Just as Sibos delegates gathered in Vienna in September 2008, Lehman Brothers filed for bankruptcy, sending shockwaves through the global financial system. “There’s no point refining our existing risk models,” Till Guldimann, then Senior Vice President of Strategy at Sungard Capital Corporation, told fellow attendees as they questioned how the world’s biggest institutions failed to see it coming. “We were all looking at the nodes,” Guldimann said, “when the problem was the network.”
The finance world won’t fail to see the network for the nodes again. Institutions of all sizes, from the smallest local bank to the United Nations, have devised new ways to handle the sea of data that is growing at exponential rates. There is a “recognition that we now live in this hyper-connected world where information moves at the speed of light, and a crisis can be all around the world very, very quickly,” Robert Kirkpatrick, the director of Global Pulse, the Big Data research agency the UN established as a result of the crisis, told the International Peace Institute.
This year big data is once again at the forefront at Sibos, as experts discuss what has changed in five years and how better data, better algorithms and better analysis could help banks create early warning systems that could prevent another meltdown.
When Lehman Brothers collapsed, everybody was without a map,” explains Kimmo Soramäki, the founder and CEO of Financial Network Analytics, or FNA, a scheduled speaker at Sibos 2013, an annual conference that gathers over 8,000 banks from around the world. After 15 years in policy-making and multidisciplinary research at several central banks, including the European Central Bank and the Federal Reserve Bank of New York, Soramäki now calls himself a financial cartographer. His job is to help central banks and other large institutions make sense of the oceans of information they collect by mapping the data and creating visualizations.
“When you have a map you’re going to make different decisions and you’re going to communicate, as a central bank, differently about how you’re proceeding,” Soramäki says. “The banks would also better know what their actual risk positions are and so forth. There was a lot of uncertainty before because there wasn’t the data. No one knew the data and no one had the connections.”
Post-financial crisis, the focus particularly among large institutions and regulators has been on how big data can spot problems and weaknesses, Soramäki says. “They are interested in managing risks. They are focusing by design on the downside.”
But big data can also be a powerful tool for spotting opportunities. “If you’re a commercial bank, or someone active in the financial markets, these maps can provide you with information others don’t have,” Soramäki says. “They can show you things that others haven’t spotted yet. That’s the idea with the correlation maps that we do. We put a lot of data into one visualization so that a trader can quickly spot opportunities, like when some assets are in locations where they shouldn’t be. That’s a trading opportunity.”
The potential for big data to improve our understanding of the world by uncovering hidden connections is only beginning to become apparent. Mobile phone data can predict household incomes, within 90% accuracy, based on the amount people use their phones; credit card companies can predict unemployment from looking at changes in when drivers routinely fill up their car with fuel; and payments data has the potential to help predict GDP in real time.
SWIFT, a global organization that each day handles financial transactions, such as wire transfers, for over 9,000 banks, is using big data to look at payment flows so they can give an early indicator for GDP (see illustration). “There are lots of opportunities like that to create information products from the big data that banks and infrastructure operators have,” says FNA’s Soramäki.
One of the best-known big data projects is Google’s Flu Trends, which uses anonymized, aggregated web search keywords to display near real-time estimates of flu outbreaks. But trying to identify the early symptom of a financial crisis to build an early warning system for the global economy is a lot more complex, says Daniel Erasmus, the founder of the Digital Thinking Network, or DTN, which works with institutions to work out scenarios and analyze data on a web scale.
“You have much looser problem sets out there,” Erasmus says, comparing the challenge of tracking financial contagion versus something like a flu outbreak. “It is much more philosophy than it is programming. How you think about the problem is really important and how you approach the problem is really important.”
“An early warning system is really a set of insights, a set of eyes and it’s a set of actions,” he adds. “It is the brain, eyes, hands. You’ve got to know what you’re looking for, you’ve got to be able to look for it and you’ve got to act based on that.”
To better detect the unpredictable and the unknown, Erasmus, a scheduled speaker at Sibos 2013, and his colleagues have spent nearly a decade developing a system called NewsConsole that monitors hundreds of thousands of news sources.
“We started asking the question of analyzing all the news in the world — and I mean all the news in the world, not just a subset, not just the financial press, but really bringing all of that together and being able to analyze specific patterns that emerge and being able to indicate early on that we’re moving in one scenario direction or another,” he says.
“A blogger in India might have very interesting things to say about what’s happening in the gold market in India, much more than The Economist at a certain point in time. There’s no newspaper that can cover the world in the level of granularity than the whole of the Web does,” Erasmus adds.
For an early warning system, the challenge is to create a system that matches the qualitative and quantitative data flowing in against scenarios to generate crucial insights, Erasmus says.
His team has already had success with building scenarios. In 2006, NewsConsole helped them construct a scenario that foretold the U.S. sub-prime mortgage crisis for a large Dutch financial institution and a sequel scenario detailing how the U.S. situation could lead international banks to retreat to more conservative territory within national borders.
“In the scenario, we described four causes that could lead to a financial crisis and any combination of them would result in a financial crisis … one of them was the high levels of debt which were being held in the U.S.,” Eramus says. “So we walked through the steps, the initial issues, the responses and, then finally, a much less globalized, much more inward-looking society that would emerge in political and possibly economic realms. We’re certainly seeing the outlines of that in European politics in the past three years.”
But a blanket tracking of keywords based on scenarios of what might happen would not be enough to constitute an early warning system, he stresses. Computers, for now, would not be able to effectively find strong correlations.
“You need a social technical system: something in which people are somewhat involved in working together with technical capabilities, on the one hand, and on the other hand, you need a system which is able to understand at some level.”
Financial Network Analytics’s Soramäki says the red flags for an eventual early warning system could lie in the networks themselves. FNA’s Chief Scientist Samantha Cook was previously a member of the Google Flu Trends team and now works on problems such has predicting the magnitude of disruption a bank failure would have on payment systems. Soramäki believes monitoring key data like payment timings — late payments in particular — could underpin an early warning system.
“A lot of these maps we create are designed to detect anomalies, detect outliers; you can start to see things out of place once you see how that map of that data set should normally look like, you’re very quick to spot there’s something odd here so they alert you to take a look,” Soramäki says.
But for now, Soramäki agrees computers alone can’t provide an early warning defense system for banks. “Intuition is very important here. Intuition is a very specific skill, you can only have intuition about things you know very well. People who are familiar with the data that is being visualized immediately see things: some (of the data) is familiar which lets them know the map is correct, some information is new,” he says. Those in the know will immediately ask: What is this doing here?
In sum, “the more you’re tracking, the likelihood that you as an organization are going to be better prepared is certainly going to be higher,” concludes Erasmus.