¹ Contabilidad, Universidad Autonoma del Peru, Lima District, Lima Region, 15001, Peru
² Facultad de Ciencias Administrativas y Económicas, Institución Universitaria Tecnológico de Antioquia, Medellín, Antioquia, 050034, Colombia

Jorge A. Restrepo Morales
Roles: Conceptualization, Formal Analysis, Methodology, Writing – Original Draft Preparation, Writing – Review & Editing

Rosa Ysabel Moreno Rodriguez
Roles: Conceptualization, Supervision, Writing – Review & Editing

Freddy Zea Restrepo
Roles: Data Curation, Formal Analysis, Investigation, Visualization, Writing – Review & Editing

Emerson Andrés Giraldo Betancur
Roles: Formal Analysis, Investigation, Methodology, Software, Writing – Review & Editing

The nexus between market structure and systemic risk has been a central theme in financial economics, particularly since the 2008 global financial crisis. Foundational work by Adrian and Brunnermeier (2016) introduced the CoVaR methodology to measure systemic risk contributions, while Billio et al. (2012) pioneered network-based measures to map financial interconnectedness. While seminal, these studies primarily focused on inter-firm linkages within the banking sector. The epicenter of systemic risk analysis has since expanded to encompass threats originating from market-based finance, particularly those related to market concentration.

Recent scholarship has rigorously documented this shift. De Franco (2021) show the dramatic increase in U.S. equity market concentration, a trend exacerbated by the massive inflows into passive investment vehicles. This has profound implications for financial stability. Federal Reserve researchers have warned that the structural shift toward passive investing may introduce new vulnerabilities into the system. Anadu et al. (2020) argue that while passive investing has benefits, it may also increase the risk of fire sales during periods of stress and reduce the incentives for market-stabilizing arbitrage. This mechanism provides a direct link between concentrated, index-based investing and systemic fragility.

Furthermore, the literature on common ownership has evolved to consider its impact on risk. Azar, Schmalz, and Tecu (2018) highlighted the anticompetitive effects, but recent work explores financial stability implications. Proposing the influential ‘inelastic markets hypothesis’, Gabaix and Koijen (2021) demonstrate that shocks to institutional demand have a greatly amplified impact on asset prices, creating volatility that is disconnected from fundamentals. This aligns with the findings of Koijen and Yogo (2019), who model the price impact of large-scale institutional flows, suggesting that demand shocks from these mega-investors can be a primary driver of prices.

The growth of passive index funds and exchange-traded funds (ETFs) adds a structural dimension to this concentration dynamic. As passive ownership increases, portfolio holdings converge toward the same large-cap constituents (Appel, Gormley, and Keim, 2016; Anadu et al., 2020), mechanically amplifying concentration in equity indices. This convergence reduces the market’s capacity for price discovery and increases co-movement across assets (Koijen and Yogo, 2019; Gabaix and Koijen, 2021), creating the conditions under which idiosyncratic shocks can propagate as systemic events.

The COVID-19 pandemic served as a real-world stress test for these dynamics. Haddad, Moreira, and Muir (2021) found that the market crash of March 2020 was characterized by a flight to cash rather than a typical flight to quality, with unprecedented selling pressure even on safe assets like U.S. Treasuries. This suggests that traditional diversification benefits may erode precisely when market concentration is high and large institutions are forced to de-lever simultaneously. Concurrently, studies on cross-asset connectedness, such as that by Yousaf and Ali (2020), have documented intensified volatility spillovers between equities and emerging asset classes like cryptocurrencies during the pandemic, highlighting new channels for contagion.

Three structural disruptions within our study period further sharpen the empirical urgency of the concentration-fragility nexus. The COVID-19 market crash (February–March 2020) demonstrated how concentrated index exposure amplified the initial shock, with the top-10 S&P 500 holdings accounting for a disproportionate share of index-level losses. The “meme stock” episode (January–February 2021) illustrated how passive-induced concentration creates asymmetric volatility transmission. The 2022 rate-shock cycle exposed how cross-asset correlations spike precisely when diversification is most needed, a pattern our framework models explicitly.

Two competing theoretical perspectives frame the debate. The concentration-stability hypothesis (Diamond, 1984; Beck et al., 2013) argues that fewer, larger institutions are more resilient due to superior diversification and economies of scale. The concentration-fragility hypothesis, which we adopt, posits instead that high concentration amplifies systemic risk through reduced market depth, correlated liquidations, and cross-asset contagion (Adrian and Brunnermeier, 2016; Billio et al., 2012). Our empirical results support the latter view across the full 2020–2024 period and both volatility regimes.

We ground our analysis in the theoretical framework of financial contagion established by Allen and Gale (2000), which we extend to incorporate market concentration dynamics through three primary channels:

Concentration Risk Channel: Elevated market concentration increases the probability that an idiosyncratic shock to a single large-cap firm can propagate into a systemic event. When a small number of firms dominate an index, their individual volatilities disproportionately influence the entire market’s risk profile.

Liquidity Amplification Channel: As theorized by Brunnermeier and Pedersen (2009), crowded trades in a few popular assets can create precarious liquidity conditions. During periods of market stress, a rush to exit these concentrated positions can trigger a liquidity spiral, dramatically amplifying price declines.

Cross-Asset Contagion Channel: Extreme concentration in the equity market can initiate volatility spillovers to other asset classes. As described by Kodres and Pritsker (2002), forced selling or portfolio rebalancing from concentrated equity positions can transmit shocks to fixed income, commodity, and even cryptocurrency markets.

Critically, these three channels operate not as independent mechanisms but as a sequential transmission cascade. Rising concentration (Concentration Risk Channel) reduces effective market depth, increasing the price impact of portfolio adjustments. This amplifies mark-to-market losses for leveraged institutions, triggering forced sales (Liquidity Amplification Channel). The resulting volatility then propagates to correlated assets via portfolio rebalancing and contagion (Cross-Asset Contagion Channel). This cascade is the core theoretical prediction tested by our four hypotheses.

Our dataset spans from January 1, 2020, to October 31, 2024, comprising 1,218 daily observations. The data encompasses a wide range of asset classes to provide a holistic view of the financial system:

Equity Markets: S&P 500, NASDAQ-100, and Russell 2000 indices; individual stock data for market capitalization calculations; and sector-specific ETFs (XLK, XLF, XLE, XLV, XLI, XLP, XLY, XLU, XLB, XLRE, XLC).

Fixed Income: U.S. Treasury yields (2-year, 5-year, 10-year, 30-year), investment-grade corporate bond spreads (LQD), high-yield spreads (HYG), and TIPS breakeven inflation rates.

Commodities: Gold (GLD), Silver (SLV), Crude Oil (USO), Natural Gas (UNG), and a broad agricultural commodity index (DBA).

Cryptocurrencies: Bitcoin and Ethereum prices, total cryptocurrency market capitalization, and indices for Decentralized Finance (DeFi) and Non-Fungible Tokens (NFTs).

Sample construction note. The Zenodo dataset contains 1,261 weekday sessions (January 2, 2020 to October 31, 2024). The estimation sample for VECM and Markov-Switching models is 1,218 observations, reflecting the removal of 43 sessions with zero or missing trading volume in thin cryptocurrency markets, checked against Bloomberg cross-reference. All robustness results in Extended Data Table 6 use the same 1,218-observation sample.

3.2.1 Market Concentration Measures

We employ three distinct but related measures to capture market concentration robustly:

Herfindahl-Hirschman Index (HHI): Calculated as:

HHI_t=Σw2_{i,t}

where w_{i,t} is the market capitalization weight of firm i at time t.

Concentration Ratio (CR10): The combined market weight of the top 10 firms:

CR10_t=Σ(i=1to10)w_{i,t}

Entropy-Based Concentration Index (ECI): A measure of market diversity:

ECI_t=−Σw_{i,t}×ln(w_{i,t})

A lower ECI value indicates higher market concentration.

Our approach uniquely combines a VECM for long-run relationships with a Markov-Switching model to capture nonlinear dynamics.

3.3.1 Vector Error Correction Model (VECM)

The VECM allows us to model both short-term dynamics and long-term cointegrating relationships between the variables. The model is specified as:

ΔY_t=αβ′Y_{t−1}+ΣΓ_iΔY_{t−i}+ε_t

where: - Y_t is the vector of endogenous variables - α is the vector of adjustment coefficients measuring the speed of convergence to equilibrium - β contains the cointegrating vectors representing the long-run equilibrium relationships - Γ_i are matrices capturing short-run dynamics - ε_t ~ N(0, Ω) is the vector of stochastic disturbances.

Prior to VECM estimation, ADF and PP unit root tests confirm that VaR, HHI, CR10, and ECI are integrated of order one [I(1)], while VIX is stationary [I(0)] and enters as an exogenous control (Extended Data Table 6, Panel A). The Johansen trace test identifies two cointegrating vectors at the 5% significance level, supporting the 4-variable VECM specification. Lag length is selected by the Akaike Information Criterion (AIC), yielding p = 2 lags for the baseline specification.

Post-estimation diagnostic tests are reported in Extended Data Table 6. Model stability is confirmed by the characteristic polynomial analysis: all non-unit roots of the VECM companion matrix lie strictly inside the unit circle. Residuals show no serial autocorrelation (LM test, 12 lags) and conditional heteroskedasticity is addressed by the Markov-Switching specification in columns (3)–(4) of Table 2.

3.3.2 Market Fragility Index (MFI) Construction

We construct the MFI using Principal Component Analysis (PCA) to synthesize information from a broad set of risk indicators into a single, comprehensive measure. The index is defined as:

MFI_t=Σλ_k×PC_k(X_t)

The component variables for the PCA include: - Market Concentration: HHI, CR10, ECI - Cross-Asset Correlations: Rolling 21-day correlations between equities, bonds, and commodities - Volatility and Tail Risk: Volatility clustering indicators (GARCH), VIX, and the VIX/SKEW ratio - Liquidity Measures: Bid-ask spreads and trading volume for key assets - Sentiment Indicators: Put-call ratios and investor sentiment surveys.

The PCA retains PC1 and PC2, jointly explaining approximately 83% of total variance. PC1 loads on concentration measures (HHI, CR10, ECI_inv) capturing the concentration risk dimension; PC2 loads on VIX and VaR, capturing the volatility-tail risk dimension. The MFI is the linear combination of PC1 and PC2 weighted by their respective eigenvalues. This composite index is designed to be a leading indicator, computed from t–1 information only, precluding look-ahead bias in the predictive performance evaluation.

Preregistration statement: This study did not involve preregistration of the research design or data analysis plan at an independent registry.

Table 1 presents the summary statistics for all key variables over the full sample period from January 2020 to October 2024. The statistics reveal significant non-normality in asset returns, characterized by negative skewness and high kurtosis (fat tails), justifying the use of advanced risk models. The HHI and CR10 measures show considerable variation, reaching historically high levels during the sample period.

Table 2 presents the results of the regression analyses examining the impact of market concentration on tail risk. The results provide robust evidence for our core hypothesis. The baseline model indicates that a 1% increase in the HHI is associated with a 2.31% increase in tail risk (1% VaR). This effect remains significant after including standard controls for volatility (VIX) and economic conditions (Term Spread). Crucially, the regime-switching model reveals that this relationship is nonlinear: the impact of concentration on tail risk nearly doubles from the low-volatility regime (β = 1.23) to the high-volatility regime (β = 2.67). This suggests that concentration acts as a powerful risk amplifier, particularly during periods of market stress.

The economic intuition behind the estimated coefficients warrants elaboration. The positive and statistically significant coefficient on HHI (β = 2.31 in the baseline specification, column 1; β = 2.67 in the high-volatility regime, column 4) implies that a one-unit increase in HHI—moving from a perfectly competitive market to a monopoly—would increase daily VaR by 231 to 267 basis points. More practically, the observed HHI range during our sample period (0.09 to 0.19, Table 1) implies a concentration-induced tail risk premium of approximately 23–27 basis points across the full sample, rising to 40+ basis points during the high-volatility regimes identified by the Markov filter.

The predictive performance of the MFI relative to traditional indicators is summarized in Table 3. The MFI significantly outperforms traditional risk indicators in predicting market downturns. With an Area Under the Curve (AUC) of 0.891 and an average lead time of over seven days, the MFI serves as a superior early-warning system for policymakers and investors by holistically capturing the risks stemming from concentration, correlations, and volatility.

Table 4 reports the volatility spillover matrix estimated via the Diebold-Yilmaz method. The spillover analysis reveals that the S&P 500 is the primary net transmitter of systemic risk at the individual series level (Net Spillover: +62.4% across 10 series). Among the four aggregate asset classes, equities are the largest net transmitters (Net: –12.7%), while commodities (+7.0%) and cryptocurrencies (+3.3%) are net receivers. The high TCI of 40.6% confirms that systemic risk propagates broadly across asset classes during our study period.

The mean-variance optimized portfolio allocations under varying concentration regimes are presented in Table 5. The optimization results show a clear mandate for defensive asset allocation as market concentration increases. In a high-concentration regime (HHI > 0.15), the optimal allocation to equities falls by over 20 percentage points compared to a low-concentration environment, while allocations to traditional safe havens like bonds and gold more than double. This highlights the failure of traditional 60/40 portfolio models in the current market structure.

The Sharpe ratio deteriorates from 1.47 in low-concentration periods to 1.18 in high-concentration periods, while the maximum drawdown worsens from −12.8% to −21.3%. This demonstrates that higher concentration is associated with lower risk-adjusted returns for a diversified investor.

Our findings strongly support a proactive macroprudential policy stance:

Macroprudential Oversight: Regulators should consider implementing concentration-based capital requirements for financial institutions. Our results suggest that an HHI threshold of 0.15 could serve as a warning level, with a level of 0.18 triggering mandatory stress testing under high-concentration scenarios.

Index Fund Regulation: Given the role of passive funds in driving concentration, policymakers could mandate enhanced diversification requirements or consider concentration limits for single-security holdings within funds marketed as “diversified.”

Market Structure Reforms: The implementation of circuit breakers triggered not only by price declines but also by rapid increases in concentration metrics could be an effective tool to curb concentration feedback loops during periods of market stress.

Our empirical coefficient for the concentration-tail risk relationship can be conceptualized within a simple risk framework:

Systemic Risk=α+γ×Concentration+ε

where our estimate for the amplification factor γ is 2.31. This shows that concentration has a powerful, nonlinear effect on magnifying idiosyncratic shocks. During stress periods, we find that concentrated assets experience 3.4 times higher bid-ask spreads and a 67% reduction in market depth, confirming the liquidity amplification channel as a key transmission mechanism.

This study provides a robust multi-asset framework for quantifying the concentration-fragility nexus using a comprehensive dataset spanning the COVID-19 crisis and recovery period. A key strength is the integration of VECM and Markov-Switching models, enabling the joint modeling of long-run equilibria and nonlinear regime dynamics.

Limitations include the reliance on daily data, which may not capture intraday liquidity dynamics, and the geographic focus on U.S. markets. Future research could extend the analysis to higher-frequency intraday data or apply the framework to global equity markets to assess cross-country heterogeneity in concentration effects.

The underlying data required to reproduce all findings reported in this article are deposited in the Zenodo open-access repository under a Creative Commons Zero (CC0 1.0) public domain dedication licence, permitting unrestricted reuse. The dataset includes:

The dataset is openly available at the Zenodo repository:

https://doi.org/10.5281/zenodo.19288616 (Restrepo Morales et al., 2026).

The dataset is freely accessible without login requirements. Raw input data were sourced from Bloomberg Terminal, Yahoo Finance, and the Federal Reserve Economic Data (FRED) database, which are subject to their respective terms of use but whose outputs are redistributed here in aggregated, non-proprietary form under Creative Commons Zero v1.0 Universal.

No personal data, patient data, or ethically restricted data are involved in this research.

The authors thank the anonymous reviewers for their constructive comments and suggestions.

•  Adrian T, Brunnermeier MK: CoVaR. Am. Econ. Rev. 2016; 106(7): 1705–1741. Publisher Full Text
•  Allen F, Gale D: Financial contagion. J. Polit. Econ. 2000; 108(1): 1–33. Publisher Full Text
•  Anadu K, Kruttli M, McCabe P, et al.: The shift from active to passive investing: Risks to financial stability?. Financ. Anal. J. 2020; 76(4): 23–39. Publisher Full Text
•  Appel I, Gormley TA, Keim DB: Passive investors, not passive owners. J. Financ. Econ. 2016; 121(1): 111–141.
•  Azar J, Schmalz MC, Tecu I: Anticompetitive effects of common ownership. J. Financ. 2018; 73(4): 1513–1565. Publisher Full Text
•  Beck T, De Jonghe O , Schepens G: Bank competition and stability: Cross-country heterogeneity. J Financ Intermed. 2013; 22(2): 218–244. Publisher Full Text
•  Billio M, Getmansky M, Lo AW, et al.: Econometric measures of connectedness and systemic risk in the finance and insurance sectors. J. Financ. Econ. 2012; 104(3): 535–559. Publisher Full Text
•  Brunnermeier MK, Pedersen LH: Market liquidity and funding liquidity. Rev. Financ. Stud. 2009; 22(6): 2201–2238. Publisher Full Text
•  De Franco C: Stock picking in the US market and the effect of passive investments: C. De Franco. J. Asset Manag. 2021; 22(1): 1–10.
•  Diamond DW: Financial intermediation and delegated monitoring. Rev Econ Stud. 1984; 51(3): 393–414. Publisher Full Text
•  Fink L: Larry Fink’s 2024 Annual Chairman’s Letter to Investors. BlackRock; 2024.
•  Gabaix X, Koijen RS: In search of the origins of financial fluctuations: The inelastic markets hypothesis (No. w28967). National Bureau of Economic Research; 2021.
•  Haddad V, Moreira A, Muir T: When selling becomes viral: Disruptions in debt markets in the COVID-19 crisis and the Fed’s response. Rev. Financ. Stud. 2021; 34(11): 5309–5351. Publisher Full Text
•  Kodres LE, Pritsker M: A rational expectations model of financial contagion. J. Financ. 2002; 57(2): 769–799. Publisher Full Text
•  Koijen RSJ, Yogo M: A demand system approach to asset pricing. J. Polit. Econ. 2019; 127(4): 1475–1515. Publisher Full Text
•  Markowitz H: Modern portfolio theory. J. Financ. 1952; 7(11): 77–91.
•  Pastor L, Stambaugh RF: Liquidity risk and expected stock returns. J. Polit. Econ. 2003; 111(3): 642–685. Publisher Full Text
•  Restrepo Morales JA, Moreno Rodríguez RY, Zea Restrepo F, et al.: The Concentration-Fragility Nexus: Early-Warning Systems and Portfolio Implications in Concentrated Markets. [Dataset]. Zenodo. 2026. Publisher Full Text
•  Yousaf I, Ali S: The COVID-19 outbreak and high frequency information transmission between major cryptocurrencies: Evidence from the VAR-DCC-GARCH approach. Borsa Istanbul Rev. 2020; 20: S1–S10. Publisher Full Text

Current Reviewer Status: ?

VERSION 1

PUBLISHED 18 Apr 2026

Reviewer Report 24 Jul 2026

Katarzyna Czech, Warsaw University of Life Sciences-SGGW, Warsaw, Poland 

Not Approved

VIEWS 0

Reviewer Report 16 Jul 2026

Godfrey Marozva, University of South Africa, Pretoria, South Africa 

Not Approved

VIEWS 0

Reviewer Report 09 Jul 2026

ENE CEZAR CATALIN, Craiova University, Craiova, Romania 

Approved with Reservations

VIEWS 0