A cauchy MESS

摘要

The matrix exponential spatial specification (MESS) recovers its dependence parameter when disturbances exhibit large impulses, as occur under Cauchy and log-Cauchy distributions. The zero diagonal of the weight matrix W eliminates the leading contamination term from the gradient, leaving an O ( 1 / n ) bias under standard constructions. Monte Carlo experiments without regression signal show the standard deviation under Cauchy errors is 24% smaller than under t ( 200 ) errors, with robustness across weight matrices and gradual degradation under nuisance regressors and spatial fixed effects.