Central-moment conjecture of Frongillo and Kash
For every integer , let denote the th central moment, . On every open convex set of positive distributions supported on distinct real outcomes, is exactly -elicitable through an identifiable intermediary, i.e. does ? Equivalently, are report coordinates necessary and sufficient for a one-observation loss to uniquely elicit an identifiable intermediary from which the th central moment can be recovered? At the same time, the mean and the th central moment are indirectly identifiable from two expectation equations, so .
References
Source: arXiv
Progress summary
A new preprint claims to settle how hard it is to recover central moments from data, but the claim has not been independently checked.
The conjecture of Frongillo and Kash concerns the exact elicitation complexity of the th central moment and the gap between identifiable elicitation and indirect identification. Earlier literature attributes the basic non-elicitation lower bound to Frongillo and Kash (2015).
Known results
- The th central moment is -elicitable for .
- It is also -elicitable.
- Before the new preprint, the cited lower bound was only that the th central moment is not -elicitable (Frongillo and Kash, 2015).
October 2026 claimed resolution
Jingyi Jia’s preprint claims sharp lower and upper bounds for identifiable elicitation of the mean and th central moment, an unbounded separation from indirect identification, and extensions to standardized moments of order at least and certain convex distribution classes. The claim is unrefereed and unverified in the retrieved evidence.
Current status (as of October 2026): The conjecture is claimed solved by Jia’s preprint, but that resolution remains unverified; no independent assessment or correction was retrieved.
Solutions 0
No solutions have been posted yet.