The Shape of a Crash: Why the Speed of a Drawdown Says More Than Its Depth

A crash is not one thing. The word covers a market that loses a meaningful share of its value in a quarter of an hour and recovers most of it by the close, a market that grinds lower for a year as the earnings it was priced for fail to arrive, and a market that falls for three weeks and then turns on the afternoon an official announcement lands. These episodes share a direction and little else. Their speed differs by orders of magnitude, they take liquidity away in different ways, they change the relationships between assets differently, and they recover on different clocks. To a systematic investor the differences matter more than the common label, because the systems that survive one kind of crash are not automatically the systems that survive another.

This piece sets out three archetypes, the fast liquidity crash, the slow fundamental repricing, and the policy-driven episode, and compares their microstructure: how quickly prices move, what happens to the order book, how correlations behave, how volatility evolves, and how the shock travels from one market to the next. The historical references are deliberately qualitative. The 2010 flash crash unfolded over minutes; the bear markets of the early 2000s unfolded over years; the March 2020 episode unfolded over weeks and ended at the moment policy arrived. Every series we plot is a stylized construction built from a simple model stated beneath the figure, chosen to show the shape of each archetype rather than to reproduce any particular date.

Three Archetypes

The fast crash is a failure of the market's plumbing. Nothing about the value of the assets has changed on the timescale of the move; what has changed is that the participants who normally stand ready to buy have stepped away at the same moment, and the orders that arrive find no one on the other side. Prices move faster than information can, which is why a fast crash reverses: once the sellers are exhausted and the buyers return, the price has nowhere to go but back toward where the information says it should be. The episode that made this archetype famous lasted well under an hour from the first sign of trouble to the recovery of most of the loss.

The slow crash is the opposite. Information arrives faster than prices adjust to it, and the adjustment is stretched over months because the information itself arrives in installments: a quarter of earnings, a revision to guidance, a default, another quarter. Liquidity does not vanish. It thins, because the participants who would normally provide it are themselves reducing risk, and it stays thin for as long as the repricing continues. There is no single moment to point to, and no mechanical reason for the decline to reverse. It ends when the news does.

The policy-driven episode borrows from both. It begins like a fast crash, with a shock that outruns the market's ability to absorb it, and continues like a slow one, because the shock has a fundamental component that takes time to assess. What distinguishes it is the ending. An official actor changes the rules, by intervening, by guaranteeing, by suspending, and the path turns on the day, sometimes on the hour. The recovery is a step, not a slope. Figure 1 draws the three paths, each normalized to its prior peak, on a logarithmic time axis so that an episode measured in minutes and one measured in months can be seen on the same chart.

Two features of the picture deserve attention. The first is that depth of loss and speed of loss are only loosely related; in the stylized model the fast crash is the shallowest of the three at its trough and the quickest to recover, while the slow repricing is the deepest and never recovers within the window shown. The second is that the policy-driven path is the only one with a discontinuity in its slope. That kink is the fingerprint of an external decision, and it is what makes the archetype so hard for a purely statistical model to anticipate: nothing in the price history before the intervention contains the information that the intervention is coming.

Where the Liquidity Goes

Price is what a crash looks like from outside. From inside the order book, what changes first is depth: the quantity that can be bought or sold within a given distance of the current price. Depth is a promise made by market makers and other liquidity providers, and it is a promise they can withdraw in microseconds. In a fast crash they withdraw it together, not because they have coordinated but because their models are similar and their risk limits bind at similar moments. The order book does not so much empty as evaporate, and a market order that would have moved the price a few basis points a minute earlier moves it by percentage points.^[1]

Figure 2 traces the visible depth near the touch through each archetype, indexed to its pre-shock level and drawn on a logarithmic vertical axis so that the loss of most of the book and the loss of half of it can both be read. In the fast archetype, depth in the stylized model falls by close to an order of magnitude within minutes and recovers most of the way within the day, with a residue that takes weeks to refill as providers rebuild confidence in their own models. In the slow archetype it thins gradually toward half its former level and stays there. In the policy-driven archetype it falls with each leg of the decline, then refills abruptly when the intervention removes the uncertainty that kept providers away.

The asymmetry between the fast and slow paths is the practical lesson. A fast crash destroys depth on a timescale no human can respond to, but the destruction is temporary, and a system that can distinguish a liquidity event from a fundamental one is rewarded for waiting, and for providing liquidity when others will not. A slow crash destroys less depth but destroys it durably, and the cost of trading in it is not a spike but a tax, paid on every order for as long as the episode lasts. The two require different responses. The first is a problem of latency and of not being the last to notice; the second is a problem of transaction-cost modeling and patience.

Correlation Is a Symptom

Investors speak of correlations rising in a crash as though correlation were a force. It is a symptom. Correlation rises when a single factor, the need to reduce risk, comes to dominate the returns of assets whose ordinary drivers are unrelated. Which assets join the common move, and which stay outside it, depends on why the risk is being reduced, and this is where the archetypes separate most cleanly.^[2]

Figure 3 shows the correlation between selected asset pairs in a calm baseline and in each archetype, generated from a one-factor model in which every asset has a loading on a common stress factor and the loadings are set per regime. In the fast archetype the loadings are high across nearly everything, including assets that ordinarily act as shelter: when the driver is the need for cash, whatever can be sold is sold, and gold and government bonds are sold alongside equities. Only the currency in which the cash is needed moves the other way. In the slow archetype the risk being reduced is exposure to growth, and the traditional diversifiers work as advertised, because falling growth expectations pull rates down and pull the safest bonds up. The policy-driven episode sits between the two until the intervention arrives, at which point the common factor loses its grip and the correlations break, often within a session.

The point is not that any one of these columns is right. It is that a single estimate of correlation, however carefully measured, describes at most one archetype, and a portfolio built to be diversified under that estimate is diversified against that kind of crash and not necessarily against the others. The useful question is not what the correlation is but what it becomes under each of the ways the market can break, and a portfolio that answers all three tolerably is more robust than one that answers any one of them precisely.

How a Shock Travels

Crashes propagate, and they propagate along the channels that connect markets: arbitrage relationships, shared liquidity providers, common risk models, and the funding arrangements that turn a loss in one asset into a forced sale in another. What the archetypes differ on is the order in which those channels carry the shock. Figure 4 lays out the sequence for a fast episode. The shock lands in the deepest and fastest market, usually index futures, because that is where large risk is transferred first. Arbitrage carries it into exchange-traded funds and the cash market within milliseconds. Liquidity providers, seeing a move they cannot explain, withdraw, and the single-stock order books thin until trades print at stub quotes far from any sensible value. Volatility reprices and hedging flows add to the selling. Systematic strategies with volatility targets and rules-based risk limits reduce exposure across asset classes at once, and margin calls extend the selling to whatever the affected participants hold. Only at the end of the chain do fundamentals or policy enter, and by then the fast episode is usually over.

The slow archetype runs the same chain in roughly the reverse order. It begins in fundamentals, reaches the risk models and the funding arrangements over weeks, and arrives at the order book last, as a gradual thinning rather than a collapse. The policy-driven episode runs the fast chain first and then waits, sometimes for weeks, for the final link to be pulled by someone other than the market. Volatility follows the same logic. In the fast archetype it is extreme and brief, a spike that the options market prices in and out within days. In the slow archetype it is elevated for months and arrives in waves, each timed to a fresh installment of bad news. In the policy-driven archetype it is high until the announcement and collapses on it, which is why the days around an intervention are as dangerous for anyone short volatility as the days before it were for anyone long risk.

Reading the Shape

Figure 5 collects the comparison. The columns are the dimensions along which the archetypes differ most, and the entries are qualitative, because the differences are of kind rather than degree, and a number attached to any of them would describe one episode rather than the class.

Classifying a drawdown while it is happening is harder than classifying it afterward, and the classification matters, because the right response to one archetype is the wrong response to another. Supplying liquidity is what ends a fast crash; supplying it into a slow one is a way to lose money in installments. Reducing risk on the first day of a policy-driven episode and reducing it on the day of the intervention are opposite trades with the same name. The observable that separates the archetypes soonest is not the size of the move but its microstructure: how fast depth is disappearing relative to how fast prices are moving, whether the assets joining the decline are the ones a cash shortage would sell or the ones a growth shock would sell, and whether the volatility surface is pricing an event or a process.

This is how we think about crashes at Oak St.: as shapes rather than sizes. We ask of any strategy how it behaves under each archetype separately, we treat the order book and the cross-asset correlation structure as the earliest reliable signals of which one is under way, and we build systems whose responses are conditional on that reading rather than on the size of the loss alone. None of this makes a crash less costly when it comes. It makes the response a decision rather than a reflex, and in an episode that lasts minutes, or in one that lasts a year, that is the difference that can be prepared for in advance.^[3]