Opening Cross: From Predictive to Productive: Practical Applications of Analytics

A case in point is predictive analytics, such as those of Boston-based EidoSearch, which scans historical data for patterns that match current events (or specific events or timeframes specified by users), identifies the impact of those same patterns in the past, and computes a potential outcome for the current pattern based on those. While a trader may recall the correlation between certain movements and outcomes in specific stocks, only vast computing power can run trillions of comparisons over a multitude of stocks, looking at data over many years.

So powerful is the potential of this analysis that EidoSearch has begun selling its underlying analytics engine in addition to the front-end and datafeed that carry its signals, so hardcore quantitative analysts can interrogate it directly, and even upload their own proprietary datasets to analyze using the vendor’s platform.

While EidoSearch’s main focus is on financial markets, there are distinctly valuable applications of this technology outside of Wall Street. For example, its technology is used by hospitals to analyze vast amounts of patient data that are collected but not analyzed statistically, or against historical information from a related peer group. To put this in perspective, imagine being able to whisk someone to hospital before they have a heart attack by analyzing patterns of events that led up to heart attacks in other patients.

And even within the financial markets, there are similar applications beyond pattern analysis to support trading and investment strategies, such as monitoring the beating heart of exchanges and marketplaces—that is, the technology platforms on which they run.

There are plenty of tools to monitor systems and performance in real time, allowing network administrators to see when traffic on a network or server is becoming dangerously high. In fact, interdealer broker Icap has invested significant efforts in expanding its data quality function to ensure it can handle ever-increasing volumes of incoming and internally generated data, setting up an offshore center in Manila to monitor data quality globally, in conjunction with implementing various in-house and third-party analysis tools.

But there’s also a case for predictive analysis to pre-empt even these tools—for example, looking for historical patterns in real time that would, for example, correlate market activity and events with infrastructure performance and show someone the likelihood of an interruption, slowdown or outage based on current market conditions, before the issue arises, allowing them to provision additional resources, or change or upgrade a component before it becomes overwhelmed. For instance, if you can spot telltale signs that a server—despite appearing to function perfectly well—may fail under a certain perfect storm of circumstances, a firm could change the server without interruption while the storm is still brewing, rather than waiting until it’s already at full force and the impact becomes critical.

This approach may help prevent incidents like those at the Moscow Exchange, which suffered two data glitches on consecutive days last week. On Monday, the exchange spotted erroneous data being displayed in its FORTS derivatives trading platform, such as open interest and volumes. And on Tuesday, one of six servers used to distribute data from the exchange’s foreign exchange market suffered interruptions over a seven-minute period. The exchange had backup servers to minimize any impact on clients, but in theory, in a situation like this, an exchange could use predictive analytics to failover to backup servers before an incident, and repair then seamlessly switch back to the production servers.

By making best use of predictive tools, incidents like this should last seven seconds—or, even better, never arise—rather than seven minutes.