Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Observable Matrix Dynamics of Stocks

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that the equity market's correlation geometry never relaxes: across three crisis decades the distance-matrix spectrum stays in the un-relaxed, pre-learning regime, collapsing in dimension at crises without ever forming a…

desk verdict A careful, reproducible empirical toolkit for reading equity correlation geometry and rank dynamics, but the central 'market never learns' claim needs an independent benchmark before it should be taken as established. read the letter →

arxiv 2607.19005 v2 pith:5G6NWWWW submitted 2026-07-21 q-fin.ST cs.CEq-fin.GNq-fin.PM

Igor Halperin This is my paper · ORCID
classification q-fin.STcs.CEq-fin.GNq-fin.PM
keywords ObservableMatrixDynamicsdistance-matrixspectrumrandomtheoryparticipationratioentropyproductiontransfervolatilityclusteringmarketcrises
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

This paper argues that the large-cap US equity market, watched through a rolling distance-matrix trajectory, is a permanently non-equilibrium object: its correlation geometry collapses at the 2008 and 2020 crises and never settles onto a low-dimensional manifold, the state a learning system occupies before learning rather than after. The evidence is the spectrum of the arccos distance matrix, which keeps a smooth power-law decay with exponent near 0.7 and no shoulder or quasi-multiplets across three crisis decades. The same fixed-size matrices also build two market-neutral Markov chains on daily return and volatility rankings: the return chain mixes in about a week and stays near time-reversible, while the volatility chain mixes in a month or more and carries a weak, episodic arrow of time that flares at market stress. If the paper is right, crises are at once a collapse of effective dimension and a burst of rank migration, and the reason an endogenous crisis like 2008 was forecastable while an exogenous shock like 2020 was not becomes a readout of the same geometry.

What carries the argument

The central object is the trajectory of the fixed-size arccos distance matrix $M_{ij}(t)=\arccos C_{ij}(t)$, built from rolling return correlations, together with two rank-ordered transition matrices $P^R(t)$ and $P^V(t)$ for daily return and volatility rankings. The distance-matrix spectrum is read through the Perron eigenvalue, the delocalised power-law slope $\beta$ from BBS theory, and the participation ratio of the correlation matrix, while the ranking chains are read through their second-eigenvalue mixing times, entropy production against a reversible null, and transfer entropy. The machinery's work is to compactify a high-dimensional evolving cross section into fixed-size objects whose spectral and geometric changes mark crises and distinguish co-movement from dispersion.

What would settle it

Recompute the distance-matrix spectrum on a much larger universe, with thousands of names, using a lookback long enough that $T\gg N$ and a noise-cleaned correlation estimator; if a shoulder appears near $K\approx\sqrt{N}$, quasi-multiplets form, or $\beta$ rises above one, the market would be relaxing to a manifold and the never-learns claim would be false.

Watch

Extended reading notes

Core claim

The central claim is that the market never learns its own correlation structure. Read against distance-matrix spectra from learning systems at initialisation, the market's spectrum stays in the un-relaxed regime: a smooth power-law decay with delocalised exponent $\beta\approx 0.7$, no quasi-multiplets, no shoulder at $K\approx\sqrt{N}$, and no relaxation to an equilibrium $M$-matrix. Supporting observations come from three fixed-size observables: the arccos distance matrix $M_{ij}(t)=\arccos C_{ij}(t)$ shows the participation ratio collapsing from roughly nine effective factors to four at Covid and from eight to five at 2008, while the 2001 bust runs the other way as a dispersed, decorrelated unwind; removing the market factor leaves a coherent sector rotation led by utilities; and the return and volatility ranking chains show fast, near-reversible performance dynamics beside slow, stress-flaring volatility dynamics. The forecastability result follows from the same machinery: fragility signals built only from trailing returns predict the endogenously built-up 2008 crisis with a per-period area-under-the-ROC-curve near 0.72 at the 63-day horizon, but not the exogenous 2020 shock or the 2001 unwind.

Load-bearing premise

The claim that the market never relaxes rests on treating the smooth, shoulder-free distance-matrix spectrum as a genuine property of the market rather than an artifact of the finite sample, and the paper itself lists the short lookback, the small universe, and the few-factor-plus-noise cross section as effects that could wash out relaxation in a system that actually does relax.

Editorial extensions

If this is right

  • If the market never relaxes, covariance and risk models that assume a stationary long-run correlation structure are misspecified; the correlation geometry should be treated as a non-equilibrium trajectory.
  • The spectral signature separates crisis types: correlated crashes such as 2008 and 2020 collapse the effective factor count and raise the market-factor share, while the 2001 dot-com period is a dispersed unwind with the opposite signature.
  • Fragility signals that use only trailing data can give months of early warning for endogenously building crises but not for exogenous shocks, so early-warning skill is regime-dependent rather than universally present or absent.
  • Once the market factor is removed, the sector geometry follows its own timetable, as the post-Covid sector rotation continued for a year or more past the March 2020 crash.
  • The volatility ranking is five to six times more persistent than the return ranking, and its weak arrow of time flares at sustained directional stress, matching the known volatility-clustering asymmetry.

Reading between the lines

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

  • If the claim holds, the same spectrum can serve as a generic test for whether any complex system has a learnable low-dimensional structure: a distance-matrix spectrum that never develops a shoulder or quasi-multiplets would mark a system with no relaxed geometry.
  • The forecastability contrast between 2008 and 2020 suggests a crisis taxonomy by origin that could guide when to trust fragility indicators: endogenous build-ups leave a spectral trail, exogenous shocks do not, and the two should not be pooled into one average skill score.
  • The volatility-chain arrow of time could be developed into a real-time regime indicator that ignores brief symmetric spikes and fires only on sustained directional reshuffling; testing it on intraday data or other asset classes would show whether the selectivity is general.
  • If markets are truly non-equilibrium, structural changes such as new trading mechanisms or regulatory constraints should transiently steepen the delocalised exponent toward or above one if any learning-like relaxation were possible; a null result would reinforce the never-learns claim.
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.

Referee Report

3 major / 4 minor

Summary. This paper applies the Observable Matrix Dynamics (OMD) framework to three S&P 500 crisis periods (2001, 2008, 2020), using a fixed-universe panel and three fixed-size matrix observables: an arccos distance matrix built from rolling return correlations, and two rank-ordered Markov transition matrices for performance and volatility. The distance-matrix analysis reports a collapse of the effective factor count at the 2008 and 2020 crises, a dispersed unwind in 2001, a coherent sector-eigenvector rotation after market-factor removal, and a causal trailing-window ROC that finds early-warning skill only for the 2008 crisis. The ranking-chain analysis reports persistence and mixing-time differences between the return and volatility rankings, near-reversibility of the return chain, a weak episodic arrow of time in the volatility chain, and sector-level transfer-entropy leadership, including a utilities-led performance channel and a financials-led risk channel. The paper interprets the persistent, shoulder-free distance-matrix spectrum with beta near 0.7 as evidence that the market never learns or relaxes to a stationary low-dimensional geometry.

Significance. If the central interpretation is accepted, the paper makes a substantive empirical contribution. It shows that trajectory-level diagnostics (Perron scale, participation ratio, eigenbasis rotation, commutator norm) add information beyond a single correlation matrix, separates correlated crashes from dispersion events, and provides crisis-specific sector composition and name-level attribution. The ranking-space analysis offers a new market-neutral view with entropy production and transfer entropy that reproduces volatility clustering and the Zumbach effect in a discrete representation. The empirical execution is careful in several respects: fixed-universe panels keep the cross section stable, the lookback windows are chosen to keep N < L, the ROC analysis uses trailing data only, an external stress-event calendar is used for validation, and the limitations are stated explicitly. The principal weakness is that the load-bearing 'never learns' interpretation depends on benchmarks calibrated in the author's own companion preprints, and the paper's own caveats permit a relaxed-but-noisy alternative.

major comments (3)
  1. [Section 3, Figure 2] The central claim that the market 'never learns' or 'never relaxes to an equilibrium M-matrix' depends on identifying the measured delocalised exponent beta near 0.7 with the 'un-relaxed, pre-learning regime' whose benchmark range beta in [0.65, 0.81] is taken from the OMD preprint [7], and on the I-BBS finite-N correction from preprint [8] to explain why the measured beta lies below the BBS threshold beta = 1. Because both sources are the author's own preprints, the calibration is not independently verifiable from this manuscript. The paper itself lists three effects that plausibly wash out the strict BBS fine structure - finite lookback with T of order N, small universe with sqrt(N) about 16 to 18, and a few-factor-plus-noise cross section - and calls the absence of a shoulder 'an open point.' Under these qualifications, the observed smooth spectrum is equally consistent with a relaxed-but-noisy system of small size. Please add an independent calibration: for example, simulate stationary factor-plus-noise returns with the same N and L and known latent dimension, and show that the resulting spectrum and beta distribution differ from the market's, or provide external replication of the I-BBS finite-N correction.
  2. [Section 4.1 and Section 4.8] The claim of no stationary limit sits in tension with the paper's own evidence that the observables revert after each crisis. Figure 4 shows the Covid trajectories plateau for approximately the lookback length and then 'steps back down', and Section 4.8 describes the MDS cloud re-expanding to 'its calm size' with 'the market back to normal.' A process that repeatedly returns to a pre-crisis baseline is mean-reverting and non-stationary, but that is not the same as never relaxing to a stationary geometry. The statement that the market 'never relaxes' needs an operational definition (for example, no convergence of the full spectral shape to a fixed attractor over decade-long windows, or beta never crossing a threshold) together with a control showing what observed trajectories would look like under a relaxational alternative. As written, the claim is stronger than the evidence.
  3. [Section 3, Eq. (2)] The power-law exponent beta is fit on at most sqrt(N) - 2, or about 14 to 16, eigenvalues per date for N between 244 and 309, and no confidence intervals or goodness-of-fit diagnostics are reported for the fitted beta. Given that the diagnostic distinction between the market's beta near 0.7 and the benchmark un-relaxed range near 0.81 is numerically narrow, the absence of error bars makes it difficult to assess whether the market's trajectory is genuinely distinguishable from any of the reference regimes. Please report standard errors or bootstrap intervals for beta at each date, and ideally for the Perron eigenvalue as well.
minor comments (4)
  1. [Section 2] Please state how missing CRSP returns, trading halts, and any suspended or delisted names are treated in the fixed-universe panel; the current description of a 'fixed universe' does not fully specify the handling of incomplete daily records.
  2. [Section 4.4] The random-subspace null for the projector drift uses the assumption that the two K-frames are independent and uniformly random; this should be stated explicitly, and at the smallest short-lookback universe with N = 94 a finite-N correction to the expectation E||V1^T V2||^2_F = K^2/N may be worth quantifying.
  3. [Section 5.1] The detailed-balance surrogate used for the entropy-production z-scores is described only as 'a reversible chain matched to the block's marginals and sample size'; the exact construction (for example, a Metropolis-filtered symmetrized transition matrix) should be specified so that the reported z-scores in Figure 17 are reproducible.
  4. [Section 4.7, Table 1] The per-period AUC values (0.46 to 0.72) are in-sample and based on only three crisis periods, so the conclusion of regime-dependent forecastability would be strengthened by bootstrap confidence intervals and by a statement of how sensitive the AUC is to the drawdown threshold and the 63-day horizon.

Circularity Check

2 steps flagged · score 4.0 of 10

The 'market never learns' claim rests on the author's own OMD/I-BBS calibrations, but the core empirical findings are independently anchored; moderate self-citation burden.

  1. self citation load bearing [Section 3, Figure 2 (also summarized in the Abstract)]
    "The un-relaxed, high-dimensional representation cloud of a learning system has a shared spectral signature at initialisation, before any training: a smooth power-law decay with β between 0.65 and 0.81, no quasi-multiplets, and no shoulder at K=√N. Grokking, MNIST, and sparse parity all begin with their untrained exponent β in this narrow range (panel c), so the value is a property of the un-relaxed cloud rather than of any one task..."

    The 'un-relaxed, pre-learning regime' is not independently derived in this paper; its defining interval β∈[0.65,0.81] and the interpretation that this interval means 'before training' come from the author's own OMD preprint [7]. The market's fitted β≈0.7 falls inside this interval, so the conclusion 'the market never learns' is an application of a self-supplied label from a self-citation, not a derivation from market data alone. The paper's own caveat that finite T~N, small N, and a few-factor-plus-noise cross section could wash out the fine structure makes the smooth spectrum compatible with a relaxed-but-noisy system, so the data do not force the 'never learns' reading without [7]'s calibration.

  2. self citation load bearing [Section 3, finite-N correction paragraph]
    "The delocalised reading is asymptotic in the sample size, valid in the BBS regime N≫d², and at finite N the measured exponent falls below its large-N asymptotic d/(d−1) by an I-BBS finite-N correction ∆β(N,d) that grows as N shrinks toward d² [8]. At our fixed N of a few hundred names, β<1 is therefore consistent either with a finite latent dimension once that correction is restored, or with an embedding whose points depart from a uniform sample on the latent hyper-sphere."

    The finite-N correction ∆β(N,d) is imported from the author's own I-BBS preprint [8] and is the only quantitative bridge that keeps β≈0.7 from implying a negative embedding dimension. The paper then uses this same β value as evidence of the 'un-relaxed' regime. If ∆β is inaccurate or inapplicable at N≈94–309, the interpretation of β as an un-relaxed initialisation signature loses its quantitative support. This is a load-bearing self-citation in the spectral-interpretation chain, though it does not itself manufacture the market data.

full rationale

The strongest claim—that the S&P 500 'never learns' its correlation structure and 'never relaxes to an equilibrium M-matrix'—is not a mathematical derivation from the market data alone. It is an interpretive classification: the market's fitted delocalised exponent β≈0.7 is compared with the 'un-relaxed initialisation' band β∈[0.65,0.81] taken from the author's own OMD preprint [7], and with the finite-N correction ∆β from the author's I-BBS preprint [8]. These are load-bearing self-citations: if the ML calibration in [7] or the correction in [8] is not independently verified, the 'never learns' reading loses its quantitative anchor. The paper itself concedes that finite T~N, small N with √N≈16–18, and a few-factor-plus-noise cross section could wash out the BBS fine structure, and states that the absence of that structure is 'an open point'; it also says β<1 is 'consistent either with a finite latent dimension once that correction is restored, or with an embedding whose points depart from a uniform sample'. That honest limitation means the central distinction between a never-learning market and a relaxed-but-noisy small-N system is not identified by the data. However, most of the paper's empirical findings are not circular: the spectral collapse, eigenvector rotation, ranking-chain persistence, entropy-production arrow (tested against a reversible null), and the sector leadership network are measured on market data and checked against independent anchors (external stress-event calendar, Zumbach effect, surrogate nulls), and the ROC caveat is disclosed as in-sample. No equation in the paper reduces to its inputs by construction; the circularity burden is limited to the benchmark calibration for the headline interpretation. Score 4.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The paper is empirical, so the ledger is dominated by analysis choices (lookback, binning, windows, hand-chosen anchors) rather than physical postulates. No invented entities are introduced. The principal content-dependent axiom is the transferability of the OMD 'un-relaxed regime' spectral signature from the author's own machine-learning experiments to equity correlation matrices; this anchors the central market-never-learns claim and is not independently calibrated. The Markov-chain and entropy-production foundations are standard.

free parameters (6)
  • lookback length L = 126, 256, and 504 trading days
    The spectral and ranking diagnostics, including the forecast AUC, are reported at three lookbacks; the 126-day window is emphasized because it sharpens onset and precursor resolution (Section 4.2, Table 1).
  • ranking rolling window tau_MC = 10 days for return chain, 21 days for volatility chain
    Used to build the daily rankings in Eq. (6); the entropy production peak tracks this window (Section 5.1, Figure 16a).
  • decile binning = 10 classes
    Coarse-grains the N-name ranking into a small Markov chain state space for the transition matrices (Section 5).
  • crisis onset dates = March 2001, September 2008, March 2020
    Hand-chosen anchors used to align crises and to define the rotation diagnostic reference (Section 4.4: 'The onset itself is a fixed, hand-chosen date for each period').
  • ROC labels and thresholds = 7 percent drawdown within 21 days, 10 percent within 63 days
    Forward labels for the early-warning AUC in Section 4.7; not optimized, but chosen by the author as the causal test definition.
  • entropy production block length = 63-day and quarterly windows
    Resolution at which the arrow-of-time z-scores are computed (Figure 17); shorter windows were not systematically swept.
assumptions (5)
  • domain assumption The BBS spectrum of a uniform sample on S^{d-1} gives beta = d/(d-1), and the I-BBS finite-N correction applies to the equity distance matrix.
    Used in Section 3 to interpret the measured beta near 0.7 as either a finite latent dimension or a finite-N artifact; the correction is from the author's own I-BBS preprint [8].
  • domain assumption Rankings of daily returns and volatilities are first-order Markov chains on decile classes.
    Section 5: 'We treat these rankings as Markov chains only as a first approximation'; higher-order memory and covariates are acknowledged but neglected at leading order.
  • ad hoc to paper The 'un-relaxed' spectral signature (beta in 0.65 to 0.81, no shoulder, no multiplets) is a benchmark property of un-trained representation clouds.
    Figure 2 claims this range is shared by grokking, MNIST, and sparse parity at initialization, from the author's own OMD experiments [7]; no independent calibration for equity correlation matrices is provided.
  • domain assumption Spectral deflation Eq. (3) isolates the market factor, and the residual is a meaningful sector-level correlation matrix.
    Standard practice, but the residual has rank N-1 with a zeroed direction; the arccos renormalization of a non-PSD residual changes the geometry (Section 4.3).
  • standard math Entropy production from Eq. (9) is a valid time-reversibility diagnostic when tested against a detailed-balance surrogate.
    KL divergence non-negativity and surrogate testing are standard (Section 5.1), and the surrogate is matched to the chain's marginals and sample size.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Observable Matrix Dynamics of Stocks." pith.science (2026). https://pith.science/paper/5G6NWWWW

@misc{pith2026260719005,
  author       = {Pith},
  title        = {Pith review of: Observable Matrix Dynamics of Stocks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5G6NWWWW}},
  note         = {Machine review of arXiv:2607.19005}
}
read the original abstract

The Observable Matrix Dynamics (OMD) approach monitors the time development of complex non-linear systems through the trajectory of a fixed-size distance matrix and its spectrum. We apply it to the S\&P 500 cross section over three crisis decades, the 2001 dot-com bust, the 2007--2008 financial crisis, and the 2020 Covid crash, with three fixed-size observables on a fixed universe. The arccos distance matrix of the rolling return correlations reads the correlation geometry: its effective dimension collapses at the 2008 and 2020 crises, while the 2001 bust is a dispersed unwind. Read against machine-learning distance matrices, its spectrum stays in the un-relaxed, pre-learning regime with no low-dimensional manifold, so the market never learns its correlation structure or relaxes to a stationary geometry. Subtracting the market factor exposes a coherent sector rotation, whose name-level attribution identifies which stocks drive each crisis and in what order. At a short lookback these signals resolve precursors and forecast the endogenous 2008 crisis, though not the exogenous 2020 shock. The other two observables model the daily return and volatility rankings as Markov chains on their ranking spaces. The return chain has persistent, defensive-led bellwethers and near-reversible dynamics. The volatility chain is far more persistent, led by the financial sector, and is the only one to carry a weak, episodic arrow of time, flaring at market stress and matching volatility clustering and the Zumbach effect. All three matrices show coherent changes during market crashes.

Figures

Figures reproduced from arXiv: 2607.19005 by the authors.

Figure 1
Figure 1. shows the spectrum itself, the eigenvalue magnitudes of M(t) ranked on a log-log scale at three phases of each crisis, one year before onset, the crisis peak, and one year after, for the raw distance matrix (top) and the market-removed residual (bottom). The single Perron eigenvalue Λ1 sits well above the rest at rank one, and the delocalised branch decays as the power law |λK| ∼ K−β over the shaded window K ∈ [2, √… view at source ↗
Figure 2
Figure 2. A learning system can relax its distance-matrix geometry, and the market does [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Three crises aligned on time-to-onset (years). Top: market-factor share [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (19 more)
Figure 4
Figure 4. Figure 4: Covid period, two-year lookback, N = 309. From top: mean correlation, market￾factor share and effective factor count of C(t), then the Perron eigenvalue and the delocalised β of the distance matrix M(t). The dashed line marks the onset and the shaded band the two-year …
Figure 5
Figure 5. Figure 5: Lookback comparison at 126, 256, and 504 days, all in the [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Raw versus market-factor-removed spectral dynamics, two-year lookback. Columns: [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Eigenvector rotation through each crisis, two-year lookback. Blue is the raw corre [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Eigenbasis reorganization measured by the commutator norm [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Sector composition of the rotating market-removed leading eigenvector, two-year [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: Name-level attribution of the rotating market-removed eigenvector, two-year look [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: Early-warning ROC for the 126-day signals against a forward market drawdown, [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: MDS embedding of the distance matrix M(t) through the three crises, 126-day lookback. Rows: the three periods. Columns: a calm baseline, the onset, the crisis peak, and one year after. Each name is a point, grey outside the crisis and, from the onset to three months p…
Figure 13
Figure 13. Figure 13: Decile rank transition matrices for the Covid universe, at [PITH_FULL_IMAGE:figures/full_fig_p019_13.png]
Figure 14
Figure 14. Figure 14: Spectra and mixing of the two ranking chains. (a) Moduli of the decile transition [PITH_FULL_IMAGE:figures/full_fig_p019_14.png]
Figure 15
Figure 15. Figure 15: Ranking-space dynamics through the crises, return and volatility. In both columns [PITH_FULL_IMAGE:figures/full_fig_p020_15.png]
Figure 16
Figure 16. Figure 16: Entropy production of the ranking-space Markov chains, return and volatility. (a) [PITH_FULL_IMAGE:figures/full_fig_p021_16.png]
Figure 17
Figure 17. Figure 17: Diagnostics over 1996–2024, resolved in time, against the market-crisis regimes [PITH_FULL_IMAGE:figures/full_fig_p024_17.png]
Figure 18
Figure 18. Figure 18: Entropy production along single-firm trajectories, volatility chain, Covid decade. [PITH_FULL_IMAGE:figures/full_fig_p025_18.png]
Figure 19
Figure 19. Figure 19: Directed transfer entropy between sectors, each represented by the tercile of its mean [PITH_FULL_IMAGE:figures/full_fig_p025_19.png]
Figure 20
Figure 20. Figure 20: Stock-level lead-lag network in ranking space, Covid decade. (a) Return ranking: the [PITH_FULL_IMAGE:figures/full_fig_p026_20.png]
Figure 21
Figure 21. Figure 21: Co-movement and dispersion, the two components of a crash. (a) Aligned on onset, [PITH_FULL_IMAGE:figures/full_fig_p027_21.png]
Figure 22
Figure 22. Figure 22: Co-dependence of the matrix views across a crisis, aligned on onset (crimson dashed [PITH_FULL_IMAGE:figures/full_fig_p029_22.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Are Three Matrices All You Need To Beat the Market? Observable Matrix Dynamics for Portfolio Optimization

    q-fin.PM 2026-07 conditional novelty 6.0 of 10

    A three-matrix, rank-based Markov-chain portfolio reports out-of-sample Sharpes of 1.06 and 1.32 vs the market's 0.78 and 1.14, net of costs, even though its return rank is itself 'close to unforecastable.'

Reference graph

Works this paper leans on

28 extracted references · 23 canonical work pages · cited by 1 Pith paper

  1. [7]

    Learning as Observable Matrix Dynamics: Diffusive Relaxations versus Phase Transitions

    I. Halperin.Learning as Observable Matrix Dynamics: Diffusive Relaxations versus Phase Transi- tions.arXiv:2606.29679, 2026

    work page Pith review arXiv 2026

  2. [8]

    I-BBS: Coordinate-Free Inference of Latent Sub-Manifolds Using Random Distance Matrix Theory

    I. Halperin.I-BBS: Coordinate-Free Inference of Latent Sub-Manifolds Using Random Distance Matrix Theory.arXiv:2606.29675, 2026

    work page Pith review arXiv 2026

  3. [1]

    V. A. Marˇ cenko and L. A. Pastur.Distribution of Eigenvalues for Some Sets of Random Matrices. Mathematics of the USSR-Sbornik1, 457–483 (1967)

    work page 1967

  4. [2]

    Laloux, P

    L. Laloux, P. Cizeau, J.-P. Bouchaud, and M. Potters.Noise Dressing of Financial Correlation Matrices. Physical Review Letters83, 1467–1470 (1999)

    work page 1999

  5. [3]

    Plerou, P

    V. Plerou, P. Gopikrishnan, B. Rosenow, L. A. N. Amaral, T. Guhr, and H. E. Stanley.Random Matrix Approach to Cross Correlations in Financial Data. Physical Review E65, 066126 (2002)

    work page 2002

  6. [4]

    Potters and J.-P

    M. Potters and J.-P. Bouchaud.A First Course in Random Matrix Theory: for Physicists, Engineers and Data Scientists.Cambridge University Press, Cambridge (2020)

    work page 2020

  7. [5]

    Onnela, A

    J.-P. Onnela, A. Chakraborti, K. Kaski, J. Kert´ esz, and A. Kanto.Dynamics of Market Correlations: Taxonomy and Portfolio Analysis. Physical Review E68, 056110 (2003)

    work page 2003

  8. [6]

    Kritzman, Y

    M. Kritzman, Y. Li, S. Page, and R. Rigobon.Principal Components as a Measure of Systemic Risk. Journal of Portfolio Management37(4), 112–126 (2011). 31

    work page 2011

Show all 28 references
  1. [9]

    Bogomolny, O

    E. Bogomolny, O. Bohigas, and C. Schmit.Spectral Properties of Distance Matrices. Journal of Physics A: Mathematical and General36, 3595–3616 (2003). arXiv:nlin/0301044

    2003 arXiv

  2. [10]

    Bogomolny, O

    E. Bogomolny, O. Bohigas, and C. Schmit.Distance Matrices and Isometric Embeddings. arXiv:0710.2063, 2007

    2007 arXiv

  3. [11]

    M´ ezard, G

    M. M´ ezard, G. Parisi, and A. Zee.Spectra of Euclidean Random Matrices. Nuclear Physics B559, 689–701 (1999)

    1999

  4. [12]

    Goetschy and S

    A. Goetschy and S. E. Skipetrov.Euclidean Random Matrices and Their Applications in Physics. arXiv:1303.2880, 2013

    2013 arXiv

  5. [13]

    Halperin.Frustrated Dynamics of Distance Matrices.arXiv:2605.05376, 2026

    I. Halperin.Frustrated Dynamics of Distance Matrices.arXiv:2605.05376, 2026

    2026 arXiv

  6. [14]

    Paninski.Estimation of Entropy and Mutual Information

    L. Paninski.Estimation of Entropy and Mutual Information. Neural Computation15, 1191–1253 (2003)

    2003

  7. [15]

    C. M. B¨ uth, K. Acharya, and M. Zanin.infomeasure: A Comprehensive Python Package for Infor- mation Theory Measures and Estimators. Scientific Reports15, 14053 (2025). arXiv:2505.14696

    2025 arXiv

  8. [16]

    Wollstadt, J

    P. Wollstadt, J. T. Lizier, R. Vicente, C. Finn, M. Mart´ ınez-Zarzuela, P. Mediano, L. Novelli, and M. Wibral.IDTxl: The Information Dynamics Toolkit xl. Journal of Open Source Software4(34), 1081 (2019)

    2019

  9. [17]

    J. C. Jackwerth and A. Slavutskaya.Relative Alpha.Working paper, University of Konstanz, SSRN 2439145 (2018)

    2018

  10. [18]

    Forsberg, D

    D. Forsberg, D. R. Gallagher, and G. J. Warren.Identifying Hedge Fund Skill by Using Peer Cohorts. Financial Analysts Journal77(2), 97–123 (2021)

    2021

  11. [19]

    E. R. Fernholz.Stochastic Portfolio Theory.Springer, New York (2002)

    2002

  12. [20]

    A. D. Banner, R. Fernholz, and I. Karatzas.Atlas Models of Equity Markets. Annals of Applied Probability15(4), 2296–2330 (2005)

    2005

  13. [21]

    Seifert.Entropy Production along a Stochastic Trajectory and an Integral Fluctuation Theorem

    U. Seifert.Entropy Production along a Stochastic Trajectory and an Integral Fluctuation Theorem. Physical Review Letters95, 040602 (2005)

    2005

  14. [22]

    Seifert.Stochastic Thermodynamics, Fluctuation Theorems and Molecular Machines

    U. Seifert.Stochastic Thermodynamics, Fluctuation Theorems and Molecular Machines. Reports on Progress in Physics75, 126001 (2012)

    2012

  15. [23]

    Georgiev.Stochastic Thermodynamics of Financial Markets I: Entropy Production and the Sec- ond Law.SSRN preprint 5454994, 2025

    G. Georgiev.Stochastic Thermodynamics of Financial Markets I: Entropy Production and the Sec- ond Law.SSRN preprint 5454994, 2025

    2025

  16. [24]

    Halperin.Information Bottleneck, Semantic Faithfulness and Entropy Production Measures for Analysis of Large Language Models as Information Engines.2026

    I. Halperin.Information Bottleneck, Semantic Faithfulness and Entropy Production Measures for Analysis of Large Language Models as Information Engines.2026

    2026

  17. [25]

    J. M. R. Parrondo, C. Van den Broeck, and R. Kawai.Entropy Production and the Arrow of Time. New Journal of Physics11, 073008 (2009)

    2009

  18. [26]

    Zumbach.Time Reversal Invariance in Finance

    G. Zumbach.Time Reversal Invariance in Finance. Quantitative Finance9(5), 505–515 (2009)

    2009

  19. [27]

    Schreiber.Measuring Information Transfer

    T. Schreiber.Measuring Information Transfer. Physical Review Letters85, 461–464 (2000)

    2000

  20. [28]

    Kraskov, H

    A. Kraskov, H. St¨ ogbauer, and P. Grassberger.Estimating Mutual Information. Physical Review E 69, 066138 (2004). 32

    2004

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.