Quantitative Portfolio Intelligence Platform

RiskFecta is a Python-based portfolio analytics system that combines machine-learning return forecasting with Modern Portfolio Theory optimization. It ingests institutional-grade market data from Bloomberg Terminal, trains a hybrid LSTM + XGBoost ensemble to predict 30-day forward returns, and feeds those signals into a SciPy Efficient Frontier optimizer—then surfaces results through an interactive Streamlit demo and a professional Tableau report.

Status (May 2026): Early development — Phase 0 (pre-build) complete. Repository scaffold, database schema, configuration, and documentation are in place. Data pipeline, models, optimizer, and deployment are planned per the locked PRD and build plan (Phases 1–9).

Professional portfolio tools (e.g. Bloomberg PORT, FactSet) cost tens of thousands of dollars per year and are out of reach for most retail investors. Free alternatives typically rely on historical averages for expected returns and do not combine deep learning, factor models, and rigorous optimization in one reproducible, deployable stack.

RiskFecta’s goal: Build an accessible, interview-defensible quant platform that:

1. Sources real Bloomberg data (BQL / Excel export → CSV → PostgreSQL)
2. Forecasts 30-day returns with a dual-model ensemble (sequential LSTM + tabular XGBoost)
3. Optimizes long-only portfolios on the Efficient Frontier using ensemble expected returns and realized-return covariance
4. Presents outcomes interactively (Streamlit + Plotly) and in a recruiter-facing static report (Tableau)

RiskFecta is not a live trading or order-execution system. It is a research and analytics platform with strict out-of-sample validation and look-ahead controls.

flowchart LR
    subgraph ingest [Phase 1–2]
        BB[Bloomberg BQL / Excel]
        CSV[CSV exports]
        PG[(PostgreSQL)]
        BB --> CSV --> PG
    end

    subgraph ml [Phase 3–5]
        LSTM[LSTM — price sequences]
        XGB[XGBoost — factors / macro]
        ENS[50/50 ensemble]
        LSTM --> ENS
        XGB --> ENS
    end

    subgraph opt [Phase 6]
        MPT[SciPy MPT optimizer]
        EF[Efficient Frontier]
        ENS --> MPT --> EF
    end

    subgraph viz [Phase 7–8]
        ST[Streamlit Cloud]
        TB[Tableau report]
        EF --> ST
        EF --> TB
    end

    PG --> LSTM
    PG --> XGB
    PG --> MPT

Validation (locked): Rolling window — 252 trading-day train, 21-day step, 30-day forecast horizon; final 3 months held out for final evaluation. No expanding window; no regime classifier. Risk-free rate for Sharpe: USGG10YR / 252 everywhere.

MVP-A (minimum viable product) is complete when: real Bloomberg data is ingested; rolling-window LSTM and XGBoost run without leakage; ensemble and Efficient Frontier are computed; Streamlit Cloud demo runs on that data. See PRD_and_buildplan/Phase0_Plan.md for criteria.

Aligned with the locked Build Plan (~85–90 hrs total):

Live links will be added here after Phase 7–8:

• Streamlit demo: coming soon
• Tableau report: coming soon

RiskFecta/
├── app/                 # Streamlit app (Phase 7)
├── pipeline/            # ingest.py, features.py (Phases 1–2)
├── models/              # lstm.py, xgboost_model.py, ensemble.py (Phases 3–5)
├── optimizer/           # portfolio.py (Phase 6)
├── tests/               # Pipeline & model tests (Phase 9)
├── tableau/             # Tableau workbook / exports (Phase 8)
├── data/raw/            # Bloomberg CSVs (gitignored)
├── config.py            # Universe, features, rolling-window constants
├── schema.sql           # PostgreSQL DDL
├── requirements.txt
├── docs/
│   ├── SETUP.md
│   └── Bloomberg_export_spec.md
└── PRD_and_buildplan/   # Locked PRD v1.0 + build plan (reference)

1. Clone the repository and use Python 3.10–3.13 (see docs/SETUP.md).
2. Create a virtual environment and install dependencies:

   py -3.13 -m venv venv
   .\venv\Scripts\Activate.ps1
   pip install -r requirements.txt

3. PostgreSQL: Create database riskfecta, apply schema.sql, add .env from .env.example:

   DATABASE_URL=postgresql://user:password@localhost:5432/riskfecta
   

4. Verify config:

   python -c "import config; print(config.TRAIN_WINDOW, config.STEP)"

Full Windows/PostgreSQL steps: docs/SETUP.md.
Bloomberg export field list for Phase 1: docs/Bloomberg_export_spec.md.

• Bloomberg CSVs live under data/raw/ and are gitignored.
• Never commit .env, credentials, or raw market exports.
• Missing Bloomberg values remain NULL in the database; forward-fill happens only in feature engineering (Phase 2), not at ingest.

RiskFecta intentionally does not include:

• Live trading or paper trading
• Synthetic or Yahoo Finance substitutes for MVP (real Bloomberg export required)
• Bloomberg API scripting from a personal laptop
• SQLite / Power BI (those belong to a separate project, PlainCents)

Kapil Iyer — University of Waterloo
Academic / internship-tier quant portfolio project (2026). Architecture and ML decisions are pre-validated in the PRD for reproducibility and technical interviews.

For internal planning detail, see PRD_and_buildplan/ (PRD v1.0, build plan v1.0, Phase 0 checklist).

License TBD. Contact the repository owner before reuse or redistribution.