Browsing by Author "Villegas Rodríguez, Andrés Fernando"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Essays on simulation methods
    (2018) Villegas Rodríguez, Andrés Fernando; Bobenrieth H., Eugenio S.; Pontificia Universidad Católica de Chile. Facultad de Agronomía e Ingeniería Forestal
    This dissertation consists of two essays in which I use simulation methods to study the structural parameters estimates from econometric models considering the complexity of water and commodity markets. In the first chapter, I implement the double bootstrap to non-parametric Data Envelopment Analysis with the purpose to estimate the efficiency of Chilean water and sewerage companies. The relevance of applied this bootstrap technique, is that allows statistical inferences that cannot be drawn directly from such non-parametric model. This feature is important in the framework of water utilities performance comparisons since it is well-known that several exogenous variables influence the water utilities efficiency. My results show that the ranking of water companies changes notably whether efficiency scores are computed applying conventional or double bootstrap DEA models. Moreover, I found that the percentage of non-revenue water and customer density are factors that influencing the efficiency of Chilean water and sewerage companies. In the second chapter, I design a Monte Carlo experiment in the context of storage model to compare finite sample performance of the Simulated Methods of Moments estimator of Duffie and Singleton (1993), the Indirect Inference estimator of Gourieroux et al. (1993), the Efficient Method of Moments estimator of Gallant and Tauchen (1996), the Pseudo Maximum Likelihood estimator (PML) of Deaton and Laroque (1995), The Conditional Maximum Likelihood estimator of Cafiero et al. (2015) and the Unconditional Maximum Likelihood of Gouel and Legrand (2017). My results suggest that for parameterizations that imply low average storage and frequent stockouts, the PML estimator for small sample presents low bias and is more efficient than Simulations estimators. However, for parameterizations that imply a more significant role of storage, the Simulations estimators present bias that decrease with sample size increase, while the PML estimator biases do not disappear but instead tend tostabilize. I prove theoretically and numerically that Maximum Likelihood estimator isconsistent and achieves better small sample performance than the others.