Optimal Periodic Double-Barrier Strategies for Spectrally Negative L\'{e}vy Processes

Kazutoshi Yamazaki; Qingyuan Zhang

REVIEW 3 major objections 4 minor 23 references

Optimal Periodic Double-Barrier Strategies for Spectrally Negative L\'{e}vy Processes

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper proves that, for a spectrally negative Lévy process with continuous upward control but downward control restricted to Poisson arrival times, the optimal policy is a double-barrier strategy — reflect upward at a lower barrier…

desk verdict Solid mixed-frequency two-sided control paper with a real bounded-variation gap in the smooth-fit/verification argument; deserves review but needs a fix. read the letter →

arxiv 2505.23080 v1 pith:3EVPFLDO submitted 2025-05-29 math.OC

Kazutoshi Yamazaki , Qingyuan Zhang This is my paper

classification math.OC MSC 60G5193E2090B05

keywords stochasticcontrolspectrallynegativeLévyprocessesscalefunctionsperiodicobservationsdouble-barrierstrategyPoissonsingularinventory

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper considers a system whose state follows a spectrally negative Lévy process — a random path with downward jumps — and whose controller can push the state upward at any time but can only push it downward at random opportunities arriving on an independent Poisson clock. It claims, and proves by a guess-and-verify argument, that an optimal policy has a two-barrier form: reflect the state upward whenever it would fall below a lower level $a^*$, and at each Poisson time, if the state is above an upper level $b^*$, push it down to $b^*$. The paper derives a semi-explicit formula for the expected cost of this strategy in terms of scale functions, identifies the optimal barriers $(a^*,b^*)$ as the unique solution of two equations, and verifies that no other admissible strategy does better. The result matters because it gives an explicitly solvable model of two-sided control under random discrete control opportunities, as in inventory management where replenishment is easy but selling is possible only at random times.

What carries the argument

The argument runs on scale functions of spectrally negative Lévy processes, the standard fluctuation kernels for these processes: $W^{(q)}$, $Z^{(q)}$, and the modified objects $Z^{(q,r)}_{a,b}$, $W^{(q,r)}_{a,b}$, etc. encode the behavior of a process reflected from below and randomly pushed down at Poisson times. The load-bearing identities are (3.12) and (3.13), defining $\Gamma(a,b)$ and its partial derivative $\gamma(a,b)$; $\Gamma=0$ makes the candidate value smooth at the lower barrier and $\gamma=0$ does the same at the upper barrier, and together they are exactly the probabilistic first-order conditions used to select $(a^*,b^*)$. The verification lemma uses the operator $M h(x)=\inf_{l\ge 0}\{C_D l + h(x-l)\}$ to check that no downward jump or upward reflection can improve the candidate value.

What would settle it

For a concrete spectrally negative Lévy process with computable scale functions, solve the equations $\Gamma(a,b)=0$ and $\gamma(a,b)=0$ from (3.12)–(3.13) and check the verification-lemma inequalities pointwise for $v_{a^*,b^*}$; if any admissible strategy is found whose simulated cost beats $v_{a^*,b^*}$ by more than Monte Carlo error, Theorem 4.7 fails. A sharper boundary test is to take $f(x)=\sqrt{1+x^2}$ with $qC_D>1$, so Assumption 2.1(3) fails: the paper predicts no finite $b^*$ and optimality of the single-barrier policy, so discovering that a two-barrier policy beats the single-barrier policy in that case would refute the classification.

Watch

Extended reading notes

Core claim

The paper's central claim is Theorem 4.7: under a convex running cost $f$ with suitable growth, and under the condition that $f'(x)-qC_D>0$ for all large $x$, the periodic-classical double-barrier strategy $\pi_{a^*,b^*}$ is optimal among all admissible strategies, and its value function equals the semi-explicit expression $v_{a^*,b^*}(x)$ in (3.11). The optimal pair $(a^*,b^*)$ is characterized probabilistically: the expected discounted integral of $f'$ under the controlled process equals $-C_U$ at the lower barrier and $C_D$ at the upper barrier. The paper proves this is equivalent to the two smoothness equations $\Gamma(a^*,b^*)=\gamma(a^*,b^*)=0$, establishes the existence and uniqueness of the pair, and confirms optimality by a verification lemma. When the slope condition fails, the conclusion degenerates: the optimal strategy is a single-barrier policy with no downward control.

Load-bearing premise

The double-barrier conclusion rests on Assumption 2.1(3): the running cost's slope eventually exceeds the discounted unit downward-control cost $qC_D$, so that exercising downward control is genuinely worthwhile; if that condition fails, the authors show the optimal policy becomes a single upward-reflection barrier with no downward control.

Editorial extensions

If this is right

If the theorem is right, a controller never needs a more complicated state-dependent policy: the optimal continuous action is always to reflect at $a^*$, and the optimal action at each Poisson opportunity is always to push down to $b^*$.
The cost under the optimal policy is computable in semi-closed form from (3.11), so for any spectrally negative Lévy process with a known scale function the optimal barriers can be found by solving the two equations $\Gamma(a,b)=0$ and $\gamma(a,b)=0$.
When the running-cost slope never exceeds $qC_D$, the optimal policy has only the lower barrier and never activates downward control, so the double-barrier form is tied to the regime where downward control pays for itself.
The paper's numerical experiments show that as the Poisson arrival rate $r$ grows, the barriers and the value function move toward those of the fully continuous two-sided control problem, giving a periodic-to-continuous bridge.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper: the probabilistic barrier conditions — expected discounted marginal running cost equal to $-C_U$ below and $C_D$ above — look transferable; the same two-equation recipe may identify optimal barriers for spectrally negative Markov additive or regime-switching processes, where scale-function identities also exist.
Beyond the paper: for deterministic equally spaced control times no exact scale-function solution is known; the Poisson model is the natural approximation, and one could test numerically whether the Poisson double-barrier value with matching mean interarrival time approximates the constant-period optimum.
Beyond the paper: since $C_U$ and $C_D$ are allowed to be negative, the theorem also covers dividend-style models where downward control earns a reward; this suggests periodic dividend-barrier strategies remain optimal in this two-sided Lévy framework.
Beyond the paper: the convergence observed as $r\to\infty$ raises a quantitative question the paper does not address — the rate at which $(a^*,b^*)$ and $v_{a^*,b^*}$ approach the classical continuous two-sided solution; a sensitivity analysis of $\Gamma$ and $\gamma$ near the classical optimizer could supply such a rate.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: the result is new and mostly right, but the proof as written does not cover bounded-variation Lévy processes. The stress-test note checks out. In Proposition 3.8, the expression for v'_{a,b}(b) drops the term -r(C_U+C_D)W^{(q+r)}(0). For bounded variation, W^{(q+r)}(0)=\delta^{-1}>0, so the right derivative at b is strictly less than C_D and the candidate has a kink. Lemma 3.6 only checks smoothness at a, not at b, so the “sufficiently smooth iff Γ=0” claim is false in that case. Consequently Theorem 4.7's verification lemma, which needs the candidate to be C1 (or C2), does not cover bounded variation. The theorem is certainly true for unbounded variation, where W^{(q+r)}(0)=0, and it is probably fixable by a relaxed verification argument or a modified barrier condition, but that patch is not in the paper.

What is genuinely good: the problem is a natural two-sided extension of the one-sided periodic results, the candidate double-barrier strategy and the semi-explicit value formula are new, and the probabilistic construction of the barriers plus the uniqueness proof is a real contribution. The equivalence between the probabilistic conditions and smooth-fit is a nice observation. Most of the technical work is in the appendices and the statements I checked are consistent apart from the missing boundary term. The numerical example is thin, but that is acceptable for a theory paper.

Other soft spots, in decreasing order: Lemma 4.5 (polynomial growth) is stated without proof; the strict comparison in Lemma 3.11 is only sketched; Proposition A.1's proof of part (i) is more of an assertion than an argument. These are fixable. The reliance on the authors' earlier papers for resolvent and control-cost formulas is heavy but legitimate, since those are previously published results.

Bottom line: this deserves a serious referee, but not acceptance as is. The bounded-variation case must be either excluded honestly or repaired. If the authors restrict Theorem 4.7 to unbounded variation and note the bounded-variation case as open, I would be satisfied in principle. I would not rely on the theorem for bounded-variation processes until the fix appears.

Referee Report

3 major / 4 minor

Summary. The paper studies a singular stochastic control problem for a spectrally negative Lévy process with continuous upward control and downward control available only at the arrival times of an independent Poisson process. The authors propose a periodic-classical double-barrier strategy, compute its expected discounted cost in semi-explicit form using scale functions, select candidate barriers via a probabilistic argument that is shown to be equivalent to smooth fit, prove existence and uniqueness of the candidate pair, and give a verification theorem claiming optimality. A numerical example illustrates the strategy and its convergence to the classical continuous-control solution as the Poisson intensity grows.

Significance. If the verification can be made fully rigorous, the result is a valuable extension of the existing literature on periodic barrier strategies: it provides explicit scale-function formulas, connects the probabilistic candidate-selection method with the smooth-fit principle, covers a general spectrally negative Lévy setting, and supplies a uniqueness/existence proof for the barrier pair. The numerical study supports the expected convergence to the classical doubly reflected solution. However, because the verification argument currently has a load-bearing gap in the bounded-variation case and two auxiliary lemmas are not fully proved, the paper needs substantive revision before the main theorem is established.

major comments (3)

[Sections 3.2 and 4; Proposition 3.8, Eq. (3.14), Lemma 4.1] The smooth-fit verification at the upper barrier is only carried out for the left derivative. Evaluating the derivative expression (3.14) from the right at x=b gives the additional boundary term -r(C_U+C_D)W^{(q+r)}(0), while for x<b the term W^{(q+r)}(x-b) is zero, so the left limit has no such term. Under conditions C' and (4.2), the left derivative equals C_D, but the right derivative equals C_D - r(C_U+C_D)W^{(q+r)}(0). By Remark 3.1(3), W^{(q+r)}(0)=δ^{-1}>0 for every bounded-variation process admitted by Assumption 2.2. Hence v_{a*,b*} is not C^1 at b* in the bounded-variation case, Lemma 4.1's 'sufficiently smooth' hypothesis is not satisfied, and Theorem 4.7 does not cover this case. The expression displayed in the proof of Proposition 3.8 omits exactly this boundary contribution.
[Section 4, Lemma 4.5] The polynomial-growth property of v_{a*,b*} is stated without proof, with the sentence 'the proof of the following result is omitted.' This property is used essentially by Lemma 4.1 and Lemma 4.6, and it is not immediate that the proof of Lemma 4.7 of [21] transfers unchanged to the current two-sided strategy with upper reflection at Poisson times, especially when W^{(q+r)}(0)>0. The proof should be supplied or its hypotheses verified against the present model.
[Section 3.4, Lemma 3.11(ii) and Proposition 3.13] The strict comparison argument in Lemma 3.11(ii) is only sketched: the inequality vf'_{a2,b}(a2) < vf'_{a2,∞}(a2) is justified by asserting that the set of times on which the two processes differ has positive Lebesgue measure with positive probability. A rigorous proof is needed, and the same strict comparison is used to assert uniqueness of b* in Proposition 3.13. The argument should cover all processes allowed by Assumption 2.2, including bounded-variation and finite-activity cases.

minor comments (4)

[Section 3, Eq. (3.10)-(3.11)] The formulas use the oriented integral ∫_x^b for x>b, where the intended value is the integral over [b,x] with the opposite sign. This convention should be stated explicitly to avoid sign ambiguity.
[Appendix B.8, Lemma B.4] The verification of the cases where |f| is bounded on one side is compressed; in particular, the construction of the dominating functions h and the treatment of the event {X(e_q)<0} should be spelled out with the relevant finite bounds.
[Appendix A, Proposition A.1] The condition 'f' ≤ qC_D a.e. on R' is asserted as the failure of Assumption 2.1(3); it would be clearer to state this as f'(x) ≤ qC_D for all sufficiently large x.
[Appendix B.7, proof of Lemma 4.4] The proof for x>a* invokes Lemma 7.2 of [17] without explaining why that lemma, proved for a different model, remains valid for the present periodic-classical strategy; a short translation of the argument would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the optimality proof is a self-contained guess-and-verify argument, and the cited prior results are external support rather than built-in conclusions.

full rationale

The paper selects candidate barriers via the probabilistic condition C, proves the equivalent smooth-fit condition C' (Gamma=gamma=0) in Proposition 3.8, establishes existence and uniqueness of (a*,b*) in Proposition 3.13, and then verifies the HJB inequality in Theorem 4.7 through Lemmas 4.1-4.6. The candidate is not defined to be optimal; optimality is shown after the fact. Dependencies on [17], [18], and [19] are published prior results with proofs or proof techniques, not self-citations that assume the present theorem. Lemma 4.5's proof is omitted as nearly identical to [21], and Proposition B.1 is imported from [17], but neither omission injects the target result into the derivation. Any concern about smoothness of v_{a,b} at b* for bounded-variation processes is a correctness issue, not a circularity issue, because the paper's own equations are used to derive the claimed smoothness rather than assuming it. The numerical section is illustrative and does not disguise fitted parameters as predictions. Therefore no circular step is exhibited, and the honest score is 0.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No data are fitted; all quantities are derived from the model primitives (q, r, C_U, C_D, Lévy characteristics, f). The main inputs are standard scale-function machinery and regularity and growth assumptions on f.

assumptions (5)

domain assumption Assumption 2.1: f is convex, piecewise C^1, and slowly or regularly varying at ±∞ when unbounded, with conditions (2) and (3) on the asymptotic slopes of f relative to qC_U and qC_D.
Imposed in Section 2; ensures well-posedness and the existence of finite barriers.
domain assumption Assumption 2.2: exponential tail condition on the Lévy measure, guaranteeing finite mean and integrability of discounted f(X).
Section 2, Assumption 2.2; needed for the verification theorem and the transversality condition.
standard math Scale function theory: W^{(q)} exists, is smooth enough (Remark 3.1), and the fluctuation identities from [11], [16], [17] hold.
Used throughout Section 3 for the semi-explicit formulas.
domain assumption The process X is not the negative of a subordinator, i.e., δ>0 in (2.2).
Needed for well-defined reflection from below at the lower barrier.
standard math Itô's lemma and the verification lemma (Lemma 4.1) with local martingale property from [20].
Used in Section 4 to prove optimality.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Optimal Periodic Double-Barrier Strategies for Spectrally Negative L\'{e}vy Processes." pith.science (2026). https://pith.science/paper/3EVPFLDO

@misc{pith2026250523080,
  author       = {Pith},
  title        = {Pith review of: Optimal Periodic Double-Barrier Strategies for Spectrally Negative L\'evy Processes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3EVPFLDO}},
  note         = {Machine review of arXiv:2505.23080}
}

read the original abstract

We study a stochastic control problem where the underlying process follows a spectrally negative L\'{e}vy process. A controller can continuously increase the process but only decrease it at independent Poisson arrival times. We show the optimality of the double-barrier strategy, which increases the process whenever it would fall below some lower barrier and decreases it whenever it is observed above a higher barrier. An optimal strategy and the associated value function are written semi-explicitly using scale functions. Numerical results are also given.

Figures

Figures reproduced from arXiv: 2505.23080 by the authors.

**Figure 2.** Plots of va,b (for r = 0.1) against initial position x. Left: Plot of va ∗,b (blue dashed) for b = b ∗ − 5, b∗ − 4, . . . , b∗ − 1, b∗ + 1, . . . , b∗ + 5, with (a ∗ , va ∗,b(a ∗ )) (lime triangle), (a ∗ , va ∗,b∗ (a ∗ )) (red triangle), (b, va ∗,b(b)) (lime square), and (b ∗ , va ∗,b∗ (b ∗ )) (red square). Right: Plot of va,b∗ (blue dashed) for a = a ∗ − 5, a∗ − 4, . . . , a∗ − 1, a∗ + 1, . . . , a∗ + 5, with (a, v… view at source ↗

**Figure 3.** Left: Plot of periodic-classical barriers ( [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

[21]

P´ erez, K

J.-L. P´ erez, K. Yamazaki, and A. Bensoussan. Optimal periodic replenishment policies for spec- trally positive l´ evy demand processes. SIAM Journal on Control and Optimization , 58(6):3428– 3456, Jan. 2020

work page 2020
[1]

Albrecher, N

H. Albrecher, N. B¨ auerle, and S. Thonhauser. Optimal dividend-payout in random discrete time. Statistics & Risk Modeling , 28(3):251–276, Sept. 2011

work page 2011
[2]

Albrecher, J

H. Albrecher, J. Ivanovs, and X. Zhou. Exit identities for l´ evy processes observed at Poisson arrival times. Bernoulli, 22(3), Aug. 2016

work page 2016
[3]

Avanzi, E

B. Avanzi, E. C. Cheung, B. Wong, and J.-K. Woo. On a periodic dividend barrier strategy in the dual model with continuous monitoring of solvency. Insurance: Mathematics and Economics , 52(1):98–113, Jan. 2013

work page 2013
[4]

Avram, A

F. Avram, A. E. Kyprianou, and M. R. Pistorius. Exit problems for spectrally negative l´ evy processes and applications to (Canadized) Russian options. The Annals of Applied Probability , 14(1), Feb. 2004

work page 2004
[5]

E. J. Baurdoux and K. Yamazaki. Optimality of doubly reflected l´ evy processes in singular control. Stochastic Processes and their Applications , 125(7):2727–2751, July 2015

work page 2015
[6]

N. H. Bingham, C. M. Goldie, and J. L. Teugels. Karamata Theory, page 1–60. Encyclopedia of Mathematics and its Applications. Cambridge University Press, 1987

work page 1987
[7]

J. G. Dai and D. Yao. Brownian Inventory Models with Convex Holding Cost, Part 1: Average- Optimal Controls. Stochastic Systems, 3(2):442–499, Dec. 2013

work page 2013

Show all 23 references

[8]

J. G. Dai and D. Yao. Brownian Inventory Models with Convex Holding Cost, Part 2: Discount- Optimal Controls. Stochastic Systems, 3(2):500–573, Dec. 2013

2013
[9]

H. Dong, C. Yin, and H. Dai. Spectrally negative l´ evy risk model under Erlangized barrier strategy. Journal of Computational and Applied Mathematics , 351:101–116, May 2019. 25

2019
[10]

Egami and K

M. Egami and K. Yamazaki. Phase-type fitting of scale functions for spectrally negative l´ evy processes. Journal of Computational and Applied Mathematics , 264:1–22, July 2014

2014
[11]

Kuznetsov, A

A. Kuznetsov, A. E. Kyprianou, and V. Rivero. The theory of scale functions for spectrally negative l´ evy processes. In S. Cohen, A. Kuznetsov, A. E. Kyprianou, and V. Rivero, editors,L´ evy Matters II: Recent Progress in Theory and Applications: Fractional L´ evy Fields, and...

2013
[12]

A. E. Kyprianou. Fluctuations of L´ evy processes with applications: introductory lectures. Univer- sitext. Springer, Heidelberg, second edition edition, 2014

2014
[13]

Leung, K

T. Leung, K. Yamazaki, and H. Zhang. An Analytic Recursive Method for Optimal Multiple Stopping: Canadization and Phase-Type Fitting. SSRN Electronic Journal , 2014

2014
[14]

M. A. Lkabous. Poissonian occupation times of spectrally negative l´ evy processes with applications. Scandinavian Actuarial Journal , 2021(10):916–935, Nov. 2021

2021
[15]

R. L. Loeffen. On optimality of the barrier strategy in de Finetti’s dividend problem for spectrally negative l´ evy processes.The Annals of Applied Probability , 18(5), Oct. 2008

2008
[16]

R. L. Loeffen, J.-F. Renaud, and X. Zhou. Occupation times of intervals until first passage times for spectrally negative l´ evy processes.Stochastic Processes and their Applications , 124(3):1408–1435, Mar. 2014

2014
[17]

D. Mata, H. A. Moreno-Franco, K. Noba, and J.-L. P´ erez. On the bailout dividend problem with periodic dividend payments for spectrally negative Markov additive processes. Nonlinear Analysis: Hybrid Systems , 48:101332, May 2023

2023
[18]

Noba, J.-L

K. Noba, J.-L. P´ erez, K. Yamazaki, and K. Yano. On optimal periodic dividend and capital injection strategies for spectrally negative l´ evy models.Journal of Applied Probability, 55(4):1272– 1286, Dec. 2018

2018
[19]

Noba and K

K. Noba and K. Yamazaki. On stochastic control under Poisson observations: optimality of a barrier strategy in a general l´ evy model.Journal of Applied Probability , 2025. To appear

2025
[20]

P´ erez and K

J.-L. P´ erez and K. Yamazaki. On the optimality of periodic barrier strategies for a spectrally positive l´ evy process.Insurance: Mathematics and Economics , 77:1–13, Nov. 2017

2017
[22]

Yamazaki

K. Yamazaki. Inventory control for spectrally positive l´ evy demand processes. Mathematics of Operations Research, 42(1):212–237, Jan. 2017

2017
[23]

Y. Zhao, P. Chen, and H. Yang. Optimal periodic dividend and capital injection problem for spectrally positive l´ evy processes.Insurance: Mathematics and Economics , 74:135–146, May 2017. 26

2017

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [21]

P´ erez, K

J.-L. P´ erez, K. Yamazaki, and A. Bensoussan. Optimal periodic replenishment policies for spec- trally positive l´ evy demand processes. SIAM Journal on Control and Optimization , 58(6):3428– 3456, Jan. 2020

work page 2020

[2] [1]

Albrecher, N

H. Albrecher, N. B¨ auerle, and S. Thonhauser. Optimal dividend-payout in random discrete time. Statistics & Risk Modeling , 28(3):251–276, Sept. 2011

work page 2011

[3] [2]

Albrecher, J

H. Albrecher, J. Ivanovs, and X. Zhou. Exit identities for l´ evy processes observed at Poisson arrival times. Bernoulli, 22(3), Aug. 2016

work page 2016

[4] [3]

Avanzi, E

B. Avanzi, E. C. Cheung, B. Wong, and J.-K. Woo. On a periodic dividend barrier strategy in the dual model with continuous monitoring of solvency. Insurance: Mathematics and Economics , 52(1):98–113, Jan. 2013

work page 2013

[5] [4]

Avram, A

F. Avram, A. E. Kyprianou, and M. R. Pistorius. Exit problems for spectrally negative l´ evy processes and applications to (Canadized) Russian options. The Annals of Applied Probability , 14(1), Feb. 2004

work page 2004

[6] [5]

E. J. Baurdoux and K. Yamazaki. Optimality of doubly reflected l´ evy processes in singular control. Stochastic Processes and their Applications , 125(7):2727–2751, July 2015

work page 2015

[7] [6]

N. H. Bingham, C. M. Goldie, and J. L. Teugels. Karamata Theory, page 1–60. Encyclopedia of Mathematics and its Applications. Cambridge University Press, 1987

work page 1987

[8] [7]

J. G. Dai and D. Yao. Brownian Inventory Models with Convex Holding Cost, Part 1: Average- Optimal Controls. Stochastic Systems, 3(2):442–499, Dec. 2013

work page 2013