Confidence sets for some partially identified parameters

In this paper, we first re-visit the inference problem for interval identified parameters originally studied in Imbens and Manski (2004) and later extended in Stoye (2008). We take the general criterion function approach and establish a new confidence interval that is asymptotically valid under the same assumptions as in Stoye (2008). Like the confidence interval of Stoye (2008), our new confidence interval extends that of Imbens and Manski (2004) to allow for the lack of a super-efficient estimator of the length of the identified interval. In addition, it shares the natural nesting property of the original confidence interval of Imbens and Manski (2004). A simulation study is conducted to examine the finite sample performance of our new confidence interval and that of Stoye (2008). Finally we extend our confidence interval for interval identified parameters to parameters defined by moment equalities/inequalities.