💰 quantitative finance

Deformation of semi-circle law for the correlated time series and Phase transition

This paper investigates how temporal correlations in financial time series deform the Wigner semicircle law of random matrices, demonstrating that correlation strength alters spectral moments and induces a phase transition between finite and divergent higher-order moments under power-law decay.

Original authors: Masato Hisakado, Takuya Kaneko

Published 2026-07-14

📖 5 min read🧠 Deep dive

CC0 1.0

Original authors: Masato Hisakado, Takuya Kaneko

Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have a giant, magical mirror made of a grid of tiny squares. If you fill this mirror with completely random, chaotic noise—like static on an old TV—the reflection you see follows a very predictable, smooth shape called the "semicircle law." It's like a perfect, gentle hill: low on the sides, high in the middle, and nothing too wild happening.

But what happens if that noise isn't random? What if the squares in your mirror are actually whispering secrets to their neighbors, creating a pattern of temporal correlations? That's exactly what Masato Hisakado and Takuya Kaneko investigated. They took financial time series (like stock prices or currency rates) and built these "Wigner matrices" to see how the whispers changed the shape of the hill.

The Big Twist: The Hill Gets a Spiky Crown

Their main discovery is that when these financial numbers are correlated, the perfect semicircle gets deformed. It doesn't just wiggle; it transforms. The center of the hill gets a much higher, sharper peak, and the tails (the sides) stretch out further, becoming "fatter."

Think of it like a crowd of people standing in a circle. If everyone is independent, they spread out evenly. But if they are all holding hands and reacting to each other's movements (correlation), they bunch up tighter in the middle and stretch out further at the edges. The authors found that the stronger the connection (correlation) between the numbers, the spikier the peak and the fatter the tails become.

The "Magic Number" and the Phase Transition

Here is where it gets really wild. The authors looked at two types of "whispers" or correlations:

Exponential Decay: The whispers fade away quickly, like a secret told down a line of friends who stop listening after a few steps. In this case, the hill gets spiky, but it stays stable.
Power Law Decay: The whispers are stubborn. They fade very slowly, meaning a number from long ago still influences the present. This is where a phase transition happens.

The authors discovered a critical "magic number" called $\gamma = 1/2$ .

If the stubbornness of the whispers is greater than 1/2 ( $\gamma > 1/2$ ), the system is stable. The "fourth moment" (a fancy math way of measuring how fat the tails are) stays finite, and the biggest eigenvalue (the highest point of the hill) stays finite.
If the stubbornness is 1/2 or less ( $\gamma \le 1/2$ ), things go crazy. The fourth moment and the largest eigenvalue become infinite.

It's like a bridge that holds up fine under normal traffic, but if the traffic gets too "sticky" (too much long-range correlation), the bridge suddenly collapses into infinity. This isn't just a guess; in their numerical simulations, they saw this transition clearly, with the data behaving differently on either side of that $1/2$ line.

Real-World Detective Work

The team didn't just play with math; they tested this on real financial data. They looked at 25 different time series, including cryptocurrencies, commodities, government bonds, and foreign exchange (FX) rates.

They set up a "null hypothesis" test: What if these markets were just random noise (normal i.i.d.)?

The Result: Most markets, like stocks and commodities, fit the semicircle law pretty well. They are mostly random.
The Outliers: However, specific foreign exchange markets (like USD/CHF, USD/CAD, EUR/CHF, and EUR/GBP) and some crypto assets rejected the idea that they are just random noise. Their eigenvalue distributions had those tell-tale signs: higher peaks and fatter tails.

The authors calculated that the deviation in these specific markets is largely due to temporal correlation. For example, in the USD/CHF market, the correlation at the shortest time lag was strong enough to push the fourth moment up to 2.128, well outside the range expected for random noise (which hovers around 2).

What They Ruled Out

It's important to note what this paper says is not the main culprit. While financial data is known for having "fat tails" (extreme events), the authors argue that for these specific Wigner matrices, the temporal correlation is the primary driver of the deformation. They explicitly show that the difference from the semicircle law depends on the correlation structure, not just the fat-tailed nature of the returns alone.

Also, they clarify that this behavior is different from other types of random matrices (like Wishart matrices). In those other systems, the "phase transition" is controlled by the second moment. But in this Wigner matrix world, it's all about the fourth moment. This suggests that different types of random matrices belong to different "universality classes"—they have their own unique rules for how they react to correlations.

How Sure Are They?

The authors are confident in their analytical math, which predicts that the fourth moment increases with correlation strength. They back this up with numerical simulations using synthetic data (like fractional Brownian motion) and real financial data. These simulations show the finite-size scaling behavior near the transition point, supporting the idea that the phase transition at $\gamma = 1/2$ is real.

However, they are careful to state that while the moment calculations are solid, a rigorous mathematical proof of "weak convergence" (a deep statistical guarantee) would require more advanced tools like the Stieltjes transform, which they didn't use in this specific paper. So, while the evidence from simulations and moment analysis is strong, the absolute mathematical proof of the convergence is still a work in progress.

In short, the paper reveals that financial markets, especially currency ones, aren't just random noise. They have a hidden "memory" that warps the mathematical landscape, turning a gentle hill into a spiky, stretched-out mountain range, and if that memory gets too strong, the whole structure hits a critical tipping point.

Technical Summary: Deformation of Semicircle Law for Correlated Time Series and Phase Transition

Problem Statement
Random Matrix Theory (RMT) provides a universal framework for analyzing the spectral behavior of large complex systems. Classical results, such as the Wigner semicircle law, describe the eigenvalue distributions of symmetric random matrices with independent entries. However, financial time series often exhibit two distinct properties: temporal correlations and fat-tailed distributions. While deviations from classical laws in Wishart matrices (used for covariance estimation) have been linked to temporal correlations and phase transitions, it remains an open question whether these correlation-induced spectral transitions are universal across different random matrix ensembles. Specifically, it is unclear if the mechanisms governing Wishart matrices extend to Wigner-type matrices, which differ fundamentally in symmetry and construction. This paper investigates the deformation of the Wigner semicircle law when matrix entries are constructed from time series with temporal correlations.

Methodology
The authors construct Wigner-type random matrices from stochastic time series $A_t$ with temporal correlations. The time series are normalized to have zero mean and unit variance. The correlation structure is defined by a correlation matrix where the temporal correlation function $d_t = \text{Cov}(A_i, A_{i+t})$ follows either an exponential decay ( $d_t = r^t$ ) or a power-law decay ( $d_t \sim t^{-\gamma}$ ).

To form the $N \times N$ Wigner matrix $S$ , the time series of length $T = N^2$ is "folded" such that matrix elements $S_{ij}$ inherit the temporal correlation structure of the underlying series. The study focuses on the limit $N \to \infty$ .

The primary analytical tool is a moment analysis of the eigenvalue distribution. The authors calculate the $k$ -th moment $\mu_k$ of the spectral density, defined as $\mu_k = \frac{1}{N} \langle \text{Tr}(S^k) \rangle$ . They specifically analyze the second, fourth, and sixth moments to characterize deviations from the standard semicircle law. The theoretical predictions are validated through:

Analytical Derivation: Calculating moments for both exponential and power-law decay cases.
Numerical Simulations: Using synthetic data (including fractional Brownian motion) and empirical financial time series (25 assets including FX, crypto, commodities, and bonds).
Statistical Testing: Comparing the moments of empirical data against a null hypothesis of normal independent and identically distributed (i.i.d.) data using bootstrap methods to establish confidence intervals (5% and 95% quantiles).

Key Contributions and Results

Deformation of the Semicircle Law: The study demonstrates that temporal correlations deform the Wigner semicircle law. The deformed distribution exhibits a higher central peak and heavier tails compared to the standard semicircle law.
Moment Analysis:
- Second Moment: Remains invariant ( $\mu_2 = 1$ ) regardless of the correlation decay type, consistent with the standard semicircle law.
- Fourth Moment: Increases monotonically with the strength of temporal correlations. For exponential decay, the fourth moment is finite. For power-law decay, the fourth moment depends on the power index $\gamma$ .
Phase Transition in Power-Law Decay: A critical phase transition is identified in the power-law decay case at a critical exponent $\gamma_c = 1/2$ $γ_{c} = 1/2$ .
- When $\gamma > 1/2$ , the fourth moment (and the largest eigenvalue) is finite.
- When $\gamma \le 1/2$ , the fourth moment diverges, leading to an infinite largest eigenvalue.
- The authors note a secondary scaling transition at $\gamma = 1/4$ related to finite-size corrections induced by the folding construction, though this is subleading to the bulk divergence at $\gamma = 1/2$ .
Universality Class Distinction: The paper establishes that correlation-induced spectral transitions are not universal across all random matrix ensembles. Unlike the Wishart ensemble, where the transition is governed by the divergence of the second moment, the Wigner ensemble transition is controlled by the divergence of the fourth moment. This difference arises from the distinct combinatorial structures in the moment expansions of the two ensembles (pair contractions in Wishart vs. diagonal-crossing correlations in Wigner).
Empirical Application: Analysis of 25 financial time series reveals that most assets (commodities, stocks, bonds) fit the semicircle law well. However, specific foreign exchange (FX) pairs (e.g., USD/CHF, USD/CAD, EUR/CHF, EUR/GBP) and certain crypto assets show significant deviations. These deviations correlate with higher temporal autocorrelations in the shortest time lags, confirming that temporal correlation is the primary origin of the spectral deformation in these markets.

Significance
The paper claims to resolve the question of universality regarding correlation-induced spectral transitions. It demonstrates that Wigner and Wishart ensembles belong to distinct universality classes concerning correlation effects. The findings highlight that the mechanism of spectral deformation is sensitive to the underlying matrix structure. Specifically, the shift from second-moment control (Wishart) to fourth-moment control (Wigner) suggests that different combinatorial factors dictate how correlations propagate through the spectrum. The identification of a phase transition at $\gamma = 1/2$ in the Wigner ensemble provides a theoretical basis for understanding extreme spectral behaviors in systems with long-range temporal correlations, such as certain financial markets. The results are supported by finite-size scaling behavior observed in numerical simulations near the transition point.

The Big Twist: The Hill Gets a Spiky Crown

The "Magic Number" and the Phase Transition

Real-World Detective Work

What They Ruled Out

How Sure Are They?

More like this