Why markets crowd: how a shared yardstick pushes everyone into the same trade

What is it about?

For quants the phenomenon is familiar even when its source isn't: the violent unwind. Strategies that look uncorrelated in normal conditions turn out to be the same trade in different clothing, so when the first players deleverage, the exit becomes a stampede — forced selling pushes prices against everyone still holding, triggering more forced selling in a self-reinforcing liquidity spiral. The August 2007 quant event is the archetype. This paper's move is to locate the hidden correlation upstream of the usual suspects. It isn't imitation, shared data feeds, or a common signal library. Tt argues that when many strategies optimise against the same objective structure — the same benchmarks, risk metrics, and definitions of a good result — they converge on the same positioning even when built independently and sharing no code. The shared evaluation layer is itself the crowding mechanism. The book looks diversified at the strategy level while being dangerously concentrated at the level of what those strategies are all trying to satisfy — and that concentration is invisible to any risk model that inspects positions or strategies one at a time rather than the objective they hold in common.

Featured Image

Photo by Yashowardhan Singh on Unsplash

Photo by Yashowardhan Singh on Unsplash

Why is it important?

The practical sting for a risk desk is that the diversification you think you have may not exist. If independent strategies crowd because they share an objective, then correlation estimated from calm-period returns systematically understates tail risk — the strategies decorrelate on the way up and re-correlate violently on the way down, exactly when it hurts. A covariance matrix built on quiet data prices this risk at zero. Worse, the exposure is invisible to the standard tools: inspecting each strategy, model, or book in isolation shows nothing, because the shared dependency doesn't live in any one of them — it lives in the objective they all answer to. The paper's argument implies crowding must be monitored one level up, at the objective structure itself — the benchmarks, risk metrics, and reward definitions many desks hold in common — rather than at the level of positions or signals. And the exposure is growing. As machine-learning and increasingly automated strategies proliferate, more of the market optimises against a shrinking set of shared objective structures — sometimes literally the same benchmarks, factor definitions, or foundation models. That makes this a systemic-risk question, not just a single-fund one: the shared evaluation layer is becoming a common dependency across the whole market, and a common dependency is precisely what turns many independent bets into one correlated position waiting to unwind.

Perspectives