Native C++23 analytics for portfolios, factors, performance and risk — one typed
API over a local DataFrame, with no Python dependency at run time.
用于投资组合、因子、业绩与风险分析的原生 C++23 库。同一份文档提供两种语言： English · 简体中文。
fincore_cpp is a native C++23 library for quantitative portfolio, factor,
performance and risk analytics. It ships its own typed API over a local
DataFrame type and does not depend on Python, pandas or NumPy at runtime — nor
does it promise pandas object compatibility.
#include <fincore/dataframe.hpp>
#include <fincore/metrics/drawdown.hpp>
#include <fincore/metrics/ratios.hpp>
const auto returns = fincore::make_series(
{fincore::utc_timestamp(2025, 1, 2), fincore::utc_timestamp(2025, 1, 3)},
"returns", {0.01, -0.02});
const double sharpe = fincore::sharpe_ratio(returns, "returns");
const double drawdown = fincore::max_drawdown(returns, "returns");| Area | What you get |
|---|---|
| Core data | fincore::dataframe (index + typed columns), UTC nanosecond timestamps, calendar and frequency helpers |
| Metrics | Returns, ratios, drawdowns, rolling statistics, tables, stability and consecutive-run statistics |
| Risk | Historical and parametric VaR/ES, GARCH/EGARCH, EVT (GPD/GEV), Bayesian cone, backtests and walk-forward validation |
| Portfolio | Positions, transactions, round trips, turnover, capacity, cash-flow-aware performance |
| Factor research | Analysis pipeline, quantile/IC studies, tears, artifacts, point-in-time joins, export contracts |
| Attribution | Brinson (BHB/Fachler), factor and style regression, exposures, prediction, performance attribution |
| Optimization | Mean-variance and risk-parity style solvers, projected gradient, box/sector constraints |
| Reporting | One report model rendered to HTML, XLSX, PDF, Plotly, Bokeh, Matplotlib-style axes and ImPlot |
| Runtime | Operation catalog and dispatch, analysis context caching, extension snapshots, serialization |
| Simulation | GBM, bootstrap, Latin hypercube, antithetic variates, correlation breakdowns |
- CMake 3.25+ and a C++23 compiler (MSVC 19.4+, Clang 17+, GCC 13+).
- Pinned git submodules (
third_party/DataFrame,third_party/implot).
Python is used only for tooling, benchmarks and documentation, never by the library.
git clone https://github.com/cloudQuant/fincore_cpp.git
cd fincore_cpp
git submodule update --init --recursive
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --parallel
ctest --test-dir build --output-on-failureInstall and consume from CMake:
cmake --install build --prefix /path/to/prefixfind_package(fincore CONFIG REQUIRED COMPONENTS fincore)
target_link_libraries(my_target PRIVATE fincore::fincore)Build options: FINCORE_BUILD_TESTING (default ON), FINCORE_BUILD_IMPLOT
(default OFF, GUI adapter), FINCORE_BUILD_IMPLOT_DEMO (default OFF),
FINCORE_ENABLE_ASAN and FINCORE_ENABLE_UBSAN (default OFF).
One bilingual documentation site, with a language switch in the top-right corner of every page and your choice remembered between visits.
- Online: https://cloudquant.github.io/fincore_cpp/en/latest/ · https://cloudquant.github.io/fincore_cpp/zh-cn/latest/
- New to the library: the tutorial goes from prices to a rendered report in six steps; the cookbook answers single tasks.
- Start here: installation, quickstart, conventions
- Guides: metrics, portfolio and risk, factor research, reports and runtime
- Reference: one page per public module, generated from the headers
- Repository sources:
docs/site/, build steps indocs/site/README.md
Public coverage is tracked by docs/site/api-ledger.tsv: every public symbol is
documented, and the strict Sphinx builds (-W -n --keep-going) run for both
languages on every change.
The repository includes a paired benchmark harness that compares the native implementation against the pinned Python reference on identical inputs. On the Alpha003 A-share research workload the 500-file development batch reached 41.20x over the reference runner; separate 1,000/2,000-file diagnostic pairs reached 34.46x / 35.63x. These are measurements of that workload, not a guarantee for every API. See the iteration 9 closeout for methodology, memory results and the evidence levels behind each number.
include/fincore/ public headers, one directory per module
src/ implementation
tests/ unit, consumer and oracle tests (1092 cases)
examples/ render and dashboard demos (excluded from the default build)
benchmarks/ alpha003 research benchmark and layout experiments
docs/site/ bilingual documentation sources and build tooling
docs/iterations/ iteration records and acceptance evidence
tools/ Python tooling, oracles and replay harnesses
third_party/ pinned submodules
- 1092 C++ tests, plus strict bilingual documentation builds, are part of the routine gate.
- Supported compilers: MSVC, Clang and GCC on their C++23 settings; the CI matrix builds Windows and macOS, with a known GCC 13 constexpr-limit failure tracked in the iteration record.
- The documentation is live on GitHub Pages (English and Chinese) and on Read the Docs (one project, English: https://fincore-cpp.readthedocs.io/en/latest/); the language switch links the two hosts at the same page. The current state is always recorded in the publication page.
Issues and pull requests are welcome. Please read
CONTRIBUTING for the build, test and documentation
rules, and keep changes comment-only when they claim to be. Historical planning
and acceptance records live under docs/iterations/ and
docs/acceptance/.
MIT — see LICENSE.
fincore_cpp 是一个原生 C++23 库，用于量化投资组合、因子、业绩与风险分析。它在
本地 DataFrame 类型之上提供自己的类型化 API，运行时不依赖 Python、pandas 或
NumPy，也不承诺与 pandas 对象兼容。
#include <fincore/dataframe.hpp>
#include <fincore/metrics/drawdown.hpp>
#include <fincore/metrics/ratios.hpp>
const auto returns = fincore::make_series(
{fincore::utc_timestamp(2025, 1, 2), fincore::utc_timestamp(2025, 1, 3)},
"returns", {0.01, -0.02});
const double sharpe = fincore::sharpe_ratio(returns, "returns");
const double drawdown = fincore::max_drawdown(returns, "returns");| 领域 | 提供的能力 |
|---|---|
| 核心数据 | fincore::dataframe（索引 + 类型化列）、UTC 纳秒时间戳、日历与频率工具 |
| 指标 | 收益、比率、回撤、滚动统计、表格、稳定性与连续段统计 |
| 风险 | 历史法与参数法 VaR/ES、GARCH/EGARCH、EVT（GPD/GEV）、贝叶斯锥、回测与滚动前推验证 |
| 组合 | 持仓、交易、往返交易、换手、容量、含现金流的业绩计算 |
| 因子研究 | 分析流水线、分位/IC 研究、tears、artifacts、point-in-time 连接、导出契约 |
| 归因 | Brinson（BHB/Fachler）、因子与风格回归、暴露、预测、业绩归因 |
| 优化 | 均值方差与风险平价类求解器、投影梯度、箱式与行业约束 |
| 报表 | 单一报表模型渲染到 HTML、XLSX、PDF、Plotly、Bokeh、Matplotlib 风格坐标轴与 ImPlot |
| 运行时 | 操作目录与分发、分析上下文缓存、扩展快照、序列化 |
| 模拟 | GBM、bootstrap、拉丁超立方、对偶变量、相关性分解 |
- CMake 3.25+ 与支持 C++23 的编译器（MSVC 19.4+、Clang 17+、GCC 13+）。
- 固定在
third_party/DataFrame、third_party/implot下的 git 子模块。
Python 只用于工具、基准与文档，库本身不需要它。
git clone https://github.com/cloudQuant/fincore_cpp.git
cd fincore_cpp
git submodule update --init --recursive
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --parallel
ctest --test-dir build --output-on-failure安装并在 CMake 中消费：
cmake --install build --prefix /path/to/prefixfind_package(fincore CONFIG REQUIRED COMPONENTS fincore)
target_link_libraries(my_target PRIVATE fincore::fincore)构建选项：FINCORE_BUILD_TESTING（默认 ON）、FINCORE_BUILD_IMPLOT
（默认 OFF，图形界面适配器）、FINCORE_BUILD_IMPLOT_DEMO（默认 OFF）、
FINCORE_ENABLE_ASAN 与 FINCORE_ENABLE_UBSAN（默认 OFF）。
同一套双语文档站点，每页右上角都有语言切换按钮，并会记住你的选择。
- 在线站点：https://cloudquant.github.io/fincore_cpp/en/latest/ · https://cloudquant.github.io/fincore_cpp/zh-cn/latest/
- 第一次使用：教程 用六步从价格走到渲染好的报表； 配方集 回答单个具体任务。
- 入门：安装、快速开始、数据与数值约定
- 指南：指标、组合与风险、因子研究、报表与运行时
- 参考：每个公共模块一页，直接由头文件生成
- 仓库内源码：
docs/site/，构建步骤见docs/site/README.md
公共覆盖由 docs/site/api-ledger.tsv 跟踪：每个公共符号都有说明，且两种语言的
严格 Sphinx 构建（-W -n --keep-going）在每次改动上都会执行。
仓库内有一套成对基准，用相同输入比较原生实现与固定的 Python 参考实现。在 Alpha003 A 股研究负载上，500 文件的开发批达到 41.20x；另外的 1,000/2,000 文件诊断对达到 34.46x / 35.63x。这些是该负载的测量结果，不构成对所有 API 的保证。方法与内存 结果见迭代 9 收尾记录。
include/fincore/ 公共头文件，每个模块一个目录
src/ 实现
tests/ 单元、消费者与 oracle 测试（1092 个用例）
examples/ 报表与仪表盘示例（默认构建不包含）
benchmarks/ alpha003 研究基准与内存布局实验
docs/site/ 双语文档源码与构建工具
docs/iterations/ 迭代记录与验收证据
tools/ Python 工具、oracle 与回放脚本
third_party/ 固定子模块
- 1092 个 C++ 测试与双语严格文档构建都属于例行门禁。
- 支持的编译器：MSVC、Clang、GCC 的 C++23 配置；CI 覆盖 Windows 与 macOS，
GCC 13 的 constexpr 限制导致的
ubuntu-latest失败已在迭代记录中登记。 - 文档已上线 GitHub Pages（中英文）与 Read the Docs（单一项目，英文： https://fincore-cpp.readthedocs.io/en/latest/）；语言切换会在同一页面上连接 两个宿主。当前状态始终记录在发布说明。
欢迎提交 issue 与 pull request。构建、测试与文档规则见
CONTRIBUTING；历史计划与验收记录位于
docs/iterations/ 与 docs/acceptance/。
MIT，详见 LICENSE。