Evaluating distribution forecasts with observation error (work in progress)

Abstract

Given that distribution forecasts for macroeconomic variables are often evaluated against realizations observed with time-varying measurement error. I examine the sensitivity of the log score, the quadratic score and the continuous ranked probability score to the presence of noise in the observations and investigate to what extent these scoring rules can be robustified in the sense of making their forecast rankings less prone to measurement error.

Publication
Date
Links