Group Co.
The Need for Trade Surveillance
Traders Magazine Online News, December 12, 2017
The financial industry has experienced great change and extraordinary challenges in the wake of the 2008 global financial crisis. The last several years in particular have seen unprecedented regulatory reforms across all aspects of the banking and trading industries, both within the United States and worldwide.
One of the biggest challenges that both regulators and compliance staff face is consistently ensuring that the market participants they supervise stay fully compliant with current market regulations. Regulators want to ensure that markets are fair and efficient, and compliance staff understand that the risks of failing to detect disruptive or manipulative trading behavior can be extremely costly.
As a result, rapid and accurate detection of prohibited market practices has become of paramount importance to all business in the financial industry.
TRADITIONAL PARAMETER-BASED SURVEILLANCE TOOLS
Traditional surveillance tools, often developed in house or available through various third-party vendors, are almost exclusively parameter- or rule-based. They function by applying if-then logic overlaid with configurable thresholds over arbitrary windows of time. In theory, the ability to configure these parameters lets users customize the alert output to their business needs; however, in practice, such configurations are often set to fit the capacity of the department reviewing the alerts, not on the validity of the results. This poses serious risk for compliance staff because the reasoning behind each threshold must stand up to regulatory scrutiny and be defensible in light of the firm’s business model and customer base. If the parameters are too tight, there is a risk of missing abusive conduct, and if they are too loose, then the reports are inundated with false positives.
Complicating this parameter-setting further, most every entity that requires surveillance, be it a small prop shop or an exchange, has a variety of trading strategies functioning at any given moment. One strategy might trade 1,000 contracts at a time, while another might never enter an order larger than a single contract. No matter where an account falls in the spectrum of trading volume or style, they all need to be surveilled for the same potential abusive behavior. This is where parameter-based surveillance tools can fall down. When a surveillance tool is comprised of a series of inflexible thresholds for things like volume totals, order size or cancellation rates, its effectiveness often will be limited to a very specific type of activity in products with a specific liquidity profile, leading to false negatives in products outside a specific profile.
The inflexible thresholds in traditional surveillance tools cause false negatives in other ways as well. For example, very few spoofing schemes are perfect. Most schemes are riddled with noise, unintended or opportunistic fills, late cancels, varying order sizes and general randomness outside of the trader’s control. Invariably, spoofing activity that would appear obvious even to an untrained eye will not be flagged due to some element of the activity not meeting an arbitrary threshold, such as the spoof order was only partially cancelled, or fills occurred on both sides of the market even if one side received the bulk of the beneficial executions. Any small deviation from the pre-set spoofing thresholds can break the continuous chain of precise events required to generate a parameter-based alert.
A MACHINE LEARNING APPROACH TO TRADE SURVEILLANCE
Machine learning is a field of artificial intelligence (AI) that involves the training of algorithms that don’t follow strict “if-then” logic or decision trees. Rather, the machine learning algorithms make use of “features.” These features allow for the algorithm to automatically aggregate clusters of activity based upon behavioral intent and score them based upon their similarity to trading patterns from known regulatory cases. See Figure A.
The features are applied such that every trade and transaction is surveilled, not just the worst offenders. In addition, regulators and compliance staff no longer have the burden of configuring parameter thresholds upon initial installation or the ongoing maintenance of those thresholds.
For the Full Paper and Illustrations, please click here:
For more information on related topics, visit the following channels:
Comments (0)