Monday, August 10, 2020

Density Forecasts and Density Realizations

Check out the interesting new paper below.  Of course I am a fan of predictive densities (e.g., GDP fan charts).  Usually I compare them to the "realization" (e.g., "published GDP").  But of course, as the authors emphasize, the realization is just a point estimate of true GDP, whereas in reality true GDP is itself uncertain and has its own density, which can be estimated (maybe...) from revisions.  Very interesting.  I wonder, for example, how predictive density assessments based on the probability integral transform (e.g., here) would need to be modified to account for uncertainty in the measured realization.

  1. By:Galvao, Ana Beatriz (University of Warwick); Mitchell, James (University of Warwick)
    Abstract:GDP is measured with error. But data uncertainty is rarely communicated quantitatively in real-time. An exception are the fan charts for historical GDP growth published by the Bank of England. To assess how well understood data uncertainty is, we first evaluate the accuracy of the historical fan charts and compare them with models of past revisions data. Secondly, to gauge perceptions of GDP data uncertainty across a wider set of experts, we conduct a new online survey. Our results call for greater communication of data uncertainties, to anchor dispersed expectations of data uncertainty. But they suggest that transitory data uncertainties can be adequately quantified, even without judgement, using past revisions data.
    Keywords:data revisions; fan charts; expectations survey; backcasts; density forecast calibration
    JEL:C53 E32

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.