Hypothesis Testing in Unidentified Models
指出统计推断不一定需要模型可识别，但会带来歧义。通过简单例子，区分了未识别模型中检验的“反驳”与“确认”两方面，并讨论了过度识别约束的可检验性问题。
An identified model is not necessary for statistical inference, but ambiguities can arise. This paper examines some simple examples and proposes a framework that distinguishes between the "refutation" and "confirmation" aspects of testing in an unidentified model. One particular problem is the interpretation given to overidentifying restrictions: a common view is that these are somehow not properly testable.