What the desk declines: a register of forty-five rejected strategies

Quantitative Strategies has reviewed 64 strategy proposals since the division was formalised in 2022. Nineteen are in production. Forty-five were declined. The declined proposals are the more useful record, because a strategy that fails in review costs a few weeks of work and a strategy that fails in production costs capital and a conversation with a client. Every declined proposal stays in a register with its reason attached, so the same idea does not return in a new wrapper two years later.

Fourteen were declined on capacity. A strategy that cannot absorb US$15 million without moving the price it depends on is not a strategy the division will run, whatever its record in a sample. Capacity is measured before performance, and it is measured against the market as it is rather than as it was during the test window. Three of the fourteen came back at a smaller size and were declined again, because the fixed cost of running a strategy does not shrink with it.

Eleven were declined on decay. The division requires a decay half-life above nine months, measured out of sample, before it will fund the infrastructure a strategy needs. An edge that halves every four months is a research result rather than a business. This test removes most of what arrives from outside the firm, because an edge that survives circulation is rare, and an edge shown to several institutions is priced by the time anyone implements it.

Eight were declined because a dataset had no documented origin or lawful basis, which is the third Charter principle applied to research rather than to reporting. Seven were declined because modelled execution cost exceeded half the gross edge. The second of those is the most common error in submitted work. A test priced at the midpoint of a spread is not slightly wrong in a thin market. It is wrong by the whole of the answer it produces.

Three were declined because they inferred behaviour about identifiable individuals. The fourth Charter principle limits inference to what a mandate requires, and a mandate to trade a market does not carry a mandate to model the people inside it. Two were declined because nobody on the desk could explain them to a client in plain language. That is not a test of elegance. A client entitled to require a human decision is entitled to a human explanation.

The register is reviewed by the Risk and Valuation Committee twice a year and by the Ethics and Technology Council once. It is not a graveyard. Four strategies declined on capacity in 2023 were revived in 2025 when the underlying market deepened, and two of those now run in production. What the register prevents is the quiet return of an idea whose reason for rejection has not changed, which is the most expensive research a desk can conduct.

Published 2026-03-17 by the Quantitative Strategies division. Research is prepared for eligible counterparties and does not constitute advice.