English · 中文 · Documentation · Contributing · Changelog · Migration Guide
fincore is a Python library for quantitative finance analytics — 150+ financial metrics, portfolio optimization, Monte Carlo simulation, and performance attribution. It continues the empyrical stack under active maintenance by cloudQuant.
Current version: 0.3.0 (Beta). Python 3.11+ is required; this is a documented breaking change relative to empyrical, which supports older interpreters.
fincore 0.3.0 exposes clearly separated surfaces. One name does not silently switch semantics between them:
| Surface | What it is | Guarantee |
|---|---|---|
Strict compatibility — fincore.empyrical |
Frozen empyrical 0.6.0 surface: 54/54 public symbols (C0), 49/49 callables (C1), core callables numerically verified (C3) | Pinned by tests/compat/fixtures/ manifests and enforced by the tests/compat/ gates |
pyfolio façade — fincore.pyfolio |
Frozen pyfolio 0.9.6 profile of 11 tear-sheet workflows: all entries C1, risk/returns/perf-attrib/full-sheet main chains C4 | Requires the fincore[pyfolio] extra |
Enhanced semantics — fincore.metrics, flat API, AnalysisContext |
fincore's own, documented divergences (e.g. week_year="iso", explicit validation exceptions) |
Recommended API; enhanced, not empyrical-identical |
Alphalens migration — fincore.alphalens / fincore.factor_analysis |
A source-shaped strict façade and a separate enhanced prepare/analyze/render workflow | Beta integration; use the tested APIs documented in the migration guide, not a top-level alphalens import |
See the compatibility matrix, empyrical matrix, and pyfolio profile.
| Feature | Description |
|---|---|
| 150+ Metrics | Returns, risk, drawdown, alpha/beta, capture ratios, timing, streaks |
| AnalysisContext | fincore.analyze() — lazy, cached computation with JSON/HTML export |
| RollingEngine | Batch rolling metrics (sharpe, volatility, max_drawdown, beta) in one call |
| Pluggable Viz | Matplotlib, HTML, Plotly, Bokeh backends via VizBackend protocol |
| Portfolio Optimization | Efficient frontier, risk parity, constrained optimization |
| Monte Carlo | Bootstrap, scenario testing, path simulation |
| Performance Attribution | Brinson, Fama-French, style analysis |
| Lazy Imports | Heavy dependencies are loaded on first use |
| PEP 561 | py.typed marker for type checker support |
pip install fincore # Core metrics
pip install "fincore[pyfolio]" # + Pyfolio tear sheets (matplotlib, seaborn, ipython)
pip install "fincore[factor-analysis]" # + Compute-only enhanced factor analysis
pip install "fincore[alphalens]" # + Factor-analysis rendering and strict Alphalens migration APIs
pip install "fincore[interactive]" # + Plotly, Bokeh backends
pip install "fincore[report-pdf]" # + Playwright PDF rendering
pip install "fincore[report-xlsx]" # + XLSX report export
pip install "fincore[bayesian]" # + Bayesian tear sheets (pymc)
pip install "fincore[data-yahoo]" # + Yahoo Finance provider
pip install "fincore[data-pandas-datareader]" # + pandas-datareader provider
pip install "fincore[data-alphavantage]" # + Alpha Vantage provider
pip install "fincore[data-cn]" # + Tushare, AkShare providers
pip install "fincore[all]" # Everything above
pip install "fincore[dev]" # Development toolsdatareader and viz are 0.3.x compatibility aliases for the functional extras above.
From source:
git clone https://github.com/cloudQuant/fincore # International
git clone https://gitee.com/yunjinqi/fincore # China mirror
cd fincore && pip install -e ".[dev,viz]"import fincore
import pandas as pd
# A self-contained return series
returns = pd.Series([0.01, -0.005, 0.002, 0.004])
print(f"Sharpe: {fincore.sharpe_ratio(returns):.4f}")
print(f"Max DD: {fincore.max_drawdown(returns):.4f}")The strict empyrical module is available directly:
from fincore import empyrical
print(empyrical.sharpe_ratio(returns))
print(empyrical.max_drawdown(returns))AnalysisContext (recommended stateful API — lazy, cached, exportable):
index = pd.date_range("2024-01-02", periods=5, freq="B")
returns = pd.Series([0.01, -0.005, 0.002, 0.004, -0.001], index=index)
benchmark = pd.Series([0.008, -0.003, 0.001, 0.002, 0.0], index=index)
ctx = fincore.analyze(returns, factor_returns=benchmark)
print(ctx.sharpe_ratio, ctx.max_drawdown)
ctx.to_json(path="report.json") # write files
ctx.to_html(path="report.html")
ctx.plot(backend="matplotlib") # -> ReportArtifactsPyfolio main chain (from fincore import Pyfolio requires the pyfolio extra):
from fincore import Pyfolio
pyfolio = Pyfolio(returns=returns, benchmark_rets=benchmark)
pyfolio.create_returns_tear_sheet(returns, benchmark_rets=benchmark)RollingEngine (batch rolling metrics):
import numpy as np
from fincore.core.engine import RollingEngine
rng = np.random.default_rng(7)
index = pd.date_range("2024-01-02", periods=60, freq="B")
returns = pd.Series(rng.normal(0.001, 0.02, 60), index=index)
benchmark = pd.Series(rng.normal(0.0005, 0.015, 60), index=index)
engine = RollingEngine(returns, factor_returns=benchmark, window=30)
results = engine.compute(['sharpe', 'volatility', 'max_drawdown', 'beta'])Portfolio Optimization:
from fincore.optimization import efficient_frontier, risk_parity, optimize
returns_df = pd.DataFrame(
{
"asset_a": [0.01, -0.005, 0.004, 0.002],
"asset_b": [0.003, 0.002, -0.001, 0.005],
}
)
ef = efficient_frontier(returns_df, n_points=5)
rp = risk_parity(returns_df)
w = optimize(returns_df, objective="max_sharpe")Every Python code block above is executed with the same data and arguments
by a matching test in tests/docs/test_examples.py.
The repository includes an executable, deterministic migration example. It builds local synthetic data with a fixed seed, uses no network, writes no output files, renders only with Agg, and closes its returned figures:
pip install "fincore[alphalens]"
MPLBACKEND=Agg python examples/factor_analysis_quickstart.pyUse fincore.alphalens for source-shaped strict calls and
fincore.factor_analysis for new prepare/analyze/render workflows. Do not use
import alphalens: fincore intentionally does not install or expose a
top-level standalone-compatible package. See the migration guide
for the API map and limitations.
fincore/
├── __init__.py # Lazy exports (Empyrical, Pyfolio, analyze)
├── empyrical.py # Strict empyrical 0.6.0 compatibility facade
├── pyfolio.py # Pyfolio 0.9.6-profile workflow facade
├── core/
│ ├── context.py # AnalysisContext — lazy cached metrics
│ └── engine.py # RollingEngine — batch rolling metrics
├── metrics/ # 17 metric modules (returns, risk, ratios, ...)
├── viz/ # VizBackend protocol + backends
├── optimization/ # Efficient frontier, risk parity
├── simulation/ # Monte Carlo, bootstrap
├── attribution/ # Brinson, Fama-French, style analysis
├── risk/ # EVT, GARCH
├── report/ # HTML/PDF report generation
├── tearsheets/ # Pyfolio-style plotting
└── utils/ # Shared helpers
Quality numbers are machine-generated, never hand-written into this README.
The current snapshot is docs/quality/current-baseline.md
(JSON: docs/quality/current-baseline.json),
regenerated by scripts/collect_quality_baseline.py on each release-gate run.
The release-candidate checklist is docs/quality/release-candidate-checklist.md.
pytest tests/ # Default selector, parallel via xdist
pytest -o addopts='' tests/compat -q # Empyrical/pyfolio compatibility gates
pytest -o addopts='' tests/docs -q # Executable documentation examples
pytest tests/ --cov=fincore # With coveragefincore is part of the cloudQuant quantitative-research ecosystem. The other members cover backtesting, strategy authoring, tooling, and web workflows:
| Project | Description |
|---|---|
| backtrader | Professional Python algorithmic-trading framework for backtesting and live trading, actively maintained by cloudQuant. |
| backtrader-skills | Offline author/review/test tooling for the backtrader fork: turns local datasets and typed StrategySpec v1 into pytest strategies or three-file bundles, reviews candidates without importing them, and runs approved candidates in isolated child processes. |
| backtrader-mcp | Local-first MCP server for building and running reproducible backtrader strategies: immutable datasets, private strategy drafts, and bounded subprocess runs with durable status and reports. Offline and backtest-only. |
| backtrader_web | "AI for Investor": a web-based full-cycle backtrader strategy management platform (Vue 3 + FastAPI) covering research, strategy generation, backtesting analysis, paper trading, live execution, and data management. |
| backtrader-agent | Offline-first strategy-authoring agent runtime: content-addressed data storage, canonical strategy specifications, static review, hash-bound approvals, and a fixed child-process execution profile with recoverable session provenance. |
| fincore | This repository: unified Python toolkit for financial metrics (150+), performance analysis, backtesting support, AI-driven insights, and multi-database/data source integration. |
More details on how these projects fit together: Ecosystem in the online docs.
The fincore repository declares Apache License 2.0; see LICENSE. Adapted-source provenance and unresolved upstream notice questions are tracked in docs/upstream-provenance.md. The required human Alphalens license/NOTICE decision remains a release blocker; no third-party notice or legal conclusion is implied by this integration.
fincore 是面向量化金融的 Python 分析库 — 150+ 金融指标、组合优化、蒙特卡洛模拟和绩效归因。它延续 empyrical 分析栈，由 cloudQuant 持续维护。
当前版本 0.3.0（Beta）。要求 Python 3.11+；这是相对 empyrical（支持更老解释器）的明确 breaking change。
fincore 0.3.0 暴露三个严格分离的界面，同名函数不会在界面之间静默切换语义：
| 界面 | 内容 | 保证 |
|---|---|---|
严格兼容 — fincore.empyrical |
冻结的 empyrical 0.6.0 表面：54/54 公共符号（C0）、49/49 callable（C1）、核心 callable 数值级验证（C3） | 由 tests/compat/fixtures/ 清单固定，tests/compat/ 门禁强制执行 |
pyfolio 门面 — fincore.pyfolio |
冻结的 pyfolio 0.9.6 profile（11 个 tear-sheet 工作流）：全部 C1，risk/returns/perf-attrib/full-sheet 主链 C4 | 需要 fincore[pyfolio] extra |
增强语义 — fincore.metrics、flat API、AnalysisContext |
fincore 自有、已文档化的分歧（如 week_year="iso"、显式校验异常） |
推荐 API；是增强语义，不承诺与 empyrical 完全一致 |
参见兼容矩阵、empyrical 矩阵与 pyfolio profile。
| 特性 | 说明 |
|---|---|
| 150+ 指标 | 收益、风险、回撤、Alpha/Beta、捕获比率、择时、连续统计 |
| AnalysisContext | fincore.analyze() — 惰性计算、自动缓存，支持 JSON/HTML 导出 |
| RollingEngine | 批量滚动指标（sharpe、volatility、max_drawdown、beta）一次调用 |
| 可插拔可视化 | Matplotlib、HTML、Plotly、Bokeh 后端，基于 VizBackend 协议 |
| 组合优化 | 有效前沿、风险平价、约束优化 |
| 蒙特卡洛 | Bootstrap、情景测试、路径模拟 |
| 绩效归因 | Brinson、Fama-French、风格分析 |
| 惰性导入 | 重型依赖在首次使用时加载 |
| PEP 561 | py.typed 标记，支持类型检查器 |
pip install fincore # 核心指标
pip install "fincore[pyfolio]" # + Pyfolio tear sheets（matplotlib、seaborn、ipython）
pip install "fincore[interactive]" # + Plotly、Bokeh 后端
pip install "fincore[report-pdf]" # + Playwright PDF 渲染
pip install "fincore[report-xlsx]" # + XLSX 报告导出
pip install "fincore[bayesian]" # + Bayesian tear sheets（pymc）
pip install "fincore[data-yahoo]" # + Yahoo Finance 数据源
pip install "fincore[data-pandas-datareader]" # + pandas-datareader 数据源
pip install "fincore[data-alphavantage]" # + Alpha Vantage 数据源
pip install "fincore[data-cn]" # + Tushare、AkShare 数据源
pip install "fincore[all]" # 以上全部
pip install "fincore[dev]" # 开发工具datareader 与 viz 是 0.3.x 的兼容别名，指向上述功能性 extras。
从源码安装：
git clone https://gitee.com/yunjinqi/fincore # 中国用户
git clone https://github.com/cloudQuant/fincore # 国际用户
cd fincore && pip install -e ".[dev,viz]"import fincore
import pandas as pd
# 自包含的收益率序列
returns = pd.Series([0.01, -0.005, 0.002, 0.004])
print(f"夏普比率: {fincore.sharpe_ratio(returns):.4f}")
print(f"最大回撤: {fincore.max_drawdown(returns):.4f}")严格兼容的 empyrical 模块可直接导入：
from fincore import empyrical
print(empyrical.sharpe_ratio(returns))
print(empyrical.max_drawdown(returns))AnalysisContext（推荐的有状态 API — 惰性、缓存、可导出）：
index = pd.date_range("2024-01-02", periods=5, freq="B")
returns = pd.Series([0.01, -0.005, 0.002, 0.004, -0.001], index=index)
benchmark = pd.Series([0.008, -0.003, 0.001, 0.002, 0.0], index=index)
ctx = fincore.analyze(returns, factor_returns=benchmark)
print(ctx.sharpe_ratio, ctx.max_drawdown)
ctx.to_json(path="report.json") # 写入文件
ctx.to_html(path="report.html")
ctx.plot(backend="matplotlib") # -> ReportArtifactsPyfolio 主链（from fincore import Pyfolio 需要 pyfolio extra）：
from fincore import Pyfolio
pyfolio = Pyfolio(returns=returns, benchmark_rets=benchmark)
pyfolio.create_returns_tear_sheet(returns, benchmark_rets=benchmark)RollingEngine（批量滚动指标）：
import numpy as np
from fincore.core.engine import RollingEngine
rng = np.random.default_rng(7)
index = pd.date_range("2024-01-02", periods=60, freq="B")
returns = pd.Series(rng.normal(0.001, 0.02, 60), index=index)
benchmark = pd.Series(rng.normal(0.0005, 0.015, 60), index=index)
engine = RollingEngine(returns, factor_returns=benchmark, window=30)
results = engine.compute(['sharpe', 'volatility', 'max_drawdown', 'beta'])组合优化：
from fincore.optimization import efficient_frontier, risk_parity, optimize
returns_df = pd.DataFrame(
{
"asset_a": [0.01, -0.005, 0.004, 0.002],
"asset_b": [0.003, 0.002, -0.001, 0.005],
}
)
ef = efficient_frontier(returns_df, n_points=5)
rp = risk_parity(returns_df)
w = optimize(returns_df, objective="max_sharpe")以上每个 Python 代码块都以相同的数据和参数被
tests/docs/test_examples.py 中对应的测试执行。
质量数字由机器生成，绝不手写进本 README。当前快照见
docs/quality/current-baseline.md
（JSON: docs/quality/current-baseline.json），
由 scripts/collect_quality_baseline.py 在每次发布门禁运行时重新生成。
发布候选清单见 docs/quality/release-candidate-checklist.md。
pytest tests/ # 默认选择器，xdist 并行
pytest -o addopts='' tests/compat -q # empyrical/pyfolio 兼容门禁
pytest -o addopts='' tests/docs -q # 可执行文档示例
pytest tests/ --cov=fincore # 含覆盖率fincore 是 cloudQuant 量化研究生态的一部分。生态中的其他成员覆盖回测、 策略编写、工具链与 Web 工作流：
| 项目 | 简介 |
|---|---|
| backtrader | 专业的 Python 量化交易框架，支持回测与实盘交易，由 cloudQuant 持续维护。 |
| backtrader-skills | backtrader 分支的离线编写/审查/测试工具：将本地数据集与类型化 StrategySpec v1 转换为 pytest 策略或三文件策略包，不导入即可静态审查候选，并在隔离子进程中运行已批准候选。 |
| backtrader-mcp | 本地优先的 MCP 服务器，用于构建和运行可复现的 backtrader 策略：不可变数据集、私有策略草稿、有界子进程运行及持久化状态与报告。完全离线、仅限回测。 |
| backtrader_web | "AI for Investor"：基于 Web 的 backtrader 全周期策略管理平台（Vue 3 + FastAPI），覆盖研究、策略生成、回测分析、模拟盘、实盘执行与数据管理。 |
| backtrader-agent | 离线优先的策略编写智能体运行时：内容寻址数据存储、规范化策略规格、静态审查、哈希绑定的审批，以及固定子进程执行配置与可恢复的会话溯源。 |
| fincore | 本仓库：统一的 Python 金融工具包，提供 150+ 金融指标、绩效分析、回测支持、AI 驱动的洞察以及多数据库/数据源集成。 |
更多项目间协作关系见在线文档的 Ecosystem 页面。
欢迎贡献！详见 CONTRIBUTING.md。
fincore 仓库声明采用 Apache License 2.0，详见 LICENSE。 改编来源与尚待人工确认的上游 notice 问题记录于 docs/upstream-provenance.md。