Procrastination is a widespread type of self-regulation failure that can have serious negative effects on people's health (e.g., because people delay or omit important health behaviors) and well-being. New 'e-coaching' technologies make it possible in principle to offer tailored support to individuals in their efforts to change their self-undermining behavior. In practice, however, such automated support is currently unfeasible because the causal mechanisms behind procrastination are complex and poorly understood. This paper presents a new agent-based model of procrastination that integrates insights from economic models about the dynamics of procrastination with psychological concepts that can help explain the behavior on an individual level. The model is validated by using 5-fold cross validation with simulated annealing to fit and test the parameters on an existing dataset on academic procrastination (n=293). Results show that the agent displays realistic behavior and that the model with the fitted parameters performs significantly better (p<0.01) than the model with randomly selected parameters.