A note on global optimization in adaptive control, econometrics and macroeconomics

One of the most cited examples in the literature on global optimization in econometrics is Maddala and Nelson's (Econometrica 42 (1974) 1013) attempt to maximize the likelihood function of a disequilibrium model. On p. 1026, they write “in ... all cases the ... hill climbing method converged but in each case to a different value, thus suggesting the existence of multiple maxima. Of all these peaks we picked the highest”. EZGRAD, the algorithm presented in these pages, works in the same way. It starts the gradient procedure from several points and picks the optimum (highest or lowest) peak. Its main advantages are: (i) ease to use, (ii) adaptability to the function under investigation and (iii) suitability for parallel computing.