GitHub

Architecture

This page provides a high-level overview of the OnlinePortfolioAnalytics.jl architecture, showing how data flows through the system and how components are organized.

High-Level Architecture

The diagram below shows the main components of OnlinePortfolioAnalytics and how they interact:

flowchart TB subgraph Input["Data Sources"] PRICES[Price Stream] RETURNS[Return Stream] PAIRED[Asset/Benchmark Pairs] end subgraph Core["OnlinePortfolioAnalytics Core"] direction TB BASE[PortfolioAnalytics Abstract Type] SINGLE[PortfolioAnalyticsSingleOutput] MULTI[PortfolioAnalyticsMultiOutput] BASE --> SINGLE BASE --> MULTI end subgraph Metrics["Portfolio Metrics"] direction LR RET[Return Metrics] RISK[Risk Metrics] CAPM[CAPM Metrics] RATIO[Risk-Adjusted Ratios] CAPTURE[Capture Ratios] ROLLING[Rolling Windows] end subgraph Integration["Integrations"] TABLES[Tables.jl] DF[DataFrames] TSF[TSFrames] ROCKET[Rocket.jl Reactive] end Input --> Core Core --> Metrics Metrics --> Integration TABLES --> DF TABLES --> TSF

Component Overview

  • Data Sources: The package accepts three types of input data:

    • Price Stream: Raw asset prices that can be converted to returns
    • Return Stream: Pre-calculated periodic returns
    • Asset/Benchmark Pairs: Paired observations for relative metrics (e.g., Beta, Tracking Error)

  • Core Types: All analytics inherit from the PortfolioAnalytics abstract type:

    • PortfolioAnalyticsSingleOutput: Metrics that return a single scalar value (e.g., Sharpe ratio)
    • PortfolioAnalyticsMultiOutput: Metrics that return a NamedTuple (e.g., AssetReturnMoments)

  • Portfolio Metrics: The package provides 40+ metrics organized into categories:

    • Return calculations, volatility, drawdown analysis
    • Risk-adjusted ratios, CAPM metrics, relative performance
    • Market capture ratios, risk metrics, rolling windows

  • Integrations: Seamless integration with the Julia data ecosystem:

    • Tables.jl: Works with any Tables.jl-compatible data source
    • DataFrames: Direct support for DataFrame operations
    • TSFrames: Time series support
    • Rocket.jl: Reactive stream processing

Key Design Principles

  1. Online Algorithms: All metrics process data incrementally using the fit! function, enabling constant-memory operation on unlimited data streams.

  2. Composability: Complex metrics are built by composing simpler online statistics (e.g., Sharpe ratio composes Mean and StdDev).

  3. Parallelizability: The merge! operation supports distributed computation by combining statistics from different data partitions.

  4. Type Safety: Metrics are parameterized by their input type T, ensuring type stability and optimal performance.

See the Internals page for detailed diagrams of the type hierarchy and data flow patterns.