Inference of high-dimensional weak instrumental variable regression models without ridge-regularization

Inference of instrumental variable regression models with many weak instruments attracts many attentions recently. To extend the classical Anderson-Rubin test to high-dimensional setting, many procedures adopt ridge-regularization. However, we show that it is not necessary to consider ridge-regularization. Actually we propose a new quadratic-type test statistic which does not involve tuning parameters. Our quadratic-type test exhibits high power against dense alternatives. While for sparse alternatives, we derive the asymptotic distribution of an existing maximum-type test, enabling the use of less conservative critical values. To achieve strong performance across a wide range of scenarios, we further introduce a combined test procedure that integrates the strengths of both approaches. This combined procedure is powerful without requiring prior knowledge of the underlying sparsity of the first-stage model. Compared to existing methods, our proposed tests are easy to implement, free of tuning parameters, and robust to arbitrarily weak instruments as well as heteroskedastic errors. Simulation studies and empirical applications demonstrate the advantages of our methods over existing approaches.