Failure to Account for Correlation Breakdowns - Detailed Explanation

Correlation Breakdown is the phenomenon where historical correlation coefficients between assets, strategies, or risk factors suddenly move to $+1$ or $-1$ under stress.

In normal markets: Your model says Equity and High-Yield Credit correlation is 0.3, so holding both diversifies risk. Your Value-at-Risk (VaR) looks low.

In stress: Flight to quality, margin calls, and forced deleveraging make everything sell off together. Correlation jumps to 0.9-1.0. Diversification disappears, VaR explodes by 5x-10x, and hedges fail exactly when you need them.

This is not a tail event - it is a stylized fact of markets. Regulatory guidance explicitly calls it out:

Basel II/III, SR 11-7 Model Risk Guidance, and FCA SYSC: Firms must not rely on historical correlation alone. Must stress-test correlation breakdowns, wrong-way risk, and model risk.

Why Models Fail to Account for It

1. Gaussian Copula Assumption: The pre-2008 CDO pricing model assumed mortgage defaults were largely independent. It treated correlation as a stable, low input.
2. VaR Based on Short Window: VaR models use 1-2 years of data. They perceive securitization as reducing correlation and diversification.
3. Over-reliance on Historical Convergence: LTCM's convergence trades assumed historical relationships between securities would continue. When Russia defaulted, the historical correlation between Russian and Mexican default was assumed to be zero, but it went to 1.
4. Liquidity and Crowded Trade: In stress, all leveraged players sell the same asset to meet margin.

Major Case Laws & Enforcement Actions

1. Long-Term Capital Management (LTCM) - 1998

The textbook correlation breakdown failure.

Facts: LTCM ran highly leveraged convergence trades - e.g., long on-the-run vs short off-the-run bonds, assuming spread correlation of 90-95%. Models were built on 100 years of data.

After Russian default in August 1998, flight to liquidity caused all risk spreads to widen. Correlations that were assumed to be uncorrelated converged to 1.. The portfolio benefit of diversification vanished and large losses occurred.

Legal / Regulatory Outcome: No direct securities fraud case because LTCM was unregulated hedge fund, but it led to Federal Reserve Bank of New York organized $3.625B bailout . It is cited by SEC, Fed and BIS as the foundational case for:

• Model Risk, Unexpected correlation or the breakdown of historical correlations, Need for stress-testing
• Resulted in Basel II counterparty risk and stress-testing reforms.

2. The 2008 Crisis - CDOs and Gaussian Copula

Failure: Ignoring correlation of defaults

Facts: Rating agencies and banks treated mortgage defaults as largely independent events . Using David Li's Gaussian Copula, CDOs were built on a flawed understanding of correlation risk - the models underestimated potential for widespread defaults .

During stress, assets previously negatively correlated with equities, such as commodities and real estate, began to move in tandem . Diversification benefits were illusory.

Case Law:

a) SEC v. Goldman Sachs & Co. and Fabrice Tourre - ABACUS 2007-AC1 (2010)
Goldman created CDO ABACUS, referencing subprime mortgages, while Paulson & Co. - a party with adverse interests - played significant role in selecting portfolio. Goldman failed to disclose that correlation assumptions were being driven by a short party. Goldman settled for $550M, largest SEC settlement at that time. Tourre was found liable for securities fraud.

Legal Principle Established: Failure to disclose material facts about correlation construction and adverse selection in CDO portfolio violates Section 17(a) of Securities Act and Section 10(b) of Exchange Act.

b) In re Citigroup Inc. Shareholder Derivative Litigation, 964 A.2d 106 (Del. Ch. 2009)
Shareholders sued directors for failing to monitor correlation risk in CDOs. Court applied business judgment rule but noted that while directors were not liable for bad business outcome, failure to implement any system to monitor correlation breakdown and super-senior CDO exposure could be a breach of duty of care.

c) ACA Financial Guaranty Corp v. Goldman Sachs
Insurer sued Goldman over ABACUS, arguing Goldman misrepresented correlation risk modeling. Court allowed fraud claims to proceed.

3. JPMorgan Chase - The London Whale (2012)

Direct regulatory finding of model manipulation to hide correlation breakdown.

Facts: Chief Investment Office's synthetic credit portfolio grew to $157B in Q1 2012 . The portfolio was meant to hedge, but internal VaR model was changed from VaR to a new model that understated risk by 50%. Correlations between IG and HY indices broke down.

Traders disregarded risk limits, manipulated risk assessment models, and hid losses .

Enforcement:

a) In re JPMorgan Chase & Co., SEC Order Instituting Cease-and-Desist Proceedings (Sept 19, 2013)
SEC found JPMorgan misstated financial results and lacked effective internal controls to prevent traders from fraudulently overvaluing investments . Violated securities laws by filing false books. Settlement: $200M to SEC, $300M to OCC, $100M to CFTC, total $920M with UK FCA .

b) FCA Final Notice: JPMorgan Chase Bank N.A. - £137.6M fine (2013)
FCA found serious failings relating to CIO's London Whale trades , including failure to inform senior management and regulators when correlation breakdown caused losses to mount .

c) CFTC Order (Oct 16, 2013)
Found reckless conduct and violation of Commodity Exchange Act .

Legal Principle: Deficiencies in Model Development and Approval processes allowed traders to trade while underestimating risks - classified as operational risk loss under Basel II . Violates Sarbanes-Oxley internal controls.

4. Archegos Capital & Credit Suisse / Nomura - 2021

Modern correlation breakdown: Total Return Swaps on 5 tech stocks were treated as diversified basket. In reality, they were highly correlated single-name exposure with no disclosure of concentrated, overlapping holdings. Correlation went to 1 on margin call, causing $10B+ losses.

SEC and DOJ charged Archegos founder Bill Hwang with market manipulation and fraud (SDNY 2022). Credit Suisse was found by FINMA to have failed to manage correlation and concentration risk.

Legal Liability Framework

If you are a risk manager, director, or fiduciary, failure to account for correlation breakdown can lead to:

1. Securities Fraud (10b-5): Misstating VaR or risk if you know correlation model is flawed - London Whale precedent.
2. Breach of Fiduciary Duty: In re Citigroup - failure to implement risk monitoring system for correlated exposures is not protected by Business Judgment Rule.
3. Regulatory Action under Model Risk (SR 11-7 / OCC 2011-12): Regulators can fine for inadequate stress-testing of correlation assumptions to 1.0 / -1.0.
4. Professional Negligence: For rating agencies, auditors, and risk consultants who rely solely on historical correlation without stress scenario.

How to Defend / Prevent - Best Practice Required by Courts

Courts and regulators now expect:

• Correlation Stress to 1: Run VaR and Expected Shortfall with correlation = 1 and 0.9 for all risk factors. Loretan & English (2000) methodology is cited by Fed.
• Wrong-Way Risk: Identify when exposure and counterparty default are positively correlated.
• Liquidity-Adjusted VaR: LTCM lesson - assume correlation breakdown coincides with liquidity drying up.
• Governance: Independent Model Validation team must approve any correlation change, as JPMorgan failed to do.

Diversification is not a risk control mechanism. It is a model that operates under a specific mathematical assumption: that instruments exhibit sufficiently low correlations under the conditions in which they are held. When that assumption fails, the entire risk framework fails.