Persistence, Randomization, and Spatial Noise

Historical persistence studies and other regressions using spatial data commonly have severely inflated t statistics, and different standard error adjustments to correct for this return markedly different estimates. This paper proposes a simple randomization inference procedure where the significance level of an explanatory variable is measured by its ability to outperform synthetic noise with the same estimated spatial structure. Spatial noise, in other words, acts as a treatment randomization in an artificial experiment based on correlated observational data. Combined with Müller and Watson (2021), randomization gives a way to estimate credible confidence intervals for spatial regressions. The performance of twenty persistence studies relative to spatial noise is examined.