About Spillover Lab

Project Overview

Spillover Lab is a modular platform for time series analysis and forecasting, built around financial and econometric models. It reimagines my master's thesis on GARCH-based volatility spillovers in global markets.

The original research modeled how a collapse in China's real estate sector could ripple through global real estate ETFs—like the 2008 Financial Crisis, but with the shock centered in Beijing.

With Spillover Lab, the epicenter is yours to choose. Where's the next bubble forming?

Features

Data Sources

Access real-time market data from Stooq and Yahoo Finance, or generate synthetic R&D datasets.

  • Stooq and Yahoo Finance: Real-time stock, ETF, and index data
  • Synthetic Data: Controlled testing environments

Econometric Models

Industry-standard time series models for financial analysis and forecasting. Implementation details available at ReadTheDocs.

  • ARIMA Models: Price forecasting and trend analysis
  • GARCH Models: Volatility clustering
  • Spillover Analysis: Interconnection measurement

Visualizations

Charts and analysis tools for comprehensive market research.

  • Time Series Plots: Price and volatility
  • Spillover Networks: interconnection diagrams
  • Risk Metrics: Various point estimates

Implementation

Architecture

The following sequence diagram illustrates the project's architecture:

sequenceDiagram actor User actor AIAgent as AI Agent participant Frontend as Frontend participant Proxy as Proxy Server participant API as API participant Package as Compute Package participant DB as Database User->>Frontend: Submit analysis request Frontend->>Proxy: Forward request parameters Proxy->>API: Forward request to backend API API->>Package: Execute computations Package-->>API: Return computation results API->>DB: Write results to database API-->>Proxy: Return processed results Proxy-->>Frontend: Return results to frontend Frontend-->>User: Display visualization %% Direct API interaction User-->>Proxy: Direct REST/GraphQL request Proxy-->>User: Direct API response %% AI Agent interaction AIAgent->>Proxy: Submit LLM requests via MCP Server Proxy->>API: Forward request to backend API API-->>Proxy: Return analysis results Proxy-->>AIAgent: Return results to AI Agent

Tech Stack

Frontend

A Django web app provides an intuitive UI for analysis. Why Django?

  • less boilerplate
  • faster prototyping
  • batteries-included (auth, forms, routing)
GitHub Repository

API

A FastAPI service handles the computational tasks. Why FastAPI?

  • Fast, easy to write
  • auto-docs with Swagger/OpenAPI
  • GraphQL wrapper for flexible data querying
GitHub Repository
Docker Hub
GraphQL Playground
Swagger
ReDoc
OpenAPI Spec

Compute Package

Python package implements statistical models. Why package?

  • package once, use everywhere (API, CLI)
  • Easy to publish and integrate with CI/CD
GitHub Repository
Docker Hub
PyPI Package
ReadTheDocs

Database

A Postgresql database stores pipeline run metadata and model results. Why postgres?

  • scales
  • good Python libraries (psycopg2, SQLAlchemy)
Home page waves photo by Thierry Meier under Unsplash License