Patterns of Consensus Failure

We estimate probability by how easily examples come to mind. The most available precedent — the last crisis we lived through — becomes the template for the current one, regardless of structural differences. Availability is not the same as relevance.

Representativeness bias causes us to match current situations to familiar categories based on superficial features. The IMF classified the Asian crisis as a Latin American-style fiscal problem. The Fed classified 2021 inflation as a demand-pull problem. Both were wrong because the surface matched but the structure didn't.

Humans are story-processing machines. A coherent narrative — Japan as Number One, the New Era of productivity, ESG companies are better run — feels explanatory and is socially rewarded. The narrative substitutes for the model. When the story is good enough, the data is adjusted to fit it rather than the reverse.

Once an anchor is set — a stock price, a GDP forecast, a rate path — subsequent estimates adjust from it but insufficiently. In financial crises this produces systematic underreaction in the calm phase and systematic overreaction in the panic phase. The anchor switches from optimistic to catastrophic, and adjustment from the new anchor is again insufficient.

The 'inside view' focuses on the specific features of the current situation — our bank, this country, this asset class. The 'outside view' asks: of all situations that looked like this, what was the distribution of outcomes? Institutions almost universally use the inside view because it feels more relevant and produces more confident forecasts.

For an institutional investor or analyst, being wrong with the consensus is survivable. Being wrong against the consensus is career-ending. This asymmetry produces systematic herding — not because individuals don't see the risk, but because the institutional incentive is to be wrong in company. Keynes called it: 'Worldly wisdom teaches that it is better for reputation to fail conventionally than to succeed unconventionally.'

Motivated reasoning is System 1 producing a desired conclusion and recruiting System 2 to rationalize it. In financial institutions it is structurally embedded: the conclusion that serves the institution's interests is the one that gets the best reasoning. Chuck Prince had to dance because Citigroup's revenue depended on it. Arthur Burns accommodated Nixon because the Fed's institutional independence depended on presidential goodwill.

Moral hazard occurs when one party takes risks while another bears the consequences. Deposit insurance protects depositors — which is its purpose — but also removes their incentive to monitor their bank's risk-taking. The combination of upside capture and downside socialization is not an accident; it is the structural consequence of incomplete risk transfer. It accumulates silently until the losses crystallize.

Fund managers are judged quarterly. Politicians face election cycles. Executives earn bonuses annually. The risks that kill institutions — duration mismatch, leverage accumulation, credit cycle deterioration — play out over years or decades. The incentive horizon is structurally shorter than the risk horizon. This is not irrationality; it is rational response to the tournament structure of institutional finance.

Correct early warnings are systematically dismissed not because they are wrong but because they are early. The costs of acting on a warning are immediate and certain; the benefits are future and probabilistic. For an institution facing tournament competition, an early warning that doesn't materialize immediately is a performance drag. The result: correct warnings are suppressed until the crisis is no longer avoidable.

Every institutional crisis playbook was written for a slower information environment. Bank runs required physical presence — there were natural friction limits. In the social media era, information moves at the speed of a tweet and capital at the speed of a mobile app. The 2023 SVB bank run was the first Twitter bank run. The playbooks assumed hours or days; the reality was minutes. Institutions are structurally unable to adapt faster than the information that kills them.

Risk models are built on historical correlations measured during normal times. In a crisis, the mechanism that drives correlations to zero — independent fundamental analysis — is replaced by a single mechanism: liquidity. When everyone is selling to raise cash, all assets fall together regardless of their fundamental independence. The diversification that looked robust in the model disappears precisely when it is needed most.

Complex adaptive systems respond to interventions. Market participants adjust their behavior based on the new environment created by the intervention. The Greenspan Put taught markets that the Fed would cut rates on any significant decline — producing the leverage accumulation that required QE. Zero rates after the GFC produced the asset price inflation that required aggressive hiking. The intervention is rational; the second-order response is predictable in direction but not in timing.