Riccardo Rebonato on Risk Management and the Crisis

Episode Summary

Executive Summary: Russ Roberts and Ricardo Rebonato examine how pre-crisis banking relied on quantitative models, value-at-risk, ratings, and Basel rules that gave a false sense of precision. Rebonato argues that risk estimates were overconfident, incentives pushed banks toward leverage and short-term returns, and regulators often underestimated systemic risk and liquidity breakdowns.

Main Topics: Limits of quantitative risk models: Rebonato argues banks and regulators overtrusted models calibrated on too little and too recent data to estimate rare, crisis-level losses. Value at Risk and capital regulation: They discuss how VaR became popular because it translated risk into dollars, but it was used beyond its proper horizon and supplemented with multipliers that still relied on shaky assumptions. Economic capital, shareholders, and bondholders: The conversation explores the tension between holding enough capital to satisfy creditors and preserving returns for equity holders, with banks gravitating toward a compromise rating level. Regulatory arbitrage and Basel II: Basel II aimed to align capital with risk, but its complexity and model dependence created openings for banks to optimize around the rules rather than reduce true risk. Ratings agencies and securitization: Rebonato criticizes the conflict of interest in issuer-paid ratings and the way structured products were engineered to achieve high ratings through cooperative model fitting. Systemic risk, too big to fail, and intervention: The hosts debate whether private actors internalized bailout expectations and whether regulators should have constrained risk-taking earlier to prevent contagion. Housing bubbles and global liquidity: The transcript closes by arguing the crisis was a perfect storm of global capital flows, cheap liquidity, leverage, and housing speculation, not a single U.S.-only cause.

Key Arguments: Rare-event models are unreliable because the data available are too limited and too regime-specific to support claims about once-in-1,000- or once-in-4,000-year losses. VaR is useful for routine trading volatility but is mismatched to solvency and capital planning, which require protection against much rarer tail events. Banks settled around a double-A-type capital compromise because stronger buffers would reduce equity returns and weaken competitiveness. Shareholders could rationally tolerate more risk than regulators because shareholders capture upside while systemic downside is borne partly by society. Institutions that entered structured-credit and housing-related trades later were often hurt more because they joined after the easiest profits had already been made. Basel II improved on Basel I but still relied too heavily on statistical modeling and encouraged regulatory arbitrage. Ratings agencies had structural conflicts because they were paid by issuers and helped structure deals to obtain desired ratings. Mark-to-market accounting is not itself the cause of losses, but in illiquid markets it can accelerate forced sales and distort balance-sheet signals. Too-big-to-fail expectations may have mattered more at the institutional level than at the level of individual executives, who personally faced ruin and humiliation. The crisis was global and driven by multiple factors: compressed risk premia, abundant liquidity, trade surpluses, leverage, and housing demand across several countries.

Data Points: 99.975 percentile one-year horizon: once every 4,000 years - Rebonato cites the industry-style economic-capital target used to signal extreme solvency strength. 95th percentile one-day VaR: $20 million - Used as an example of a simple VaR estimate meaning losses should exceed that amount only 5 days out of 100. VaR 95th percentile frequency: 1 in 20 days - Illustrates the everyday-trading interpretation of VaR. VaR 99th percentile frequency: 1 in 100 days - Shows how the statistic remains focused on relatively short horizons. Capital multiplier: 3x - Regulators applied a safety multiplier to bank VaR estimates under Basel-style rules. Engineering safety factor example: 10x for steel ropes; 1.1x for aircraft landing gear - Used as analogies for why regulators add conservative buffers beyond model output. LTCM/Russia default shock: 1998 - Presented as a contained financial-sector event rather than a real-economy crisis. System turning point: end of 2006 - Rebonato identifies this as the major sea change before the crisis fully unfolded. Conversation date: May 22, 2009 - The episode date when Roberts and Rebonato discuss the aftermath of the crisis.

Pivotal Quotes: "we really are not in a position of saying anything about extremely rare events" — Ricardo Rebonato: He explains why tail-risk models are unreliable given the small amount and limited relevance of available data. "the regulator's goal is to prevent insolvency or disaster because if it spreads, that's the entire financial system could be at risk" — Ricardo Rebonato: Roberts clarifies, and Rebonato agrees, on why bank regulation must consider systemic effects, not just firm-level risk. "if I didn't have a thermometer, this patient wouldn't have a fever" — Ricardo Rebonato: His analogy for why mark-to-market accounting may reveal volatility but does not create the underlying problem.

Implications: Risk management must be humble about model limits, capital rules must account for liquidity and systemic spillovers, and investors/regulators should expect crises when leverage and incentives align. Better supervision means less faith in precision and more attention to buffers, incentives, and contagion.