Quantitative Risk Analyst

The problem.

A limit is only as good as the measure underneath it. Most risk measures are estimated from too little data, assume a distribution the market does not have, and are quoted to more decimal places than they deserve. Making them useful means knowing exactly where each one breaks.

What you would do.

• Build and maintain the tail measures the firm uses, with estimation error stated alongside the estimate.
• Run the scenario library: historical episodes, hypothetical shocks, and operational ones where the position is fine and the venue is not.
• Attribute exposure to the decisions that produced it, so a limit breach is traceable rather than aggregate.
• Backtest the risk models themselves and report where they were wrong, at a cadence that makes it uncomfortable.
• Give the risk manager numbers in a form that supports a decision under time pressure.

What we would want evidence of.

• You can explain what a tail estimate assumes and what it does when the assumption fails.
• You have backtested a risk model and reported that it underestimated.
• You are careful about annualisation. You do not apply a square root of two hundred and fifty-two to a market that trades continuously.
• You have built something used in a live decision rather than in a report.
• You are precise about the difference between an estimate and a limit.

What would make this a bad fit.

This role is a poor fit if you want the modelling without the operational scenarios. Half of this role is the boring half. It is the half that has saved firms.

This role is a poor fit if producing a number is the deliverable. Here the deliverable is a number somebody is willing to act on.