Methodology
How the Risk score works
Every asset on BlackNull gets two numbers from one published method: a Risk score that says how well it has paid for the risk taken, and a Risk level that says how much it can swing and fall. The AI explains them; it never sets them.
What the Risk score means
The Risk score runs from 0 to 100, and higher is better. It answers one question: over the last ten years, did this asset reward its holders for the ups and downs they lived through, compared with simply holding cash?
- 50 means it did about as well as cash, once the swings are taken into account.
- Above 50 means it beat cash by more than its swings and falls would suggest. The closer to 100, the more consistent that reward has been.
- Below 50 means the swings and falls were not paid for: cash would have done better for the risk.
The score is a record of the past, not a forecast. A high score does not mean an asset cannot fall; the Risk level says how far it can.
Inputs
Each score uses the total return of the underlying asset over the last ten years (or its whole history when it is shorter): price changes plus dividends or interest where they exist, from end-of-day data. A position or a vault is scored on what it holds, so trading costs and funding payments are left out of the score; trading costs count in the Risk level instead.
Returns are measured against cash: every day, the value is divided by what the same money would have earned in three-month US Treasury bills. A result of 0 therefore means "no better than cash", whatever the level of interest rates.
- Index products use a widely held fund that tracks the same index, with dividends reinvested. Sector baskets are rebuilt from their holdings, equally weighted.
- Single holdings use the stock, the metal or energy futures price, or the digital asset itself.
- Yield vaults use the vault's own value or yield history; liquid staking also includes the price of the staked asset.
The three measures
Three well-known ways of judging return for risk are averaged, so no single statistic decides the score.
1. Risk-adjusted return (risk aversion 2)
The monthly returns above cash are averaged the way a cautious investor feels them: a bad month counts for more than a good month of the same size. With a risk aversion of 2, an asset that swings a lot needs a clearly higher return to score the same as a steady one. The result is an annual return, in percent.
2. Downside-deviation ratio
The yearly growth above cash divided by the size of the losing months only (gains are not counted as risk). The size of the losses is floored at 5% a year, so very smooth histories cannot produce extreme ratios.
3. Drawdown depth-and-duration ratio
The yearly growth above cash divided by a measure of how deep and how long the falls from previous highs were (the root-mean-square of every day's distance below the last high). It rewards assets that recover quickly. The drawdown measure is floored at 2.
Formula and constants
Each measure is squashed into the range −1 to +1 with a hyperbolic tangent, so one extreme measure cannot dominate. The three are averaged and placed on the 0–100 scale around 50.
Example: cash itself has R, D and M of 0, so it scores 50.
Risk level
The Risk level describes how rough the ride can be, on seven steps. Each component of risk is placed on a scale from 1 (smallest) to 8 (largest); the overall class is the worst of them, because one weak point is enough to lose money. It is then turned into a level score, 100 × (8 − class) ÷ 7, and named with these bands:
| Level score | Risk level | Profile |
|---|---|---|
| 86 and above | Very low | Conservative |
| 72 to 85 | Low | Conservative |
| 58 to 71 | Medium-low | Conservative |
| 43 to 57 | Medium | Balanced |
| 29 to 42 | Medium-high | Balanced |
| 15 to 28 | High | Aggressive |
| Below 15 | Very high | Aggressive |
Components
- Volatility and tail loss. From five years of daily returns: the volatility a normal distribution would need to produce the same one-in-forty yearly loss, allowing for skew and fat tails, or the recent volatility when it is higher. Steps at 0.5%, 5%, 12%, 20%, 30% and 80% a year. It counts for 70% of the market component.
- Drawdown. The largest fall from a high over up to twenty years, on matching steps. It counts for 30% of the market component.
- Liquidity. The cost of buying and then selling $10,000 on the best market, with steps at 0.1%, 0.3%, 1%, 3%, 10% and 30%.
- Leverage. Borrowing raises the class by one step for every doubling of exposure.
- Trading venue and price feed. Where the position is held and how its price is set; a market with no reference price ranks worst.
- Issuer. Who stands behind a token, a stablecoin or a fund share, and whether that backing is independently rated.
- Smart contracts, curator and depeg. For vaults: the contracts, the manager who sets the strategy, a staking token drifting from its value, and withdrawals that can be gated when a lending reserve is fully used.
How they combine
- Worst-of: the class is the highest of the components.
- When two or more non-market components are at 5 or worse, the class rises by another half step.
- Under two years of price history, the market component rises by one step.
- A vault is never safer than what it holds: a vault that holds a digital asset is at least as risky as that asset on its own.
Bad day and bad year
Each asset page also shows two plain losses: a bad day, the daily loss exceeded on one day in twenty over the last five years, and a bad year, the yearly loss the volatility measure above expects once in forty years.
Provisional scores and shrinkage
A short history can look better or worse than it really is, so the score is only trusted fully with three years of data.
- Under three years: the score is pulled toward 50 in proportion to the history, score = 50 + (years ÷ 3) × (measured score − 50), and it is marked Provisional.
- Under one year: the score is also capped at 50.
- No usable history: the score is 50, Provisional, until there is.
Structural cap
Good past returns cannot outweigh a serious structural weakness. When any non-market component of the Risk level is at class 6 or worse (for example a token with no market history, or withdrawals that can be gated), the Risk score is capped at 50 and the page says why.
The curated list
BlackNull's curated lists (such as YIELD on Invest) only show assets that pass every one of these tests:
- Risk score of 60 or more;
- Risk level of Medium-high or lower;
- not Provisional;
- no structural cap.
Investor profiles follow the Risk level: Conservative up to Medium-low, Balanced for Medium and Medium-high, Aggressive for High and Very high.
Refresh and as-of date
Scores are recomputed by BlackNull's server whenever an asset's report is rebuilt: every day, and sooner when a vault's yield moves by one percentage point or a price moves by 1%. Until an asset's first report is built, the server uses a table computed with the same method from the same end-of-day market data. Every score carries the date of the data it was computed from.
The AI explains, it does not score
The Risk score and the Risk level are calculated by the formula on this page and nothing else. BlackNull's AI reads the numbers, the history and recent news, and explains them in plain words; it cannot raise or lower a score.
Limitations
- The past does not repeat exactly. Ten years may not include the kind of crisis that hurts an asset most.
- Total-return data can miss small costs, taxes and the exact timing of payments.
- Young assets and vaults have little history; their scores are Provisional and can change a lot.
- Structural risks such as contracts, issuers and venues are judged on fixed steps, not on a full audit.
- The score is not investment advice and does not know your goals. Read the risks on each asset page before investing.