pith. sign in

arxiv: 2606.20406 · v1 · pith:TDOCSYQGnew · submitted 2026-06-18 · 📊 stat.ME · stat.CO

Flexible modeling of bimodal distributions via skewed-t mixtures

Marco Bee , Flavio Santi This is my paper

Pith reviewed 2026-06-26 16:06 UTC · model grok-4.3

classification 📊 stat.ME stat.CO
keywords skewed-t distributionmixture modelbimodal distributionEM algorithmfinancial returnsS&P 500heavy-tailed dataskewness test
0
0 comments X

The pith

Mixtures of skewed-t distributions model bimodal skewed heavy-tailed data and confirm bimodality in S&P 500 returns.

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

The paper develops a mixture model built from location-scale skewed-t components to handle data that is simultaneously bimodal, asymmetric, and heavy-tailed. It derives maximum likelihood estimates through the EM algorithm and supplies a likelihood ratio test for the hypothesis of zero skewness in any component. Simulations establish that the model recovers parameters accurately when the data truly follow the assumed form and outperforms a competing g-and-h mixture when the data-generating process differs. Applied to S&P 500 returns, the fitted mixture supports two distinct modes, implying the market has spent time in bearish or bullish states rather than hovering near a single fundamental value.

Core claim

A finite mixture of location-scale skewed-t distributions, with parameters obtained by EM and skewness tested by likelihood ratio, recovers parameters precisely in correctly specified simulations, fits misspecified data better than g-and-h mixtures, and applied to S&P 500 distortion data establishes bimodality indicating the US stock market has historically been in bearish or bullish conditions rather than near its fundamental value.

What carries the argument

Mixture of Fernández-Steel skewed-t distributions estimated by EM algorithm together with a likelihood ratio test for component skewness.

If this is right

  • The model can be extended by replacing the skewed-t base distribution with any other symmetric distribution while retaining the same EM estimation and skewness test.
  • Parameter estimates remain accurate when the data truly follow the mixture and the model still captures the shape of data generated from a different heavy-tailed bimodal family.
  • The likelihood ratio test can be used to decide whether asymmetry is needed in one or both components of the mixture.
  • Fitting the mixture to financial returns supplies statistical support for the existence of two market regimes rather than a single central tendency.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same construction could be used to test for regime shifts in other economic series that display apparent bimodality.
  • If the two components are interpreted as distinct economic states, the model supplies a parametric way to assign observations to bull or bear periods.
  • Stability of the mixture parameters over long samples would justify treating the bear-bull split as a time-invariant feature of the market.

Load-bearing premise

The observed data, including S&P 500 returns, are generated from or closely approximated by a finite mixture of skewed-t components whose parameters remain stable over the sample period.

What would settle it

A Monte Carlo experiment matching the sample size and moments of the S&P 500 series in which the likelihood ratio test for skewness rejects at rates far from the nominal level, or in which the fitted two-component skewed-t mixture collapses to a single mode.

Figures

Figures reproduced from arXiv: 2606.20406 by Flavio Santi, Marco Bee.

Figure 1
Figure 1. Figure 1: Absolute median-bias and MAD-RMSE in setups 2, 5 an [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: QQ-plots of the simulated null distribution of the [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The S&P 500 distortion with the estimated skewed- [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The S&P 500 distortion with the estimated g-and-h m [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Shiller’s distortion (histogram), estimated pos [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The stamps thickness data with the estimated skewe [PITH_FULL_IMAGE:figures/full_fig_p023_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The stamp thickness data with the estimated skewed [PITH_FULL_IMAGE:figures/full_fig_p024_7.png] view at source ↗
read the original abstract

We propose a mixture of location-scale skewed-$t$ distributions to fit bimodal, skewed and heavy-tailed data. In particular, the mixture is based on the skewed-$t$ distribution by Fern\'andez and Steel (1998), so that the model-building procedure can be easily extended to mixtures of other symmetric distributions. After studying the properties of the mixture, we develop a maximum likelihood estimation approach via the EM algorithm and a likelihood ratio test of the null hypothesis of no skewness in any given component. A simulation-based comparison to a recently proposed mixture of g-and-h distributions suggests that the performance of the proposed model is excellent, in terms of both estimation precision in well-specified setups and modeling capability in mis-specified frameworks. Fitting the model to the Standard & Poor's 500 distortion allows us to confirm the bimodality of its distribution, with the implication that the US stock market has historically been in bearish or bullish conditions, rather than near its fundamental value.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes a finite mixture of location-scale skewed-t distributions (based on the Fernández-Steel 1998 skewed-t) for bimodal, skewed, and heavy-tailed data. It derives properties of the mixture, develops an EM algorithm for maximum likelihood estimation, and introduces a likelihood ratio test for the null of no skewness in any component. Simulations compare the model favorably to a g-and-h mixture in both well-specified and misspecified settings. An application to S&P 500 returns is used to confirm bimodality and interpret the data as arising from distinct bearish/bullish regimes rather than a single central value.

Significance. If the central claims hold, the work supplies a practical, extensible mixture framework that leverages an established skewed-t family and standard EM machinery. The simulation evidence of good estimation precision and robustness to misspecification would strengthen the case for the model as a flexible alternative for such data; the empirical illustration would then provide a concrete demonstration of its ability to recover economically interpretable structure in financial returns.

major comments (2)
  1. [empirical application] Empirical application: the conclusion that S&P 500 returns exhibit bimodality corresponding to stable bearish/bullish regimes rests on the assumption that the mixture parameters (mixing proportions, component location, scale, skewness, and degrees of freedom) remain constant over the full historical sample. No structural-break diagnostics, rolling-window fits, or checks for time-series dependence are reported, yet financial returns routinely violate stationarity; this directly affects the validity of the fitted model and the LRT for skewness.
  2. [likelihood ratio test] Likelihood ratio test section: the LRT for the null of no skewness in any component is presented without finite-sample size or power results under the actual sample size, possible serial dependence, or the mixture setting of the S&P 500 data. Because the test is invoked to support the skewness assessment in the application, its operating characteristics under these conditions are load-bearing for the reported conclusions.
minor comments (1)
  1. [model definition] Notation for the component-specific parameters (location, scale, skewness, df) should be introduced once with a clear table or list to avoid repeated redefinition across sections.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the referee's thoughtful comments. We provide point-by-point responses to the major comments below, indicating where revisions will be made.

read point-by-point responses

  1. Referee: Empirical application: the conclusion that S&P 500 returns exhibit bimodality corresponding to stable bearish/bullish regimes rests on the assumption that the mixture parameters (mixing proportions, component location, scale, skewness, and degrees of freedom) remain constant over the full historical sample. No structural-break diagnostics, rolling-window fits, or checks for time-series dependence are reported, yet financial returns routinely violate stationarity; this directly affects the validity of the fitted model and the LRT for skewness.

    Authors: The model is formulated for i.i.d. data from a mixture distribution, which is appropriate for modeling the unconditional distribution of returns. The application to S&P 500 is an illustration of how the model can identify bimodality, with the two components interpretable as bear and bull regimes. We recognize that assuming constant parameters over the entire period is a simplification and that financial returns may exhibit non-stationarity. In the revision, we will explicitly state the i.i.d. assumption and note that extensions to time-varying mixtures or regime-switching models could address time dependence, while emphasizing that the current work focuses on the cross-sectional mixture properties. revision: partial

  2. Referee: Likelihood ratio test section: the LRT for the null of no skewness in any component is presented without finite-sample size or power results under the actual sample size, possible serial dependence, or the mixture setting of the S&P 500 data. Because the test is invoked to support the skewness assessment in the application, its operating characteristics under these conditions are load-bearing for the reported conclusions.

    Authors: We derive the LRT under the mixture model framework with the null hypothesis that all components are symmetric (skewness parameter zero). The test statistic's asymptotic distribution accounts for the boundary problem. While the paper includes simulation studies for parameter estimation accuracy, we have not reported dedicated finite-sample simulations for the LRT's size and power specifically under serial correlation or for the S&P 500 sample characteristics. We will revise the manuscript to include a clearer statement on the asymptotic nature of the test and its potential limitations in the presence of dependence, but we maintain that the test provides useful evidence in the application under the model assumptions. revision: partial

Circularity Check

0 steps flagged

No circularity: model, estimation, and conclusions are independent of self-defined quantities or self-citation chains.

full rationale

The paper defines a mixture of Fernández-Steel (1998) skewed-t components, derives EM-based MLE and an LRT for skewness, runs new simulations comparing to g-and-h mixtures, and fits the model to S&P 500 returns to assess bimodality. None of these steps reduce by the paper's own equations to a fitted parameter defined in terms of itself, nor does any central claim rest on a load-bearing self-citation whose content is unverified. The 1998 reference is external; the simulation design and data application are new and falsifiable against external benchmarks. The derivation chain is therefore self-contained.

Axiom & Free-Parameter Ledger

3 free parameters · 2 axioms · 0 invented entities

Only the abstract is available; therefore the ledger is populated from information explicitly stated or necessarily implied by the modeling approach described. No invented entities are introduced. Free parameters are the usual mixture parameters whose values are estimated from data.

free parameters (3)
  • mixing proportions
    Weights for each skewed-t component; fitted by EM to match observed bimodality.
  • component-specific location, scale, skewness, and degrees of freedom
    Parameters of each skewed-t component; estimated via maximum likelihood.
  • number of mixture components
    Implicitly chosen as two to capture the reported bimodality; not derived from data.
axioms (2)
  • domain assumption The observed data are i.i.d. draws from a finite mixture of skewed-t distributions.
    Required for the likelihood and EM procedure to be correctly specified; stated implicitly by the model proposal.
  • standard math The EM algorithm converges to a local maximum of the observed-data likelihood.
    Standard assumption for EM-based MLE; invoked by the estimation approach.

pith-pipeline@v0.9.1-grok · 5688 in / 1646 out tokens · 57757 ms · 2026-06-26T16:06:07.893414+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

29 extracted references · 1 linked inside Pith

  1. [1]

    M., and Rodr \' guez-Casal, A

    Ameijeiras-Alonso, J., Crujeiras, R. M., and Rodr \' guez-Casal, A. (2019). Mode testing, critical bandwidth and excess mass. TEST , 28(3):900--919

    2019

  2. [2]

    Azzalini, A. (2005). The skew-normal distribution and related multivariate families. Scandinavian Journal of Statistics , 32(2):159--188

    2005

  3. [3]

    Bee, M., Hambuckers, J., and Trapin, L. (2019). Estimating V alue-at- R isk for the g-and-h distribution: an indirect inference approach. Quantitative Finance , 19(8):1255--1266

    2019

  4. [4]

    Bee, M., Hambuckers, J., and Trapin, L. (2021). Estimating large losses in insurance analytics and operational risk using the g-and-h distribution. Quantitative Finance , 21(7):1207--1221

    2021

  5. [5]

    and Faires, J

    Burden, R. and Faires, J. (2010). Numerical Analysis . Brooks/Cole, ninth edition

    2010

  6. [6]

    and Steel, M

    Fern\'andez, C. and Steel, M. F. J. (1998). On B ayesian modeling of fat tails and skewness. Journal of the American Statistical Association , 93(441):359--371

    1998

  7. [7]

    Flury, B. (1997). A First Course in Multivariate Statistics . Springer

    1997

  8. [8]

    Garcia, R., Renault, E., and Veredas, D. (2011). Estimation of stable distributions by indirect inference. Journal of Econometrics , 161(2):325 -- 337

    2011

  9. [9]

    Hoaglin, D. C. (1985). Summarizing Shape Numerically: The g-and-h Distributions , chapter 11, pages 461--513. Wiley

    1985

  10. [10]

    Izenman, A. J. and Sommer, C. J. (1988). Philatelic mixtures and multimodal densities. Journal of the American Statistical Association , 83(404):941--953

    1988

  11. [11]

    Kim, H.-J. (2008). Moments of truncated S tudent-t distribution. Journal of the Korean Statistical Society , 37(1):81--87

    2008

  12. [12]

    L., Little, R

    Lange, K. L., Little, R. J. A., and Taylor, J. M. G. (1989). Robust statistical modeling using the t distribution. Journal of the American Statistical Association , 84(408):881--896

    1989

  13. [13]

    and McLachlan, G

    Lee, S. and McLachlan, G. (2013). On mixtures of skew normal and skew t distributions . Advances in Data Analysis and Classification , 7:241--266

    2013

  14. [14]

    and McLachlan, G

    Lee, S. and McLachlan, G. (2014). Finite mixtures of multivariate skew t-distributions: some recent and new results. Statistics and Computing , 24:181--202

    2014

  15. [15]

    Lehmann, E. L. and Casella, G. (1998). Theory of Point Estimation . Springer, second edition

    1998

  16. [16]

    Lin, T.-I. (2010). Robust mixture modeling using multivariate skew t distributions. Statistics and Computing , 20:343--356

    2010

  17. [17]

    Lin, T.-I., Lee, J., and Hsieh, W. (2007). Robust mixture modeling using the skew t distribution. Statistics and Computing , 17:81--92

    2007

  18. [18]

    and Krishnan, T

    McLachlan, G. and Krishnan, T. (2008). The EM Algorithm and Extensions . Wiley, second edition

    2008

  19. [19]

    and Peel, D

    McLachlan, G. and Peel, D. (2000). Finite Mixture Models . Wiley

    2000

  20. [20]

    Pearson, K. (1894). Contribution to the mathematical theory of evolution. Phil. Trans. Roy. Soc. A , 185:71--110

  21. [21]

    Robertson, C. A. and Fryer, J. G. (1969). Some descriptive properties of normal mixtures. Scandinavian Actuarial Journal , 1969(3-4):137--146

    1969

  22. [22]

    and Westerhoff, F

    Schmitt, N. and Westerhoff, F. (2017). On the bimodality of the distribution of the S&P 500's distortion: E mpirical evidence and theoretical explanations. Journal of Economic Dynamics and Control , 80:34--53

    2017

  23. [23]

    Shiller, R. J. (2015). Irrational Exuberance: Revised and Expanded Third Edition . Princeton University Press

    2015

  24. [24]

    Titterington, D., Smith, A., and Makov, U. (1985). Statistical Analysis of Finite Mixture Distributions . Wiley

    1985

  25. [25]

    van Zyl, J. M. (2015). The efficiency of the likelihood ratio to choose between a t-distribution and a normal distribution. arXiv preprint arXiv:1505.01979

    Pith/arXiv arXiv 2015

  26. [26]

    and Ripley, B

    Venables, W. and Ripley, B. (2002). Modern Applied Statistics with S . Springer, fourth edition

    2002

  27. [27]

    and Walker, S

    Villa, C. and Walker, S. G. (2014). Objective prior for the number of degrees of freedom of a t distribution. Bayesian Analysis , 9(1):197--220

    2014

  28. [28]

    Wu, C. F. J. (1983). On the convergence properties of the EM algorithm. Annals of Statistics , 11(1):95--103

    1983

  29. [29]

    R., Rui, H., and Chervoneva, I

    Zhan, T., Misung, Y., Peck, A. R., Rui, H., and Chervoneva, I. (2025). Estimation and model selection for finite mixtures of T ukey's g-&-h distributions. Statistics and Computing , 35:67

    2025