Product-ControlHard

3 min read · 507 words

What the seat actually does

A trading desk marks its own positions. Somebody independent has to check those marks against the market, produce the profit and loss from them, and be able to explain the number in terms of what actually moved. That is product control, and it is the reason a bank's reported earnings are not simply what the desks said they were.

The core exercise is the explain. Yesterday's risk positions plus today's market moves should predict today's profit and loss almost exactly. When they do not, either the risk was mismeasured, the marks are wrong, or something is in the book that nobody described — and all three are worth finding on the same day.

• Independent price verification — testing the desk's marks against sources the desk does not control.
• Daily profit and loss, split into what the market did and what the desk did.
• Valuation adjustments and reserves — for illiquidity, model uncertainty and counterparty credit.
• Reporting, into the accounts that eventually become the annual report.

A day, and where it goes

• The explain — yesterday's risk against today's moves, and the residual nobody can account for.
• Marks. Positions where the desk's price and the independent one differ, and by how much.
• New trades — anything booked yesterday that the systems do not value the way the desk does.
• Month end, when reserves are set and the arguments about them happen.

What it is measured on

• Unexplained profit and loss, which should be small and random rather than small and one-directional.
• Verification coverage — the share of the book actually tested against an independent price.
• Adjustments found, and whether they were found here or by somebody outside the firm.
• Timeliness. A correct number produced three days late has not controlled anything.

What it touches on this site

• What is being valued — valuation and the payoff pages behind every structured position.
• Where the marks are hardest — alternatives and anything without a screen price.
• The desk on the other side — trading and structuring.
• When it goes wrong — 1995, where an account nobody controlled held the losses.

How it goes wrong

• Verifying against a source the desk supplied. Independence that runs through the same broker is not independence.
• A residual that is always the same sign. Noise averages to nothing; a bias is a message.
• Reserves negotiated rather than calculated, which turns a control into a discussion about somebody's year.
• Marking the model instead of the market. A position with no observable price is valued by an assumption, and the assumption is where the profit is.

Concepts to master

• Profit and loss should be explainable by risk, and the part that is not is the finding.
• Fair value has levels. An observable price, an observable input, and a number that came from a model are three different claims.
• A reserve is an admission of uncertainty, taken up front rather than discovered later.
• Marking is not settling — see valuation and liquidity.