The risky side of statistics

Natalia Nolde's risk assessment research has a wide range of applications.

Natalia Nolde is a UBC statistician with a risky specialty: predicting and assessing extreme events. Her work faces two major challenges. Extreme events, by definition, don’t happen often and so related data are scarce. They’re also incredibly complex. But if Nolde and colleagues are able to overcome the challenges, their research could help improve everything from dyke design to financial models.

What kind areas does risk assessment touch on?

It covers a wide range of areas such as hydrology and finance. In both cases, we’re interested in extreme events, and extreme events don’t occur in isolation. Think about a dyke in the Netherlands. You have to consider not only the level of the sea, but trends in wind patterns. You also have to consider river networks and model extremes at multiple locations. Conceptually it’s very similar to modeling multiple stocks in a portfolio. 

How did you become interested in risk assessment issues?

The subprime mortgage crisis took place while I was conducting my PhD. The crisis has been blamed on the misuse of simple mathematical models that account for interdependencies. A lot of my work was motivated by this event.

What do people misunderstand about your research?

Statistics is often viewed as a service discipline. Some statisticians do work in consultant roles but many, including myself, work on the theoretical side, developing new estimation theories and methodologies.

Scotiabank recently funded a $2-million cybersecurity and risk analytics initiative at UBC. How will the initiative assist your research?

The data we typically use is publicly available, and there are limitations to it. The new partnership with Scotiabank will allow us to access high-quality, real-world data and enhance our knowledge of risk related questions faced by the banking industry. UBC students will also be able to intern at Scotiabank and tackle concrete problems affecting the bank.

Geoff Gilliard