Flexible estimation of skill formation models

TL;DR

This paper introduces a flexible, likelihood-based estimation method for skill formation models that improves accuracy and computational efficiency, accommodating nonlinearities and measurement complexities.

Contribution

It presents a novel iterative likelihood approach that estimates latent distributions and incorporates model restrictions over time, outperforming existing methods.

Findings

Outperforms existing estimators in simulations and empirical tests.

Reduces computational complexity compared to traditional methods.

Provides more accurate and less biased estimates of skill formation models.

Abstract

This paper examines estimation of skill formation models, a critical component in understanding human capital development and its effects on individual outcomes. Existing estimators are either based on moment conditions and only applicable in specific settings or rely on distributional approximations that often do not align with the model. Our method employs an iterative likelihood-based procedure, which flexibly estimates latent variable distributions and recursively incorporates model restrictions across time periods. This approach reduces computational complexity while accommodating nonlinear production functions and measurement systems. Inference can be based on a bootstrap procedure that does not require re-estimating the model for bootstrap samples. Monte Carlo simulations and an empirical application demonstrate that our estimator outperforms existing methods, whose estimators can be substantially biased or noisy.