Skip to content

Roadmap

PortPy is built in stages. This page is the plain-language status; see Architecture for the full design document each stage is built from.

Stage 1 - Metrics & Core - done (v0.1.0, hardened in v0.1.2, cost/currency realism extended in v0.2.0)

  • portpy.metrics: returns, risk, performance, drawdowns, rolling/expanding windows, distributions, benchmark-relative, covariance/portfolio-level risk decomposition, one-shot summaries, transaction costs (a blended flat rate, plus itemized spread/commission/FX-spread/tax-drag estimators - see Costs & taxes).
  • portpy.core: asset-class tagging, calendar alignment, currency conversion (with an FX-spread estimator), weight normalization (including long/short books).
  • portpy.explain: the explainability layer.
  • Cross-validated against empyrical and quantstats on synthetic and real market data.

Stage 2 - Visualization - not started (deferred out of v0.2.0)

Plotly-first charts, one per metric family (price_chart, drawdown_chart, correlation_heatmap, monthly_returns_heatmap, risk_return_scatter, ...), plus a full dashboard() and a multi-portfolio comparison_dashboard(). Each chart recycles its underlying metric's/model's registered Explanation (fig.explain()) rather than registering its own - no chart ever registers a category="chart" explanation. Originally planned alongside Stage 3 for v0.2.0, but hasn't started yet and now ships in a later version.

Stage 3 - Portfolio Construction, Optimization & Management - done (v0.2.0)

  • Estimators: expected returns (mean-historical, EWMA, CAPM-implied, James-Stein shrinkage), covariance (sample, EWMA, shrinkage, Ledoit-Wolf, robust/MCD), factor models (CAPM, Fama-French 3/5-factor, linear/rolling regression, factor attribution).
  • Optimization: mean-variance, max Sharpe, min variance, target return/volatility, efficient frontier, Black-Litterman, hierarchical risk parity, risk parity/budgeting, maximum diversification - solved via multi-start SLSQP (scipy-only, no cvxpy).
  • Construction: constraint objects (weight bounds, group/sector caps, turnover caps, net/gross exposure), plus the optimize/build/efficient_frontier recipe layer.
  • Management: rebalance (trade lists), monitor (drift/limit checks), compare (before/after or portfolio-vs-portfolio).
  • Cross-validated against PyPortfolioOpt, Riskfolio-Lib, and skfolio (plus a plain-numpy OLS closed form for the factor regressions) on synthetic and real market data, including negative-weight/short-position solutions and a mixed-timeframe (equities + crypto) universe.

Stage 4 - Simulation & Forecasting Toolkit - not started

  • Stochastic processes: GBM, Ornstein-Uhlenbeck, Heston, jump-diffusion, CIR - calibration + simulation.
  • Monte Carlo: bootstrap resampling, parametric/historical MC VaR & CVaR, path statistics.
  • Time series: ARIMA, GARCH, cointegration, stationarity tests.
  • ML toolkit: PCA factor extraction, regime detection (HMM).

Doesn't block Stage 3 and shouldn't hold up shipping it - that's why the two were split into separate stages.

Stage 5 - Strategies - not started

A BaseStrategy interface, a backtest() engine, position sizing (Kelly, volatility targeting), and a library of concrete strategies: buy-and-hold, equal-weight, minimum variance, risk parity, time-series/cross-sectional/dual momentum, mean-reversion (z-score, Bollinger), pairs trading, moving-average crossover, breakout, volatility targeting, factor tilt, sector rotation, plus more advanced statistical strategies (Random Matrix Theory portfolios, regime-switching, PCA vector-space tracking, econophysical approaches) - and evaluation tooling (walk-forward analysis, Monte Carlo robustness).

Contributions and corrections toward any stage are welcome - open an issue on GitHub to discuss scope.