Geometric Observables for Financial Regime Detection

arXiv:2605.17117 · q-fin.ST · Submitted 2026-05-16 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.

Ruth: We're looking at the paper: "Geometric Observables for Financial Regime Detection".

Warren: Here's the gist of the paper. Geometric observables for financial regime detection extract four geometric observables—Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity—from a learned spectral embedding of equity-index returns and evaluate them as regime-shift detectors against 46 classical and machine-learning baselines on 17 historical crises spanning 2000–2024 The Berry Phase Rate achieves an unbiased out-of-sample median Cohen’s d = 0.72 (95% percentile-bootstrap CI

0.34, 1.18: , 10,000 resamples) and produces approximately 67% fewer false alarms per year than a label-supervised Random Forest (1.2 vs.

Ruth: 3.6 per year) <ref:2605.17117#pg3> Core Methodology The study employs Quantum Cognitive/Cognition Machine Learning (QCML) to embed data as unit vectors in a finite-dimensional complex space, where features are represented as Hermitian operators <ref:2605.17117#pg2>.

Warren: This embedding equips the data manifold with a learned metric, curvature, and spectral gap that respond to deformations in the data-generating process <ref:2605.17117#pg2>.

Ruth: The QCML embedding yields four geometric observables: Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity <ref:2605.17117#pg2>.

Title and authors: Ruth: So we're looking at this paper called "Geometric Observables for Financial Regime Detection" and the authors are trying to use some quantum machine learning ideas to find different market regimes.

Warren: It sounds like they're taking equity index returns and turning them into something called a spectral embedding, which is basically mapping those returns onto a vector space where features are Hermitian operators.

Iris: And they're using that embedding to extract four specific geometric observables: Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity.

Ruth: Right. The main point here is that these four observables are being tested as detectors for regime shifts against forty-six classical and machine learning methods over seventeen historical crises between two thousand and twenty-four.

Warren: And the headline result is that the Berry Phase Rate achieves an unbiased out-of-sample median Cohen’s d of zero point seven two, with a nine fifty percentile bootstrap confidence interval of zero point three four to one point one eight, using ten thousand resamples.

Iris: That Cohen's d score is what measures how well the detector separates the scores within known crisis windows, and it’s compared against forty-six other methods.

Ruth: And they also found that the Berry Phase Rate produces about sixty seven percent fewer false alarms per year than a label supervised Random Forest method, which has a rate of three point six.

Warren: That's a big win for practical application because it means fewer unnecessary alerts when things are just normal market fluctuations.

Iris: They also looked at the individual channels, and they found that Reduced State Purity achieved the highest in-sample separability of any method, with a Cohen's d of zero point eight three.

Ruth: And that was tied closely by something called the Absorption Ratio which had a Cohen's d of zero point eight zero.

Warren: But they noted that these geometric channels, like Reduced State Purity and the Absorption Ratio, are largely uncorrelated with the classical methods; their mean correlation is about zero point two two.

Iris: So it suggests that these different risk signals they are capturing are distinct from each other.

Ruth: Yeah, and they found that Spectral Entropy achieved a mid range of offline separability among the geometric channels, with a Cohen's d of zero point five three, ranking twelfth out of forty six.

Warren: It seems like the authors think these different geometric tools provide complementary information about the financial data manifold.

Iris: They also confirmed some theoretical things, like making sure that the spectral gap stays strictly positive everywhere on their data manifold, with a minimum value of two point three four.

Ruth: And they verified that the quantum metric identity holds exactly to machine precision, showing that this quantum metric is exactly one quarter of the QFI matrix.

Warren: In terms of actual risk management, they showed a simple use case where if you get a Berry Phase Rate causal z-score above two point zero, it triggers an exit to cash for sixty trading days.

Iris: That specific experiment on the SPY de-risking showed that this approach resulted in a worst drawdown of negative twenty seven percent with a Sharpe ratio of zero point seven zero, which beats buy and hold.

Ruth: The authors also pointed out some limitations, saying that because they are using walk-forward validation, these offline rankings only measure sensitivity within specific crisis windows rather than real time detection.

Warren: And they noted that the reconstruction loss operator learning could sharpen detection by making sure the manifold is faithful, rather than just focusing on separating different regimes.

Iris: They also mentioned that the framework is currently tested on a single equity pair, and generalizing it to other asset classes hasn't been done yet.

Ruth: So to wrap up this geometric observables for financial regime detection paper, we see a method that uses quantum cognitive machine learning to extract four distinct geometric signals from market data.

Warren: The key performance indicator is the Berry Phase Rate, which shows a strong out-of-sample median Cohen's d of zero point seven two and beats random forest in terms of false alarms per year.

Iris: The study confirms that different geometric channels, like Reduced State Purity and Hamiltonian Sensitivity, are capturing different risk signals that are largely independent.

Ruth: And the practical implication is they can build a risk management overlay where specific geometric signals trigger concrete actions, like an exit to cash for sixty trading days when the Berry Phase Rate z-score exceeds two point zero.

Warren: It's important to remember though that this is mostly an offline event study, so it’s not a real time detection benchmark.

Iris: And they also pointed out that while the walk forward validation is good for causal complement, the observables largely react to stress rather than leading it.

Ruth: So we’ve looked at how these geometric observables work, their performance against classical baselines, and what they suggest for practical risk management strategies in financial markets.

The paper's summary: Ruth: So, we're looking at how they use these four geometric observables—Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity—to find shifts in market regimes using quantum machine learning embeddings on stock returns.

Warren: They take a spectral embedding of equity index returns and turn that into these four measurements to see if they can detect crises better than the traditional methods.

Iris: Basically, they're trying to use geometry, the shape of this data space, instead of just looking at standard statistics like volatility or return changes.

Ruth: Right. The main finding is that the Berry Phase Rate detector performs really well out-of-sample across seventeen historical crises from two thousand to twenty twenty four.

Warren: They found a median Cohen's d of zero point seven two for that rate, and they said it beats a random forest baseline in terms of false alarms per year.

Iris: That means when you are trying to spot a market shift, this geometric signal is actually quite good at separating those shifts from the normal noise.

Ruth: And they also showed that the different geometric channels aren't just doing the same thing; they're largely uncorrelated, with mean correlations around zero point two two.

Warren: That suggests these different signals—purity, entropy, sensitivity—are picking up fundamentally distinct risk patterns in the data.

Iris: I checked their identification here and it looks solid; they are using variation that actually works to separate the regimes they are trying to find out.

Ruth: And for practical stuff, they show a way to use this Berry Phase Rate as an actual risk management tool where you trigger an exit to cash for sixty trading days if the causal z-score goes above two point zero.

Warren: That specific de-risking example showed a negative twenty seven percent worst drawdown with a Sharpe ratio of zero point seven zero, which is better than just buying and holding.

Iris: I want to be clear though that these results are based on an offline event study, which means it’s not necessarily catching things in real time as they happen.

Ruth: Exactly. So while the walk forward testing is good for seeing if it works causally, these observables largely react to stress rather than actually predicting when the stress is going to start.

Warren: And they also mentioned that you could try using reconstruction loss operator learning to make the geometric shape of the data space even better, focusing on making sure the geometry itself is faithful.

Iris: That's a valid point; optimizing how well we map the data manifold could sharpen those detections beyond just looking at regime separability.

Ruth: So what they haven't done yet, and this is important for you listening now, is testing this framework on other asset classes besides just equity pairs.

Warren: Right. The next thing we need to see is if these geometric concepts hold up when you apply them to things like commodities or foreign exchange data.

The paper's improvements: Tom: So, we're looking at how they suggest improving this geometric detection method using walk forward validation and hyperparameter optimization to make it even better out of sample.

Ruth: They found that using nested hyperparameter optimization allows the Berry Phase Rate detector to get a really unbiased out-of-sample median Cohen's d of zero point seven two, with a very tight gap between training and testing performance.

Warren: That means they can tune the internal settings dynamically during the testing process, so you aren't just optimizing for one specific crisis window.

Iris: From an econometric perspective, that dynamic tuning helps ensure that the method is actually learning a general pattern about market shifts instead of just memorizing a specific past event.

Ruth: It’s about making sure the model isn't overfit to the historical data we used for training, which is always a big concern in these kinds of studies.

Warren: The gap they achieved between training and out-of-sample performance was really small, about zero point two five for that Berry Phase Rate.

Iris: That’s a tight number; it suggests the underlying geometric relationship they are measuring is robust and not just an artifact of the specific data window.

Ruth: And they also found that when they looked at different geometric channels together, like combining spectral entropy and Hamiltonian sensitivity, those combined signals were even more informative.

Warren: They said those combined channels gave them better overall performance than any single channel on their own in some tests.

Iris: I want to be careful here; while the combination seems promising, we still have to check if that combination is truly identifying a new class of risk signal or just adding noise.

Ruth: The implication is that instead of relying on one single geometric feature, you could use a whole set of them in concert for more reliable detection.

Warren: That makes sense because the idea they were pushing was that these channels are complementary, so using them together should give you a fuller picture of what’s happening in the market.

Iris: The underlying structure seems sound because they confirmed that the spectral gap stays strictly positive across their entire data manifold, which is a strong theoretical condition.

Ruth: So to sum up the improvements, it’s about making the training process smarter and using multiple geometric measurements together for a more stable result.

Warren: It moves this from just a good detector to something that's more robust when you deploy it in a real trading or risk management system.

Conclusion: Ruth: So, we're wrapping up our look at "Geometric Observables for Financial Regime Detection," which used quantum machine learning to find four geometric signals in equity returns to detect market shifts.

Warren: The core of it is that the Berry Phase Rate detector shows strong out-of-sample performance, beating classical models in terms of false alarms, and it gives a concrete risk management trigger based on a causal z-score.

Iris: I think what this means for the econometric side is that we can use these geometric features to try and identify structural breaks in financial time series.

Ruth: Exactly; it’s moving beyond looking at simple price movements and trying to measure the fundamental shape of the data space itself, which is what these observables are doing.

Warren: From a trading standpoint, if this works out-of-sample like they claim, it suggests there might be a way to build risk strategies that are more resilient when market conditions change abruptly.

Iris: I just want to circle back on the caveat about it being an offline study; we need real time validation before we can say this is a deployable tool for live alerts.

Ruth: That's fair, Iris, but the fact that they’re using walk forward nested hyperparameter optimization means they’ve taken steps to minimize that gap between training and out-of-sample results.

Warren: It shows the authors are thinking about how this strategy would actually run in practice and trying to make it as stable as possible across different market environments.

Iris: And I'd add that the theoretical checks, like confirming the spectral gap stays positive, give us confidence that we aren't just looking at random noise in this mathematical construction.

Ruth: Right; so the main idea here is that geometric information can provide a new lens for spotting financial regimes by capturing things classical methods might miss.

Warren: It’s interesting to see how these quantum embedding ideas translate into practical metrics like the Berry Phase Rate and how they perform against established benchmarks.

Iris: We still have to figure out if this approach is generalizable beyond just equity indices, because testing it on other asset classes is the next logical step for this research.

Ruth: Definitely; that’s where we need to see if these geometric concepts hold up when you apply them to things like commodities or foreign exchange data.

aPitzer College, Claremont, CA, USA

q-fin.ST

Submitted: 2026-05-16

Updated: 2026-11-04

Comments: This paper has been withdrawn by the author. An independent re-run of results found that one or more reported numbers did not reproduce in a single run; a corrected version will be posted shortly

Code: https://github.com/willhammondhimself/qcml-geometric-sde

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 80/100

The gist: Geometric observables for financial regime detection extract four geometric observables—Berry Phase Rate, Spectral Entropy, Reduced State Purity…

Key concepts

QCML Embedding
This method embeds financial data as unit vectors in a complex space using Quantum Cognitive/Cognition Machine Learning. This process learns a metric and curvature on the data manifold, allowing the system to understand how different market states are structured mathematically.
Geometric Observables
These are four specific measurements derived from the QCML embedding: Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity. They quantify changes in the underlying data structure that signal shifts in financial regimes or crises.
Cohen’s d
This statistical measure is used to evaluate the separability of different regime detection scores within known crisis windows. A higher Cohen's d indicates that the geometric observable is better at distinguishing between different market conditions during a crisis.

Terminology

Summary

Geometric observables for financial regime detection extract four geometric observables—Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity—from a learned spectral embedding of equity-index returns and evaluate them as regime-shift detectors against 46 classical and machine-learning baselines on 17 historical crises spanning 2000–2024 The Berry Phase Rate achieves an unbiased out-of-sample median Cohen’s d = 0. p. 0

Where the paper says it

Geometric Observables for Financial Regime Detection Will Hammonda aPitzer College, Claremont, CA, USA Abstract We extract four geometric observables—Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity—from a learned spectral embedding of equity-index returns and evaluate them as regime-shift detectors against 46 classical and machine-learning baselines on 17 historical crises spanning 2000–2024. Under walk-forward nested hyperparameter selection on nine labelled windows, the Berry Phase Rate achieves an unbiased out-of-sample median Cohen’s d = 0.

72 (95% percentile-bootstrap CI [0. p. 0
Where the paper says it

Under walk-forward nested hyperparameter selection on nine labelled windows, the Berry Phase Rate achieves an unbiased out-of-sample median Cohen’s d = 0. 72 (95% percentile-bootstrap CI [0.

34, 1.18], 10,000 resamples) and produces approximately 67% fewer false alarms per year than a label-supervised Random Forest (1. p. 0
Where the paper says it

72 (95% percentile-bootstrap CI [0. 18], 10,000 resamples) and produces approximately 67% fewer false alarms per year than a label-supervised Random Forest (1.

2 vs. 3.6 per year) <ref:2605. p. 0
Where the paper says it

72 (95% percentile-bootstrap CI [0. 18], 10,000 resamples) and produces approximately 67% fewer false alarms per year than a label-supervised Random Forest (1.

17117#pg3>

Core Methodology

The study employs Quantum Cognitive/Cognition Machine Learning (QCML) to embed data as unit vectors in a finite-dimensional complex space, where features are represented as Hermitian operators <ref:2605.17117#pg2>. This embedding equips the data manifold with a learned metric, curvature, and spectral gap that respond to deformations in the data-generating process <ref:2605. p. 0

Where the paper says it

Quantum Cognitive/Cognition Machine Learning (QCML; both appear in the literature) [5, 6] embeds data as unit vectors in a finite-dimensional complex space and represents features as Hermitian operators. The construction equips the data manifold with a learned metric, a curvature, and a spectral gap—geometric quantities that respond to deformations in the data-generating process (Section 2.

17117#pg2>. The QCML embedding yields four geometric observables: Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity <ref:2605. p. 0
Where the paper says it

Everything runs on classical hardware; “quantum” refers to the Hilbert-space formalism, not quantum computing. The QCML embedding yields four geometric observables, Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity, which we compare against classical baselines including Absorption Ratio, Hamilton Markov-Switching, and GARCH (Table 2).

17117#pg2>. These observables are computed at every time step t and converted to z-scores via Algorithm 1 <ref:2605. p. 3
Where the paper says it

[5] show that gradient-descent-trained operators give the cleanest manifold approximation; our fixed operator basis preserves the spectral-gap structure that actually matters for detection, with the caveat that absolute dimension estimates are only approximate. Geometric Observables Four geometric observables are extracted from the QCML embedding, each computed at every time step t and converted to z-scores via Algorithm 1 (Table 2).

17117#pg4>.

Geometric Observables and Separability

The four geometric channels studied are the Absorption Ratio [1], Spectral Entropy, Hamiltonian Sensitivity, and Reduced State Purity <ref:2605.17117#pg4>. Reduced State Purity attains the highest in-sample separability of any method (d = 0. p. 0

Where the paper says it

6 per year). Reduced State Purity attains the highest in-sample separability of any method (d = 0.

83), tied closely by the Absorption Ratio [1] (d = 0. p. 0
Where the paper says it

Reduced State Purity attains the highest in-sample separability of any method (d = 0. 83), tied closely by the Absorption Ratio [1] (d = 0.

80) <ref:2605. p. 0
Where the paper says it

Email address: whammond@pitzer. edu (Will Hammond) arXiv:2605.

17117#pg3>. The geometric and classical channels are largely uncorrelated, with mean ρ ≈ 0. p. 0
Where the paper says it

83), tied closely by the Absorption Ratio [1] (d = 0. 80); geometric and classical channels are largely uncorrelated (mean |ρ| ≈ 0.

22, suggesting they capture distinct risk signals <ref:2605. p. 0
Where the paper says it

80); geometric and classical channels are largely uncorrelated (mean |ρ| ≈ 0. 22), suggesting they capture distinct risk signals.

17117#pg3>. The Spectral Entropy achieves mid-range offline separability among geometric channels (d = 0. p. 3
Where the paper says it

Spectral Entropy (Section 3. 2) tracks the full eigenvalue distribution and achieves mid-range offline separability among geometric channels (d = 0.

53, rank 12 of 46) <ref:2605. p. 0
Where the paper says it

Email address: whammond@pitzer. edu (Will Hammond) arXiv:2605.

17117#pg10>.

Walk-Forward Validation and Performance

The walk-forward nested HPO yields a Berry Phase Rate OOS median d = 0.72 (CI [0.34, 1.18]), the highest per-crisis Cohen’s d among three walk-forward detectors on five of nine crises <ref:2605. p. 1

Where the paper says it

Nested HPO yields a Berry Phase Rate OOS median d = 0. 18]), the highest per-crisis Cohen’s d among three walk-forward detectors on five of nine crises, with ∼67% fewer false alarms than Random Forest (1.

17117#pg3>. The Berry Phase Rate achieves the highest OOS median d = 0. p. 9
Where the paper says it

Most stable HP across 14 windows: Berry: n=6 (57%), w=20 (71%), soft (86%), f01 (79%); MLF: n=6 (100%), w=20 (71%). Berry Phase Rate achieves the highest OOS median d = 0.

72 (95% percentile bootstrap CI [0. p. 0
Where the paper says it

Under walk-forward nested hyperparameter selection on nine labelled windows, the Berry Phase Rate achieves an unbiased out-of-sample median Cohen’s d = 0. 72 (95% percentile-bootstrap CI [0.

34, 1.18] on the 9 windows; 10,000 resamples) and wins 5 of 9 crises <ref:2605. p. 9
Where the paper says it

72 (95% percentile bootstrap CI [0. 18] on the 9 windows; 10,000 resamples) and wins 5 of 9 crises, particularly rate-driven events (Rate Hikes 10

17117#pg3>. The best detector changes every crisis, meaning the channels carry complementary information rather than ranking stably <ref:2605. p. 9
Where the paper says it

73 — Berry wins 5/9 crises (Flash Crash 2010, China 2015, Q4 Selloff 2018, Rate Hikes 2022, SVB 2023); MLF wins 4/9 (Euro 2011, Volmageddon 2018, Repo 2019, COVID 2020). The best detector changes every crisis; no single observable dominates the whole panel.

17117#pg10>.

Theoretical Validation and Risk Overlay

The spectral gap remains strictly positive throughout the observed data manifold (∆min = 2.34), confirming that the smoothness condition holds everywhere in the observed data manifold <ref:2605. p. 2

Where the paper says it

(4) Equation (4) holds in the parallel-transport gauge ⟨ψ|∂aψ⟩ = 0; our numerical implementation uses the gauge-invariant plaquette formula (Section 3. Theorem 1 (Smoothness [5]).

17117#pg19>. The QFI–metric identity is verified to machine precision, confirming that the quantum metric is exactly one-quarter of the QFI matrix <ref:2605. p. 3
Where the paper says it

Consequently: (i) The QFI pseudo-determinant det+(g) is proportional to the volume element of the Fisher information metric on the statistical manifold: det+(g) = 4−r det+(FQ|r), where r = rank(g). (ii) The quantum Cramér–Rao bound [27, 28] Var(ˆxa) ≥ 1 4 [g −1 ]aa implies that the quantum metric controls the ultimate precision for estimating data coordinates from state measurements.

17117#pg21>. A simple use case for the geometric channel is de-risking, where a Berry Phase Rate causal z-score exceeding 2. p. 4
Where the paper says it

Table 2: Geometric observatory channels studied in this paper. Channel Family Geometric property Equation Berry Phase Rate Holonomy U(1) holonomy change rate (5) Spectral Entropy Spectral Excitation distribution (7) Hamiltonian Sensitivity State Perturbation variance (8) Reduced Purity State Subsystem separability (9) 3.

0 triggers an exit to cash for 60 trading days <ref:2605. p. 6
Where the paper says it

We evaluate on 17 historical crises spanning 2000–2024, classified a priori as 9 Conventional (with historical parallels) and 8 Novel (unprecedented mechanisms). Crisis windows are extended by ±10 trading days.

17117#pg14>. The Berry overlay returns 443% with Sharpe 0. p. 1
Where the paper says it

2/yr for the Berry channel are low enough to support a practitioner risk-management overlay; a SPY de-risking experiment (Section 6. 3) cuts maximum drawdown by ∼51% with no degradation in Sharpe ratio.

70 and worst drawdown −27%, which beats buy-and-hold on Sharpe <ref:2605. p. 1
Where the paper says it

2/yr for the Berry channel are low enough to support a practitioner risk-management overlay; a SPY de-risking experiment (Section 6. 3) cuts maximum drawdown by ∼51% with no degradation in Sharpe ratio.

17117#pg14>.

Limitations and Future Work

The primary metric (Cohen’s d) measures separability of score distributions within known crisis windows, which is an offline event study rather than a real-time detection benchmark <ref:2605.17117#pg14>. The walk-forward results provide a causal complement, but the observables largely react to rather than lead stress (17/45 reverse vs. p. 8

Where the paper says it

2) shows reverse causality (market → QCML) dominating forward (17/45 vs. 6/45); the observables respond to stress rather than leading it.

6/45 forward) <ref:2605. p. 0
Where the paper says it

Email address: whammond@pitzer. edu (Will Hammond) arXiv:2605.

17117#pg12>. Reconstruction-loss operator learning may sharpen detection by optimizing manifold faithfulness rather than regime separability <ref:2605. p. 3
Where the paper says it

Why fixed operators suffice for detection. The QCML framework [5] normally learns operators via reconstruction loss to estimate intrinsic dimension and generate predictions.

17117#pg15>. The framework is currently tested on a single equity pair, and generalization to other asset classes remains untested <ref:2605.17117#pg15>.

Data and Code Availability

Market data from Yahoo Finance via the open-source yfinance library (v0.2.x), freely available at no cost. p. 7

Where the paper says it

Detection Pipeline Algorithm 3 QCML Geometric Regime Detection Require: Feature matrix X ∈ R T ×d , Hilbert dimension n, PCA components p Ensure: Z-scored observable time series 1: Standardize X; PCA to p components; L 2 -normalize 2: Fit operators {Ak} via PCA-inspired method (or expanding-window refit) 3: for t = 1, . , T do 4: Solve |ψ(xt)⟩ = ground state of H(xt) (1) 5: Compute raw observables (Table 2) 6: end for 7: Convert to z-scores via Algorithm 1 Computational cost.

All results are fully reproducible without any paid subscription. p. 8
Where the paper says it

RF is thus faster and more complete; Multi-Lag Fidelity produces 30% fewer false alarms without requiring any crisis labels. Note that RF produces probability scores ∈ [0, 1], not z-scores, making the fixed z > 2 threshold dimensionally inapplicable; the FAR-calibrated results are the appropriate benchmark for RF comparisons.

The qcml geometry/ library and all experiment scripts, together with crisis definitions, hyperparameter configurations, and canonical result JSON files, are available at https://github. p. 6
Where the paper says it

Experimental Protocol 4. Data and Crisis Definitions We use daily OHLCV data for SPY and DIA (1998–2024) from Yahoo Finance via the yfinance library (v0.

com/willhammondhimself/qcml-geometric-sde. p. 0
Where the paper says it

Score construction is unsupervised; hyperparameter selection is the only supervised step. Keywords: regime detection, financial crises, walk-forward validation, false-alarm rate, Berry curvature, Fubini–Study metric, quantum Fisher information, spectral metric learning, geometric observables, QCML 1.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Berry Phase Rate Detection: An AI system can be deployed to detect acute shocks, such as Flash Crash events with a median Cohen’s d of 0.77 during walk-forward validation on specific crises, achieving a false alarm rate of 1.2 per year compared to 3.6 for Random Forest.

  2. Reduced State Purity Monitoring: The system can identify prolonged structural crises by monitoring the loss of separability, as evidenced by its offline separability score of d = 0.83, signaling when the ground state cannot be decomposed into independent factor components.

  3. Geometric Channel Orthogonality Analysis: The AI can assess whether different geometric observables capture distinct risk signals by measuring their mean correlation, which is found to be approximately 0.22 across channels like the Fubini–Study metric and Berry curvature.

  4. Adaptive Thresholding for Real-time Alerts: An AI system can implement a detection pipeline that uses a rolling quantile threshold that sets the alarm level at the 95th percentile of scores over a trailing 252-day window combined with a score velocity trigger to fire alarms only when both mechanisms are triggered.

  5. Walk-Forward Hyperparameter Optimization: The system can dynamically optimize its internal settings using Nested HPO, re-optimizing hyperparameters at each expanding window using only past crises to achieve unbiased out-of-sample performance, minimizing the train–OOS gap to 0.25 for the Berry Phase Rate.

  6. Risk Management Overlay: The AI can generate concrete risk management rules by using a Berry overlay where an exit to cash for 60 trading days is triggered when the causal z-score of the Berry Phase Rate exceeds 2.0, achieving a worst drawdown of -27% with a Sharpe ratio of 0.70.

Abstract

We extract four geometric observables -- Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity -- from a learned spectral embedding of equity-index returns and evaluate them as regime-shift detectors against 46 classical and machine-learning baselines on 17 historical crises spanning 2000-2024. Under walk-forward nested hyperparameter selection on nine labelled windows, the Berry Phase Rate achieves an unbiased out-of-sample median Cohen's d = 0.72 (95% percentile-bootstrap CI [0.34, 1.18], 10,000 resamples) and produces approximately 67% fewer false alarms per year than a label-supervised Random Forest (1.2 vs. 3.6 per year). Reduced State Purity attains the highest in-sample separability of any method (d = 0.83), tied closely by the Absorption Ratio (d = 0.80); geometric and classical channels are largely uncorrelated (mean ρ about 0.22), suggesting they capture distinct risk signals. Score construction is unsupervised; hyperparameter selection is the only supervised step.

Sources

Related papers