Estimation of common change point and isolation of changed panels after sequential detection

Yanhong Wu

arxiv: 1907.02097 · v1 · pith:5ZPIX4HUnew · submitted 2019-07-03 · 🧮 math.ST · stat.TH

Estimation of common change point and isolation of changed panels after sequential detection

Yanhong Wu This is my paper

Pith reviewed 2026-05-25 09:17 UTC · model grok-4.3

classification 🧮 math.ST stat.TH

keywords change point estimationCUSUMfalse discovery ratesequential detectionmulti-stream datapanel isolationcommon change

0 comments

The pith

After detecting a common change via combined CUSUM-SR in multi-stream data, the BH procedure isolates changed panels using the asymptotic exponential property of the CUSUM process to control FDR, then estimates the common change point from

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a post-detection procedure for multi-panel sequential monitoring. Once a common change is detected, the joint distribution of the CUSUM process and detection delay time on unchanged panels is analyzed to establish an asymptotic exponential property. This property allows the CUSUM values to be treated as p-values in the Benjamini-Hochberg procedure for isolating changed panels while controlling the false discovery rate. The common change point is then estimated using only the isolated changed panels. Simulations indicate that proper choice of the FDR level also controls the false non-discovery rate.

Core claim

After a common change is detected by using a combined CUSUM-SR procedure, the BH method by using the asymptotic exponential property for the CUSUM process is developed to isolate the changed panels with the control on FDR. The common change point is then estimated based on the isolated changed panels.

What carries the argument

Benjamini-Hochberg procedure applied to CUSUM statistics justified by their asymptotic exponential distribution for unchanged panels

Load-bearing premise

The CUSUM process on unchanged panels after the detection time follows the stated asymptotic exponential distribution that justifies treating the statistics as p-values inside the BH procedure.

What would settle it

A dataset or simulation where the realized false discovery rate among declared changed panels exceeds the nominal level under the BH threshold would show the exponential approximation is insufficient.

read the original abstract

Quick detection of common changes is critical in sequential monitoring of multi-stream data where a common change is referred as a change that only occurs in a portion of panels. After a common change is detected by using a combined CUSUM-SR procedure, we first study the joint distribution for values of the CUSUM process and the estimated delay detection time for the unchanged panels. The BH method by using the asymptotic exponential property for the CUSUM process is developed to isolate the changed panels with the control on FDR. The common change point is then estimated based on the isolated changed panels. Simulation results show that the proposed method can also control the FNR by properly selecting FDR.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a post-detection isolation step that feeds CUSUM values into BH under a claimed exponential tail, then estimates the common change point from the selected panels; the justification for that tail after the random stopping time is the part that needs checking.

read the letter

The paper takes a combined CUSUM-SR detector for a common change across panels and adds an isolation stage. After detection it looks at the joint distribution of the CUSUM process and the delay estimate on the unchanged panels, invokes an asymptotic exponential tail to turn those values into p-values, runs BH to control FDR, and then estimates the common change point from the panels that pass the threshold. Simulations are used to show that a suitable FDR level can also keep the false-negative rate reasonable. That combination of steps is the concrete addition; it is not just a restatement of earlier CUSUM work but a practical way to move from detection to identification in the multi-panel setting. The approach is straightforward to implement once the asymptotics are accepted. The soft spot is exactly the one flagged in the stress-test note. The abstract states that the exponential limit holds jointly with the estimated delay, yet the detection time is a stopping time built from the pooled statistic across all panels. No derivation or set of conditions is supplied to show that the marginal tail on the null panels remains exponential once that dependence is taken into account, or that the local-alternative regime used for the changed panels does not distort the null tail. Without those steps the FDR guarantee rests on an unverified claim. The simulation results are mentioned but not described in enough detail to judge whether they cover the relevant regimes. The work is aimed at people who already use sequential methods for panel or multi-stream data and need a follow-up isolation rule. A reader who wants a ready-to-use procedure and is willing to check the asymptotics themselves could get value from it. It is coherent enough on its own terms to deserve a serious referee who can verify the joint-distribution argument and the simulation design.

Referee Report

2 major / 1 minor

Summary. The paper proposes a sequential procedure for multi-panel data: a combined CUSUM-SR statistic detects a common change; the joint distribution of the post-detection CUSUM process and estimated delay is derived for unchanged panels; the asymptotic exponential tail of this CUSUM is used to construct p-values for the Benjamini-Hochberg procedure that isolates changed panels while controlling FDR; the common change point is then estimated from the isolated panels. Simulations are reported to show that FNR can also be controlled by suitable choice of the nominal FDR level.

Significance. If the asymptotic exponential marginal and the joint distribution hold under the stopping time induced by the global CUSUM-SR detector, the method supplies a theoretically justified route to FDR-controlled panel isolation after detection, followed by change-point estimation on the reduced set. This addresses a practical need in high-dimensional sequential monitoring where only a subset of streams change. The explicit use of asymptotic tail behavior to license p-values and the simulation check on FNR are positive features.

major comments (2)

[Abstract and the section stating the joint distribution result] The central justification for treating post-detection CUSUM values on null panels as p-values inside BH is the claim that these values remain asymptotically exponential (jointly with the estimated delay) even though the detection time is a stopping time constructed from the aggregate statistic across all panels. No explicit argument is supplied showing that the marginal tail is unaffected by the dependence induced by the common stopping time under the local-alternative regime used for the changed panels. This step is load-bearing for the FDR-control guarantee.
[Abstract] The abstract asserts the existence of the joint distribution result for the CUSUM process and estimated delay on unchanged panels, yet supplies neither the derivation steps nor the precise regularity conditions (e.g., on the local-alternative strength or the number of panels) under which the exponential limit holds. Without these, it is impossible to verify whether the subsequent BH application is valid.

minor comments (1)

[Abstract] Simulation design details (number of panels, change magnitudes, signal-to-noise ratios, number of Monte Carlo replications) are referenced but not described in the abstract; these should be stated explicitly so that the reported FDR/FNR behavior can be reproduced.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and the constructive comments, which identify key points where the theoretical justification requires explicit expansion. We address each major comment below.

read point-by-point responses

Referee: [Abstract and the section stating the joint distribution result] The central justification for treating post-detection CUSUM values on null panels as p-values inside BH is the claim that these values remain asymptotically exponential (jointly with the estimated delay) even though the detection time is a stopping time constructed from the aggregate statistic across all panels. No explicit argument is supplied showing that the marginal tail is unaffected by the dependence induced by the common stopping time under the local-alternative regime used for the changed panels. This step is load-bearing for the FDR-control guarantee.

Authors: We agree that an explicit argument establishing the asymptotic exponential marginal (jointly with the delay estimator) under the global stopping time is necessary to justify the p-values for the BH procedure. The current manuscript states the result but does not supply the technical steps showing that the dependence through the aggregate detector does not disturb the exponential tail for unchanged panels under the local-alternative regime. In the revision we will add a self-contained appendix deriving this limit, including the required martingale arguments and uniform integrability conditions that isolate the effect of the stopping time. revision: yes
Referee: [Abstract] The abstract asserts the existence of the joint distribution result for the CUSUM process and estimated delay on unchanged panels, yet supplies neither the derivation steps nor the precise regularity conditions (e.g., on the local-alternative strength or the number of panels) under which the exponential limit holds. Without these, it is impossible to verify whether the subsequent BH application is valid.

Authors: We accept that the abstract and the statement of the joint-distribution result should be accompanied by the main regularity conditions and an outline of the derivation. The revision will (i) augment the abstract with a parenthetical reference to the conditions (local-alternative strength of order o(1) but large enough for consistent detection, and panel count p = o(T^α) for suitable α), and (ii) insert a short proof sketch in the main text together with the full argument in the appendix. This will make the validity of the subsequent BH step verifiable. revision: yes

Circularity Check

0 steps flagged

No circularity: asymptotic justification and simulations are external to the procedure

full rationale

The abstract and described method invoke an asymptotic exponential tail for the post-detection CUSUM on null panels to license BH p-values, then estimate the change point from isolated panels. This is presented as a derived theoretical property rather than a quantity fitted or defined inside the same procedure. No equation reduces a claimed performance metric to a self-referential fit or renaming. Simulations provide separate empirical checks. The derivation chain therefore remains self-contained against external asymptotic benchmarks and does not match any enumerated circularity pattern.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract supplies no explicit free parameters, no listed axioms beyond the implicit asymptotic property, and no new postulated entities.

axioms (1)

domain assumption Asymptotic exponential property holds for the CUSUM process on unchanged panels after detection
Invoked to justify p-value construction inside the BH procedure

pith-pipeline@v0.9.0 · 5630 in / 1283 out tokens · 29934 ms · 2026-05-25T09:17:55.123865+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

P[M>x]=e^{-δx}; BH p_i=exp(-δ(T_τ(i)+ρ))
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

continuous-time Brownian model for unchanged panels

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.