Liquidity vigilance: what the last three drawdowns rhymed on.

We keep a standing internal file we call the "rhyme sheet" — the recurring structural features across drawdowns that were technically distinct events but broke portfolios in the same shape. It's built off February 2018 (vol-selling unwind), March 2020 (Covid liquidity shock), and March 2023 (regional bank contagion / rates repricing). Three different triggers, three different macro narratives, and yet the mechanics of who actually lost money — and why — were close enough to the same pattern that we now treat the pattern, not the trigger, as the thing worth underwriting against.

The trigger is never the lesson

It's tempting to build risk process around the last crisis's specific trigger — short-vol positioning after 2018, illiquid credit after 2020, duration mismatch after 2023. That's a losing game, because the next trigger is by definition something the last cycle's process wasn't built to see. What's useful isn't the trigger inventory. It's the transmission mechanism, and across all three events the transmission mechanism was liquidity, specifically the gap between:

• —the liquidity a position was priced and risk-managed as if it had, and
• —the liquidity it actually had when everyone wanted out at the same time.

That gap is the single most persistent thread across the three events, and it's the thing a standing risk office is built to monitor continuously rather than discover after the fact.

Rhyme 1: liquidity is a correlation, not a constant

In each event, instruments that traded independently in calm markets converged to trading as one liquidity factor under stress. In Feb 2018, short-vol ETPs and their underlying futures book, which had looked like uncorrelated instruments serving different investor bases, collapsed into a single redemption-driven feedback loop within a trading session. In March 2020, investment-grade credit — theoretically among the most liquid corporate credit — traded at bid-ask spreads and price dislocations more associated with high yield, because everyone holding "liquid" credit tried to sell it at once and dealer balance sheets weren't there to absorb it. In March 2023, regional bank equity and short-duration Treasuries, instruments with essentially no fundamental linkage, moved together because both were being sold by the same forced sellers into the same thin dealer book.

The risk-office implication: liquidity-adjusted VaR built on trailing average bid-ask spreads and trading volumes is measuring calm-market liquidity, which is close to useless for the tail scenario it's nominally meant to price. What we monitor instead is a stressed-liquidity multiplier by instrument class — an explicit haircut to both spread and depth calibrated off the worst historical liquidity conditions for that instrument type, refreshed against each new episode, applied on top of the calm-market number rather than instead of it.

Rhyme 2: crowding hid inside "diversified" books

Each event caught managers who believed their books were diversified across names, sectors, or even strategies, but who were secretly expressing a single crowded factor bet at the position-sizing level. Before Feb 2018, "sell volatility" was expressed through a dozen different instruments and strategies that all had the same tail behavior. Before March 2020, "reach for yield in investment-grade and BBB credit" was a crowded trade across managers with otherwise unrelated equity books. Before March 2023, "own regional bank equity and hold-to-maturity duration risk on the balance sheet" was crowded within the sector even among institutions with no direct linkage to each other.

The rhyme here isn't about any one factor — it's that position-level diversification and factor-level diversification diverge exactly when it matters most, and standard correlation matrices built on 60–90 day trailing windows systematically understate that divergence because crowded trades look uncorrelated in calm regimes and only reveal their true correlation under stress. A risk office earns its keep here by running crowding checks that are regime-aware rather than trailing-correlation-based: what fraction of the book's risk, under a stress scenario correlation matrix rather than the calm one, comes down to a single factor.

Rhyme 3: the exit was gated, throttled, or gapped — never smooth

In each case, the actual mechanism of loss for managers who got hurt wasn't that the position moved against them slowly and they chose to hold. It's that the exit itself became unavailable or radically repriced at the moment they wanted it. ETP unwind mechanics gapped in 2018. Credit ETF NAV-to-price dislocation and outright market closures in parts of the credit market defined March 2020. Deposit and repo market freezes, and forced asset sales at a loss to meet redemptions, defined the bank failures in 2023.

This is the strongest argument for why liquidity risk needs to be modeled as a distribution of exit costs under stress, not a point estimate of current bid-ask spread. The relevant risk-office deliverable is a liquidation cost curve per position, run under progressively worse assumptions about available depth and time-to-exit, cross-referenced against the fund's actual redemption terms and gate provisions — because the liquidity a manager needs is a function of their liability structure, not an abstract property of the assets.

What this means for how we build the monitoring stack

Concretely, our standing liquidity framework for a client book runs on four legs, refreshed continuously rather than reconstructed after the fact:

1. Stressed liquidity haircuts by instrument, calibrated against the worst observed historical conditions for that instrument class, not trailing averages — updated whenever a new stress episode adds data. 2. Factor crowding under a stress-regime correlation matrix, run in parallel with the standard trailing-window matrix, specifically to surface the divergence between the two. 3. Liability-matched liquidation curves — how long it would actually take to raise a given fraction of the book's cash needs under stress, mapped against the fund's actual redemption and gate terms, not a generic assumption. 4. A standing scenario library that includes all three of these episodes explicitly, re-run on the current book on a fixed schedule, so the question "how would this book have performed in March 2020's liquidity conditions" always has a current, dated answer rather than a hypothetical one produced only when someone thinks to ask.

The honest caveat

None of this predicts the next trigger. We don't know what the fourth event's proximate cause will be, and anyone who claims their framework does is overselling it. What this buys a manager and their allocators is a book whose liquidity risk has been priced against the actual historical behavior of liquidity under stress rather than its calm-market appearance — which is the one thing that has been true, in some form, across every liquidity-driven drawdown of the last decade, regardless of what set it off.

The rhyme isn't the trigger. It's the gap between assumed and realized liquidity, and it's measurable in advance if someone is actually looking for it on a standing basis rather than reconstructing it afterward.