SPORT

SPORT (Scalable Portfolio Optimization Research Tool) provides a scalable architecture to calculate the optimal asset allocation for multi-asset-class portfolios.

It incorporates several optimization packages (scipy.optimize, cvxopt). It provides a flexible syntax to formulate constraints and can handle a variety of target functions, including mean-variance, Sharpe ratio, volatility, risk parity, and maximum drawdown.

How to Run

1. Run batch_download_data.py to download daily price data to the data folder.
2. In the data/etf/ folder, run the shell script clean_yfinance_data.sh to clean up data.
3. Run btach_port_covar.py to load daily price data obtained in Step 1, and then generate covariance and correlation matrices.
4. Manually update the SEC_NM column of data/etf_covar_matrix.csv to enrich output information (if not updated, code will run without issue).
5. Manually update template_{cn,en,etf}_input.xlsx to generate inputs.
6. Run batch_port_main.py to generate optimal weights for various objectives and constraints; output will be saved to template_{cn,en,etf}_output.xlsx.
7. Run risk_analysis.ipynb for interactive risk analysis of any portfolio. View risk_analysis.html for the result.

TODO

implement more sophisticated methods of covariance estimation, including linear shrinkage, eigenvalue clipping, eigenvalue substitution, and rotationally invariant optimal shrinkage (based on Random Matrix Theory).

Class Documentation

security.py

An abstraction of single security, single index etc. that carries an ID and a list of attributes.

portfolio.py

A portfolio class that carries a list of Security objects and their covariance and weights.

function.py

1. constr_avg_max_drawdown(x: List[float], params_constr: Dict[str: object]) -> float: Calculate average maximum drawdown of a portfolio.
2. constr_risk(x: List[float], params_constr: Dict[str: object]) -> float: Calculate risk of a portfolio.
3. obj_avg_max_drawdown(x: List[float], params_obj: Dict[str, object]) -> float: Average maximum drawdown.
4. obj_neg_rtrn(x: List[float], params_obj: Dict[str, object]) -> float: Negative of portfolio return.
5. obj_neg_sharpe_ratio(x: List[float], params_obj: Dict[str, object]) -> float: Negative of the Sharpe ratio.
6. obj_qp(x: List[float], params_obj: Dict[str, object]) -> float: Objective function for quadratic programming (minimization).
7. obj_risk(x: List[float], params_obj: Dict[str, object]) -> float: Objective function for risk.
8. obj_risk_parity(x: List[float], params_obj: Dict[str, object]) -> float: Objective function for risk parity optimization https://en.wikipedia.org/wiki/Risk_parity.
9. util_covar_to_corr_matrix(covar_matrix: pd.DataFrame) -> pd.DataFrame: convert a covar matrix to a correlation matrix.
10. util_is_valid_covar(covar_matrix: pd.DataFrame) -> tuple: validate if a matrix is a valid covar matrix (symmetric and positive definite).
11. util_md_Qn(x: float) -> float: Qn function used in maximum drawdown.
12. util_md_Qp(x: float) -> float: Qp function used in maximum drawdown.

constraint.py

A factory class that generates constrains for portfolio optimization.

objective.py

A factory class that generates objective functions for optimizers.

optimizer.py

Optimization calculator.

covar_estimator.py

Estimator of covariance matrix and average returns of assets.

dao.py

Data access object class to get security attributes.