A new way to spot the next financial crisis
Tech revolution means better financial system models. This will help identify risks
A more realistic way to model financial risk is emerging. It could help big banks and regulators spot potholes, even if it can’t stop people falling into them. The previous method for modeling the financial world was shown to have huge blindspots during the 2008 crisis. People didn’t behave in the rational ways economists had assumed and supposedly freak events appeared with alarming frequency. Investors and regulators have been hunting for alternatives ever since.
The latest candidate is “agentbased” modeling. U.K.-based startup Simudyne has joined with U.S. federally-funded research company Mitre Corp to turn an agent-based model of asset fire-sales and investor flight from banks and funds into a commercial product. They are using the building blocks of a model of the U.S. financial system that Mitre built for the U.S. Treasury.
Traditional financial models assumed that rational, well-informed people acted in efficient markets, allowing economists to analyze markets relatively simply with a few generalized rules. Agent-based modeling in- stead simulates market activity by creating dynamic computer programs with lots of agents, such as investors and banks, and seeing what happens when they interact.
Agents can be rational or irrational, greedy or fearful, and have all sorts of restrictions and motivations. Their actions and reactions are played out thousands of times over different time scales.
Such models have long been used in ecological research and military war games, but they faced a restriction: The more agents in the program and the faster they interact, the more computing power is needed. Until recently, it was impossible to create a large, complex model that could run quickly enough to be commercially useful, espe- cially for finance. Cloud computing has changed that, according to Concentric, a U.S.based simulations company, because model makers can now rent computing power and run multiple calculations in parallel.
The Simudyne-Mitre system, which is meant for central banks, regulators and perhaps the very largest global banks, can handle hundreds of millions of agents that represent real-world bank trading desks, hedge funds and other kinds of investors and lenders.
It is designed to show how panic and a sudden draining of liquidity could rush through markets and identify who would suffer most and who might fail.
Although this kind of model will be better at finding vulnerabilities in markets, it can never say exactly when or where panic will hit. But the sophistication of agent-based models could dupe users into thinking they have a complete view of all that can go wrong.
Better models are always welcome, but investors should be wary of assuming financial risk is fully known. One of the key lessons of 2008 is that overconfidence creates its own fragilities. That will never change.