Research: Using Advanced Analytics to Predict Future Financial Fault LinesNovember 2013
Why is this important? Consider that traditional methods of testing financial stress failed to predict the 2008 economic crisis, which exploded in September when:
- One of the country's largest investment banks, Lehman Brothers, declared bankruptcy.
- Bank of America announced its fire-sale purchase of another of the country's largest investment banks, Merrill Lynch.
- The multinational insurance firm American International Group (AIG) received an $85 billion bailout from the Federal Reserve Bank to avoid failure.
The U.S. Treasury Department's Office of Financial Research (OFR) recently asked MITRE to develop new ways to evaluate market dynamics in response to real and hypothetical shocks. (The OFR is responsible for developing analytical methods for assessing risks to the financial system and sharing its knowledge with the financial community.) The result of MITRE’s research is a prototype of various components of a financial crisis based on "agent-based modeling" techniques. These techniques can provide government financial regulators with powerful new analytic tools.
Tested on the Battlefield
Our prototype builds on capabilities MITRE originally developed, tested, and implemented for the Department of Defense. "Over the years, we have worked with the military to come up with better ways to model combat," explains Matt Koehler, a MITRE engineer and modeling and simulation expert. "The military's needs pushed FFRDCs [federally funded research and development centers], national labs, and academics to come up with ways of representing large numbers of tanks and planes, for example. This work taught us how to develop models of immense scale and complexity, which we can now apply to other areas of the government, including finance and emergency response."
Agent-based modeling differs from classic economic modeling techniques, which depend on examining historical patterns in market events. These classical models also rely on the assumption that historical patterns will repeat themselves and that people's actions are always rational and based on perfect knowledge of financial markets.
"But this isn't how real life works or real people act, as evidenced by the 2008 economic crash," says Shaun Brady, a MITRE systems engineer who is the research project leader. "Agent-based models incorporate human behavior into possible scenarios and are one tool we can use to improve our understanding of the impact of regulations on market dynamics and the economy."
"With agent-based modeling, regulators can develop models that better account for uncertainty—and can better measure the connections among different market entities," adds Brian Tivnan, a MITRE engineer and team lead in advanced simulation technologies. "We believe agent-based modeling will allow the regulatory community to measure cascading effects across the financial system."
Recent advances in computing power and simulation capabilities have made possible much more accurate analysis of market dynamics. For example, our models can answer questions such as:
- How will an individual's decision affect the rest of the group?
- How will it react?
- How far does the chain reaction go?
Analyzing the Dynamic Financial System
MITRE brought together a multi-disciplinary team from across the corporation to demonstrate the capabilities of agent-based modeling to the OFR—pulling from our work with such agencies as the Census Bureau and Department of Homeland Security to develop, apply, and vet our agent-based modeling capabilities in different areas. For example, in one research project, we examined the possible impact of disease pandemics on business-continuity planning.
The MITRE team works closely with OFR, transitioning information and technology it the evolves, such as our work on modeling the interactions between banks and leveraged asset managers that shows how "fire sales" develop as asset values plunge.
MITRE engineers also advised the OFR on its development of a white paper titled "Using Agent-Based Models for Analyzing Threats to Financial Stability." The paper, authored by OFR Research Principal Richard Bookstaber, discusses the shortcomings of classical economic models and the potential improvements offered by agent-based models. (The New York Times cited the important possibilities of these models, along with MITRE's contribution, in a January 2013 story, "Clouds Seen in Regulators' Crystal Ball for Banks."
"The only way to understand systems like the financial system is to build models and study them," Koehler says. "The only way to have analytically grounded policies on financial markets is to build tools to analyze them. Now we have the technology to scale to the level of something as large and complex as the global financial marketplace. Using complexity science and agent-based models, we can represent the entire market."
With all the government's new data and tools—including ABM—it hopes that in the future it can spot potential trouble before it happens and give organizations time to take steps that will stabilize the situation and avert disaster.
—by Maria S. Lee