Statistics Theory · arXiv:2611.17016 v1 · submitted 12 Nov 2026 · 18 references
Control Variates to reduce the variance of the empirical Moment Generating Function: an extreme value perspective
The paper says
“Accurate estimation of Moment Generating Functions is fundamental in many applications, but empirical estimators have high variances. Within a control variate setup, Extreme Value Theory is leveraged to identify how extreme values affect the variance reduction method's optimality.”
from the abstract
Pith says
Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
Does not hold up as stated. Interesting qualitative insight, but the displayed formulas (4) and (5) are wrong as written due to mishandling of slowly varying functions. This line is wrong
This review says the paper's mathematics is wrong. Pith did not check that against the paper's source.
5 major objections · 8 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 18 references say.
The paper says, at Appendix A.3, Eq. (4) and surrounding derivation
The review did not quote this passage.
Pith says: Objection 1 of 5 · major
The derivation of the limit in Theorem 1 contains a critical error in the handling of slowly varying functions. In the correlation expression, the terms involving L_T and L_S do not cancel to produce the constants c_T and c_S in the denominator of (4) under the stated assumptions.
more of objection 1
Specifically, the proof in A.3 writes Corr ~ ... and then replaces L_T(u) and L_S(u) by their limits c_T and c_S, but the correlation involves ratios like L_T(u)/√(L_T(2u)L_S(2u)). For slowly varying functions, L_T(u)/L_T(2u) → 1 as u → ∞, so the constants c_T and c_S should cancel out, not appear in the final limit. The paper's Eq. (4) includes c_T c_S in the denominator, which is inconsistent with the regular variation properties. The authors assume lim L_T(u) = c_T > 0 and lim L_S(u) = c_S > 0 (b127, b131), but this is not implied by the hypothesis (2) which only states that L_T and L_S are slowly varying. A slowly varying function need not converge to a positive constant; it can oscillate. Even if it converges, the ratio L_T(u)/L_T(2u) → 1, so the constants cancel. The claimed limit ν(B)²/(c_T c_S) 2^{-2/γ_T} is therefore not justified. This is a load-bearing error because the main result (4) is the paper's central claim. The authors need to either correct the limit expression or provide additional justification that the constants survive in the limit.
The paper says, at Appendix A.2, derivation of E[exp(u(T+S))]
The review did not quote this passage.
Pith says: Objection 2 of 5 · major
The derivation of the asymptotic equivalent for E[exp(u(T+S))] in (A.5) is not rigorous. The step from the multivariate regular variation condition (3) to the statement that P(1/(1/Z_T + 1/Z_S) > -u ln t) ~ ν(B) P(Z_T > -u ln t) (or similarly for Z_S) assumes that the set B in (3) corresponds to the event {1/(1/Z_T + 1/Z_S) > -u ln t} after appropriate scaling.
more of objection 2
The paper defines B = {(x,y) ∈ (0,∞)²: 1/x + 1/y < 1}. With x = 1/(u(t*-T)), y = 1/(u(s*-S)), the condition (x,y) ∈ B becomes u(t*-T) + u(s*-S) < 1, i.e., V_T + V_S < 1/u. But the event needed for the MGF expansion is V_T + V_S < -ln t / u, which is of the same form with 1 replaced by -ln t. The scaling in (3) uses u in the denominator of the arguments, but the event in the MGF involves -ln t / u. The derivation implicitly assumes that the regular variation property extends to arguments scaled by -ln t, which is plausible but requires justification because -ln t can be arbitrarily large as t→0. The paper does not discuss how the slowly varying function g(u) interacts with the factor (-ln t)^{α*}. The final expression (A.5) may be correct, but the derivation is sketchy and relies on unverified steps. This is a major concern because the joint MGF expansion is essential for both theorems.
The paper says, at Section 3, Theorem 1 and Appendix A.3
The review did not quote this passage.
Pith says: Objection 3 of 5 · major
The paper states in Theorem 1 that if γ_T ≠ γ_S or ν((0,∞)²)=0, the limit of 1-Corr² is 1. However, the derivation in A.3 for the case γ_T ≠ γ_S appears to assume that the dominant term in the numerator comes from the joint MGF, but when γ_T ≠ γ_S, the product of marginal MGFs may dominate the joint MGF, leading to a different asymptotic.
more of objection 3
The paper's calculation shows the correlation tends to 0, which gives 1-Corr² → 1. This is plausible, but the derivation should be checked carefully because the numerator involves a difference of two terms that may be of the same order. The paper's approach of 'considering only the higher-order terms' is not fully justified. A more rigorous treatment using the exact asymptotic expansions with slowly varying functions is needed.
The paper says, at Appendix A.3, line b128
The review did not quote this passage.
Pith says: Objection 4 of 5 · major
In the correlation derivation, the paper writes: Corr ~ u^{-max(-1/γ_T, -1/γ_S)} C_TS / (u^{1/(2γ_S)+1/(2γ_T)} sqrt(2^{1/γ_S} C_S 2^{1/γ_T} C_T)). This step seems to drop the exponential factors exp(u(t*+s*)) from the numerator and denominator, but those factors should cancel because they appear in both the joint MGF and the product of marginal MGFs.
more of objection 4
However, the derivation is not clearly explained. The cancellation of the exponential factors is critical; if they do not cancel exactly, the correlation would be 0 or diverge. The paper should explicitly show the cancellation and state that t* and s* are the same (or that the exponential factors cancel regardless). In the case γ_T = γ_S, the limit is claimed to be ν(B)/√(c_T c_S) 2^{-1/γ_T}. But as noted in major comment 1, the constants c_T and c_S should not appear if the slowly varying functions are handled correctly.
The paper says, at Section 4, Simulation study
The review did not quote this passage.
Pith says: Objection 5 of 5 · major
The simulation study uses sample size n=1,000,000 and u up to 150. The paper claims to illustrate the theoretical limits, but it does not provide any numerical comparison between the estimated limits and the theoretical expressions (4) and (5). The figures only show qualitative behavior (curves approaching a constant, diverging, or going to zero).
more of objection 5
To convincingly support the theory, the authors should report the estimated limit values and compare them to the theoretical predictions, perhaps in a table. Without such quantitative validation, the simulation is merely illustrative and does not confirm the specific formulas.
8 minor objections
The paper says, at General
Pith says: minor objection 1 of 8
The paper contains several typographical errors and formatting issues. For example, in the abstract: 'estim ators' should be 'estimators'. In the introduction: 'th e' should be 'the'. In Section 2: 'occ urs' should be 'occurs'. The paper should be carefully proofread.
The paper says, at Section 2, Definition 1
Pith says: minor objection 2 of 8
The definition of the CV estimator uses w_opt(u) = Cov(˜T(u), ˜S(u))/Var(˜S(u)). This is correct for minimizing variance, but the paper should note that this assumes the source MGF G(u) is known. In practice, w_opt(u) must be estimated, which introduces additional variability. The paper acknowledges this later, but it would be helpful to mention it earlier.
show 6 more
The paper says, at Section 3, Equation (2)
Pith says: minor objection 3 of 8
The notation in (2) uses P(1/(t*-T) > u) ~ u^{1/γ_T} L_T(u). Since γ_T is negative, 1/γ_T is negative, so the tail decays polynomially. This is correct for the Weibull domain. However, the paper should explicitly state that γ_T < 0 and γ_S < 0, which is implied by the finite upper endpoints. It is stated in A.1 but should be stated earlier in Section 3.
The paper says, at Appendix A.1, equation (A.1)
Pith says: minor objection 4 of 8
The derivation from the regular variation condition to (A.1) is somewhat confusing. The paper writes '⇐⇒ ⋆' without explaining the star. The steps could be clarified. Also, the final result F_{x*-X}(1/v) ~ v^{1/γ} L(v) is correct, but the notation L(v) is reused; it should be noted that this is a different slowly varying function from the one in (2).
The paper says, at Appendix A.2, line b124
Pith says: minor objection 5 of 8
The paper writes 'g(u) = { P(Z_T > u) when Z_T has lighter tail than Z_S; P(Z_S > u) when Z_S has lighter tail than Z_T }'. This is not generally true for multivariate regular variation; the normalizing function g(u) is determined by the marginal tails and the dependence structure.
more of objection 5
The statement that g(u) equals the lighter tail probability is a specific assumption that should be justified or referenced. The paper cites Basrak et al. (2025), but the connection is not explained.
The paper says, at Appendix A.3, line b128
Pith says: minor objection 6 of 8
The derivation uses the notation C_TS = ν(B)Γ(1+α*). But α* = max(α_T, α_S) = max(-1/γ_T, -1/γ_S). When γ_T = γ_S, α* = -1/γ_T. Then Γ(1+α*) = Γ(1 - 1/γ_T). This is consistent with the final expression. However, the paper should note this simplification explicitly.
The paper says, at Section 3, Theorem 2
Pith says: minor objection 7 of 8
The statement of Theorem 2 includes the case γ_T ≤ γ_S for the limit when t* = s*. But the condition in the theorem is 'if γ_T = γ_S, t* = s* and ν((0,∞)²) > 0'. The proof in A.4 gives a more detailed case analysis.
more of objection 7
The theorem statement should be consistent with the proof. Also, the limit in (5) is given as ν(B) c_S 2^{-1/γ_S}, but the proof shows ν(B)/(2^{1/γ_S} c_S) (with c_S in denominator). There is a discrepancy: Eq. (5) says ν(B)c_S 2^{-1/γ_S}, while the proof says C_TS/(2^{1/γ_S} C_S) = ν(B)Γ(1+α*)/(2^{1/γ_S} c_S Γ(1-1/γ_S)). For γ_T = γ_S, α* = -1/γ_T = -1/γ_S, so Γ(1+α*) = Γ(1-1/γ_S). Then the limit becomes ν(B)/(2^{1/γ_S} c_S). This is different from Eq. (5) which has c_S in the numerator. This is a critical inconsistency that must be resolved.
The paper says, at References
Pith says: minor objection 8 of 8
The reference 'Basrak, B., Milinčević, N., and Molchanov, I. (2025). Foundations of regular variation on topological spaces. arXiv preprint arXiv:2503.00921' is cited for the multivariate regular variation condition. This is a very recent preprint; the authors should ensure that the specific condition (3) is indeed covered by that reference. Also, the paper should include a reference for the standard result on regular variation of the Weibull domain (e.g., de Haan and Ferreira, 2006).
What would settle it
Recompute the limit of 1-Corr(exp(uT),exp(uS))^2 for the pair T=S where P(1-T<v)=2v, v∈(0,1/2) (Weibull domain, γ=-1, L_T≡L_S≡2, t*=1, g(u)=2/u, ν(B)=1/2). (a) Direct: since T=S, Corr≡1 and the limit is 0 for every u. (b) The paper's Eq. (4) with c_T=c_S=2, γ_T=-1, ν(B)=1/2 gives 1-3/4·... i.e. 1-(1/4)·2^2/(2·2)=3/4. (c) Re-run the chain A.2-A.3 without dropping the slowly-varying factor: E[exp(u(T+S))]=E[exp(2uT)]~e^{2u}·L_T·(2u)^{-1}Γ(2)=e^{2u}/u (carries L_T≡2), not the paper's e^{2u}/(2u); plugging this into cov/√(var·var) with Var(exp(uT))~e^{2u}(2u)^{-1}·2 yields Corr→1 and limit 0. If step (c) reproduces 0, Theorem 1's formula (4) is false under its own hypotheses; if it reproduces 3/4, the discrepancy lies elsewhere and the stated Eq. (4) would need its own counterexample.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 13 Nov 2026
- Reader does not hold up
- Skeptic partly agrees
- Editor wrote the verdict above
- Referee major revision: 5 major objections, 8 minor
Not run on this paper: the simulated authors' defence.
Pith did not check this review's objections against the paper.
The review in full
The review in brief, machine-written
The paper tackles a genuinely useful question: when does a control variate reduce variance of the empirical MGF for large u? The qualitative answer—only when the extreme value indices match and asymptotic dependence holds—is plausible and practically relevant. The coupling of EVT with control variates for the EMGF is new, and the paper is clearly written.
However, the proofs have a load-bearing problem that the stress-test note correctly identifies. The slowly varying functions LT and LS are dropped in the joint tail expansion (A.5) but kept in the marginals (A.3). The paper then assumes they converge to constants cT, cS—an assumption not stated in the hypotheses (2)–(3). This is not a minor oversight. For the T=S counterexample given in the stress test (uniform on (0,1/2)), every stated hypothesis holds, yet Eq. (4) gives 3/4 instead of the true limit 0. The same mishandling corrupts Theorem 2: the displayed formula (5) has cS in the numerator, while the proof (A.4) gives cS in the denominator—a factor-c_S2 discrepancy.
The reader's primary objection about the set B being the wrong region is incorrect; the stress-test note shows that B correctly captures the near-endpoint region. So the proof structure is not fundamentally broken, but the L-handling is.
The qualitative dichotomy (reduction iff matched EVI plus asymptotic dependence) likely survives a corrected derivation, but the specific formulas (4) and (5) are wrong as printed. The simulation study is illustrative only and does not test the formulas quantitatively.
Who this is for: researchers working on variance reduction for MGF estimation who want a qualitative boundary. The paper deserves a serious referee because the question is important and the qualitative insight is valuable, but the referee will need to insist on a corrected derivation that properly tracks the slowly varying functions through the correlation and optimal coefficient limits.
Recommendation: send to peer review, but flag that the displayed formulas need correction and the assumptions need to be stated explicitly.
The strongest objection
The reader's primary objection is incorrect. The set B in (3) with variables x=1/(u(t*-T)), y=1/(u(s*-S)) is the event {VT+VS<1/u}, i.e., both T and S near their right endpoints — precisely the near-endpoint region A.2 needs, not the 'small-Z, far-from-endpoint' region. Checking against T=S uniform: P((ZT/u,ZS/u)∈B)=P(2(1-T)<1/u)=1/(2u), g(u)=1/u, so ν(B)=1/2 and A.5 gives E[exp(u(T+S))]~e^{2u}(1/2)u^{-1}Γ(2)=e^{2u}/(2u), the correct value. The paper's intermediate scaling (b124), P~ν(B)u^{-α*}(-1/ln t)^{-α*}, is also correct since (-1/ln t)^{-α*}=(-ln t)^{α*} matches the w^{α*} from homogeneity.
The real, load-bearing defect is the treatment of slowly varying functions. In A.2/A.5 the proof drops the L-factor (the normalization g(u)=u^{-α*}L(u) of (3)) in P(VT+VS<-ln t/u)~ν(B)u^{-α*}(-ln t)^{α*}L(u), while A.3 keeps L_T in the marginals. Tracking L through the correlation gives Corr ~ ν(B)2^{-1/γT}(L_T(u)/√(L_T(2u)L_S(2u)))-ratios; for matched marginal L's the factors cancel via L(u)/L(2u)→1, so no c_T c_S belongs in Eq. (4). The paper instead asserts lim L_T=c_T, lim L_S=c_S (b127/b131) — an assumption absent from the hypotheses (2)-(3) — and inserts c_T c_S in the denominator of (4).
Concrete falsification under the paper's own hypotheses: take T=S with P(1-T<v)=2v on (0,1/2); then t*=s*=1, γ_T=γ_S=-1, L_T≡L_S≡2, g(u)=2/u, and ν(B)=1/2 with ν((0,∞)^2)>0, so every hypothesis of Theorem 1 holds. But since T=S, Corr(exp(uT),exp(uS))≡1 and the true limit of 1-Corr^2 is 0, whereas Eq. (4) gives 1-(1/4)·2^{2}/(2·2)=3/4. The same L-mishandling corrupts Theorem 2: A.4 yields lim w_opt=ν(B)2^{-1/γS}/c_S while the statement (Eq. (5)) has c_S in the numerator — a factor-c_S^2 discrepancy the reader flagged as secondary. The qualitative dichotomy (reduction iff matched EVI plus asymptotic dependence) likely survives a correct re-derivation, but the displayed formulas (4) and (5) are wrong as written.
I checked the reader's weakest_assumption in detail and found it does not land: with x=1/(u(t*-T)), y=1/(u(s*-S)), the set B={1/x+1/y<1} in (3) is precisely {VT+VS<1/u}, the joint near-endpoint upper-tail event, which is exactly the regime used in Appendix A.2 (P(VT+VS<-ln t/u)). The reader's identification of B as the 'small-Z, far-from-endpoint' region is backwards — {ZT+ZS<1/u} would mean VT,VS > u, which is the complement region. I verified the paper's A.2/A.5 arithmetic against a T=S uniform example, where it reproduces E[exp(u(T+S))]~e^{2u}/(2u) correctly. So the rejection must rest elsewhere.
The genuine load-bearing problem is the mishandling of the slowly varying functions. The proof of A.2 drops the L-factor carried by the normalization g(u) in condition (3) (see b124 → A.5, and b127/b131 which explicitly impose lim L_T=c_T, lim L_S=c_S, a condition not among the theorem's hypotheses), while A.3 retains L_T. Because the correlation is a ratio, the correct computation carries L_T(u)/√(L_T(2u)L_S(2u))-type factors that, for L_T=L_S, cancel to 1 by slow variation rather than producing the c_T c_S in Eq. (4). This is not merely a hypothesis-gap: I constructed T=S with P(1-T<v)=2v on (0,1/2), which satisfies conditions (2) and (3) with γ_T=γ_S=-1, L_T≡L_S≡2, ν(B)=1/2 and ν((0,∞)^2)>0. Since T=S, Corr(exp(uT),exp(uS))≡1, so the true limit of 1-Corr^2 is 0, yet Eq. (4) predicts 3/4. That is a direct falsification of the paper's central displayed formula under its own assumptions, independent of any convergence conventions. The same defect produces the c_S discrepancy in Theorem 2: the proof gives ν(B)2^{-1/γS}/c_S (Eq. (A.4) algebra), the statement (Eq. (5)) has c_S in the numerator.
Verdict: REJECT stands, but for reasons different from the reader's primary argument. The qualitative dichotomy (variance reduction persists iff γ_T=γ_S plus asymptotic dependence) is plausibly salvageable — the counterexample preserves it (the limit is 0 only because ν(B)^2 2^{-2/γ}=1 there, i.e., perfect dependence gives full reduction) — so a revised version tracking the L-factors through A.5 and correcting Eqs. (4) and (5) could well restore the paper. As printed, however, the headline formula is numerically wrong, so UNCHANGED (REJECT, moderate confidence) is appropriate. I credit the paper for citing the multivariate regular variation framework (Basrak et al., 2025) and for the transparent hand-written proofs that made the error auditable; the simulations are illustrative only and, as the reader noted, cannot adjudicate the asymptotics.
The referee's report
If correct, the results provide a clear theoretical boundary for when control variates can reduce variance of empirical MGF estimators at large u. The finding that variance reduction is only possible when the target and source share the same extreme value index and are asymptotically dependent is practically relevant for choosing source variables. The paper is well-structured and addresses a gap in the literature on variance reduction for MGF estimation. However, the significance is somewhat limited by the restrictive assumptions (finite upper endpoints, Weibull domain, existence of MGF, known or estimable source MGF). The paper does not provide finite-sample guarantees or practical guidance on estimating the optimal coefficient in the extreme regime, which is acknowledged as a limitation.
Pith's reading of the paper (a machine's summary, not the authors')
This paper investigates when control variates can reduce the variance of empirical moment generating function (MGF) estimators for large parameters u. The authors prove that asymptotic variance reduction is possible only if the target and source variables share the same extreme value index (EVI) and exhibit asymptotic dependence. If these conditions are not met, the correlation between exponentiated variables vanishes as u grows, and the control variate estimator offers no improvement over the plain empirical mean. The result provides a theoretical boundary for transfer learning in MGF estimation under extreme parameters.
What carries the argument. Multivariate regular variation of the joint tail of (T,S) with the set B = {(x,y) in (0,∞)^2 : 1/x + 1/y < 1} and the extreme value indices γ_T, γ_S. The asymptotic expansions of E[exp(uT)], E[exp(uS)], and E[exp(u(T+S))] derived from Weibull domain-of-attraction conditions are used to compute the limiting correlation and optimal coefficient.
Core claim. The paper proves Theorem 1: under multivariate regular variation of the joint tail, the squared correlation between exponentiated target and source variables converges to a non-zero constant if and only if the two variables have equal extreme value indices and are asymptotically dependent. Otherwise the correlation tends to zero, implying no variance reduction from control variates at large u. Theorem 2 shows that the optimal control variate coefficient stabilises at a finite constant only when endpoints match and the source tail is not lighter than the target tail. These results establish necessary and sufficient conditions for the effectiveness of control variates in the extreme regime.
The premise it rests on. The joint tail of (T,S) satisfies multivariate regular variation with the specific set B = {(x,y) : 1/x + 1/y < 1}, and the slowly varying functions converge to positive constants.
If the paper is right (Pith's inferences, not the authors')
Practitioners can determine in advance whether control variates will help for large u by checking whether the target and source have the same EVI and are asymptotically dependent.
The results apply to both exact and approximate control variate estimators, and extend to u → -∞ by symmetry.
The analysis reveals that endpoint mismatch causes the optimal coefficient to diverge or vanish, leading to extreme variance of plug-in estimates and potential variance inflation.
The paper provides a rigorous foundation for future work on robust coefficient selection strategies in extreme regimes.
The condition of equal EVIs is restrictive; many practical applications involve heavy-tailed distributions where MGFs do not exist. The paper's framework is limited to bounded variables with finite endpoints (Weibull domain).
The simulation study uses a Gumbel copula with high dependence; real applications may have weaker dependence structures that make asymptotic dependence harder to detect.
The proof in Appendix A.2 contains a potential issue: the multivariate regular variation condition (3) is stated for the set B = {1/x + 1/y < 1}, but the derivation of E[exp(u(T+S))] relies on the event Z_T + Z_S < 1/u, which corresponds to the complement of B. The authors may have intended a different normalisation or set.
Check this review
Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.
How often reviews like this one are right
Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.
How often objections from reviews like this one are right: not yet measured
Reviews by people
This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.
The paper
Abstract
Accurate estimation of Moment Generating Functions is fundamental in many applications, but empirical estimators have high variances. Within a control variate setup, Extreme Value Theory is leveraged to identify how extreme values affect the variance reduction method's optimality.
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.
abstract compiled · 2026-11-13
The authors' main claim, as the classifier picked it out 0.98
“Within a control variate setup, Extreme Value Theory is leveraged to identify how extreme values affect the variance reduction method's optimality.”
How the main claim is stated: asserted 0.99
What the abstract itself states, as probabilities: claims first 0.04 · states limitation 0.16 · human participants 0.03 · controlled comparison 0.15 · randomized assignment 0.02 · states quantitative result 0.04 · compares with prior results 0.09 · mentions code or data availability 0.02
The abstract, sentence by sentence
This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| Accurate estimation of Moment Generating Functions is fundamental in many applications, but empirical estimators have high variances. | background 1.00 | 0.06 | 0.05 | asserted 1.00 |
| Within a control variate setup, Extreme Value Theory is leveraged to identify how extreme values affect the variance reduction method's optimality. | method 0.65 | 0.67 | 0.22 | asserted 0.99 |
What kind of paper the classifier reports this is: method or tool 0.52 · evidence from proof or derivation 0.69
What was examined, and what was not
Sources read
- arXiv listing abstract v1
Not examined
- full text
- figures and tables
- references and citing sentences
- relations to other papers
Levels run
- P abstract
Not run
- S sections
- T body sentences
- C citations
- R cross-paper relations
How the labels were made
Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 2,242 input tokens.
Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON
Figures (1)
Figures are the authors', reproduced from the paper.
References (18)
- [1] Asmussen, Søren and Glynn, Peter W. , year =. Variance-. Stochastic. doi:https://doi.org/10.1007/978-0-387-69033-9\_5 , pages= on Pithdoi:10.1007/978-0-387-69033-9
- [2] Extreme. doi:https://doi.org/10.1007/0-387-34471-3 , publisher = on Pithdoi:10.1007/0-387-34471-3
- [3] Semiparametric doubly robust targeted double machine learning: a review , author=. Handbook of statistical methods for precision medicine , pages=. 2024 , publisher= on Pith
- [4] Owen, Art B. , year =. Variance reduction , booktitle = on Pith
- [5] Fisher, Ronald Aylmer and Tippett, Leonard Henry Caleb , year =. Limiting forms of the frequency distribution of the largest or smallest member of a sample , volume =. Mathematical proceedings of the on Pith
- [6] Gnedenko, Boris , year =. Sur la distribution limite du terme maximum d'une s\'erie al\'eatoire , journal = on Pith
- [7] Selim Orhun Susam , title =. Statistics , pages =. 2026 , publisher = on Pith
- [8] Edward E. Gbur and Robert A. Collins , title =. 1989 , publisher = on Pith
- [9] Nonparametric recursive method for moment generating function kernel-type estimators , journal =. 2022 , issn =. doi:https://doi.org/10.1016/j.spl.2022.109422 , author = on Pithdoi:10.1016/j.spl.2022.109422
- [10] An alternative approach to decisions under uncertainty using the empirical moment-generating function , author=. 1986 , publisher= on Pith
- [11] Using the empirical moment generating function in testing for the Weibull and the type I extreme value distributions , author=. Test , volume=. 2005 , publisher= on Pith
- [12] Efficient exponential tilting with applications , author=. 2024 , publisher= on Pith
- [13] Meintanis, Simos G , journal=. A. 2007 , publisher= on Pith
- [14] Gorodetsky, Alex A. and Geraci, Gianluca and Eldred, Michael S. and Jakeman, John D. , year =. A generalized approximate control variate framework for multifidelity uncertainty quantification , volume =. Journal of on Pith
- [15] Les nombres remarquables , author=. 1983 , publisher= on Pith
- [16] Foundations of regular variation on topological spaces , author=. arXiv preprint arXiv:2503.00921 , year= on PithPith's reviewarXiv
- [17] Zhu, Zhengyu and Yan, Yibo and Li, Gefei and Zhang, Riquan , year =. Recent. doi:http://doi.org/10.1111/insr.12613 , journal = on Pithdoi:10.1111/insr.12613
- [18] Lavenberg, Stephen S and Welch, Peter D , year =. A. doi:http://doi.org/10.1287/mnsc.27.3.322 , journal = on Pithdoi:10.1287/mnsc.27.3.322
Challenges and discussion
Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.
No comments yet.
Sign in to comment