In this paper, we study the relationship among schooling, youth employment and youth crime. The framework, a multinomial discrete choice vector autoregression, provides a comprehensive analysis of the dynamic interactions among a youth's schooling, work and crime decisions and arrest and incarceration outcomes. We allow for observable initial conditions, unobserved heterogeneity, measurement error and missing data. We use data from the NLSY97 on black male youths starting from age 14. The estimates indicate important roles both for heterogeneity in initial conditions and for stochastic events that arise during one's youth in determining outcomes as young adults.