Conditional value-at-risk (CVaR) Portfolio Optimization in High Dimensions.
Our algothim is implemented in opt_algo.py; utils.py contains utils functions.
- Estimation error bound:
estimation_err.py - Simulation:
run_simu.pywhich requires a sample covariance matrixSigma.npy. - Real data:
run_real.py, which requires the S&P stock data. See the subsection at the end for more details about the data.
The script summary.py outputs the results and reproduces the figures.
This code is delivered via the files described above.
Python (version 3.6 or later) is required to run the files, and it has only been tested on the Linux (6 Xeon(R) CPU E5-2690 @ 2.90GHz and 128 GB memory) and the MacOS platforms.
Python packages to run reproducible code:
- cvxopt=1.2.7
- cvxpy=1.2.0
- joblib=1.1.0
- nonlinshrink=0.7
- numba=0.51.1
- numpy=1.21.2
- pandas=1.3.4
- statsmodels=0.13.2
- scikit-learn=1.0.2
- scipy=1.7.1
- tqdm=4.62.3
The S&P dataset and the constituent information are proprietary, purchased through WRDS and Siblis Research, Inc. The contract heavily restricts even characteristics of the data (for example, information on stock prices that appear in the datasets). Please refer to repo for obtaining the S&P 500 dataset.