Search Results

Author: Nigmatullin, Eldar Ayratovich
Resulting in 1 citation.
1. Nigmatullin, Eldar Ayratovich
Estimation of Markov Decision Processes in the Presence of Model Uncertainty
Ph.D. Dissertation, The University of Wisconsin - Madison, 2003.
Cohort(s): NLSY79
Publisher: UMI - University Microfilms, Bell and Howell Information and Learning
Keyword(s): Adolescent Fertility; Bayesian; Data Analysis; Demography; Markov chain / Markov model; Modeling; Neighborhood Effects; Statistical Analysis

The Bayesian approach provides a coherent framework for accounting for model uncertainty and it takes form of Bayesian model averaging (BMA) when there are finitely many models under consideration. In my first essay I develop the BMA technique for moment conditions models. My work extends the existing theory on BMA that has been limited to the parametric case. I develop a methodology, grounded in information theoretic and Bayesian arguments, which allows implementation of the BMA technique by means of nonparametric likelihood methods in situations when all information that is available comes in the form of moments conditions. Moment conditions models arise naturally in the context of estimation of Markov decision processes because first-order conditions for agents' optimization problem produce population moment conditions. I consequently consider an application of my methodology to the problem of optimal portfolio choice in the presence of partial predictability of assets returns. In my second essay I analyze the effects of neighborhood interactions on the pre-marital fertility decisions by women of NLSY79 in the framework of Cox proportional hazards. The empirical analysis finds that out-of-wedlock fertility dynamics vary systematically with neighborhood characteristics when counties of residence are taken as individuals' neighborhoods. I find the contextual effects to be well pronounced. The data do not reveal the presence of significant endogenous social interactions. The empirical analysis indicates the presence of significant uncertainty in the choice of explanatory variables. Bayesian model averaging is applied both to account for model uncertainty and the subsequent inference.
Bibliography Citation
Nigmatullin, Eldar Ayratovich. Estimation of Markov Decision Processes in the Presence of Model Uncertainty. Ph.D. Dissertation, The University of Wisconsin - Madison, 2003..