The Researcher-PM Contract: Who Decides What

A researcher builds a signal, the backtest looks good, and it gets handed to a portfolio manager to trade. Six months later the PM has cut its weight in half, added a volatility filter the researcher never tested, and stopped using it on Fridays — and the researcher has no idea why, or whether any of it is a problem. This friction is common enough that it's less a personality clash and more a missing document: nobody wrote down which decisions belong to whom.

The idea

A researcher and a PM are optimizing for related but not identical things. The researcher usually owns signal quality — is the underlying edge real, is it correctly measured, is it likely to persist. The PM usually owns capital allocation — how much to bet on this signal relative to everything else in the book, how to size it given the portfolio's overall risk budget, and when to override a signal based on context the backtest never saw, like a name going into an index rebalance or an upcoming binary event. Without an explicit contract, both sides end up making decisions that belong to the other: a researcher who insists on a specific position size is quietly doing the PM's job, and a PM who redesigns a signal's inputs without looping in the researcher is quietly doing the researcher's.

The contract doesn't need to be a legal document, but it does need to answer a small number of concrete questions in advance: who can change the signal's weight in the book, and within what range without asking; who decides if a signal gets turned off during a drawdown, and based on what trigger; who owns explaining performance to risk or to investors when something goes wrong; and what obligations the PM has to tell the researcher about overrides they've made, so the researcher's live-performance tracking doesn't silently diverge from what was actually traded.

A concrete example

A quant team formalizes this as a one-page agreement for each new signal going live: the researcher retains the right to pull the signal entirely if a data-quality issue is found, no PM approval required. The PM retains discretion to size the signal anywhere between 0.5x and 1.5x of the backtested target weight without consulting the researcher, but any change outside that band, or any structural override (like excluding a sector), triggers a joint review within a week. Both sides agree the PM will log every override in a shared file so the researcher's post-trade analysis reflects what was actually done, not what the model recommended. When a dispute arises later — the PM wants to cut the signal after a bad month the researcher thinks is just normal variance — the document doesn't resolve the disagreement automatically, but it makes clear whose call it ultimately is, which turns an emotional standoff into a scoped conversation.

What this means in practice

Most researcher-PM friction traces back to an assumption one side made about their own authority that the other side never agreed to. Writing the contract down before a signal goes live — even informally, in a shared document — costs an afternoon and prevents months of quiet resentment, and it gives both people a neutral reference to point to instead of relitigating the relationship every time a disagreement comes up.