REVIEW 3 major objections 4 minor 4 references
Equilibrium Policy on Dividend and Capital Injection under Time-inconsistent Preferences
T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read This paper establishes an explicit threshold equilibrium for dividend payments and capital injections under pseudo-exponential discounting, with three regimes depending on surplus level.
desk verdict Plausible and worth a revision, but the central theorem's case (i) is not actually proved and the sign in equation (19) is wrong. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The machinery is the continuous-time weak equilibrium definition, in which an $\epsilon$-perturbation of the controls cannot improve the objective in the limit, and the associated extended HJB equation. For pseudo-exponential discounting, the equilibrium value function splits into a weighted sum of two components, each satisfying a linear ODE on each of the three surplus intervals; the thresholds $x_1$ and $x_2$ are fixed by smooth pasting together with $V'(x_1)=\phi$ and $V'(x_2)=1$. Concavity, proved through a transformed ODE lemma, makes the threshold feedback policy internally consistent.
What would settle it
For parameters in the both-threshold case, for example $\mu=\sigma=1$, $\omega=0.3$, $\rho_1=0.6$, $\rho_2=1$, $\bar l=\bar r=1$, $\phi=1.2$, substitute the explicit $V$ from equations (32)-(33) into the simplified extended HJB equation (19) at surplus levels spanning $(x_1,x_2)$; a nonzero residual or a failure of the supremum to be attained by the claimed feedback rule would refute the equilibrium claim.
Extended reading notes
Core claim
Under the pseudo-exponential discount function $\delta(s,t)=\omega e^{-\rho_1(t-s)}+(1-\omega)e^{-\rho_2(t-s)}$, the paper claims that the equilibrium dividend and capital injection strategy for the diffusion surplus process is of threshold form: for $0<x_1\le x_2$, inject capital at rate $\bar r$ when the surplus $X<x_1$, do nothing when $x_1\le X<x_2$, and pay dividends at rate $\bar l$ when $X\ge x_2$. If the equation for $x_1$ has no positive solution, then $x_1=0$ and no capital is ever injected; if the equation for $x_2$ also has no positive solution, dividends are always paid at the maximal rate. Writing the equilibrium value as $V(x)=\omega V_1(x)+(1-\omega)V_2(x)$, each $V_i$ solves a linear second-order ODE on the three surplus intervals with smooth-fit boundary conditions, and Theorem 4.1 asserts that $V$ is increasing and concave.
Load-bearing premise
The proof assumes, rather than checks, that the proposed value function actually solves the equilibrium equation in the case with both thresholds positive; if this check fails, the threshold strategy could fail to be an equilibrium.
Editorial extensions
If this is right
- If the equilibrium claim holds, an insurance company with non-exponential discounters and a restricted injection rate follows a simple three-regime policy, with the exact thresholds obtained by solving the paper's equation (33).
- When capital injection cost exceeds a cutoff, the equilibrium strategy never injects, reducing the problem to a pure dividend equilibrium.
- When the maximal dividend rate is low relative to discounting, the equilibrium pays dividends continuously at the cap, accepting possible ruin.
- Both thresholds increase with the maximal dividend rate; the dividend threshold increases with injection cost, while the injection threshold decreases with it.
- A larger spread between the two group discount rates, with fixed weighted average, raises the equilibrium value function and both thresholds.
Reading between the lines
- If one completes the missing verification by direct substitution for the both-threshold case, the same splitting technique should produce explicit equilibria for any discount function that is a finite convex combination of exponentials.
- The threshold pattern suggests that equilibrium capital-injection policies in other risk models, such as dual or regime-switching models, may also be characterized by two thresholds with the same no-injection and always-dividend degenerations.
- The disappearance of capital injection under high cost is an equilibrium phenomenon; a precommitted planner might still inject, so the model offers a comparative test of commitment versus sophistication.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies a dividend and capital injection problem for a drifted Brownian motion surplus process under general, possibly non-exponential, discounting. The control is the dividend payment rate and the capital injection rate, each bounded in [0,\bar l] and [0,\bar r], with a proportional cost \phi>1 on injected capital, and the objective is expected discounted net cash flow until ruin. For the time-consistent exponential case the authors derive closed-form threshold solutions. For general discount functions they adopt the weak equilibrium concept of Bjork and Murgoci (2010), derive an extended HJB system with a verification theorem, and then specialize to pseudo-exponential discounting \delta(s,t)=\omega e^{-\rho_1(t-s)}+(1-\omega)e^{-\rho_2(t-s)}. The main claim is an explicit three-case equilibrium classification: (i) inject at the maximal rate below x_1 and pay dividends at the maximal rate above x_2; (ii) never inject and pay dividends above x_2; (iii) always pay dividends at the maximal rate. The equilibrium value function is claimed to be V(x)=\omega V_1(x)+(1-\omega)V_2(x), with V_i solving a system of ODEs and thresholds determined by a smooth-fit system.
Significance. If the main theorem is fully established, the paper would be a useful contribution: it is the first systematic treatment of equilibrium dividend and capital injection policy under time-inconsistent preferences, it provides a verification theorem for the weak equilibrium, and it gives explicit closed-form value functions and threshold strategies for pseudo-exponential discounting. The concavity analysis, in particular Lemma A.5, is a nontrivial technical ingredient, and the numerical examples illustrate economically reasonable comparative statics, such as high injection costs eliminating capital injection. The paper also makes a concrete falsifiable prediction about the three regimes. However, the central equilibrium verification for case (i) is not proved, and the printed extended HJB equation contains a sign error, so the paper cannot be accepted in its current form.
major comments (3)
- [§3.2, Eq. (19) and Proposition 3.4] The simplified extended HJB equation (19) has the wrong sign on the right-hand side. In the proof of Proposition 3.4 the authors compute f_s(s,x)-g_s(s,x;s)=E_{s,x}[\int_s^{\tau} \delta_s(s,t)(\hat l_t-\phi\hat r_t)\,dt], and substitution into (14) yields sup_{l,r}\{V_s+(\mu-l+r)V_x+\frac12\sigma^2V_{xx}+(l-\phi r)\}=E_{s,x}[\int_s^{\tau}\delta_s(s,t)(\hat l_t-\phi\hat r_t)\,dt]. Equation (19) as printed has a minus sign before the expectation, which is inconsistent with this derivation. The sign is not merely cosmetic: with the pseudo-exponential discount function (22), the positive sign gives the terms \omega\rho_1V_1+(1-\omega)\rho_2V_2 that match the ODE system (25), whereas the printed negative sign would not. This must be corrected in the statement of (19) and in Proposition 3.4.
- [§4, Theorem 4.2(i), Appendix A] Case (i) of Theorem 4.2 is asserted but not proved. The proof of Theorem 4.2 in Appendix A begins with 'It remains to prove (ii) and (iii)' and then proves only those two cases; case (i), the regime 0<x_1\le x_2 with both capital injection and dividends, is left without verification. Theorem 4.1 cannot fill this gap because it assumes from the outset that an equilibrium strategy of the form (23) with 0<x_1\le x_2 exists. That is precisely the statement that needs to be proved in case (i). The manuscript therefore does not establish that the constructed V(x)=\omega V_1(x)+(1-\omega)V_2(x) with thresholds solving (33) is a weak equilibrium in the sense of Definition 2.2; it only proves, conditionally, the payoff representation and concavity once the equilibrium form is assumed.
- [§4, Eq. (33) and Theorem 4.2(i)] The paper gives no existence or uniqueness result for the two-dimensional system (33), nor does it prove that any solution (x_1,x_2) of (33) yields the required derivative ordering V'(x)>\phi on (0,x_1), 1\le V'(x)\le\phi on [x_1,x_2), and V'(x)\le1 on [x_2,\infty). The numerical example (x_1=0.1916, x_2=0.317) is evidence but not a proof. Since the equilibrium verification in case (i) depends on this derivative ordering to ensure that (23) realizes the supremum in the extended HJB equation, the theorem as stated is conditional on an unverified existence/ordering claim. The authors should either prove existence-uniqueness for (33) under explicit parameter conditions, or state Theorem 4.2(i) as a conditional result with the existence of such thresholds as a hypothesis.
minor comments (4)
- [§3.1, Proof of Proposition 3.1] The concavity proof for V_c uses the equality \tau_{u^\xi}_{0,\lambda x_1+(1-\lambda)x_2}=\tau_{u^{(1)}}_{0,x_1}\vee\tau_{u^{(2)}}_{0,x_2}. This equality is not generally correct: the convex combination of two positive processes can hit zero before both component processes have hit zero, and the processes are not defined after their individual ruin times. The argument needs revision, for example by a comparison argument or a different coupling.
- [§1, Introduction] There is a typo in the sentence 'Not that for the time-consistent counterpart...'; it should be 'Note that'.
- [Throughout] The word 'càdlàg' appears in a broken form ('c` agl` ad') in the introduction and should be typeset correctly.
- [§5, Figures 1–6] The captions and discussion of the numerical examples would be clearer if the text stated that the displayed regions and threshold curves are computed numerically and are not claimed as analytic existence regions; this would avoid any appearance of substituting numerics for the missing proof of existence in (33).
Circularity Check
Case (i) of the three-threshold equilibrium is assumed rather than derived: Theorem 4.1 postulates the equilibrium strategy (23) to prove the concavity used to certify (23).
-
self definitional
[Section 4, Theorem 4.1 and Theorem 4.2; proof in Appendix A]
"Theorem 4.1 Suppose that there exist 0 < x1 ≤ x2 such that the equilibrium strategy is given by (23). ... In addition, V (x) = ωV1(x) + (1 − ω)V2(x) is increasing and concave. ... Proof of Theorem 4.2 It remains to prove (ii) and (iii)."
Case (i) of Theorem 4.2 asserts: if 0<x1≤x2 solve the smooth-fit system (33), then the three-region strategy (23) is the equilibrium. The only concavity statement used to convert (33) into the derivative ordering required by (21) is Theorem 4.1, whose hypothesis is exactly that (23) is already an equilibrium strategy. Since Appendix A's proof of Theorem 4.2 explicitly says 'It remains to prove (ii) and (iii)', no independent verification of the extended HJB equation (19) is supplied for case (i). Thus the equilibrium property of the threshold strategy is an input to the derivation, not an output of it.
-
self definitional
[Section 4, Proposition 4.2]
"Suppose that there exist 0 < x1 ≤ x2 such that the equilibrium strategy is given by (23) and there exist functions vi(x) ∈ C2 with bounded first-order derivatives that solve (25)-(26). Then vi = Vi which satisfies the probabilistic interpretation (24) and V (x) = ωv1(x) + (1 − ω)v2(x) is the equilibrium value function."
This proposition is the bridge from the ODE system (25) to the equilibrium value function, but it is conditional on the existence of thresholds for which (23) is already the equilibrium strategy. That is precisely the conclusion needed for Theorem 4.2(i). The later equations (33) only enforce the smooth-pasting conditions V'(x1)=ϕ and V'(x2)=1; they do not, by themselves, establish that the strategy satisfies the perturbation condition of Definition 2.2, because no independent check of (19) is provided for case (i).
full rationale
The non-circular parts of the derivation are substantial: the time-consistent benchmark in Section 3.1 is solved from the model, the ODE system (25) is derived from the controlled diffusion and the pseudo-exponential discount structure, and the threshold equations (33) are a standard smooth-fit fixed-point system rather than a fitted input. The paper also draws on external, non-self-cited support for cases (ii)-(iii) via Zhao et al. (2014, Lemma 4.1). No data are fitted and no parameter is renamed as a prediction. The circularity is concentrated in case (i), the genuinely new capital-injection equilibrium regime: the proof of concavity, which is needed to justify the derivative ordering behind (23), assumes the existence of the very equilibrium strategy that case (i) purports to establish. The appendix confirms this by proving only cases (ii) and (iii) for Theorem 4.2. This makes the central three-region classification partially circular, though the formulas themselves are not constructed from the conclusion. The sign issue in the printed equation (19) is a correctness concern, not a circularity, and does not affect this score.
Assumptions & free parameters
assumptions (5)
- domain assumption Surplus evolves as a drifted Brownian motion with constant coefficients μ and σ.
- domain assumption Dividend and injection rates are bounded, lt∈[0,¯l] and rt∈[0,¯r], and controls are Ft-measurable with unique solution.
- domain assumption The weak equilibrium definition of Bjork and Murgoci is appropriate for time-inconsistent preferences.
- domain assumption Pseudo-exponential discounting δ(s,t)=ωe^{-ρ1(t-s)}+(1−ω)e^{-ρ2(t-s)} is used for the explicit solution.
- standard math Smooth fit and the existence/uniqueness of solutions to linear systems for the constants.
Cite this review
Pith. "Pith review of Equilibrium Policy on Dividend and Capital Injection under Time-inconsistent Preferences." pith.science (2026). https://pith.science/paper/PMZIP4BF
@misc{pith2026250523511,
author = {Pith},
title = {Pith review of: Equilibrium Policy on Dividend and Capital Injection under Time-inconsistent Preferences},
year = {2026},
howpublished = {\url{https://pith.science/paper/PMZIP4BF}},
note = {Machine review of arXiv:2505.23511}
}
read the original abstract
This paper studies the dividend and capital injection problem under a diffusion risk model with general discount functions. A proportional cost is imposed when injecting capitals. For exponential discounting as time-consistent benchmark, we obtain the closed-form solutions and show that the optimal strategies are of threshold type. Under general discount function which leads to time-inconsistency, we adopt the definition of weak equilibrium and obtain the extended HJB equation system. An explicit solution is derived under pseudo-exponential discounting where three cases of the dividend and capital injection thresholds are obtained. Numerical examples show that large capital injection cost may lead to no capital injection at all, while larger difference in group discount rate leads to higher equilibrium value function.
Figures
Figures from the paper (3 more)
Reference graph
Works this paper leans on
-
[1]
Asmussen, S. and Taksar, M. (1997). Controlled diffusion models for optimal dividend pay-out, Insurance: Mathematics and Economics 20(1): 1–15. Avanzi, B., Shen, J. and Wong, B. (2011). Optimal dividends and capital injections in the dual model with diffusion,ASTIN Bulletin: The Journal of the IAA 41(2): 611–
work page 1997
-
[3]
The proof of the case of 0 < x1 = x2 is similar. Q.E.D. Because Bi3 < 0, Vi is concave on interval [ x2, ∞). The concavity of Vi on the whole domain is proved with the help of Lemma A.5 inspired by Shreve et al. (1984, Lemma 4.1) and Bai et al. (2023, Proposition 3.2). Lemma A.5 Suppose that vi(x), i = 1, 2, are solutions to the linear ordinary differ- en...
work page 1984
-
[4]
(1984, Lemma 4.1)), it indicates that l′ 2(y) > 0, ∀y ∈ [0, ¯y)
Combining with the fact that l2(¯y−) = v′ 2(b−) > 0 and l′ 2 has no zero in [0 , ¯y) (thanks to Shreve et al. (1984, Lemma 4.1)), it indicates that l′ 2(y) > 0, ∀y ∈ [0, ¯y). Therefore, ωρ1l′ 1(y) + (1 − ω)ρ2l′ 2(y) > ωρ1l′ 1(y) + (1 − ω)ρ1l′ 2(y) = ρ1l′(y). (A.26) We then claim that ωρ1l1(y) + (1 − ω)ρ2l2(y) < 0 (i.e., l′′(y) <
work page 1984
-
[644]
Z τ ˜ u 0,x 0 e−ρt ˜lt − ϕ˜rt dt # = E
Bai, L., Gamage, T., Ma, J. and Xie, P. (2023). Reinforcement learning for optimal dividend problem under diffusion model, arXiv preprint arXiv:2309.10242 . Bjork, T. and Murgoci, A. (2010). A general theory of markovian time inconsistent stochastic control problems, SSRN 1694759 . Bjork, T., Murgoci, A. and Zhou, X. Y. (2014). Mean–variance portfolio opt...
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.