Loading...

Estimating a Mixed Profile MDCEV:Case of Daily Activity Type and Duration

Shamshiripour, Ali | 2015

647 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 48261 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Samimi, Amir
  7. Abstract:
  8. The closed-form formulation of MDCEV has become popular in past years, as it accounts for satiation effects through a novel consumption-based utility structure. The model, however, suffers from an “empirical identification” problem. The problem is mainly due to entangled roles of two model parameters in capturing satiation effects. Thus far, the best known solution to the problem is fixing either or both of the parameters to pre-defined values and estimate a restricted satiation profile. This paper, presents a new hybrid optimization algorithm (named HEMPME) to eliminate the need for imposing such restrictions, and disentangle the two effects through estimating a “mixed profile” model. HELPME benefits from a coarse-to-fine search strategy, in which a customized version of Electromagnetism-like (EML) meta-heuristic precedes a gradient-based approach, like BFGS. The Atlanta Regional Travel Survey is used to empirically analyze performance of HELPME as well as significance of MDCEV accuracy improvement. To this end, four MDCEV models are developed for nine outdoor activity types and durations, on 11,121 observations. The models, employing different profiles, use identical explanatory variables. As evidenced by results of model accuracy tests, the mixed-profile MDCEV outperforms conventional models. According to the results, out-of-sample prediction power is evolved up to 11 percent in a 90-percent significance level. In addition, goodness-of in-sample fit is subject to an improvement in a 99-percent level of significance
  9. Keywords:
  10. Maximum Likelihood Estimation ; Optimization ; Electromagnetic ; Activity ; Joint Decision ; Hybrid Electromagnetism-Like Paradigm for Maximum-likelihood Estimation (HELPME)Algorithm

 Digital Object List

 Bookmark

No TOC