PortPy

Portfolio analysis that explains itself.

PortPy is a Python library for portfolio performance measurement and risk analysis, built around one object — Portfolio — and one idea: every number it gives you can explain, in plain language, what it is, how to read it, and whether it's good or bad.

import pandas as pd
from portpy import Portfolio

prices = pd.read_csv("prices.csv", index_col=0, parse_dates=True)  # columns = tickers

portfolio = Portfolio(
    prices,
    weights={"AAPL": 0.4, "MSFT": 0.35, "GOOGL": 0.25},
    name="Tech Portfolio",
    risk_free_rate=0.04,
)

portfolio.metrics.sharpe_ratio(as_result=True).explain()

sharpe_ratio (metric)
=====================

What it is:
  Measures excess return earned per unit of total volatility taken, using the risk-free rate as the return hurdle.

Formula:
  mean(r - rf) / std(r - rf, ddof=1) * sqrt(periods_per_year)

How to read it:
  A Sharpe of 1.0 means the strategy generated approximately one unit of excess return for each unit of volatility. Higher values indicate better risk-adjusted performance.

Good vs. bad:
  Higher is generally better. Values above 1 are commonly considered strong, but interpretation depends on the asset class, time period, and strategy complexity.

Caveats:
  Sharpe treats all volatility as bad, including upside volatility. It can also overstate strategies with asymmetric downside risk, illiquidity, or short backtests. The sqrt(periods_per_year) annualization assumes i.i.d., serially uncorrelated returns - positive autocorrelation (illiquid/infrequently-priced assets) means the annualized figure is overstated (Lo, 2002).

This result:
  very good

Why PortPy

• One object, one namespace. portfolio.metrics.<name>() auto-fills returns, weights, risk-free rate, and annualization frequency from the portfolio itself — no re-threading the same five arguments through every call.
• A real portfolio, not just a return stream. Weights (negative/short included), asset-class tags, and explicit calendar-alignment / currency-conversion helpers for combining assets that don't trade on the same schedule (crypto, equities, bonds) or in the same currency.
• Explains itself. .explain() on any result — a plain-language card covering what it is, how to read it, good vs. bad, and known caveats.
• Numerically validated. Cross-checked against empyrical and quantstats for metrics, and PyPortfolioOpt, Riskfolio-Lib, and skfolio for every optimizer — including negative-weight (short) solutions, not just long-only — on real market data, see tests/validation/.
• Builds portfolios, not just measures them. portfolio.models estimates expected returns/covariance, solves for weights (mean-variance, max Sharpe, risk parity, hierarchical risk parity, Black-Litterman, ...), and manages a book afterward (rebalancing, drift monitoring, before/after comparisons) — every result explains itself the same way metrics do.

Install

pip install portpy-quant               # core: numpy, pandas, scipy, statsmodels
pip install "portpy-quant[viz]"        # + plotly, matplotlib, seaborn
pip install "portpy-quant[data]"       # + yfinance, alpaca-py (for the examples)
pip install "portpy-quant[all]"        # everything

Documentation

• Getting Started
• User Guide — the Portfolio object, weights & shorts, calendar/currency alignment, the explainability layer
• API Reference — every function, by module
• Roadmap — what's implemented today vs. planned (visualization and strategies are not yet built)
• examples/ — runnable scripts, metrics_debug.ipynb/metrics_tutorial.ipynb exercising every metric against live Alpaca + Fed (FRED) data, and models_debug.ipynb/models_tutorial.ipynb doing the same for .models

Status

PortPy is pre-1.0 (Development Status :: 4 - Beta). Metrics and core (calendar/currency alignment, weights, the explainability layer) and .models (estimators, optimization, construction, management) are implemented and tested. Visualization and strategies/backtesting are designed but not yet built — see the Roadmap.

License

MIT © Arthur Faugeron