TECHNICAL PORTFOLIO // ENGINEERING & MODELING

Pairs Trading with Machine Learning on B3 (USP/ESALQ & Revista E&S)

public

MBA thesis in Data Science and Analytics at USP/ESALQ and scientific paper published in Revista Estratégias e Soluções (Rev. E&S). Implement...

Organization / Context: USP / ESALQ
Timeline: 2024 - 2026
Domain Category: Quantitative Finance
Technologies: Python, Statsmodels, Scikit-Learn, Pandas, NumPy, Matplotlib

Executive Overview

MBA thesis in Data Science and Analytics at USP/ESALQ and scientific paper published in Revista Estratégias e Soluções (Rev. E&S). Implemented a rigorous empirical benchmark comparing classical Engle-Granger econometric cointegration models against unsupervised and supervised Machine Learning algorithms (PCA + DBSCAN) for statistical arbitrage pairs trading on B3 equities.

Architecture & Data Flow

The diagram below illustrates the sanitized high-level component topology and data processing pipeline:

flowchart TD
  A[Dados Históricos B3] --> B[Pré-processamento & Alinhamento Temporal]
  B --> C1[Filtros Econométricos Engle-Granger / ADF]
  B --> C2[Classificadores Supervisionados Machine Learning]
  C1 --> D[Mecanismo de Backtesting & Sinais de Arbitragem]
  C2 --> D
  D --> E[Métricas de Risco & Retorno: Sharpe / Drawdown / Sortino]

Key Engineering Highlights

  • Peer-Reviewed Journal Publication: Research article published in Revista Estratégias e Soluções (Rev. E&S, ISSN 2675-6528, DOI: 10.22167/2675-6528-202602302) with open access and walk-forward empirical validation.
  • Rigorous Empirical Benchmark: Comparative analysis between classical econometric cointegration models and Machine Learning pipelines (PCA + DBSCAN).
  • Portfolio Backtesting: Historical backtesting on B3 equities with 202.48% cumulative return, 0.85 Sharpe ratio, and 4.29% annualized volatility, above the benchmark.
  • Academic Defense: Approved by the USP/ESALQ examination board.

References & Verification