Linear Approximation Using MOTAD and Separable Programming: Should It Be Done?

Linear approximation techniques have often been applied to nonlinear mathematical programming models for computational efficiency reasons. Price-endogenous agricultural sector models and risk models have found numerous applications. This article addresses the issue of approximation efficiency. Based on computational experience with a series of moderate and large-scale sector and risk models, it is concluded that direct nonlinear solution is more efficient than using linear approximations for sector and risk models having objective function nonlinearities. On the other hand, the experimental results indicate that approximation should continue in case of models with nonlinearities in their constraints.