Information Overload: Making Big Data Scalable
Last week, I had the pleasure of speaking at the SkyBridge Alternatives (SALT) Conference on a panel entitled, "Information Overload: Managing Risk & Data to Maximize Returns." I was joined by leaders from SkyBridge, Quadratic Capital, Wilshire Associates, and State Street Global Exchange for what evolved into an insightful and thought-provoking conversation. My panel lasted 30 minutes, but given the importance of risk management to securing investor confidence, increasing AUM, and driving returns in today's environment, the conversation could have continued for hours.
Below are additional thoughts I've put together on some of the questions that came up during our discussion.
Big data and risk management
What's the biggest change in risk management today?
There's a lot more regulation than there was 5-10 years ago. Markets move faster, and there's more data than ever to contend with. This is a mixed blessing. Of course, a good risk manager always wants to have as much information as possible. On the other hand, with more data to scrub and organize, there's a greater chance of a bad data point getting through.
The financial markets have become increasingly complex as turnaround times have become increasingly compressed. An example of this would be the new regulations that require FCMs to perform super-fast, complex calculations across diverse sets of securities. By merging risk management with innovative technologies, we are able to utilize a nonrelational database and parallelize our calculations across a server farm of computers in order to provide real-time information to our clients.
As increasing amounts of data become available, how can people avoid losing sight of the forest for the trees?
Big data is now part and parcel of providing risk management solutions. When we built our risk aggregation product, we had to develop powerful data visualizations (DVs) to display large volumes of data in meaningful ways.
Good DVs help draw your attention to what really matters. Think of it as a picture that's worth a thousand words -- or, rather, thousands of numbers and positions reacting to market factors, captured in real-time. In the past, risk managers would manually tackle the painstaking process of generating a huge table of numbers and go through them. Fortunately, today's DVs are well equipped to sort an increasing amount of data and identify the crucial points through interactive visualizations. Color coding, block segments, and nifty bubble charts help draw your attention to the important information. These features make it easy to drill down, look at the raw numbers, and perform further calculations without losing your focus. This combination of speed and analytics is critical in today's marketplace.
Risk management always has an issue of "garbage in, garbage out." Is there more that technology can do to help people make sure bad data points never get in?
Lots of filtering can be done to clean the data, but at some point, you'll always need a human to look at the results and give the final signoff. Zero errors is the goal, but it's also an abstraction.
The goal of DVs is to make errors stick out like sore thumbs. Then you can evaluate whether the sore thumb -- such as an outlying event that gets generated into a historical VaR calculation -- is a real possibility or just a bad data point. Ultimately, you want to have fewer sore thumbs.
Can you expand on the various measures that can be taken to avoid "bad data," such as Historical VaR and Parametric VaR?
Parametric VaR takes a bunch of data and does a statistical calculation. Bad data points will get you bad numbers, but you may not know where the bad data point is. With historical VaR, you see everything graphically -- making it easy to identify and investigate a bad number when it appears.
Stress testing
What constitutes a "realistic and elaborate" stress test?
A realistic stress test accounts for the market environment if a particular event occurs. Often, stress testing looks at the question of "How much will I lose?" or "How bad will the mark to market be?" A realistic and elaborate stress test goes a step further.
For instance, if you test what the world will look like if the market drops 10%, you need to take into account that it will be harder to trade, liquidity will dry up, and there could be a flight to quality in which assets that normally track each other suddenly move sharply in opposite directions. This happened to Long-Term Capital Management when it had hedged a bet on Russian debt by shorting US Treasuries to minimize interest rate risk. What would have been a hedge in normal situations became a disaster when the market environment changed. The worst time to trade is after a sharp movement in the market, when liquidity is at its lowest.
Monitor your leverage so that, if an event occurs, you can minimize the extent to which you have to take action. Stress tests can be utilized to estimate your margin requirements in the event of such an event ("stress margin").
Money is lost (and made on the other side) when firms are forced to liquidate to meet margin calls. The best case is that you can go back to your foxhole and wait things out. If you're overleveraged, however, you'll have no choice but to do battle.
Is there a way to bulletproof your stress testing?
I wouldn't dare say yes. More money has been lost by complacency or a false sense of security than just about anything else. You have to be respectful of the market. For every scenario you can come up with, there will be one you haven't considered. You should have a lot of people review and discuss your stress test. Play out what the world looks like if a black swan event occurs. Include potential margin calls, and have a plan to satisfy them. As Dirty Harry didn't say, "A man's got to know his potential limitations."
Dr. Smith brings a rare combination of business and academic experience to his role as CEO and co-founder at Imagine Software, with twenty-five-plus years of direct involvement in trading and risk management practices, mathematical modeling, and system development. Before ... View Full Bio
Tweet This