Mixture Quantile Estimation via Constrained Linear Regression

Back to main page
Case study background and problem formulations


The optimization problems are solved using standard Python libraries (e.g., SciPy) and advanced convex/mixed-integer solvers (e.g., GUROBI).
Please refer to the Environment and Python Framework Instruction.

PROBLEM: Mixture Quantile Estimation for Electricity Price Data

Minimize L1 / L2 error with Non-negativity, L1 Selection Penalty, and Smoothness Constraints.

————————————————————–
This case study demonstrates the formulation and solution of the Mixture Quantile (MQ) model applied to daily U.S. electricity prices. Unlike traditional Maximum Likelihood Estimation, this framework reduces distribution fitting to a constrained linear regression problem. The Python implementation utilizes parametric bases (Standard Gaussian, Student’s t), Tail-specific Generalized Pareto Distributions (GPD) for extreme price spikes, and non-parametric I-splines for flexible body fitting.

Data and solution in Python Environment

Problem Dataset# of Variables# of Samples
U.S. Electricity Prices (2001-2024)Python CodeSolution19 (Basis functions)79,651


References

• Peng, C., Li, Y. and S. Uryasev. Mixture Quantiles Estimated by Constrained Linear Regression. Annals of Operations Research. 2026, Accepted for publication. https://uryasev.ams.stonybrook.edu/wp-content/uploads/2026/04/Mixture_quantiles_constrained_linear_regression.pdf.