The classification of term structure shapes in the two-factor Vasicek model -- a total positivity approach

Martin Keller-Ressel

The classification of term structure shapes in the two-factor Vasicek model -- a total positivity approach

T0 review · 0 major / 6 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 The two-factor Vasicek model can produce exactly nine term-structure shapes, and no others, with the reachable set shrinking or growing depending on the mean-reversion gap and correlation.

desk verdict First complete classification of two-factor Vasicek term structure shapes, using total positivity; solid math, with a few typos and one boundary Wronskian argument that should be cleaned up before publication. read the letter →

arxiv 1908.04667 v3 pith:C5EH3GQB submitted 2019-08-13 q-fin.MF

Martin Keller-Ressel This is my paper · ORCID

classification q-fin.MF MSC 91G3015B4891G80

keywords termstructureshapestwo-factorVasicekmodelyieldcurveforwardtotalpositivityDescartessystemssignsequencesaffineinterestratemodels

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 gives a complete classification of every shape the forward curve and yield curve can take in the two-factor Vasicek interest-rate model. It proves that the only possible shapes are normal, inverse, humped, dipped, hump-dip, and, in certain parameter regimes, up to four additional dip/hump alternations; no other shape can be produced by any parameter and state choice. The boundary between regimes is set by how much faster the second factor mean-reverts than the first, together with the sign of the correlation between the two Brownian motions. The classification matters because it turns a model question into a list: any observed term structure whose shape is not on the relevant list cannot be generated by the model, and every shape on the list can be generated, often with extrema placed anywhere the modeler wants.

What carries the argument

The carrying object is a Descartes system in the curve derivatives. The forward-curve derivative is a linear combination of five exponentials $e^{-\alpha x}$ with $\alpha\in\{2\lambda_2,\lambda_1+\lambda_2,\lambda_2,2\lambda_1,\lambda_1\}$, and the ordering of these exponentials by $\lambda_1,\lambda_2$ makes them a Descartes system: a family of functions with the property that no linear combination can change sign more often than the sign sequence of its coefficients. The yield-curve derivative is the analogous combination with each exponential replaced by $g_\alpha(x)=x^{-2}\int_0^x y e^{-\alpha y}\,dy$, which forms a Descartes system on $[0,\infty)$ including the boundary at $x=0$. The variation-diminishing property of these systems yields the impossibility half, and a D-polynomial interpolation theorem—D-polynomial meaning a linear combination of the system's functions—yields the attainability half, by prescribing the zeroes of the derivative and then solving for model parameters.

What would settle it

Evaluate the closed-form formulas for $\partial_x f(x;z,p)$ and $\partial_x Y(x;z,p)$ on a dense parameter search; a single parameter vector and state vector producing a derivative sign sequence outside the theorem's list, for example $[-+-]$ in the scale-proximal $\rho\ge 0$ case, would disprove the classification. The boundary extension at $x=0$ is the delicate point, so a counterexample with an extra extremum arbitrarily close to maturity zero would specifically indict the Wronskian argument for the Descartes property.

Watch

Extended reading notes

Core claim

Define a hump as a local maximum and a dip as a local minimum; H and D are used as shorthands. The paper's central claim, Theorem 2.3, is that in the two-factor Vasicek model the attainable forward- and yield-curve shapes coincide and are exactly the following. If $2\lambda_1<\lambda_2$ (scale-separated: the fast factor mean-reverts more than twice as fast as the slow factor), the attainable shapes are normal, inverse, humped, dipped, HD, DH, and HDH. If $2\lambda_1>\lambda_2$ (scale-proximal) and $\rho\ge 0$, only normal, inverse, humped, dipped, and HD are attainable; if $\rho<0$, DH, HDH, DHD, and HDHD are additionally attainable. At the critical value $2\lambda_1=\lambda_2$, the positive- or negative-correlation list applies according to the sign of $\rho$. The impossibility half shows no other sign pattern of the curve derivative can occur; the attainability half constructs a parameter vector and state vector for every listed shape, in several cases with arbitrarily prescribed locations for the humps and dips.

Load-bearing premise

The impossibility part rests on the claim that every attainable forward- or yield-curve derivative changes sign at most as often as its coefficient sequence does, and that this sign-variation bound still holds on the closed half-line $[0,\infty)$, including the boundary at maturity zero.

Editorial extensions

If this is right

In the scale-proximal, nonnegatively correlated regime the model can produce at most one dip; hump-dip patterns such as DH or HDH are structurally impossible there.
Every shape on the listed sets is attainable, so observing a listed shape never requires fine-tuned parameters; in most cases the hump and dip locations can be chosen freely.
The same shape list governs forward and yield curves, so a model that is rejected by one curve's shape is rejected by the other.
The sign of the slowly reverting factor's coefficient controls the long end of the forward curve, so long-run monotonicity is decided by a single combination of state and parameters.
Observed shape alone can serve as a regime diagnostic: DH or HDH indicates either scale separation or negative correlation with scale proximity, and DHD or HDHD requires negative correlation in the scale-proximal regime.

Reading between the lines

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

The same exponential ordering argument should extend to $n$-factor affine short-rate models, where the obstacles are not the sign-variation bound but the combinatorial growth of possible sign sequences.
Since hump and dip locations are often freely prescribable, matching a qualitative curve shape or even extremum positions gives little information about the underlying parameters; shape lists are a decisive model-selection filter but not a calibration tool.
Empirically, the four-extremum shapes DHD and HDHD are fingerprints of the negative-correlation scale-proximal regime; a historical scan of yield curves for such patterns would give a direct, if informal, check of whether that regime is ever realized.

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

Colleague,

This is a solid paper and a definite contribution. It gives the first complete classification of attainable forward and yield curve shapes in the two-factor Vasicek model, and it does it with the right tool. Theorem 2.3 is the real result: depending on scale separation (2λ1<λ2 vs 2λ1>λ2) and the sign of correlation, the attainable shapes are exactly the listed ones. The proof via Descartes systems is clean and almost self-contained. I particularly like the sufficiency part: construct D-polynomials with prescribed extrema, then solve the coefficient system to realize them in the Vasicek dynamics. The paper goes beyond the bare classification, too, with state-contingent regions, strict/strong/Σ-attainability, and the observation that varying only state and covariance suffices. That is more than the abstract promises.

Where are the soft spots? The necessity direction enumerates cases and then says "after iterating through all cases" for the yield curve. A displayed table would have been nicer, but the case structure is low-dimensional and the pattern is credible. Lemma 4.7 has a typo in the displayed formula for ρ: the denominator should be √(a_{2λ1}a_{2λ2}), not √(a_{2λ1}a_{2λ1}). More substantially, the boundary Descartes property for the E systems — which rules out extra short-end extrema for yield curves — rests on the Wronskian argument in Appendix A.2. As printed, the Wronskian definition is not a square determinant and the factor in (A.8) is off by a power: for the standard Wronskian you would get (k+1)!^{-1}, not (k+1)!^{-k}. Since the factor is positive, the conclusion W>0 survives, so this is a presentation bug, not a gap in the logic. Still, because this is the single point where a failure would change the yield-curve lists, I would want the author to rewrite A.2 cleanly before I trust it without redoing the computation. I found no circularity; the Karlin results are used properly and the author's own earlier papers appear only as background.

Bottom line: this is a definite result that will likely become a standard reference for two-factor Vasicek calibration and model selection. The math is sound in its main lines; the remaining issues are exposition and a few typos. It deserves a serious referee and, after minor revision, acceptance. I would take it for a reading group if anyone in the group works on affine term structure.

Referee Report

0 major / 6 minor

Summary. The paper studies the two-factor Vasicek model and classifies all possible shapes of the forward and yield curves as the model parameters (including correlation and the two mean-reversion speeds) and the current factor state vary. The main theorem (Theorem 2.3) states that the attainable shapes are exactly: in the scale-separated regime 2λ1<λ2, the seven shapes normal, inverse, humped, dipped, HD, DH, and HDH; in the scale-proximal regime 2λ1>λ2 with ρ≥0, the five shapes normal, inverse, humped, dipped, and HD; and in the scale-proximal regime with ρ<0, nine shapes including DHD and HDHD, with the scale-critical case 2λ1=λ2 covered by the corresponding cases. The proof represents the derivatives of the forward and yield curves as D-polynomials in explicitly identified Descartes systems (Lemmas 4.1 and 4.2), uses the variation-diminishing property to bound the possible sign sequences (Theorem 4.5), and then constructs realizations via interpolation D-polynomials and by solving an algebraic system (Section 4.3). Additional sections provide state-contingent shape regions, strict/strong/Σ-attainability refinements, and asymptotic results.

Significance. The classification is a definitive answer for a benchmark multi-factor short-rate model and is likely to be a standard reference. Its main strength is that it is a genuine existence/characterization theorem rather than an empirical or simulation-based statement: no data are fitted and no parameter is introduced ad hoc. The total-positivity machinery is applied in a new way to term-structure shape analysis, and the paper gives explicit constructive sufficiency via interpolation polynomials, including a nontrivial boundary-interpolation lemma (Lemma A.2) for the scale-proximal negative-correlation regime. The refinements on strict, strong, and Σ-attainability add practical value and are proved from the same construction. If the proof is correct, the paper also produces falsifiable predictions: in each regime, any shape outside the lists is impossible for every parameter and state vector.

minor comments (6)

[Section 4.2, proof of Theorem 2.3 (necessity)] The proof writes "after iterating through all cases" and "the same lists of shapes are obtained" for the yield curve, but provides only two examples for the forward curve. Since the yield-curve case uses only the weaker ⊆ constraint rather than the tail relation, the omitted enumeration is not completely transparent. Please add a table covering all sign combinations of (w1,w2), the three scale regimes, and the forward/yield cases, so that the necessity part is fully checkable.
[Appendix A.2, Eq. (A.3)] The displayed Wronskian is not a square matrix (it has k rows and k+1 columns). It should be W(φ1,...,φk)(x)=det([φ_i^{(j-1)}(x)]_{i,j=1}^k). Please correct the definition.
[Appendix A.2, Eq. (A.8)] The constant factor is wrong: inserting (A.7) into the standard Wronskian gives (k+1)!^{-1}, not (k+1)!^{-k}. The positivity conclusion is unaffected, but the formula should be corrected so that the boundary Descartes property of E is properly documented.
[Lemma 4.7(c), Eq. (4.6)] The square root in the denominator has a_{2λ1}a_{2λ1}; it should be a_{2λ1}a_{2λ2}. As written, the condition is unverifiable and likely a typo.
[Abstract and Section 2.3] The abstract says "up to four additional shapes can be produced," but Section 2.3 correctly states that the number of additional shapes can grow up to six (scale-proximal, ρ<0; Theorem 2.3(c)). Please align the abstract with the theorem.
[Throughout] Typos: "Auxilliary" should be "Auxiliary" (Appendix A title); "descreasing" should be "decreasing" (Table 1); "satsifies" should be "satisfies" (Theorem 4.5); "is is given in section 4" should be "is given in Section 4" (end of Section 2.3); "in Sec. 1.1" should be "in Sec. 3.1" in the proof of Corollary 4.6. Also "withϵ in some small set [0,δ)" in Section 5.2 should use the same symbol ε consistently.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the classification is derived self-contained from explicit bond-price formulas and standard external theorems on Descartes systems.

full rationale

The paper's derivation chain is self-contained. The classification in Theorem 2.3 is obtained from explicit bond-price formulas (2.2)–(2.6), the computed derivatives in Lemma 4.3, and standard variation-diminishing results for Descartes systems quoted from Karlin and Karlin–Studden. Sufficiency is established by constructing D-polynomials with prescribed sign sequences via interpolation theorems (Theorem 3.8, Lemma A.2) and then explicitly solving the coefficient-to-parameter system (4.5). No empirical quantity is fitted to a subset of data and renamed as a prediction; no parameter is defined in terms of a target shape; and no load-bearing step reduces to a self-citation. The author's earlier one-factor results are used only for background comparison, not as justification for the two-factor classification. The one potentially nonstandard ingredient, the boundary Descartes property of the E-systems at x=0, is proven in Appendix A.2 via a Wronskian argument; whether that proof is flawlessly executed is a correctness question, not a circularity one. The paper is therefore free of definitional, fitted-prediction, or self-citation circularity.

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

The theorem quantifies over all model parameters rather than fitting them, so no empirical constants enter. The only inputs are standard total positivity theorems and the affine Vasicek model structure.

assumptions (6)

standard math Karlin's variation-diminishing theorem for totally positive kernels (Theorem 3.3)
Used to relate sign sequences of integrals to sign sequences of integrands; cited to Karlin 1968, Chapter 5.
standard math Variation-diminishing property of Descartes systems (Theorem 3.6)
Core bound on sign changes of the forward and yield curve derivatives in terms of coefficient signs; cited to Karlin-Studden and Borwein-Erdelyi.
standard math Existence of extremal D-polynomials with prescribed zeros (Theorem 3.8)
Used in the sufficiency part to build forward and yield derivatives with required shapes; a proof is supplied in Appendix A.1.
domain assumption The two-factor Vasicek affine bond pricing formula (Eqs. 2.2 to 2.6)
Model setup: rt equals the sum of two OU factors, with bond prices exponential-affine as in Dai-Singleton. This is the mathematical starting point for all derivative formulas.
domain assumption Shape of a smooth curve is captured by the sign sequence of its derivative
Definition in Section 2.1; smoothness of Vasicek curves and the finite number of sign changes are used throughout the classification.
domain assumption Mean-reversion speeds are strictly positive and ordered lambda1 < lambda2
Convention in Section 2.2; factor labeling makes this assumption without loss of generality.

how reviews work

0 comments

Cite this review

Pith. "Pith review of The classification of term structure shapes in the two-factor Vasicek model -- a total positivity approach." pith.science (2026). https://pith.science/paper/C5EH3GQB

@misc{pith2026190804667,
  author       = {Pith},
  title        = {Pith review of: The classification of term structure shapes in the two-factor Vasicek model -- a total positivity approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C5EH3GQB}},
  note         = {Machine review of arXiv:1908.04667}
}

read the original abstract

We provide a full classification of all attainable term structure shapes in the two-factor Vasicek model of interest rates. In particular, we show that the shapes normal, inverse, humped, dipped and hump-dip are always attainable. In certain parameter regimes up to four additional shapes can be produced. Our results apply to both forward and yield curves and show that the correlation and the difference in mean-reversion speeds of the two factor processes play a key role in determining the scope of attainable shapes. The key mathematical tool is the theory of total positivity, pioneered by Samuel Karlin and others in the 1950ies.

Figures

Figures reproduced from arXiv: 1908.04667 by the authors.

**Figure 1.** State-contingent term structure shapes in the two-dimensional Vasicek model. The attainable shapes of the forward curve (panel A) and the yield curve (panel B) in different regions (delimited in black) of the state space R 2 of the factors (z1, z2) are shown. The underlying model is assumed to be scaleproximal and positively correlated, see case (b) of Theorem 2.3. The contour lines of the risk-neutral stationary d… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

[1]

Totally positive matrices

Tsuyoshi Ando. Totally positive matrices. Linear algebra and its applications , 90:165--219, 1987

work page 1987
[2]

Polynomials and polynomial inequalities , volume 161

Peter Borwein and Tam \'a s Erd \'e lyi. Polynomials and polynomial inequalities , volume 161. Springer Science & Business Media, 1995

work page 1995
[3]

Interest rate models-theory and practice: with smile, inflation and credit

Damiano Brigo and Fabio Mercurio. Interest rate models-theory and practice: with smile, inflation and credit . Springer Science & Business Media, 2007

work page 2007
[4]

A theory of the term structure of interest rates

John C Cox, Jonathan E Ingersoll Jr, and Stephen A Ross. A theory of the term structure of interest rates. Econometrica , 53(2):385--408, 1985

work page 1985
[5]

Moderne Finanzmathematik -- Theorie und praktische Anwendung, Band 2

Sascha Desmettre and Ralf Korn. Moderne Finanzmathematik -- Theorie und praktische Anwendung, Band 2 . Springer, 2018

work page 2018
[6]

Specification analysis of affine term structure models

Qiang Dai and Kenneth J Singleton. Specification analysis of affine term structure models. The Journal of Finance , 55(5):1943--1978, 2000

work page 1943
[7]

Handbook of linear algebra

Leslie Hogben. Handbook of linear algebra . Chapman and Hall/CRC, 2013

work page 2013
[8]

Pricing interest-rate-derivative securities

John Hull and Alan White. Pricing interest-rate-derivative securities. The review of financial studies , 3(4):573--592, 1990

work page 1990

Show all 16 references

[9]

Total positivity , volume 1

Samuel Karlin. Total positivity , volume 1. Stanford University Press, 1968

1968
[10]

Monotonicity and convexity of option prices revisited

Masaaki Kijima. Monotonicity and convexity of option prices revisited. Mathematical Finance , 12(4):411--425, 2002

2002
[11]

Correction to: Yield curve shapes and the asymptotic short rate distribution in affine one-factor models

Martin Keller-Ressel. Correction to: Yield curve shapes and the asymptotic short rate distribution in affine one-factor models. Finance and Stochastics , 22(2):503--510, 2018

2018
[12]

Yield curve shapes and the asymptotic short rate distribution in affine one-factor models

Martin Keller-Ressel and Thomas Steiner. Yield curve shapes and the asymptotic short rate distribution in affine one-factor models. Finance and Stochastics , 12(2):149 -- 172, 2008

2008
[13]

Tchebycheff systems: with applications in analysis and statistics

Samuel Karlin and William J Studden. Tchebycheff systems: with applications in analysis and statistics . Interscience, 1966

1966
[14]

Level--slope--curvature--fact or artefact? Applied Mathematical Finance , 14(2):105--130, 2007

Roger Lord and Antoon Pelsser. Level--slope--curvature--fact or artefact? Applied Mathematical Finance , 14(2):105--130, 2007

2007
[15]

Correlation matrices of yields and total positivity

Ernesto Salinelli and Carlo Sgarra. Correlation matrices of yields and total positivity. Linear algebra and its applications , 418(2-3):682--692, 2006

2006
[16]

An equilibrium characterization of the term structure

Oldrich Vasi c ek. An equilibrium characterization of the term structure. Journal of Financial Economics , 5:177--188, 1977

1977

Pith tools

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

[1] [1]

Totally positive matrices

Tsuyoshi Ando. Totally positive matrices. Linear algebra and its applications , 90:165--219, 1987

work page 1987

[2] [2]

Polynomials and polynomial inequalities , volume 161

Peter Borwein and Tam \'a s Erd \'e lyi. Polynomials and polynomial inequalities , volume 161. Springer Science & Business Media, 1995

work page 1995

[3] [3]

Interest rate models-theory and practice: with smile, inflation and credit

Damiano Brigo and Fabio Mercurio. Interest rate models-theory and practice: with smile, inflation and credit . Springer Science & Business Media, 2007

work page 2007

[4] [4]

A theory of the term structure of interest rates

John C Cox, Jonathan E Ingersoll Jr, and Stephen A Ross. A theory of the term structure of interest rates. Econometrica , 53(2):385--408, 1985

work page 1985

[5] [5]

Moderne Finanzmathematik -- Theorie und praktische Anwendung, Band 2

Sascha Desmettre and Ralf Korn. Moderne Finanzmathematik -- Theorie und praktische Anwendung, Band 2 . Springer, 2018

work page 2018

[6] [6]

Specification analysis of affine term structure models

Qiang Dai and Kenneth J Singleton. Specification analysis of affine term structure models. The Journal of Finance , 55(5):1943--1978, 2000

work page 1943

[7] [7]

Handbook of linear algebra

Leslie Hogben. Handbook of linear algebra . Chapman and Hall/CRC, 2013

work page 2013

[8] [8]

Pricing interest-rate-derivative securities

John Hull and Alan White. Pricing interest-rate-derivative securities. The review of financial studies , 3(4):573--592, 1990

work page 1990