Quantitative Portfolio Analysis & Deep Learning Trading Signals

An end-to-end quantitative finance project analyzing a real 22-asset portfolio through classical time series methods, mean-variance optimization, and deep learning-based trading signal generation — built as a direct extension of Shumway & Stoffer's Time Series Analysis and Its Applications (Chapters 1-3). Data sourced from Yahoo Finance, WRDS/OptionMetrics, LSEG Machine Readable News, and Bloomberg Terminal.

---

Project Overview

Phase 1 — Buy & Hold Portfolio Analysis

• Log return computation and stationarity testing (ADF test)
• Annualized return, volatility, Sharpe ratio, Sortino ratio, max drawdown
• Correlation matrix and diversification analysis
• Benchmark comparison against S&P 500 (VOO)
• ACF/PACF analysis connecting to Shumway & Stoffer Ch. 1-3

Phase 2 — Portfolio Optimization

• Monte Carlo simulation of 10,000 random portfolios
• Analytical mean-variance optimization via scipy (SLSQP)
• Efficient frontier visualization
• Maximum Sharpe and minimum volatility portfolio construction
• Current vs optimized weight comparison

Phase 3 — Quant-Grade Signal Research (In Progress)

• Feature library construction: price, volume, options, sentiment, macro
• Information coefficient (IC) analysis at multiple lags
• Signal decay and regime analysis
• XGBoost as primary model for tabular financial features
• LSTM and Transformer for sequential price modeling
• Walk-forward validation — no single train/test split
• Data sources: OptionMetrics via WRDS, LSEG Machine Readable News, Bloomberg Terminal exports

Phase 4 — Risk Management & Position Sizing

• Kelly criterion for optimal position sizing based on model edge
• Volatility targeting — scale positions by predicted volatility
• Maximum drawdown constraints and stop-loss framework
• Walk-forward backtest with explicit transaction cost modeling

Phase 5 — Black-Litterman Optimization with Deep Learning Views

• Market equilibrium prior from market cap weights
• Deep learning return forecasts as quantitative investor views
• Confidence intervals derived from model validation error
• Black-Litterman adjusted expected returns fed into mean-variance optimizer
• Dynamic portfolio rebalancing pipeline

---

Motivation

This project bridges classical time series econometrics and modern quantitative finance. ARIMA models from Shumway & Stoffer Ch. 3 serve as interpretable baselines, while XGBoost and deep learning architectures (LSTM, Transformer) extend the analysis into nonlinear territory. The ultimate goal is a fully data-driven portfolio management pipeline — from raw price data through signal research, deep learning forecasting, and Bayesian portfolio optimization — consistent with industry-standard quantitative research practice.

---

Portfolio

22 real holdings spanning:

• High-growth speculative tech (QBTS, OKLO, PLTR, SOLS)
• Blue-chip equities (HON, IBM, NVDA, MSFT, META, AMZN)
• Diversified ETFs (VOO, VTI, VEA, VB, VO, VWO, VCIT)
• Sector-specific plays (QTUM, SKYT, JD, WDAY, PSKY)

---

Tech Stack

• Python 3.10+
• yfinance — baseline price and volume data
• pandas, numpy — data manipulation
• matplotlib, seaborn — visualization
• statsmodels — ARIMA, ADF testing, ACF/PACF
• scipy — portfolio optimization
• PyTorch — LSTM and Transformer models
• xgboost — gradient boosted trees for tabular signal modeling
• scikit-learn — preprocessing, evaluation, walk-forward validation
• WRDS/OptionMetrics — historical options data (IV, put/call ratio, skew)
• LSEG Machine Readable News — professional news sentiment scores
• Bloomberg Terminal — implied volatility history, options flow

---

Repository Structure

quantitative-portfolio-analysis/
│
├── notebooks/
│   ├── 01_portfolio_analysis.ipynb         # Phase 1
│   ├── 02_portfolio_optimization.ipynb     # Phase 2
│   ├── 03_deep_learning_signals.ipynb      # Phase 3 (prototype)
│   ├── 04_qbts_feature_study.ipynb         # Phase 3 feature research
│   ├── 05_quant_signal_research.ipynb      # Phase 3 redesign (in progress)
│   ├── 06_risk_management.ipynb            # Phase 4 (planned)
│   └── 07_black_litterman.ipynb            # Phase 5 (planned)
│
├── data/                                   # Cached and exported data
├── models/                                 # Saved model weights
├── results/                                # Charts, tables, backtest results
├── requirements.txt
└── README.md

---

Data Sources

Source	Access	Data
Yahoo Finance	Free, remote	Price, volume
WRDS/OptionMetrics	Stevens HFSC	Historical options, IV, put/call
LSEG Machine Readable News	Stevens HFSC	News sentiment scores
Bloomberg Terminal	Stevens HFSC (on-site)	IV history, options flow
S&P Capital IQ	Stevens HFSC	Earnings surprise, short interest

---

Current Status

Phase	Status
Phase 1 — Portfolio Analysis	Complete
Phase 2 — Portfolio Optimization	Complete
Phase 3 — Signal Research	In Progress — awaiting WRDS access
Phase 4 — Risk Management	Planned
Phase 5 — Black-Litterman	Planned

---

Getting Started

git clone https://github.com/YOUR_USERNAME/quantitative-portfolio-analysis.git
cd quantitative-portfolio-analysis
pip install -r requirements.txt
jupyter notebook notebooks/

---

References

• Shumway, R.H. & Stoffer, D.S. (2017). Time Series Analysis and Its Applications. Springer.
• Markowitz, H. (1952). Portfolio Selection. Journal of Finance.
• Black, F. & Litterman, R. (1992). Global Portfolio Optimization. Financial Analysts Journal.
• Hochreiter, S. & Schmidhuber, J. (1997). Long Short-Term Memory. Neural Computation.
• Vaswani et al. (2017). Attention Is All You Need. NeurIPS.
• Chen, T. & Guestrin, C. (2016). XGBoost: A Scalable Tree Boosting System. KDD.