This paper investigates the weekly evolution of skills as measured by unique data from a widely-emulated early childhood home-visiting program in rural China. The design of the study avoids input endogeneity issues and lack of comparable measures of skills that plague previous studies. Skills, nominally classified as the same, in fact, do not appear to share a common unit scale across levels. They are produced by skill-lifecycle-stage-specific learning processes. A novel dynamic stochastic skill production model for multiple skills is developed, aligning with empirical evidence. The model explains the "fadeout" of measures of learning through forgetting or depreciation of skills.

This paper estimates the causal impacts on child skills and the mechanisms producing these impacts using data from a randomized control experiment. We study a widely emulated early-childhood home visiting program and show the feasibility of replicating it at scale. We go beyond reporting treatment effects as unweighted item scores and assess item difficulties. To interpret treatment effects, we estimate individual-level latent skills and compare treatments and controls. The program substantially improves multiple skills. We decompose the source of treatment effects and find that enhancements in latent skills explain most of the conventional treatment effects for language and cognition.

We model the time-varying probability of consumption disasters with international risk interactions and estimate the model using national accounts data of 42 countries back to 1833. The estimated world and country-specific disaster probabilities accord well with historical macroeconomic disasters. A match of the equity premium requires a relative risk aversion coefficient of approximately 5, which is significantly lower than previous estimates. Furthermore, the model provides notably better fits for equity volatility compared with alternative rare-disaster models. Finally, the disaster probability index estimated from the model demonstrates significant out-of-sample predictive power over long horizons, performing well not only over time but also across countries.