The Informational Content in Lepto-Variance and Its Relation to Higher Moments
stat.ME · cs.LG · q-fin.RM · stat.ML
Quick verdict
Tree self-split variance decomposition: lepto-ratio tracks kurtosis; but 36.3% normal limit is uncredited Lloyd–Max (1−2/π) and 'no factor' claims overreach.
Summary
This paper investigates the informational content of 'lepto-variance' — the residual variance left after a regression tree is fit to a sample using the target itself as the splitting feature ('lepto-regression', from the author's 2022/2024 work) — and its ratio to total sample variance. After restating the prior finding that 58% of 96 years of US daily stock-return variance is 1-bit lepto-variance (Section 3), the paper derives the continuous-uniform lepto-ratio analytically (25%, Section 4) and links it to the 6 dB/bit result of PCM quantization theory. Monte Carlo simulation of 10,000 normal N(0,1) samples of n=100 (Table 2) shows lepto-variance correlates strongly with sample variance (0.831) while lepto-ratio correlates with excess kurtosis (0.782; slope 5.25 percentage points per kurtosis unit, R² = 61.2%), and both are orthogonal to skewness. Large-sample simulations (n = 100,000 and 10⁷) converge to a normal lepto-ratio of 36.3% (Tables 3–4). Limitations: no code or seeds disclosed; the 36.3% limit is not recognized as the closed form 1 − 2/π of the Lloyd–Max 1-bit Gaussian quantizer; orthogonality claims lack formal tests; and the abstract's 'cannot be explained by any financial factor' overreaches what 1-bit sorted-split tree bounds can establish.
Review complete
Assessment at a glance
5
risk flags identified
● high 1
● medium 2
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- The lepto-ratio of a normal distribution converges to 36.3% of total variance (≈4.4 dB), established by simulations at n = 100,000 and n = 10⁷.
- In small (n = 100) normal samples, lepto-variance is highly correlated with sample variance (0.831), while lepto-ratio is highly correlated with excess kurtosis (0.782; slope 5.25 percentage points per kurtosis unit, R² = 61.2%).
- Both lepto-variance and lepto-ratio are orthogonal to sample skewness, and lepto-ratio is orthogonal to sample variance despite its correlation to lepto-variance, for normal samples.
- The continuous-uniform 1-bit lepto-ratio is exactly 25%, connecting lepto-regression to the 6 dB/bit result of PCM quantization theory.
- Carried from prior work: 58% of the variance of 96 years of US daily stock returns is 1-bit lepto-variance that 'cannot be explained by any financial factor'.
Strengths
Tidy analytic anchor case
Section 4 proves the continuous-uniform lepto-ratio is exactly 25%, and Table 1 shows fast discrete-uniform convergence (ratio → 0.25, SNR → −6.02 dB), connecting RT self-splitting to the classic PCM 6 dB/bit result (Bennett 1948; Oliver et al. 1948) — a verifiable bridge to quantization theory.
Sensible small-sample design
10,000 independent N(0,1) samples at n = 100 (Table 2) exploit natural sampling variability of the moments — a reasonable device for studying how lepto statistics co-vary with variance, skew and kurtosis when the population values are degenerate (0 and 0).
Multi-scale convergence checks
Internally consistent arithmetic
Honest scoping of open questions
Weaknesses
Headline constant is a known closed form, uncredited
Abstract overclaims explanatory scope
No inferential statistics
Reproducibility gaps
Loose terminology and reporting inconsistencies
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper (iBusiness 17(3), 202–217) asks what "lepto-variance" — the residual variance after a regression tree (RT) is fit to a sample with the target variable itself as the splitting feature ("lepto-regression", Polimenis 2022, 2024) — actually measures, and how it relates to classical sample statistics. Central claims: (i) for continuous uniforms the 1-bit lepto-ratio is exactly 25%, connecting to PCM quantization's 6 dB/bit rule (Section 4); (ii) for normal samples, lepto-variance tracks sample variance (r = 0.831) while lepto-ratio tracks excess kurtosis (r = 0.782; slope 5.25 pp/unit; R² = 61.2%), both orthogonal to skew (Sections 5–6, Table 2); (iii) the normal's lepto-ratio converges to 36.3% (Tables 3–4); (iv) restating prior work, 58% of 96 years of US daily-return variance (25,272 observations, 1926–2022, total variance 1.167 %²) is 1-bit lepto-variance "that cannot be explained by any financial factor." Claims (i)–(iii) are simulation results; claim (iv) is imported and overreaches, as detailed below.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is lepto-variance?
It is the portion of a sample's variance that remains after fitting a regression tree to the data using the target variable itself as the only splitting feature ('lepto-regression'). Because the best possible single split of a sample is a sorted split on the target, this residual is the smallest MSE any one-split (1-bit) regression tree can achieve; the 'lepto-ratio' expresses it as a fraction of total sample variance.
What is the lepto-ratio of a normal distribution according to the paper?
The paper's simulations (10 replications at n = 10,000,000) converge to 36.3% of total variance, about 4.4 dB. A reviewer observation: this equals the closed form 1 − 2/π ≈ 36.34% obtained by splitting a unit normal at the median (leaf means ±√(2/π)), which is also the classic Lloyd–Max 1-bit Gaussian quantizer distortion — an identity the paper itself does not point out.
Where does the 58% figure for US stocks come from?
It is inherited from the author's earlier work: for 25,272 daily US market returns from July 1926 to June 2022 (total variance 1.167 in %²), the optimal 1-bit split leaves a lepto-variance of 0.678, i.e. a lepto-ratio of about 58%. The present paper does not re-derive it; it studies simulated normal samples instead. Note the abstract's phrasing that this variance 'cannot be explained by any financial factor' is stronger than what a 1-bit tree bound can justify.
How does the lepto-ratio relate to kurtosis?
In 10,000 simulated samples of size n = 100 from a standard normal, the lepto-ratio correlates 0.782 with sample excess kurtosis; a linear regression gives a slope of 5.25 percentage points per unit of kurtosis with R² ≈ 61%. Both lepto-variance and lepto-ratio show essentially zero correlation with skewness, though the paper reports no formal significance tests.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1–5 (regression scatter plots and trees) are referenced but their contents cannot be verified from the extracted text; only the in-text summaries are reviewable.
- Tables 2–4 were extracted but original decimal alignment could not be cross-checked; minor reporting discrepancies (e.g. dB sign flip) are noted in the review.
- The review covers the published iBusiness version (17(3), 202–217); no supplementary material or code accompanies the extracted text.
- Some prior-work context (Polimenis 2022 conference paper) is not publicly verifiable from this text alone.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.