Six systematic macro alpha signals: C++26 hot-path engines + Python 3.13 research interface — production-grade for Cubist/Citadel/Millennium quant pods.

• 🔍 0 Strategy Deep Dives
• 📂 1 Repository Structure
• 📖 2 Solution Overview
• ✨ 3 Technology Features
  • 🐍 Python 3.13
  • 🔷 C++26
  • ⚛️ nanobind (C++ bindings)
• 📐 4 Mathematical Concepts
  • 🏛️ PDRRM Signal (Policy Divergence × Real Rate Momentum)
  • 🎢 TPMCR Signal (Term Premium Momentum & Curve Regime Signal
  • 📊 MAERM Signal (Macro-Adjusted Earnings Revision)
  • 🎁 ISRC Signal (Inventory Surprise × Roll Return Composite)
  • ⚖️ VSRA Signal (Volatility Surface Regime Arbitrage)
  • 💥 FDSP Signal (Fiscal Dominance Shock Propagation)
  • ⏱️ Signal Decay Monitor
  • 🛡️ Vol-Targeted Positions
  • 🏗️ Architecture
• 🚀 Quick Start: Build & Run
• ⚙️ 5 Prerequisites & Installation
  • 🐧 Linux (Ubuntu 24.04)
  • 💻 Windows 11
• 📓 6 Notebook: research.ipynb
• 🏗️ 7 Compile, Build & Run
  • 🤖 GitHub Actions
• 🗺️ 8. Improvement Roadmap
  • 🗓️ Remainder of 2026
  • 🚀 2027

⬆ Back to Top

systematic_macro_2026_alpha_research/
├── .bazelrc                     # C++26 -O3 -march=native -mavx2 flags
├── .gitignore
├── .github/workflows/ci.yml     # Bazel gateway: build, test, daily decay monitor + Slack
├── BUILD.bazel                  # Root targets: backtest monitor research docker_*
├── WORKSPACE                    # Deps: Eigen3, GoogleTest, nanobind, rules_python
├── CMakeLists.txt               # Installs alpha_cpp.so into Python site-packages
├── Dockerfile                   # Multi-stage: cpp-builder → python:3.13-slim
├── docker-compose.yml           # Services: jupyter, backtest, monitor
├── pyproject.toml               # Hatchling: alpha_research pkg + all Python deps
├── run.sh / run.bat             # Ubuntu/Win11 launchers (Bazel gateway)
├── cmake/alpha_research-config.cmake.in
├── data/                        # Market data CSVs (or synthetic)
├── notebooks/research.ipynb     # 28-cell production research notebook
├── scripts/runner.py            # Bazel dispatcher (backtest/monitor/jupyter/docker)
├── src/
│   ├── cpp/                     # C++26 engines
│   │   ├── math_utils.hpp       # EWMA, RingBuffer(lock-free), ridge, Sharpe, IC, HL
│   │   ├── pdrrm_engine.hpp     # PDRRM: RRDM+PSS+RAC, ridge weights, vol-targeted pos
│   │   ├── strategies_engine.hpp# TPMCR, MAERM, ISRC, VSRA, FDSP online tick() engines
│   │   ├── portfolio_optimizer.hpp # BacktestEngine, KPIBundle, SignalDecayMonitor
│   │   ├── bindings.cpp         # nanobind Python↔C++26 bridge (zero-copy ndarray)
│   │   └── signal_demo.cpp      # Standalone CLI demo binary
│   └── python/alpha_research/   # Python 3.13 package (importable as alpha_research)
│       ├── data.py              # Polars ingestion + 10-yr synthetic generator
│       ├── signals.py           # Signal orchestration → C++26 engines
│       └── backtest.py          # BacktestOrchestrator + black-swan stress tests
├── tests/
│   ├── cpp/test_engines.cpp     # Google Test: 40+ tests, 100% engine coverage
│   └── python/test_pipeline.py  # pytest: 50+ tests, 100% module coverage
└── third_party/                 # Bazel BUILD files for Eigen3 + nanobind

⬆ Back to Top

All six signals are combined via PCA risk-budgeting in a master QP with TC penalty.

⬆ Back to Top

• dataclasses(slots=True, frozen=True) — cache-line-efficient config structs
• typing.NamedTuple result bundles; match/case dispatcher
• Polars (no pandas) — Arrow/SIMD backend, LazyFrame query plans, .to_numpy() zero-copy → C++
• __future__.annotations PEP 563 for forward refs

• std::expected<T,E> — monadic error handling, no exceptions on hot path
• std::print / std::println (P2093R14)
• Designated initialisers for self-documenting DayData{.tp_acm_10y=1.2, …}
• [[nodiscard]], [[likely]], [[unlikely]], std::assume for vectoriser hints
• alignas(64) RingBuffer<N> with power-of-2 branchless modular indexing
• Eigen column-major matrices; -mavx2 -mfma via .bazelrc for SIMD BLAS

• ndarray<double, ndim<2>, c_contig> → Eigen::Map zero-copy bridge
• nb::capsule custom deleter for safe numpy buffer ownership
• Full type-safe exposure of all six engines + KPIBundle + BacktestEngine

⬆ Back to Top

$$\boxed{S_{i,t} = \hat{\alpha}_1 \cdot \text{RRDM}_{i,t} + \hat{\alpha}_2 \cdot \text{PSS}_{i,t} + \hat{\alpha}_3 \cdot \text{RAC}_{i,t}}$$

$$\boxed{\hat{\boldsymbol{\alpha}} = (X^\top X + \lambda I)^{-1} X^\top y \quad \text{(ridge, } \lambda=0.1\text{)}}$$

$$\boxed{\text{RRDM}_{i,t} = z_{cs}!\left[(r^{\text{nom}}_i - \pi^{\text{be}}_i)_t - (r^{\text{nom}}_i - \pi^{\text{be}}_i)_{t-20}\right]}$$

$$\boxed{S^{\text{TPMCR}}_{i,t} = \hat{\beta}_1 \cdot \text{TPM}_{i,t} + \hat{\beta}_2 \cdot \text{CRS}_{i,t} + \hat{\beta}_3 \cdot \text{FiscalStress}_{t} \cdot \mathbb{1}[\text{USD bond}]}$$

$$\boxed{\text{TPM}_{t} = z_{cs}!\left(\text{TP}^{\text{ACM}}_{10Y,t} - \text{TP}^{\text{ACM}}_{10Y,t-\tau}\right)}$$

$$\boxed{\text{CRS}_{t} = \sum_{k=1}^{4} P(\text{State}_t = k | \mathbf{o}_{1:t}) \cdot v_k}$$

$$\boxed{\text{FSO}_{t} = z!\left(-\text{SwapSpread}^{30Y}_{t} + \delta \cdot \Delta\text{SwapSpread}^{30Y}_{t-5:t}\right) \cdot \mathbb{1}[\text{USD bond}]}$$

$$\boxed{S^{\text{MAERM}}_{i,t} = \hat{\gamma}_1 \cdot \text{ERB}_{i,t} + \hat{\gamma}_2 \cdot \text{MRF}_{t} \cdot \text{ERB}_{i,t} + \hat{\gamma}_3 \cdot \text{SD}_{i,t}}$$

$$\boxed{\text{ERB}_{i,t} = z!\left(\frac{N^{+}_{i,t-5:t} - N^{-}_{i,t-5:t}}{N^{+}_{i,t-5:t} + N^{-}_{i,t-5:t} + N^{0}_{i,t-5:t}}\right)}$$

$$\boxed{\text{MRF}_{t} = \sigma!\left(\alpha_0 + \alpha_1 \cdot \text{ISM}_{t} + \alpha_2 \cdot \Delta\text{ISM}_{t-1:t}\right)}$$

$$\boxed{\text{SD}_{i,t} = \sum_{q} \left(\frac{\text{EPS}_{\text{actual},q} - \text{EPS}_{\text{consensus},q}}{\text{EPS}_{\text{consensus},q}}\right) \cdot e^{-\frac{(t - t_q) \ln 2}{30}}}$$

$$\boxed{S^{\text{ISRC}}_{i,t} = \hat{\delta}_1 \cdot \text{IS}_{i,t} + \hat{\delta}_2 \cdot \text{RRM}_{i,t} + \hat{\delta}_3 \cdot \text{IS}_{i,t} \cdot \text{RRM}_{i,t}}$$

$$\boxed{\text{IS}_{i,t} = z!\left(-\frac{\text{EIA}^{\text{actual}}_{i,t} - \text{EIA}^{\text{consensus}}_{i,t}}{\sigma^{52W}_{\text{surprise},i}}\right)}$$

$$\boxed{\text{RRM}_{i,t} = z!\left(\text{RR}_{i,t} - \text{RR}_{i,t-20}\right)}$$

$$\boxed{S^{\text{VSRA}}_{t} = \hat{\eta}_1 \cdot \text{TSS}_{t} + \hat{\eta}_2 \cdot \text{SKA}_{t} + \hat{\eta}_3 \cdot \text{VRP}_{t}}$$

$$\boxed{\text{TSS}_{t} = z!\left(\ln!\left(\frac{\text{VIX3M}_{t}}{\text{VIX}_{t}}\right)\right)}$$

$$\boxed{\text{SKA}_{t} = z!\left((\sigma^{25P}_{t} - \sigma^{25C}_{t}) - \widehat{\text{RSkew}}_{t}(21d)\right)}$$

$$\boxed{\text{VRP}_{t} = z!\left(\text{VIX}_{t}^2/100 - \text{RV}_{21,t}\right)}$$

$$\boxed{S^{\text{FDSP}}_{\text{asset},t} = \hat{\phi}_{\text{asset}} \cdot \text{FCI}_{t} + \hat{\psi}_{\text{asset}} \cdot \text{PropagationLag}_{\text{asset},t}}$$

$$\boxed{\text{FCI}_{t} = z!\left(w_1 \cdot (-\text{SS}_{30Y,t}) + w_2 \cdot \text{CDS}_{5Y,t} + w_3 \cdot \text{TBill Spike}_{t}\right)}$$

$$\boxed{\text{PropagationLag}_{\text{asset},t} = \sum_{k=0}^{K} w_{\text{asset},k} \cdot \text{FCI}_{t-k}}$$

$$\boxed{\Delta y_t = \theta \cdot y_{t-1} + \varepsilon \implies \text{HL} = -\ln(2)/\theta \quad \text{Alert if HL} < 30\text{d}}$$

$$\boxed{w_{i,t} = \frac{\sigma^{\text{target}}}{\hat{\sigma}^{\text{ann}}_{i,t}} \cdot S_{i,t}, \qquad w_{i,t} \in [-0.25,, 0.25]}$$

flowchart TD
    D["Polars DataFrames"] --> PY["Python 3.13 Orchestrator"]
    PY -->|nanobind zero-copy| CPP["C++26 Engines AVX2/SIMD"]
    CPP --> P["PDRRM"] & T["TPMCR"] & M["MAERM"] & I["ISRC"] & V["VSRA"] & F["FDSP"]
    P & T & M & I & V & F --> BT["BacktestEngine KPIBundle"]
    BT --> MON["SignalDecayMonitor"]
    MON -->|HL lt 30d| SLK["Slack Alert"]
    BT --> NB["research.ipynb"]

⬆ Back to Top

sudo apt-get install -y build-essential cmake ninja-build \
    python3.13 python3.13-dev libeigen3-dev git curl

# Bazelisk
curl -fsSL https://github.com/bazelbuild/bazelisk/releases/latest/download/bazelisk-linux-amd64 \
  -o /usr/local/bin/bazel && chmod +x /usr/local/bin/bazel

# Python package (editable, no absolute paths)
pip install -e ".[dev]"

# C++ extension
cmake -B build -S . -DCMAKE_BUILD_TYPE=Release -G Ninja
cmake --build build --parallel && cmake --install build

python -c "import alpha_research; import alpha_cpp; print('All OK')"

winget install Kitware.CMake Microsoft.VisualStudio.2022.BuildTools
pip install -e .[dev]
cmake -B build -S . -DCMAKE_BUILD_TYPE=Release
cmake --build build --config Release
cmake --install build

⬆ Back to Top

28-cell production notebook covering:

Run (Ubuntu local): ./run.sh notebook → http://localhost:8888 Run (Docker): ./run.sh docker-notebook → http://localhost:8888 Run (Windows local): run.bat notebook Run (Windows Docker): run.bat docker-notebook

⬆ Back to Top

# Ubuntu — all via Bazel (common gateway)
./run.sh build              # bazel build //...
./run.sh test               # bazel test //:test_all
./run.sh backtest           # bazel run //:backtest
./run.sh monitor            # bazel run //:monitor

# Windows
run.bat build
run.bat test
run.bat backtest

# Docker
./run.sh docker-build
./run.sh docker-notebook    # http://localhost:8888
./run.sh docker-down

# CMake alternative
cmake -B build -S . -DCMAKE_BUILD_TYPE=Release -G Ninja
cmake --build build --parallel
ctest --test-dir build --output-on-failure
pytest tests/python/ -v

Trigger by pushing to main or develop. Force-trigger:

gh workflow run ci.yml

Required secret: SLACK_WEBHOOK (Settings → Secrets → Actions).

Jobs: cpp-build-test · python-lint-test (every push) · signal-monitor (daily 06:00 UTC) · full-backtest (push to main).

⬆ Back to Top

• TPMCR: Replace rule-based regime with full Baum-Welch HMM (hmmlearn in deps, ready to swap)
• PDRRM: Add BOJ JGB purchase tapering as 4th signal component
• FDSP: Re-calibrate propagation kernels with 2026 debt-ceiling episode data
• Monitoring: Upgrade from daily to 4-hour IC cadence; add BOCPD changepoint detection
• Risk: Add cross-signal TC optimisation (shared QP minimises total portfolio turnover, not per-strategy)

• Replace nanobind with native CPython 3.14 limited API for sub-100µs bridge latency
• Migrate covariance estimation to GPU (CUDA 13 + cuSOLVER) for 200+ instrument universes
• Alternative data: Satellite shipping data × ISRC; LLM-parsed FOMC transcript sentiment × PDRRM PSS
• Meta-model: HMM + gradient-boosted regime classifier to dynamically reweight 6 strategies
• Cross-signal alpha: Research PDRRM × TPMCR multiplicative interaction (JPY rally amplified when US term premium also rising)
• Variance swap replication for cleaner VRP capture in VSRA; gamma overlay for FDSP during stress

⬆ Back to Top