Different owner strategies can still share one market exposure

Owner strategy diversity becomes economically meaningful when it survives the observations the agents actually receive. Different mandates can converge on the same instrument, side and decision window if a shared renderer repeatedly supplies the same salient opportunity. We treat this as a population risk question: how much diversity reaches executed exposure after the common harness has shaped every decision?

This question extends our historical continuous record. That research studied two fleets and found selection effects associated with rendered opportunities. The agents had no inter-agent communication channel in the studied setup. Shared behavior therefore provides a reason to investigate common inputs, while intentional coordination would require separate evidence. A population can produce correlated actions through a common environment alone.

Where diversity can disappear

Consider three illustrative owners. One wants short momentum trades, one seeks liquid opportunities and one uses a cautious entry mandate. A common panel highlights an asset whose price and volume have both risen. Each mandate can select that asset for a different reason. Their explanations remain distinct while their exposure overlaps. If the panel refreshes on a shared schedule, their entries may also cluster in time.

The mechanism has three stages. First, the renderer determines the inspected opportunity set. Second, each mandate maps those opportunities into proposals. Third, execution constraints and account states determine which proposals settle. Strategy descriptions sit mainly in the second stage. Common inputs can concentrate the first, and common runtime semantics can concentrate the third.

We can express the distinction with the law of total covariance. For two agents' signed proposed exposures A and B with finite second moments, and a shared observation C, their covariance equals the expected covariance conditional on C plus the covariance of their conditional expectations. The second term can be positive even when the agents act independently after seeing C. Their average responses move together as the common context changes.

This is a mathematical decomposition, rather than an estimated DXRG correlation. It identifies why removing communication leaves a meaningful dependence pathway. It also clarifies why a large count of owner strategies supplies little assurance about fleet diversification by itself.

The population comparison we would run

Our proposed experiment crosses mandate diversity with observation diversity. Keep accounts, market snapshots and permitted actions fixed in an offline replay. Compare a shared candidate renderer against separately assigned renderers whose eligibility rules remain identical. Within each observation condition, compare a repeated mandate against a declared collection of different mandates. Assignment should preserve the information available at each historical cutoff.

The important outcome is the change in concentration at each stage. Measure instrument-and-side overlap among proposals, overlap among admitted orders, and capital-weighted exposure after execution. Raw counts can make a fleet of many tiny positions look similar to a fleet dominated by one large account. Report both account-weighted and capital-weighted views, with the studied decision interval explicit.

If diverse mandates lower concentration only under varied observations, the shared renderer limits their expression. If concentration persists across varied observations, common model responses or market conditions become stronger candidate explanations. If proposal diversity disappears during admission, inspect the execution path. These contrasts locate the dependence rather than naming every shared action herding.

The next economic question concerns crowding. Correlated choices can create exposure concentration while remaining too small to affect venue prices. Demonstrating market impact requires order sizes, liquidity and counterfactual execution evidence. The historical directional-edge null also remains intact: behavioral dependence supplies no return advantage.

For a builder evaluating an agent population, we would ask for the mandate distribution and the observation distribution together. Owner variety is a useful input to diversification. The measured result is how much variety remains after common context, action validation and capital weighting have done their work.