Incluant une Journée industrielle de l'optimisation
HEC Montréal, 7  9 mai 2012
JOPT2012
HEC Montréal, 7 — 9 mai 2012
TD7 Analyse de données / Data Analysis
8 mai 2012 15h30 – 17h10
Salle: Sony
Présidée par Vicente CollSerrano
4 présentations

15h30  15h55
Efficiency Evaluation with Imprecise Data  An application to the Spanish Textile Industry
Efficiency results obtained by applying conventional DEA models are usually used for Decision Making. It is common to rank the assessed units and to show the percentages of improvement that should be promoted. However, these assessments are based on the assumption of certain data and, as noted by some authors in the academic literature, this assumption is not always acceptable. In these cases, and when we search the robustness of our statements, it is preferable to consider some kind of imprecision in the data. In this study we propose a possibilistic model, which we apply to the Spanish textile firms.

15h55  16h20
GARCH Parameter Estimation using Chebyshev Polynomials
The objective of the paper is to calibrate the parameters in a GARCH model from real data. The usual approach is to minimize a meansquare dollar objective function as in Duan and Zhang (2009), Christoffersen (2004) among others. We propose a method (which is applicable to a general GARCH model) based on Chebyshev interpolation. In a first step, after choosing a particular GARCH specification, we build a matrix of contract prices calculated with enough precision for different parameters values including initial volatility and riskfree interest rate. This has to be done just once. Afterwards, LeastSquares estimation is carried out using Chebyshev interpolated values of the objective and Jacobian functions. The method is fast and performs the calibration in a fraction of the time that a direct implementation needs.

16h20  16h45
Nonsmooth Nonconvex Approach to Bilevel Programs in Machine Learning
We develop a nonsmooth nonconvex optimization approach for solving bilevel programs resulting from support vector machine model selection problems. By expressing lower level problems as penalized nonsmooth nonconvex constraints, the problem can be optimized using a decomposable Lagrangian method. The proposed approach is more efficient than prior bilevel approaches.

16h45  17h10
Inexact Sample Average Approximation for the Stochastic ConFL Problem
We consider the ConFL problem where there is uncertainty on the assignment costs, and propose an inexact sample average approximation (SAA) approach. This inexact SAA relies on a heuristic to solve the sample problems yet yields tight confidence intervals at a fraction of the time required by the standard SAA approach.