« Back to Results
Pennsylvania Convention Center, 106-B
Hosted By:
Econometric Society
Discrete Choice Models in Action
Paper Session
Sunday, Jan. 7, 2018 1:00 PM - 3:00 PM
- Chair: Aviv Nevo, University of Pennsylvania
A Semi-nonparametric Estimator for Random Coefficient Demand Models
Abstract
This paper proposes a two-step semi-nonparametric estimator for the random coefficients BLP model that is computationally easy to implement and allows for nonparametric specification of preference heterogeneity. In the first step, we transform the original demand system into a partial linear model that can be estimated easily by two-stage least-squares (2SLS). In the second step, we substitute the estimated mean utility from the first step back into the original demand system and estimate the distribution of random coecients nonparametrically. We develop the asymptotic theory for our estimator and demonstrate the its effectiveness with simulations and real data applications.Three-step CCP Estimation of Dynamic Programming Discrete Choice Models with a Large State Space
Abstract
The existing estimation methods of structural dynamic discrete choice models mostly require knowing or estimating the state transition distributions. When some or all state variables are continuous and/or the dimension of the vector of state variables is moderately large, the estimation of state transition distributions becomes difficult and has to rely on tight distributional assumptions. We show that state transition distributions are indeed not necessary to identify and estimate the flow utility functions in the presence of panel data and excluded variables. A state variable is called excluded variable if it does not affect the flow utility but affects the choice probability by affecting the future payoff of current choice. We propose a new estimator of flow utility functions without estimating or specifying the state transition law or solving agents' dynamic programming problems. The estimator can be applied to both infinite horizon stationary model or general dynamic discrete choice models with time varying flow utility functions and state transition law.Sufficient Statistics for Unobserved Heterogeneity in Structural Dynamic Logit Models
Abstract
We study the identification and estimation of panel data structural dynamic logit models with a nonparametric specification of the joint distribution of time-invariant unobserved heterogeneity and observable state variables, i.e., a fixed-effects structural dynamic logit model. We consider multinomial models with two endogenous state variables: the lagged decision variable, and the duration in the last choice. This class of models includes as particular cases important economic applications such as models of market entry-exit, occupational choice, machine replacement, inventory and investment decisions, or demand of differentiated storable products. The main challenge is to find a sufficient statistic that (i) controls for the contribution of the fixed-effect not only to current utility but also to the continuation values in the forward-looking decision, and (ii) still leaves information on the parameters of interest. We characterize the minimum sufficient statistics for the structural parameters. Based on our identification results, we propose a Conditional Maximum Likelihood estimator. We apply this estimator to the bus engine replacement data in Rust (1987) and to the consumer scanner data on demand of a differentiated storable product in Erdem, Imai, and Keane (2003).JEL Classifications
- C25 - Discrete Regression and Qualitative Choice Models; Discrete Regressors; Proportions; Probabilities
- C14 - Semiparametric and Nonparametric Methods: General