2019 Fiscal Year Research-status Report
A sample selection model with a monotone selection correction function
| Project/Area Number |
19K13666
|
|---|
| Research Institution | University of Tsukuba |
|---|
Principal Investigator |
YU ZHENGFEI 筑波大学, 人文社会系, 助教 (40774758)
|
|---|
| Project Period (FY) |
2019-04-01 – 2021-03-31
|
| Keywords | Shape restrictions / Semiparametric / Sample selection model / Ordered-response model / Tuning parameter free |
|---|
| Outline of Annual Research Achievements |
This research incorporates a monotonicity restriction to the sample selection model that has been widely used in empirical researches of economics. The research proposes a dependence condition on the latent errors that is sufficient to yield a monotone control function. The monotonicity of the control function is a feature shared by the celebrated Heckman selection model that assumes joint normal errors and other parametric selection models that specify non-normal distributions (for example, the joint t-distribution) for the errors.
Then the research proposes a novel semiparametric estimator that takes into account the monotonicity restriction and establishes its root-n consistency and asymptotic normality. This shape-restricted estimator does not rely on parametric distributions assumptions and meanwhile frees the practitioners from choosing any tuning parameters. This framework is particularly appealing in the scenarios where researchers have certain prior knowledge regarding the dependence between latent errors. Monte Carlo simulations and an empirical example confirm good performances of the shape-restricted estimator. The proposed estimator is also fast to compute and can be implemented using existing R packages. As extensions of the current research, the shape-restricted estimation procedure is also applied to the ordered response model and the accelerated failure model with log-concave errors.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
The research has been generating desired results. One of the main advantages of the proposed estimation procedure is that it does not require practitioners to choose any running parameters. This practical benefit poses a theoretical challenge, that is, to handle the two-step estimation with the generated regressor when the estimated control function is piece-wise constant and thus not differentiable. Fortunately, this technical problem can be solved by combining the characterization of isotonic regression and the the empirical process theory. This clears the major task in the theoretical aspect of the research. In terms of implementation, the proposed estimator performs fairly well in Monte Carlo simulations and in real-data examples. A paper produced by the early stage of this research has been accepted by The Econometrics Journal.
|
| Strategy for Future Research Activity |
The future plan will focus on the following issues: 1. From the theoretical perspective, the requirement of sub-exponential tails on the error terms is to be replaced by weaker moment conditions. 2. Other shape restrictions such as the convexity/concavity are going to be investigated. 3. The shape-restricted estimators is going to be compared to kernel estimators with various bandwidth selectors. Also, papers produced by the researches will be submitted to academic journals and presented at conference to receive comments.
|
| Causes of Carryover |
A purchase plan was postponed due to the delay in the publication of some research handbooks.
|
Research Products
(3 results)
