Statistical Inference And Application Of Discrete Choice Model

The improvement of China's statistical system and the rapid development of the Internet allow researchers to have more accessibility to micro data,especially micro panel data.Therefore,the theoretical and empirical research on individual choice behavior by using micro panel data is becoming a focus of researchers' attention and one of the forefront issues.The nonlinear discrete choice model is usually adopted to study the individual choice behavior for its distinctive structure,which makes it irreplaceable by the traditional estimation method of the linear model.Compared with that of linear model,the estimation method of the nonlinear discrete choice model is more complicated,and many aspects are left for discussion in its theoretical research.In this paper,we study the parameter estimation of discrete choice model in theory,and analyze the application of discrete choice model in our research through the actual data.In this paper,our main research work and innovations are described as follows:(1)Given the dynamic binary panel choice logit models with individual heterogeneity,and when the time T is fixed,individual N approaches to infinity,we propose a new estimation method.The simulation results show that the proposed estimator performs very well with small samples.In addition,theoretically,we prove the consistency and asymptotic normality of the estimator.Compared with many estimation methods of the existing dynamic binary logit model,the simulation shows that our estimation has smaller mean deviation and root mean square error.(2)Given the multinomial choice model with individual and option heterogeneity,when the time T is fixed and individual N approaches to infinity,and the error term does not specify a particular distribution,we propose a semi-parametric method to estimate the multinomial choice model,and deduce the large sample property of the estimator,which is proved to be consistent in theory.Compared with the maximum likelihood estimation method,our estimation method turns out to be more robust.(3)Given the random regret minimization model,we use the rank-ordered choice data to estimate the parameters of the model.We find that the efficiency of the estimator increases significantly with the increase of the order choice data length,and we prove this conclusion in theory.Compared with the standard random regret minimization model,the simulation shows that the estimator of the rank-ordered random regret minimization model has smaller mean deviation and root mean square error.(4)This paper proposes a consideration set model.In the first stage,the choice set is reduced to a smaller set to form the consideration set by using some elimination rules,and in the second stage,the optimal product is selected by the consumers.The simulation results show that our consideration set model is superior to the single-stage multinomial logit model with random utility and the random regret minimization model.In addition,we also give a criterion to determine the size of the consideration set.Under certain conditions,our method can accurately predict the size of the consideration set.(5)As for the empirical analysis,we have done the following research work.First of all,this paper uses the panel data from consumer's purchase of washing products,and study consumer's purchase habits through the binary dynamic logit model.We find an obvious inertia exists in consumers' multiple purchases,that is,the last purchase decision significantly affects the next purchase decision;then,we use the panel data from households' purchase of margarine,and study the consumers' purchase behavior of this product through the rank-ordered random regret minimization model;finally,we use the panel data from households' purchase of tomato sauce to study the consideration set model,and find that the promotion method of featured advertising has a significant impact on the composition of consumers' consideration set.