Skip to main content

For AI agents: the complete documentation index is at llms.txt. Markdown versions of pages are available by appending .md to the URL or sending Accept: text/markdown.

Financial Machine Learning

Orion supports financial machine learning (FML) as one approach to portfolio construction and rebalancing. FML can simplify allocation decisions, inform systematic rebalancing, and automate intent generation under defined rules, though model output remains a candidate allocation, not a guarantee of future performance.

Orion enables two core approaches:

  • Passive portfolio construction, powered by DeFi-native indexing and onchain metrics;
  • Active, ML-driven management, where models inform allocation and rebalancing.

This dual system lets vaults range from hands-off, rules-based strategies to model-informed allocations within Orion's unified framework.

DeFi-Native Indexing​

Orion enables a marketplace for onchain indexes: passive portfolios based on predefined rulesets, rebalanced automatically, and tokenized as ERC-20s.

In this model:

  • Managers define index weights based on asset classes, themes, or rulesets;
  • Managers may use APIs from skfolio to define quantitative finance metrics (e.g., covariance, tail risk, train-test split) that inform strategist intents submitted to the Protocol;
  • Index logic may be referenced onchain via IPFS CIDs pointing to dockerized portfolio management logic, supporting design goals for verifiable index construction.
Example of a ruleset
 from skfolio import RiskMeasure
 from skfolio.optimization import MeanRisk, ObjectiveFunction

 X = get_universe_returns()
 model = MeanRisk(
    risk_measure=RiskMeasure.STANDARD_DEVIATION,
    objective_function=ObjectiveFunction.MAXIMIZE_RATIO,
    portfolio_params=dict(name="Max Sharpe"),
 )

 model.fit(X)
 model.weights_

This supports user-defined indexes based on financial machine learning rulesets, with methodology pinned for reproducibility.

Cards

Trust Guarantees​

Each step of this process is verifiable and reproducible:

  • Provenance: Methodology + data pinned on IPFS and CID pinned onchain;
  • Reproducibility: Anyone can rerun the pipeline and validate the output;
  • Audit Trail: All updates and rebalances are logged transparently onchain.

Active Management: ML-Optimized Vaults​

For managers seeking more sophisticated exposure, Orion supports financial machine learning (FML) pipelines that inform allocation, rebalancing, and dynamic optimization.

Common model families include:

  • Hierarchical risk parity;
  • Expected return models;
  • Online learning for portfolio allocation.

These models are either:

  • Hardcoded into the vault strategy, based on parameterized ML outputs, or
  • Queried dynamically via oracles or offchain computation layers (e.g., zkML, verifiable compute).

Orion abstracts this complexity for users, providing data, SDKs and tooling, while allowing managers and strategists to tune model parameters, override logic, or compose hybrid strategies.

Last updated on