Risk Methodology¶

Version¶

Current methodology version:

crypto-scout-canonical-v1.1

The mini-product methodology is aligned with the canonical risk model from crypto-scout-analytics for the implemented BTC risk surface.

Input Rows¶

Each daily input row must contain:

date
open
high
low
close
volume
market cap
circulating supply

Prices, market cap, and circulating supply must be positive. Rows are sorted by date and duplicate dates are rejected.

Feature Construction¶

HLC3 Price¶

The model price is:

HLC3 = (high + low + close) / 3

When the model price is displayed in the product or API docs, label it as Model price or HLC3 rather than a generic spot or close price. If daily high and low are displayed alongside it, High and Low refer to the same completed daily candle used to compute HLC3.

Trend Deviation¶

Trend deviation uses a 365-day EMA:

trend_dev = ln(HLC3 / EMA365(HLC3))

Volatility Regime¶

Volatility is the rolling population standard deviation of daily log returns:

log_return[t] = ln(HLC3[t] / HLC3[t-1])
vol_regime = rolling_std(log_return, 30 days)

Turnover¶

Turnover is enabled when valid market cap is available across the source rows:

turnover = ln(volume / market_cap)

Robust Z-Scores¶

Each feature is converted to a robust rolling z-score with:

window: 1460 days
minimum periods: 365 days
center: median
scale: 1.4826 * MAD + 1e-12
clipping: [-6.0, 6.0]

When a window has fewer than 365 valid observations, the z-score is 0.0.

Score Weights¶

When turnover is enabled:

Feature	Weight
trend deviation	`0.60`
volatility regime	`0.25`
turnover	`0.15`

When turnover is disabled:

Feature	Weight
trend deviation	`0.70`
volatility regime	`0.30`

Risk Value¶

The score is mapped to risk with a sigmoid:

risk = sigmoid(score)

The stored risk value is always expected to be in [0.0, 1.0].

Risk States¶

Range	State
`< 0.30`	`low`
`>= 0.30` and `< 0.70`	`neutral`
`>= 0.70`	`high`

Risk Levels¶

/api/risk/levels solves hypothetical BTC prices for target risk values from 0.0 to 1.0 in 0.025 increments.

The solver:

builds a context from the full OHLCV history;
keeps non-price components fixed for the latest day;
solves prices that would move the trend component to each target risk;
verifies target prices through the same risk model.

The latest risk from /api/risk/latest and meta.current_risk from /api/risk/levels are expected to match when the database is consistent.

The /api/risk/levels response keeps the full 0.00 to 1.00 scenario ladder. The public frontend chart displays the practical 0.20 to 0.80 window so clipped solver endpoints remain available for diagnostics without stretching the user-facing price scale.

Interpretation Limits¶

The risk metric and risk levels are analytics outputs. They are not financial advice, investment advice, or trading recommendations. Risk levels are scenario estimates based on the current methodology and input data quality.