Optimal Investment with Correlated Stochastic Volatility Factors

Jean-Pierre Fouque; Maxim Bichuch

REVIEW 3 major objections 4 minor 20 references

Optimal Investment with Correlated Stochastic Volatility Factors

T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper proves that for a power-utility investor facing two correlated stochastic volatility factors, expanding around the perfectly correlated case yields an ε²-accurate approximation of the optimal value and a nearly optimal trading…

desk verdict Novel correlation-based perturbation with a promising accuracy theorem, but the proof as printed has a sign inconsistency in the central cancellation and several key verifications are omitted. read the letter →

arxiv 1908.07626 v3 pith:OCFC7K7B submitted 2019-08-20 q-fin.MF math.PR

Maxim Bichuch , Jean-Pierre Fouque This is my paper · ORCID

classification q-fin.MFmath.PR MSC 91G8060H30

keywords optimalinvestmentstochasticvolatilitypowerutilitydistortiontransformationasymptoticexpansionHJBequationnear-optimalstrategysub-andsuper-solutions

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

The paper treats a portfolio allocation problem in which a stock's return and volatility are driven by two correlated random factors, with the investor maximizing power utility of terminal wealth. When there is only one factor, a classical distortion transformation linearizes the Hamilton–Jacobi–Bellman equation, but with several factors the equation stays nonlinear. The paper's idea is to perturb around the limiting case where the two factors are perfectly correlated, because that limit reduces to the solvable single-factor problem and produces an explicit zeroth-order solution. The main result is a theorem saying that the value function is approximated to order ε² uniformly in wealth and factor values, and that the simple zeroth-order strategy is nearly optimal in the same sense. What makes this useful is that the approximation only requires solving two linear equations in lower dimension, not the fully nonlinear HJB equation.

What carries the argument

The load-bearing object is the distortion transformation $v(t,x,z_1,z_2) = (x^p/p)(\Psi(t,z_1,z_2))^q$ with $q = 1/(1+\Gamma\rho^2)$ and $\Gamma = p/(1-p)$, which converts the power-utility HJB equation into an equation for $\Psi$. In the perfectly correlated limit $\rho_1=\rho_2=\rho$, $\rho_{12}=1$, the choice of $q$ cancels the nonlinear terms and $\Psi^{(0)}$ solves a linear parabolic equation with a Feynman–Kac representation. The perturbation in $\varepsilon$ uses the ansatz $\Psi \approx \Psi^{(0)}+\varepsilon\Psi^{(1)}$; $\Psi^{(1)}$ solves the same linear operator applied to a source $f_1$ built from $\Psi^{(0)}$ and its derivatives. The accuracy proof then forms $v^\pm = (x^p/p)(\Psi^{(0)}+\varepsilon\Psi^{(1)}\pm\varepsilon^2 M(T-t))^q$ and shows that $v^-$ is a submartingale along $\pi^0$ while $v^+$ is a supermartingale along every admissible strategy, which sandwiches the true value function within $x^pO(\varepsilon^2)$.

What would settle it

Choose a two-factor model whose coefficients satisfy Lemma 1 and Theorem 3, solve the HJB equation numerically to high accuracy, and compute $|v - (x^p/p)(\Psi^{(0)}+\varepsilon\Psi^{(1)})^q|/x^p$ for $\varepsilon = 0.1, 0.05, 0.025$; if this normalized error does not decay like a constant times $\varepsilon^2$ uniformly in time and the factors, the theorem is false. Testing the square-root example instead would not settle the theorem, since the paper explicitly excludes it from the hypotheses.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central result is Theorem 3. Under boundedness, smoothness, and non-degeneracy assumptions on the coefficients, with the return-to-volatility ratio $\lambda$ bounded away from zero and $0<p<1$, the value function satisfies $v(t,x,z_1,z_2) = (x^p/p)(\Psi^{(0)}(t,z_1,z_2)+\varepsilon\Psi^{(1)}(t,z_1,z_2))^q + x^pO(\varepsilon^2)$, uniformly in time and factors. The same theorem asserts that the strategy $\pi^0$ built from the zeroth-order solution is nearly optimal: $0 \le v(t,x,z_1,z_2) - E_{t,x,z_1,z_2}[(1/p)(X_T^{\pi^0})^p] = x^pO(\varepsilon^2)$. The function $\Psi^{(0)}$ is the solution of a linear equation inherited from the perfectly correlated limit, $\Psi^{(1)}$ solves a companion linear equation whose source term is built from $\Psi^{(0)}$ and its derivatives, and the proof works by constructing sub- and super-solutions whose difference is of order $\varepsilon^2$. The same expansion is carried out formally for two stocks and two factors.

Load-bearing premise

The theorem assumes the Sharpe ratio, the factor drifts, and the factor volatilities are bounded with bounded derivatives, with the volatilities and the Sharpe ratio bounded away from zero; the entire sub- and super-solution argument and the Feynman–Kac representations depend on those bounds, and the paper's own square-root example does not satisfy them.

Editorial extensions

If this is right

The optimal value function in a multi-factor power-utility problem can be computed to order ε² by solving two linear parabolic equations, one for Ψ⁽⁰⁾ and one for Ψ⁽¹⁾, rather than the fully nonlinear HJB equation.
The zeroth-order strategy π⁰ delivers expected terminal utility within O(ε²) of the true optimum, so an investor can act on a simple explicit rule without solving the nonlinear control problem.
The approximation error is uniform over the factor domain covered by the hypotheses, so the method is not limited to a particular calibrated point.
The same ansatz and linear-equation structure extend to two stocks and two factors, indicating the approach is not tied to a single risky asset.

Reading between the lines

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

If the stopping-and-truncation argument mentioned in Remark 2 can be made uniform in the truncation parameter, the ε² accuracy should extend to square-root volatility factors, bringing the paper's explicit example under the theorem.
A likely extension, not pursued in the paper, is to apply the same perfect-correlation perturbation to other settings that admit a distortion transformation, such as long-run or infinite-horizon problems, where the limiting one-factor model is still linear.
A practical consequence the authors do not state: the error bound's constant depends on how far the Sharpe ratio and factor volatilities stay away from zero, so the approximation should be trusted least in volatility states near zero; a finite-horizon stopping version could quantify that degradation.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

The correlation-based perturbation is the genuinely new piece here. Expanding around fully correlated volatility factors rather than using time-scale separation is a real contribution, and it gives a tractable way to handle multiple same-timescale factors. The first-order correction Ψ^(1) solving a linear PDE is new, and the sub/super-solution approach is appropriate for the nonlinear HJB equation. If Theorem 3 is correct, the near-optimality of π0 is a genuine result.

But the proof as printed does not deliver. In Section 4.2.2, formula (31) has the order-ε bracket written with −f1, while Ψ^(1) is defined in (17) with +f1. Substituting (17) leaves an O(ε) residual; the claimed cancellation does not occur. The same −f1 appears in the multi-asset residual in Section 5.1.2. This looks like a fixable sign typo, but it is in the central step of the main theorem, so the accuracy result is unproven as written.

There are other soft spots, roughly in decreasing order of concern. The proof of Theorem 3 punts on the O(ε^3) sign estimate and on the martingale verification, both needed to justify the absorption by M. The paper's own Remark 2 concedes that the square-root example in Section 3 does not satisfy the theorem's assumptions, so the 'errors of order O(ε) and O(ε^2)' in Figures 1–2 are heuristic; the numerical method behind the reference solution is also undocumented. In Section 4.2.5 (p<0), the claim that v− < v+ appears to have the sign backwards: with p<0, x^p/p is negative, so the defined functions satisfy the opposite inequality. None of this is fatal to the method—the 0

I'd send this to a serious referee, with instructions to check the sign and ask for the missing estimates. The idea is fresh and the intended audience—people doing asymptotic expansions in stochastic portfolio theory—will care. The current version should not be accepted as is.

Referee Report

3 major / 4 minor

Summary. The paper studies the Merton optimal investment problem with power utility in a market with one risky asset and two stochastic volatility factors. Since the multi-factor HJB equation is not tractable, the authors expand around the case of perfectly correlated factors. They perturb the correlations as rho_i = rho + epsilon rho_i^(1) and rho_12 = 1 + epsilon rho_12^(1), posit the ansatz v(t,x,z1,z2) ~ (x^p/p)(Psi^(0)+epsilon Psi^(1))^q, and derive linear PDEs (12) and (17) for Psi^(0) and Psi^(1). The paper gives a closed-form example in a Chacko-Viceira-type model and presents numerical error plots. The central theoretical claim is Theorem 3: under boundedness and regularity assumptions, v differs from the first-order approximation by x^p O(epsilon^2), uniformly in the state variables, and the proposed strategy pi^0 in (19) is nearly optimal in the sense of (30). An extension to two assets and two factors is sketched in Section 5.

Significance. If correct, the paper offers a tangible reduction in computational complexity: the nonlinear HJB equation is replaced by two linear equations, and an explicit nearly optimal strategy is identified. The sub- and super-solution framework is a natural and potentially reusable way to turn a heuristic asymptotic expansion into a uniform error bound. The main strengths are the explicit derivation of Psi^(0) and Psi^(1), the closed-form example, and the clear architecture of the accuracy proof. However, as printed, the proof of Theorem 3 contains a sign inconsistency in the order-epsilon cancellation, and it delegates the decisive O(epsilon^3) sign control and the martingale verification to omitted computations. These issues are local and likely fixable, but they prevent the rigorous accuracy claim from being fully supported in the current version.

major comments (3)

[Section 4.2.2, Eq. (31); Section 4.2.3, Eq. (33)] The order-epsilon bracket in (31) is printed as partial_t Psi^(1) + L^{1,rho} Psi^(1) + (Gamma/(2q)) lambda^2 Psi^(1) - f_1(Psi^(0), nabla Psi^(0), H(Psi^(0))), and the same expression appears in (33). But the defining equation (17) has this expression with + f_1. Substituting (17) therefore gives -2 f_1, not 0, so the claimed cancellation of order-epsilon terms does not occur. This is not a harmless sign in an intermediate step: it leaves an O(epsilon) residual that cannot be absorbed by the M-dependent epsilon^2 term for small epsilon, and the proof of Theorem 3 and of (30) fails as written. The same mismatch occurs in the multi-asset formal computation in Section 5.1.2 following Eq. (42). One of the signs in (17), (31), (33), or in the definition of f_1 must be corrected, and the cancellation must be verified with the corrected sign.
[Section 4.2.2 and Section 4.2.3, proof of Theorem 3] The proof relies on two claims that are explicitly not shown: that the O(epsilon^3) terms in (31) and (33) do not change the sign of the epsilon^2 term for sufficiently small epsilon, and that the martingale parts in the Ito arguments leading to (26) and (27) are true martingales. The manuscript says 'We omit here this lengthy but straightforward computation' and 'We omit the details.' These are load-bearing for the advertised rigorous accuracy result. The O(epsilon^3) remainder must be shown to be uniformly controlled in x, t, z, and M, and the martingale property must be checked using the stated boundedness and admissibility hypotheses. As printed, the theorem is not fully proven.
[Section 4.2.4, Remark 2, and Section 3] The theorem's hypotheses exclude the paper's only fully worked example. The square-root processes have alpha_i(z)=m-z and beta_i(z)=beta-bar sqrt(2z), so beta_i is neither bounded nor bounded away from zero; hence Lemma 1 and Theorem 3 do not apply, as Remark 2 admits. The numerical O(epsilon) and O(epsilon^2) errors in Figures 1 and 2 are therefore not justified by the theorem. The authors should either extend the accuracy result to the example with the stopping argument they mention, or explicitly state that the numerical illustration is heuristic and outside the scope of the rigorous theorem. The abstract's unqualified claim of a rigorous accuracy result should be qualified accordingly.

minor comments (4)

[Throughout] There are several typographical errors, including 'pratical' in Section 3, 'begining' in Section 4.2, 'exemple' in Section 5.1.3, and 'there exits a constant M' in Theorem 3. These should be corrected in a final revision.
[Section 5.1.2] The displayed computation after (43) contains a confusing fraction 'L^{pi0,rho^W,rho}_{x,z} v / v' where the division by v appears to be a typographical artifact; the equation should be presented without that notation.
[Remark 2] The text 'Psi^(0' is missing a closing brace; it should read 'Psi^(0)'.
[Section 2.2, Eq. (15)] The paper assumes rho_i = rho + rho_i^(1) epsilon for i=1,2 with the same base rho. It would be helpful to state explicitly that the base rho satisfies |rho|<1 and that the perturbed coefficients preserve the covariance condition (3) uniformly in epsilon, not merely for fixed epsilon.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the approximation functions are constructed by substituting the ansatz into the HJB equation and then verified against the full nonlinear equation via sub- and super-solutions.

full rationale

The central derivation is not circular. The functions Ψ^(0) and Ψ^(1) are not fitted to the value function; they are defined by the linear equations (12) and (17), obtained by substituting ansatz (16) into the HJB equation (10), with explicit Feynman–Kac representations (14) and (18). The accuracy claim in Theorem 3 is then tested independently against the original nonlinear HJB operator: the paper computes Q^{π0}[v−] in (31) and sup_π Q^π[v+] in (33) and invokes (12) and (17) to cancel the leading terms, so the claimed order of accuracy is not assumed as an input. Lemma 1 rests on external results (Oleinik [18], Karatzas–Shreve [14]), not on the authors' own theorems. The self-citations in the paper, such as Bichuch [2], Bichuch–Sircar [3], and Fouque et al. [13], are methodological or background citations and are not load-bearing; the sub- and super-solution construction is carried out explicitly in the text. Remark 2 in Section 4.2.4 explicitly states: "The model used in our example given in Section 3 based on square-root processes does not satisfy the assumptions of Theorem 3." This is an honest limitation that narrows the rigorous scope of the paper, but it does not make the derivation circular. Similarly, the proof delegates some O(ε^3) sign checks and martingale verifications to omitted routine computations, and the skeptic's sign discrepancy between (17) and the ε terms in (31)/(33) would be a correctness or consistency risk, not a circularity: even if that cancellation fails, it fails because of algebra, not because the theorem's conclusion is built into its hypotheses.

Assumptions & free parameters 1 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the regularity/boundedness class of Lemma 1 and Theorem 3, which the paper's own Section 3 example violates (Remark 2); on the affine-in-epsilon correlation perturbation (15); on classical Feynman-Kac and verification tools; and on the positivity of the constructed v+/-. No parameters are fitted to data; the Section 3.1 numbers are hand-picked for the illustration only. No new entities are introduced.

free parameters (1)

Example parameters and correlation expansion coefficients (Section 3.1) = mu-bar=0.05, m=26, sigma-bar=0.2, beta-bar=5, p=-1, T=1, rho=0.5, rho_1^(1)=0, rho_2^(1)=-0.5, rho_12^(1)=-1…
Hand-chosen for the numerical illustration only; they are not fitted to data and do not enter the statement of Theorem 3.

assumptions (6)

domain assumption Standing classical hypotheses on the coefficients of (1)-(2) ensuring existence and uniqueness of a strong solution
Invoked in Section 2 immediately before equation (3) without being stated explicitly.
domain assumption Lemma 1 assumptions: lambda, alpha_i, beta_i bounded, twice differentiable with bounded derivatives; sigma, beta_i bounded away from zero
Section 4.1. This is the load-bearing regularity class; it buys smoothness of Psi^(0), Psi^(1), and the Feynman-Kac representations, and is needed by Theorem 3.
domain assumption Theorem 3 additional assumptions: lambda bounded away from zero and 0 < p < 1
Section 4.2. The case p<0 is treated separately in Section 4.2.5 with a sketch.
ad hoc to paper Correlation structure (15): rho_i = rho + rho_i^(1) epsilon, rho_12 = 1 + rho_12^(1) epsilon, with epsilon small enough that the covariance condition (3) holds
The paper's perturbation device; all results are relative to this affine-in-epsilon correlation family.
domain assumption Well-definedness of the sub/super-solutions: Psi^(0) + epsilon Psi^(1) +/- epsilon^2 M(T-t) > 0
Section 4.2, used to define v+/- as powers; requires epsilon small and Psi^(0) bounded away from zero.
standard math Classical Feynman-Kac and degenerate-elliptic regularity results ([14][Theorem 5.7.6], [18][Theorem 6])
Used in Lemma 1 and the verification argument in Section 4.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Optimal Investment with Correlated Stochastic Volatility Factors." pith.science (2026). https://pith.science/paper/OCFC7K7B

@misc{pith2026190807626,
  author       = {Pith},
  title        = {Pith review of: Optimal Investment with Correlated Stochastic Volatility Factors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OCFC7K7B}},
  note         = {Machine review of arXiv:1908.07626}
}

read the original abstract

The problem of portfolio allocation in the context of stocks evolving in random environments, that is with volatility and returns depending on random factors, has attracted a lot of attention. The problem of maximizing a power utility at a terminal time with only one random factor can be linearized thanks to a classical distortion transformation. In the present paper, we address the situation with several factors using a perturbation technique around the case where these factors are perfectly correlated reducing the problem to the case with a single factor. Our proposed approximation requires to solve numerically two linear equations in lower dimension instead of a fully non-linear HJB equation. A rigorous accuracy result is derived by constructing sub- and super- solutions so that their difference is at the desired order of accuracy. We illustrate our result with a particular model for which we have explicit formulas for the approximation. In order to keep the notations as explicit as possible, we treat the case with one stock and two factors and we describe an extension to the case with two stocks and two factors.

Figures

Figures reproduced from arXiv: 1908.07626 by the authors.

**Figure 2.** Left: graph of the numerical solution of Ψ (blue), zero order and first order approximations Ψ [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages

[1]

Construction of Forward Performance Processes in Stochastic Factor Models and an Extension of Widder's Theorem

L. Avanesyan, M. Shkolnikov, and R. Sircar. Construction of forward performance processes in stochastic factor models and an extension of widder’s theorem. arXiv:1805.04535v1, 2018

work page Pith review arXiv 2018
[2]

M. Bichuch. Asymptotic analysis for optimal investment in ﬁnite time with transaction costs. SIAM Journal on Financial Mathematics, 3(1):433–458, 2012

work page 2012
[3]

Bichuch and R

M. Bichuch and R. Sircar. Optimal investment with transaction costs and stochastic volatility part ii: Finite horizon. SIAM Journal on Control and Optimization , 57(1):437–467, 2019

work page 2019
[4]

Chacko and L

G. Chacko and L. M. Viceira. Dynamic consumption and portfolio choice with stochastic volatility in incomplete markets. Review of Financial Studies , 18(4):1369–1402, 2005

work page 2005
[5]

J. Feng, M. Forde, and J.-P. Fouque. Short-maturity asymptotics for a fast mean-reverting heston stochastic volatility model. SIAM Journal on Financial Mathematics , 1(1):126–141, 2010

work page 2010
[6]

Fouque and R

J.-P. Fouque and R. Hu. Asymptotic optimal strategy for portfolio optimization in a slowly varying stochastic environment. SIAM Journal on Control and Optimization , 5(3), 2017

work page 2017
[7]

Fouque and R

J.-P. Fouque and R. Hu. Optimal portfolio under fractional stochastic environment. Mathematical Finance, 29(3), 2019

work page 2019
[8]

Fouque and N

J.-P. Fouque and N. Ning. Uncertain volatility models with stochastic bounds. SIAM Journal on Financial Mathematics, 9(4):1175–1207, 2018

work page 2018

Show all 20 references

[9]

Fouque, G

J.-P. Fouque, G. Papanicolaou, and R. Sircar. Derivatives in ﬁnancial markets with stochastic volatility . Cam- bridge University Press, 2000

2000
[10]

Fouque, G

J.-P. Fouque, G. Papanicolaou, and R. Sircar. Mean-reverting stochastic volatility. International Journal of theoretical and applied ﬁnance, 3(01):101–142, 2000

2000
[11]

Fouque, G

J.-P. Fouque, G. Papanicolaou, R. Sircar, and K. Solna. Multiscale stochastic volatility asymptotics. Multiscale Modeling & Simulation , 2(1):22–42, 2003

2003
[12]

Fouque, G

J.-P. Fouque, G. Papanicolaou, R. Sircar, and K. Solna. Multiscale Stochatic Volatility for Equity, Interest-Rate and Credit Derivatives. Cambridge University Press, 2011

2011
[13]

Fouque, R

J.-P. Fouque, R. Sircar, and T. Zariphopoulou. Portfolio optimization & stochastic volatility asymptotics. Mathematical Finance, 2016

2016
[14]

Karatzas and S

I. Karatzas and S. Shreve. Brownian motion. In Brownian Motion and Stochastic Calculus , pages 47–127. Springer, 1998

1998
[15]

Kramkov and W

D. Kramkov and W. Schachermayer. Necessary and suﬃcient conditions in the problem of optimal investment in incomplete markets. The Annals of Applied Probability , 13(4):1504–1516, 2003

2003
[16]

R. C. Merton. Lifetime portfolio selection under uncertainty: The continuous-time case. Review of Economics and statistics, 51:247–257, 1969

1969
[17]

R. C. Merton. Optimum consumption and portfolio rules in a continuous-time model. Journal of economic theory, 3(4):373–413, 1971

1971
[18]

O. Oleinik. On the smoothness of solutions of degenerating elliptic and parabolic equations. In Doklady Akademii Nauk, volume 163, pages 577–580. Russian Academy of Sciences, 1965

1965
[19]

Sircar and G

R. Sircar and G. Papanicolaou. Stochastic volatility, smile & asymptotics. Applied Mathematical Finance , 6:107–145, 1999

1999
[20]

Zariphopoulou

T. Zariphopoulou. Optimal investment and consumption models with non-linear stock dynamics. Mathematical Methods of Operations Research, 50(2):271–296, 1999. 17

1999

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Construction of Forward Performance Processes in Stochastic Factor Models and an Extension of Widder's Theorem

L. Avanesyan, M. Shkolnikov, and R. Sircar. Construction of forward performance processes in stochastic factor models and an extension of widder’s theorem. arXiv:1805.04535v1, 2018

work page Pith review arXiv 2018

[2] [2]

M. Bichuch. Asymptotic analysis for optimal investment in ﬁnite time with transaction costs. SIAM Journal on Financial Mathematics, 3(1):433–458, 2012

work page 2012

[3] [3]

Bichuch and R

M. Bichuch and R. Sircar. Optimal investment with transaction costs and stochastic volatility part ii: Finite horizon. SIAM Journal on Control and Optimization , 57(1):437–467, 2019

work page 2019

[4] [4]

Chacko and L

G. Chacko and L. M. Viceira. Dynamic consumption and portfolio choice with stochastic volatility in incomplete markets. Review of Financial Studies , 18(4):1369–1402, 2005

work page 2005

[5] [5]

J. Feng, M. Forde, and J.-P. Fouque. Short-maturity asymptotics for a fast mean-reverting heston stochastic volatility model. SIAM Journal on Financial Mathematics , 1(1):126–141, 2010

work page 2010

[6] [6]

Fouque and R

J.-P. Fouque and R. Hu. Asymptotic optimal strategy for portfolio optimization in a slowly varying stochastic environment. SIAM Journal on Control and Optimization , 5(3), 2017

work page 2017

[7] [7]

Fouque and R

J.-P. Fouque and R. Hu. Optimal portfolio under fractional stochastic environment. Mathematical Finance, 29(3), 2019

work page 2019

[8] [8]

Fouque and N

J.-P. Fouque and N. Ning. Uncertain volatility models with stochastic bounds. SIAM Journal on Financial Mathematics, 9(4):1175–1207, 2018

work page 2018