Pith. sign in

REVIEW 5 minor 33 references

On the Weak Error for Local Stochastic Volatility Models

T0 review · 0 major / 5 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 the half-step particle scheme for local stochastic volatility models has a total weak error of order $h + \delta^{1-} + (1/\sqrt{N})\delta^{-1} \exp((T/2h)\log(C\delta^{-1} h^{-1/2}))$, with each term traced to a…

desk verdict A genuinely new and well-supported weak-error decomposition for LSV particle schemes; the core math deserves refereeing, with a typo and the numerics needing cleanup. read the letter →

arxiv 2506.10817 v1 pith:POJWGYWA submitted 2025-06-12 math.PR q-fin.CP

Peter K. Friz , Benjamin Jourdain , Thomas Wagenhofer , Alexandre Zhou This is my paper · ORCID
classification math.PRq-fin.CP MSC 91G6060H3591G8065C30
keywords LocalstochasticvolatilityweakerrorrateMcKean-Vlasovdynamicsparticlemethodhalf-stepschemeEulerdiscretizationpropagationofchaosconditionalexpectation
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

Local stochastic volatility models are the standard way to calibrate option prices on the fly, but their natural particle approximation is tied to an ill-posed McKean–Vlasov equation. This paper sidesteps the ill-posedness by discretizing first: it starts from a well-defined Euler scheme, adds a half-step construction that makes conditional expectations computable in closed form through Gaussian convolutions, and only then introduces particles. The main result is a complete weak-error decomposition: the difference between the particle price and the target price is bounded by the Euler bias of order $h$, a regularization bias of order $\delta^{1-}$, and a statistical error of order $1/\sqrt{N}$ multiplied by an explicit exponential factor. A companion theorem proves that the underlying Euler scheme with exact conditional expectations already converges with weak order one. The practical payoff is a parameter recipe: to reach a desired tolerance, one can choose the step size, the smoothing, and the number of particles according to explicit balances.

What carries the argument

The half-step scheme (Equation (14)) is the load-bearing object: each Euler step is split into two half-steps, and the second half-step adds an independent Brownian increment with variance $\lambda = c_{\min}^2 \bar{\rho}^2 h$. This turns the conditional expectation into a ratio of Gaussian convolutions, $E[\xi_{jh}^2 | \widetilde{X}_{jh} = y] = \frac{E[\xi_{jh}^2 \varphi_\lambda(y - \widetilde{X}_{jh-h/2})]}{E[\varphi_\lambda(y - \widetilde{X}_{jh-h/2})]}$ (Lemma 4.5), which a particle system approximates by replacing the expectations with empirical averages plus a $\delta$ regularization. Because this map and its inverse are Lipschitz with explicit constants (Lemma 5.5), a discrete Gronwall argument controls the interacting-particle error, and a parabolic-PDE-regularity argument of the classical weak-error-expansion type controls the Euler bias.

What would settle it

A direct Monte Carlo estimate of $\|\widehat{E}_\delta(\xi^2|X) - E[\xi^2|X]\|_{L^p}$ for $\rho$ very close to $\pm 1$ (for example $\rho = 1 - 10^{-6}$ with a rough-Bergomi-style $\xi$) should decay like $(\delta^p / (-\log \delta))^{1/p}$ according to Lemma 4.8, but the proof relies on a density lower bound $E[\varphi_\lambda(x - X)] \gtrsim \lambda^{-1/2}\int f$ that degenerates as $\lambda = c_{\min}^2 \bar{\rho}^2 h$ shrinks; if the measured $L^p$ error fails to follow this rate as $\delta \to 0$, the error decomposition behind Theorem 5.4 collapses.

Watch

Extended reading notes

Core claim

The central claim is Theorem 5.4: for a test function that is three times weakly differentiable with polynomial growth, under the ellipticity and boundedness Assumption 2 and the non-perfect-correlation Assumption 1 ($\rho^2 < 1$), the weak error of the half-step particle scheme $\widetilde{X}^{h,(1),N,\delta}_T$ against the target $Y_T$ is bounded by $h + \delta^{1-} + \frac{1}{\sqrt{N}}\delta^{-1}\exp\bigl(\frac{T}{2h}\log(C\delta^{-1}h^{-1/2})\bigr)$. The proof splits the error into exactly three traceable parts: Theorem 3.3 bounds the Euler discretization with exact conditional expectations by order $h$ (with rates $\sqrt{h}$, $h\log(1/h)$, or $h$ depending on the test-function regularity); Theorem 4.6 charges the $\delta$-regularized estimator only $\delta^{1/p}$; and Theorem 5.2 gives a propagation-of-chaos estimate with an explicit exponential constant. The paper additionally shows that a generic Nadaraya–Watson kernel (Theorem 7.2) yields a similar but strictly worse bound, with an extra $\varepsilon h^{-1}$ bias and a much larger exponential factor, which makes the half-step construction the preferred route whenever $\rho$ is away from $\pm 1$.

Load-bearing premise

The load-bearing premise is Assumption 1, $\rho^2 < 1$: the volatility factor $\xi$ must not be perfectly correlated with the Brownian motion driving the price, because the half-step scheme's closed-form conditional expectation needs the extra independent Gaussian variance $\lambda = c_{\min}^2 \bar{\rho}^2 h$ to be positive; at $\rho = \pm 1$ that smoothing vanishes and, as the paper notes, an analogous bound would require an unverifiable Hölder assumption on $x \mapsto E[\xi^2 | X = x]$.

Editorial extensions

If this is right

  • Any combination of $h$, $\delta$, and $N$ that satisfies the bound of Theorem 5.4 achieves a prescribed total error; the paper gives the analogous explicit recipe for the kernel variant in Remark 7.12.
  • The Euler scheme with exact conditional expectations has weak order one for test functions that are only weakly differentiable, with rates $\sqrt{h}$ for one derivative and $h\log(1/h)$ for two derivatives.
  • The half-step scheme strictly outperforms generic kernel regression on parameter dependence: with $h = \delta$ and $\varepsilon = h^2$, the exponential factor in the statistical error is $h^{-3/2}$ in the log-exponent instead of $h^{-9}$ without the half-step construction.
  • The scheme matches one-dimensional marginals, not paths: numerical experiments with fake Brownian motion show the variance of the quadratic variation does not vanish as $N$ grows, which is consistent with convergence in law only.
  • For smooth payoffs the numerical error sits inside the theoretical bound, and even non-smooth call-type payoffs show the absolute error decreasing quickly, though with a rougher convergence pattern.

Reading between the lines

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

  • The explicit exponential factor suggests a practical rule: the smoothing $\delta$ should not be chosen smaller than about $h^{1/2}$ times a constant, otherwise the statistical error inflates; an optimization over $(h, \delta, N)$ under a fixed computational budget would make this trade-off quantitative.
  • Assumption 1 excludes path-dependent volatility models, where the volatility factor is driven by the same Brownian motion as the price; the paper's Remark 7.13 indicates that without $\rho^2 < 1$ one would need an unverifiable Hölder regularity condition on $x \mapsto E[\xi^2 | X = x]$, so the method as stated is tied to models with a separate, not perfectly correlated, volatility driver.
  • The 'Euler first, particle second' ordering cleanly separates two sources of error that earlier analyses entangled; the same ordering could be exported to other McKean–Vlasov equations whose coefficients involve conditional expectations, provided a Gaussian smoothing step can be inserted.
  • The comparison between Theorem 5.4 and Theorem 7.2 shows that the half-step Gaussian smoothing is doing genuine work rather than being a cosmetic choice, which suggests that implementations should prefer it whenever the correlation parameter is not too close to $\pm 1$.
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.

Referee Report

0 major / 5 minor

Summary. The paper derives explicit weak-error bounds for a class of local stochastic volatility models. Starting from an Euler scheme with the exact conditional expectation, it proves (Theorem 3.3) first-order weak convergence to the Gyöngy-mimicking Markov process for test functions with three weak derivatives, with degraded rates for rougher test functions. It then introduces a half-step Euler scheme with a Gaussian-regularized estimator of the conditional expectation, proves a propagation-of-chaos estimate (Theorem 5.2), and combines the ingredients into a full parameter-explicit error bound for the interacting particle scheme (Theorems 1 and 5.4). A second part treats Nadaraya-Watson kernel estimators (Theorem 7.2) and compares the resulting rates. Numerical experiments illustrate the scaling in h, delta, and N.

Significance. The contribution is methodologically solid and novel in its 'Euler first, particle second' ordering. The proof of Theorem 3.3 is a coherent Talay-Tubaro expansion with honest tracking of the final grid interval and of the singular weights; Theorem 5.2's Gronwall argument splits the empirical fluctuation into an i.i.d. variance term and a Lipschitz interaction term. The half-step scheme yields an exact conditional-expectation identity (Lemma 4.5), which is an elegant tool and a practical advantage. All error constants are derived rather than fitted, and the dependence on h, delta, and N is stated explicitly. The main limitation, Assumption 1 (rho^2 < 1), is structural and clearly acknowledged in Remark 7.13; it excludes fully correlated or path-dependent volatility, but this is a scope restriction rather than an internal inconsistency.

minor comments (5)
  1. [Theorem 5.4 and its proof] The printed bound in Theorem 5.4 does not match the proof. Theorem 5.2 gives ||eX_T^{h,(1),delta} - eX_T^{h,(1),N,delta}||_{L2} <= C N^{-1/2} h^{1/4} delta^{-1/2} exp((T/2) h^{-1} log(C delta^{-1} h^{-1/2})), so the displayed prefactor N^{-1/2} delta^{-1} in Theorem 5.4 should be N^{-1/2} delta^{-1/2} (the factor h^{1/4} can be dropped since h < 1). In the proof of Theorem 5.4, the logarithm contains h^{-3/2}; this should be h^{-1/2} to agree with Theorem 5.2. Theorem 1 also omits the delta^{-1/2} prefactor; please either state Theorem 1 as an informal summary or align it with the precise bound in Theorem 5.4.
  2. [Theorem 5.2 statement] The right-hand side 'c/N h^{1/2} delta^{-1} exp(...)' is ambiguous: it should be written as (c/N) h^{1/2} delta^{-1} exp(...) or with explicit parentheses, since the current typesetting could be read as c/(N h^{1/2} delta^{-1}).
  3. [Section 2, notation] In the definition of the C1 norm there is a stray 'and .' at the end of the sentence; this appears to be a typographical artifact and should be removed.
  4. [Section 6.2] In the description of the reference Monte-Carlo simulation, '90 2 Monte Carlo steps' should presumably be '90^2 Monte Carlo steps' (or an explicit number); please correct this typo.
  5. [Lemma 3.4 statement] The bound in Lemma 3.4 is stated for all (t,x) in [0,T) x R with no positive-part convention on the exponent (2i+j-l)/2; when 2i+j < l the right-hand side would blow up as t approaches T. Since the applications only use cases with 2i+j >= l, this is not a blocking issue, but the statement should clarify that the exponent is meant as max((2i+j-l)/2, 0) or that only the needed parameter ranges are asserted.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the weak-error bounds are derived from explicit estimates; the sole self-citation is re-proved in the appendix.

full rationale

The paper's central claims are derived, not fitted. Theorem 3.3 is a standard Talay-Tubaro telescoping argument with PDE regularity estimates (Lemmas 3.4 and 3.5) and does not presuppose its conclusion. The half-step scheme is introduced with reference to the author's thesis [28], but the load-bearing identity, Lemma 4.5, is proved in the paper via Lemma 9.5, so the self-citation is not the actual support. The regularization bias is quantified in Lemma 4.8 by a direct Gaussian-density estimate, and Theorem 4.6 combines this with Theorem 4.1, whose proof is self-contained. Theorem 5.2 is a propagation-of-chaos argument that splits the empirical error into an i.i.d. variance term of order 1/(N lambda) and a Lipschitz interaction term closed by a discrete Gronwall lemma; the constants and exponential prefactors are computed from Lemmas 5.5, 9.8, and 9.9, not imposed by the desired rate. Theorem 5.4 then assembles these pieces, and every parameter h, delta, N appears with an explicit error cost. No fitted constant is renamed as a prediction, and no uniqueness theorem or ansatz is imported from the authors' prior work to force the choice of scheme. The only self-referential element is the half-step scheme from [28], but it is used as a tool whose needed property is re-established inside the paper, so it does not make the argument circular. The discrepancy between the printed prefactor/log-exponent in Theorem 5.4 and the proof's L2 estimate appears to be a typo or a conservative restatement and does not affect the derivation structure.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

No new particles, forces, mediators, or dimensions are introduced. The half-step scheme and the regularised estimators bE_delta and bE^epsilon_delta are numerical constructions whose approximation error is explicitly quantified rather than postulated. Tuning parameters h, delta, epsilon, and N all carry explicit error terms in the bounds, so they are not hidden free parameters; only c_min in the numerics is hand-chosen.

free parameters (1)
  • c_min (numerics only) = 0.05 (hand-picked, rounded down)
    In the theory c_min = ac/(2b) is derived from Assumption 2. In the fake-Brownian-motion experiment (Section 6.1, footnote) the rough-Bergomi factor xi is unbounded, violating Assumption 2, and c_min is chosen by hand so that the square roots in (14) stay positive. All error constants blow up as c_min approaches 0, so this hand choice is load-bearing for the numerical illustration.
assumptions (5)
  • domain assumption Gyongy's mimicking theorem (Theorem 3.1, quoted from [15])
    Defines the Markov target Y with the same one-dimensional marginals as the LSV dynamics; the paper measures all errors against E[f(Y_T)]. Cited, not reproven.
  • standard math Feynman-Kac formula and parabolic regularity estimates (PDE (6) and Lemma 9.1 from Friedman [11])
    Provides the derivative bounds partial_t^i partial_x^j u <= C (T-t)^{-(2i+j-l)/2} used in every increment of the telescoping expansion in Theorem 3.3.
  • domain assumption Assumption 1 (rho^2 < 1)
    Needed for the half-step scheme (14): lambda = c_min^2 rhobar^2 h is positive only when rho^2 < 1, so the extra Brownian increment has positive variance and the conditional expectation becomes a Gaussian convolution (Lemma 4.5).
  • domain assumption Assumption 2 (sigma bounded, C^4, uniformly elliptic; a <= xi <= b)
    Uniform ellipticity gives the lower bound c_min and bounds the square roots in (14); sigma regularity gives the PDE derivative estimates; bounded xi gives the L-infinity bounds used in Lemmas 4.8 and 5.5.
  • standard math Exponential tail bound for processes with bounded quadratic variation (Lemma 9.3, from Revuz and Yor)
    Used to control P(|eX^{h,delta}| > M) in Lemma 4.8 and to control moment growth in the proof of Theorem 3.3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of On the Weak Error for Local Stochastic Volatility Models." pith.science (2026). https://pith.science/paper/POJWGYWA

@misc{pith2026250610817,
  author       = {Pith},
  title        = {Pith review of: On the Weak Error for Local Stochastic Volatility Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/POJWGYWA}},
  note         = {Machine review of arXiv:2506.10817}
}
read the original abstract

Local stochastic volatility refers to a popular model class in applied mathematical finance that allows for "calibration-on-the-fly", typically via a particle method, derived from a formal McKean-Vlasov equation. Well-posedness of this limit is a well-known problem in the field; the general case is largely open, despite recent progress in Markovian situations. Our take is to start with a well-defined Euler approximation to the formal McKean-Vlasov equation, followed by a newly established half-step-scheme, allowing for good approximations of conditional expectations. In a sense, we do Euler first, particle second in contrast to previous works that start with the particle approximation. We show weak order one for the Euler discretization, plus error terms that account for the said approximation. The case of particle approximation is discussed in detail and the error rate is given in dependence of all parameters used.

Figures

Figures reproduced from arXiv: 2506.10817 by the authors.

Figure 1
Figure 1. shows that the observed error is well within our theoretical error range. For this chosen function the error does not significantly change depending on the number of steps. It can also be seen that, as expected, a higher number of paths leads to a smaller error. In [PITH_FULL_IMAGE:figures/full_fig_p022_1.png] view at source ↗
Figure 3
Figure 3. Logplot of error for cos(x) [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
Figure 5
Figure 5. Plot of the variance of the quadratic variation. [PITH_FULL_IMAGE:figures/full_fig_p023_5.png] view at source ↗
Figures from the paper (2 more)
Figure 6
Figure 6. Figure 6: Error plot for differente δ. 6.2 Tanh model In this section we use a non-trivial volatility, namely we choose σ(x) = 3 4 + tanh(x) 4 . Such a toy model was previously introduced by [10] and has similar properties as required in Assumption 2, due to boundedness and smoo…
Figure 7
Figure 7. Figure 7: Error for payoff function cos(x) [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 25 canonical work pages

  1. [1]

    Abergel and R

    F. Abergel and R. Tachet. A nonlinear partial integro-differential equation from mathematical finance. Discrete and Continuous Dynamical Systems - Series A , 27(3):907–917, 2010. doi:10.3934/dcds.2010.27.907

    work page doi:10.3934/dcds.2010.27.907 2010

  2. [2]

    Bayer, D

    C. Bayer, D. Belomestny, O. Butkovsky, and J. Schoenmakers. A reproducing kernel Hilbert space approach to singular local stochastic volatility McKean–Vlasov models, 2024. doi:10.1007/s00780-024-00541-5

    work page doi:10.1007/s00780-024-00541-5 2024

  3. [3]

    Beiglb¨ ock, G

    M. Beiglb¨ ock, G. Lowther, G. Pammer, and W. Schachermayer. Faking Brownian motion with continuous Markov martingales. Finance and Stochastics , 28(1):259–284, 2024. doi:10.1007/s00780-023-00526-w

    work page doi:10.1007/s00780-023-00526-w 2024

  4. [4]

    Mimicking the marginal distributions of a semimartingale

    A. Bentata and R. Cont. Mimicking the marginal distributions of a semimartingale, 2012. arXiv preprint. arXiv:0910.3992

    work page Pith review arXiv 2012

  5. [5]

    Bouchard and N

    B. Bouchard and N. Touzi. Discrete-time approximation and Monte-Carlo simulation of backward stochastic differential equations. Stochastic Processes and their Applications , 111(2):175–206, 2004. doi:10.1016/j.spa.2004.01.001

    doi:10.1016/j.spa.2004.01.001 2004

  6. [6]

    Brunick and S

    G. Brunick and S. Shreve. Mimicking an Itˆ o process by a solution of a stochastic differential equation. The Annals of Applied Probability , 23(4):1584 – 1628, 2013. doi:10.1214/12-AAP881

    doi:10.1214/12-aap881 2013

  7. [7]

    M. F. Djete. Non–regular McKean–Vlasov equations and calibration problem in local stochastic volatility models, 2022. arXiv preprint. arXiv:2208.09986. 33

    arXiv 2022

  8. [8]

    B. Dupire. Pricing with a smile. Risk, 7:18–20, 1994

    work page 1994

Show all 33 references
  1. [9]

    B. Dupire. Pricing and hedging with smiles. In M. A. Dempster and S. R. Pliska, editors, Mathematics of Derivative Securities , pages 103–111. Cambridge University Press, 1997

    1997

  2. [10]

    Forde and A

    M. Forde and A. Jacquier. Small-time asymptotics for an uncorrelated local-stochastic volatility model. Appl. Math. Finance , 18(6):517–535, 2011. doi:10.1080/1350486X.2011.591159

    2011

  3. [11]

    Friedman

    A. Friedman. Partial differential equations of parabolic type . Prentice-Hall, Inc., Englewood Cliffs, NJ, 1964

    1964

  4. [12]

    Guyon and P

    J. Guyon and P. Henry-Labordere. Nonlinear Option Pricing . Chapman and Hall/CRC, 1st edition, 2013. doi:10.1201/b16332

    2013 doi

  5. [13]

    Guyon and P

    J. Guyon and P. Henry-Labord` ere. Being particular about calibration. Risk, 25(1):88–93, 01

  6. [14]

    Guyon and J

    J. Guyon and J. Lekeufack. Volatility is (mostly) path-dependent. Quant. Finance, 23(9):1221–1258, 2023. doi:10.1080/14697688.2023.2221281

    2023

  7. [15]

    Gy¨ ongy

    I. Gy¨ ongy. Mimicking the one-dimensional marginal distributions of processes having an Itˆ o differential. Probability Theory and Related Fields , 71:501–516, 1986. doi:10.1007/BF00699039

    1986 doi

  8. [16]

    Hamza and F

    K. Hamza and F. C. Klebaner. A family of non-Gaussian martingales with Gaussian marginals. Journal of Applied Mathematics and Stochastic Analysis , 2007:19 p., 2007. doi:10.1155/2007/92723

    2007 doi

  9. [17]

    Hirsch, C

    F. Hirsch, C. Profeta, B. Roynette, and M. Yor. Peacocks and Associated Martingales, with Explicit Constructions . Bocconi & Springer Series. Springer Milano, 1 edition, 2011. doi:10.1007/978-88-470-1908-9

    2011 doi

  10. [18]

    Jourdain and A

    B. Jourdain and A. Zhou. Existence of a calibrated regime switching local volatility model. Math. Finance, 30(2):501–546, 2020. doi:10.1111/mafi.12231

    2020 doi

  11. [19]

    H. G. Kellerer. Markov-Komposition und eine Anwendung auf Martingale. Mathematische Annalen, 198:99–122, 1972. doi:10.1007/BF01432281

    1972 doi

  12. [20]

    Kohatsu-Higa and S

    A. Kohatsu-Higa and S. Ogawa. Weak rate of convergence for an Euler scheme of nonlinear SDE’s. Monte Carlo Methods Appl. , 3(4):327–345, 1997. doi:10.1515/mcma.1997.3.4.327

    1997 doi

  13. [21]

    Konakov and E

    V. Konakov and E. Mammen. Edgeworth type expansions for Euler schemes for stochastic differential equations. Monte Carlo Methods Appl. , 8(3):271–285, 2002. doi:10.1515/mcma.2002.8.3.271

    2002 doi

  14. [22]

    Lacker, M

    D. Lacker, M. Shkolnikov, and J. Zhang. Inverting the Markovian projection, with an application to local stochastic volatility models. The Annals of Probability , 48(5):2189 – 2211, 2020. doi:10.1214/19-AOP1420

    2020 doi

  15. [23]

    Li and J

    Q. Li and J. S. Racine. Nonparametric Econometrics: Theory and Practice . Princeton University Press, Princeton, NJ, 2007

    2007

  16. [24]

    Muguruza

    A. Muguruza. Not so particular about calibration: Smile problem resolved, 2019. arXiv preprint. arXiv:1909.13366

    2019 arXiv

  17. [25]

    Reisinger and M

    C. Reisinger and M. O. Tsianni. Convergence of the Euler–Maruyama particle scheme for a regularised McKean–Vlasov equation arising from the calibration of local-stochastic volatility models, 2024. doi:10.1007/978-3-031-59762-6_28 . 34

    2024 doi

  18. [26]

    Revuz and M

    D. Revuz and M. Yor. Continuous Martingales and Brownian Motion , volume 293 of Grundlehren der mathematischen Wissenschaften . Springer-Verlag, Berlin, third edition,

  19. [27]

    Talay and L

    D. Talay and L. Tubaro. Expansion of the global error for numerical schemes solving stochastic differential equations. Stochastic Anal. Appl. , 8(4):483–509, 1990. doi:10.1080/07362999008809220

    1990 doi

  20. [28]

    A. Zhou. Theoretical and numerical study of problems nonlinear in the sense of McKean in finance. Phd thesis, Universit´ e Paris Est,´Ecole des Ponts Paris Tech, Oct. 2018. URL: https://inria.hal.science/tel-01957638. 35 9 Appendix 9.1 Proof of Lemma 3.4 The following lemma is...

    2018

  21. [31]

    ∀x ∈ R, ∀γ ∈ (0, 1], |x|φλ(x) ≤ (eγ)− 1 2 λ 1−γ 2 φ1−γ λ (x),

  22. [32]

    ∀x, y∈ R with |x − y| < ε,∀γ ∈ (0, 1], φλ(y) ≥ (2πλ) γ 2 φλ(x)1+γe− ε2 γλ ,

  23. [33]

    ∀x ∈ R, ∀γ >0, φ1+γ λ (x) ≤ (2πλ)− γ 2 φλ(x). Proof. (1): We use the fact that R ∋ y 7→ |y|e− y2 2 is bounded by e− 1 2 and see that for x >0 xφλ(x) = xe− x2 γ 2λ 1√ 2πλ γ φ1−γ λ = x √γ√ λ e− x2 γ 2λ λ 1−γ 2 γ− 1 2 1√ 2π γ φ1−γ λ ≤ (eγ)− 1 2 λ 1−γ 2 φ1−γ λ . By symmetry the fi...

  24. [1999]

    doi:10.1007/978-3-662-06400-9

    doi

  25. [2012]

    URL: https://www.risk.net/markets/2135540/being-particular-about-calibration

Pith tools

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