Research Interests Computational Statistics, Econometrics, Financial Economics.Current research My current research focuses on building numerically and computationally efficient procedures for parameter estimation, state smoothing and filtering in non-linear and non-Gaussian state space models. PapersThe HESSIAN Method with Conditional Dependence (Job Market Paper) For state space models where the state vector is Gaussian and the observed vector need not be, we propose a close approximation of the state posterior density. Thus, we achieve higher numerical efficiency in posterior simulations with Markov chain Monte Carlo methods, given that acceptance probability are closer to 1, or with Importance Sampling methods due to the very low variance of importance weights. Illustrations with stochastic volatility models with the leverage effect (Harvey and Shephard,
Journal of Business and Economic Statistics, 1996), on artificial and real data, show that our procedure is more numerically and
computationally efficient than similar procedure in the literature, like that
of Omori, Chib, Shephard and Nakajima (Journal of Econometrics, 2007).
Generalized HESSIAN for Nonlinear NonGaussian State Space Model Abstract. This paper develops method for simulation smoothing and parameters estimation in nonlinear and non-Gaussian state space models. The state vector is required to be a first order Markov process but not necessary Gaussian. State and observation vectors may exhibit conditional independence or dependence. We illustrate using Asymmetric Stochastic Conditional Duration models (ASCD), a slight modified version of the stochastic conditional duration models with the 'leverage effect' introduced by Feng, Jiang and Song (Journal of Financial Econometrics, 2004). Simulation on artificial data and real data show that this method is numerically and computationally efficient. The method is based on a closed approximation of the conditional density of states given observations. The approximate density can be used as importance density in importance sampling or proposal density in Markov chain Monte Carlo methods for Bayesian posterior simulations. Our methodology is related to similar work in Djegnene and McCausland (unpublished manuscript, 2010) where the state vector is required to be Gaussian. Reputation and Endogenous Membership in Mixed Duopsony
This paper deals with organizational aspects of commodity markets. The liberalization of agriculture markets in Africa raises the issue of commodity risk management for small farmers. A necessary condition for small farmers to use financial derivatives markets to hedge their risk exposure is organizing themselves into viable farmer-owned cooperatives. The paper builds a model that explains why some small farmers in Africa fail to do so. |