The goal of this project was to understand and explore various pairs trading strategies on stocks and portfolios (stocks + risk-free assets). Further details of each of the project components are outlined in the specific README.md files in the respective folders. Here we present a high-level overview for help in navigation.
1. ML-based Trading
- Equal Weights Naive trading strategy
- MVP Optimization by selecting weights based on risk tolerance, expected return and sharpe ratio
- LSTM based prediction and trading
- LSTM + GRU based prediction and trading
- LSTM + GRU + Attention based prediction and trading
- LSTM + CNN based trading
2. Pairs Trading
- "Buying the yesterday's loser strategy"
- Logistic Regression Heuristic
- Decision Trees strategy with Feature Engineering
- Future Work: Hyperparameter Tuning, Ensembling, Cointegration, and Kalman Filters
3. Portfolio Construction and Risk Analysis
- Descriptive Statistics and Distributions Fitting
- MVP construction with and without shorting
- Tangency Portfolio construction with and without shorting
- Asset Allocation in various scenarios
- PCA and Factor Analysis [Valuable Insight: Found out that VISA and Mastercard were clubbed in a factor, and all the Tech companies were clubbed into a factor in our results]
- Risk Analysis through Value at Risk (VaR) and Expected Shortfall (ES)
- Fitting Copulas to model join distribution of returns [Spoiler: T-Copula was the best fit by a large margin because it was able to capture tail dependencies better than Gumbel Copula]
Base packages needed in order to execute the logic in this repository are listed with respective versions in requirements.txt.