Interior point algorithm for LSAD and LMAD designs (Q1297852)

The model under consideration is a multiple regression model \(y=Xb+\epsilon\), where \(y\in R^n\) is a vector of response variables, \(X\in R^{n\times k}\), \(n\geq k\), is the matrix of unknown parameter vectors, \(b \in R^k\) is an unknown parameter vector and \(\epsilon\) is a random noise. The authors suggested to use the primal-dual interior algorithms for the least sums of absolute deviations (LSAD) and the least maximum absolute deviation (LMAD) estimates. The computational experiments have been carried out for investigating the computational performance of these algorithms both for small and large regression models and their comparison with simplex-based algorithms. It occurs that the interior point algorithm is more efficient for LSAD estimation, but not for LMAD estimation. It is shown how two interior point algorithms can be combined in one framework.