The Physics of Price Discovery: Deconvolving Information, Volatility, and the Critical Breakdown of Signal during Retail Herding

Sungwoo Kang · 2026

Read The Physics of Price Discovery: Deconvolving Information, Volatility, and the Critical Breakdown of Signal during Retail Herding by Sungwoo Kang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-11602.txt.

Introduction

Sungwoo Kang studies the physics of price discovery in markets. The 2026 preprint tracks how an informed signal travels and breaks. A prior 2025 finding set the stage for this work. Market cap normalization isolates a pure directional signal there.

That signal reflects informed trading in the market. Kang asks how it is transmitted through price action. He also asks how it breaks down under stress. Retail herding is the breakdown case in the title. When crowds herd, the clean signal can fail at once. Deconvolving information and volatility is the method's core. Such work speaks to traders who read order flow signals.

Catalog record

CREATORSungwoo Kang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-11602

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The Physics of Price Discovery: Deconvolving Information, Volatility, and the Critical Breakdown of Signal during Retail Herding Sungwoo Kang Abstract Building on the finding that Market Cap Normalization ( S MC S_{\text{MC}} ) isolates the “pure” directional signal of informed trading ( Kang, 2025 ) , this paper investigates the physics of how that signal is transmitted—and how it breaks down. We employ Tikhonov-regularized deconvolution to recover the impulse response kernels of investor flows, revealing a dual-channel market structure: Foreign and Institutional investors act as “architects” of price discovery (positive permanent impact), while Individual investors act as liquidity providers (negative total impact). However, using Multivariate Hawkes Processes , we demonstrate that this structure is fragile. We find that individual investor order flow exhibits near-critical self-excitation (Branching Ratio ≈ \approx 0.998). During periods of high retail herding, the market undergoes a phase transition into a “critical state.” In this regime, the signal-to-noise ratio collapses, causing the price impact of sophisticated investors to reverse from positive to negative. These findings suggest that retail contagion acts as a physical barrier that temporarily disables efficient price discovery. Keywords: Market microstructure, Price discovery, Order flow, Deconvolution, Hawkes processes, Herding behavior, Information geometry JEL Classification: G14, G12, C58 1 Introduction The central tenet of market microstructure theory is the distinction between “informed” trading, which facilitates permanent price discovery, and “uninformed” or noise trading, which generates transient volatility. Since Kyle, 1985 , researchers have sought to decouple these forces by analyzing the relationship between order flow imbalance and asset returns. However, a critical methodological ambiguity remains: how should order flow be normalized to extract the purest informational signal? Standard practice typically employs volume normalization ( S_{\text{TV}} ), effectively measuring the execution cost relative to current liquidity. However, this conflates information arrival with volatility. In Kang, 2025 , we established that Market Cap Normalization ( S_{\text{MC}} ) is the superior metric for isolating the static “pure” information signal from liquidity noise. Building on that foundation, this paper shifts focus from the measurement of signal to the physics of its transmission. We investigate the dynamic mechanics of price discovery, specifically modeling how the “Phase Transition” into a “Critical State” of retail herding—quantified via “Hawkes Processes”—can disrupt the standard impulse response functions recovered through “Deconvolution.” We approach this problem through two advanced physical modeling frameworks. First, we employ Tikhonov-regularized deconvolution to extract the “Impulse Response Functions” of investor flows. This allows us to visualize the temporal shape of liquidity: whether a unit of flow creates a permanent step-function in price (characteristic of fundamental information) or a transient spike-and-decay pattern (characteristic of noise). Second, we model the systemic stability of these flows using Multivariate Hawkes Processes , treating large trading surges as self-exciting point processes. This allows us to quantify the “criticality” of the market—specifically, whether trading activity is exogenous (news-driven) or endogenous (feedback-driven). Our analysis yields three distinct contributions to the physics of price discovery. First, using information-theoretic measures, we resolve the debate between volume and market-cap normalization. We show that while volume-normalized flows ( S_{\text{TV}} ) appear to have high distributional divergence, this is largely an artifact of volatility clustering. Once returns are standardized by their rolling volatility, the informational content of institutional S_{\text{MC}} flows is shown to be 4.85 times greater than that of S_{\text{TV}} flows (representing a 14-fold improvement over the unadjusted baseline). This confirms that S_{\text{MC}} isolates the directional “signal,” while S_{\text{TV}} largely captures the volatility “noise.” Second, our deconvolution analysis reveals a stark physical segmentation of market participants. We identify a “Global Market Kernel” where Foreign and Institutional flows exert a positive, persistent price impact, acting as the architects of price discovery. Conversely, Individual investor flows exhibit a negative cumulative impact , effectively acting as liquidity providers whose buying pressure systematically precedes mean reversion. Third, and perhaps most significantly, we document a regime-dependent breakdown in market efficiency. We find that individual investor order flow exhibits a Hawkes Branching Ratio of \approx 0.998, indicating a system in a near-critical state of self-excitation. During periods of high retail herding, the standard laws of price impact are suspended. We show that the price impact of sophisticated institutional investors, typically positive and permanent, reverses to negative during these high-herding regimes. This suggests that “swarming” behavior by noise traders creates a noise barrier so dense that it temporarily inverts the signal-to-noise ratio, rendering sophisticated flow toxic. Collectively, these findings suggest that price discovery is not a constant mechanical property of the market, but a state variable dependent on the “temperature” of retail herding. When the crowd is quiet, the signal ( S_{\text{MC}} ) transmits clearly; when the crowd herds, the physics of information transmission breaks down. The remainder of this paper is organized as follows. Section 2 reviews the theoretical framework and related literature. Section 3 describes our data and empirical methodology. Section 4 presents our main results across the three experiments. Section 5 discusses the implications and Section 6 concludes. 2 Theoretical Framework and Literature Review 2.1 Market Microstructure Theory The foundation of market microstructure theory lies in distinguishing between informed and uninformed trading. Kyle, 1985 developed the seminal continuous-time trading model where informed traders possess private information about an asset’s fundamental value, while noise traders generate non-informational order flow. This framework predicts that informed trading should have permanent price impact, while noise trading generates only transient volatility. The empirical challenge is measuring this distinction. Traditional approaches normalize order flow by trading volume to construct imbalance metrics. However, as Hasbrouck, 1991 and Hasbrouck, 1993 demonstrate, volume itself is endogenous and varies with information arrival. This creates a confound: high volume-normalized imbalance may indicate informed trading or simply reflect low liquidity conditions. We build on recent work proposing market capitalization as an alternative normalization ( Kang, 2025 ) . The intuition is that market cap represents the fundamental “scale” of a security, independent of transient liquidity conditions. A $1 million buy order in a $100 million market cap stock represents the same fundamental signal as a $10 million buy in a $1 billion stock, but volume-based normalization would treat these differently if liquidity varies. 2.2 Information Geometry and KL Divergence To quantify the informational content of different normalization schemes, we employ tools from information geometry. The Kullback-Leibler (KL) divergence ( Kullback and Leibler, 1951 ) measures the separation between two probability distributions. For return distributions conditional on buying ( P_{\text{buy}} ) versus selling ( P_{\text{sell}} ), the KL divergence quantifies how distinguishable the two regimes are: D_{KL}(P_{\text{buy}}\|P_{\text{sell}})=\int P_{\text{buy}}(r)\log\frac{P_{\text{buy}}(r)}{P_{\text{sell}}(r)}dr (1) A key insight is that return distributions exhibit fat tails ( Mandelbrot, 1963 ) , requiring Student- t distributions for accurate characterization. Moreover, volatility clustering ( Engle, 1982 ) creates artifacts: high S_{\text{TV}} periods may simply coincide with high volatility regimes, inflating apparent distributional separation without genuine informational content. We address this by standardizing returns using rolling volatility: R_{\text{adj},t+1}=\frac{R_{t,t+1}}{\sigma_{t,\text{rolling}}} (2) where \sigma_{t,\text{rolling}} is the 20-day rolling standard deviation. This volatility adjustment isolates the directional signal from the volatility artifact. 2.3 Impulse Response and Deconvolution The price impact literature typically estimates contemporaneous or short-lag effects ( Evans and Lyons, 2002 ) . However, the full temporal structure of impact—the “impulse response function”—provides richer physical insight. Consider the discrete-time convolution model: R_{t+k}=\sum_{\tau=0}^{L}\psi_{\tau}\cdot I_{t+k-\tau}+\varepsilon_{t+k}

Read the complete text