Blog article
See all stories »

Deploying Predictive Analytics? Check Your Data Integrity First

The explosion of predictive analytics in financial services is undeniable. Layering on advancements in big data, billions of dollars are now dedicated to analytics spend annually. Dozens of startups dive in every year to develop the tools to tackle increasingly complex datasets at mind-blowing volumes, often deploying some variation or combination of artificial intelligence (AI) methods to shine new lights on new insights.

Once seen as a proprietary “secret sauce”, the industry’s largest banks and institutions now eye new ways to weave analytics directly into their operational processes, tapping into big data and driving digitalisation initiatives.  Solution providers are likewise arming our platforms with tools to derive business intelligence from granular data points like anomalies, policy exceptions, user behaviour and process performance.

There is a drive to realise data’s value at every turn, and it truly runs the gamut. This has been called a generational, or revolutionary, trend by industry observers. But perhaps one of the greatest and more recent signs of that progress is the directness with which analytics are now embedded into trade execution.

Unless competing purely on speed or ultra high-frequency trading (HFT) strategies, most firms have preferred to keep at least some layer of control—or human eyes—between fresh analytics being delivered into the trading desk, and their trading programs. For one thing, analytics engines can make mistakes, the data being analysed might be incomplete or inaccurate and so many firms will run their own data verification processes before actioning.

Even if the data has the highest possible integrity, the information is often complex and nuanced and requires further analysis against current portfolio holdings, risk tolerance and even current market conditions and pricing. Often, there are good reasons to wait.

But with more trading desks raising their execution technology’s sophistication, and analytics becoming even more reliable, a wider segment of traders are now rethinking that stance—at least for certain situations.

An illustration: a large equity option position whose expiry is set in anticipation of an earnings or dividend announcement from the company whose stock underlies the option. The timing of those announcements is fairly predictable, but can just as easily change for a number of reasons. If that change upsets the timing of the option, that can spell either wasted money putting the option on to begin with or worse, the option becoming vulnerable to being picked off by aforementioned HFT players. Analytics using natural-language processing (NLP) can now read through company news to flag these potential failed-trade situations—and they are relatively straightforward, without much need for interpretation, and sometimes demand urgent attention (for instance, if timing changes occur on the day of).

Coding analytics directly into trade execution won’t play all the time. But spots like these, where it can be done effectively, are increasingly under consideration. As firms do that, the bigger problem isn’t the veracity of the information coming in, so much as being able to link it up with past transactional data and current positional data to accurately inform how and where a trade order should be placed (or modified). Indeed, with as much institutional energy and investment put into analytics and front-end trading platforms, the ability to achieve effective data integrity still often lacks.

To construct more predictive analytics-driven automated trading, this second leg of the process is crucial. Take the simple—but again, potentially costly—options trade: for a highly-fragmented market, traders’ systems must be able to confirm not only the price, size, and timing, but also execution venue availability and any specific constraints related to the option contract at hand. You can’t unload or fix what you don’t know, and tiny mistakes in this area can add up to six-figure losses or savings on a single trade. That is why data integrity—even if it lies more in the background than the latest new-fangled analytics do—needs attention.

As derivative and credit markets continue towards electronification, boosting efforts at automated and algorithmic trading, the same can be said for them.  Predictive analytics will only take us so far, and so fast, if a firm’s internal systems are incapable of storing critical ancillary transaction information with pinpoint accuracy.

The data must be timely too, being updated in close to real-time, and complete - integrating with more nimble order and execution management systems. All sides must be in sync.

Getting the trade right starts with the right data—however new or fashionable the source of the signal.

Deploying Predictive Analytics? Check Your Data Integrity
4804

Comments: (0)

Blog group founder

Retired Member

Member since

19 Mar 2009

Location

Blog posts

5,573

Comments

6,020

More from Retired

This post is from a series of posts in the group:

Fintech

Fintech discussions and conversations around the development of fintech.


See all