Forecast Distribution API — Full Probabilistic Ranges

Q: Can I use the p05 percentile as a VaR estimate?

The p05 final return can serve as a model-conditional tail estimate at the endpoint's horizon. For example, if p05 final return is -3%, the model assigns ~5% probability to a loss exceeding 3% at that horizon. This is not a standard VaR figure — it lacks portfolio context, return convention, and conditioning. Use it as an input to your own risk framework, not as a standalone VaR number.

Q: How much does the forecast distribution API cost?

$0.01 per timeframe, capped at $0.04 for all 5 (20% off). Upfront payment — settled before the handler runs. Pay with USDC on Solana or Base via x402. Pair with /predict for $0.01 per timeframe to get both direction and full distribution.

The /forecast-distribution endpoint returns the full sample distribution from Kronos probabilistic inference — percentiles at each prediction step, plus the final return distribution.

Endpoint

GET /api/feeds/kronos/forecast-distribution/:symbolKey?timeframes=5m,1h

Parameters

Param	Type	Required	Description
`symbolKey`	path	Yes	Asset symbol, e.g. `btc_usdt`
`timeframes`	query	No	Comma-separated (default: `5m`)

Pricing

$0.01 per timeframe, capped at $0.04 (20% off when requesting all 5). Upfront payment — settled before the handler runs.

Example response (truncated)

{
  "symbol": "btc_usdt",
  "timeframe": "1h",
  "steps": [
    {
      "step": 1,
      "p05": 66800, "p10": 67000, "p25": 67300,
      "p50": 67600, "p75": 67900, "p90": 68200, "p95": 68500
    },
    {
      "step": 2,
      "p05": 66500, "p10": 66800, "p25": 67200,
      "p50": 67700, "p75": 68200, "p90": 68700, "p95": 69100
    }
  ],
  "final_return_distribution": {
    "p05": -0.03, "p50": 0.012, "p95": 0.058
  }
}

Use cases

Risk Management

Use the p05/p95 percentiles as inputs to your own risk model. The distribution is model-conditional (Kronos forecast at a given horizon) — combine with your portfolio context and conventions before deriving risk metrics.

Options Analysis

The percentile spreads across timeframes reflect the model's uncertainty at each horizon. These can inform your options analysis, but are not a direct substitute for market-implied volatility surfaces from option prices.

Scenario Analysis

Walk through best/worst/median case price paths. Each sample path is a plausible future — stress-test your portfolio against hundreds of them.

Portfolio Optimization

Distribution-aware position sizing — size positions based on tail risk, not just expected return. The full distribution lets you optimize for CVaR, not just variance.

Frequently asked questions

What is a forecast distribution API for crypto?

It returns the full probabilistic sample distribution from Kronos inference — p05, p10, p25, p50, p75, p90, p95 percentiles at each prediction step, plus the final return distribution. Useful for tail risk estimation, scenario analysis, and distribution-aware modeling.

Can I use the p05 percentile as a VaR estimate?

The p05 final return can serve as a model-conditional tail estimate at the endpoint's horizon. For example, if p05 final return is -3%, the model assigns ~5% probability to a loss exceeding 3% at that horizon. This is not a standard VaR figure — it lacks portfolio context, return convention, and conditioning. Use it as an input to your own risk framework, not as a standalone VaR number.

Can I extract implied volatility from the forecast distribution?

The percentile spreads reflect the model's forecast uncertainty at each horizon. They can inform options analysis, but they are not option-implied volatility — IV is derived from market option prices, not from a forecast model's internal distribution. Use the distribution as a complementary input alongside market IV.

How much does the forecast distribution API cost?

$0.01 per timeframe, capped at $0.04 for all 5 (20% off). Upfront payment — settled before the handler runs. Pay with USDC on Solana or Base via x402. Pair with /predict for $0.01 per timeframe to get both direction and full distribution.