Hypothesis/Goals
General Goals
- Strategy must have successful held out performance. This is defined as meeting the (naively set) thresholds of Sharpe ratio > 2.0, Alpha (Annualized) > 1.5, and Max Drawdown < 15% for all held-out periods used in the walk-forward analysis. There must be a minimum of 30 completed trades overall across all periods.
- Successful implementation of common strategy classes to analyze the performance of each class. Successful implementation's definition varies because of the intention of the user, but broadly speaking, means the strategy executes as the user intends it.
- Successful classification of market regimes as strategy performance has been proven to depend on market regime. Successful classification means the strategy performances well upon the assumption the strategy itself is implemented correctly. It also means the classified regime matches with historical data trends.
- Efficient implementation of all operations. Operations are data extraction and organization, strategy implementation, and regime classification. Efficient implementation means at maximum linear-log performance for time complexity.
-
All operations must be scalable. Initial dataset will have a maximum of 5000 data points.
List of Strategies and Filters
Strategy classes
Each class is determined by various parameters associated with that class. Standard definitions are used. For further details, see here
- Cointegration
- Mean reversion time based
- Mean reversion cross-sectional
- Momentum, time based
- Momentum, cross-sectional
- Learning, supervised and unsupervised, based models
- Ensemble models
Statistical Approaches
- Beta neutrality/analysis
- Dollar neutrality
- Sector neutrality/analysis
- PC neutrality/PCA
- Hidden regime variables
ML Approaches
- Linear regression
- Logistic regression
- Lasso/Ridge/Elastic Net
- LDA, QDA
- Kernel methods
- Neural Networks/Transformers
- Decision Trees/Forests
- Support Vector Machines
- Bayesian Filtering/Kalman Filters/HMMs
Tools
Software Packages
- Yfinance
- Pandas Series/Dataframe
- Optuna
- VectorBt
- Multiprocessing.pool
- Statsmodels
- DuckDB
- Plotly
- Numpy
- PyTorch
- scikit-learn
AI Agents/Programming Workflows/Environments