Post-selection adaptive inference for least angle regression and the

We propose inference tools for least angle regression and the lasso, from the joint distribution of suitably normalized spacings of the LARS algorithm. From this we extend the results of the asymptotic null distribution of the “covariance test ” of Lockhart et al. (2013). But we go much further, deriving exact finite sample results for a new asymptotically equivalent procedure called the “spacing test”. This provides exact conditional tests at any step of the LAR algorithm as well as “selection intervals ” for the appropriate true underlying regression parameter. Remarkably, these tests and intervals account correctly for the adaptive selection done by LARS