Financial Instability as a Structural Property

Every account of a crash has a proximate cause, and the proximate causes are usually unremarkable: a mid-sized default, a repricing in a market that few people watched, a fund that could not meet a margin call. Similar events happen regularly without consequence. The interesting variable is not the trigger but the condition of the system when the trigger arrives.

Endogenous fragility

Financial systems build up fragility during calm periods, and they do it through mechanisms that look prudent at the level of each individual institution. Low measured volatility reduces risk metrics, which permits more leverage on the same capital. Rising asset prices increase the value of collateral, which permits more borrowing against it. Both are individually rational responses. Together they mean the amount of leverage in the system is highest precisely when measured risk is lowest.

This is why the volatility paradox is not a paradox at all once you model it: the risk measure and the risk-taking are coupled, so the measure stops being informative exactly when you need it. No external shock is required to reach this state. The system walks there on its own, which is what "endogenous" means in this context.

Fire sales and the mechanism of amplification

The amplification loop is mechanical. An institution takes a loss and breaches a leverage constraint. To restore the ratio it sells assets. If it is large relative to the market, or if many institutions hold the same assets and breach at the same time, the selling moves prices. Lower prices mean further losses for everyone holding those assets, which triggers further constraint breaches and further selling.

The crucial feature is that this loop runs on overlapping portfolios, not on direct contractual links. Two institutions with no exposure to each other can be tightly coupled if they hold the same assets and face the same constraints. Regulators who mapped only contractual exposure systematically underestimated how connected the system was.

What network models contributed

Modelling the interbank system as a network produced results that were not available from institution-level analysis:

• Position beats size. A moderately sized institution sitting between two otherwise disconnected parts of the network can matter more than a much larger one at the periphery. This is the analytical basis for designating institutions as systemically important on structural grounds.
• Connectivity has two regimes. Adding links initially spreads losses thinly and makes the system more robust. Past a threshold, the same connectivity spreads them everywhere at once. The relationship is not monotonic, which means "more integration is safer" and "more integration is riskier" are both true, in different ranges.
• Diversification differs from diversity. If every institution diversifies into the same portfolio, each is individually safer and the system is far more fragile, because every balance sheet now responds identically to the same shock.

Why forecasting the date is the wrong goal

These models are structural, not predictive. They identify states of elevated vulnerability - high leverage, compressed spreads, concentrated overlapping exposure - in which the arrival of an ordinary disturbance has a high probability of producing a large outcome. They do not tell you when the disturbance arrives, and models claiming otherwise should be treated with the scepticism described in validation and calibration. The distinction matters: a fire-risk assessment is useful without predicting the date of the fire.

The behavioural layer

Balance-sheet mechanics explain amplification but not the initial build-up of exposure. That requires something about how people form beliefs under genuine uncertainty, where probabilities are not available and narratives substitute for them. When a narrative is widely shared, positions become correlated across institutions that never coordinated - which returns the problem to how beliefs spread through a network, and to the non-equilibrium framing that permits an economy to be in a state nobody chose.