown Nov 21 → Dec 22 | -17.9% / -21% | -46.7% / -54% | -71.3% / -77% | +1.8% / 0% |
Recovery Dec 22 → Jan 24 | +8.9% / -0% | +54.1% / -8% | +81.4% / -15% | +5.1% / 0% |
Recent Feb 24 → today | +1.2% / -29% | +14.8% / -31% | -0.5% / -55% | +4.3% / 0% |
Each cell: annualized CAGR over the phase / max drawdown within the phase. Phase boundaries are calendar-defined.
The composite delivers its largest separation during the 2022 drawdown, where it holds capital that the HODL strategies are forced to give back. Through the 2021 mania the composite keeps pace with the blend. In benign phases (Recovery, Recent), the diversified HODL captures more upside while the composite holds defensive cash positions and earns the T-bill yield on that portion. The combination across all phases is what produces the 8-year outperformance.
Considerations
- Sample window. 2018–2026 covers roughly two full crypto cycles. A larger sample would tighten confidence on the magnitudes.
- Backtest, not deployment. Real execution introduces slippage, gas costs, custody overhead, and behavioural drift. Realized performance will differ from backtested.
- In-sample / out-of-sample. The in-sample window (through Jan 2024) contains the 2022 drawdown; the out-of-sample window (Feb 2024 onward) has been comparatively benign. The composite’s value is concentrated in periods that contain tail events.
- Methodology is fixed, not tuned. Thresholds (0.33 / 0.67 percentile bucket, 5-day confirmation, 2σ fast-track) are chosen from published convention and from the data’s own noise distribution, not from backtest CAGR.
What this is
A public research artifact: open data, open code, deterministic computation, one-page methodology. Not a fund, a managed account, a token offering, a financial product, or a recommendation. The full mechanics — indicators, weights, regime smoothing, correlation analysis — are inspectable at /regime. Source code and CSV history are in the public repository linked from the same page.