An estimator is incentive-compatible (for a given prior belief regarding the model's true parameters) if it does not give an agent an incentive to misreport the value of his covariates. Eliaz and Spiegler (2019) studied incentive-compatibility of estimators in a setting with a single binary explanatory variable. We extend this analysis to penalized-regression estimation in a simple multi-variable setting. Our results highlight the incentive problems that are created by the element of variable selection/shrinkage in the estimation procedure.
- Incentive-compatible estimators
- Online platforms
- Penalized regression