Bias-Corrected Peaks-Over-Threshold Estimation of the CVaR (UAI 2021)

https://proceedings.mlr.press/v161/troop21a.html

This code will generate all data used in simulations presented in the paper and produce plots. Parameters can be adjusted to produce results for other scenarios.

Requirements:

• Python v3.xx
• Modules:
  • numpy
  • scipy
  • matplotlib
  • multiprocessing

Replicating results

Executing run_sim.py will generate all samples from all distributions given in the paper, and compute estimated CVaRs (alpha=0.998) using UPOT, BPOT, and SA. All samples and CVaR values will be stored in the data folder. The following files can be used to reproduce plots in the paper:

• dist_plots.py Produces RMSE and absolute bias plots from the generated CVaR data
• coverage_prob.py Produces plots of coverage probability from the generated CVaR data
• asymp_var.py Comparison of UPOT and SA asymptotic variance for the Frechet distribution