Dissecting Mechanisms of Financial Crises: Intermediation and Sentiment

Financial crises have a recognizable pattern. Before the crisis, credit expands, asset prices look strong, risk spreads are low, and the economy often appears healthy. Then the transition is abrupt: asset values fall, intermediaries pull back, credit contracts, and output declines. The recovery is slow. The paper asks what combination of mechanisms can explain the whole cycle, not just the crash.

Two explanations dominate discussions of financial crises. One emphasizes financial intermediation. Banks and other intermediaries take on leverage, and when losses hit, their reduced balance sheet capacity amplifies the shock. This view explains why crises are severe and why recovery can be slow. The other explanation emphasizes sentiment or beliefs. Investors become too comfortable after a long good period, underestimate risk, and then suddenly reassess when bad news arrives. This view explains why markets can look frothy before the crisis.

Our paper shows that both mechanisms are needed. A model with financial frictions can generate amplification during the crisis and persistence afterward, but it struggles to explain why the pre-crisis period looks so optimistic: low spreads, high credit, and strong asset prices. A model with time-varying beliefs helps explain that froth. When investors go a long time without seeing an illiquidity event, they come to view such events as less likely. Intermediaries then take on more liquidity risk. If a bad shock arrives, beliefs change and the fragile financial structure amplifies the damage.

The model has two central state variables. One captures the financial strength of intermediaries. The other captures beliefs about the likelihood of an illiquidity state. An illiquidity state is a situation in which markets freeze up or funding becomes hard to roll over. When the financial sector is fragile and investors are complacent about liquidity risk, the economy is vulnerable. The crisis is not simply a big productivity shock; it is a financial event whose effect depends on the balance sheet and belief state before the shock.

A key contribution is that we confront the model with facts across the entire crisis cycle: pre-crisis asset markets and credit, the sharp crisis transition, and post-crisis output recovery. The model with both intermediation and belief variation gives a parsimonious account of these patterns. The paper also compares Bayesian learning and diagnostic, or overreactive, beliefs. Both can match broad patterns, although they differ in some quantitative dimensions.

For policy, the paper has an important message. Regulators do not need to know exactly what is inside investors' minds to see when the system is vulnerable. If leverage, credit growth, spreads, and intermediary balance sheets indicate fragility, then lean-against-the-wind policies can be evaluated using observable financial conditions. The paper finds that policy effects can be similar across different belief models once the same observed state is matched.

The takeaway is that crises are not caused only by bad luck or only by irrational exuberance. They arise when financial structure and beliefs interact. A calm period can make the system more fragile because it encourages balance sheets to expand and liquidity risk to be underpriced. When the shock finally arrives, the same structure that looked efficient in good times becomes an amplifier.

This perspective is useful for thinking about modern financial stability. Low spreads and abundant credit are not automatically signs of safety. They can also be signs that risk has migrated into tail states. The challenge for policy is to recognize when market confidence is being supported by fragile intermediation rather than by genuine resilience.