MIT 2017

A more realistic look at risk

Standard methods of risk management don’t work in a crisis but a revolutionary approach called agent-based modelling, which the University of California uses, is appropriate for investors, says chief risk officer for the university’s $110 billion pension and endowment, Rick Bookstaber.

The failings of standard methods of risk management, which use historical data and stress testing, are rooted in the fact that they deal with static portfolios. But crises stem from a series of events, not just one, with many agents interacting and changing the environment constantly, Bookstaber said.

“Why do we need an agent-based model for risk management? Because we are human, not automatons, and we interact with our environment,” he argued. “A lot of economics assumes there’s only one agent, and everyone behaves the same. But there are a set of agents with heterogeneous heuristics, and they interact and that changes the environment and you get a cascade period by period.”

“Because we are human, we create and innovate, so the world is always filled with surprises and things we couldn’t anticipate. This leads to radical uncertainty, but this is not central to economics. Not only does the future not look like the past but the future crises won’t look like past crises.”

Bookstaber, who was speaking at the Fiduciary Investors Symposium at the Massachusetts Institute of Technology, said the financial system consists of “computational irreducibility”; in other words, it’s a system that can’t be reduced to a set of equations. Yet most of the economic theories the industry practices are based on trying to solve problems with equations.

“Interactions create dynamic complexity and we can’t solve it, we have to live with it,” he said. “The world is not ergodic.”

The agent-based model uses heuristics instead of optimisation, and an open, simulated environment where many paths can emerge.

“In standard economics, you have an atomistic equilibrium world – whatever you do, or others do, doesn’t change the environment. That doesn’t make sense. In an agent-based model, you have an interactive, changing world that evolves over time.”

In a financial crisis, liquidity, leverage and concentration are the key areas that need to be dealt with, he said. In practice at the University of California, this plays out by logging different levels of liquidity and leverage on a heat map and determining where the fund’s holdings are at any point in time.

“Then we can see how bad things could be with a stress scenario,” he said. “Each square shows the distribution of assets and thus a portfolio over time. This is not symmetric, movement into the tails is not smooth, risk doesn’t resolve at a constant rate.

“Value at Risk [VaR] doesn’t deal with a crisis. A tail event or a regime change is not a bad draw from the urn, it’s a different urn. To say it’s a tail event or a black swan is a cop out. It’s saying something bad happened, but not looking at which market dynamics led to that and being prepared to deal with that.”

The objectives of the agent-based approach are not only managing risk better but taking advantage of opportunities that might arise because of crises.

“It’s to manage risk better, and to understand any market dislocation and how it could affect me and the narrative for how it might cascade out,” Bookstaber said. “Then, seeing the crisis, and knowing it and how it works through the narrative, you can buy in and generate returns. [If] someone else is screaming for liquidity and you can provide that, it’s a social value to the market. If enough people with enough capital are involved, the liquidity can dampen the crisis. You can’t solve for life, you have to live it.”

Bookstaber’s book on this subject is called The End of Theory.

Join the discussion