Comment on "Investigating Nonlinearity"

The paper by Bond, Harrison, and O'Brien illustrates the role that convergence criteria, search algorithms, and starting values can play in influencing the success of numerical optimization. One aspect contributing to the importance of these choices in their results appears to be a multivariate generalization of the "pile-up" phenomenon for maximum likelihood estimation of moving average processes. Their results suggest advantages in some settings of using either Bayesian methods or an alternative specification of the random field.