This dissertation consists of three independent essays on earnings dynamics, educational production function, and retirement. Each chapter explains labor supply and human capital formation from a life-cycle perspective. In the first chapter, I investigate how two different kinds of uncertainty jointly affect young workers' decisions. This paper introduces the possibility of multidimensional learning about worker ability and job match quality into a model of work decisions. This mechanism has a unique prediction, negative sorting into job mobility that fades away over time, which is verified in the NLSY79 data if the AFQT score carries over some information unused by workers and employers. I estimate the structural model, which also has flexible skill accumulation, by indirect inference. From simulation results on earnings dynamics, I find that the contribution of job shopping to average earnings growth is higher than previous estimates; also, individual heterogeneity in earnings growth is mostly explained by the process of resolving uncertainties. In the second chapter, which is joint work with Keunkwan Ryu, we estimate the effects of high school class size on college entrance exam scores, using Korean administrative data. For the identification, we exploit quasi-experimental variation in class size arising from distinct institutional settings in Korea: especially, students are separately educated by major from grade 11 with different class sizes between majors. By using multi-level differencing and instrumental variable techniques, we find the effects of high school class size reduction on the test scores are positive but small. In the third chapter, I examine the effects of life expectancy on retirement and related decisions. I construct a structural model which has a realistic description of complicated dynamic incentives facing the elderly, including Social Security. Furthermore, individual heterogeneity in survival beliefs are flexibly modeled, directly using subjective survival probabilities in the Health and Retirement Study (HRS) data. The estimated model suggests that many people in the data believed their wealth was over-annuitized; they would have chosen to work and save less if their average life expectancy had increased. This result partially explains the early retirement puzzle in the last century.