Identification and Estimation of Dynamic Structural Models with Unobserved Choices
summary
Claims that dynamic discrete-choice models remain identifiable when actions are unobserved choices by using time variation in a finite-horizon setting as an exclusion restriction; derives closed-form recovery of the two-action state-response functions (M1, M0) from observable moment differences and then back out choice probabilities and the shock distribution. Implements a sieve maximum likelihood estimator, shows good Monte Carlo performance, and fits a 142-governor panel to find higher shirking probability near term end.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:25Motivation
- 3:20Research Question
- 4:02Existing Literature on DDC
- 10:15A Basic Model
- 16:06Dynamic Process of
- 21:35Intuition
- 24:05Assumption on the State Transition Process
- 26:54Identification
- 31:29Closed-Form Solution
- 34:30Ordering Assumption
- 36:08Outline
- 36:53Extensions
- 39:21Discussion - Assumption 8
- 41:49Summary Statistics
- 42:15Probabilities of Shirking
- 42:26Summary of Findings
related talks
48mLearning Preferences with Latent State
30mBuying Data Over Time: Approximately Optimal Strategies for Dynamic Data-Driven Decisions
30mNon-parametric Causal Inference in Dynamic Thresholding Designs
28mConfounding-Robust Policy Evaluation In Infinite-Horizon Reinforcement Learning
45mPreference Alignment from Choices and Response Times
18mMichaelWeylandt
43mLearning from Dynamics
09mTowards Credible and Effective Data-Driven Decision-Making