Matrix-IFS

New hope for CIOs and their resources

From a compliance standpoint, the 2008 financial crisis spawned a proliferation of third-party solutions and in-house developments offering various systems to automate the Market Surveillance process. At the time, these rule-based solutions presented a well-received architecture and strong blend of analytics and scalability, with the flexibility to add detection scenarios and data feeds to crunch millions of daily transactions.

However, the issue of debilitating False Positive (FP) was born when these systems went online. Compliance officers worldwide became overwhelmed by the amount of noise these systems generated. The problem further intensified as detection scenarios grew more elaborate and intertwined with legitimate trading-related activity.

Like a violent single-cell organism, this problem has exponentially multiplied, far exceeding the volume of true positives. Today, “well-tuned” market surveillance systems can generate 70-90 percent FP, universally causing a painful drain on both resources and budgets. Regardless of your chosen system, whether developed by a solution provider, in-house, and/or a possible hybrid approach, false positives appeared the nanosecond you defined a threshold and ran your data against it.

False positives are by far the most efficiency-inhibiting factor of the investigation process. FPs pose a lingering risk as valid reports and alerts await analysis. Every firm faces this issue. Some have surrendered to it and treat the resulting overhead as part of the cost of market surveillance. Others are boldly trying to tune/calibrate/validate their thresholds—a costly, long, and tediously excruciating iterative process where every FP percent reduction is hailed. But, this calibration exercise must be extended for each scenario and periodically repeated for the entire program, as FPs demonstrate the superiority of their digital DNA, aided by additional data feeds and/or changing market conditions.

  Today, ’’ ‘well-tuned’ market surveillance systems can generate 70-90 percent false positives, universally causing a painful drain on both resources and budgets  

Today’s financial institutions face a constant decline in compliance resources and related budgets, therefore the FP axiom now attracts more attention. But, as mentioned above, dealing with this issue is complex at best. So, are CIOs and Compliance officers condemned to follow Dante’s path? (If you feel this is an exaggeration, talk to your compliance officer, look at their face, and you’ll get it.) Are you doomed? Is resistance truly futile–or is it?

As in medicine, dealing with an outbreak or uncontrollable cell multiplication is accomplished by tackling multiple parallel aspects: the symptoms and the source. Try to prevent the existing situation from deteriorating further while keeping an eye on the prize. Isolate the source (your “Patient Zero”) and take control. Make sure people can live with it at the very least, develop a vaccine at best.

So far, the financial industry has reactively dealt with the symptoms of FPs, trying to reduce the phenomenon by calibrating thresholds per different personas monitored statically. However, a sustainable and ongoing calibration program requires a more dynamic approach.

Your thousands of FPs (hopefully labeled as such) hold the key to better understand their origin and decipher their digital DNA. Most importantly, sophisticated examination could uncover the correlations governing them under different circumstances (i.e., market conditions, financial instruments, personas, etc.) But how can you efficiently address this Herculean task?

Advanced analytics, such as data mining, predictive analytics, and true machine learning has been around for years. For instance, industrial engineering utilized various regression models to predict seasonal sales and inventory management 20 years ago, and early CRM solutions predicted campaign-driven customer conversions. Today, these terms are often greatly abused and exaggerated, especially in FI, with some vendors claiming to deliver AI capabilities while merely providing advanced regressions, which creates more confusion and skepticism toward the technology. But there is hope.

Although the underlying methodology has not vastly changed, the various predictive/learning engines and coding languages provide astonishing speed, delivering record breaking time-to-production at a reasonable price. Today’s tools require minimal data setup but can deliver true insights and actionable metrics.

Approaching the FP issue in a predictive manner to understand the nature of an alert as it is generated can provide immediate transparency of your system, ahead of any investigation by a compliance officer. This technique affords the ability to control the population and prioritize accordingly. Existing tools provide excellent classification models, so after a relatively short time, accuracy becomes reliable to a point where you can extract relevant thresholds and calibrate accordingly.

After linking these thresholds to the appropriate correlations, the next step is to establish a dynamic mechanism to automatically adjust them according to the various impactful parameters/conditions.

By applying this methodology, we have discovered that some data used for market surveillance has little or no impact on the validity of FP alerts, while other parameters not currently used have a crucial weight. These discoveries significantly impact performance and effectiveness.

Tying it all up in a dedicated “Health & Welfare” dashboard provides complete transparency and control to various stakeholders (business, IT, etc.) over the reports/alerts being generated, as well as the ability to predict alert validity around certain news and market conditions.

Applying such a dashboard driven by advanced analytics tools will dramatically shorten the calibration cycle and provide the ability to run “what if” analysis on recommended thresholds. This approach will enable your compliance team to uncover and respond to illicit activity far faster and more accurately.

We are year(s) away from replacing rules-based engines with pure machine learning algorithms. Even if deployed now, training takes time. But you will also gain valuable experience with different tools and methodologies, and learn the importance of your newest team member–the data scientist, who will help you initiate advanced analytics to achieve tangible results, unparalleled efficiency, and excellent ROI.

Now you can stop wasting critical resources and budget on false positives and step into the fast lane of efficiency with your own, advanced analytics-driven FP market surveillance approach.