Skip to content

Factor Risk Model Boundary

Language: English · 简体中文

quant-market-research maintains a lightweight, framework-neutral factor-risk capability. It converts precomputed factor-return and asset-specific-return histories into a point-in-time risk estimate.

Model

For exposure matrix X, factor covariance F, and diagonal specific variance D, asset covariance is projected as:

Sigma_asset = X F X' + D

The implementation does not estimate factor returns from stock returns. Factor construction is a research assumption supplied explicitly by the caller. This function validates and summarizes historical inputs already produced by the research layer.

Inputs

  • Current or as-of asset-by-factor exposures
  • Historical factor returns
  • Historical asset-specific returns
  • An explicit as_of timestamp
  • A covariance estimate with shrinkage toward the factor-variance diagonal
  • A minimum number of common observations

Every return observation must be on or before as_of; future rows are rejected.

Output

FactorRiskModelEstimate contains:

  • Factor covariance
  • Specific risk for each asset
  • Asset exposures
  • History start/end and observation count
  • Estimator configuration
  • An asset_covariance() projection
  • Versioned receipt metadata

The covariance projection uses specific variance, computed by squaring the stored specific-risk values.

Future work

  • Evaluate industry/style factor sets and estimation windows.
  • Add estimator comparisons and stability diagnostics.
  • Publish risk-model evidence through research artifacts.
  • Integrate an authoritative estimate into portfolio_backtester optimization requests.
  • Compare against licensed RQData risk-model outputs when authorized data is available.

Third-party risk-model objects must remain inside adapters. Promotion evidence must distinguish estimator quality from optimizer quality.