A genetic method of LAD estimation for models with censored data

The objective function of the censored least absolute deviations estimation is neither differentiable, nor convex. Therefore, it is very difficult to get the global optimal solution, especially when the censoring is heavy. To improve the computational efficiency, a genetic algorithm is designed, taking the asymptotic property of the objective function into consideration. By the proposed algorithm, the global optimal solution can be obtained with a very large probability. Simulation results show that the algorithm works well, even when a lot of censoring is present.