Flexibility and regularity properties of the asymptotically ideal production model
通过模拟数据研究渐近理想模型（AIM）逼近未知技术的效果，发现高阶AIM的预测误差远小于广义里昂惕夫和超越对数模型，但非负约束会严重限制其灵活性。
This paper uses simulations to investigate the ability of the Asymptotically Ideal Model(AIM) to approximate unknown technologies. AIM was proposed by Barnett and Jonas (1983) as an expansion which could approximate any indirect utility or cost function arbitrarily well at all prices. By applying AIM to simulated data, we fmd a significant improvement in prediction by higher orders of AIM relative to the generalized Leontief and the translog. The largest error in approximating some CES technologies with AIM is 125 times smaller than that of the generalized Leontief and we find a translog approximation error 1000 times that of AIM in others. We also fmd that non-negativity constraints used by Barnett, Geweke, and Wolfe (1991) to impose concavity on the model greatly restrict the flexibility of AIM, even when all the inputs are substitutes.