Risk Builds in Calm, Not in Storms

The largest events in financial history are usually described afterward as sudden. They rarely are. What is sudden is the resolution. The accumulation that produced the vulnerability took place over the preceding months or years, in plain sight, through decisions that each looked prudent at the moment they were taken. The trigger, when one is identified, is usually trivial compared to the size of the move it appears to have caused.

The mechanism is straightforward once it is named. Low trailing volatility supports higher leverage under any risk model that uses volatility as its central input. Higher leverage produces higher returns during the continuing calm, which reinforces the impression that the strategy is working and encourages further leverage. Positions become crowded because the same signal produces the same decision across independent participants. The system reaches a configuration in which any moderate perturbation forces a response larger than the original perturbation, and the response then becomes the story.

The proof case that keeps recurring

The 2018 Japanese government bond market experienced a six per cent move against positioning that had been sized for a trailing volatility reading of roughly one per cent. Under the risk framework in use at the time, the sizing was authorised because the trailing volatility supported it. The move that arrived did not exceed any historical distribution; it was a six-sigma event only against the trailing sample the model had access to, and a rather ordinary event against the full distribution of Japanese bond moves over longer histories.

What broke was not the market. What broke was the assumption that recent calm was evidence of underlying safety rather than evidence of accumulated positioning against the possibility of a move. The same pattern has recurred in credit markets, in equity volatility, in cross-currency basis, and in the 2019 repo episode. Each event carries its own local specifics. The underlying arithmetic is the same across all of them.

Why the standard framework misreads calm

Volatility-based risk measures are procyclical by construction. Falling volatility authorises higher positions. Rising volatility forces smaller ones. The framework therefore produces its largest positions in the periods immediately preceding the largest events, and its smallest positions in the periods immediately following them. The framework is behaving exactly as designed. The design is calibrated on the assumption that recent volatility is informative about near-future volatility, which is approximately correct on average and severely wrong at exactly the moments that matter for portfolio survival.

Adjusting the framework to read calm as accumulation risk rather than as safety produces a different sizing discipline, one that caps position sizes in absolute terms rather than in terms of a trailing volatility measure. The trade-off is that the capped framework foregoes some of the return available during genuine periods of calm, in exchange for structurally limiting the sensitivity to the calm periods that end badly. Whether the trade-off is worth taking depends on the size of the tail events the portfolio would otherwise be exposed to, which is exactly the variable the volatility framework is calibrated not to see.

What this implies for household balance sheets

The pattern reproduces at the household level. Two decades of stable Australian employment supported two decades of accumulated household leverage against residential property. Each borrowing decision was reasonable at the moment it was taken. The trailing environment supported it. Risk models used by the banks that provided the credit read the trailing volatility of household income and property values as evidence of safety.

The system has therefore reached a configuration in which the sensitivity of the household sector to any moderate shock (a rate rise, a labour market softening, an income disruption) is meaningfully higher than the trailing volatility of household indicators would suggest. This is not a forecast that the shock arrives. It is an observation about the state of the system, which is the useful variable regardless of whether any specific shock is anticipated.