Confidence intervals for nonlinear regression extrapolation

The various methods of confidence intervals construction for nonlinear regression are considered. The new method named by a method of associated simulation (the AS-method) is proposed. Using computerized simulation, it is shown on the example that only two methods, the bootstrap and the AS-method, give a satisfactory accuracy. The advantage of the AS-method is the speed. In comparison with the bootstrap, the prize is at least 10 000 times. This method may be applied when regression parameters estimates are obtained by the maximum likelihood method. It was proposed to use the AS-method when extrapolation of complicated physico-chemical model is performed to predict the behavior of the model in the area far from observation.