Research

The Shift from Traditional Pensions to 401(k)s: Retirement Risks and the Timing of Retirement with Rosemary Kaiser and Xiaohui Sun 

U.S. retirement coverage has shifted from defined benefit (DB) to defined contribution (DC) plans. We build a life-cycle model in which DB and DC plans differ in exposure to longevity and return risk and households respond to the additional risks by saving more and working longer. These two insurance margins interact with one another: additional saving before retirement reduces the need for precautionary labor supply at older ages. The model matches observed differences in retirement timing across plan types and decomposes the effects of the transition into plan structure and generosity, return risk, and longevity risk. Return risk is the main driver of delayed retirement, while longevity risk is the main source of the welfare loss attributable to uncertainty. We then evaluate annuitization as a way to improve the insurance provided by accumulated assets. Allowing retirees to convert assets into actuarially fair annuities raises welfare and induces earlier retirement, with larger gains after the shift toward DC plans.

Unequal Recovery from Recessions: Skills Learning Among Young Workers

Graduating during a recession leads to wage losses—about 2% per one-point rise in the unemployment rate—with effects that diminish but persist for over a decade. However, these averages mask substantial heterogeneity: using NLSY data and quantile regression, I find that young workers in higher residual wage quantiles recover significantly faster. This heterogeneity can be attributed to some unobservable characteristics of workers. To quantify the role of endogenous skill investment in recovery and the sources of heterogeneity, I develop a life-cycle model featuring two-dimensional Ben-Porath skill accumulation with workers differ along three unobserved dimensions relevant for wage determination: true abilities, initial self-beliefs about true abilities, and initial stock of skills. This model extends the classical Ben-Porath model by accounting for workers' uncertainties about their skills accumulation and introducing aggregate fluctuations in the demand for skills. Counterfactual analysis shows that the endogenous change in workers' investment in skills accumulation due to the recession explains up to 23% of wage losses every year throughout the recovery period and 92% from the recovery periods even after the recession is over. Heterogeneity in wage losses is primarily driven by differences in true skill accumulation ability, with low-ability workers suffering the most—mainly due to slower recovery in the time they devote to skill accumulation.

Working with AI: Required Skill Bundles and Wage Premium in the Chinese Labor Market with Fa-Hsiang Chang

This paper studies how artificial intelligence (AI) skill requirements are associated with skill bundles and posted wages in China. We define AI skill requirements as job-ad requirements to use completed AI tools, integrate AI models or systems into products and business processes, or develop, train, optimize, and deploy AI models. Using more than eight million postings from 51job.com, we construct a context-aware taxonomy that distinguishes AI user, applied technical, and core technical skills, and identify seven categories of nontechnical skills from job-description text. AI skill requirements remain rare but are concentrated in highly educated and technically intensive jobs. Postings requiring any AI skill offer a conditional wage premium of about 17 percent, with the premium increasing from AI user to applied technical and core technical skills. Postings requiring AI capabilities are also more likely to require cognitive, executive, strategic, and self-management skills, especially among AI user and applied technical jobs. However, wage interaction estimates do not show broad superadditive complementarity between AI and nontechnical skills. The results further reveal substantial industry heterogeneity in both AI-related skill bundles and wage premia.