The Measurement Problem
Fat Tails, Estimation, and What Standard Finance Gets Wrong
Preface
This is a working monograph. It is incomplete by design — new chapters are added as the research develops. Every version is dated. Errors, when found, are corrected in place and noted in the session log.
0.1 What this book argues
Some standard financial methods rely on moment, dependence, stationarity, and measurement assumptions that can fail under fat-tailed return distributions. The consequences are not determined by a fitted tail exponent alone. Each chapter now distinguishes a valid mathematical result, a model-conditional simulation, a survivorship-conditioned historical finding, and a registered prospective test. The 2009–2025 panel of 715 U.S. equities remains discovery evidence; it is not a point-in-time market universe or prospective confirmation.
The through-line is epistemological: the method you choose determines what you can see, and the standard toolkit was not designed with fat tails in mind. This does not mean the standard toolkit is always wrong — it means there is an identifiable class of problems for which it systematically misleads, and that class is larger than commonly acknowledged.
0.2 Four central research claims
The working-paper series and Part IV are undergoing claim-level mathematical and empirical revalidation. The review identified corrections required in the fixed-jump Hill argument, sample-variance bias language, stable-law parameterization, and sample-mean convergence claims. Chapters now distinguish established results, exploratory historical findings, and registered prospective tests. Existing SSRN papers remain working papers; their presence on SSRN is not peer review or proof of correctness.
A fixed operational Hill bandwidth can conflate tail sources. A bounded discrete jump does not invalidate Hill consistency under a proper intermediate sequence when the structural tail is unbounded and regularly varying. At a fixed top-decile bandwidth, however, jumps can persist inside the fitted threshold region and materially change the finite-sample estimate. The size and economic relevance of that contamination are now registered empirical questions, not an established asymptotic theorem.
Newey-West HAC standard errors address one of two distinct problems introduced by overlapping windows. They correct serial correlation inflation. They do not correct directional bias in the point estimate. The size of that separate measurement problem must be estimated with a clean counterfactual rather than inferred from the tail exponent.
The ergodicity deficit, as formalized by Peters (2019), has not been safely calibrated across empirical tail regimes. Infinite-variance models, finite-variance models with unstable higher moments, and bounded-return/log-growth constraints must be treated separately. A sample variance can be highly unstable without being upward biased.
The minimum track record required to distinguish skill from luck depends on more than a Gaussian standard error. Tail behavior can change Sharpe-estimator limits when the fourth moment does not exist, but a numerical track-record claim also requires the limiting scale, dependence, and nuisance parameters. The previously reported headline point estimate is under replication and is not treated here as established.
0.3 Who this is for
Anyone willing to think carefully about probability and measurement. The mathematics is precise but the motivation is accessible: these are questions about what we can and cannot know from the data we have, under the distributional conditions that actually obtain.
0.5 How to cite
Choufani, J.-M. (2026). The Measurement Problem: Fat Tails, Estimation, and What Standard Finance Gets Wrong (working monograph). Retrieved from [URL].
Individual chapters have separate SSRN preprint identifiers where available; these are listed at the top of each chapter.
0.6 License
This work is released under CC BY 4.0. You may share and adapt the material for any purpose, provided appropriate credit is given. No permission is needed to read, share, or build on this work.
First chapter written: 2026. This version: August 9, 2026.