Skip to content

01_overview.md

Latest commit

 

History

History
142 lines (109 loc) · 5.15 KB

01_overview.md

File metadata and controls

142 lines (109 loc) · 5.15 KB
Raw

QuantYield - Platform Overview

QuantYield is an institutional-grade fixed income analytics platform providing bond pricing, yield curve modelling, portfolio risk management, and AI-powered forecasting through a clean REST API built on Django.

Platform Capabilities

Domain Capabilities
Bond Pricing Dirty/clean price, YTM solving, accrued interest, cash flows
Duration Analytics Macaulay, Modified, DV01, Key Rate Duration (10 tenors)
Spread Analytics Z-spread, OAS (Monte Carlo), benchmark spread matrix
Yield Curves Nelson-Siegel, Svensson, Bootstrap, Cubic Spline
Portfolio Risk Duration, convexity, VaR, CVaR, scenario analysis
AI Forecasting Transformer, LSTM, AR(1) rate forecasting with uncertainty bands
Regime Detection ML ensemble (Random Forest + XGBoost) regime classification
Volatility GARCH(1,1), EGARCH, historical volatility term structure
Credit Risk XGBoost credit spread prediction with feature importances
Curve Factors PCA factor decomposition (Level, Slope, Curvature)

System Architecture

QuantYield/
 +-- code/
 |    +-- backend/          Django REST API (persistence, admin, CRUD)
 |    +-- ml_services/      AI and ML models (framework-agnostic)
 +-- docs/                  Full documentation suite
 +-- docker-compose.yml     Production stack (Postgres + Redis + Nginx)
 +-- nginx.conf             Reverse proxy configuration

The backend and ML services are deliberately decoupled. The ML services directory is a standalone Python package that can be used independently of Django. The Django backend imports from it but does not depend on any ML-specific framework at the web layer.

Technology Stack

Layer Technology
Web Framework Django 5 + Django REST Framework
Database SQLite (development) / PostgreSQL (production)
Cache Local memory (development) / Redis (production)
Task Scheduling Management commands (cron-friendly)
API Documentation drf-spectacular (OpenAPI 3.0, Swagger, ReDoc)
ML - Deep Learning PyTorch (Transformer, LSTM) - optional
ML - Ensemble scikit-learn, XGBoost
ML - Volatility arch (GARCH, EGARCH)
ML - Statistics scipy, statsmodels
Numerical Core numpy, scipy, pandas
Deployment Docker Compose, uvicorn (ASGI), nginx

Key Design Decisions

Service Layer Separation

The services/ package in the backend and the ml_services/ package contain pure Python with no Django dependencies. This means:

  • Services can be tested without a running Django instance.
  • ML models can be replaced, updated, or extended without touching API code.
  • The same pricing engine can be embedded in notebooks or scripts.

Graceful Degradation

Every ML feature has a fallback:

Primary Fallback
Transformer forecaster LSTM
LSTM forecaster AR(1) autoregressive
GARCH volatility Historical rolling vol
XGBoost credit model Random Forest
ML regime classifier Rule-based detection
FRED live data Representative fallback curve

Caching Strategy

Resource TTL Backend
Treasury curve 300 seconds Redis / local memory
ML model weights Process lifetime In-process dict
Yield history Not cached Fetched on demand

Quick Start

# 1. Install dependencies
cd code/backend
pip install -r requirements.txt

# 2. Configure environment
cp .env.example .env
# Edit .env: set SECRET_KEY, optionally FRED_API_KEY

# 3. Run migrations and seed data
python manage.py migrate
python manage.py seed_data

# 4. Start the server
python manage.py runserver

# API docs: http://localhost:8000/docs/
# Admin UI: http://localhost:8000/admin/

Docker (Full Stack)

cp .env.example .env
docker-compose up --build
# API:   http://localhost:8000
# Admin: http://localhost:8000/admin/

API Base URLs

Resource URL
API Root /api/v1/
Bonds /api/v1/bonds/
Portfolios /api/v1/portfolios/
Yield Curves /api/v1/curves/
Analytics /api/v1/analytics/
Swagger UI /docs/
ReDoc /redoc/
Admin /admin/
Health Check /api/v1/health/