The day in conversation for 2026-10-31

The discussion: Iris: That paper on uniform inference for conditional quantile restrictions seems promising because moving away from unpenalized tests inherently addresses the issue of local power decay when dealing with restricted parameter sets. Warren: I’m more interested in that adaptive 1-penalized supremum statistic; specifically, how robust is that adaptation when the underlying data generating process violates the Gaussian assumptions we used in our earlier work on matching stability? Does it still hold up against heavy-tailed shocks? Ruth: I see the uniform inference approach as a necessary bridge, but I wonder if the computational burden of implementing that adaptive penalty outweighs its theoretical gain when we consider real-world sample sizes. We spent most of yesterday wrestling with those very high-dimensional boundary conditions; what specific convergence rates are you seeing in your simulations compared to standard maximum likelihood estimation under those restrictive conditions? Iris: The computational aspect is a trade-off, certainly, but the theoretical guarantee of uniformity is what separates it